1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * NET3 Protocol independent device support routines.
5 * Derived from the non IP parts of dev.c 1.0.19
7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
8 * Mark Evans, <evansmp@uhura.aston.ac.uk>
11 * Florian la Roche <rzsfl@rz.uni-sb.de>
12 * Alan Cox <gw4pts@gw4pts.ampr.org>
13 * David Hinds <dahinds@users.sourceforge.net>
14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
15 * Adam Sulmicki <adam@cfar.umd.edu>
16 * Pekka Riikonen <priikone@poesidon.pspt.fi>
19 * D.J. Barrow : Fixed bug where dev->refcnt gets set
20 * to 2 if register_netdev gets called
21 * before net_dev_init & also removed a
22 * few lines of code in the process.
23 * Alan Cox : device private ioctl copies fields back.
24 * Alan Cox : Transmit queue code does relevant
25 * stunts to keep the queue safe.
26 * Alan Cox : Fixed double lock.
27 * Alan Cox : Fixed promisc NULL pointer trap
28 * ???????? : Support the full private ioctl range
29 * Alan Cox : Moved ioctl permission check into
31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
32 * Alan Cox : 100 backlog just doesn't cut it when
33 * you start doing multicast video 8)
34 * Alan Cox : Rewrote net_bh and list manager.
35 * Alan Cox : Fix ETH_P_ALL echoback lengths.
36 * Alan Cox : Took out transmit every packet pass
37 * Saved a few bytes in the ioctl handler
38 * Alan Cox : Network driver sets packet type before
39 * calling netif_rx. Saves a function
41 * Alan Cox : Hashed net_bh()
42 * Richard Kooijman: Timestamp fixes.
43 * Alan Cox : Wrong field in SIOCGIFDSTADDR
44 * Alan Cox : Device lock protection.
45 * Alan Cox : Fixed nasty side effect of device close
47 * Rudi Cilibrasi : Pass the right thing to
49 * Dave Miller : 32bit quantity for the device lock to
50 * make it work out on a Sparc.
51 * Bjorn Ekwall : Added KERNELD hack.
52 * Alan Cox : Cleaned up the backlog initialise.
53 * Craig Metz : SIOCGIFCONF fix if space for under
55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
56 * is no device open function.
57 * Andi Kleen : Fix error reporting for SIOCGIFCONF
58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
59 * Cyrus Durgin : Cleaned for KMOD
60 * Adam Sulmicki : Bug Fix : Network Device Unload
61 * A network device unload needs to purge
63 * Paul Rusty Russell : SIOCSIFNAME
64 * Pekka Riikonen : Netdev boot-time settings code
65 * Andrew Morton : Make unregister_netdevice wait
66 * indefinitely on dev->refcnt
67 * J Hadi Salim : - Backlog queue sampling
68 * - netif_rx() feedback
71 #include <linux/uaccess.h>
72 #include <linux/bitops.h>
73 #include <linux/capability.h>
74 #include <linux/cpu.h>
75 #include <linux/types.h>
76 #include <linux/kernel.h>
77 #include <linux/hash.h>
78 #include <linux/slab.h>
79 #include <linux/sched.h>
80 #include <linux/sched/mm.h>
81 #include <linux/mutex.h>
82 #include <linux/string.h>
84 #include <linux/socket.h>
85 #include <linux/sockios.h>
86 #include <linux/errno.h>
87 #include <linux/interrupt.h>
88 #include <linux/if_ether.h>
89 #include <linux/netdevice.h>
90 #include <linux/etherdevice.h>
91 #include <linux/ethtool.h>
92 #include <linux/skbuff.h>
93 #include <linux/bpf.h>
94 #include <linux/bpf_trace.h>
95 #include <net/net_namespace.h>
97 #include <net/busy_poll.h>
98 #include <linux/rtnetlink.h>
99 #include <linux/stat.h>
101 #include <net/dst_metadata.h>
102 #include <net/pkt_sched.h>
103 #include <net/pkt_cls.h>
104 #include <net/checksum.h>
105 #include <net/xfrm.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/module.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #include <linux/delay.h>
112 #include <net/iw_handler.h>
113 #include <asm/current.h>
114 #include <linux/audit.h>
115 #include <linux/dmaengine.h>
116 #include <linux/err.h>
117 #include <linux/ctype.h>
118 #include <linux/if_arp.h>
119 #include <linux/if_vlan.h>
120 #include <linux/ip.h>
122 #include <net/mpls.h>
123 #include <linux/ipv6.h>
124 #include <linux/in.h>
125 #include <linux/jhash.h>
126 #include <linux/random.h>
127 #include <trace/events/napi.h>
128 #include <trace/events/net.h>
129 #include <trace/events/skb.h>
130 #include <linux/inetdevice.h>
131 #include <linux/cpu_rmap.h>
132 #include <linux/static_key.h>
133 #include <linux/hashtable.h>
134 #include <linux/vmalloc.h>
135 #include <linux/if_macvlan.h>
136 #include <linux/errqueue.h>
137 #include <linux/hrtimer.h>
138 #include <linux/netfilter_ingress.h>
139 #include <linux/crash_dump.h>
140 #include <linux/sctp.h>
141 #include <net/udp_tunnel.h>
142 #include <linux/net_namespace.h>
143 #include <linux/indirect_call_wrapper.h>
144 #include <net/devlink.h>
146 #include "net-sysfs.h"
148 #define MAX_GRO_SKBS 8
150 /* This should be increased if a protocol with a bigger head is added. */
151 #define GRO_MAX_HEAD (MAX_HEADER + 128)
153 static DEFINE_SPINLOCK(ptype_lock);
154 static DEFINE_SPINLOCK(offload_lock);
155 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
156 struct list_head ptype_all __read_mostly; /* Taps */
157 static struct list_head offload_base __read_mostly;
159 static int netif_rx_internal(struct sk_buff *skb);
160 static int call_netdevice_notifiers_info(unsigned long val,
161 struct netdev_notifier_info *info);
162 static int call_netdevice_notifiers_extack(unsigned long val,
163 struct net_device *dev,
164 struct netlink_ext_ack *extack);
165 static struct napi_struct *napi_by_id(unsigned int napi_id);
168 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
171 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
173 * Writers must hold the rtnl semaphore while they loop through the
174 * dev_base_head list, and hold dev_base_lock for writing when they do the
175 * actual updates. This allows pure readers to access the list even
176 * while a writer is preparing to update it.
178 * To put it another way, dev_base_lock is held for writing only to
179 * protect against pure readers; the rtnl semaphore provides the
180 * protection against other writers.
182 * See, for example usages, register_netdevice() and
183 * unregister_netdevice(), which must be called with the rtnl
186 DEFINE_RWLOCK(dev_base_lock);
187 EXPORT_SYMBOL(dev_base_lock);
189 static DEFINE_MUTEX(ifalias_mutex);
191 /* protects napi_hash addition/deletion and napi_gen_id */
192 static DEFINE_SPINLOCK(napi_hash_lock);
194 static unsigned int napi_gen_id = NR_CPUS;
195 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
197 static seqcount_t devnet_rename_seq;
199 static inline void dev_base_seq_inc(struct net *net)
201 while (++net->dev_base_seq == 0)
205 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
207 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
209 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
212 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
214 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
217 static inline void rps_lock(struct softnet_data *sd)
220 spin_lock(&sd->input_pkt_queue.lock);
224 static inline void rps_unlock(struct softnet_data *sd)
227 spin_unlock(&sd->input_pkt_queue.lock);
231 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
234 struct netdev_name_node *name_node;
236 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
239 INIT_HLIST_NODE(&name_node->hlist);
240 name_node->dev = dev;
241 name_node->name = name;
245 static struct netdev_name_node *
246 netdev_name_node_head_alloc(struct net_device *dev)
248 struct netdev_name_node *name_node;
250 name_node = netdev_name_node_alloc(dev, dev->name);
253 INIT_LIST_HEAD(&name_node->list);
257 static void netdev_name_node_free(struct netdev_name_node *name_node)
262 static void netdev_name_node_add(struct net *net,
263 struct netdev_name_node *name_node)
265 hlist_add_head_rcu(&name_node->hlist,
266 dev_name_hash(net, name_node->name));
269 static void netdev_name_node_del(struct netdev_name_node *name_node)
271 hlist_del_rcu(&name_node->hlist);
274 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
277 struct hlist_head *head = dev_name_hash(net, name);
278 struct netdev_name_node *name_node;
280 hlist_for_each_entry(name_node, head, hlist)
281 if (!strcmp(name_node->name, name))
286 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
289 struct hlist_head *head = dev_name_hash(net, name);
290 struct netdev_name_node *name_node;
292 hlist_for_each_entry_rcu(name_node, head, hlist)
293 if (!strcmp(name_node->name, name))
298 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
300 struct netdev_name_node *name_node;
301 struct net *net = dev_net(dev);
303 name_node = netdev_name_node_lookup(net, name);
306 name_node = netdev_name_node_alloc(dev, name);
309 netdev_name_node_add(net, name_node);
310 /* The node that holds dev->name acts as a head of per-device list. */
311 list_add_tail(&name_node->list, &dev->name_node->list);
315 EXPORT_SYMBOL(netdev_name_node_alt_create);
317 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
319 list_del(&name_node->list);
320 netdev_name_node_del(name_node);
321 kfree(name_node->name);
322 netdev_name_node_free(name_node);
325 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
327 struct netdev_name_node *name_node;
328 struct net *net = dev_net(dev);
330 name_node = netdev_name_node_lookup(net, name);
333 /* lookup might have found our primary name or a name belonging
336 if (name_node == dev->name_node || name_node->dev != dev)
339 __netdev_name_node_alt_destroy(name_node);
343 EXPORT_SYMBOL(netdev_name_node_alt_destroy);
345 static void netdev_name_node_alt_flush(struct net_device *dev)
347 struct netdev_name_node *name_node, *tmp;
349 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
350 __netdev_name_node_alt_destroy(name_node);
353 /* Device list insertion */
354 static void list_netdevice(struct net_device *dev)
356 struct net *net = dev_net(dev);
360 write_lock_bh(&dev_base_lock);
361 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
362 netdev_name_node_add(net, dev->name_node);
363 hlist_add_head_rcu(&dev->index_hlist,
364 dev_index_hash(net, dev->ifindex));
365 write_unlock_bh(&dev_base_lock);
367 dev_base_seq_inc(net);
370 /* Device list removal
371 * caller must respect a RCU grace period before freeing/reusing dev
373 static void unlist_netdevice(struct net_device *dev)
377 /* Unlink dev from the device chain */
378 write_lock_bh(&dev_base_lock);
379 list_del_rcu(&dev->dev_list);
380 netdev_name_node_del(dev->name_node);
381 hlist_del_rcu(&dev->index_hlist);
382 write_unlock_bh(&dev_base_lock);
384 dev_base_seq_inc(dev_net(dev));
391 static RAW_NOTIFIER_HEAD(netdev_chain);
394 * Device drivers call our routines to queue packets here. We empty the
395 * queue in the local softnet handler.
398 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
399 EXPORT_PER_CPU_SYMBOL(softnet_data);
401 /*******************************************************************************
403 * Protocol management and registration routines
405 *******************************************************************************/
409 * Add a protocol ID to the list. Now that the input handler is
410 * smarter we can dispense with all the messy stuff that used to be
413 * BEWARE!!! Protocol handlers, mangling input packets,
414 * MUST BE last in hash buckets and checking protocol handlers
415 * MUST start from promiscuous ptype_all chain in net_bh.
416 * It is true now, do not change it.
417 * Explanation follows: if protocol handler, mangling packet, will
418 * be the first on list, it is not able to sense, that packet
419 * is cloned and should be copied-on-write, so that it will
420 * change it and subsequent readers will get broken packet.
424 static inline struct list_head *ptype_head(const struct packet_type *pt)
426 if (pt->type == htons(ETH_P_ALL))
427 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
429 return pt->dev ? &pt->dev->ptype_specific :
430 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
434 * dev_add_pack - add packet handler
435 * @pt: packet type declaration
437 * Add a protocol handler to the networking stack. The passed &packet_type
438 * is linked into kernel lists and may not be freed until it has been
439 * removed from the kernel lists.
441 * This call does not sleep therefore it can not
442 * guarantee all CPU's that are in middle of receiving packets
443 * will see the new packet type (until the next received packet).
446 void dev_add_pack(struct packet_type *pt)
448 struct list_head *head = ptype_head(pt);
450 spin_lock(&ptype_lock);
451 list_add_rcu(&pt->list, head);
452 spin_unlock(&ptype_lock);
454 EXPORT_SYMBOL(dev_add_pack);
457 * __dev_remove_pack - remove packet handler
458 * @pt: packet type declaration
460 * Remove a protocol handler that was previously added to the kernel
461 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
462 * from the kernel lists and can be freed or reused once this function
465 * The packet type might still be in use by receivers
466 * and must not be freed until after all the CPU's have gone
467 * through a quiescent state.
469 void __dev_remove_pack(struct packet_type *pt)
471 struct list_head *head = ptype_head(pt);
472 struct packet_type *pt1;
474 spin_lock(&ptype_lock);
476 list_for_each_entry(pt1, head, list) {
478 list_del_rcu(&pt->list);
483 pr_warn("dev_remove_pack: %p not found\n", pt);
485 spin_unlock(&ptype_lock);
487 EXPORT_SYMBOL(__dev_remove_pack);
490 * dev_remove_pack - remove packet handler
491 * @pt: packet type declaration
493 * Remove a protocol handler that was previously added to the kernel
494 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
495 * from the kernel lists and can be freed or reused once this function
498 * This call sleeps to guarantee that no CPU is looking at the packet
501 void dev_remove_pack(struct packet_type *pt)
503 __dev_remove_pack(pt);
507 EXPORT_SYMBOL(dev_remove_pack);
511 * dev_add_offload - register offload handlers
512 * @po: protocol offload declaration
514 * Add protocol offload handlers to the networking stack. The passed
515 * &proto_offload is linked into kernel lists and may not be freed until
516 * it has been removed from the kernel lists.
518 * This call does not sleep therefore it can not
519 * guarantee all CPU's that are in middle of receiving packets
520 * will see the new offload handlers (until the next received packet).
522 void dev_add_offload(struct packet_offload *po)
524 struct packet_offload *elem;
526 spin_lock(&offload_lock);
527 list_for_each_entry(elem, &offload_base, list) {
528 if (po->priority < elem->priority)
531 list_add_rcu(&po->list, elem->list.prev);
532 spin_unlock(&offload_lock);
534 EXPORT_SYMBOL(dev_add_offload);
537 * __dev_remove_offload - remove offload handler
538 * @po: packet offload declaration
540 * Remove a protocol offload handler that was previously added to the
541 * kernel offload handlers by dev_add_offload(). The passed &offload_type
542 * is removed from the kernel lists and can be freed or reused once this
545 * The packet type might still be in use by receivers
546 * and must not be freed until after all the CPU's have gone
547 * through a quiescent state.
549 static void __dev_remove_offload(struct packet_offload *po)
551 struct list_head *head = &offload_base;
552 struct packet_offload *po1;
554 spin_lock(&offload_lock);
556 list_for_each_entry(po1, head, list) {
558 list_del_rcu(&po->list);
563 pr_warn("dev_remove_offload: %p not found\n", po);
565 spin_unlock(&offload_lock);
569 * dev_remove_offload - remove packet offload handler
570 * @po: packet offload declaration
572 * Remove a packet offload handler that was previously added to the kernel
573 * offload handlers by dev_add_offload(). The passed &offload_type is
574 * removed from the kernel lists and can be freed or reused once this
577 * This call sleeps to guarantee that no CPU is looking at the packet
580 void dev_remove_offload(struct packet_offload *po)
582 __dev_remove_offload(po);
586 EXPORT_SYMBOL(dev_remove_offload);
588 /******************************************************************************
590 * Device Boot-time Settings Routines
592 ******************************************************************************/
594 /* Boot time configuration table */
595 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
598 * netdev_boot_setup_add - add new setup entry
599 * @name: name of the device
600 * @map: configured settings for the device
602 * Adds new setup entry to the dev_boot_setup list. The function
603 * returns 0 on error and 1 on success. This is a generic routine to
606 static int netdev_boot_setup_add(char *name, struct ifmap *map)
608 struct netdev_boot_setup *s;
612 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
613 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
614 memset(s[i].name, 0, sizeof(s[i].name));
615 strlcpy(s[i].name, name, IFNAMSIZ);
616 memcpy(&s[i].map, map, sizeof(s[i].map));
621 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
625 * netdev_boot_setup_check - check boot time settings
626 * @dev: the netdevice
628 * Check boot time settings for the device.
629 * The found settings are set for the device to be used
630 * later in the device probing.
631 * Returns 0 if no settings found, 1 if they are.
633 int netdev_boot_setup_check(struct net_device *dev)
635 struct netdev_boot_setup *s = dev_boot_setup;
638 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
639 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
640 !strcmp(dev->name, s[i].name)) {
641 dev->irq = s[i].map.irq;
642 dev->base_addr = s[i].map.base_addr;
643 dev->mem_start = s[i].map.mem_start;
644 dev->mem_end = s[i].map.mem_end;
650 EXPORT_SYMBOL(netdev_boot_setup_check);
654 * netdev_boot_base - get address from boot time settings
655 * @prefix: prefix for network device
656 * @unit: id for network device
658 * Check boot time settings for the base address of device.
659 * The found settings are set for the device to be used
660 * later in the device probing.
661 * Returns 0 if no settings found.
663 unsigned long netdev_boot_base(const char *prefix, int unit)
665 const struct netdev_boot_setup *s = dev_boot_setup;
669 sprintf(name, "%s%d", prefix, unit);
672 * If device already registered then return base of 1
673 * to indicate not to probe for this interface
675 if (__dev_get_by_name(&init_net, name))
678 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
679 if (!strcmp(name, s[i].name))
680 return s[i].map.base_addr;
685 * Saves at boot time configured settings for any netdevice.
687 int __init netdev_boot_setup(char *str)
692 str = get_options(str, ARRAY_SIZE(ints), ints);
697 memset(&map, 0, sizeof(map));
701 map.base_addr = ints[2];
703 map.mem_start = ints[3];
705 map.mem_end = ints[4];
707 /* Add new entry to the list */
708 return netdev_boot_setup_add(str, &map);
711 __setup("netdev=", netdev_boot_setup);
713 /*******************************************************************************
715 * Device Interface Subroutines
717 *******************************************************************************/
720 * dev_get_iflink - get 'iflink' value of a interface
721 * @dev: targeted interface
723 * Indicates the ifindex the interface is linked to.
724 * Physical interfaces have the same 'ifindex' and 'iflink' values.
727 int dev_get_iflink(const struct net_device *dev)
729 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
730 return dev->netdev_ops->ndo_get_iflink(dev);
734 EXPORT_SYMBOL(dev_get_iflink);
737 * dev_fill_metadata_dst - Retrieve tunnel egress information.
738 * @dev: targeted interface
741 * For better visibility of tunnel traffic OVS needs to retrieve
742 * egress tunnel information for a packet. Following API allows
743 * user to get this info.
745 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
747 struct ip_tunnel_info *info;
749 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
752 info = skb_tunnel_info_unclone(skb);
755 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
758 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
760 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
763 * __dev_get_by_name - find a device by its name
764 * @net: the applicable net namespace
765 * @name: name to find
767 * Find an interface by name. Must be called under RTNL semaphore
768 * or @dev_base_lock. If the name is found a pointer to the device
769 * is returned. If the name is not found then %NULL is returned. The
770 * reference counters are not incremented so the caller must be
771 * careful with locks.
774 struct net_device *__dev_get_by_name(struct net *net, const char *name)
776 struct netdev_name_node *node_name;
778 node_name = netdev_name_node_lookup(net, name);
779 return node_name ? node_name->dev : NULL;
781 EXPORT_SYMBOL(__dev_get_by_name);
784 * dev_get_by_name_rcu - find a device by its name
785 * @net: the applicable net namespace
786 * @name: name to find
788 * Find an interface by name.
789 * If the name is found a pointer to the device is returned.
790 * If the name is not found then %NULL is returned.
791 * The reference counters are not incremented so the caller must be
792 * careful with locks. The caller must hold RCU lock.
795 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
797 struct netdev_name_node *node_name;
799 node_name = netdev_name_node_lookup_rcu(net, name);
800 return node_name ? node_name->dev : NULL;
802 EXPORT_SYMBOL(dev_get_by_name_rcu);
805 * dev_get_by_name - find a device by its name
806 * @net: the applicable net namespace
807 * @name: name to find
809 * Find an interface by name. This can be called from any
810 * context and does its own locking. The returned handle has
811 * the usage count incremented and the caller must use dev_put() to
812 * release it when it is no longer needed. %NULL is returned if no
813 * matching device is found.
816 struct net_device *dev_get_by_name(struct net *net, const char *name)
818 struct net_device *dev;
821 dev = dev_get_by_name_rcu(net, name);
827 EXPORT_SYMBOL(dev_get_by_name);
830 * __dev_get_by_index - find a device by its ifindex
831 * @net: the applicable net namespace
832 * @ifindex: index of device
834 * Search for an interface by index. Returns %NULL if the device
835 * is not found or a pointer to the device. The device has not
836 * had its reference counter increased so the caller must be careful
837 * about locking. The caller must hold either the RTNL semaphore
841 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
843 struct net_device *dev;
844 struct hlist_head *head = dev_index_hash(net, ifindex);
846 hlist_for_each_entry(dev, head, index_hlist)
847 if (dev->ifindex == ifindex)
852 EXPORT_SYMBOL(__dev_get_by_index);
855 * dev_get_by_index_rcu - find a device by its ifindex
856 * @net: the applicable net namespace
857 * @ifindex: index of device
859 * Search for an interface by index. Returns %NULL if the device
860 * is not found or a pointer to the device. The device has not
861 * had its reference counter increased so the caller must be careful
862 * about locking. The caller must hold RCU lock.
865 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
867 struct net_device *dev;
868 struct hlist_head *head = dev_index_hash(net, ifindex);
870 hlist_for_each_entry_rcu(dev, head, index_hlist)
871 if (dev->ifindex == ifindex)
876 EXPORT_SYMBOL(dev_get_by_index_rcu);
880 * dev_get_by_index - find a device by its ifindex
881 * @net: the applicable net namespace
882 * @ifindex: index of device
884 * Search for an interface by index. Returns NULL if the device
885 * is not found or a pointer to the device. The device returned has
886 * had a reference added and the pointer is safe until the user calls
887 * dev_put to indicate they have finished with it.
890 struct net_device *dev_get_by_index(struct net *net, int ifindex)
892 struct net_device *dev;
895 dev = dev_get_by_index_rcu(net, ifindex);
901 EXPORT_SYMBOL(dev_get_by_index);
904 * dev_get_by_napi_id - find a device by napi_id
905 * @napi_id: ID of the NAPI struct
907 * Search for an interface by NAPI ID. Returns %NULL if the device
908 * is not found or a pointer to the device. The device has not had
909 * its reference counter increased so the caller must be careful
910 * about locking. The caller must hold RCU lock.
913 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
915 struct napi_struct *napi;
917 WARN_ON_ONCE(!rcu_read_lock_held());
919 if (napi_id < MIN_NAPI_ID)
922 napi = napi_by_id(napi_id);
924 return napi ? napi->dev : NULL;
926 EXPORT_SYMBOL(dev_get_by_napi_id);
929 * netdev_get_name - get a netdevice name, knowing its ifindex.
930 * @net: network namespace
931 * @name: a pointer to the buffer where the name will be stored.
932 * @ifindex: the ifindex of the interface to get the name from.
934 * The use of raw_seqcount_begin() and cond_resched() before
935 * retrying is required as we want to give the writers a chance
936 * to complete when CONFIG_PREEMPTION is not set.
938 int netdev_get_name(struct net *net, char *name, int ifindex)
940 struct net_device *dev;
944 seq = raw_seqcount_begin(&devnet_rename_seq);
946 dev = dev_get_by_index_rcu(net, ifindex);
952 strcpy(name, dev->name);
954 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
963 * dev_getbyhwaddr_rcu - find a device by its hardware address
964 * @net: the applicable net namespace
965 * @type: media type of device
966 * @ha: hardware address
968 * Search for an interface by MAC address. Returns NULL if the device
969 * is not found or a pointer to the device.
970 * The caller must hold RCU or RTNL.
971 * The returned device has not had its ref count increased
972 * and the caller must therefore be careful about locking
976 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
979 struct net_device *dev;
981 for_each_netdev_rcu(net, dev)
982 if (dev->type == type &&
983 !memcmp(dev->dev_addr, ha, dev->addr_len))
988 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
990 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
992 struct net_device *dev;
995 for_each_netdev(net, dev)
996 if (dev->type == type)
1001 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
1003 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
1005 struct net_device *dev, *ret = NULL;
1008 for_each_netdev_rcu(net, dev)
1009 if (dev->type == type) {
1017 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1020 * __dev_get_by_flags - find any device with given flags
1021 * @net: the applicable net namespace
1022 * @if_flags: IFF_* values
1023 * @mask: bitmask of bits in if_flags to check
1025 * Search for any interface with the given flags. Returns NULL if a device
1026 * is not found or a pointer to the device. Must be called inside
1027 * rtnl_lock(), and result refcount is unchanged.
1030 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1031 unsigned short mask)
1033 struct net_device *dev, *ret;
1038 for_each_netdev(net, dev) {
1039 if (((dev->flags ^ if_flags) & mask) == 0) {
1046 EXPORT_SYMBOL(__dev_get_by_flags);
1049 * dev_valid_name - check if name is okay for network device
1050 * @name: name string
1052 * Network device names need to be valid file names to
1053 * to allow sysfs to work. We also disallow any kind of
1056 bool dev_valid_name(const char *name)
1060 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1062 if (!strcmp(name, ".") || !strcmp(name, ".."))
1066 if (*name == '/' || *name == ':' || isspace(*name))
1072 EXPORT_SYMBOL(dev_valid_name);
1075 * __dev_alloc_name - allocate a name for a device
1076 * @net: network namespace to allocate the device name in
1077 * @name: name format string
1078 * @buf: scratch buffer and result name string
1080 * Passed a format string - eg "lt%d" it will try and find a suitable
1081 * id. It scans list of devices to build up a free map, then chooses
1082 * the first empty slot. The caller must hold the dev_base or rtnl lock
1083 * while allocating the name and adding the device in order to avoid
1085 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1086 * Returns the number of the unit assigned or a negative errno code.
1089 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1093 const int max_netdevices = 8*PAGE_SIZE;
1094 unsigned long *inuse;
1095 struct net_device *d;
1097 if (!dev_valid_name(name))
1100 p = strchr(name, '%');
1103 * Verify the string as this thing may have come from
1104 * the user. There must be either one "%d" and no other "%"
1107 if (p[1] != 'd' || strchr(p + 2, '%'))
1110 /* Use one page as a bit array of possible slots */
1111 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1115 for_each_netdev(net, d) {
1116 if (!sscanf(d->name, name, &i))
1118 if (i < 0 || i >= max_netdevices)
1121 /* avoid cases where sscanf is not exact inverse of printf */
1122 snprintf(buf, IFNAMSIZ, name, i);
1123 if (!strncmp(buf, d->name, IFNAMSIZ))
1127 i = find_first_zero_bit(inuse, max_netdevices);
1128 free_page((unsigned long) inuse);
1131 snprintf(buf, IFNAMSIZ, name, i);
1132 if (!__dev_get_by_name(net, buf))
1135 /* It is possible to run out of possible slots
1136 * when the name is long and there isn't enough space left
1137 * for the digits, or if all bits are used.
1142 static int dev_alloc_name_ns(struct net *net,
1143 struct net_device *dev,
1150 ret = __dev_alloc_name(net, name, buf);
1152 strlcpy(dev->name, buf, IFNAMSIZ);
1157 * dev_alloc_name - allocate a name for a device
1159 * @name: name format string
1161 * Passed a format string - eg "lt%d" it will try and find a suitable
1162 * id. It scans list of devices to build up a free map, then chooses
1163 * the first empty slot. The caller must hold the dev_base or rtnl lock
1164 * while allocating the name and adding the device in order to avoid
1166 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1167 * Returns the number of the unit assigned or a negative errno code.
1170 int dev_alloc_name(struct net_device *dev, const char *name)
1172 return dev_alloc_name_ns(dev_net(dev), dev, name);
1174 EXPORT_SYMBOL(dev_alloc_name);
1176 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1181 if (!dev_valid_name(name))
1184 if (strchr(name, '%'))
1185 return dev_alloc_name_ns(net, dev, name);
1186 else if (__dev_get_by_name(net, name))
1188 else if (dev->name != name)
1189 strlcpy(dev->name, name, IFNAMSIZ);
1195 * dev_change_name - change name of a device
1197 * @newname: name (or format string) must be at least IFNAMSIZ
1199 * Change name of a device, can pass format strings "eth%d".
