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
149 #define MAX_NEST_DEV 8
151 /* This should be increased if a protocol with a bigger head is added. */
152 #define GRO_MAX_HEAD (MAX_HEADER + 128)
154 static DEFINE_SPINLOCK(ptype_lock);
155 static DEFINE_SPINLOCK(offload_lock);
156 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
157 struct list_head ptype_all __read_mostly; /* Taps */
158 static struct list_head offload_base __read_mostly;
160 static int netif_rx_internal(struct sk_buff *skb);
161 static int call_netdevice_notifiers_info(unsigned long val,
162 struct netdev_notifier_info *info);
163 static int call_netdevice_notifiers_extack(unsigned long val,
164 struct net_device *dev,
165 struct netlink_ext_ack *extack);
166 static struct napi_struct *napi_by_id(unsigned int napi_id);
169 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
172 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
174 * Writers must hold the rtnl semaphore while they loop through the
175 * dev_base_head list, and hold dev_base_lock for writing when they do the
176 * actual updates. This allows pure readers to access the list even
177 * while a writer is preparing to update it.
179 * To put it another way, dev_base_lock is held for writing only to
180 * protect against pure readers; the rtnl semaphore provides the
181 * protection against other writers.
183 * See, for example usages, register_netdevice() and
184 * unregister_netdevice(), which must be called with the rtnl
187 DEFINE_RWLOCK(dev_base_lock);
188 EXPORT_SYMBOL(dev_base_lock);
190 static DEFINE_MUTEX(ifalias_mutex);
192 /* protects napi_hash addition/deletion and napi_gen_id */
193 static DEFINE_SPINLOCK(napi_hash_lock);
195 static unsigned int napi_gen_id = NR_CPUS;
196 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
198 static seqcount_t devnet_rename_seq;
200 static inline void dev_base_seq_inc(struct net *net)
202 while (++net->dev_base_seq == 0)
206 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
208 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
210 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
213 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
215 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
218 static inline void rps_lock(struct softnet_data *sd)
221 spin_lock(&sd->input_pkt_queue.lock);
225 static inline void rps_unlock(struct softnet_data *sd)
228 spin_unlock(&sd->input_pkt_queue.lock);
232 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
235 struct netdev_name_node *name_node;
237 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
240 INIT_HLIST_NODE(&name_node->hlist);
241 name_node->dev = dev;
242 name_node->name = name;
246 static struct netdev_name_node *
247 netdev_name_node_head_alloc(struct net_device *dev)
249 struct netdev_name_node *name_node;
251 name_node = netdev_name_node_alloc(dev, dev->name);
254 INIT_LIST_HEAD(&name_node->list);
258 static void netdev_name_node_free(struct netdev_name_node *name_node)
263 static void netdev_name_node_add(struct net *net,
264 struct netdev_name_node *name_node)
266 hlist_add_head_rcu(&name_node->hlist,
267 dev_name_hash(net, name_node->name));
270 static void netdev_name_node_del(struct netdev_name_node *name_node)
272 hlist_del_rcu(&name_node->hlist);
275 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
278 struct hlist_head *head = dev_name_hash(net, name);
279 struct netdev_name_node *name_node;
281 hlist_for_each_entry(name_node, head, hlist)
282 if (!strcmp(name_node->name, name))
287 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
290 struct hlist_head *head = dev_name_hash(net, name);
291 struct netdev_name_node *name_node;
293 hlist_for_each_entry_rcu(name_node, head, hlist)
294 if (!strcmp(name_node->name, name))
299 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
301 struct netdev_name_node *name_node;
302 struct net *net = dev_net(dev);
304 name_node = netdev_name_node_lookup(net, name);
307 name_node = netdev_name_node_alloc(dev, name);
310 netdev_name_node_add(net, name_node);
311 /* The node that holds dev->name acts as a head of per-device list. */
312 list_add_tail(&name_node->list, &dev->name_node->list);
316 EXPORT_SYMBOL(netdev_name_node_alt_create);
318 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
320 list_del(&name_node->list);
321 netdev_name_node_del(name_node);
322 kfree(name_node->name);
323 netdev_name_node_free(name_node);
326 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
328 struct netdev_name_node *name_node;
329 struct net *net = dev_net(dev);
331 name_node = netdev_name_node_lookup(net, name);
334 __netdev_name_node_alt_destroy(name_node);
338 EXPORT_SYMBOL(netdev_name_node_alt_destroy);
340 static void netdev_name_node_alt_flush(struct net_device *dev)
342 struct netdev_name_node *name_node, *tmp;
344 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
345 __netdev_name_node_alt_destroy(name_node);
348 /* Device list insertion */
349 static void list_netdevice(struct net_device *dev)
351 struct net *net = dev_net(dev);
355 write_lock_bh(&dev_base_lock);
356 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
357 netdev_name_node_add(net, dev->name_node);
358 hlist_add_head_rcu(&dev->index_hlist,
359 dev_index_hash(net, dev->ifindex));
360 write_unlock_bh(&dev_base_lock);
362 dev_base_seq_inc(net);
365 /* Device list removal
366 * caller must respect a RCU grace period before freeing/reusing dev
368 static void unlist_netdevice(struct net_device *dev)
372 /* Unlink dev from the device chain */
373 write_lock_bh(&dev_base_lock);
374 list_del_rcu(&dev->dev_list);
375 netdev_name_node_del(dev->name_node);
376 hlist_del_rcu(&dev->index_hlist);
377 write_unlock_bh(&dev_base_lock);
379 dev_base_seq_inc(dev_net(dev));
386 static RAW_NOTIFIER_HEAD(netdev_chain);
389 * Device drivers call our routines to queue packets here. We empty the
390 * queue in the local softnet handler.