1202 int dev_change_name(struct net_device *dev, const char *newname)
1204 unsigned char old_assign_type;
1205 char oldname[IFNAMSIZ];
1211 BUG_ON(!dev_net(dev));
1215 /* Some auto-enslaved devices e.g. failover slaves are
1216 * special, as userspace might rename the device after
1217 * the interface had been brought up and running since
1218 * the point kernel initiated auto-enslavement. Allow
1219 * live name change even when these slave devices are
1222 * Typically, users of these auto-enslaving devices
1223 * don't actually care about slave name change, as
1224 * they are supposed to operate on master interface
1227 if (dev->flags & IFF_UP &&
1228 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1231 write_seqcount_begin(&devnet_rename_seq);
1233 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1234 write_seqcount_end(&devnet_rename_seq);
1238 memcpy(oldname, dev->name, IFNAMSIZ);
1240 err = dev_get_valid_name(net, dev, newname);
1242 write_seqcount_end(&devnet_rename_seq);
1246 if (oldname[0] && !strchr(oldname, '%'))
1247 netdev_info(dev, "renamed from %s\n", oldname);
1249 old_assign_type = dev->name_assign_type;
1250 dev->name_assign_type = NET_NAME_RENAMED;
1253 ret = device_rename(&dev->dev, dev->name);
1255 memcpy(dev->name, oldname, IFNAMSIZ);
1256 dev->name_assign_type = old_assign_type;
1257 write_seqcount_end(&devnet_rename_seq);
1261 write_seqcount_end(&devnet_rename_seq);
1263 netdev_adjacent_rename_links(dev, oldname);
1265 write_lock_bh(&dev_base_lock);
1266 netdev_name_node_del(dev->name_node);
1267 write_unlock_bh(&dev_base_lock);
1271 write_lock_bh(&dev_base_lock);
1272 netdev_name_node_add(net, dev->name_node);
1273 write_unlock_bh(&dev_base_lock);
1275 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1276 ret = notifier_to_errno(ret);
1279 /* err >= 0 after dev_alloc_name() or stores the first errno */
1282 write_seqcount_begin(&devnet_rename_seq);
1283 memcpy(dev->name, oldname, IFNAMSIZ);
1284 memcpy(oldname, newname, IFNAMSIZ);
1285 dev->name_assign_type = old_assign_type;
1286 old_assign_type = NET_NAME_RENAMED;
1289 pr_err("%s: name change rollback failed: %d\n",
1298 * dev_set_alias - change ifalias of a device
1300 * @alias: name up to IFALIASZ
1301 * @len: limit of bytes to copy from info
1303 * Set ifalias for a device,
1305 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1307 struct dev_ifalias *new_alias = NULL;
1309 if (len >= IFALIASZ)
1313 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1317 memcpy(new_alias->ifalias, alias, len);
1318 new_alias->ifalias[len] = 0;
1321 mutex_lock(&ifalias_mutex);
1322 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1323 mutex_is_locked(&ifalias_mutex));
1324 mutex_unlock(&ifalias_mutex);
1327 kfree_rcu(new_alias, rcuhead);
1331 EXPORT_SYMBOL(dev_set_alias);
1334 * dev_get_alias - get ifalias of a device
1336 * @name: buffer to store name of ifalias
1337 * @len: size of buffer
1339 * get ifalias for a device. Caller must make sure dev cannot go
1340 * away, e.g. rcu read lock or own a reference count to device.
1342 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1344 const struct dev_ifalias *alias;
1348 alias = rcu_dereference(dev->ifalias);
1350 ret = snprintf(name, len, "%s", alias->ifalias);
1357 * netdev_features_change - device changes features
1358 * @dev: device to cause notification
1360 * Called to indicate a device has changed features.
1362 void netdev_features_change(struct net_device *dev)
1364 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1366 EXPORT_SYMBOL(netdev_features_change);
1369 * netdev_state_change - device changes state
1370 * @dev: device to cause notification
1372 * Called to indicate a device has changed state. This function calls
1373 * the notifier chains for netdev_chain and sends a NEWLINK message
1374 * to the routing socket.
1376 void netdev_state_change(struct net_device *dev)
1378 if (dev->flags & IFF_UP) {
1379 struct netdev_notifier_change_info change_info = {
1383 call_netdevice_notifiers_info(NETDEV_CHANGE,
1385 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1388 EXPORT_SYMBOL(netdev_state_change);
1391 * netdev_notify_peers - notify network peers about existence of @dev
1392 * @dev: network device
1394 * Generate traffic such that interested network peers are aware of
1395 * @dev, such as by generating a gratuitous ARP. This may be used when
1396 * a device wants to inform the rest of the network about some sort of
1397 * reconfiguration such as a failover event or virtual machine
1400 void netdev_notify_peers(struct net_device *dev)
1403 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1404 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1407 EXPORT_SYMBOL(netdev_notify_peers);
1409 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1411 const struct net_device_ops *ops = dev->netdev_ops;
1416 if (!netif_device_present(dev))
1419 /* Block netpoll from trying to do any rx path servicing.
1420 * If we don't do this there is a chance ndo_poll_controller
1421 * or ndo_poll may be running while we open the device
1423 netpoll_poll_disable(dev);
1425 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1426 ret = notifier_to_errno(ret);
1430 set_bit(__LINK_STATE_START, &dev->state);
1432 if (ops->ndo_validate_addr)
1433 ret = ops->ndo_validate_addr(dev);
1435 if (!ret && ops->ndo_open)
1436 ret = ops->ndo_open(dev);
1438 netpoll_poll_enable(dev);
1441 clear_bit(__LINK_STATE_START, &dev->state);
1443 dev->flags |= IFF_UP;
1444 dev_set_rx_mode(dev);
1446 add_device_randomness(dev->dev_addr, dev->addr_len);
1453 * dev_open - prepare an interface for use.
1454 * @dev: device to open
1455 * @extack: netlink extended ack
1457 * Takes a device from down to up state. The device's private open
1458 * function is invoked and then the multicast lists are loaded. Finally
1459 * the device is moved into the up state and a %NETDEV_UP message is
1460 * sent to the netdev notifier chain.
1462 * Calling this function on an active interface is a nop. On a failure
1463 * a negative errno code is returned.
1465 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1469 if (dev->flags & IFF_UP)
1472 ret = __dev_open(dev, extack);
1476 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1477 call_netdevice_notifiers(NETDEV_UP, dev);
1481 EXPORT_SYMBOL(dev_open);
1483 static void __dev_close_many(struct list_head *head)
1485 struct net_device *dev;
1490 list_for_each_entry(dev, head, close_list) {
1491 /* Temporarily disable netpoll until the interface is down */
1492 netpoll_poll_disable(dev);
1494 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1496 clear_bit(__LINK_STATE_START, &dev->state);
1498 /* Synchronize to scheduled poll. We cannot touch poll list, it
1499 * can be even on different cpu. So just clear netif_running().
1501 * dev->stop() will invoke napi_disable() on all of it's
1502 * napi_struct instances on this device.
1504 smp_mb__after_atomic(); /* Commit netif_running(). */
1507 dev_deactivate_many(head);
1509 list_for_each_entry(dev, head, close_list) {
1510 const struct net_device_ops *ops = dev->netdev_ops;
1513 * Call the device specific close. This cannot fail.
1514 * Only if device is UP
1516 * We allow it to be called even after a DETACH hot-plug
1522 dev->flags &= ~IFF_UP;
1523 netpoll_poll_enable(dev);
1527 static void __dev_close(struct net_device *dev)
1531 list_add(&dev->close_list, &single);
1532 __dev_close_many(&single);
1536 void dev_close_many(struct list_head *head, bool unlink)
1538 struct net_device *dev, *tmp;
1540 /* Remove the devices that don't need to be closed */
1541 list_for_each_entry_safe(dev, tmp, head, close_list)
1542 if (!(dev->flags & IFF_UP))
1543 list_del_init(&dev->close_list);
1545 __dev_close_many(head);
1547 list_for_each_entry_safe(dev, tmp, head, close_list) {
1548 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1549 call_netdevice_notifiers(NETDEV_DOWN, dev);
1551 list_del_init(&dev->close_list);
1554 EXPORT_SYMBOL(dev_close_many);
1557 * dev_close - shutdown an interface.
1558 * @dev: device to shutdown
1560 * This function moves an active device into down state. A
1561 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1562 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1565 void dev_close(struct net_device *dev)
1567 if (dev->flags & IFF_UP) {
1570 list_add(&dev->close_list, &single);
1571 dev_close_many(&single, true);
1575 EXPORT_SYMBOL(dev_close);
1579 * dev_disable_lro - disable Large Receive Offload on a device
1582 * Disable Large Receive Offload (LRO) on a net device. Must be
1583 * called under RTNL. This is needed if received packets may be
1584 * forwarded to another interface.
1586 void dev_disable_lro(struct net_device *dev)
1588 struct net_device *lower_dev;
1589 struct list_head *iter;
1591 dev->wanted_features &= ~NETIF_F_LRO;
1592 netdev_update_features(dev);
1594 if (unlikely(dev->features & NETIF_F_LRO))
1595 netdev_WARN(dev, "failed to disable LRO!\n");
1597 netdev_for_each_lower_dev(dev, lower_dev, iter)
1598 dev_disable_lro(lower_dev);
1600 EXPORT_SYMBOL(dev_disable_lro);
1603 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1606 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1607 * called under RTNL. This is needed if Generic XDP is installed on
1610 static void dev_disable_gro_hw(struct net_device *dev)
1612 dev->wanted_features &= ~NETIF_F_GRO_HW;
1613 netdev_update_features(dev);
1615 if (unlikely(dev->features & NETIF_F_GRO_HW))
1616 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1619 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1622 case NETDEV_##val: \
1623 return "NETDEV_" __stringify(val);
1625 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1626 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1627 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1628 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1629 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1630 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1631 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1632 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1633 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1637 return "UNKNOWN_NETDEV_EVENT";
1639 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1641 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1642 struct net_device *dev)
1644 struct netdev_notifier_info info = {
1648 return nb->notifier_call(nb, val, &info);
1651 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1652 struct net_device *dev)
1656 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1657 err = notifier_to_errno(err);
1661 if (!(dev->flags & IFF_UP))
1664 call_netdevice_notifier(nb, NETDEV_UP, dev);
1668 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1669 struct net_device *dev)
1671 if (dev->flags & IFF_UP) {
1672 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1674 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1676 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1679 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1682 struct net_device *dev;
1685 for_each_netdev(net, dev) {
1686 err = call_netdevice_register_notifiers(nb, dev);
1693 for_each_netdev_continue_reverse(net, dev)
1694 call_netdevice_unregister_notifiers(nb, dev);
1698 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1701 struct net_device *dev;
1703 for_each_netdev(net, dev)
1704 call_netdevice_unregister_notifiers(nb, dev);
1707 static int dev_boot_phase = 1;
1710 * register_netdevice_notifier - register a network notifier block
1713 * Register a notifier to be called when network device events occur.
1714 * The notifier passed is linked into the kernel structures and must
1715 * not be reused until it has been unregistered. A negative errno code
1716 * is returned on a failure.
1718 * When registered all registration and up events are replayed
1719 * to the new notifier to allow device to have a race free
1720 * view of the network device list.
1723 int register_netdevice_notifier(struct notifier_block *nb)
1728 /* Close race with setup_net() and cleanup_net() */
1729 down_write(&pernet_ops_rwsem);
1731 err = raw_notifier_chain_register(&netdev_chain, nb);
1737 err = call_netdevice_register_net_notifiers(nb, net);
1744 up_write(&pernet_ops_rwsem);
1748 for_each_net_continue_reverse(net)
1749 call_netdevice_unregister_net_notifiers(nb, net);
1751 raw_notifier_chain_unregister(&netdev_chain, nb);
1754 EXPORT_SYMBOL(register_netdevice_notifier);
1757 * unregister_netdevice_notifier - unregister a network notifier block
1760 * Unregister a notifier previously registered by
1761 * register_netdevice_notifier(). The notifier is unlinked into the
1762 * kernel structures and may then be reused. A negative errno code
1763 * is returned on a failure.
1765 * After unregistering unregister and down device events are synthesized
1766 * for all devices on the device list to the removed notifier to remove
1767 * the need for special case cleanup code.
1770 int unregister_netdevice_notifier(struct notifier_block *nb)
1775 /* Close race with setup_net() and cleanup_net() */
1776 down_write(&pernet_ops_rwsem);
1778 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1783 call_netdevice_unregister_net_notifiers(nb, net);
1787 up_write(&pernet_ops_rwsem);
1790 EXPORT_SYMBOL(unregister_netdevice_notifier);
1792 static int __register_netdevice_notifier_net(struct net *net,
1793 struct notifier_block *nb,
1794 bool ignore_call_fail)
1798 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1804 err = call_netdevice_register_net_notifiers(nb, net);
1805 if (err && !ignore_call_fail)
1806 goto chain_unregister;
1811 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1815 static int __unregister_netdevice_notifier_net(struct net *net,
1816 struct notifier_block *nb)
1820 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1824 call_netdevice_unregister_net_notifiers(nb, net);
1829 * register_netdevice_notifier_net - register a per-netns network notifier block
1830 * @net: network namespace
1833 * Register a notifier to be called when network device events occur.
1834 * The notifier passed is linked into the kernel structures and must
1835 * not be reused until it has been unregistered. A negative errno code
1836 * is returned on a failure.
1838 * When registered all registration and up events are replayed
1839 * to the new notifier to allow device to have a race free
1840 * view of the network device list.
1843 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1848 err = __register_netdevice_notifier_net(net, nb, false);
1852 EXPORT_SYMBOL(register_netdevice_notifier_net);
1855 * unregister_netdevice_notifier_net - unregister a per-netns
1856 * network notifier block
1857 * @net: network namespace
1860 * Unregister a notifier previously registered by
1861 * register_netdevice_notifier(). The notifier is unlinked into the
1862 * kernel structures and may then be reused. A negative errno code
1863 * is returned on a failure.
1865 * After unregistering unregister and down device events are synthesized
1866 * for all devices on the device list to the removed notifier to remove
1867 * the need for special case cleanup code.
1870 int unregister_netdevice_notifier_net(struct net *net,
1871 struct notifier_block *nb)
1876 err = __unregister_netdevice_notifier_net(net, nb);
1880 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1882 int register_netdevice_notifier_dev_net(struct net_device *dev,
1883 struct notifier_block *nb,
1884 struct netdev_net_notifier *nn)
1889 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1892 list_add(&nn->list, &dev->net_notifier_list);
1897 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1899 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1900 struct notifier_block *nb,
1901 struct netdev_net_notifier *nn)
1906 list_del(&nn->list);
1907 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1911 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1913 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1916 struct netdev_net_notifier *nn;
1918 list_for_each_entry(nn, &dev->net_notifier_list, list) {
1919 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb);
1920 __register_netdevice_notifier_net(net, nn->nb, true);
1925 * call_netdevice_notifiers_info - call all network notifier blocks
1926 * @val: value passed unmodified to notifier function
1927 * @info: notifier information data
1929 * Call all network notifier blocks. Parameters and return value
1930 * are as for raw_notifier_call_chain().
1933 static int call_netdevice_notifiers_info(unsigned long val,
1934 struct netdev_notifier_info *info)
1936 struct net *net = dev_net(info->dev);
1941 /* Run per-netns notifier block chain first, then run the global one.
1942 * Hopefully, one day, the global one is going to be removed after
1943 * all notifier block registrators get converted to be per-netns.
1945 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1946 if (ret & NOTIFY_STOP_MASK)
1948 return raw_notifier_call_chain(&netdev_chain, val, info);
1951 static int call_netdevice_notifiers_extack(unsigned long val,
1952 struct net_device *dev,
1953 struct netlink_ext_ack *extack)
1955 struct netdev_notifier_info info = {
1960 return call_netdevice_notifiers_info(val, &info);
1964 * call_netdevice_notifiers - call all network notifier blocks
1965 * @val: value passed unmodified to notifier function
1966 * @dev: net_device pointer passed unmodified to notifier function
1968 * Call all network notifier blocks. Parameters and return value
1969 * are as for raw_notifier_call_chain().
1972 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1974 return call_netdevice_notifiers_extack(val, dev, NULL);
1976 EXPORT_SYMBOL(call_netdevice_notifiers);
1979 * call_netdevice_notifiers_mtu - call all network notifier blocks
1980 * @val: value passed unmodified to notifier function
1981 * @dev: net_device pointer passed unmodified to notifier function
1982 * @arg: additional u32 argument passed to the notifier function
1984 * Call all network notifier blocks. Parameters and return value
1985 * are as for raw_notifier_call_chain().
1987 static int call_netdevice_notifiers_mtu(unsigned long val,
1988 struct net_device *dev, u32 arg)
1990 struct netdev_notifier_info_ext info = {
1995 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
1997 return call_netdevice_notifiers_info(val, &info.info);
2000 #ifdef CONFIG_NET_INGRESS
2001 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2003 void net_inc_ingress_queue(void)
2005 static_branch_inc(&ingress_needed_key);
2007 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2009 void net_dec_ingress_queue(void)
2011 static_branch_dec(&ingress_needed_key);
2013 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2016 #ifdef CONFIG_NET_EGRESS
2017 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2019 void net_inc_egress_queue(void)
2021 static_branch_inc(&egress_needed_key);
2023 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2025 void net_dec_egress_queue(void)
2027 static_branch_dec(&egress_needed_key);
2029 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2032 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2033 #ifdef CONFIG_JUMP_LABEL
2034 static atomic_t netstamp_needed_deferred;
2035 static atomic_t netstamp_wanted;
2036 static void netstamp_clear(struct work_struct *work)
2038 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2041 wanted = atomic_add_return(deferred, &netstamp_wanted);
2043 static_branch_enable(&netstamp_needed_key);
2045 static_branch_disable(&netstamp_needed_key);
2047 static DECLARE_WORK(netstamp_work, netstamp_clear);
2050 void net_enable_timestamp(void)
2052 #ifdef CONFIG_JUMP_LABEL
2056 wanted = atomic_read(&netstamp_wanted);
2059 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
2062 atomic_inc(&netstamp_needed_deferred);
2063 schedule_work(&netstamp_work);
2065 static_branch_inc(&netstamp_needed_key);
2068 EXPORT_SYMBOL(net_enable_timestamp);
2070 void net_disable_timestamp(void)
2072 #ifdef CONFIG_JUMP_LABEL
2076 wanted = atomic_read(&netstamp_wanted);
2079 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2082 atomic_dec(&netstamp_needed_deferred);
2083 schedule_work(&netstamp_work);
2085 static_branch_dec(&netstamp_needed_key);
2088 EXPORT_SYMBOL(net_disable_timestamp);
2090 static inline void net_timestamp_set(struct sk_buff *skb)
2093 if (static_branch_unlikely(&netstamp_needed_key))
2094 __net_timestamp(skb);
2097 #define net_timestamp_check(COND, SKB) \
2098 if (static_branch_unlikely(&netstamp_needed_key)) { \
2099 if ((COND) && !(SKB)->tstamp) \
2100 __net_timestamp(SKB); \
2103 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2107 if (!(dev->flags & IFF_UP))
2110 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
2111 if (skb->len <= len)
2114 /* if TSO is enabled, we don't care about the length as the packet
2115 * could be forwarded without being segmented before
2117 if (skb_is_gso(skb))
2122 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2124 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2126 int ret = ____dev_forward_skb(dev, skb);
2129 skb->protocol = eth_type_trans(skb, dev);
2130 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2135 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2138 * dev_forward_skb - loopback an skb to another netif
2140 * @dev: destination network device
2141 * @skb: buffer to forward
2144 * NET_RX_SUCCESS (no congestion)
2145 * NET_RX_DROP (packet was dropped, but freed)
2147 * dev_forward_skb can be used for injecting an skb from the
2148 * start_xmit function of one device into the receive queue
2149 * of another device.
2151 * The receiving device may be in another namespace, so
2152 * we have to clear all information in the skb that could
2153 * impact namespace isolation.
2155 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2157 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2159 EXPORT_SYMBOL_GPL(dev_forward_skb);
2161 static inline int deliver_skb(struct sk_buff *skb,
2162 struct packet_type *pt_prev,
2163 struct net_device *orig_dev)
2165 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2167 refcount_inc(&skb->users);
2168 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2171 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2172 struct packet_type **pt,
2173 struct net_device *orig_dev,
2175 struct list_head *ptype_list)
2177 struct packet_type *ptype, *pt_prev = *pt;
2179 list_for_each_entry_rcu(ptype, ptype_list, list) {
2180 if (ptype->type != type)
2183 deliver_skb(skb, pt_prev, orig_dev);
2189 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2191 if (!ptype->af_packet_priv || !skb->sk)
2194 if (ptype->id_match)
2195 return ptype->id_match(ptype, skb->sk);
2196 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2203 * dev_nit_active - return true if any network interface taps are in use
2205 * @dev: network device to check for the presence of taps
2207 bool dev_nit_active(struct net_device *dev)
2209 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2211 EXPORT_SYMBOL_GPL(dev_nit_active);
2214 * Support routine. Sends outgoing frames to any network
2215 * taps currently in use.
2218 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2220 struct packet_type *ptype;
2221 struct sk_buff *skb2 = NULL;
2222 struct packet_type *pt_prev = NULL;
2223 struct list_head *ptype_list = &ptype_all;
2227 list_for_each_entry_rcu(ptype, ptype_list, list) {
2228 if (ptype->ignore_outgoing)
2231 /* Never send packets back to the socket
2232 * they originated from - MvS (miquels@drinkel.ow.org)
2234 if (skb_loop_sk(ptype, skb))
2238 deliver_skb(skb2, pt_prev, skb->dev);
2243 /* need to clone skb, done only once */
2244 skb2 = skb_clone(skb, GFP_ATOMIC);
2248 net_timestamp_set(skb2);
2250 /* skb->nh should be correctly
2251 * set by sender, so that the second statement is
2252 * just protection against buggy protocols.
2254 skb_reset_mac_header(skb2);
2256 if (skb_network_header(skb2) < skb2->data ||
2257 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2258 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2259 ntohs(skb2->protocol),
2261 skb_reset_network_header(skb2);
2264 skb2->transport_header = skb2->network_header;
2265 skb2->pkt_type = PACKET_OUTGOING;
2269 if (ptype_list == &ptype_all) {
2270 ptype_list = &dev->ptype_all;
2275 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2276 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2282 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2285 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2286 * @dev: Network device
2287 * @txq: number of queues available
2289 * If real_num_tx_queues is changed the tc mappings may no longer be
2290 * valid. To resolve this verify the tc mapping remains valid and if
2291 * not NULL the mapping. With no priorities mapping to this
2292 * offset/count pair it will no longer be used. In the worst case TC0
2293 * is invalid nothing can be done so disable priority mappings. If is
2294 * expected that drivers will fix this mapping if they can before
2295 * calling netif_set_real_num_tx_queues.
2297 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2300 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2302 /* If TC0 is invalidated disable TC mapping */
2303 if (tc->offset + tc->count > txq) {
2304 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2309 /* Invalidated prio to tc mappings set to TC0 */
2310 for (i = 1; i < TC_BITMASK + 1; i++) {
2311 int q = netdev_get_prio_tc_map(dev, i);
2313 tc = &dev->tc_to_txq[q];
2314 if (tc->offset + tc->count > txq) {
2315 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2317 netdev_set_prio_tc_map(dev, i, 0);
2322 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2325 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2328 /* walk through the TCs and see if it falls into any of them */
2329 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2330 if ((txq - tc->offset) < tc->count)
2334 /* didn't find it, just return -1 to indicate no match */
2340 EXPORT_SYMBOL(netdev_txq_to_tc);
2343 struct static_key xps_needed __read_mostly;
2344 EXPORT_SYMBOL(xps_needed);
2345 struct static_key xps_rxqs_needed __read_mostly;
2346 EXPORT_SYMBOL(xps_rxqs_needed);
2347 static DEFINE_MUTEX(xps_map_mutex);
2348 #define xmap_dereference(P) \
2349 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2351 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2354 struct xps_map *map = NULL;
2358 map = xmap_dereference(dev_maps->attr_map[tci]);
2362 for (pos = map->len; pos--;) {
2363 if (map->queues[pos] != index)
2367 map->queues[pos] = map->queues[--map->len];
2371 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2372 kfree_rcu(map, rcu);
2379 static bool remove_xps_queue_cpu(struct net_device *dev,
2380 struct xps_dev_maps *dev_maps,
2381 int cpu, u16 offset, u16 count)
2383 int num_tc = dev->num_tc ? : 1;
2384 bool active = false;
2387 for (tci = cpu * num_tc; num_tc--; tci++) {
2390 for (i = count, j = offset; i--; j++) {
2391 if (!remove_xps_queue(dev_maps, tci, j))
2401 static void reset_xps_maps(struct net_device *dev,
2402 struct xps_dev_maps *dev_maps,
2406 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2407 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2409 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2411 static_key_slow_dec_cpuslocked(&xps_needed);
2412 kfree_rcu(dev_maps, rcu);
2415 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask,
2416 struct xps_dev_maps *dev_maps, unsigned int nr_ids,
2417 u16 offset, u16 count, bool is_rxqs_map)
2419 bool active = false;
2422 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids),
2424 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset,
2427 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2430 for (i = offset + (count - 1); count--; i--) {
2431 netdev_queue_numa_node_write(
2432 netdev_get_tx_queue(dev, i),
2438 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2441 const unsigned long *possible_mask = NULL;
2442 struct xps_dev_maps *dev_maps;
2443 unsigned int nr_ids;
2445 if (!static_key_false(&xps_needed))
2449 mutex_lock(&xps_map_mutex);
2451 if (static_key_false(&xps_rxqs_needed)) {
2452 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2454 nr_ids = dev->num_rx_queues;
2455 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids,
2456 offset, count, true);
2460 dev_maps = xmap_dereference(dev->xps_cpus_map);
2464 if (num_possible_cpus() > 1)
2465 possible_mask = cpumask_bits(cpu_possible_mask);
2466 nr_ids = nr_cpu_ids;
2467 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count,
2471 mutex_unlock(&xps_map_mutex);
2475 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2477 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2480 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2481 u16 index, bool is_rxqs_map)
2483 struct xps_map *new_map;
2484 int alloc_len = XPS_MIN_MAP_ALLOC;
2487 for (pos = 0; map && pos < map->len; pos++) {
2488 if (map->queues[pos] != index)
2493 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2495 if (pos < map->alloc_len)
2498 alloc_len = map->alloc_len * 2;
2501 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2505 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2507 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2508 cpu_to_node(attr_index));
2512 for (i = 0; i < pos; i++)
2513 new_map->queues[i] = map->queues[i];
2514 new_map->alloc_len = alloc_len;
2520 /* Must be called under cpus_read_lock */
2521 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2522 u16 index, bool is_rxqs_map)
2524 const unsigned long *online_mask = NULL, *possible_mask = NULL;
2525 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2526 int i, j, tci, numa_node_id = -2;
2527 int maps_sz, num_tc = 1, tc = 0;
2528 struct xps_map *map, *new_map;
2529 bool active = false;
2530 unsigned int nr_ids;
2533 /* Do not allow XPS on subordinate device directly */
2534 num_tc = dev->num_tc;
2538 /* If queue belongs to subordinate dev use its map */
2539 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2541 tc = netdev_txq_to_tc(dev, index);
2546 mutex_lock(&xps_map_mutex);
2548 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2549 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2550 nr_ids = dev->num_rx_queues;
2552 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2553 if (num_possible_cpus() > 1) {
2554 online_mask = cpumask_bits(cpu_online_mask);
2555 possible_mask = cpumask_bits(cpu_possible_mask);
2557 dev_maps = xmap_dereference(dev->xps_cpus_map);
2558 nr_ids = nr_cpu_ids;
2561 if (maps_sz < L1_CACHE_BYTES)
2562 maps_sz = L1_CACHE_BYTES;
2564 /* allocate memory for queue storage */
2565 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2568 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2569 if (!new_dev_maps) {
2570 mutex_unlock(&xps_map_mutex);
2574 tci = j * num_tc + tc;
2575 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) :
2578 map = expand_xps_map(map, j, index, is_rxqs_map);
2582 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2586 goto out_no_new_maps;
2589 /* Increment static keys at most once per type */
2590 static_key_slow_inc_cpuslocked(&xps_needed);
2592 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2595 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2597 /* copy maps belonging to foreign traffic classes */
2598 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) {
2599 /* fill in the new device map from the old device map */
2600 map = xmap_dereference(dev_maps->attr_map[tci]);
2601 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2604 /* We need to explicitly update tci as prevous loop
2605 * could break out early if dev_maps is NULL.
2607 tci = j * num_tc + tc;
2609 if (netif_attr_test_mask(j, mask, nr_ids) &&
2610 netif_attr_test_online(j, online_mask, nr_ids)) {
2611 /* add tx-queue to CPU/rx-queue maps */
2614 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2615 while ((pos < map->len) && (map->queues[pos] != index))
2618 if (pos == map->len)
2619 map->queues[map->len++] = index;
2622 if (numa_node_id == -2)
2623 numa_node_id = cpu_to_node(j);
2624 else if (numa_node_id != cpu_to_node(j))
2628 } else if (dev_maps) {
2629 /* fill in the new device map from the old device map */
2630 map = xmap_dereference(dev_maps->attr_map[tci]);
2631 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2634 /* copy maps belonging to foreign traffic classes */
2635 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2636 /* fill in the new device map from the old device map */
2637 map = xmap_dereference(dev_maps->attr_map[tci]);
2638 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2643 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps);
2645 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps);
2647 /* Cleanup old maps */
2649 goto out_no_old_maps;
2651 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2653 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2654 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2655 map = xmap_dereference(dev_maps->attr_map[tci]);
2656 if (map && map != new_map)
2657 kfree_rcu(map, rcu);
2661 kfree_rcu(dev_maps, rcu);
2664 dev_maps = new_dev_maps;
2669 /* update Tx queue numa node */
2670 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2671 (numa_node_id >= 0) ?
2672 numa_node_id : NUMA_NO_NODE);
2678 /* removes tx-queue from unused CPUs/rx-queues */
2679 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2681 for (i = tc, tci = j * num_tc; i--; tci++)
2682 active |= remove_xps_queue(dev_maps, tci, index);
2683 if (!netif_attr_test_mask(j, mask, nr_ids) ||
2684 !netif_attr_test_online(j, online_mask, nr_ids))
2685 active |= remove_xps_queue(dev_maps, tci, index);
2686 for (i = num_tc - tc, tci++; --i; tci++)
2687 active |= remove_xps_queue(dev_maps, tci, index);
2690 /* free map if not active */
2692 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2695 mutex_unlock(&xps_map_mutex);
2699 /* remove any maps that we added */
2700 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2702 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2703 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2705 xmap_dereference(dev_maps->attr_map[tci]) :
2707 if (new_map && new_map != map)
2712 mutex_unlock(&xps_map_mutex);
2714 kfree(new_dev_maps);
2717 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2719 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2725 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
2730 EXPORT_SYMBOL(netif_set_xps_queue);
2733 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2735 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2737 /* Unbind any subordinate channels */
2738 while (txq-- != &dev->_tx[0]) {
2740 netdev_unbind_sb_channel(dev, txq->sb_dev);
2744 void netdev_reset_tc(struct net_device *dev)
2747 netif_reset_xps_queues_gt(dev, 0);
2749 netdev_unbind_all_sb_channels(dev);
2751 /* Reset TC configuration of device */
2753 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2754 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2756 EXPORT_SYMBOL(netdev_reset_tc);
2758 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2760 if (tc >= dev->num_tc)
2764 netif_reset_xps_queues(dev, offset, count);
2766 dev->tc_to_txq[tc].count = count;
2767 dev->tc_to_txq[tc].offset = offset;
2770 EXPORT_SYMBOL(netdev_set_tc_queue);
2772 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2774 if (num_tc > TC_MAX_QUEUE)
2778 netif_reset_xps_queues_gt(dev, 0);
2780 netdev_unbind_all_sb_channels(dev);
2782 dev->num_tc = num_tc;
2785 EXPORT_SYMBOL(netdev_set_num_tc);
2787 void netdev_unbind_sb_channel(struct net_device *dev,
2788 struct net_device *sb_dev)
2790 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2793 netif_reset_xps_queues_gt(sb_dev, 0);
2795 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2796 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2798 while (txq-- != &dev->_tx[0]) {
2799 if (txq->sb_dev == sb_dev)
2803 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2805 int netdev_bind_sb_channel_queue(struct net_device *dev,
2806 struct net_device *sb_dev,
2807 u8 tc, u16 count, u16 offset)
2809 /* Make certain the sb_dev and dev are already configured */
2810 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2813 /* We cannot hand out queues we don't have */
2814 if ((offset + count) > dev->real_num_tx_queues)
2817 /* Record the mapping */
2818 sb_dev->tc_to_txq[tc].count = count;
2819 sb_dev->tc_to_txq[tc].offset = offset;
2821 /* Provide a way for Tx queue to find the tc_to_txq map or
2822 * XPS map for itself.
2825 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2829 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2831 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2833 /* Do not use a multiqueue device to represent a subordinate channel */
2834 if (netif_is_multiqueue(dev))
2837 /* We allow channels 1 - 32767 to be used for subordinate channels.
2838 * Channel 0 is meant to be "native" mode and used only to represent
2839 * the main root device. We allow writing 0 to reset the device back
2840 * to normal mode after being used as a subordinate channel.
2842 if (channel > S16_MAX)
2845 dev->num_tc = -channel;
2849 EXPORT_SYMBOL(netdev_set_sb_channel);
2852 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2853 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2855 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2860 disabling = txq < dev->real_num_tx_queues;
2862 if (txq < 1 || txq > dev->num_tx_queues)
2865 if (dev->reg_state == NETREG_REGISTERED ||
2866 dev->reg_state == NETREG_UNREGISTERING) {
2869 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2875 netif_setup_tc(dev, txq);
2877 dev->real_num_tx_queues = txq;
2881 qdisc_reset_all_tx_gt(dev, txq);
2883 netif_reset_xps_queues_gt(dev, txq);
2887 dev->real_num_tx_queues = txq;
2892 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2896 * netif_set_real_num_rx_queues - set actual number of RX queues used
2897 * @dev: Network device
2898 * @rxq: Actual number of RX queues
2900 * This must be called either with the rtnl_lock held or before
2901 * registration of the net device. Returns 0 on success, or a
2902 * negative error code. If called before registration, it always
2905 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2909 if (rxq < 1 || rxq > dev->num_rx_queues)
2912 if (dev->reg_state == NETREG_REGISTERED) {
2915 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2921 dev->real_num_rx_queues = rxq;
2924 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2928 * netif_get_num_default_rss_queues - default number of RSS queues
2930 * This routine should set an upper limit on the number of RSS queues
2931 * used by default by multiqueue devices.
2933 int netif_get_num_default_rss_queues(void)
2935 return is_kdump_kernel() ?
2936 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2938 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2940 static void __netif_reschedule(struct Qdisc *q)
2942 struct softnet_data *sd;
2943 unsigned long flags;
2945 local_irq_save(flags);
2946 sd = this_cpu_ptr(&softnet_data);
2947 q->next_sched = NULL;
2948 *sd->output_queue_tailp = q;
2949 sd->output_queue_tailp = &q->next_sched;
2950 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2951 local_irq_restore(flags);
2954 void __netif_schedule(struct Qdisc *q)
2956 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2957 __netif_reschedule(q);
2959 EXPORT_SYMBOL(__netif_schedule);
2961 struct dev_kfree_skb_cb {
2962 enum skb_free_reason reason;
2965 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2967 return (struct dev_kfree_skb_cb *)skb->cb;
2970 void netif_schedule_queue(struct netdev_queue *txq)
2973 if (!netif_xmit_stopped(txq)) {
2974 struct Qdisc *q = rcu_dereference(txq->qdisc);
2976 __netif_schedule(q);
2980 EXPORT_SYMBOL(netif_schedule_queue);
2982 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2984 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2988 q = rcu_dereference(dev_queue->qdisc);
2989 __netif_schedule(q);
2993 EXPORT_SYMBOL(netif_tx_wake_queue);
2995 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2997 unsigned long flags;
3002 if (likely(refcount_read(&skb->users) == 1)) {
3004 refcount_set(&skb->users, 0);
3005 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3008 get_kfree_skb_cb(skb)->reason = reason;
3009 local_irq_save(flags);
3010 skb->next = __this_cpu_read(softnet_data.completion_queue);
3011 __this_cpu_write(softnet_data.completion_queue, skb);
3012 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3013 local_irq_restore(flags);
3015 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3017 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3019 if (in_irq() || irqs_disabled())
3020 __dev_kfree_skb_irq(skb, reason);
3024 EXPORT_SYMBOL(__dev_kfree_skb_any);
3028 * netif_device_detach - mark device as removed
3029 * @dev: network device
3031 * Mark device as removed from system and therefore no longer available.
3033 void netif_device_detach(struct net_device *dev)
3035 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3036 netif_running(dev)) {
3037 netif_tx_stop_all_queues(dev);
3040 EXPORT_SYMBOL(netif_device_detach);
3043 * netif_device_attach - mark device as attached
3044 * @dev: network device
3046 * Mark device as attached from system and restart if needed.
3048 void netif_device_attach(struct net_device *dev)
3050 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3051 netif_running(dev)) {
3052 netif_tx_wake_all_queues(dev);
3053 __netdev_watchdog_up(dev);
3056 EXPORT_SYMBOL(netif_device_attach);
3059 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3060 * to be used as a distribution range.
3062 static u16 skb_tx_hash(const struct net_device *dev,
3063 const struct net_device *sb_dev,
3064 struct sk_buff *skb)
3068 u16 qcount = dev->real_num_tx_queues;
3071 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3073 qoffset = sb_dev->tc_to_txq[tc].offset;
3074 qcount = sb_dev->tc_to_txq[tc].count;
3077 if (skb_rx_queue_recorded(skb)) {
3078 hash = skb_get_rx_queue(skb);
3079 while (unlikely(hash >= qcount))
3081 return hash + qoffset;
3084 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3087 static void skb_warn_bad_offload(const struct sk_buff *skb)
3089 static const netdev_features_t null_features;
3090 struct net_device *dev = skb->dev;
3091 const char *name = "";
3093 if (!net_ratelimit())
3097 if (dev->dev.parent)
3098 name = dev_driver_string(dev->dev.parent);
3100 name = netdev_name(dev);
3102 skb_dump(KERN_WARNING, skb, false);
3103 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3104 name, dev ? &dev->features : &null_features,
3105 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3109 * Invalidate hardware checksum when packet is to be mangled, and
3110 * complete checksum manually on outgoing path.
3112 int skb_checksum_help(struct sk_buff *skb)
3115 int ret = 0, offset;
3117 if (skb->ip_summed == CHECKSUM_COMPLETE)
3118 goto out_set_summed;
3120 if (unlikely(skb_shinfo(skb)->gso_size)) {
3121 skb_warn_bad_offload(skb);
3125 /* Before computing a checksum, we should make sure no frag could
3126 * be modified by an external entity : checksum could be wrong.
3128 if (skb_has_shared_frag(skb)) {
3129 ret = __skb_linearize(skb);
3134 offset = skb_checksum_start_offset(skb);
3135 BUG_ON(offset >= skb_headlen(skb));
3136 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3138 offset += skb->csum_offset;
3139 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
3141 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3145 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3147 skb->ip_summed = CHECKSUM_NONE;
3151 EXPORT_SYMBOL(skb_checksum_help);
3153 int skb_crc32c_csum_help(struct sk_buff *skb)
3156 int ret = 0, offset, start;
3158 if (skb->ip_summed != CHECKSUM_PARTIAL)
3161 if (unlikely(skb_is_gso(skb)))
3164 /* Before computing a checksum, we should make sure no frag could
3165 * be modified by an external entity : checksum could be wrong.
3167 if (unlikely(skb_has_shared_frag(skb))) {
3168 ret = __skb_linearize(skb);
3172 start = skb_checksum_start_offset(skb);
3173 offset = start + offsetof(struct sctphdr, checksum);
3174 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3179 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3183 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3184 skb->len - start, ~(__u32)0,
3186 *(__le32 *)(skb->data + offset) = crc32c_csum;
3187 skb->ip_summed = CHECKSUM_NONE;
3188 skb->csum_not_inet = 0;
3193 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3195 __be16 type = skb->protocol;
3197 /* Tunnel gso handlers can set protocol to ethernet. */
3198 if (type == htons(ETH_P_TEB)) {
3201 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3204 eth = (struct ethhdr *)skb->data;
3205 type = eth->h_proto;
3208 return __vlan_get_protocol(skb, type, depth);
3212 * skb_mac_gso_segment - mac layer segmentation handler.
3213 * @skb: buffer to segment
3214 * @features: features for the output path (see dev->features)
3216 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3217 netdev_features_t features)
3219 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3220 struct packet_offload *ptype;
3221 int vlan_depth = skb->mac_len;
3222 __be16 type = skb_network_protocol(skb, &vlan_depth);
3224 if (unlikely(!type))
3225 return ERR_PTR(-EINVAL);
3227 __skb_pull(skb, vlan_depth);
3230 list_for_each_entry_rcu(ptype, &offload_base, list) {
3231 if (ptype->type == type && ptype->callbacks.gso_segment) {
3232 segs = ptype->callbacks.gso_segment(skb, features);
3238 __skb_push(skb, skb->data - skb_mac_header(skb));
3242 EXPORT_SYMBOL(skb_mac_gso_segment);
3245 /* openvswitch calls this on rx path, so we need a different check.
3247 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3250 return skb->ip_summed != CHECKSUM_PARTIAL &&
3251 skb->ip_summed != CHECKSUM_UNNECESSARY;
3253 return skb->ip_summed == CHECKSUM_NONE;
3257 * __skb_gso_segment - Perform segmentation on skb.
3258 * @skb: buffer to segment
3259 * @features: features for the output path (see dev->features)
3260 * @tx_path: whether it is called in TX path
3262 * This function segments the given skb and returns a list of segments.
3264 * It may return NULL if the skb requires no segmentation. This is
3265 * only possible when GSO is used for verifying header integrity.
3267 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
3269 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3270 netdev_features_t features, bool tx_path)
3272 struct sk_buff *segs;
3274 if (unlikely(skb_needs_check(skb, tx_path))) {
3277 /* We're going to init ->check field in TCP or UDP header */
3278 err = skb_cow_head(skb, 0);
3280 return ERR_PTR(err);
3283 /* Only report GSO partial support if it will enable us to
3284 * support segmentation on this frame without needing additional
3287 if (features & NETIF_F_GSO_PARTIAL) {
3288 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3289 struct net_device *dev = skb->dev;
3291 partial_features |= dev->features & dev->gso_partial_features;
3292 if (!skb_gso_ok(skb, features | partial_features))
3293 features &= ~NETIF_F_GSO_PARTIAL;
3296 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
3297 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3299 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3300 SKB_GSO_CB(skb)->encap_level = 0;
3302 skb_reset_mac_header(skb);
3303 skb_reset_mac_len(skb);
3305 segs = skb_mac_gso_segment(skb, features);
3307 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3308 skb_warn_bad_offload(skb);
3312 EXPORT_SYMBOL(__skb_gso_segment);
3314 /* Take action when hardware reception checksum errors are detected. */
3316 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3318 if (net_ratelimit()) {
3319 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3320 skb_dump(KERN_ERR, skb, true);
3324 EXPORT_SYMBOL(netdev_rx_csum_fault);
3327 /* XXX: check that highmem exists at all on the given machine. */
3328 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3330 #ifdef CONFIG_HIGHMEM
3333 if (!(dev->features & NETIF_F_HIGHDMA)) {
3334 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3335 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3337 if (PageHighMem(skb_frag_page(frag)))
3345 /* If MPLS offload request, verify we are testing hardware MPLS features
3346 * instead of standard features for the netdev.
3348 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3349 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3350 netdev_features_t features,
3353 if (eth_p_mpls(type))
3354 features &= skb->dev->mpls_features;
3359 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3360 netdev_features_t features,
3367 static netdev_features_t harmonize_features(struct sk_buff *skb,
3368 netdev_features_t features)
3373 type = skb_network_protocol(skb, &tmp);
3374 features = net_mpls_features(skb, features, type);
3376 if (skb->ip_summed != CHECKSUM_NONE &&
3377 !can_checksum_protocol(features, type)) {
3378 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3380 if (illegal_highdma(skb->dev, skb))
3381 features &= ~NETIF_F_SG;
3386 netdev_features_t passthru_features_check(struct sk_buff *skb,
3387 struct net_device *dev,
3388 netdev_features_t features)
3392 EXPORT_SYMBOL(passthru_features_check);
3394 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3395 struct net_device *dev,
3396 netdev_features_t features)
3398 return vlan_features_check(skb, features);
3401 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3402 struct net_device *dev,
3403 netdev_features_t features)
3405 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3407 if (gso_segs > dev->gso_max_segs)
3408 return features & ~NETIF_F_GSO_MASK;
3410 /* Support for GSO partial features requires software
3411 * intervention before we can actually process the packets
3412 * so we need to strip support for any partial features now
3413 * and we can pull them back in after we have partially
3414 * segmented the frame.
3416 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3417 features &= ~dev->gso_partial_features;
3419 /* Make sure to clear the IPv4 ID mangling feature if the
3420 * IPv4 header has the potential to be fragmented.
3422 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3423 struct iphdr *iph = skb->encapsulation ?
3424 inner_ip_hdr(skb) : ip_hdr(skb);
3426 if (!(iph->frag_off & htons(IP_DF)))
3427 features &= ~NETIF_F_TSO_MANGLEID;
3433 netdev_features_t netif_skb_features(struct sk_buff *skb)
3435 struct net_device *dev = skb->dev;
3436 netdev_features_t features = dev->features;
3438 if (skb_is_gso(skb))
3439 features = gso_features_check(skb, dev, features);
3441 /* If encapsulation offload request, verify we are testing
3442 * hardware encapsulation features instead of standard
3443 * features for the netdev
3445 if (skb->encapsulation)
3446 features &= dev->hw_enc_features;
3448 if (skb_vlan_tagged(skb))
3449 features = netdev_intersect_features(features,
3450 dev->vlan_features |
3451 NETIF_F_HW_VLAN_CTAG_TX |
3452 NETIF_F_HW_VLAN_STAG_TX);
3454 if (dev->netdev_ops->ndo_features_check)
3455 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3458 features &= dflt_features_check(skb, dev, features);
3460 return harmonize_features(skb, features);
3462 EXPORT_SYMBOL(netif_skb_features);
3464 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3465 struct netdev_queue *txq, bool more)
3470 if (dev_nit_active(dev))
3471 dev_queue_xmit_nit(skb, dev);
3474 trace_net_dev_start_xmit(skb, dev);
3475 rc = netdev_start_xmit(skb, dev, txq, more);
3476 trace_net_dev_xmit(skb, rc, dev, len);
3481 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3482 struct netdev_queue *txq, int *ret)
3484 struct sk_buff *skb = first;
3485 int rc = NETDEV_TX_OK;
3488 struct sk_buff *next = skb->next;
3490 skb_mark_not_on_list(skb);
3491 rc = xmit_one(skb, dev, txq, next != NULL);
3492 if (unlikely(!dev_xmit_complete(rc))) {
3498 if (netif_tx_queue_stopped(txq) && skb) {
3499 rc = NETDEV_TX_BUSY;
3509 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3510 netdev_features_t features)
3512 if (skb_vlan_tag_present(skb) &&
3513 !vlan_hw_offload_capable(features, skb->vlan_proto))
3514 skb = __vlan_hwaccel_push_inside(skb);
3518 int skb_csum_hwoffload_help(struct sk_buff *skb,
3519 const netdev_features_t features)
3521 if (unlikely(skb->csum_not_inet))
3522 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3523 skb_crc32c_csum_help(skb);
3525 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3527 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3529 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3531 netdev_features_t features;
3533 features = netif_skb_features(skb);
3534 skb = validate_xmit_vlan(skb, features);
3538 skb = sk_validate_xmit_skb(skb, dev);
3542 if (netif_needs_gso(skb, features)) {
3543 struct sk_buff *segs;
3545 segs = skb_gso_segment(skb, features);
3553 if (skb_needs_linearize(skb, features) &&
3554 __skb_linearize(skb))
3557 /* If packet is not checksummed and device does not
3558 * support checksumming for this protocol, complete
3559 * checksumming here.
3561 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3562 if (skb->encapsulation)
3563 skb_set_inner_transport_header(skb,
3564 skb_checksum_start_offset(skb));
3566 skb_set_transport_header(skb,
3567 skb_checksum_start_offset(skb));
3568 if (skb_csum_hwoffload_help(skb, features))
3573 skb = validate_xmit_xfrm(skb, features, again);
3580 atomic_long_inc(&dev->tx_dropped);
3584 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3586 struct sk_buff *next, *head = NULL, *tail;
3588 for (; skb != NULL; skb = next) {
3590 skb_mark_not_on_list(skb);
3592 /* in case skb wont be segmented, point to itself */
3595 skb = validate_xmit_skb(skb, dev, again);
3603 /* If skb was segmented, skb->prev points to
3604 * the last segment. If not, it still contains skb.
3610 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3612 static void qdisc_pkt_len_init(struct sk_buff *skb)
3614 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3616 qdisc_skb_cb(skb)->pkt_len = skb->len;
3618 /* To get more precise estimation of bytes sent on wire,
3619 * we add to pkt_len the headers size of all segments
3621 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3622 unsigned int hdr_len;
3623 u16 gso_segs = shinfo->gso_segs;
3625 /* mac layer + network layer */
3626 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3628 /* + transport layer */
3629 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3630 const struct tcphdr *th;
3631 struct tcphdr _tcphdr;
3633 th = skb_header_pointer(skb, skb_transport_offset(skb),
3634 sizeof(_tcphdr), &_tcphdr);
3636 hdr_len += __tcp_hdrlen(th);
3638 struct udphdr _udphdr;
3640 if (skb_header_pointer(skb, skb_transport_offset(skb),
3641 sizeof(_udphdr), &_udphdr))
3642 hdr_len += sizeof(struct udphdr);
3645 if (shinfo->gso_type & SKB_GSO_DODGY)
3646 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3649 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3653 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3654 struct net_device *dev,
3655 struct netdev_queue *txq)
3657 spinlock_t *root_lock = qdisc_lock(q);
3658 struct sk_buff *to_free = NULL;
3662 qdisc_calculate_pkt_len(skb, q);
3664 if (q->flags & TCQ_F_NOLOCK) {
3665 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3668 if (unlikely(to_free))
3669 kfree_skb_list(to_free);
3674 * Heuristic to force contended enqueues to serialize on a
3675 * separate lock before trying to get qdisc main lock.
3676 * This permits qdisc->running owner to get the lock more
3677 * often and dequeue packets faster.
3679 contended = qdisc_is_running(q);
3680 if (unlikely(contended))
3681 spin_lock(&q->busylock);
3683 spin_lock(root_lock);
3684 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3685 __qdisc_drop(skb, &to_free);
3687 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3688 qdisc_run_begin(q)) {
3690 * This is a work-conserving queue; there are no old skbs
3691 * waiting to be sent out; and the qdisc is not running -
3692 * xmit the skb directly.
3695 qdisc_bstats_update(q, skb);
3697 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3698 if (unlikely(contended)) {
3699 spin_unlock(&q->busylock);
3706 rc = NET_XMIT_SUCCESS;
3708 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3709 if (qdisc_run_begin(q)) {
3710 if (unlikely(contended)) {
3711 spin_unlock(&q->busylock);
3718 spin_unlock(root_lock);
3719 if (unlikely(to_free))
3720 kfree_skb_list(to_free);
3721 if (unlikely(contended))
3722 spin_unlock(&q->busylock);
3726 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3727 static void skb_update_prio(struct sk_buff *skb)
3729 const struct netprio_map *map;
3730 const struct sock *sk;
3731 unsigned int prioidx;
3735 map = rcu_dereference_bh(skb->dev->priomap);
3738 sk = skb_to_full_sk(skb);
3742 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3744 if (prioidx < map->priomap_len)
3745 skb->priority = map->priomap[prioidx];
3748 #define skb_update_prio(skb)
3752 * dev_loopback_xmit - loop back @skb
3753 * @net: network namespace this loopback is happening in
3754 * @sk: sk needed to be a netfilter okfn
3755 * @skb: buffer to transmit
3757 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3759 skb_reset_mac_header(skb);
3760 __skb_pull(skb, skb_network_offset(skb));
3761 skb->pkt_type = PACKET_LOOPBACK;
3762 skb->ip_summed = CHECKSUM_UNNECESSARY;
3763 WARN_ON(!skb_dst(skb));
3768 EXPORT_SYMBOL(dev_loopback_xmit);
3770 #ifdef CONFIG_NET_EGRESS
3771 static struct sk_buff *
3772 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3774 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3775 struct tcf_result cl_res;
3780 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3781 mini_qdisc_bstats_cpu_update(miniq, skb);
3783 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3785 case TC_ACT_RECLASSIFY:
3786 skb->tc_index = TC_H_MIN(cl_res.classid);
3789 mini_qdisc_qstats_cpu_drop(miniq);
3790 *ret = NET_XMIT_DROP;
3796 *ret = NET_XMIT_SUCCESS;
3799 case TC_ACT_REDIRECT:
3800 /* No need to push/pop skb's mac_header here on egress! */
3801 skb_do_redirect(skb);
3802 *ret = NET_XMIT_SUCCESS;
3810 #endif /* CONFIG_NET_EGRESS */
3813 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3814 struct xps_dev_maps *dev_maps, unsigned int tci)
3816 struct xps_map *map;
3817 int queue_index = -1;
3821 tci += netdev_get_prio_tc_map(dev, skb->priority);
3824 map = rcu_dereference(dev_maps->attr_map[tci]);
3827 queue_index = map->queues[0];
3829 queue_index = map->queues[reciprocal_scale(
3830 skb_get_hash(skb), map->len)];
3831 if (unlikely(queue_index >= dev->real_num_tx_queues))
3838 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3839 struct sk_buff *skb)
3842 struct xps_dev_maps *dev_maps;
3843 struct sock *sk = skb->sk;
3844 int queue_index = -1;
3846 if (!static_key_false(&xps_needed))
3850 if (!static_key_false(&xps_rxqs_needed))
3853 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map);
3855 int tci = sk_rx_queue_get(sk);
3857 if (tci >= 0 && tci < dev->num_rx_queues)
3858 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3863 if (queue_index < 0) {
3864 dev_maps = rcu_dereference(sb_dev->xps_cpus_map);
3866 unsigned int tci = skb->sender_cpu - 1;
3868 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3880 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3881 struct net_device *sb_dev)
3885 EXPORT_SYMBOL(dev_pick_tx_zero);
3887 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
3888 struct net_device *sb_dev)
3890 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3892 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3894 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
3895 struct net_device *sb_dev)
3897 struct sock *sk = skb->sk;
3898 int queue_index = sk_tx_queue_get(sk);
3900 sb_dev = sb_dev ? : dev;
3902 if (queue_index < 0 || skb->ooo_okay ||
3903 queue_index >= dev->real_num_tx_queues) {
3904 int new_index = get_xps_queue(dev, sb_dev, skb);
3907 new_index = skb_tx_hash(dev, sb_dev, skb);
3909 if (queue_index != new_index && sk &&
3911 rcu_access_pointer(sk->sk_dst_cache))
3912 sk_tx_queue_set(sk, new_index);
3914 queue_index = new_index;
3919 EXPORT_SYMBOL(netdev_pick_tx);
3921 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
3922 struct sk_buff *skb,
3923 struct net_device *sb_dev)
3925 int queue_index = 0;
3928 u32 sender_cpu = skb->sender_cpu - 1;
3930 if (sender_cpu >= (u32)NR_CPUS)
3931 skb->sender_cpu = raw_smp_processor_id() + 1;
3934 if (dev->real_num_tx_queues != 1) {
3935 const struct net_device_ops *ops = dev->netdev_ops;
3937 if (ops->ndo_select_queue)
3938 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
3940 queue_index = netdev_pick_tx(dev, skb, sb_dev);
3942 queue_index = netdev_cap_txqueue(dev, queue_index);
3945 skb_set_queue_mapping(skb, queue_index);
3946 return netdev_get_tx_queue(dev, queue_index);
3950 * __dev_queue_xmit - transmit a buffer
3951 * @skb: buffer to transmit
3952 * @sb_dev: suboordinate device used for L2 forwarding offload
3954 * Queue a buffer for transmission to a network device. The caller must
3955 * have set the device and priority and built the buffer before calling
3956 * this function. The function can be called from an interrupt.