393 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
394 EXPORT_PER_CPU_SYMBOL(softnet_data);
396 /*******************************************************************************
398 * Protocol management and registration routines
400 *******************************************************************************/
404 * Add a protocol ID to the list. Now that the input handler is
405 * smarter we can dispense with all the messy stuff that used to be
408 * BEWARE!!! Protocol handlers, mangling input packets,
409 * MUST BE last in hash buckets and checking protocol handlers
410 * MUST start from promiscuous ptype_all chain in net_bh.
411 * It is true now, do not change it.
412 * Explanation follows: if protocol handler, mangling packet, will
413 * be the first on list, it is not able to sense, that packet
414 * is cloned and should be copied-on-write, so that it will
415 * change it and subsequent readers will get broken packet.
419 static inline struct list_head *ptype_head(const struct packet_type *pt)
421 if (pt->type == htons(ETH_P_ALL))
422 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
424 return pt->dev ? &pt->dev->ptype_specific :
425 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
429 * dev_add_pack - add packet handler
430 * @pt: packet type declaration
432 * Add a protocol handler to the networking stack. The passed &packet_type
433 * is linked into kernel lists and may not be freed until it has been
434 * removed from the kernel lists.
436 * This call does not sleep therefore it can not
437 * guarantee all CPU's that are in middle of receiving packets
438 * will see the new packet type (until the next received packet).
441 void dev_add_pack(struct packet_type *pt)
443 struct list_head *head = ptype_head(pt);
445 spin_lock(&ptype_lock);
446 list_add_rcu(&pt->list, head);
447 spin_unlock(&ptype_lock);
449 EXPORT_SYMBOL(dev_add_pack);
452 * __dev_remove_pack - remove packet handler
453 * @pt: packet type declaration
455 * Remove a protocol handler that was previously added to the kernel
456 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
457 * from the kernel lists and can be freed or reused once this function
460 * The packet type might still be in use by receivers
461 * and must not be freed until after all the CPU's have gone
462 * through a quiescent state.
464 void __dev_remove_pack(struct packet_type *pt)
466 struct list_head *head = ptype_head(pt);
467 struct packet_type *pt1;
469 spin_lock(&ptype_lock);
471 list_for_each_entry(pt1, head, list) {
473 list_del_rcu(&pt->list);
478 pr_warn("dev_remove_pack: %p not found\n", pt);
480 spin_unlock(&ptype_lock);
482 EXPORT_SYMBOL(__dev_remove_pack);
485 * dev_remove_pack - remove packet handler
486 * @pt: packet type declaration
488 * Remove a protocol handler that was previously added to the kernel
489 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
490 * from the kernel lists and can be freed or reused once this function
493 * This call sleeps to guarantee that no CPU is looking at the packet
496 void dev_remove_pack(struct packet_type *pt)
498 __dev_remove_pack(pt);
502 EXPORT_SYMBOL(dev_remove_pack);
506 * dev_add_offload - register offload handlers
507 * @po: protocol offload declaration
509 * Add protocol offload handlers to the networking stack. The passed
510 * &proto_offload is linked into kernel lists and may not be freed until
511 * it has been removed from the kernel lists.
513 * This call does not sleep therefore it can not
514 * guarantee all CPU's that are in middle of receiving packets
515 * will see the new offload handlers (until the next received packet).
517 void dev_add_offload(struct packet_offload *po)
519 struct packet_offload *elem;
521 spin_lock(&offload_lock);
522 list_for_each_entry(elem, &offload_base, list) {
523 if (po->priority < elem->priority)
526 list_add_rcu(&po->list, elem->list.prev);
527 spin_unlock(&offload_lock);
529 EXPORT_SYMBOL(dev_add_offload);
532 * __dev_remove_offload - remove offload handler
533 * @po: packet offload declaration
535 * Remove a protocol offload handler that was previously added to the
536 * kernel offload handlers by dev_add_offload(). The passed &offload_type
537 * is removed from the kernel lists and can be freed or reused once this
540 * The packet type might still be in use by receivers
541 * and must not be freed until after all the CPU's have gone
542 * through a quiescent state.
544 static void __dev_remove_offload(struct packet_offload *po)
546 struct list_head *head = &offload_base;
547 struct packet_offload *po1;
549 spin_lock(&offload_lock);
551 list_for_each_entry(po1, head, list) {
553 list_del_rcu(&po->list);
558 pr_warn("dev_remove_offload: %p not found\n", po);
560 spin_unlock(&offload_lock);
564 * dev_remove_offload - remove packet offload handler
565 * @po: packet offload declaration
567 * Remove a packet offload handler that was previously added to the kernel
568 * offload handlers by dev_add_offload(). The passed &offload_type is
569 * removed from the kernel lists and can be freed or reused once this
572 * This call sleeps to guarantee that no CPU is looking at the packet
575 void dev_remove_offload(struct packet_offload *po)
577 __dev_remove_offload(po);
581 EXPORT_SYMBOL(dev_remove_offload);
583 /******************************************************************************
585 * Device Boot-time Settings Routines
587 ******************************************************************************/
589 /* Boot time configuration table */
590 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
593 * netdev_boot_setup_add - add new setup entry
594 * @name: name of the device
595 * @map: configured settings for the device
597 * Adds new setup entry to the dev_boot_setup list. The function
598 * returns 0 on error and 1 on success. This is a generic routine to
601 static int netdev_boot_setup_add(char *name, struct ifmap *map)
603 struct netdev_boot_setup *s;
607 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
608 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
609 memset(s[i].name, 0, sizeof(s[i].name));
610 strlcpy(s[i].name, name, IFNAMSIZ);
611 memcpy(&s[i].map, map, sizeof(s[i].map));
616 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
620 * netdev_boot_setup_check - check boot time settings
621 * @dev: the netdevice
623 * Check boot time settings for the device.