3958 * A negative errno code is returned on a failure. A success does not
3959 * guarantee the frame will be transmitted as it may be dropped due
3960 * to congestion or traffic shaping.
3962 * -----------------------------------------------------------------------------------
3963 * I notice this method can also return errors from the queue disciplines,
3964 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3967 * Regardless of the return value, the skb is consumed, so it is currently
3968 * difficult to retry a send to this method. (You can bump the ref count
3969 * before sending to hold a reference for retry if you are careful.)
3971 * When calling this method, interrupts MUST be enabled. This is because
3972 * the BH enable code must have IRQs enabled so that it will not deadlock.
3975 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
3977 struct net_device *dev = skb->dev;
3978 struct netdev_queue *txq;
3983 skb_reset_mac_header(skb);
3985 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3986 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3988 /* Disable soft irqs for various locks below. Also
3989 * stops preemption for RCU.
3993 skb_update_prio(skb);
3995 qdisc_pkt_len_init(skb);
3996 #ifdef CONFIG_NET_CLS_ACT
3997 skb->tc_at_ingress = 0;
3998 # ifdef CONFIG_NET_EGRESS
3999 if (static_branch_unlikely(&egress_needed_key)) {
4000 skb = sch_handle_egress(skb, &rc, dev);
4006 /* If device/qdisc don't need skb->dst, release it right now while
4007 * its hot in this cpu cache.
4009 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4014 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4015 q = rcu_dereference_bh(txq->qdisc);
4017 trace_net_dev_queue(skb);
4019 rc = __dev_xmit_skb(skb, q, dev, txq);
4023 /* The device has no queue. Common case for software devices:
4024 * loopback, all the sorts of tunnels...
4026 * Really, it is unlikely that netif_tx_lock protection is necessary
4027 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4029 * However, it is possible, that they rely on protection
4032 * Check this and shot the lock. It is not prone from deadlocks.
4033 *Either shot noqueue qdisc, it is even simpler 8)
4035 if (dev->flags & IFF_UP) {
4036 int cpu = smp_processor_id(); /* ok because BHs are off */
4038 if (txq->xmit_lock_owner != cpu) {
4039 if (dev_xmit_recursion())
4040 goto recursion_alert;
4042 skb = validate_xmit_skb(skb, dev, &again);
4046 HARD_TX_LOCK(dev, txq, cpu);
4048 if (!netif_xmit_stopped(txq)) {
4049 dev_xmit_recursion_inc();
4050 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4051 dev_xmit_recursion_dec();
4052 if (dev_xmit_complete(rc)) {
4053 HARD_TX_UNLOCK(dev, txq);
4057 HARD_TX_UNLOCK(dev, txq);
4058 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4061 /* Recursion is detected! It is possible,
4065 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4071 rcu_read_unlock_bh();
4073 atomic_long_inc(&dev->tx_dropped);
4074 kfree_skb_list(skb);
4077 rcu_read_unlock_bh();
4081 int dev_queue_xmit(struct sk_buff *skb)
4083 return __dev_queue_xmit(skb, NULL);
4085 EXPORT_SYMBOL(dev_queue_xmit);
4087 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
4089 return __dev_queue_xmit(skb, sb_dev);
4091 EXPORT_SYMBOL(dev_queue_xmit_accel);
4093 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4095 struct net_device *dev = skb->dev;
4096 struct sk_buff *orig_skb = skb;
4097 struct netdev_queue *txq;
4098 int ret = NETDEV_TX_BUSY;
4101 if (unlikely(!netif_running(dev) ||
4102 !netif_carrier_ok(dev)))
4105 skb = validate_xmit_skb_list(skb, dev, &again);
4106 if (skb != orig_skb)
4109 skb_set_queue_mapping(skb, queue_id);
4110 txq = skb_get_tx_queue(dev, skb);
4114 HARD_TX_LOCK(dev, txq, smp_processor_id());
4115 if (!netif_xmit_frozen_or_drv_stopped(txq))
4116 ret = netdev_start_xmit(skb, dev, txq, false);
4117 HARD_TX_UNLOCK(dev, txq);
4121 if (!dev_xmit_complete(ret))
4126 atomic_long_inc(&dev->tx_dropped);
4127 kfree_skb_list(skb);
4128 return NET_XMIT_DROP;
4130 EXPORT_SYMBOL(dev_direct_xmit);
4132 /*************************************************************************
4134 *************************************************************************/
4136 int netdev_max_backlog __read_mostly = 1000;
4137 EXPORT_SYMBOL(netdev_max_backlog);
4139 int netdev_tstamp_prequeue __read_mostly = 1;
4140 int netdev_budget __read_mostly = 300;
4141 unsigned int __read_mostly netdev_budget_usecs = 2000;
4142 int weight_p __read_mostly = 64; /* old backlog weight */
4143 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4144 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4145 int dev_rx_weight __read_mostly = 64;
4146 int dev_tx_weight __read_mostly = 64;
4147 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
4148 int gro_normal_batch __read_mostly = 8;
4150 /* Called with irq disabled */
4151 static inline void ____napi_schedule(struct softnet_data *sd,
4152 struct napi_struct *napi)
4154 list_add_tail(&napi->poll_list, &sd->poll_list);
4155 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4160 /* One global table that all flow-based protocols share. */
4161 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4162 EXPORT_SYMBOL(rps_sock_flow_table);
4163 u32 rps_cpu_mask __read_mostly;
4164 EXPORT_SYMBOL(rps_cpu_mask);
4166 struct static_key_false rps_needed __read_mostly;
4167 EXPORT_SYMBOL(rps_needed);
4168 struct static_key_false rfs_needed __read_mostly;
4169 EXPORT_SYMBOL(rfs_needed);
4171 static struct rps_dev_flow *
4172 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4173 struct rps_dev_flow *rflow, u16 next_cpu)
4175 if (next_cpu < nr_cpu_ids) {
4176 #ifdef CONFIG_RFS_ACCEL
4177 struct netdev_rx_queue *rxqueue;
4178 struct rps_dev_flow_table *flow_table;
4179 struct rps_dev_flow *old_rflow;
4184 /* Should we steer this flow to a different hardware queue? */
4185 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4186 !(dev->features & NETIF_F_NTUPLE))
4188 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4189 if (rxq_index == skb_get_rx_queue(skb))
4192 rxqueue = dev->_rx + rxq_index;
4193 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4196 flow_id = skb_get_hash(skb) & flow_table->mask;
4197 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4198 rxq_index, flow_id);
4202 rflow = &flow_table->flows[flow_id];
4204 if (old_rflow->filter == rflow->filter)
4205 old_rflow->filter = RPS_NO_FILTER;
4209 per_cpu(softnet_data, next_cpu).input_queue_head;
4212 rflow->cpu = next_cpu;
4217 * get_rps_cpu is called from netif_receive_skb and returns the target
4218 * CPU from the RPS map of the receiving queue for a given skb.
4219 * rcu_read_lock must be held on entry.
4221 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4222 struct rps_dev_flow **rflowp)
4224 const struct rps_sock_flow_table *sock_flow_table;
4225 struct netdev_rx_queue *rxqueue = dev->_rx;
4226 struct rps_dev_flow_table *flow_table;
4227 struct rps_map *map;
4232 if (skb_rx_queue_recorded(skb)) {
4233 u16 index = skb_get_rx_queue(skb);
4235 if (unlikely(index >= dev->real_num_rx_queues)) {
4236 WARN_ONCE(dev->real_num_rx_queues > 1,
4237 "%s received packet on queue %u, but number "
4238 "of RX queues is %u\n",
4239 dev->name, index, dev->real_num_rx_queues);
4245 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4247 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4248 map = rcu_dereference(rxqueue->rps_map);
4249 if (!flow_table && !map)
4252 skb_reset_network_header(skb);
4253 hash = skb_get_hash(skb);
4257 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4258 if (flow_table && sock_flow_table) {
4259 struct rps_dev_flow *rflow;
4263 /* First check into global flow table if there is a match */
4264 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4265 if ((ident ^ hash) & ~rps_cpu_mask)
4268 next_cpu = ident & rps_cpu_mask;
4270 /* OK, now we know there is a match,
4271 * we can look at the local (per receive queue) flow table
4273 rflow = &flow_table->flows[hash & flow_table->mask];
4277 * If the desired CPU (where last recvmsg was done) is
4278 * different from current CPU (one in the rx-queue flow
4279 * table entry), switch if one of the following holds:
4280 * - Current CPU is unset (>= nr_cpu_ids).
4281 * - Current CPU is offline.
4282 * - The current CPU's queue tail has advanced beyond the
4283 * last packet that was enqueued using this table entry.
4284 * This guarantees that all previous packets for the flow
4285 * have been dequeued, thus preserving in order delivery.
4287 if (unlikely(tcpu != next_cpu) &&
4288 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4289 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4290 rflow->last_qtail)) >= 0)) {
4292 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4295 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4305 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4306 if (cpu_online(tcpu)) {
4316 #ifdef CONFIG_RFS_ACCEL
4319 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4320 * @dev: Device on which the filter was set
4321 * @rxq_index: RX queue index
4322 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4323 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4325 * Drivers that implement ndo_rx_flow_steer() should periodically call
4326 * this function for each installed filter and remove the filters for
4327 * which it returns %true.
4329 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4330 u32 flow_id, u16 filter_id)
4332 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4333 struct rps_dev_flow_table *flow_table;
4334 struct rps_dev_flow *rflow;
4339 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4340 if (flow_table && flow_id <= flow_table->mask) {
4341 rflow = &flow_table->flows[flow_id];
4342 cpu = READ_ONCE(rflow->cpu);
4343 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4344 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4345 rflow->last_qtail) <
4346 (int)(10 * flow_table->mask)))
4352 EXPORT_SYMBOL(rps_may_expire_flow);
4354 #endif /* CONFIG_RFS_ACCEL */
4356 /* Called from hardirq (IPI) context */
4357 static void rps_trigger_softirq(void *data)
4359 struct softnet_data *sd = data;
4361 ____napi_schedule(sd, &sd->backlog);
4365 #endif /* CONFIG_RPS */
4368 * Check if this softnet_data structure is another cpu one
4369 * If yes, queue it to our IPI list and return 1
4372 static int rps_ipi_queued(struct softnet_data *sd)
4375 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4378 sd->rps_ipi_next = mysd->rps_ipi_list;
4379 mysd->rps_ipi_list = sd;
4381 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4384 #endif /* CONFIG_RPS */
4388 #ifdef CONFIG_NET_FLOW_LIMIT
4389 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4392 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4394 #ifdef CONFIG_NET_FLOW_LIMIT
4395 struct sd_flow_limit *fl;
4396 struct softnet_data *sd;
4397 unsigned int old_flow, new_flow;
4399 if (qlen < (netdev_max_backlog >> 1))
4402 sd = this_cpu_ptr(&softnet_data);
4405 fl = rcu_dereference(sd->flow_limit);
4407 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4408 old_flow = fl->history[fl->history_head];
4409 fl->history[fl->history_head] = new_flow;
4412 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4414 if (likely(fl->buckets[old_flow]))
4415 fl->buckets[old_flow]--;
4417 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4429 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4430 * queue (may be a remote CPU queue).
4432 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4433 unsigned int *qtail)
4435 struct softnet_data *sd;
4436 unsigned long flags;
4439 sd = &per_cpu(softnet_data, cpu);
4441 local_irq_save(flags);
4444 if (!netif_running(skb->dev))
4446 qlen = skb_queue_len(&sd->input_pkt_queue);
4447 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4450 __skb_queue_tail(&sd->input_pkt_queue, skb);
4451 input_queue_tail_incr_save(sd, qtail);
4453 local_irq_restore(flags);
4454 return NET_RX_SUCCESS;
4457 /* Schedule NAPI for backlog device
4458 * We can use non atomic operation since we own the queue lock
4460 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4461 if (!rps_ipi_queued(sd))
4462 ____napi_schedule(sd, &sd->backlog);
4471 local_irq_restore(flags);
4473 atomic_long_inc(&skb->dev->rx_dropped);
4478 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4480 struct net_device *dev = skb->dev;
4481 struct netdev_rx_queue *rxqueue;
4485 if (skb_rx_queue_recorded(skb)) {
4486 u16 index = skb_get_rx_queue(skb);
4488 if (unlikely(index >= dev->real_num_rx_queues)) {
4489 WARN_ONCE(dev->real_num_rx_queues > 1,
4490 "%s received packet on queue %u, but number "
4491 "of RX queues is %u\n",
4492 dev->name, index, dev->real_num_rx_queues);
4494 return rxqueue; /* Return first rxqueue */
4501 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4502 struct xdp_buff *xdp,
4503 struct bpf_prog *xdp_prog)
4505 struct netdev_rx_queue *rxqueue;
4506 void *orig_data, *orig_data_end;
4507 u32 metalen, act = XDP_DROP;
4508 __be16 orig_eth_type;
4514 /* Reinjected packets coming from act_mirred or similar should
4515 * not get XDP generic processing.
4517 if (skb_is_tc_redirected(skb))
4520 /* XDP packets must be linear and must have sufficient headroom
4521 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4522 * native XDP provides, thus we need to do it here as well.
4524 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4525 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4526 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4527 int troom = skb->tail + skb->data_len - skb->end;
4529 /* In case we have to go down the path and also linearize,
4530 * then lets do the pskb_expand_head() work just once here.
4532 if (pskb_expand_head(skb,
4533 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4534 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4536 if (skb_linearize(skb))
4540 /* The XDP program wants to see the packet starting at the MAC
4543 mac_len = skb->data - skb_mac_header(skb);
4544 hlen = skb_headlen(skb) + mac_len;
4545 xdp->data = skb->data - mac_len;
4546 xdp->data_meta = xdp->data;
4547 xdp->data_end = xdp->data + hlen;
4548 xdp->data_hard_start = skb->data - skb_headroom(skb);
4549 orig_data_end = xdp->data_end;
4550 orig_data = xdp->data;
4551 eth = (struct ethhdr *)xdp->data;
4552 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4553 orig_eth_type = eth->h_proto;
4555 rxqueue = netif_get_rxqueue(skb);
4556 xdp->rxq = &rxqueue->xdp_rxq;
4558 act = bpf_prog_run_xdp(xdp_prog, xdp);
4560 /* check if bpf_xdp_adjust_head was used */
4561 off = xdp->data - orig_data;
4564 __skb_pull(skb, off);
4566 __skb_push(skb, -off);
4568 skb->mac_header += off;
4569 skb_reset_network_header(skb);
4572 /* check if bpf_xdp_adjust_tail was used. it can only "shrink"
4575 off = orig_data_end - xdp->data_end;
4577 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4582 /* check if XDP changed eth hdr such SKB needs update */
4583 eth = (struct ethhdr *)xdp->data;
4584 if ((orig_eth_type != eth->h_proto) ||
4585 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4586 __skb_push(skb, ETH_HLEN);
4587 skb->protocol = eth_type_trans(skb, skb->dev);
4593 __skb_push(skb, mac_len);
4596 metalen = xdp->data - xdp->data_meta;
4598 skb_metadata_set(skb, metalen);
4601 bpf_warn_invalid_xdp_action(act);
4604 trace_xdp_exception(skb->dev, xdp_prog, act);
4615 /* When doing generic XDP we have to bypass the qdisc layer and the
4616 * network taps in order to match in-driver-XDP behavior.
4618 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4620 struct net_device *dev = skb->dev;
4621 struct netdev_queue *txq;
4622 bool free_skb = true;
4625 txq = netdev_core_pick_tx(dev, skb, NULL);
4626 cpu = smp_processor_id();
4627 HARD_TX_LOCK(dev, txq, cpu);
4628 if (!netif_xmit_stopped(txq)) {
4629 rc = netdev_start_xmit(skb, dev, txq, 0);
4630 if (dev_xmit_complete(rc))
4633 HARD_TX_UNLOCK(dev, txq);
4635 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4639 EXPORT_SYMBOL_GPL(generic_xdp_tx);
4641 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4643 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4646 struct xdp_buff xdp;
4650 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4651 if (act != XDP_PASS) {
4654 err = xdp_do_generic_redirect(skb->dev, skb,
4660 generic_xdp_tx(skb, xdp_prog);
4671 EXPORT_SYMBOL_GPL(do_xdp_generic);
4673 static int netif_rx_internal(struct sk_buff *skb)
4677 net_timestamp_check(netdev_tstamp_prequeue, skb);
4679 trace_netif_rx(skb);
4682 if (static_branch_unlikely(&rps_needed)) {
4683 struct rps_dev_flow voidflow, *rflow = &voidflow;
4689 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4691 cpu = smp_processor_id();
4693 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4702 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4709 * netif_rx - post buffer to the network code
4710 * @skb: buffer to post
4712 * This function receives a packet from a device driver and queues it for
4713 * the upper (protocol) levels to process. It always succeeds. The buffer
4714 * may be dropped during processing for congestion control or by the
4718 * NET_RX_SUCCESS (no congestion)
4719 * NET_RX_DROP (packet was dropped)
4723 int netif_rx(struct sk_buff *skb)
4727 trace_netif_rx_entry(skb);
4729 ret = netif_rx_internal(skb);
4730 trace_netif_rx_exit(ret);
4734 EXPORT_SYMBOL(netif_rx);
4736 int netif_rx_ni(struct sk_buff *skb)
4740 trace_netif_rx_ni_entry(skb);
4743 err = netif_rx_internal(skb);
4744 if (local_softirq_pending())
4747 trace_netif_rx_ni_exit(err);
4751 EXPORT_SYMBOL(netif_rx_ni);
4753 static __latent_entropy void net_tx_action(struct softirq_action *h)
4755 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4757 if (sd->completion_queue) {
4758 struct sk_buff *clist;
4760 local_irq_disable();
4761 clist = sd->completion_queue;
4762 sd->completion_queue = NULL;
4766 struct sk_buff *skb = clist;
4768 clist = clist->next;
4770 WARN_ON(refcount_read(&skb->users));
4771 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4772 trace_consume_skb(skb);
4774 trace_kfree_skb(skb, net_tx_action);
4776 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4779 __kfree_skb_defer(skb);
4782 __kfree_skb_flush();
4785 if (sd->output_queue) {
4788 local_irq_disable();
4789 head = sd->output_queue;
4790 sd->output_queue = NULL;
4791 sd->output_queue_tailp = &sd->output_queue;
4795 struct Qdisc *q = head;
4796 spinlock_t *root_lock = NULL;
4798 head = head->next_sched;
4800 if (!(q->flags & TCQ_F_NOLOCK)) {
4801 root_lock = qdisc_lock(q);
4802 spin_lock(root_lock);
4804 /* We need to make sure head->next_sched is read
4805 * before clearing __QDISC_STATE_SCHED
4807 smp_mb__before_atomic();
4808 clear_bit(__QDISC_STATE_SCHED, &q->state);
4811 spin_unlock(root_lock);
4815 xfrm_dev_backlog(sd);
4818 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4819 /* This hook is defined here for ATM LANE */
4820 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4821 unsigned char *addr) __read_mostly;
4822 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4825 static inline struct sk_buff *
4826 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4827 struct net_device *orig_dev)
4829 #ifdef CONFIG_NET_CLS_ACT
4830 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4831 struct tcf_result cl_res;
4833 /* If there's at least one ingress present somewhere (so
4834 * we get here via enabled static key), remaining devices
4835 * that are not configured with an ingress qdisc will bail
4842 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4846 qdisc_skb_cb(skb)->pkt_len = skb->len;
4847 skb->tc_at_ingress = 1;
4848 mini_qdisc_bstats_cpu_update(miniq, skb);
4850 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
4852 case TC_ACT_RECLASSIFY:
4853 skb->tc_index = TC_H_MIN(cl_res.classid);
4856 mini_qdisc_qstats_cpu_drop(miniq);
4864 case TC_ACT_REDIRECT:
4865 /* skb_mac_header check was done by cls/act_bpf, so
4866 * we can safely push the L2 header back before
4867 * redirecting to another netdev
4869 __skb_push(skb, skb->mac_len);
4870 skb_do_redirect(skb);
4872 case TC_ACT_CONSUMED:
4877 #endif /* CONFIG_NET_CLS_ACT */
4882 * netdev_is_rx_handler_busy - check if receive handler is registered
4883 * @dev: device to check
4885 * Check if a receive handler is already registered for a given device.
4886 * Return true if there one.
4888 * The caller must hold the rtnl_mutex.
4890 bool netdev_is_rx_handler_busy(struct net_device *dev)
4893 return dev && rtnl_dereference(dev->rx_handler);
4895 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4898 * netdev_rx_handler_register - register receive handler
4899 * @dev: device to register a handler for
4900 * @rx_handler: receive handler to register
4901 * @rx_handler_data: data pointer that is used by rx handler
4903 * Register a receive handler for a device. This handler will then be
4904 * called from __netif_receive_skb. A negative errno code is returned
4907 * The caller must hold the rtnl_mutex.
4909 * For a general description of rx_handler, see enum rx_handler_result.
4911 int netdev_rx_handler_register(struct net_device *dev,
4912 rx_handler_func_t *rx_handler,
4913 void *rx_handler_data)
4915 if (netdev_is_rx_handler_busy(dev))
4918 if (dev->priv_flags & IFF_NO_RX_HANDLER)
4921 /* Note: rx_handler_data must be set before rx_handler */
4922 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4923 rcu_assign_pointer(dev->rx_handler, rx_handler);
4927 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4930 * netdev_rx_handler_unregister - unregister receive handler
4931 * @dev: device to unregister a handler from
4933 * Unregister a receive handler from a device.
4935 * The caller must hold the rtnl_mutex.
4937 void netdev_rx_handler_unregister(struct net_device *dev)
4941 RCU_INIT_POINTER(dev->rx_handler, NULL);
4942 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4943 * section has a guarantee to see a non NULL rx_handler_data
4947 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4949 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4952 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4953 * the special handling of PFMEMALLOC skbs.
4955 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4957 switch (skb->protocol) {
4958 case htons(ETH_P_ARP):
4959 case htons(ETH_P_IP):
4960 case htons(ETH_P_IPV6):
4961 case htons(ETH_P_8021Q):
4962 case htons(ETH_P_8021AD):
4969 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4970 int *ret, struct net_device *orig_dev)
4972 if (nf_hook_ingress_active(skb)) {
4976 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4981 ingress_retval = nf_hook_ingress(skb);
4983 return ingress_retval;
4988 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
4989 struct packet_type **ppt_prev)
4991 struct packet_type *ptype, *pt_prev;
4992 rx_handler_func_t *rx_handler;
4993 struct net_device *orig_dev;
4994 bool deliver_exact = false;
4995 int ret = NET_RX_DROP;
4998 net_timestamp_check(!netdev_tstamp_prequeue, skb);
5000 trace_netif_receive_skb(skb);
5002 orig_dev = skb->dev;
5004 skb_reset_network_header(skb);
5005 if (!skb_transport_header_was_set(skb))
5006 skb_reset_transport_header(skb);
5007 skb_reset_mac_len(skb);
5012 skb->skb_iif = skb->dev->ifindex;
5014 __this_cpu_inc(softnet_data.processed);
5016 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5020 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5023 if (ret2 != XDP_PASS)
5025 skb_reset_mac_len(skb);
5028 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
5029 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
5030 skb = skb_vlan_untag(skb);
5035 if (skb_skip_tc_classify(skb))
5041 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5043 ret = deliver_skb(skb, pt_prev, orig_dev);
5047 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5049 ret = deliver_skb(skb, pt_prev, orig_dev);
5054 #ifdef CONFIG_NET_INGRESS
5055 if (static_branch_unlikely(&ingress_needed_key)) {
5056 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
5060 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5066 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5069 if (skb_vlan_tag_present(skb)) {
5071 ret = deliver_skb(skb, pt_prev, orig_dev);
5074 if (vlan_do_receive(&skb))
5076 else if (unlikely(!skb))
5080 rx_handler = rcu_dereference(skb->dev->rx_handler);
5083 ret = deliver_skb(skb, pt_prev, orig_dev);
5086 switch (rx_handler(&skb)) {
5087 case RX_HANDLER_CONSUMED:
5088 ret = NET_RX_SUCCESS;
5090 case RX_HANDLER_ANOTHER:
5092 case RX_HANDLER_EXACT:
5093 deliver_exact = true;
5094 case RX_HANDLER_PASS:
5101 if (unlikely(skb_vlan_tag_present(skb))) {
5103 if (skb_vlan_tag_get_id(skb)) {
5104 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5107 skb->pkt_type = PACKET_OTHERHOST;
5108 } else if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
5109 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
5110 /* Outer header is 802.1P with vlan 0, inner header is
5111 * 802.1Q or 802.1AD and vlan_do_receive() above could
5112 * not find vlan dev for vlan id 0.
5114 __vlan_hwaccel_clear_tag(skb);
5115 skb = skb_vlan_untag(skb);
5118 if (vlan_do_receive(&skb))
5119 /* After stripping off 802.1P header with vlan 0
5120 * vlan dev is found for inner header.
5123 else if (unlikely(!skb))
5126 /* We have stripped outer 802.1P vlan 0 header.
5127 * But could not find vlan dev.
5128 * check again for vlan id to set OTHERHOST.
5132 /* Note: we might in the future use prio bits
5133 * and set skb->priority like in vlan_do_receive()
5134 * For the time being, just ignore Priority Code Point
5136 __vlan_hwaccel_clear_tag(skb);
5139 type = skb->protocol;
5141 /* deliver only exact match when indicated */
5142 if (likely(!deliver_exact)) {
5143 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5144 &ptype_base[ntohs(type) &
5148 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5149 &orig_dev->ptype_specific);
5151 if (unlikely(skb->dev != orig_dev)) {
5152 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5153 &skb->dev->ptype_specific);
5157 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5159 *ppt_prev = pt_prev;
5163 atomic_long_inc(&skb->dev->rx_dropped);
5165 atomic_long_inc(&skb->dev->rx_nohandler);
5167 /* Jamal, now you will not able to escape explaining
5168 * me how you were going to use this. :-)
5177 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5179 struct net_device *orig_dev = skb->dev;
5180 struct packet_type *pt_prev = NULL;
5183 ret = __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
5185 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5186 skb->dev, pt_prev, orig_dev);
5191 * netif_receive_skb_core - special purpose version of netif_receive_skb
5192 * @skb: buffer to process
5194 * More direct receive version of netif_receive_skb(). It should
5195 * only be used by callers that have a need to skip RPS and Generic XDP.
5196 * Caller must also take care of handling if (page_is_)pfmemalloc.
5198 * This function may only be called from softirq context and interrupts
5199 * should be enabled.
5201 * Return values (usually ignored):
5202 * NET_RX_SUCCESS: no congestion
5203 * NET_RX_DROP: packet was dropped
5205 int netif_receive_skb_core(struct sk_buff *skb)
5210 ret = __netif_receive_skb_one_core(skb, false);
5215 EXPORT_SYMBOL(netif_receive_skb_core);
5217 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5218 struct packet_type *pt_prev,
5219 struct net_device *orig_dev)
5221 struct sk_buff *skb, *next;
5225 if (list_empty(head))
5227 if (pt_prev->list_func != NULL)
5228 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5229 ip_list_rcv, head, pt_prev, orig_dev);
5231 list_for_each_entry_safe(skb, next, head, list) {
5232 skb_list_del_init(skb);
5233 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5237 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5239 /* Fast-path assumptions:
5240 * - There is no RX handler.
5241 * - Only one packet_type matches.
5242 * If either of these fails, we will end up doing some per-packet
5243 * processing in-line, then handling the 'last ptype' for the whole
5244 * sublist. This can't cause out-of-order delivery to any single ptype,
5245 * because the 'last ptype' must be constant across the sublist, and all
5246 * other ptypes are handled per-packet.
5248 /* Current (common) ptype of sublist */
5249 struct packet_type *pt_curr = NULL;
5250 /* Current (common) orig_dev of sublist */
5251 struct net_device *od_curr = NULL;
5252 struct list_head sublist;
5253 struct sk_buff *skb, *next;
5255 INIT_LIST_HEAD(&sublist);
5256 list_for_each_entry_safe(skb, next, head, list) {
5257 struct net_device *orig_dev = skb->dev;
5258 struct packet_type *pt_prev = NULL;
5260 skb_list_del_init(skb);
5261 __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
5264 if (pt_curr != pt_prev || od_curr != orig_dev) {
5265 /* dispatch old sublist */
5266 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5267 /* start new sublist */
5268 INIT_LIST_HEAD(&sublist);
5272 list_add_tail(&skb->list, &sublist);
5275 /* dispatch final sublist */
5276 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5279 static int __netif_receive_skb(struct sk_buff *skb)
5283 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5284 unsigned int noreclaim_flag;
5287 * PFMEMALLOC skbs are special, they should
5288 * - be delivered to SOCK_MEMALLOC sockets only
5289 * - stay away from userspace
5290 * - have bounded memory usage
5292 * Use PF_MEMALLOC as this saves us from propagating the allocation
5293 * context down to all allocation sites.
5295 noreclaim_flag = memalloc_noreclaim_save();
5296 ret = __netif_receive_skb_one_core(skb, true);
5297 memalloc_noreclaim_restore(noreclaim_flag);
5299 ret = __netif_receive_skb_one_core(skb, false);
5304 static void __netif_receive_skb_list(struct list_head *head)
5306 unsigned long noreclaim_flag = 0;
5307 struct sk_buff *skb, *next;
5308 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5310 list_for_each_entry_safe(skb, next, head, list) {
5311 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5312 struct list_head sublist;
5314 /* Handle the previous sublist */
5315 list_cut_before(&sublist, head, &skb->list);
5316 if (!list_empty(&sublist))
5317 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5318 pfmemalloc = !pfmemalloc;
5319 /* See comments in __netif_receive_skb */
5321 noreclaim_flag = memalloc_noreclaim_save();
5323 memalloc_noreclaim_restore(noreclaim_flag);
5326 /* Handle the remaining sublist */
5327 if (!list_empty(head))
5328 __netif_receive_skb_list_core(head, pfmemalloc);
5329 /* Restore pflags */
5331 memalloc_noreclaim_restore(noreclaim_flag);
5334 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5336 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5337 struct bpf_prog *new = xdp->prog;
5340 switch (xdp->command) {
5341 case XDP_SETUP_PROG:
5342 rcu_assign_pointer(dev->xdp_prog, new);
5347 static_branch_dec(&generic_xdp_needed_key);
5348 } else if (new && !old) {
5349 static_branch_inc(&generic_xdp_needed_key);
5350 dev_disable_lro(dev);
5351 dev_disable_gro_hw(dev);
5355 case XDP_QUERY_PROG:
5356 xdp->prog_id = old ? old->aux->id : 0;
5367 static int netif_receive_skb_internal(struct sk_buff *skb)
5371 net_timestamp_check(netdev_tstamp_prequeue, skb);
5373 if (skb_defer_rx_timestamp(skb))
5374 return NET_RX_SUCCESS;
5378 if (static_branch_unlikely(&rps_needed)) {
5379 struct rps_dev_flow voidflow, *rflow = &voidflow;
5380 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5383 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5389 ret = __netif_receive_skb(skb);
5394 static void netif_receive_skb_list_internal(struct list_head *head)
5396 struct sk_buff *skb, *next;
5397 struct list_head sublist;
5399 INIT_LIST_HEAD(&sublist);
5400 list_for_each_entry_safe(skb, next, head, list) {
5401 net_timestamp_check(netdev_tstamp_prequeue, skb);
5402 skb_list_del_init(skb);
5403 if (!skb_defer_rx_timestamp(skb))
5404 list_add_tail(&skb->list, &sublist);
5406 list_splice_init(&sublist, head);
5410 if (static_branch_unlikely(&rps_needed)) {
5411 list_for_each_entry_safe(skb, next, head, list) {
5412 struct rps_dev_flow voidflow, *rflow = &voidflow;
5413 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5416 /* Will be handled, remove from list */
5417 skb_list_del_init(skb);
5418 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5423 __netif_receive_skb_list(head);
5428 * netif_receive_skb - process receive buffer from network
5429 * @skb: buffer to process
5431 * netif_receive_skb() is the main receive data processing function.
5432 * It always succeeds. The buffer may be dropped during processing
5433 * for congestion control or by the protocol layers.
5435 * This function may only be called from softirq context and interrupts
5436 * should be enabled.
5438 * Return values (usually ignored):
5439 * NET_RX_SUCCESS: no congestion
5440 * NET_RX_DROP: packet was dropped
5442 int netif_receive_skb(struct sk_buff *skb)
5446 trace_netif_receive_skb_entry(skb);
5448 ret = netif_receive_skb_internal(skb);
5449 trace_netif_receive_skb_exit(ret);
5453 EXPORT_SYMBOL(netif_receive_skb);
5456 * netif_receive_skb_list - process many receive buffers from network
5457 * @head: list of skbs to process.
5459 * Since return value of netif_receive_skb() is normally ignored, and
5460 * wouldn't be meaningful for a list, this function returns void.
5462 * This function may only be called from softirq context and interrupts
5463 * should be enabled.
5465 void netif_receive_skb_list(struct list_head *head)
5467 struct sk_buff *skb;
5469 if (list_empty(head))
5471 if (trace_netif_receive_skb_list_entry_enabled()) {
5472 list_for_each_entry(skb, head, list)
5473 trace_netif_receive_skb_list_entry(skb);
5475 netif_receive_skb_list_internal(head);
5476 trace_netif_receive_skb_list_exit(0);
5478 EXPORT_SYMBOL(netif_receive_skb_list);
5480 DEFINE_PER_CPU(struct work_struct, flush_works);
5482 /* Network device is going away, flush any packets still pending */
5483 static void flush_backlog(struct work_struct *work)
5485 struct sk_buff *skb, *tmp;
5486 struct softnet_data *sd;
5489 sd = this_cpu_ptr(&softnet_data);
5491 local_irq_disable();
5493 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5494 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5495 __skb_unlink(skb, &sd->input_pkt_queue);
5497 input_queue_head_incr(sd);
5503 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5504 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5505 __skb_unlink(skb, &sd->process_queue);
5507 input_queue_head_incr(sd);
5513 static void flush_all_backlogs(void)
5519 for_each_online_cpu(cpu)
5520 queue_work_on(cpu, system_highpri_wq,
5521 per_cpu_ptr(&flush_works, cpu));
5523 for_each_online_cpu(cpu)
5524 flush_work(per_cpu_ptr(&flush_works, cpu));
5529 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */
5530 static void gro_normal_list(struct napi_struct *napi)
5532 if (!napi->rx_count)
5534 netif_receive_skb_list_internal(&napi->rx_list);
5535 INIT_LIST_HEAD(&napi->rx_list);
5539 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded,
5540 * pass the whole batch up to the stack.
5542 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb)
5544 list_add_tail(&skb->list, &napi->rx_list);
5545 if (++napi->rx_count >= gro_normal_batch)
5546 gro_normal_list(napi);
5549 INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int));
5550 INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int));
5551 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb)
5553 struct packet_offload *ptype;
5554 __be16 type = skb->protocol;
5555 struct list_head *head = &offload_base;
5558 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5560 if (NAPI_GRO_CB(skb)->count == 1) {
5561 skb_shinfo(skb)->gso_size = 0;
5566 list_for_each_entry_rcu(ptype, head, list) {
5567 if (ptype->type != type || !ptype->callbacks.gro_complete)
5570 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5571 ipv6_gro_complete, inet_gro_complete,
5578 WARN_ON(&ptype->list == head);
5580 return NET_RX_SUCCESS;
5584 gro_normal_one(napi, skb);
5585 return NET_RX_SUCCESS;
5588 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5591 struct list_head *head = &napi->gro_hash[index].list;
5592 struct sk_buff *skb, *p;
5594 list_for_each_entry_safe_reverse(skb, p, head, list) {
5595 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5597 skb_list_del_init(skb);
5598 napi_gro_complete(napi, skb);
5599 napi->gro_hash[index].count--;
5602 if (!napi->gro_hash[index].count)
5603 __clear_bit(index, &napi->gro_bitmask);
5606 /* napi->gro_hash[].list contains packets ordered by age.
5607 * youngest packets at the head of it.
5608 * Complete skbs in reverse order to reduce latencies.
5610 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5612 unsigned long bitmask = napi->gro_bitmask;
5613 unsigned int i, base = ~0U;
5615 while ((i = ffs(bitmask)) != 0) {
5618 __napi_gro_flush_chain(napi, base, flush_old);
5621 EXPORT_SYMBOL(napi_gro_flush);
5623 static struct list_head *gro_list_prepare(struct napi_struct *napi,
5624 struct sk_buff *skb)
5626 unsigned int maclen = skb->dev->hard_header_len;
5627 u32 hash = skb_get_hash_raw(skb);
5628 struct list_head *head;
5631 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list;
5632 list_for_each_entry(p, head, list) {
5633 unsigned long diffs;
5635 NAPI_GRO_CB(p)->flush = 0;
5637 if (hash != skb_get_hash_raw(p)) {
5638 NAPI_GRO_CB(p)->same_flow = 0;
5642 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5643 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5644 if (skb_vlan_tag_present(p))
5645 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb);
5646 diffs |= skb_metadata_dst_cmp(p, skb);
5647 diffs |= skb_metadata_differs(p, skb);
5648 if (maclen == ETH_HLEN)
5649 diffs |= compare_ether_header(skb_mac_header(p),
5650 skb_mac_header(skb));
5652 diffs = memcmp(skb_mac_header(p),
5653 skb_mac_header(skb),
5655 NAPI_GRO_CB(p)->same_flow = !diffs;
5661 static void skb_gro_reset_offset(struct sk_buff *skb)
5663 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5664 const skb_frag_t *frag0 = &pinfo->frags[0];
5666 NAPI_GRO_CB(skb)->data_offset = 0;
5667 NAPI_GRO_CB(skb)->frag0 = NULL;
5668 NAPI_GRO_CB(skb)->frag0_len = 0;
5670 if (!skb_headlen(skb) && pinfo->nr_frags &&
5671 !PageHighMem(skb_frag_page(frag0))) {
5672 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5673 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5674 skb_frag_size(frag0),
5675 skb->end - skb->tail);
5679 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5681 struct skb_shared_info *pinfo = skb_shinfo(skb);
5683 BUG_ON(skb->end - skb->tail < grow);
5685 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5687 skb->data_len -= grow;
5690 skb_frag_off_add(&pinfo->frags[0], grow);
5691 skb_frag_size_sub(&pinfo->frags[0], grow);
5693 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5694 skb_frag_unref(skb, 0);
5695 memmove(pinfo->frags, pinfo->frags + 1,
5696 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
5700 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head)
5702 struct sk_buff *oldest;
5704 oldest = list_last_entry(head, struct sk_buff, list);
5706 /* We are called with head length >= MAX_GRO_SKBS, so this is
5709 if (WARN_ON_ONCE(!oldest))
5712 /* Do not adjust napi->gro_hash[].count, caller is adding a new
5715 skb_list_del_init(oldest);
5716 napi_gro_complete(napi, oldest);
5719 INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *,
5721 INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *,
5723 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5725 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
5726 struct list_head *head = &offload_base;
5727 struct packet_offload *ptype;
5728 __be16 type = skb->protocol;
5729 struct list_head *gro_head;
5730 struct sk_buff *pp = NULL;
5731 enum gro_result ret;
5735 if (netif_elide_gro(skb->dev))
5738 gro_head = gro_list_prepare(napi, skb);
5741 list_for_each_entry_rcu(ptype, head, list) {
5742 if (ptype->type != type || !ptype->callbacks.gro_receive)
5745 skb_set_network_header(skb, skb_gro_offset(skb));
5746 skb_reset_mac_len(skb);
5747 NAPI_GRO_CB(skb)->same_flow = 0;
5748 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
5749 NAPI_GRO_CB(skb)->free = 0;
5750 NAPI_GRO_CB(skb)->encap_mark = 0;
5751 NAPI_GRO_CB(skb)->recursion_counter = 0;
5752 NAPI_GRO_CB(skb)->is_fou = 0;
5753 NAPI_GRO_CB(skb)->is_atomic = 1;
5754 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
5756 /* Setup for GRO checksum validation */
5757 switch (skb->ip_summed) {
5758 case CHECKSUM_COMPLETE:
5759 NAPI_GRO_CB(skb)->csum = skb->csum;
5760 NAPI_GRO_CB(skb)->csum_valid = 1;
5761 NAPI_GRO_CB(skb)->csum_cnt = 0;
5763 case CHECKSUM_UNNECESSARY:
5764 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
5765 NAPI_GRO_CB(skb)->csum_valid = 0;
5768 NAPI_GRO_CB(skb)->csum_cnt = 0;
5769 NAPI_GRO_CB(skb)->csum_valid = 0;
5772 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
5773 ipv6_gro_receive, inet_gro_receive,
5779 if (&ptype->list == head)
5782 if (PTR_ERR(pp) == -EINPROGRESS) {
5787 same_flow = NAPI_GRO_CB(skb)->same_flow;
5788 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5791 skb_list_del_init(pp);
5792 napi_gro_complete(napi, pp);
5793 napi->gro_hash[hash].count--;
5799 if (NAPI_GRO_CB(skb)->flush)
5802 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
5803 gro_flush_oldest(napi, gro_head);
5805 napi->gro_hash[hash].count++;
5807 NAPI_GRO_CB(skb)->count = 1;
5808 NAPI_GRO_CB(skb)->age = jiffies;
5809 NAPI_GRO_CB(skb)->last = skb;
5810 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
5811 list_add(&skb->list, gro_head);
5815 grow = skb_gro_offset(skb) - skb_headlen(skb);
5817 gro_pull_from_frag0(skb, grow);
5819 if (napi->gro_hash[hash].count) {
5820 if (!test_bit(hash, &napi->gro_bitmask))
5821 __set_bit(hash, &napi->gro_bitmask);
5822 } else if (test_bit(hash, &napi->gro_bitmask)) {
5823 __clear_bit(hash, &napi->gro_bitmask);
5833 struct packet_offload *gro_find_receive_by_type(__be16 type)
5835 struct list_head *offload_head = &offload_base;
5836 struct packet_offload *ptype;
5838 list_for_each_entry_rcu(ptype, offload_head, list) {
5839 if (ptype->type != type || !ptype->callbacks.gro_receive)
5845 EXPORT_SYMBOL(gro_find_receive_by_type);
5847 struct packet_offload *gro_find_complete_by_type(__be16 type)
5849 struct list_head *offload_head = &offload_base;
5850 struct packet_offload *ptype;
5852 list_for_each_entry_rcu(ptype, offload_head, list) {
5853 if (ptype->type != type || !ptype->callbacks.gro_complete)
5859 EXPORT_SYMBOL(gro_find_complete_by_type);
5861 static void napi_skb_free_stolen_head(struct sk_buff *skb)
5865 kmem_cache_free(skbuff_head_cache, skb);
5868 static gro_result_t napi_skb_finish(struct napi_struct *napi,
5869 struct sk_buff *skb,
5874 gro_normal_one(napi, skb);
5881 case GRO_MERGED_FREE:
5882 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5883 napi_skb_free_stolen_head(skb);
5897 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5901 skb_mark_napi_id(skb, napi);
5902 trace_napi_gro_receive_entry(skb);
5904 skb_gro_reset_offset(skb);
5906 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
5907 trace_napi_gro_receive_exit(ret);
5911 EXPORT_SYMBOL(napi_gro_receive);
5913 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
5915 if (unlikely(skb->pfmemalloc)) {
5919 __skb_pull(skb, skb_headlen(skb));
5920 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
5921 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
5922 __vlan_hwaccel_clear_tag(skb);
5923 skb->dev = napi->dev;
5926 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
5927 skb->pkt_type = PACKET_HOST;
5929 skb->encapsulation = 0;
5930 skb_shinfo(skb)->gso_type = 0;
5931 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5937 struct sk_buff *napi_get_frags(struct napi_struct *napi)
5939 struct sk_buff *skb = napi->skb;
5942 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
5945 skb_mark_napi_id(skb, napi);
5950 EXPORT_SYMBOL(napi_get_frags);
5952 static gro_result_t napi_frags_finish(struct napi_struct *napi,
5953 struct sk_buff *skb,
5959 __skb_push(skb, ETH_HLEN);
5960 skb->protocol = eth_type_trans(skb, skb->dev);
5961 if (ret == GRO_NORMAL)
5962 gro_normal_one(napi, skb);
5966 napi_reuse_skb(napi, skb);
5969 case GRO_MERGED_FREE:
5970 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5971 napi_skb_free_stolen_head(skb);
5973 napi_reuse_skb(napi, skb);
5984 /* Upper GRO stack assumes network header starts at gro_offset=0
5985 * Drivers could call both napi_gro_frags() and napi_gro_receive()
5986 * We copy ethernet header into skb->data to have a common layout.
5988 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
5990 struct sk_buff *skb = napi->skb;
5991 const struct ethhdr *eth;
5992 unsigned int hlen = sizeof(*eth);
5996 skb_reset_mac_header(skb);
5997 skb_gro_reset_offset(skb);
5999 if (unlikely(skb_gro_header_hard(skb, hlen))) {
6000 eth = skb_gro_header_slow(skb, hlen, 0);
6001 if (unlikely(!eth)) {
6002 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
6003 __func__, napi->dev->name);
6004 napi_reuse_skb(napi, skb);
6008 eth = (const struct ethhdr *)skb->data;
6009 gro_pull_from_frag0(skb, hlen);
6010 NAPI_GRO_CB(skb)->frag0 += hlen;
6011 NAPI_GRO_CB(skb)->frag0_len -= hlen;
6013 __skb_pull(skb, hlen);
6016 * This works because the only protocols we care about don't require
6018 * We'll fix it up properly in napi_frags_finish()
6020 skb->protocol = eth->h_proto;
6025 gro_result_t napi_gro_frags(struct napi_struct *napi)
6028 struct sk_buff *skb = napi_frags_skb(napi);
6033 trace_napi_gro_frags_entry(skb);
6035 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
6036 trace_napi_gro_frags_exit(ret);
6040 EXPORT_SYMBOL(napi_gro_frags);
6042 /* Compute the checksum from gro_offset and return the folded value
6043 * after adding in any pseudo checksum.
6045 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
6050 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
6052 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
6053 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
6054 /* See comments in __skb_checksum_complete(). */
6056 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
6057 !skb->csum_complete_sw)
6058 netdev_rx_csum_fault(skb->dev, skb);
6061 NAPI_GRO_CB(skb)->csum = wsum;
6062 NAPI_GRO_CB(skb)->csum_valid = 1;
6066 EXPORT_SYMBOL(__skb_gro_checksum_complete);
6068 static void net_rps_send_ipi(struct softnet_data *remsd)
6072 struct softnet_data *next = remsd->rps_ipi_next;
6074 if (cpu_online(remsd->cpu))
6075 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6082 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6083 * Note: called with local irq disabled, but exits with local irq enabled.
6085 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6088 struct softnet_data *remsd = sd->rps_ipi_list;
6091 sd->rps_ipi_list = NULL;
6095 /* Send pending IPI's to kick RPS processing on remote cpus. */
6096 net_rps_send_ipi(remsd);
6102 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6105 return sd->rps_ipi_list != NULL;
6111 static int process_backlog(struct napi_struct *napi, int quota)
6113 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6117 /* Check if we have pending ipi, its better to send them now,
6118 * not waiting net_rx_action() end.
6120 if (sd_has_rps_ipi_waiting(sd)) {
6121 local_irq_disable();
6122 net_rps_action_and_irq_enable(sd);
6125 napi->weight = dev_rx_weight;
6127 struct sk_buff *skb;
6129 while ((skb = __skb_dequeue(&sd->process_queue))) {
6131 __netif_receive_skb(skb);
6133 input_queue_head_incr(sd);
6134 if (++work >= quota)
6139 local_irq_disable();
6141 if (skb_queue_empty(&sd->input_pkt_queue)) {
6143 * Inline a custom version of __napi_complete().
6144 * only current cpu owns and manipulates this napi,
6145 * and NAPI_STATE_SCHED is the only possible flag set
6147 * We can use a plain write instead of clear_bit(),
6148 * and we dont need an smp_mb() memory barrier.
6153 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6154 &sd->process_queue);
6164 * __napi_schedule - schedule for receive
6165 * @n: entry to schedule
6167 * The entry's receive function will be scheduled to run.
6168 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6170 void __napi_schedule(struct napi_struct *n)
6172 unsigned long flags;
6174 local_irq_save(flags);
6175 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6176 local_irq_restore(flags);
6178 EXPORT_SYMBOL(__napi_schedule);
6181 * napi_schedule_prep - check if napi can be scheduled
6184 * Test if NAPI routine is already running, and if not mark
6185 * it as running. This is used as a condition variable
6186 * insure only one NAPI poll instance runs. We also make
6187 * sure there is no pending NAPI disable.
6189 bool napi_schedule_prep(struct napi_struct *n)
6191 unsigned long val, new;
6194 val = READ_ONCE(n->state);
6195 if (unlikely(val & NAPIF_STATE_DISABLE))
6197 new = val | NAPIF_STATE_SCHED;
6199 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6200 * This was suggested by Alexander Duyck, as compiler
6201 * emits better code than :
6202 * if (val & NAPIF_STATE_SCHED)
6203 * new |= NAPIF_STATE_MISSED;
6205 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6207 } while (cmpxchg(&n->state, val, new) != val);
6209 return !(val & NAPIF_STATE_SCHED);
6211 EXPORT_SYMBOL(napi_schedule_prep);
6214 * __napi_schedule_irqoff - schedule for receive
6215 * @n: entry to schedule
6217 * Variant of __napi_schedule() assuming hard irqs are masked
6219 void __napi_schedule_irqoff(struct napi_struct *n)
6221 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6223 EXPORT_SYMBOL(__napi_schedule_irqoff);
6225 bool napi_complete_done(struct napi_struct *n, int work_done)
6227 unsigned long flags, val, new;
6230 * 1) Don't let napi dequeue from the cpu poll list
6231 * just in case its running on a different cpu.
6232 * 2) If we are busy polling, do nothing here, we have
6233 * the guarantee we will be called later.
6235 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6236 NAPIF_STATE_IN_BUSY_POLL)))
6239 if (n->gro_bitmask) {
6240 unsigned long timeout = 0;
6243 timeout = n->dev->gro_flush_timeout;
6245 /* When the NAPI instance uses a timeout and keeps postponing
6246 * it, we need to bound somehow the time packets are kept in
6249 napi_gro_flush(n, !!timeout);
6251 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6252 HRTIMER_MODE_REL_PINNED);
6257 if (unlikely(!list_empty(&n->poll_list))) {
6258 /* If n->poll_list is not empty, we need to mask irqs */
6259 local_irq_save(flags);
6260 list_del_init(&n->poll_list);
6261 local_irq_restore(flags);
6265 val = READ_ONCE(n->state);
6267 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6269 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
6271 /* If STATE_MISSED was set, leave STATE_SCHED set,
6272 * because we will call napi->poll() one more time.
6273 * This C code was suggested by Alexander Duyck to help gcc.
6275 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6277 } while (cmpxchg(&n->state, val, new) != val);
6279 if (unlikely(val & NAPIF_STATE_MISSED)) {
6286 EXPORT_SYMBOL(napi_complete_done);
6288 /* must be called under rcu_read_lock(), as we dont take a reference */
6289 static struct napi_struct *napi_by_id(unsigned int napi_id)
6291 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6292 struct napi_struct *napi;
6294 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6295 if (napi->napi_id == napi_id)
6301 #if defined(CONFIG_NET_RX_BUSY_POLL)
6303 #define BUSY_POLL_BUDGET 8
6305 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
6309 /* Busy polling means there is a high chance device driver hard irq
6310 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6311 * set in napi_schedule_prep().
6312 * Since we are about to call napi->poll() once more, we can safely
6313 * clear NAPI_STATE_MISSED.
6315 * Note: x86 could use a single "lock and ..." instruction
6316 * to perform these two clear_bit()
6318 clear_bit(NAPI_STATE_MISSED, &napi->state);
6319 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6323 /* All we really want here is to re-enable device interrupts.
6324 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6326 rc = napi->poll(napi, BUSY_POLL_BUDGET);
6327 /* We can't gro_normal_list() here, because napi->poll() might have
6328 * rearmed the napi (napi_complete_done()) in which case it could
6329 * already be running on another CPU.
6331 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
6332 netpoll_poll_unlock(have_poll_lock);
6333 if (rc == BUSY_POLL_BUDGET) {
6334 /* As the whole budget was spent, we still own the napi so can
6335 * safely handle the rx_list.
6337 gro_normal_list(napi);
6338 __napi_schedule(napi);
6343 void napi_busy_loop(unsigned int napi_id,
6344 bool (*loop_end)(void *, unsigned long),
6347 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6348 int (*napi_poll)(struct napi_struct *napi, int budget);
6349 void *have_poll_lock = NULL;
6350 struct napi_struct *napi;
6357 napi = napi_by_id(napi_id);
6367 unsigned long val = READ_ONCE(napi->state);
6369 /* If multiple threads are competing for this napi,
6370 * we avoid dirtying napi->state as much as we can.
6372 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6373 NAPIF_STATE_IN_BUSY_POLL))
6375 if (cmpxchg(&napi->state, val,
6376 val | NAPIF_STATE_IN_BUSY_POLL |
6377 NAPIF_STATE_SCHED) != val)
6379 have_poll_lock = netpoll_poll_lock(napi);
6380 napi_poll = napi->poll;
6382 work = napi_poll(napi, BUSY_POLL_BUDGET);
6383 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
6384 gro_normal_list(napi);
6387 __NET_ADD_STATS(dev_net(napi->dev),
6388 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6391 if (!loop_end || loop_end(loop_end_arg, start_time))
6394 if (unlikely(need_resched())) {
6396 busy_poll_stop(napi, have_poll_lock);
6400 if (loop_end(loop_end_arg, start_time))
6407 busy_poll_stop(napi, have_poll_lock);
6412 EXPORT_SYMBOL(napi_busy_loop);
6414 #endif /* CONFIG_NET_RX_BUSY_POLL */
6416 static void napi_hash_add(struct napi_struct *napi)
6418 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
6419 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
6422 spin_lock(&napi_hash_lock);
6424 /* 0..NR_CPUS range is reserved for sender_cpu use */
6426 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6427 napi_gen_id = MIN_NAPI_ID;
6428 } while (napi_by_id(napi_gen_id));
6429 napi->napi_id = napi_gen_id;
6431 hlist_add_head_rcu(&napi->napi_hash_node,
6432 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6434 spin_unlock(&napi_hash_lock);
6437 /* Warning : caller is responsible to make sure rcu grace period
6438 * is respected before freeing memory containing @napi
6440 bool napi_hash_del(struct napi_struct *napi)
6442 bool rcu_sync_needed = false;
6444 spin_lock(&napi_hash_lock);
6446 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
6447 rcu_sync_needed = true;
6448 hlist_del_rcu(&napi->napi_hash_node);
6450 spin_unlock(&napi_hash_lock);
6451 return rcu_sync_needed;
6453 EXPORT_SYMBOL_GPL(napi_hash_del);
6455 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6457 struct napi_struct *napi;
6459 napi = container_of(timer, struct napi_struct, timer);
6461 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6462 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6464 if (napi->gro_bitmask && !napi_disable_pending(napi) &&
6465 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
6466 __napi_schedule_irqoff(napi);
6468 return HRTIMER_NORESTART;
6471 static void init_gro_hash(struct napi_struct *napi)
6475 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6476 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6477 napi->gro_hash[i].count = 0;
6479 napi->gro_bitmask = 0;
6482 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6483 int (*poll)(struct napi_struct *, int), int weight)
6485 INIT_LIST_HEAD(&napi->poll_list);
6486 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6487 napi->timer.function = napi_watchdog;
6488 init_gro_hash(napi);
6490 INIT_LIST_HEAD(&napi->rx_list);
6493 if (weight > NAPI_POLL_WEIGHT)
6494 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6496 napi->weight = weight;
6497 list_add(&napi->dev_list, &dev->napi_list);
6499 #ifdef CONFIG_NETPOLL
6500 napi->poll_owner = -1;
6502 set_bit(NAPI_STATE_SCHED, &napi->state);
6503 napi_hash_add(napi);
6505 EXPORT_SYMBOL(netif_napi_add);
6507 void napi_disable(struct napi_struct *n)
6510 set_bit(NAPI_STATE_DISABLE, &n->state);
6512 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6514 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6517 hrtimer_cancel(&n->timer);
6519 clear_bit(NAPI_STATE_DISABLE, &n->state);
6521 EXPORT_SYMBOL(napi_disable);
6523 static void flush_gro_hash(struct napi_struct *napi)
6527 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6528 struct sk_buff *skb, *n;
6530 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6532 napi->gro_hash[i].count = 0;
6536 /* Must be called in process context */
6537 void netif_napi_del(struct napi_struct *napi)
6540 if (napi_hash_del(napi))
6542 list_del_init(&napi->dev_list);
6543 napi_free_frags(napi);
6545 flush_gro_hash(napi);
6546 napi->gro_bitmask = 0;
6548 EXPORT_SYMBOL(netif_napi_del);
6550 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6555 list_del_init(&n->poll_list);
6557 have = netpoll_poll_lock(n);
6561 /* This NAPI_STATE_SCHED test is for avoiding a race
6562 * with netpoll's poll_napi(). Only the entity which
6563 * obtains the lock and sees NAPI_STATE_SCHED set will
6564 * actually make the ->poll() call. Therefore we avoid
6565 * accidentally calling ->poll() when NAPI is not scheduled.