624 * The found settings are set for the device to be used
625 * later in the device probing.
626 * Returns 0 if no settings found, 1 if they are.
628 int netdev_boot_setup_check(struct net_device *dev)
630 struct netdev_boot_setup *s = dev_boot_setup;
633 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
634 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
635 !strcmp(dev->name, s[i].name)) {
636 dev->irq = s[i].map.irq;
637 dev->base_addr = s[i].map.base_addr;
638 dev->mem_start = s[i].map.mem_start;
639 dev->mem_end = s[i].map.mem_end;
645 EXPORT_SYMBOL(netdev_boot_setup_check);
649 * netdev_boot_base - get address from boot time settings
650 * @prefix: prefix for network device
651 * @unit: id for network device
653 * Check boot time settings for the base address of device.
654 * The found settings are set for the device to be used
655 * later in the device probing.
656 * Returns 0 if no settings found.
658 unsigned long netdev_boot_base(const char *prefix, int unit)
660 const struct netdev_boot_setup *s = dev_boot_setup;
664 sprintf(name, "%s%d", prefix, unit);
667 * If device already registered then return base of 1
668 * to indicate not to probe for this interface
670 if (__dev_get_by_name(&init_net, name))
673 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
674 if (!strcmp(name, s[i].name))
675 return s[i].map.base_addr;
680 * Saves at boot time configured settings for any netdevice.
682 int __init netdev_boot_setup(char *str)
687 str = get_options(str, ARRAY_SIZE(ints), ints);
692 memset(&map, 0, sizeof(map));
696 map.base_addr = ints[2];
698 map.mem_start = ints[3];
700 map.mem_end = ints[4];
702 /* Add new entry to the list */
703 return netdev_boot_setup_add(str, &map);
706 __setup("netdev=", netdev_boot_setup);
708 /*******************************************************************************
710 * Device Interface Subroutines
712 *******************************************************************************/
715 * dev_get_iflink - get 'iflink' value of a interface
716 * @dev: targeted interface
718 * Indicates the ifindex the interface is linked to.
719 * Physical interfaces have the same 'ifindex' and 'iflink' values.
722 int dev_get_iflink(const struct net_device *dev)
724 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
725 return dev->netdev_ops->ndo_get_iflink(dev);
729 EXPORT_SYMBOL(dev_get_iflink);
732 * dev_fill_metadata_dst - Retrieve tunnel egress information.
733 * @dev: targeted interface
736 * For better visibility of tunnel traffic OVS needs to retrieve
737 * egress tunnel information for a packet. Following API allows
738 * user to get this info.
740 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
742 struct ip_tunnel_info *info;
744 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
747 info = skb_tunnel_info_unclone(skb);
750 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
753 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
755 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
758 * __dev_get_by_name - find a device by its name
759 * @net: the applicable net namespace
760 * @name: name to find
762 * Find an interface by name. Must be called under RTNL semaphore
763 * or @dev_base_lock. If the name is found a pointer to the device
764 * is returned. If the name is not found then %NULL is returned. The
765 * reference counters are not incremented so the caller must be
766 * careful with locks.
769 struct net_device *__dev_get_by_name(struct net *net, const char *name)
771 struct netdev_name_node *node_name;
773 node_name = netdev_name_node_lookup(net, name);
774 return node_name ? node_name->dev : NULL;
776 EXPORT_SYMBOL(__dev_get_by_name);
779 * dev_get_by_name_rcu - find a device by its name
780 * @net: the applicable net namespace
781 * @name: name to find
783 * Find an interface by name.
784 * If the name is found a pointer to the device is returned.
785 * If the name is not found then %NULL is returned.
786 * The reference counters are not incremented so the caller must be
787 * careful with locks. The caller must hold RCU lock.
790 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
792 struct netdev_name_node *node_name;
794 node_name = netdev_name_node_lookup_rcu(net, name);
795 return node_name ? node_name->dev : NULL;
797 EXPORT_SYMBOL(dev_get_by_name_rcu);
800 * dev_get_by_name - find a device by its name
801 * @net: the applicable net namespace
802 * @name: name to find
804 * Find an interface by name. This can be called from any
805 * context and does its own locking. The returned handle has
806 * the usage count incremented and the caller must use dev_put() to
807 * release it when it is no longer needed. %NULL is returned if no
808 * matching device is found.
811 struct net_device *dev_get_by_name(struct net *net, const char *name)
813 struct net_device *dev;
816 dev = dev_get_by_name_rcu(net, name);
822 EXPORT_SYMBOL(dev_get_by_name);
825 * __dev_get_by_index - find a device by its ifindex
826 * @net: the applicable net namespace
827 * @ifindex: index of device
829 * Search for an interface by index. Returns %NULL if the device
830 * is not found or a pointer to the device. The device has not
831 * had its reference counter increased so the caller must be careful
832 * about locking. The caller must hold either the RTNL semaphore
836 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
838 struct net_device *dev;
839 struct hlist_head *head = dev_index_hash(net, ifindex);
841 hlist_for_each_entry(dev, head, index_hlist)
842 if (dev->ifindex == ifindex)
847 EXPORT_SYMBOL(__dev_get_by_index);
850 * dev_get_by_index_rcu - find a device by its ifindex
851 * @net: the applicable net namespace
852 * @ifindex: index of device
854 * Search for an interface by index. Returns %NULL if the device
855 * is not found or a pointer to the device. The device has not
856 * had its reference counter increased so the caller must be careful
857 * about locking. The caller must hold RCU lock.