6568 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6569 work = n->poll(n, weight);
6570 trace_napi_poll(n, work, weight);
6573 WARN_ON_ONCE(work > weight);
6575 if (likely(work < weight))
6578 /* Drivers must not modify the NAPI state if they
6579 * consume the entire weight. In such cases this code
6580 * still "owns" the NAPI instance and therefore can
6581 * move the instance around on the list at-will.
6583 if (unlikely(napi_disable_pending(n))) {
6588 if (n->gro_bitmask) {
6589 /* flush too old packets
6590 * If HZ < 1000, flush all packets.
6592 napi_gro_flush(n, HZ >= 1000);
6597 /* Some drivers may have called napi_schedule
6598 * prior to exhausting their budget.
6600 if (unlikely(!list_empty(&n->poll_list))) {
6601 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6602 n->dev ? n->dev->name : "backlog");
6606 list_add_tail(&n->poll_list, repoll);
6609 netpoll_poll_unlock(have);
6614 static __latent_entropy void net_rx_action(struct softirq_action *h)
6616 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6617 unsigned long time_limit = jiffies +
6618 usecs_to_jiffies(netdev_budget_usecs);
6619 int budget = netdev_budget;
6623 local_irq_disable();
6624 list_splice_init(&sd->poll_list, &list);
6628 struct napi_struct *n;
6630 if (list_empty(&list)) {
6631 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6636 n = list_first_entry(&list, struct napi_struct, poll_list);
6637 budget -= napi_poll(n, &repoll);
6639 /* If softirq window is exhausted then punt.
6640 * Allow this to run for 2 jiffies since which will allow
6641 * an average latency of 1.5/HZ.
6643 if (unlikely(budget <= 0 ||
6644 time_after_eq(jiffies, time_limit))) {
6650 local_irq_disable();
6652 list_splice_tail_init(&sd->poll_list, &list);
6653 list_splice_tail(&repoll, &list);
6654 list_splice(&list, &sd->poll_list);
6655 if (!list_empty(&sd->poll_list))
6656 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6658 net_rps_action_and_irq_enable(sd);
6660 __kfree_skb_flush();
6663 struct netdev_adjacent {
6664 struct net_device *dev;
6666 /* upper master flag, there can only be one master device per list */
6669 /* lookup ignore flag */
6672 /* counter for the number of times this device was added to us */
6675 /* private field for the users */
6678 struct list_head list;
6679 struct rcu_head rcu;
6682 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6683 struct list_head *adj_list)
6685 struct netdev_adjacent *adj;
6687 list_for_each_entry(adj, adj_list, list) {
6688 if (adj->dev == adj_dev)
6694 static int ____netdev_has_upper_dev(struct net_device *upper_dev, void *data)
6696 struct net_device *dev = data;
6698 return upper_dev == dev;
6702 * netdev_has_upper_dev - Check if device is linked to an upper device
6704 * @upper_dev: upper device to check
6706 * Find out if a device is linked to specified upper device and return true
6707 * in case it is. Note that this checks only immediate upper device,
6708 * not through a complete stack of devices. The caller must hold the RTNL lock.
6710 bool netdev_has_upper_dev(struct net_device *dev,
6711 struct net_device *upper_dev)
6715 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6718 EXPORT_SYMBOL(netdev_has_upper_dev);
6721 * netdev_has_upper_dev_all - Check if device is linked to an upper device
6723 * @upper_dev: upper device to check
6725 * Find out if a device is linked to specified upper device and return true
6726 * in case it is. Note that this checks the entire upper device chain.
6727 * The caller must hold rcu lock.
6730 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6731 struct net_device *upper_dev)
6733 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6736 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6739 * netdev_has_any_upper_dev - Check if device is linked to some device
6742 * Find out if a device is linked to an upper device and return true in case
6743 * it is. The caller must hold the RTNL lock.
6745 bool netdev_has_any_upper_dev(struct net_device *dev)
6749 return !list_empty(&dev->adj_list.upper);
6751 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6754 * netdev_master_upper_dev_get - Get master upper device
6757 * Find a master upper device and return pointer to it or NULL in case
6758 * it's not there. The caller must hold the RTNL lock.
6760 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6762 struct netdev_adjacent *upper;
6766 if (list_empty(&dev->adj_list.upper))
6769 upper = list_first_entry(&dev->adj_list.upper,
6770 struct netdev_adjacent, list);
6771 if (likely(upper->master))
6775 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6777 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6779 struct netdev_adjacent *upper;
6783 if (list_empty(&dev->adj_list.upper))
6786 upper = list_first_entry(&dev->adj_list.upper,
6787 struct netdev_adjacent, list);
6788 if (likely(upper->master) && !upper->ignore)
6794 * netdev_has_any_lower_dev - Check if device is linked to some device
6797 * Find out if a device is linked to a lower device and return true in case
6798 * it is. The caller must hold the RTNL lock.
6800 static bool netdev_has_any_lower_dev(struct net_device *dev)
6804 return !list_empty(&dev->adj_list.lower);
6807 void *netdev_adjacent_get_private(struct list_head *adj_list)
6809 struct netdev_adjacent *adj;
6811 adj = list_entry(adj_list, struct netdev_adjacent, list);
6813 return adj->private;
6815 EXPORT_SYMBOL(netdev_adjacent_get_private);
6818 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6820 * @iter: list_head ** of the current position
6822 * Gets the next device from the dev's upper list, starting from iter
6823 * position. The caller must hold RCU read lock.
6825 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6826 struct list_head **iter)
6828 struct netdev_adjacent *upper;
6830 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6832 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6834 if (&upper->list == &dev->adj_list.upper)
6837 *iter = &upper->list;
6841 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6843 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6844 struct list_head **iter,
6847 struct netdev_adjacent *upper;
6849 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6851 if (&upper->list == &dev->adj_list.upper)
6854 *iter = &upper->list;
6855 *ignore = upper->ignore;
6860 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6861 struct list_head **iter)
6863 struct netdev_adjacent *upper;
6865 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6867 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6869 if (&upper->list == &dev->adj_list.upper)
6872 *iter = &upper->list;
6877 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6878 int (*fn)(struct net_device *dev,
6882 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6883 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6888 iter = &dev->adj_list.upper;
6892 ret = fn(now, data);
6899 udev = __netdev_next_upper_dev(now, &iter, &ignore);
6906 niter = &udev->adj_list.upper;
6907 dev_stack[cur] = now;
6908 iter_stack[cur++] = iter;
6915 next = dev_stack[--cur];
6916 niter = iter_stack[cur];
6926 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6927 int (*fn)(struct net_device *dev,
6931 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6932 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6936 iter = &dev->adj_list.upper;
6940 ret = fn(now, data);
6947 udev = netdev_next_upper_dev_rcu(now, &iter);
6952 niter = &udev->adj_list.upper;
6953 dev_stack[cur] = now;
6954 iter_stack[cur++] = iter;
6961 next = dev_stack[--cur];
6962 niter = iter_stack[cur];
6971 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6973 static bool __netdev_has_upper_dev(struct net_device *dev,
6974 struct net_device *upper_dev)
6978 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
6983 * netdev_lower_get_next_private - Get the next ->private from the
6984 * lower neighbour list
6986 * @iter: list_head ** of the current position
6988 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6989 * list, starting from iter position. The caller must hold either hold the
6990 * RTNL lock or its own locking that guarantees that the neighbour lower
6991 * list will remain unchanged.
6993 void *netdev_lower_get_next_private(struct net_device *dev,
6994 struct list_head **iter)
6996 struct netdev_adjacent *lower;
6998 lower = list_entry(*iter, struct netdev_adjacent, list);
7000 if (&lower->list == &dev->adj_list.lower)
7003 *iter = lower->list.next;
7005 return lower->private;
7007 EXPORT_SYMBOL(netdev_lower_get_next_private);
7010 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7011 * lower neighbour list, RCU
7014 * @iter: list_head ** of the current position
7016 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7017 * list, starting from iter position. The caller must hold RCU read lock.
7019 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7020 struct list_head **iter)
7022 struct netdev_adjacent *lower;
7024 WARN_ON_ONCE(!rcu_read_lock_held());
7026 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7028 if (&lower->list == &dev->adj_list.lower)
7031 *iter = &lower->list;
7033 return lower->private;
7035 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7038 * netdev_lower_get_next - Get the next device from the lower neighbour
7041 * @iter: list_head ** of the current position
7043 * Gets the next netdev_adjacent from the dev's lower neighbour
7044 * list, starting from iter position. The caller must hold RTNL lock or
7045 * its own locking that guarantees that the neighbour lower
7046 * list will remain unchanged.
7048 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7050 struct netdev_adjacent *lower;
7052 lower = list_entry(*iter, struct netdev_adjacent, list);
7054 if (&lower->list == &dev->adj_list.lower)
7057 *iter = lower->list.next;
7061 EXPORT_SYMBOL(netdev_lower_get_next);
7063 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7064 struct list_head **iter)
7066 struct netdev_adjacent *lower;
7068 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7070 if (&lower->list == &dev->adj_list.lower)
7073 *iter = &lower->list;
7078 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7079 struct list_head **iter,
7082 struct netdev_adjacent *lower;
7084 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7086 if (&lower->list == &dev->adj_list.lower)
7089 *iter = &lower->list;
7090 *ignore = lower->ignore;
7095 int netdev_walk_all_lower_dev(struct net_device *dev,
7096 int (*fn)(struct net_device *dev,
7100 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7101 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7105 iter = &dev->adj_list.lower;
7109 ret = fn(now, data);
7116 ldev = netdev_next_lower_dev(now, &iter);
7121 niter = &ldev->adj_list.lower;
7122 dev_stack[cur] = now;
7123 iter_stack[cur++] = iter;
7130 next = dev_stack[--cur];
7131 niter = iter_stack[cur];
7140 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7142 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7143 int (*fn)(struct net_device *dev,
7147 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7148 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7153 iter = &dev->adj_list.lower;
7157 ret = fn(now, data);
7164 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7171 niter = &ldev->adj_list.lower;
7172 dev_stack[cur] = now;
7173 iter_stack[cur++] = iter;
7180 next = dev_stack[--cur];
7181 niter = iter_stack[cur];
7191 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7192 struct list_head **iter)
7194 struct netdev_adjacent *lower;
7196 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7197 if (&lower->list == &dev->adj_list.lower)
7200 *iter = &lower->list;
7204 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7206 static u8 __netdev_upper_depth(struct net_device *dev)
7208 struct net_device *udev;
7209 struct list_head *iter;
7213 for (iter = &dev->adj_list.upper,
7214 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7216 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7219 if (max_depth < udev->upper_level)
7220 max_depth = udev->upper_level;
7226 static u8 __netdev_lower_depth(struct net_device *dev)
7228 struct net_device *ldev;
7229 struct list_head *iter;
7233 for (iter = &dev->adj_list.lower,
7234 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7236 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7239 if (max_depth < ldev->lower_level)
7240 max_depth = ldev->lower_level;
7246 static int __netdev_update_upper_level(struct net_device *dev, void *data)
7248 dev->upper_level = __netdev_upper_depth(dev) + 1;
7252 static int __netdev_update_lower_level(struct net_device *dev, void *data)
7254 dev->lower_level = __netdev_lower_depth(dev) + 1;
7258 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7259 int (*fn)(struct net_device *dev,
7263 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7264 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7268 iter = &dev->adj_list.lower;
7272 ret = fn(now, data);
7279 ldev = netdev_next_lower_dev_rcu(now, &iter);
7284 niter = &ldev->adj_list.lower;
7285 dev_stack[cur] = now;
7286 iter_stack[cur++] = iter;
7293 next = dev_stack[--cur];
7294 niter = iter_stack[cur];
7303 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7306 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7307 * lower neighbour list, RCU
7311 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7312 * list. The caller must hold RCU read lock.
7314 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7316 struct netdev_adjacent *lower;
7318 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7319 struct netdev_adjacent, list);
7321 return lower->private;
7324 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7327 * netdev_master_upper_dev_get_rcu - Get master upper device
7330 * Find a master upper device and return pointer to it or NULL in case
7331 * it's not there. The caller must hold the RCU read lock.
7333 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7335 struct netdev_adjacent *upper;
7337 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7338 struct netdev_adjacent, list);
7339 if (upper && likely(upper->master))
7343 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7345 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7346 struct net_device *adj_dev,
7347 struct list_head *dev_list)
7349 char linkname[IFNAMSIZ+7];
7351 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7352 "upper_%s" : "lower_%s", adj_dev->name);
7353 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7356 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7358 struct list_head *dev_list)
7360 char linkname[IFNAMSIZ+7];
7362 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7363 "upper_%s" : "lower_%s", name);
7364 sysfs_remove_link(&(dev->dev.kobj), linkname);
7367 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7368 struct net_device *adj_dev,
7369 struct list_head *dev_list)
7371 return (dev_list == &dev->adj_list.upper ||
7372 dev_list == &dev->adj_list.lower) &&
7373 net_eq(dev_net(dev), dev_net(adj_dev));
7376 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7377 struct net_device *adj_dev,
7378 struct list_head *dev_list,
7379 void *private, bool master)
7381 struct netdev_adjacent *adj;
7384 adj = __netdev_find_adj(adj_dev, dev_list);
7388 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7389 dev->name, adj_dev->name, adj->ref_nr);
7394 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7399 adj->master = master;
7401 adj->private = private;
7402 adj->ignore = false;
7405 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7406 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7408 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7409 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7414 /* Ensure that master link is always the first item in list. */
7416 ret = sysfs_create_link(&(dev->dev.kobj),
7417 &(adj_dev->dev.kobj), "master");
7419 goto remove_symlinks;
7421 list_add_rcu(&adj->list, dev_list);
7423 list_add_tail_rcu(&adj->list, dev_list);
7429 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7430 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7438 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7439 struct net_device *adj_dev,
7441 struct list_head *dev_list)
7443 struct netdev_adjacent *adj;
7445 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7446 dev->name, adj_dev->name, ref_nr);
7448 adj = __netdev_find_adj(adj_dev, dev_list);
7451 pr_err("Adjacency does not exist for device %s from %s\n",
7452 dev->name, adj_dev->name);
7457 if (adj->ref_nr > ref_nr) {
7458 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7459 dev->name, adj_dev->name, ref_nr,
7460 adj->ref_nr - ref_nr);
7461 adj->ref_nr -= ref_nr;
7466 sysfs_remove_link(&(dev->dev.kobj), "master");
7468 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7469 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7471 list_del_rcu(&adj->list);
7472 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7473 adj_dev->name, dev->name, adj_dev->name);
7475 kfree_rcu(adj, rcu);
7478 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7479 struct net_device *upper_dev,
7480 struct list_head *up_list,
7481 struct list_head *down_list,
7482 void *private, bool master)
7486 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7491 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7494 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7501 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7502 struct net_device *upper_dev,
7504 struct list_head *up_list,
7505 struct list_head *down_list)
7507 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7508 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7511 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7512 struct net_device *upper_dev,
7513 void *private, bool master)
7515 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7516 &dev->adj_list.upper,
7517 &upper_dev->adj_list.lower,
7521 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7522 struct net_device *upper_dev)
7524 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7525 &dev->adj_list.upper,
7526 &upper_dev->adj_list.lower);
7529 static int __netdev_upper_dev_link(struct net_device *dev,
7530 struct net_device *upper_dev, bool master,
7531 void *upper_priv, void *upper_info,
7532 struct netlink_ext_ack *extack)
7534 struct netdev_notifier_changeupper_info changeupper_info = {
7539 .upper_dev = upper_dev,
7542 .upper_info = upper_info,
7544 struct net_device *master_dev;
7549 if (dev == upper_dev)
7552 /* To prevent loops, check if dev is not upper device to upper_dev. */
7553 if (__netdev_has_upper_dev(upper_dev, dev))
7556 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7560 if (__netdev_has_upper_dev(dev, upper_dev))
7563 master_dev = __netdev_master_upper_dev_get(dev);
7565 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7568 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7569 &changeupper_info.info);
7570 ret = notifier_to_errno(ret);
7574 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7579 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7580 &changeupper_info.info);
7581 ret = notifier_to_errno(ret);
7585 __netdev_update_upper_level(dev, NULL);
7586 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7588 __netdev_update_lower_level(upper_dev, NULL);
7589 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7595 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7601 * netdev_upper_dev_link - Add a link to the upper device
7603 * @upper_dev: new upper device
7604 * @extack: netlink extended ack
7606 * Adds a link to device which is upper to this one. The caller must hold
7607 * the RTNL lock. On a failure a negative errno code is returned.
7608 * On success the reference counts are adjusted and the function
7611 int netdev_upper_dev_link(struct net_device *dev,
7612 struct net_device *upper_dev,
7613 struct netlink_ext_ack *extack)
7615 return __netdev_upper_dev_link(dev, upper_dev, false,
7616 NULL, NULL, extack);
7618 EXPORT_SYMBOL(netdev_upper_dev_link);
7621 * netdev_master_upper_dev_link - Add a master link to the upper device
7623 * @upper_dev: new upper device
7624 * @upper_priv: upper device private
7625 * @upper_info: upper info to be passed down via notifier
7626 * @extack: netlink extended ack
7628 * Adds a link to device which is upper to this one. In this case, only
7629 * one master upper device can be linked, although other non-master devices
7630 * might be linked as well. The caller must hold the RTNL lock.
7631 * On a failure a negative errno code is returned. On success the reference
7632 * counts are adjusted and the function returns zero.
7634 int netdev_master_upper_dev_link(struct net_device *dev,
7635 struct net_device *upper_dev,
7636 void *upper_priv, void *upper_info,
7637 struct netlink_ext_ack *extack)
7639 return __netdev_upper_dev_link(dev, upper_dev, true,
7640 upper_priv, upper_info, extack);
7642 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7645 * netdev_upper_dev_unlink - Removes a link to upper device
7647 * @upper_dev: new upper device
7649 * Removes a link to device which is upper to this one. The caller must hold
7652 void netdev_upper_dev_unlink(struct net_device *dev,
7653 struct net_device *upper_dev)
7655 struct netdev_notifier_changeupper_info changeupper_info = {
7659 .upper_dev = upper_dev,
7665 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7667 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7668 &changeupper_info.info);
7670 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7672 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7673 &changeupper_info.info);
7675 __netdev_update_upper_level(dev, NULL);
7676 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7678 __netdev_update_lower_level(upper_dev, NULL);
7679 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7682 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7684 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7685 struct net_device *lower_dev,
7688 struct netdev_adjacent *adj;
7690 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7694 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7699 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7700 struct net_device *lower_dev)
7702 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7705 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7706 struct net_device *lower_dev)
7708 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7711 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7712 struct net_device *new_dev,
7713 struct net_device *dev,
7714 struct netlink_ext_ack *extack)
7721 if (old_dev && new_dev != old_dev)
7722 netdev_adjacent_dev_disable(dev, old_dev);
7724 err = netdev_upper_dev_link(new_dev, dev, extack);
7726 if (old_dev && new_dev != old_dev)
7727 netdev_adjacent_dev_enable(dev, old_dev);
7733 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7735 void netdev_adjacent_change_commit(struct net_device *old_dev,
7736 struct net_device *new_dev,
7737 struct net_device *dev)
7739 if (!new_dev || !old_dev)
7742 if (new_dev == old_dev)
7745 netdev_adjacent_dev_enable(dev, old_dev);
7746 netdev_upper_dev_unlink(old_dev, dev);
7748 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7750 void netdev_adjacent_change_abort(struct net_device *old_dev,
7751 struct net_device *new_dev,
7752 struct net_device *dev)
7757 if (old_dev && new_dev != old_dev)
7758 netdev_adjacent_dev_enable(dev, old_dev);
7760 netdev_upper_dev_unlink(new_dev, dev);
7762 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7765 * netdev_bonding_info_change - Dispatch event about slave change
7767 * @bonding_info: info to dispatch
7769 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7770 * The caller must hold the RTNL lock.
7772 void netdev_bonding_info_change(struct net_device *dev,
7773 struct netdev_bonding_info *bonding_info)
7775 struct netdev_notifier_bonding_info info = {
7779 memcpy(&info.bonding_info, bonding_info,
7780 sizeof(struct netdev_bonding_info));
7781 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7784 EXPORT_SYMBOL(netdev_bonding_info_change);
7786 static void netdev_adjacent_add_links(struct net_device *dev)
7788 struct netdev_adjacent *iter;
7790 struct net *net = dev_net(dev);
7792 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7793 if (!net_eq(net, dev_net(iter->dev)))
7795 netdev_adjacent_sysfs_add(iter->dev, dev,
7796 &iter->dev->adj_list.lower);
7797 netdev_adjacent_sysfs_add(dev, iter->dev,
7798 &dev->adj_list.upper);
7801 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7802 if (!net_eq(net, dev_net(iter->dev)))
7804 netdev_adjacent_sysfs_add(iter->dev, dev,
7805 &iter->dev->adj_list.upper);
7806 netdev_adjacent_sysfs_add(dev, iter->dev,
7807 &dev->adj_list.lower);
7811 static void netdev_adjacent_del_links(struct net_device *dev)
7813 struct netdev_adjacent *iter;
7815 struct net *net = dev_net(dev);
7817 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7818 if (!net_eq(net, dev_net(iter->dev)))
7820 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7821 &iter->dev->adj_list.lower);
7822 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7823 &dev->adj_list.upper);
7826 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7827 if (!net_eq(net, dev_net(iter->dev)))
7829 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7830 &iter->dev->adj_list.upper);
7831 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7832 &dev->adj_list.lower);
7836 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
7838 struct netdev_adjacent *iter;
7840 struct net *net = dev_net(dev);
7842 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7843 if (!net_eq(net, dev_net(iter->dev)))
7845 netdev_adjacent_sysfs_del(iter->dev, oldname,
7846 &iter->dev->adj_list.lower);
7847 netdev_adjacent_sysfs_add(iter->dev, dev,
7848 &iter->dev->adj_list.lower);
7851 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7852 if (!net_eq(net, dev_net(iter->dev)))
7854 netdev_adjacent_sysfs_del(iter->dev, oldname,
7855 &iter->dev->adj_list.upper);
7856 netdev_adjacent_sysfs_add(iter->dev, dev,
7857 &iter->dev->adj_list.upper);
7861 void *netdev_lower_dev_get_private(struct net_device *dev,
7862 struct net_device *lower_dev)
7864 struct netdev_adjacent *lower;
7868 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
7872 return lower->private;
7874 EXPORT_SYMBOL(netdev_lower_dev_get_private);
7878 * netdev_lower_change - Dispatch event about lower device state change
7879 * @lower_dev: device
7880 * @lower_state_info: state to dispatch
7882 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
7883 * The caller must hold the RTNL lock.
7885 void netdev_lower_state_changed(struct net_device *lower_dev,
7886 void *lower_state_info)
7888 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
7889 .info.dev = lower_dev,
7893 changelowerstate_info.lower_state_info = lower_state_info;
7894 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
7895 &changelowerstate_info.info);
7897 EXPORT_SYMBOL(netdev_lower_state_changed);
7899 static void dev_change_rx_flags(struct net_device *dev, int flags)
7901 const struct net_device_ops *ops = dev->netdev_ops;
7903 if (ops->ndo_change_rx_flags)
7904 ops->ndo_change_rx_flags(dev, flags);
7907 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
7909 unsigned int old_flags = dev->flags;
7915 dev->flags |= IFF_PROMISC;
7916 dev->promiscuity += inc;
7917 if (dev->promiscuity == 0) {
7920 * If inc causes overflow, untouch promisc and return error.
7923 dev->flags &= ~IFF_PROMISC;
7925 dev->promiscuity -= inc;
7926 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
7931 if (dev->flags != old_flags) {
7932 pr_info("device %s %s promiscuous mode\n",
7934 dev->flags & IFF_PROMISC ? "entered" : "left");
7935 if (audit_enabled) {
7936 current_uid_gid(&uid, &gid);
7937 audit_log(audit_context(), GFP_ATOMIC,
7938 AUDIT_ANOM_PROMISCUOUS,
7939 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
7940 dev->name, (dev->flags & IFF_PROMISC),
7941 (old_flags & IFF_PROMISC),
7942 from_kuid(&init_user_ns, audit_get_loginuid(current)),
7943 from_kuid(&init_user_ns, uid),
7944 from_kgid(&init_user_ns, gid),
7945 audit_get_sessionid(current));
7948 dev_change_rx_flags(dev, IFF_PROMISC);
7951 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
7956 * dev_set_promiscuity - update promiscuity count on a device
7960 * Add or remove promiscuity from a device. While the count in the device
7961 * remains above zero the interface remains promiscuous. Once it hits zero
7962 * the device reverts back to normal filtering operation. A negative inc
7963 * value is used to drop promiscuity on the device.
7964 * Return 0 if successful or a negative errno code on error.
7966 int dev_set_promiscuity(struct net_device *dev, int inc)
7968 unsigned int old_flags = dev->flags;
7971 err = __dev_set_promiscuity(dev, inc, true);
7974 if (dev->flags != old_flags)
7975 dev_set_rx_mode(dev);
7978 EXPORT_SYMBOL(dev_set_promiscuity);
7980 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
7982 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
7986 dev->flags |= IFF_ALLMULTI;
7987 dev->allmulti += inc;
7988 if (dev->allmulti == 0) {
7991 * If inc causes overflow, untouch allmulti and return error.
7994 dev->flags &= ~IFF_ALLMULTI;
7996 dev->allmulti -= inc;
7997 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
8002 if (dev->flags ^ old_flags) {
8003 dev_change_rx_flags(dev, IFF_ALLMULTI);
8004 dev_set_rx_mode(dev);
8006 __dev_notify_flags(dev, old_flags,
8007 dev->gflags ^ old_gflags);
8013 * dev_set_allmulti - update allmulti count on a device
8017 * Add or remove reception of all multicast frames to a device. While the
8018 * count in the device remains above zero the interface remains listening
8019 * to all interfaces. Once it hits zero the device reverts back to normal
8020 * filtering operation. A negative @inc value is used to drop the counter
8021 * when releasing a resource needing all multicasts.
8022 * Return 0 if successful or a negative errno code on error.
8025 int dev_set_allmulti(struct net_device *dev, int inc)
8027 return __dev_set_allmulti(dev, inc, true);
8029 EXPORT_SYMBOL(dev_set_allmulti);
8032 * Upload unicast and multicast address lists to device and
8033 * configure RX filtering. When the device doesn't support unicast
8034 * filtering it is put in promiscuous mode while unicast addresses
8037 void __dev_set_rx_mode(struct net_device *dev)
8039 const struct net_device_ops *ops = dev->netdev_ops;
8041 /* dev_open will call this function so the list will stay sane. */
8042 if (!(dev->flags&IFF_UP))
8045 if (!netif_device_present(dev))
8048 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8049 /* Unicast addresses changes may only happen under the rtnl,
8050 * therefore calling __dev_set_promiscuity here is safe.
8052 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8053 __dev_set_promiscuity(dev, 1, false);
8054 dev->uc_promisc = true;
8055 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8056 __dev_set_promiscuity(dev, -1, false);
8057 dev->uc_promisc = false;
8061 if (ops->ndo_set_rx_mode)
8062 ops->ndo_set_rx_mode(dev);
8065 void dev_set_rx_mode(struct net_device *dev)
8067 netif_addr_lock_bh(dev);
8068 __dev_set_rx_mode(dev);
8069 netif_addr_unlock_bh(dev);
8073 * dev_get_flags - get flags reported to userspace
8076 * Get the combination of flag bits exported through APIs to userspace.
8078 unsigned int dev_get_flags(const struct net_device *dev)
8082 flags = (dev->flags & ~(IFF_PROMISC |
8087 (dev->gflags & (IFF_PROMISC |
8090 if (netif_running(dev)) {
8091 if (netif_oper_up(dev))
8092 flags |= IFF_RUNNING;
8093 if (netif_carrier_ok(dev))
8094 flags |= IFF_LOWER_UP;
8095 if (netif_dormant(dev))
8096 flags |= IFF_DORMANT;
8101 EXPORT_SYMBOL(dev_get_flags);
8103 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8104 struct netlink_ext_ack *extack)
8106 unsigned int old_flags = dev->flags;
8112 * Set the flags on our device.