860 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
862 struct net_device *dev;
863 struct hlist_head *head = dev_index_hash(net, ifindex);
865 hlist_for_each_entry_rcu(dev, head, index_hlist)
866 if (dev->ifindex == ifindex)
871 EXPORT_SYMBOL(dev_get_by_index_rcu);
875 * dev_get_by_index - find a device by its ifindex
876 * @net: the applicable net namespace
877 * @ifindex: index of device
879 * Search for an interface by index. Returns NULL if the device
880 * is not found or a pointer to the device. The device returned has
881 * had a reference added and the pointer is safe until the user calls
882 * dev_put to indicate they have finished with it.
885 struct net_device *dev_get_by_index(struct net *net, int ifindex)
887 struct net_device *dev;
890 dev = dev_get_by_index_rcu(net, ifindex);
896 EXPORT_SYMBOL(dev_get_by_index);
899 * dev_get_by_napi_id - find a device by napi_id
900 * @napi_id: ID of the NAPI struct
902 * Search for an interface by NAPI ID. Returns %NULL if the device
903 * is not found or a pointer to the device. The device has not had
904 * its reference counter increased so the caller must be careful
905 * about locking. The caller must hold RCU lock.
908 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
910 struct napi_struct *napi;
912 WARN_ON_ONCE(!rcu_read_lock_held());
914 if (napi_id < MIN_NAPI_ID)
917 napi = napi_by_id(napi_id);
919 return napi ? napi->dev : NULL;
921 EXPORT_SYMBOL(dev_get_by_napi_id);
924 * netdev_get_name - get a netdevice name, knowing its ifindex.
925 * @net: network namespace
926 * @name: a pointer to the buffer where the name will be stored.
927 * @ifindex: the ifindex of the interface to get the name from.
929 * The use of raw_seqcount_begin() and cond_resched() before
930 * retrying is required as we want to give the writers a chance
931 * to complete when CONFIG_PREEMPT is not set.
933 int netdev_get_name(struct net *net, char *name, int ifindex)
935 struct net_device *dev;
939 seq = raw_seqcount_begin(&devnet_rename_seq);
941 dev = dev_get_by_index_rcu(net, ifindex);
947 strcpy(name, dev->name);
949 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
958 * dev_getbyhwaddr_rcu - find a device by its hardware address
959 * @net: the applicable net namespace
960 * @type: media type of device
961 * @ha: hardware address
963 * Search for an interface by MAC address. Returns NULL if the device
964 * is not found or a pointer to the device.
965 * The caller must hold RCU or RTNL.
966 * The returned device has not had its ref count increased
967 * and the caller must therefore be careful about locking
971 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
974 struct net_device *dev;
976 for_each_netdev_rcu(net, dev)
977 if (dev->type == type &&
978 !memcmp(dev->dev_addr, ha, dev->addr_len))
983 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
985 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
987 struct net_device *dev;
990 for_each_netdev(net, dev)
991 if (dev->type == type)
996 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
998 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
1000 struct net_device *dev, *ret = NULL;
1003 for_each_netdev_rcu(net, dev)
1004 if (dev->type == type) {
1012 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1015 * __dev_get_by_flags - find any device with given flags
1016 * @net: the applicable net namespace
1017 * @if_flags: IFF_* values
1018 * @mask: bitmask of bits in if_flags to check
1020 * Search for any interface with the given flags. Returns NULL if a device
1021 * is not found or a pointer to the device. Must be called inside
1022 * rtnl_lock(), and result refcount is unchanged.
1025 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1026 unsigned short mask)
1028 struct net_device *dev, *ret;
1033 for_each_netdev(net, dev) {
1034 if (((dev->flags ^ if_flags) & mask) == 0) {
1041 EXPORT_SYMBOL(__dev_get_by_flags);
1044 * dev_valid_name - check if name is okay for network device
1045 * @name: name string
1047 * Network device names need to be valid file names to
1048 * to allow sysfs to work. We also disallow any kind of
1051 bool dev_valid_name(const char *name)
1055 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1057 if (!strcmp(name, ".") || !strcmp(name, ".."))
1061 if (*name == '/' || *name == ':' || isspace(*name))
1067 EXPORT_SYMBOL(dev_valid_name);
1070 * __dev_alloc_name - allocate a name for a device
1071 * @net: network namespace to allocate the device name in
1072 * @name: name format string
1073 * @buf: scratch buffer and result name string
1075 * Passed a format string - eg "lt%d" it will try and find a suitable
1076 * id. It scans list of devices to build up a free map, then chooses
1077 * the first empty slot. The caller must hold the dev_base or rtnl lock
1078 * while allocating the name and adding the device in order to avoid
1080 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1081 * Returns the number of the unit assigned or a negative errno code.
1084 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1088 const int max_netdevices = 8*PAGE_SIZE;
1089 unsigned long *inuse;
1090 struct net_device *d;
1092 if (!dev_valid_name(name))
1095 p = strchr(name, '%');
1098 * Verify the string as this thing may have come from
1099 * the user. There must be either one "%d" and no other "%"
1102 if (p[1] != 'd' || strchr(p + 2, '%'))
1105 /* Use one page as a bit array of possible slots */
1106 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1110 for_each_netdev(net, d) {
1111 if (!sscanf(d->name, name, &i))
1113 if (i < 0 || i >= max_netdevices)
1116 /* avoid cases where sscanf is not exact inverse of printf */
1117 snprintf(buf, IFNAMSIZ, name, i);
1118 if (!strncmp(buf, d->name, IFNAMSIZ))
1122 i = find_first_zero_bit(inuse, max_netdevices);
1123 free_page((unsigned long) inuse);
1126 snprintf(buf, IFNAMSIZ, name, i);
1127 if (!__dev_get_by_name(net, buf))
1130 /* It is possible to run out of possible slots
1131 * when the name is long and there isn't enough space left
1132 * for the digits, or if all bits are used.