8115 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8116 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8118 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8122 * Load in the correct multicast list now the flags have changed.
8125 if ((old_flags ^ flags) & IFF_MULTICAST)
8126 dev_change_rx_flags(dev, IFF_MULTICAST);
8128 dev_set_rx_mode(dev);
8131 * Have we downed the interface. We handle IFF_UP ourselves
8132 * according to user attempts to set it, rather than blindly
8137 if ((old_flags ^ flags) & IFF_UP) {
8138 if (old_flags & IFF_UP)
8141 ret = __dev_open(dev, extack);
8144 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8145 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8146 unsigned int old_flags = dev->flags;
8148 dev->gflags ^= IFF_PROMISC;
8150 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8151 if (dev->flags != old_flags)
8152 dev_set_rx_mode(dev);
8155 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8156 * is important. Some (broken) drivers set IFF_PROMISC, when
8157 * IFF_ALLMULTI is requested not asking us and not reporting.
8159 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8160 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8162 dev->gflags ^= IFF_ALLMULTI;
8163 __dev_set_allmulti(dev, inc, false);
8169 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8170 unsigned int gchanges)
8172 unsigned int changes = dev->flags ^ old_flags;
8175 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8177 if (changes & IFF_UP) {
8178 if (dev->flags & IFF_UP)
8179 call_netdevice_notifiers(NETDEV_UP, dev);
8181 call_netdevice_notifiers(NETDEV_DOWN, dev);
8184 if (dev->flags & IFF_UP &&
8185 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8186 struct netdev_notifier_change_info change_info = {
8190 .flags_changed = changes,
8193 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8198 * dev_change_flags - change device settings
8200 * @flags: device state flags
8201 * @extack: netlink extended ack
8203 * Change settings on device based state flags. The flags are
8204 * in the userspace exported format.
8206 int dev_change_flags(struct net_device *dev, unsigned int flags,
8207 struct netlink_ext_ack *extack)
8210 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8212 ret = __dev_change_flags(dev, flags, extack);
8216 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8217 __dev_notify_flags(dev, old_flags, changes);
8220 EXPORT_SYMBOL(dev_change_flags);
8222 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8224 const struct net_device_ops *ops = dev->netdev_ops;
8226 if (ops->ndo_change_mtu)
8227 return ops->ndo_change_mtu(dev, new_mtu);
8229 /* Pairs with all the lockless reads of dev->mtu in the stack */
8230 WRITE_ONCE(dev->mtu, new_mtu);
8233 EXPORT_SYMBOL(__dev_set_mtu);
8235 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8236 struct netlink_ext_ack *extack)
8238 /* MTU must be positive, and in range */
8239 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8240 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8244 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8245 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8252 * dev_set_mtu_ext - Change maximum transfer unit
8254 * @new_mtu: new transfer unit
8255 * @extack: netlink extended ack
8257 * Change the maximum transfer size of the network device.
8259 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8260 struct netlink_ext_ack *extack)
8264 if (new_mtu == dev->mtu)
8267 err = dev_validate_mtu(dev, new_mtu, extack);
8271 if (!netif_device_present(dev))
8274 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8275 err = notifier_to_errno(err);
8279 orig_mtu = dev->mtu;
8280 err = __dev_set_mtu(dev, new_mtu);
8283 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8285 err = notifier_to_errno(err);
8287 /* setting mtu back and notifying everyone again,
8288 * so that they have a chance to revert changes.
8290 __dev_set_mtu(dev, orig_mtu);
8291 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8298 int dev_set_mtu(struct net_device *dev, int new_mtu)
8300 struct netlink_ext_ack extack;
8303 memset(&extack, 0, sizeof(extack));
8304 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8305 if (err && extack._msg)
8306 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8309 EXPORT_SYMBOL(dev_set_mtu);
8312 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8314 * @new_len: new tx queue length
8316 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8318 unsigned int orig_len = dev->tx_queue_len;
8321 if (new_len != (unsigned int)new_len)
8324 if (new_len != orig_len) {
8325 dev->tx_queue_len = new_len;
8326 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8327 res = notifier_to_errno(res);
8330 res = dev_qdisc_change_tx_queue_len(dev);
8338 netdev_err(dev, "refused to change device tx_queue_len\n");
8339 dev->tx_queue_len = orig_len;
8344 * dev_set_group - Change group this device belongs to
8346 * @new_group: group this device should belong to
8348 void dev_set_group(struct net_device *dev, int new_group)
8350 dev->group = new_group;
8352 EXPORT_SYMBOL(dev_set_group);
8355 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8357 * @addr: new address
8358 * @extack: netlink extended ack
8360 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8361 struct netlink_ext_ack *extack)
8363 struct netdev_notifier_pre_changeaddr_info info = {
8365 .info.extack = extack,
8370 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8371 return notifier_to_errno(rc);
8373 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8376 * dev_set_mac_address - Change Media Access Control Address
8379 * @extack: netlink extended ack
8381 * Change the hardware (MAC) address of the device
8383 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8384 struct netlink_ext_ack *extack)
8386 const struct net_device_ops *ops = dev->netdev_ops;
8389 if (!ops->ndo_set_mac_address)
8391 if (sa->sa_family != dev->type)
8393 if (!netif_device_present(dev))
8395 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8398 err = ops->ndo_set_mac_address(dev, sa);
8401 dev->addr_assign_type = NET_ADDR_SET;
8402 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8403 add_device_randomness(dev->dev_addr, dev->addr_len);
8406 EXPORT_SYMBOL(dev_set_mac_address);
8409 * dev_change_carrier - Change device carrier
8411 * @new_carrier: new value
8413 * Change device carrier
8415 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8417 const struct net_device_ops *ops = dev->netdev_ops;
8419 if (!ops->ndo_change_carrier)
8421 if (!netif_device_present(dev))
8423 return ops->ndo_change_carrier(dev, new_carrier);
8425 EXPORT_SYMBOL(dev_change_carrier);
8428 * dev_get_phys_port_id - Get device physical port ID
8432 * Get device physical port ID
8434 int dev_get_phys_port_id(struct net_device *dev,
8435 struct netdev_phys_item_id *ppid)
8437 const struct net_device_ops *ops = dev->netdev_ops;
8439 if (!ops->ndo_get_phys_port_id)
8441 return ops->ndo_get_phys_port_id(dev, ppid);
8443 EXPORT_SYMBOL(dev_get_phys_port_id);
8446 * dev_get_phys_port_name - Get device physical port name
8449 * @len: limit of bytes to copy to name
8451 * Get device physical port name
8453 int dev_get_phys_port_name(struct net_device *dev,
8454 char *name, size_t len)
8456 const struct net_device_ops *ops = dev->netdev_ops;
8459 if (ops->ndo_get_phys_port_name) {
8460 err = ops->ndo_get_phys_port_name(dev, name, len);
8461 if (err != -EOPNOTSUPP)
8464 return devlink_compat_phys_port_name_get(dev, name, len);
8466 EXPORT_SYMBOL(dev_get_phys_port_name);
8469 * dev_get_port_parent_id - Get the device's port parent identifier
8470 * @dev: network device
8471 * @ppid: pointer to a storage for the port's parent identifier
8472 * @recurse: allow/disallow recursion to lower devices
8474 * Get the devices's port parent identifier
8476 int dev_get_port_parent_id(struct net_device *dev,
8477 struct netdev_phys_item_id *ppid,
8480 const struct net_device_ops *ops = dev->netdev_ops;
8481 struct netdev_phys_item_id first = { };
8482 struct net_device *lower_dev;
8483 struct list_head *iter;
8486 if (ops->ndo_get_port_parent_id) {
8487 err = ops->ndo_get_port_parent_id(dev, ppid);
8488 if (err != -EOPNOTSUPP)
8492 err = devlink_compat_switch_id_get(dev, ppid);
8493 if (!err || err != -EOPNOTSUPP)
8499 netdev_for_each_lower_dev(dev, lower_dev, iter) {
8500 err = dev_get_port_parent_id(lower_dev, ppid, recurse);
8505 else if (memcmp(&first, ppid, sizeof(*ppid)))
8511 EXPORT_SYMBOL(dev_get_port_parent_id);
8514 * netdev_port_same_parent_id - Indicate if two network devices have
8515 * the same port parent identifier
8516 * @a: first network device
8517 * @b: second network device
8519 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8521 struct netdev_phys_item_id a_id = { };
8522 struct netdev_phys_item_id b_id = { };
8524 if (dev_get_port_parent_id(a, &a_id, true) ||
8525 dev_get_port_parent_id(b, &b_id, true))
8528 return netdev_phys_item_id_same(&a_id, &b_id);
8530 EXPORT_SYMBOL(netdev_port_same_parent_id);
8533 * dev_change_proto_down - update protocol port state information
8535 * @proto_down: new value
8537 * This info can be used by switch drivers to set the phys state of the
8540 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8542 const struct net_device_ops *ops = dev->netdev_ops;
8544 if (!ops->ndo_change_proto_down)
8546 if (!netif_device_present(dev))
8548 return ops->ndo_change_proto_down(dev, proto_down);
8550 EXPORT_SYMBOL(dev_change_proto_down);
8553 * dev_change_proto_down_generic - generic implementation for
8554 * ndo_change_proto_down that sets carrier according to
8558 * @proto_down: new value
8560 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
8563 netif_carrier_off(dev);
8565 netif_carrier_on(dev);
8566 dev->proto_down = proto_down;
8569 EXPORT_SYMBOL(dev_change_proto_down_generic);
8571 u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op,
8572 enum bpf_netdev_command cmd)
8574 struct netdev_bpf xdp;
8579 memset(&xdp, 0, sizeof(xdp));
8582 /* Query must always succeed. */
8583 WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG);
8588 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op,
8589 struct netlink_ext_ack *extack, u32 flags,
8590 struct bpf_prog *prog)
8592 bool non_hw = !(flags & XDP_FLAGS_HW_MODE);
8593 struct bpf_prog *prev_prog = NULL;
8594 struct netdev_bpf xdp;
8598 prev_prog = bpf_prog_by_id(__dev_xdp_query(dev, bpf_op,
8600 if (IS_ERR(prev_prog))
8604 memset(&xdp, 0, sizeof(xdp));
8605 if (flags & XDP_FLAGS_HW_MODE)
8606 xdp.command = XDP_SETUP_PROG_HW;
8608 xdp.command = XDP_SETUP_PROG;
8609 xdp.extack = extack;
8613 err = bpf_op(dev, &xdp);
8615 bpf_prog_change_xdp(prev_prog, prog);
8618 bpf_prog_put(prev_prog);
8623 static void dev_xdp_uninstall(struct net_device *dev)
8625 struct netdev_bpf xdp;
8628 /* Remove generic XDP */
8629 WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL));
8631 /* Remove from the driver */
8632 ndo_bpf = dev->netdev_ops->ndo_bpf;
8636 memset(&xdp, 0, sizeof(xdp));
8637 xdp.command = XDP_QUERY_PROG;
8638 WARN_ON(ndo_bpf(dev, &xdp));
8640 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8643 /* Remove HW offload */
8644 memset(&xdp, 0, sizeof(xdp));
8645 xdp.command = XDP_QUERY_PROG_HW;
8646 if (!ndo_bpf(dev, &xdp) && xdp.prog_id)
8647 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8652 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
8654 * @extack: netlink extended ack
8655 * @fd: new program fd or negative value to clear
8656 * @flags: xdp-related flags
8658 * Set or clear a bpf program for a device
8660 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
8663 const struct net_device_ops *ops = dev->netdev_ops;
8664 enum bpf_netdev_command query;
8665 struct bpf_prog *prog = NULL;
8666 bpf_op_t bpf_op, bpf_chk;
8672 offload = flags & XDP_FLAGS_HW_MODE;
8673 query = offload ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG;
8675 bpf_op = bpf_chk = ops->ndo_bpf;
8676 if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) {
8677 NL_SET_ERR_MSG(extack, "underlying driver does not support XDP in native mode");
8680 if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE))
8681 bpf_op = generic_xdp_install;
8682 if (bpf_op == bpf_chk)
8683 bpf_chk = generic_xdp_install;
8688 if (!offload && __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG)) {
8689 NL_SET_ERR_MSG(extack, "native and generic XDP can't be active at the same time");
8693 prog_id = __dev_xdp_query(dev, bpf_op, query);
8694 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && prog_id) {
8695 NL_SET_ERR_MSG(extack, "XDP program already attached");
8699 prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
8700 bpf_op == ops->ndo_bpf);
8702 return PTR_ERR(prog);
8704 if (!offload && bpf_prog_is_dev_bound(prog->aux)) {
8705 NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported");
8710 /* prog->aux->id may be 0 for orphaned device-bound progs */
8711 if (prog->aux->id && prog->aux->id == prog_id) {
8716 if (!__dev_xdp_query(dev, bpf_op, query))
8720 err = dev_xdp_install(dev, bpf_op, extack, flags, prog);
8721 if (err < 0 && prog)
8728 * dev_new_index - allocate an ifindex
8729 * @net: the applicable net namespace
8731 * Returns a suitable unique value for a new device interface
8732 * number. The caller must hold the rtnl semaphore or the
8733 * dev_base_lock to be sure it remains unique.
8735 static int dev_new_index(struct net *net)
8737 int ifindex = net->ifindex;
8742 if (!__dev_get_by_index(net, ifindex))
8743 return net->ifindex = ifindex;
8747 /* Delayed registration/unregisteration */
8748 static LIST_HEAD(net_todo_list);
8749 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
8751 static void net_set_todo(struct net_device *dev)
8753 list_add_tail(&dev->todo_list, &net_todo_list);
8754 dev_net(dev)->dev_unreg_count++;
8757 static void rollback_registered_many(struct list_head *head)
8759 struct net_device *dev, *tmp;
8760 LIST_HEAD(close_head);
8762 BUG_ON(dev_boot_phase);
8765 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
8766 /* Some devices call without registering
8767 * for initialization unwind. Remove those
8768 * devices and proceed with the remaining.
8770 if (dev->reg_state == NETREG_UNINITIALIZED) {
8771 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
8775 list_del(&dev->unreg_list);
8778 dev->dismantle = true;
8779 BUG_ON(dev->reg_state != NETREG_REGISTERED);
8782 /* If device is running, close it first. */
8783 list_for_each_entry(dev, head, unreg_list)
8784 list_add_tail(&dev->close_list, &close_head);
8785 dev_close_many(&close_head, true);
8787 list_for_each_entry(dev, head, unreg_list) {
8788 /* And unlink it from device chain. */
8789 unlist_netdevice(dev);
8791 dev->reg_state = NETREG_UNREGISTERING;
8793 flush_all_backlogs();
8797 list_for_each_entry(dev, head, unreg_list) {
8798 struct sk_buff *skb = NULL;
8800 /* Shutdown queueing discipline. */
8803 dev_xdp_uninstall(dev);
8805 /* Notify protocols, that we are about to destroy
8806 * this device. They should clean all the things.
8808 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8810 if (!dev->rtnl_link_ops ||
8811 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8812 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
8813 GFP_KERNEL, NULL, 0);
8816 * Flush the unicast and multicast chains
8821 netdev_name_node_alt_flush(dev);
8822 netdev_name_node_free(dev->name_node);
8824 if (dev->netdev_ops->ndo_uninit)
8825 dev->netdev_ops->ndo_uninit(dev);
8828 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
8830 /* Notifier chain MUST detach us all upper devices. */
8831 WARN_ON(netdev_has_any_upper_dev(dev));
8832 WARN_ON(netdev_has_any_lower_dev(dev));
8834 /* Remove entries from kobject tree */
8835 netdev_unregister_kobject(dev);
8837 /* Remove XPS queueing entries */
8838 netif_reset_xps_queues_gt(dev, 0);
8844 list_for_each_entry(dev, head, unreg_list)
8848 static void rollback_registered(struct net_device *dev)
8852 list_add(&dev->unreg_list, &single);
8853 rollback_registered_many(&single);
8857 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
8858 struct net_device *upper, netdev_features_t features)
8860 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8861 netdev_features_t feature;
8864 for_each_netdev_feature(upper_disables, feature_bit) {
8865 feature = __NETIF_F_BIT(feature_bit);
8866 if (!(upper->wanted_features & feature)
8867 && (features & feature)) {
8868 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
8869 &feature, upper->name);
8870 features &= ~feature;
8877 static void netdev_sync_lower_features(struct net_device *upper,
8878 struct net_device *lower, netdev_features_t features)
8880 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8881 netdev_features_t feature;
8884 for_each_netdev_feature(upper_disables, feature_bit) {
8885 feature = __NETIF_F_BIT(feature_bit);
8886 if (!(features & feature) && (lower->features & feature)) {
8887 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
8888 &feature, lower->name);
8889 lower->wanted_features &= ~feature;
8890 netdev_update_features(lower);
8892 if (unlikely(lower->features & feature))
8893 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
8894 &feature, lower->name);
8899 static netdev_features_t netdev_fix_features(struct net_device *dev,
8900 netdev_features_t features)
8902 /* Fix illegal checksum combinations */
8903 if ((features & NETIF_F_HW_CSUM) &&
8904 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
8905 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
8906 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
8909 /* TSO requires that SG is present as well. */
8910 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
8911 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
8912 features &= ~NETIF_F_ALL_TSO;
8915 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
8916 !(features & NETIF_F_IP_CSUM)) {
8917 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
8918 features &= ~NETIF_F_TSO;
8919 features &= ~NETIF_F_TSO_ECN;
8922 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
8923 !(features & NETIF_F_IPV6_CSUM)) {
8924 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
8925 features &= ~NETIF_F_TSO6;
8928 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
8929 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
8930 features &= ~NETIF_F_TSO_MANGLEID;
8932 /* TSO ECN requires that TSO is present as well. */
8933 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
8934 features &= ~NETIF_F_TSO_ECN;
8936 /* Software GSO depends on SG. */
8937 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
8938 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
8939 features &= ~NETIF_F_GSO;
8942 /* GSO partial features require GSO partial be set */
8943 if ((features & dev->gso_partial_features) &&
8944 !(features & NETIF_F_GSO_PARTIAL)) {
8946 "Dropping partially supported GSO features since no GSO partial.\n");
8947 features &= ~dev->gso_partial_features;
8950 if (!(features & NETIF_F_RXCSUM)) {
8951 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
8952 * successfully merged by hardware must also have the
8953 * checksum verified by hardware. If the user does not
8954 * want to enable RXCSUM, logically, we should disable GRO_HW.
8956 if (features & NETIF_F_GRO_HW) {
8957 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
8958 features &= ~NETIF_F_GRO_HW;
8962 /* LRO/HW-GRO features cannot be combined with RX-FCS */
8963 if (features & NETIF_F_RXFCS) {
8964 if (features & NETIF_F_LRO) {
8965 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
8966 features &= ~NETIF_F_LRO;
8969 if (features & NETIF_F_GRO_HW) {
8970 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
8971 features &= ~NETIF_F_GRO_HW;
8978 int __netdev_update_features(struct net_device *dev)
8980 struct net_device *upper, *lower;
8981 netdev_features_t features;
8982 struct list_head *iter;
8987 features = netdev_get_wanted_features(dev);
8989 if (dev->netdev_ops->ndo_fix_features)
8990 features = dev->netdev_ops->ndo_fix_features(dev, features);
8992 /* driver might be less strict about feature dependencies */
8993 features = netdev_fix_features(dev, features);
8995 /* some features can't be enabled if they're off an an upper device */
8996 netdev_for_each_upper_dev_rcu(dev, upper, iter)
8997 features = netdev_sync_upper_features(dev, upper, features);
8999 if (dev->features == features)
9002 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9003 &dev->features, &features);
9005 if (dev->netdev_ops->ndo_set_features)
9006 err = dev->netdev_ops->ndo_set_features(dev, features);
9010 if (unlikely(err < 0)) {
9012 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9013 err, &features, &dev->features);
9014 /* return non-0 since some features might have changed and
9015 * it's better to fire a spurious notification than miss it
9021 /* some features must be disabled on lower devices when disabled
9022 * on an upper device (think: bonding master or bridge)
9024 netdev_for_each_lower_dev(dev, lower, iter)
9025 netdev_sync_lower_features(dev, lower, features);
9028 netdev_features_t diff = features ^ dev->features;
9030 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9031 /* udp_tunnel_{get,drop}_rx_info both need
9032 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9033 * device, or they won't do anything.
9034 * Thus we need to update dev->features
9035 * *before* calling udp_tunnel_get_rx_info,
9036 * but *after* calling udp_tunnel_drop_rx_info.
9038 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9039 dev->features = features;
9040 udp_tunnel_get_rx_info(dev);
9042 udp_tunnel_drop_rx_info(dev);
9046 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9047 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9048 dev->features = features;
9049 err |= vlan_get_rx_ctag_filter_info(dev);
9051 vlan_drop_rx_ctag_filter_info(dev);
9055 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9056 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9057 dev->features = features;
9058 err |= vlan_get_rx_stag_filter_info(dev);
9060 vlan_drop_rx_stag_filter_info(dev);
9064 dev->features = features;
9067 return err < 0 ? 0 : 1;
9071 * netdev_update_features - recalculate device features
9072 * @dev: the device to check
9074 * Recalculate dev->features set and send notifications if it
9075 * has changed. Should be called after driver or hardware dependent
9076 * conditions might have changed that influence the features.
9078 void netdev_update_features(struct net_device *dev)
9080 if (__netdev_update_features(dev))
9081 netdev_features_change(dev);
9083 EXPORT_SYMBOL(netdev_update_features);
9086 * netdev_change_features - recalculate device features
9087 * @dev: the device to check
9089 * Recalculate dev->features set and send notifications even
9090 * if they have not changed. Should be called instead of
9091 * netdev_update_features() if also dev->vlan_features might
9092 * have changed to allow the changes to be propagated to stacked
9095 void netdev_change_features(struct net_device *dev)
9097 __netdev_update_features(dev);
9098 netdev_features_change(dev);
9100 EXPORT_SYMBOL(netdev_change_features);
9103 * netif_stacked_transfer_operstate - transfer operstate
9104 * @rootdev: the root or lower level device to transfer state from
9105 * @dev: the device to transfer operstate to
9107 * Transfer operational state from root to device. This is normally
9108 * called when a stacking relationship exists between the root
9109 * device and the device(a leaf device).
9111 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9112 struct net_device *dev)
9114 if (rootdev->operstate == IF_OPER_DORMANT)
9115 netif_dormant_on(dev);
9117 netif_dormant_off(dev);
9119 if (netif_carrier_ok(rootdev))
9120 netif_carrier_on(dev);
9122 netif_carrier_off(dev);
9124 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9126 static int netif_alloc_rx_queues(struct net_device *dev)
9128 unsigned int i, count = dev->num_rx_queues;
9129 struct netdev_rx_queue *rx;
9130 size_t sz = count * sizeof(*rx);
9135 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9141 for (i = 0; i < count; i++) {
9144 /* XDP RX-queue setup */
9145 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i);
9152 /* Rollback successful reg's and free other resources */
9154 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9160 static void netif_free_rx_queues(struct net_device *dev)
9162 unsigned int i, count = dev->num_rx_queues;
9164 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9168 for (i = 0; i < count; i++)
9169 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9174 static void netdev_init_one_queue(struct net_device *dev,
9175 struct netdev_queue *queue, void *_unused)
9177 /* Initialize queue lock */
9178 spin_lock_init(&queue->_xmit_lock);
9179 lockdep_set_class(&queue->_xmit_lock, &dev->qdisc_xmit_lock_key);
9180 queue->xmit_lock_owner = -1;
9181 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9184 dql_init(&queue->dql, HZ);
9188 static void netif_free_tx_queues(struct net_device *dev)
9193 static int netif_alloc_netdev_queues(struct net_device *dev)
9195 unsigned int count = dev->num_tx_queues;
9196 struct netdev_queue *tx;
9197 size_t sz = count * sizeof(*tx);
9199 if (count < 1 || count > 0xffff)
9202 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9208 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9209 spin_lock_init(&dev->tx_global_lock);
9214 void netif_tx_stop_all_queues(struct net_device *dev)
9218 for (i = 0; i < dev->num_tx_queues; i++) {
9219 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9221 netif_tx_stop_queue(txq);
9224 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9226 static void netdev_register_lockdep_key(struct net_device *dev)
9228 lockdep_register_key(&dev->qdisc_tx_busylock_key);
9229 lockdep_register_key(&dev->qdisc_running_key);
9230 lockdep_register_key(&dev->qdisc_xmit_lock_key);
9231 lockdep_register_key(&dev->addr_list_lock_key);
9234 static void netdev_unregister_lockdep_key(struct net_device *dev)
9236 lockdep_unregister_key(&dev->qdisc_tx_busylock_key);
9237 lockdep_unregister_key(&dev->qdisc_running_key);
9238 lockdep_unregister_key(&dev->qdisc_xmit_lock_key);
9239 lockdep_unregister_key(&dev->addr_list_lock_key);
9242 void netdev_update_lockdep_key(struct net_device *dev)
9244 lockdep_unregister_key(&dev->addr_list_lock_key);
9245 lockdep_register_key(&dev->addr_list_lock_key);
9247 lockdep_set_class(&dev->addr_list_lock, &dev->addr_list_lock_key);
9249 EXPORT_SYMBOL(netdev_update_lockdep_key);
9252 * register_netdevice - register a network device
9253 * @dev: device to register
9255 * Take a completed network device structure and add it to the kernel
9256 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9257 * chain. 0 is returned on success. A negative errno code is returned
9258 * on a failure to set up the device, or if the name is a duplicate.
9260 * Callers must hold the rtnl semaphore. You may want
9261 * register_netdev() instead of this.
9264 * The locking appears insufficient to guarantee two parallel registers
9265 * will not get the same name.
9268 int register_netdevice(struct net_device *dev)
9271 struct net *net = dev_net(dev);
9273 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9274 NETDEV_FEATURE_COUNT);
9275 BUG_ON(dev_boot_phase);
9280 /* When net_device's are persistent, this will be fatal. */
9281 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9284 spin_lock_init(&dev->addr_list_lock);
9285 lockdep_set_class(&dev->addr_list_lock, &dev->addr_list_lock_key);
9287 ret = dev_get_valid_name(net, dev, dev->name);
9292 dev->name_node = netdev_name_node_head_alloc(dev);
9293 if (!dev->name_node)
9296 /* Init, if this function is available */
9297 if (dev->netdev_ops->ndo_init) {
9298 ret = dev->netdev_ops->ndo_init(dev);
9306 if (((dev->hw_features | dev->features) &
9307 NETIF_F_HW_VLAN_CTAG_FILTER) &&
9308 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
9309 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
9310 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
9317 dev->ifindex = dev_new_index(net);
9318 else if (__dev_get_by_index(net, dev->ifindex))
9321 /* Transfer changeable features to wanted_features and enable
9322 * software offloads (GSO and GRO).
9324 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
9325 dev->features |= NETIF_F_SOFT_FEATURES;
9327 if (dev->netdev_ops->ndo_udp_tunnel_add) {
9328 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9329 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9332 dev->wanted_features = dev->features & dev->hw_features;
9334 if (!(dev->flags & IFF_LOOPBACK))
9335 dev->hw_features |= NETIF_F_NOCACHE_COPY;
9337 /* If IPv4 TCP segmentation offload is supported we should also
9338 * allow the device to enable segmenting the frame with the option
9339 * of ignoring a static IP ID value. This doesn't enable the
9340 * feature itself but allows the user to enable it later.
9342 if (dev->hw_features & NETIF_F_TSO)
9343 dev->hw_features |= NETIF_F_TSO_MANGLEID;
9344 if (dev->vlan_features & NETIF_F_TSO)
9345 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
9346 if (dev->mpls_features & NETIF_F_TSO)
9347 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
9348 if (dev->hw_enc_features & NETIF_F_TSO)
9349 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
9351 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
9353 dev->vlan_features |= NETIF_F_HIGHDMA;
9355 /* Make NETIF_F_SG inheritable to tunnel devices.
9357 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
9359 /* Make NETIF_F_SG inheritable to MPLS.
9361 dev->mpls_features |= NETIF_F_SG;
9363 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
9364 ret = notifier_to_errno(ret);
9368 ret = netdev_register_kobject(dev);
9370 dev->reg_state = NETREG_UNREGISTERED;
9373 dev->reg_state = NETREG_REGISTERED;
9375 __netdev_update_features(dev);
9378 * Default initial state at registry is that the
9379 * device is present.