1137 static int dev_alloc_name_ns(struct net *net,
1138 struct net_device *dev,
1145 ret = __dev_alloc_name(net, name, buf);
1147 strlcpy(dev->name, buf, IFNAMSIZ);
1152 * dev_alloc_name - allocate a name for a device
1154 * @name: name format string
1156 * Passed a format string - eg "lt%d" it will try and find a suitable
1157 * id. It scans list of devices to build up a free map, then chooses
1158 * the first empty slot. The caller must hold the dev_base or rtnl lock
1159 * while allocating the name and adding the device in order to avoid
1161 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1162 * Returns the number of the unit assigned or a negative errno code.
1165 int dev_alloc_name(struct net_device *dev, const char *name)
1167 return dev_alloc_name_ns(dev_net(dev), dev, name);
1169 EXPORT_SYMBOL(dev_alloc_name);
1171 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1176 if (!dev_valid_name(name))
1179 if (strchr(name, '%'))
1180 return dev_alloc_name_ns(net, dev, name);
1181 else if (__dev_get_by_name(net, name))
1183 else if (dev->name != name)
1184 strlcpy(dev->name, name, IFNAMSIZ);
1190 * dev_change_name - change name of a device
1192 * @newname: name (or format string) must be at least IFNAMSIZ
1194 * Change name of a device, can pass format strings "eth%d".
1197 int dev_change_name(struct net_device *dev, const char *newname)
1199 unsigned char old_assign_type;
1200 char oldname[IFNAMSIZ];
1206 BUG_ON(!dev_net(dev));
1210 /* Some auto-enslaved devices e.g. failover slaves are
1211 * special, as userspace might rename the device after
1212 * the interface had been brought up and running since
1213 * the point kernel initiated auto-enslavement. Allow
1214 * live name change even when these slave devices are
1217 * Typically, users of these auto-enslaving devices
1218 * don't actually care about slave name change, as
1219 * they are supposed to operate on master interface
1222 if (dev->flags & IFF_UP &&
1223 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1226 write_seqcount_begin(&devnet_rename_seq);
1228 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1229 write_seqcount_end(&devnet_rename_seq);
1233 memcpy(oldname, dev->name, IFNAMSIZ);
1235 err = dev_get_valid_name(net, dev, newname);
1237 write_seqcount_end(&devnet_rename_seq);
1241 if (oldname[0] && !strchr(oldname, '%'))
1242 netdev_info(dev, "renamed from %s\n", oldname);
1244 old_assign_type = dev->name_assign_type;
1245 dev->name_assign_type = NET_NAME_RENAMED;
1248 ret = device_rename(&dev->dev, dev->name);
1250 memcpy(dev->name, oldname, IFNAMSIZ);
1251 dev->name_assign_type = old_assign_type;
1252 write_seqcount_end(&devnet_rename_seq);
1256 write_seqcount_end(&devnet_rename_seq);
1258 netdev_adjacent_rename_links(dev, oldname);
1260 write_lock_bh(&dev_base_lock);
1261 netdev_name_node_del(dev->name_node);
1262 write_unlock_bh(&dev_base_lock);
1266 write_lock_bh(&dev_base_lock);
1267 netdev_name_node_add(net, dev->name_node);
1268 write_unlock_bh(&dev_base_lock);
1270 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1271 ret = notifier_to_errno(ret);
1274 /* err >= 0 after dev_alloc_name() or stores the first errno */
1277 write_seqcount_begin(&devnet_rename_seq);
1278 memcpy(dev->name, oldname, IFNAMSIZ);
1279 memcpy(oldname, newname, IFNAMSIZ);
1280 dev->name_assign_type = old_assign_type;
1281 old_assign_type = NET_NAME_RENAMED;
1284 pr_err("%s: name change rollback failed: %d\n",
1293 * dev_set_alias - change ifalias of a device
1295 * @alias: name up to IFALIASZ
1296 * @len: limit of bytes to copy from info
1298 * Set ifalias for a device,
1300 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1302 struct dev_ifalias *new_alias = NULL;
1304 if (len >= IFALIASZ)
1308 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1312 memcpy(new_alias->ifalias, alias, len);
1313 new_alias->ifalias[len] = 0;
1316 mutex_lock(&ifalias_mutex);
1317 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1318 mutex_is_locked(&ifalias_mutex));
1319 mutex_unlock(&ifalias_mutex);
1322 kfree_rcu(new_alias, rcuhead);
1326 EXPORT_SYMBOL(dev_set_alias);
1329 * dev_get_alias - get ifalias of a device
1331 * @name: buffer to store name of ifalias
1332 * @len: size of buffer
1334 * get ifalias for a device. Caller must make sure dev cannot go
1335 * away, e.g. rcu read lock or own a reference count to device.
1337 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1339 const struct dev_ifalias *alias;
1343 alias = rcu_dereference(dev->ifalias);
1345 ret = snprintf(name, len, "%s", alias->ifalias);
1352 * netdev_features_change - device changes features
1353 * @dev: device to cause notification
1355 * Called to indicate a device has changed features.
1357 void netdev_features_change(struct net_device *dev)
1359 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1361 EXPORT_SYMBOL(netdev_features_change);
1364 * netdev_state_change - device changes state
1365 * @dev: device to cause notification
1367 * Called to indicate a device has changed state. This function calls
1368 * the notifier chains for netdev_chain and sends a NEWLINK message
1369 * to the routing socket.
1371 void netdev_state_change(struct net_device *dev)
1373 if (dev->flags & IFF_UP) {
1374 struct netdev_notifier_change_info change_info = {
1378 call_netdevice_notifiers_info(NETDEV_CHANGE,
1380 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1383 EXPORT_SYMBOL(netdev_state_change);
1386 * netdev_notify_peers - notify network peers about existence of @dev
1387 * @dev: network device
1389 * Generate traffic such that interested network peers are aware of
1390 * @dev, such as by generating a gratuitous ARP. This may be used when
1391 * a device wants to inform the rest of the network about some sort of
1392 * reconfiguration such as a failover event or virtual machine
1395 void netdev_notify_peers(struct net_device *dev)
1398 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1399 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1402 EXPORT_SYMBOL(netdev_notify_peers);
1404 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1406 const struct net_device_ops *ops = dev->netdev_ops;
1411 if (!netif_device_present(dev))
1414 /* Block netpoll from trying to do any rx path servicing.
1415 * If we don't do this there is a chance ndo_poll_controller
1416 * or ndo_poll may be running while we open the device
1418 netpoll_poll_disable(dev);
1420 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1421 ret = notifier_to_errno(ret);
1425 set_bit(__LINK_STATE_START, &dev->state);
1427 if (ops->ndo_validate_addr)
1428 ret = ops->ndo_validate_addr(dev);
1430 if (!ret && ops->ndo_open)
1431 ret = ops->ndo_open(dev);
1433 netpoll_poll_enable(dev);
1436 clear_bit(__LINK_STATE_START, &dev->state);
1438 dev->flags |= IFF_UP;
1439 dev_set_rx_mode(dev);
1441 add_device_randomness(dev->dev_addr, dev->addr_len);
1448 * dev_open - prepare an interface for use.
1449 * @dev: device to open
1450 * @extack: netlink extended ack
1452 * Takes a device from down to up state. The device's private open
1453 * function is invoked and then the multicast lists are loaded. Finally
1454 * the device is moved into the up state and a %NETDEV_UP message is
1455 * sent to the netdev notifier chain.
1457 * Calling this function on an active interface is a nop. On a failure
1458 * a negative errno code is returned.
1460 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1464 if (dev->flags & IFF_UP)
1467 ret = __dev_open(dev, extack);
1471 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1472 call_netdevice_notifiers(NETDEV_UP, dev);
1476 EXPORT_SYMBOL(dev_open);
1478 static void __dev_close_many(struct list_head *head)
1480 struct net_device *dev;
1485 list_for_each_entry(dev, head, close_list) {
1486 /* Temporarily disable netpoll until the interface is down */
1487 netpoll_poll_disable(dev);
1489 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1491 clear_bit(__LINK_STATE_START, &dev->state);
1493 /* Synchronize to scheduled poll. We cannot touch poll list, it
1494 * can be even on different cpu. So just clear netif_running().
1496 * dev->stop() will invoke napi_disable() on all of it's
1497 * napi_struct instances on this device.
1499 smp_mb__after_atomic(); /* Commit netif_running(). */
1502 dev_deactivate_many(head);
1504 list_for_each_entry(dev, head, close_list) {
1505 const struct net_device_ops *ops = dev->netdev_ops;
1508 * Call the device specific close. This cannot fail.
1509 * Only if device is UP
1511 * We allow it to be called even after a DETACH hot-plug
1517 dev->flags &= ~IFF_UP;
1518 netpoll_poll_enable(dev);
1522 static void __dev_close(struct net_device *dev)
1526 list_add(&dev->close_list, &single);
1527 __dev_close_many(&single);
1531 void dev_close_many(struct list_head *head, bool unlink)
1533 struct net_device *dev, *tmp;
1535 /* Remove the devices that don't need to be closed */
1536 list_for_each_entry_safe(dev, tmp, head, close_list)
1537 if (!(dev->flags & IFF_UP))
1538 list_del_init(&dev->close_list);
1540 __dev_close_many(head);
1542 list_for_each_entry_safe(dev, tmp, head, close_list) {
1543 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1544 call_netdevice_notifiers(NETDEV_DOWN, dev);
1546 list_del_init(&dev->close_list);
1549 EXPORT_SYMBOL(dev_close_many);
1552 * dev_close - shutdown an interface.
1553 * @dev: device to shutdown
1555 * This function moves an active device into down state. A
1556 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1557 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1560 void dev_close(struct net_device *dev)
1562 if (dev->flags & IFF_UP) {
1565 list_add(&dev->close_list, &single);
1566 dev_close_many(&single, true);
1570 EXPORT_SYMBOL(dev_close);
1574 * dev_disable_lro - disable Large Receive Offload on a device
1577 * Disable Large Receive Offload (LRO) on a net device. Must be
1578 * called under RTNL. This is needed if received packets may be
1579 * forwarded to another interface.
1581 void dev_disable_lro(struct net_device *dev)
1583 struct net_device *lower_dev;
1584 struct list_head *iter;
1586 dev->wanted_features &= ~NETIF_F_LRO;
1587 netdev_update_features(dev);
1589 if (unlikely(dev->features & NETIF_F_LRO))
1590 netdev_WARN(dev, "failed to disable LRO!\n");
1592 netdev_for_each_lower_dev(dev, lower_dev, iter)
1593 dev_disable_lro(lower_dev);
1595 EXPORT_SYMBOL(dev_disable_lro);
1598 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1601 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1602 * called under RTNL. This is needed if Generic XDP is installed on
1605 static void dev_disable_gro_hw(struct net_device *dev)
1607 dev->wanted_features &= ~NETIF_F_GRO_HW;
1608 netdev_update_features(dev);
1610 if (unlikely(dev->features & NETIF_F_GRO_HW))
1611 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1614 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1617 case NETDEV_##val: \
1618 return "NETDEV_" __stringify(val);
1620 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1621 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1622 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1623 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1624 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1625 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1626 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1627 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1628 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1632 return "UNKNOWN_NETDEV_EVENT";
1634 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1636 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1637 struct net_device *dev)
1639 struct netdev_notifier_info info = {
1643 return nb->notifier_call(nb, val, &info);
1646 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1647 struct net_device *dev)
1651 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1652 err = notifier_to_errno(err);
1656 if (!(dev->flags & IFF_UP))
1659 call_netdevice_notifier(nb, NETDEV_UP, dev);
1663 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1664 struct net_device *dev)
1666 if (dev->flags & IFF_UP) {
1667 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1669 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1671 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1674 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1677 struct net_device *dev;
1680 for_each_netdev(net, dev) {
1681 err = call_netdevice_register_notifiers(nb, dev);
1688 for_each_netdev_continue_reverse(net, dev)
1689 call_netdevice_unregister_notifiers(nb, dev);
1693 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1696 struct net_device *dev;
1698 for_each_netdev(net, dev)
1699 call_netdevice_unregister_notifiers(nb, dev);
1702 static int dev_boot_phase = 1;
1705 * register_netdevice_notifier - register a network notifier block
1708 * Register a notifier to be called when network device events occur.