9382 set_bit(__LINK_STATE_PRESENT, &dev->state);
9384 linkwatch_init_dev(dev);
9386 dev_init_scheduler(dev);
9388 list_netdevice(dev);
9389 add_device_randomness(dev->dev_addr, dev->addr_len);
9391 /* If the device has permanent device address, driver should
9392 * set dev_addr and also addr_assign_type should be set to
9393 * NET_ADDR_PERM (default value).
9395 if (dev->addr_assign_type == NET_ADDR_PERM)
9396 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
9398 /* Notify protocols, that a new device appeared. */
9399 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
9400 ret = notifier_to_errno(ret);
9402 rollback_registered(dev);
9405 dev->reg_state = NETREG_UNREGISTERED;
9408 * Prevent userspace races by waiting until the network
9409 * device is fully setup before sending notifications.
9411 if (!dev->rtnl_link_ops ||
9412 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
9413 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
9419 if (dev->netdev_ops->ndo_uninit)
9420 dev->netdev_ops->ndo_uninit(dev);
9421 if (dev->priv_destructor)
9422 dev->priv_destructor(dev);
9424 netdev_name_node_free(dev->name_node);
9427 EXPORT_SYMBOL(register_netdevice);
9430 * init_dummy_netdev - init a dummy network device for NAPI
9431 * @dev: device to init
9433 * This takes a network device structure and initialize the minimum
9434 * amount of fields so it can be used to schedule NAPI polls without
9435 * registering a full blown interface. This is to be used by drivers
9436 * that need to tie several hardware interfaces to a single NAPI
9437 * poll scheduler due to HW limitations.
9439 int init_dummy_netdev(struct net_device *dev)
9441 /* Clear everything. Note we don't initialize spinlocks
9442 * are they aren't supposed to be taken by any of the
9443 * NAPI code and this dummy netdev is supposed to be
9444 * only ever used for NAPI polls
9446 memset(dev, 0, sizeof(struct net_device));
9448 /* make sure we BUG if trying to hit standard
9449 * register/unregister code path
9451 dev->reg_state = NETREG_DUMMY;
9453 /* NAPI wants this */
9454 INIT_LIST_HEAD(&dev->napi_list);
9456 /* a dummy interface is started by default */
9457 set_bit(__LINK_STATE_PRESENT, &dev->state);
9458 set_bit(__LINK_STATE_START, &dev->state);
9460 /* napi_busy_loop stats accounting wants this */
9461 dev_net_set(dev, &init_net);
9463 /* Note : We dont allocate pcpu_refcnt for dummy devices,
9464 * because users of this 'device' dont need to change
9470 EXPORT_SYMBOL_GPL(init_dummy_netdev);
9474 * register_netdev - register a network device
9475 * @dev: device to register
9477 * Take a completed network device structure and add it to the kernel
9478 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9479 * chain. 0 is returned on success. A negative errno code is returned
9480 * on a failure to set up the device, or if the name is a duplicate.
9482 * This is a wrapper around register_netdevice that takes the rtnl semaphore
9483 * and expands the device name if you passed a format string to
9486 int register_netdev(struct net_device *dev)
9490 if (rtnl_lock_killable())
9492 err = register_netdevice(dev);
9496 EXPORT_SYMBOL(register_netdev);
9498 int netdev_refcnt_read(const struct net_device *dev)
9502 for_each_possible_cpu(i)
9503 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
9506 EXPORT_SYMBOL(netdev_refcnt_read);
9509 * netdev_wait_allrefs - wait until all references are gone.
9510 * @dev: target net_device
9512 * This is called when unregistering network devices.
9514 * Any protocol or device that holds a reference should register
9515 * for netdevice notification, and cleanup and put back the
9516 * reference if they receive an UNREGISTER event.
9517 * We can get stuck here if buggy protocols don't correctly
9520 static void netdev_wait_allrefs(struct net_device *dev)
9522 unsigned long rebroadcast_time, warning_time;
9525 linkwatch_forget_dev(dev);
9527 rebroadcast_time = warning_time = jiffies;
9528 refcnt = netdev_refcnt_read(dev);
9530 while (refcnt != 0) {
9531 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
9534 /* Rebroadcast unregister notification */
9535 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
9541 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
9543 /* We must not have linkwatch events
9544 * pending on unregister. If this
9545 * happens, we simply run the queue
9546 * unscheduled, resulting in a noop
9549 linkwatch_run_queue();
9554 rebroadcast_time = jiffies;
9559 refcnt = netdev_refcnt_read(dev);
9561 if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) {
9562 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
9564 warning_time = jiffies;
9573 * register_netdevice(x1);
9574 * register_netdevice(x2);
9576 * unregister_netdevice(y1);
9577 * unregister_netdevice(y2);
9583 * We are invoked by rtnl_unlock().
9584 * This allows us to deal with problems:
9585 * 1) We can delete sysfs objects which invoke hotplug
9586 * without deadlocking with linkwatch via keventd.
9587 * 2) Since we run with the RTNL semaphore not held, we can sleep
9588 * safely in order to wait for the netdev refcnt to drop to zero.
9590 * We must not return until all unregister events added during
9591 * the interval the lock was held have been completed.
9593 void netdev_run_todo(void)
9595 struct list_head list;
9597 /* Snapshot list, allow later requests */
9598 list_replace_init(&net_todo_list, &list);
9603 /* Wait for rcu callbacks to finish before next phase */
9604 if (!list_empty(&list))
9607 while (!list_empty(&list)) {
9608 struct net_device *dev
9609 = list_first_entry(&list, struct net_device, todo_list);
9610 list_del(&dev->todo_list);
9612 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
9613 pr_err("network todo '%s' but state %d\n",
9614 dev->name, dev->reg_state);
9619 dev->reg_state = NETREG_UNREGISTERED;
9621 netdev_wait_allrefs(dev);
9624 BUG_ON(netdev_refcnt_read(dev));
9625 BUG_ON(!list_empty(&dev->ptype_all));
9626 BUG_ON(!list_empty(&dev->ptype_specific));
9627 WARN_ON(rcu_access_pointer(dev->ip_ptr));
9628 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
9629 #if IS_ENABLED(CONFIG_DECNET)
9630 WARN_ON(dev->dn_ptr);
9632 if (dev->priv_destructor)
9633 dev->priv_destructor(dev);
9634 if (dev->needs_free_netdev)
9637 /* Report a network device has been unregistered */
9639 dev_net(dev)->dev_unreg_count--;
9641 wake_up(&netdev_unregistering_wq);
9643 /* Free network device */
9644 kobject_put(&dev->dev.kobj);
9648 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
9649 * all the same fields in the same order as net_device_stats, with only
9650 * the type differing, but rtnl_link_stats64 may have additional fields
9651 * at the end for newer counters.
9653 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
9654 const struct net_device_stats *netdev_stats)
9656 #if BITS_PER_LONG == 64
9657 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
9658 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
9659 /* zero out counters that only exist in rtnl_link_stats64 */
9660 memset((char *)stats64 + sizeof(*netdev_stats), 0,
9661 sizeof(*stats64) - sizeof(*netdev_stats));
9663 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
9664 const unsigned long *src = (const unsigned long *)netdev_stats;
9665 u64 *dst = (u64 *)stats64;
9667 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
9668 for (i = 0; i < n; i++)
9670 /* zero out counters that only exist in rtnl_link_stats64 */
9671 memset((char *)stats64 + n * sizeof(u64), 0,
9672 sizeof(*stats64) - n * sizeof(u64));
9675 EXPORT_SYMBOL(netdev_stats_to_stats64);
9678 * dev_get_stats - get network device statistics
9679 * @dev: device to get statistics from
9680 * @storage: place to store stats
9682 * Get network statistics from device. Return @storage.
9683 * The device driver may provide its own method by setting
9684 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
9685 * otherwise the internal statistics structure is used.
9687 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
9688 struct rtnl_link_stats64 *storage)
9690 const struct net_device_ops *ops = dev->netdev_ops;
9692 if (ops->ndo_get_stats64) {
9693 memset(storage, 0, sizeof(*storage));
9694 ops->ndo_get_stats64(dev, storage);
9695 } else if (ops->ndo_get_stats) {
9696 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
9698 netdev_stats_to_stats64(storage, &dev->stats);
9700 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
9701 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
9702 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
9705 EXPORT_SYMBOL(dev_get_stats);
9707 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
9709 struct netdev_queue *queue = dev_ingress_queue(dev);
9711 #ifdef CONFIG_NET_CLS_ACT
9714 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
9717 netdev_init_one_queue(dev, queue, NULL);
9718 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
9719 queue->qdisc_sleeping = &noop_qdisc;
9720 rcu_assign_pointer(dev->ingress_queue, queue);
9725 static const struct ethtool_ops default_ethtool_ops;
9727 void netdev_set_default_ethtool_ops(struct net_device *dev,
9728 const struct ethtool_ops *ops)
9730 if (dev->ethtool_ops == &default_ethtool_ops)
9731 dev->ethtool_ops = ops;
9733 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
9735 void netdev_freemem(struct net_device *dev)
9737 char *addr = (char *)dev - dev->padded;
9743 * alloc_netdev_mqs - allocate network device
9744 * @sizeof_priv: size of private data to allocate space for
9745 * @name: device name format string
9746 * @name_assign_type: origin of device name
9747 * @setup: callback to initialize device
9748 * @txqs: the number of TX subqueues to allocate
9749 * @rxqs: the number of RX subqueues to allocate
9751 * Allocates a struct net_device with private data area for driver use
9752 * and performs basic initialization. Also allocates subqueue structs
9753 * for each queue on the device.
9755 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
9756 unsigned char name_assign_type,
9757 void (*setup)(struct net_device *),
9758 unsigned int txqs, unsigned int rxqs)
9760 struct net_device *dev;
9761 unsigned int alloc_size;
9762 struct net_device *p;
9764 BUG_ON(strlen(name) >= sizeof(dev->name));
9767 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
9772 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
9776 alloc_size = sizeof(struct net_device);
9778 /* ensure 32-byte alignment of private area */
9779 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
9780 alloc_size += sizeof_priv;
9782 /* ensure 32-byte alignment of whole construct */
9783 alloc_size += NETDEV_ALIGN - 1;
9785 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9789 dev = PTR_ALIGN(p, NETDEV_ALIGN);
9790 dev->padded = (char *)dev - (char *)p;
9792 dev->pcpu_refcnt = alloc_percpu(int);
9793 if (!dev->pcpu_refcnt)
9796 if (dev_addr_init(dev))
9802 dev_net_set(dev, &init_net);
9804 netdev_register_lockdep_key(dev);
9806 dev->gso_max_size = GSO_MAX_SIZE;
9807 dev->gso_max_segs = GSO_MAX_SEGS;
9808 dev->upper_level = 1;
9809 dev->lower_level = 1;
9811 INIT_LIST_HEAD(&dev->napi_list);
9812 INIT_LIST_HEAD(&dev->unreg_list);
9813 INIT_LIST_HEAD(&dev->close_list);
9814 INIT_LIST_HEAD(&dev->link_watch_list);
9815 INIT_LIST_HEAD(&dev->adj_list.upper);
9816 INIT_LIST_HEAD(&dev->adj_list.lower);
9817 INIT_LIST_HEAD(&dev->ptype_all);
9818 INIT_LIST_HEAD(&dev->ptype_specific);
9819 INIT_LIST_HEAD(&dev->net_notifier_list);
9820 #ifdef CONFIG_NET_SCHED
9821 hash_init(dev->qdisc_hash);
9823 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
9826 if (!dev->tx_queue_len) {
9827 dev->priv_flags |= IFF_NO_QUEUE;
9828 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
9831 dev->num_tx_queues = txqs;
9832 dev->real_num_tx_queues = txqs;
9833 if (netif_alloc_netdev_queues(dev))
9836 dev->num_rx_queues = rxqs;
9837 dev->real_num_rx_queues = rxqs;
9838 if (netif_alloc_rx_queues(dev))
9841 strcpy(dev->name, name);
9842 dev->name_assign_type = name_assign_type;
9843 dev->group = INIT_NETDEV_GROUP;
9844 if (!dev->ethtool_ops)
9845 dev->ethtool_ops = &default_ethtool_ops;
9847 nf_hook_ingress_init(dev);
9856 free_percpu(dev->pcpu_refcnt);
9858 netdev_freemem(dev);
9861 EXPORT_SYMBOL(alloc_netdev_mqs);
9864 * free_netdev - free network device
9867 * This function does the last stage of destroying an allocated device
9868 * interface. The reference to the device object is released. If this
9869 * is the last reference then it will be freed.Must be called in process
9872 void free_netdev(struct net_device *dev)
9874 struct napi_struct *p, *n;
9877 netif_free_tx_queues(dev);
9878 netif_free_rx_queues(dev);
9880 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
9882 /* Flush device addresses */
9883 dev_addr_flush(dev);
9885 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
9888 free_percpu(dev->pcpu_refcnt);
9889 dev->pcpu_refcnt = NULL;
9890 free_percpu(dev->xdp_bulkq);
9891 dev->xdp_bulkq = NULL;
9893 netdev_unregister_lockdep_key(dev);
9895 /* Compatibility with error handling in drivers */
9896 if (dev->reg_state == NETREG_UNINITIALIZED) {
9897 netdev_freemem(dev);
9901 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
9902 dev->reg_state = NETREG_RELEASED;
9904 /* will free via device release */
9905 put_device(&dev->dev);
9907 EXPORT_SYMBOL(free_netdev);
9910 * synchronize_net - Synchronize with packet receive processing
9912 * Wait for packets currently being received to be done.
9913 * Does not block later packets from starting.
9915 void synchronize_net(void)
9918 if (rtnl_is_locked())
9919 synchronize_rcu_expedited();
9923 EXPORT_SYMBOL(synchronize_net);
9926 * unregister_netdevice_queue - remove device from the kernel
9930 * This function shuts down a device interface and removes it
9931 * from the kernel tables.
9932 * If head not NULL, device is queued to be unregistered later.
9934 * Callers must hold the rtnl semaphore. You may want
9935 * unregister_netdev() instead of this.
9938 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
9943 list_move_tail(&dev->unreg_list, head);
9945 rollback_registered(dev);
9946 /* Finish processing unregister after unlock */
9950 EXPORT_SYMBOL(unregister_netdevice_queue);
9953 * unregister_netdevice_many - unregister many devices
9954 * @head: list of devices
9956 * Note: As most callers use a stack allocated list_head,
9957 * we force a list_del() to make sure stack wont be corrupted later.
9959 void unregister_netdevice_many(struct list_head *head)
9961 struct net_device *dev;
9963 if (!list_empty(head)) {
9964 rollback_registered_many(head);
9965 list_for_each_entry(dev, head, unreg_list)
9970 EXPORT_SYMBOL(unregister_netdevice_many);
9973 * unregister_netdev - remove device from the kernel
9976 * This function shuts down a device interface and removes it
9977 * from the kernel tables.
9979 * This is just a wrapper for unregister_netdevice that takes
9980 * the rtnl semaphore. In general you want to use this and not
9981 * unregister_netdevice.
9983 void unregister_netdev(struct net_device *dev)
9986 unregister_netdevice(dev);
9989 EXPORT_SYMBOL(unregister_netdev);
9992 * dev_change_net_namespace - move device to different nethost namespace
9994 * @net: network namespace
9995 * @pat: If not NULL name pattern to try if the current device name
9996 * is already taken in the destination network namespace.
9998 * This function shuts down a device interface and moves it
9999 * to a new network namespace. On success 0 is returned, on
10000 * a failure a netagive errno code is returned.
10002 * Callers must hold the rtnl semaphore.
10005 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
10007 int err, new_nsid, new_ifindex;
10011 /* Don't allow namespace local devices to be moved. */
10013 if (dev->features & NETIF_F_NETNS_LOCAL)
10016 /* Ensure the device has been registrered */
10017 if (dev->reg_state != NETREG_REGISTERED)
10020 /* Get out if there is nothing todo */
10022 if (net_eq(dev_net(dev), net))
10025 /* Pick the destination device name, and ensure
10026 * we can use it in the destination network namespace.
10029 if (__dev_get_by_name(net, dev->name)) {
10030 /* We get here if we can't use the current device name */
10033 err = dev_get_valid_name(net, dev, pat);
10039 * And now a mini version of register_netdevice unregister_netdevice.
10042 /* If device is running close it first. */
10045 /* And unlink it from device chain */
10046 unlist_netdevice(dev);
10050 /* Shutdown queueing discipline. */
10053 /* Notify protocols, that we are about to destroy
10054 * this device. They should clean all the things.
10056 * Note that dev->reg_state stays at NETREG_REGISTERED.
10057 * This is wanted because this way 8021q and macvlan know
10058 * the device is just moving and can keep their slaves up.
10060 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10063 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
10064 /* If there is an ifindex conflict assign a new one */
10065 if (__dev_get_by_index(net, dev->ifindex))
10066 new_ifindex = dev_new_index(net);
10068 new_ifindex = dev->ifindex;
10070 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
10074 * Flush the unicast and multicast chains
10079 /* Send a netdev-removed uevent to the old namespace */
10080 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
10081 netdev_adjacent_del_links(dev);
10083 /* Move per-net netdevice notifiers that are following the netdevice */
10084 move_netdevice_notifiers_dev_net(dev, net);
10086 /* Actually switch the network namespace */
10087 dev_net_set(dev, net);
10088 dev->ifindex = new_ifindex;
10090 /* Send a netdev-add uevent to the new namespace */
10091 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
10092 netdev_adjacent_add_links(dev);
10094 /* Fixup kobjects */
10095 err = device_rename(&dev->dev, dev->name);
10098 /* Add the device back in the hashes */
10099 list_netdevice(dev);
10101 /* Notify protocols, that a new device appeared. */
10102 call_netdevice_notifiers(NETDEV_REGISTER, dev);
10105 * Prevent userspace races by waiting until the network
10106 * device is fully setup before sending notifications.
10108 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10115 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
10117 static int dev_cpu_dead(unsigned int oldcpu)
10119 struct sk_buff **list_skb;
10120 struct sk_buff *skb;
10122 struct softnet_data *sd, *oldsd, *remsd = NULL;
10124 local_irq_disable();
10125 cpu = smp_processor_id();
10126 sd = &per_cpu(softnet_data, cpu);
10127 oldsd = &per_cpu(softnet_data, oldcpu);
10129 /* Find end of our completion_queue. */
10130 list_skb = &sd->completion_queue;
10132 list_skb = &(*list_skb)->next;
10133 /* Append completion queue from offline CPU. */
10134 *list_skb = oldsd->completion_queue;
10135 oldsd->completion_queue = NULL;
10137 /* Append output queue from offline CPU. */
10138 if (oldsd->output_queue) {
10139 *sd->output_queue_tailp = oldsd->output_queue;
10140 sd->output_queue_tailp = oldsd->output_queue_tailp;
10141 oldsd->output_queue = NULL;
10142 oldsd->output_queue_tailp = &oldsd->output_queue;
10144 /* Append NAPI poll list from offline CPU, with one exception :
10145 * process_backlog() must be called by cpu owning percpu backlog.
10146 * We properly handle process_queue & input_pkt_queue later.
10148 while (!list_empty(&oldsd->poll_list)) {
10149 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
10150 struct napi_struct,
10153 list_del_init(&napi->poll_list);
10154 if (napi->poll == process_backlog)
10157 ____napi_schedule(sd, napi);
10160 raise_softirq_irqoff(NET_TX_SOFTIRQ);
10161 local_irq_enable();
10164 remsd = oldsd->rps_ipi_list;
10165 oldsd->rps_ipi_list = NULL;
10167 /* send out pending IPI's on offline CPU */
10168 net_rps_send_ipi(remsd);
10170 /* Process offline CPU's input_pkt_queue */
10171 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
10173 input_queue_head_incr(oldsd);
10175 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
10177 input_queue_head_incr(oldsd);
10184 * netdev_increment_features - increment feature set by one
10185 * @all: current feature set
10186 * @one: new feature set
10187 * @mask: mask feature set
10189 * Computes a new feature set after adding a device with feature set
10190 * @one to the master device with current feature set @all. Will not
10191 * enable anything that is off in @mask. Returns the new feature set.
10193 netdev_features_t netdev_increment_features(netdev_features_t all,
10194 netdev_features_t one, netdev_features_t mask)
10196 if (mask & NETIF_F_HW_CSUM)
10197 mask |= NETIF_F_CSUM_MASK;
10198 mask |= NETIF_F_VLAN_CHALLENGED;
10200 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
10201 all &= one | ~NETIF_F_ALL_FOR_ALL;
10203 /* If one device supports hw checksumming, set for all. */
10204 if (all & NETIF_F_HW_CSUM)
10205 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
10209 EXPORT_SYMBOL(netdev_increment_features);
10211 static struct hlist_head * __net_init netdev_create_hash(void)
10214 struct hlist_head *hash;
10216 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
10218 for (i = 0; i < NETDEV_HASHENTRIES; i++)
10219 INIT_HLIST_HEAD(&hash[i]);
10224 /* Initialize per network namespace state */
10225 static int __net_init netdev_init(struct net *net)
10227 BUILD_BUG_ON(GRO_HASH_BUCKETS >
10228 8 * sizeof_field(struct napi_struct, gro_bitmask));
10230 if (net != &init_net)
10231 INIT_LIST_HEAD(&net->dev_base_head);
10233 net->dev_name_head = netdev_create_hash();
10234 if (net->dev_name_head == NULL)
10237 net->dev_index_head = netdev_create_hash();
10238 if (net->dev_index_head == NULL)
10241 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
10246 kfree(net->dev_name_head);
10252 * netdev_drivername - network driver for the device
10253 * @dev: network device
10255 * Determine network driver for device.
10257 const char *netdev_drivername(const struct net_device *dev)
10259 const struct device_driver *driver;
10260 const struct device *parent;
10261 const char *empty = "";
10263 parent = dev->dev.parent;
10267 driver = parent->driver;
10268 if (driver && driver->name)
10269 return driver->name;
10273 static void __netdev_printk(const char *level, const struct net_device *dev,
10274 struct va_format *vaf)
10276 if (dev && dev->dev.parent) {
10277 dev_printk_emit(level[1] - '0',
10280 dev_driver_string(dev->dev.parent),
10281 dev_name(dev->dev.parent),
10282 netdev_name(dev), netdev_reg_state(dev),
10285 printk("%s%s%s: %pV",
10286 level, netdev_name(dev), netdev_reg_state(dev), vaf);
10288 printk("%s(NULL net_device): %pV", level, vaf);
10292 void netdev_printk(const char *level, const struct net_device *dev,
10293 const char *format, ...)
10295 struct va_format vaf;
10298 va_start(args, format);
10303 __netdev_printk(level, dev, &vaf);
10307 EXPORT_SYMBOL(netdev_printk);
10309 #define define_netdev_printk_level(func, level) \
10310 void func(const struct net_device *dev, const char *fmt, ...) \
10312 struct va_format vaf; \
10315 va_start(args, fmt); \
10320 __netdev_printk(level, dev, &vaf); \
10324 EXPORT_SYMBOL(func);
10326 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
10327 define_netdev_printk_level(netdev_alert, KERN_ALERT);
10328 define_netdev_printk_level(netdev_crit, KERN_CRIT);
10329 define_netdev_printk_level(netdev_err, KERN_ERR);
10330 define_netdev_printk_level(netdev_warn, KERN_WARNING);
10331 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
10332 define_netdev_printk_level(netdev_info, KERN_INFO);
10334 static void __net_exit netdev_exit(struct net *net)
10336 kfree(net->dev_name_head);
10337 kfree(net->dev_index_head);
10338 if (net != &init_net)
10339 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
10342 static struct pernet_operations __net_initdata netdev_net_ops = {
10343 .init = netdev_init,
10344 .exit = netdev_exit,
10347 static void __net_exit default_device_exit(struct net *net)
10349 struct net_device *dev, *aux;
10351 * Push all migratable network devices back to the
10352 * initial network namespace
10355 for_each_netdev_safe(net, dev, aux) {
10357 char fb_name[IFNAMSIZ];
10359 /* Ignore unmoveable devices (i.e. loopback) */
10360 if (dev->features & NETIF_F_NETNS_LOCAL)
10363 /* Leave virtual devices for the generic cleanup */
10364 if (dev->rtnl_link_ops)
10367 /* Push remaining network devices to init_net */
10368 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
10369 if (__dev_get_by_name(&init_net, fb_name))
10370 snprintf(fb_name, IFNAMSIZ, "dev%%d");
10371 err = dev_change_net_namespace(dev, &init_net, fb_name);
10373 pr_emerg("%s: failed to move %s to init_net: %d\n",
10374 __func__, dev->name, err);
10381 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
10383 /* Return with the rtnl_lock held when there are no network
10384 * devices unregistering in any network namespace in net_list.
10387 bool unregistering;
10388 DEFINE_WAIT_FUNC(wait, woken_wake_function);
10390 add_wait_queue(&netdev_unregistering_wq, &wait);
10392 unregistering = false;
10394 list_for_each_entry(net, net_list, exit_list) {
10395 if (net->dev_unreg_count > 0) {
10396 unregistering = true;
10400 if (!unregistering)
10404 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
10406 remove_wait_queue(&netdev_unregistering_wq, &wait);
10409 static void __net_exit default_device_exit_batch(struct list_head *net_list)
10411 /* At exit all network devices most be removed from a network
10412 * namespace. Do this in the reverse order of registration.
10413 * Do this across as many network namespaces as possible to
10414 * improve batching efficiency.
10416 struct net_device *dev;
10418 LIST_HEAD(dev_kill_list);
10420 /* To prevent network device cleanup code from dereferencing
10421 * loopback devices or network devices that have been freed
10422 * wait here for all pending unregistrations to complete,
10423 * before unregistring the loopback device and allowing the
10424 * network namespace be freed.
10426 * The netdev todo list containing all network devices
10427 * unregistrations that happen in default_device_exit_batch
10428 * will run in the rtnl_unlock() at the end of
10429 * default_device_exit_batch.
10431 rtnl_lock_unregistering(net_list);
10432 list_for_each_entry(net, net_list, exit_list) {
10433 for_each_netdev_reverse(net, dev) {
10434 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
10435 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
10437 unregister_netdevice_queue(dev, &dev_kill_list);
10440 unregister_netdevice_many(&dev_kill_list);
10444 static struct pernet_operations __net_initdata default_device_ops = {
10445 .exit = default_device_exit,
10446 .exit_batch = default_device_exit_batch,
10450 * Initialize the DEV module. At boot time this walks the device list and
10451 * unhooks any devices that fail to initialise (normally hardware not
10452 * present) and leaves us with a valid list of present and active devices.
10457 * This is called single threaded during boot, so no need
10458 * to take the rtnl semaphore.
10460 static int __init net_dev_init(void)
10462 int i, rc = -ENOMEM;
10464 BUG_ON(!dev_boot_phase);
10466 if (dev_proc_init())
10469 if (netdev_kobject_init())
10472 INIT_LIST_HEAD(&ptype_all);
10473 for (i = 0; i < PTYPE_HASH_SIZE; i++)
10474 INIT_LIST_HEAD(&ptype_base[i]);
10476 INIT_LIST_HEAD(&offload_base);
10478 if (register_pernet_subsys(&netdev_net_ops))
10482 * Initialise the packet receive queues.
10485 for_each_possible_cpu(i) {
10486 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
10487 struct softnet_data *sd = &per_cpu(softnet_data, i);
10489 INIT_WORK(flush, flush_backlog);
10491 skb_queue_head_init(&sd->input_pkt_queue);
10492 skb_queue_head_init(&sd->process_queue);
10493 #ifdef CONFIG_XFRM_OFFLOAD
10494 skb_queue_head_init(&sd->xfrm_backlog);
10496 INIT_LIST_HEAD(&sd->poll_list);
10497 sd->output_queue_tailp = &sd->output_queue;
10499 sd->csd.func = rps_trigger_softirq;
10504 init_gro_hash(&sd->backlog);
10505 sd->backlog.poll = process_backlog;
10506 sd->backlog.weight = weight_p;
10509 dev_boot_phase = 0;
10511 /* The loopback device is special if any other network devices
10512 * is present in a network namespace the loopback device must
10513 * be present. Since we now dynamically allocate and free the
10514 * loopback device ensure this invariant is maintained by
10515 * keeping the loopback device as the first device on the
10516 * list of network devices. Ensuring the loopback devices
10517 * is the first device that appears and the last network device
10520 if (register_pernet_device(&loopback_net_ops))
10523 if (register_pernet_device(&default_device_ops))
10526 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
10527 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
10529 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
10530 NULL, dev_cpu_dead);
10537 subsys_initcall(net_dev_init);