1709 * The notifier passed is linked into the kernel structures and must
1710 * not be reused until it has been unregistered. A negative errno code
1711 * is returned on a failure.
1713 * When registered all registration and up events are replayed
1714 * to the new notifier to allow device to have a race free
1715 * view of the network device list.
1718 int register_netdevice_notifier(struct notifier_block *nb)
1723 /* Close race with setup_net() and cleanup_net() */
1724 down_write(&pernet_ops_rwsem);
1726 err = raw_notifier_chain_register(&netdev_chain, nb);
1732 err = call_netdevice_register_net_notifiers(nb, net);
1739 up_write(&pernet_ops_rwsem);
1743 for_each_net_continue_reverse(net)
1744 call_netdevice_unregister_net_notifiers(nb, net);
1746 raw_notifier_chain_unregister(&netdev_chain, nb);
1749 EXPORT_SYMBOL(register_netdevice_notifier);
1752 * unregister_netdevice_notifier - unregister a network notifier block
1755 * Unregister a notifier previously registered by
1756 * register_netdevice_notifier(). The notifier is unlinked into the
1757 * kernel structures and may then be reused. A negative errno code
1758 * is returned on a failure.
1760 * After unregistering unregister and down device events are synthesized
1761 * for all devices on the device list to the removed notifier to remove
1762 * the need for special case cleanup code.
1765 int unregister_netdevice_notifier(struct notifier_block *nb)
1770 /* Close race with setup_net() and cleanup_net() */
1771 down_write(&pernet_ops_rwsem);
1773 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1778 call_netdevice_unregister_net_notifiers(nb, net);
1782 up_write(&pernet_ops_rwsem);
1785 EXPORT_SYMBOL(unregister_netdevice_notifier);
1788 * register_netdevice_notifier_net - register a per-netns network notifier block
1789 * @net: network namespace
1792 * Register a notifier to be called when network device events occur.
1793 * The notifier passed is linked into the kernel structures and must
1794 * not be reused until it has been unregistered. A negative errno code
1795 * is returned on a failure.
1797 * When registered all registration and up events are replayed
1798 * to the new notifier to allow device to have a race free
1799 * view of the network device list.
1802 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1807 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1813 err = call_netdevice_register_net_notifiers(nb, net);
1815 goto chain_unregister;
1822 raw_notifier_chain_unregister(&netdev_chain, nb);
1825 EXPORT_SYMBOL(register_netdevice_notifier_net);
1828 * unregister_netdevice_notifier_net - unregister a per-netns
1829 * network notifier block
1830 * @net: network namespace
1833 * Unregister a notifier previously registered by
1834 * register_netdevice_notifier(). The notifier is unlinked into the
1835 * kernel structures and may then be reused. A negative errno code
1836 * is returned on a failure.
1838 * After unregistering unregister and down device events are synthesized
1839 * for all devices on the device list to the removed notifier to remove
1840 * the need for special case cleanup code.
1843 int unregister_netdevice_notifier_net(struct net *net,
1844 struct notifier_block *nb)
1849 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1853 call_netdevice_unregister_net_notifiers(nb, net);
1859 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1862 * call_netdevice_notifiers_info - call all network notifier blocks
1863 * @val: value passed unmodified to notifier function
1864 * @info: notifier information data
1866 * Call all network notifier blocks. Parameters and return value
1867 * are as for raw_notifier_call_chain().
1870 static int call_netdevice_notifiers_info(unsigned long val,
1871 struct netdev_notifier_info *info)
1873 struct net *net = dev_net(info->dev);
1878 /* Run per-netns notifier block chain first, then run the global one.
1879 * Hopefully, one day, the global one is going to be removed after
1880 * all notifier block registrators get converted to be per-netns.
1882 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1883 if (ret & NOTIFY_STOP_MASK)
1885 return raw_notifier_call_chain(&netdev_chain, val, info);
1888 static int call_netdevice_notifiers_extack(unsigned long val,
1889 struct net_device *dev,
1890 struct netlink_ext_ack *extack)
1892 struct netdev_notifier_info info = {
1897 return call_netdevice_notifiers_info(val, &info);
1901 * call_netdevice_notifiers - call all network notifier blocks
1902 * @val: value passed unmodified to notifier function
1903 * @dev: net_device pointer passed unmodified to notifier function
1905 * Call all network notifier blocks. Parameters and return value
1906 * are as for raw_notifier_call_chain().
1909 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1911 return call_netdevice_notifiers_extack(val, dev, NULL);
1913 EXPORT_SYMBOL(call_netdevice_notifiers);
1916 * call_netdevice_notifiers_mtu - call all network notifier blocks
1917 * @val: value passed unmodified to notifier function
1918 * @dev: net_device pointer passed unmodified to notifier function
1919 * @arg: additional u32 argument passed to the notifier function
1921 * Call all network notifier blocks. Parameters and return value
1922 * are as for raw_notifier_call_chain().
1924 static int call_netdevice_notifiers_mtu(unsigned long val,
1925 struct net_device *dev, u32 arg)
1927 struct netdev_notifier_info_ext info = {
1932 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
1934 return call_netdevice_notifiers_info(val, &info.info);
1937 #ifdef CONFIG_NET_INGRESS
1938 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
1940 void net_inc_ingress_queue(void)
1942 static_branch_inc(&ingress_needed_key);
1944 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1946 void net_dec_ingress_queue(void)
1948 static_branch_dec(&ingress_needed_key);
1950 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1953 #ifdef CONFIG_NET_EGRESS
1954 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
1956 void net_inc_egress_queue(void)
1958 static_branch_inc(&egress_needed_key);
1960 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1962 void net_dec_egress_queue(void)
1964 static_branch_dec(&egress_needed_key);
1966 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1969 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
1970 #ifdef CONFIG_JUMP_LABEL
1971 static atomic_t netstamp_needed_deferred;
1972 static atomic_t netstamp_wanted;
1973 static void netstamp_clear(struct work_struct *work)
1975 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1978 wanted = atomic_add_return(deferred, &netstamp_wanted);
1980 static_branch_enable(&netstamp_needed_key);
1982 static_branch_disable(&netstamp_needed_key);
1984 static DECLARE_WORK(netstamp_work, netstamp_clear);
1987 void net_enable_timestamp(void)
1989 #ifdef CONFIG_JUMP_LABEL
1993 wanted = atomic_read(&netstamp_wanted);
1996 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1999 atomic_inc(&netstamp_needed_deferred);
2000 schedule_work(&netstamp_work);
2002 static_branch_inc(&netstamp_needed_key);
2005 EXPORT_SYMBOL(net_enable_timestamp);
2007 void net_disable_timestamp(void)
2009 #ifdef CONFIG_JUMP_LABEL
2013 wanted = atomic_read(&netstamp_wanted);
2016 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2019 atomic_dec(&netstamp_needed_deferred);
2020 schedule_work(&netstamp_work);
2022 static_branch_dec(&netstamp_needed_key);
2025 EXPORT_SYMBOL(net_disable_timestamp);
2027 static inline void net_timestamp_set(struct sk_buff *skb)
2030 if (static_branch_unlikely(&netstamp_needed_key))
2031 __net_timestamp(skb);
2034 #define net_timestamp_check(COND, SKB) \
2035 if (static_branch_unlikely(&netstamp_needed_key)) { \
2036 if ((COND) && !(SKB)->tstamp) \
2037 __net_timestamp(SKB); \
2040 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2044 if (!(dev->flags & IFF_UP))
2047 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
2048 if (skb->len <= len)
2051 /* if TSO is enabled, we don't care about the length as the packet
2052 * could be forwarded without being segmented before
2054 if (skb_is_gso(skb))
2059 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2061 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2063 int ret = ____dev_forward_skb(dev, skb);
2066 skb->protocol = eth_type_trans(skb, dev);
2067 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2072 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2075 * dev_forward_skb - loopback an skb to another netif
2077 * @dev: destination network device
2078 * @skb: buffer to forward
2081 * NET_RX_SUCCESS (no congestion)
2082 * NET_RX_DROP (packet was dropped, but freed)
2084 * dev_forward_skb can be used for injecting an skb from the
2085 * start_xmit function of one device into the receive queue
2086 * of another device.
2088 * The receiving device may be in another namespace, so
2089 * we have to clear all information in the skb that could
2090 * impact namespace isolation.
2092 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2094 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2096 EXPORT_SYMBOL_GPL(dev_forward_skb);
2098 static inline int deliver_skb(struct sk_buff *skb,
2099 struct packet_type *pt_prev,
2100 struct net_device *orig_dev)
2102 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2104 refcount_inc(&skb->users);
2105 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2108 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2109 struct packet_type **pt,
2110 struct net_device *orig_dev,
2112 struct list_head *ptype_list)
2114 struct packet_type *ptype, *pt_prev = *pt;
2116 list_for_each_entry_rcu(ptype, ptype_list, list) {
2117 if (ptype->type != type)
2120 deliver_skb(skb, pt_prev, orig_dev);
2126 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2128 if (!ptype->af_packet_priv || !skb->sk)
2131 if (ptype->id_match)
2132 return ptype->id_match(ptype, skb->sk);
2133 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2140 * dev_nit_active - return true if any network interface taps are in use
2142 * @dev: network device to check for the presence of taps
2144 bool dev_nit_active(struct net_device *dev)
2146 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2148 EXPORT_SYMBOL_GPL(dev_nit_active);
2151 * Support routine. Sends outgoing frames to any network
2152 * taps currently in use.
2155 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2157 struct packet_type *ptype;
2158 struct sk_buff *skb2 = NULL;
2159 struct packet_type *pt_prev = NULL;
2160 struct list_head *ptype_list = &ptype_all;
2164 list_for_each_entry_rcu(ptype, ptype_list, list) {
2165 if (ptype->ignore_outgoing)
2168 /* Never send packets back to the socket
2169 * they originated from - MvS (miquels@drinkel.ow.org)
2171 if (skb_loop_sk(ptype, skb))
2175 deliver_skb(skb2, pt_prev, skb->dev);
2180 /* need to clone skb, done only once */
2181 skb2 = skb_clone(skb, GFP_ATOMIC);
2185 net_timestamp_set(skb2);
2187 /* skb->nh should be correctly
2188 * set by sender, so that the second statement is
2189 * just protection against buggy protocols.
2191 skb_reset_mac_header(skb2);
2193 if (skb_network_header(skb2) < skb2->data ||
2194 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2195 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2196 ntohs(skb2->protocol),
2198 skb_reset_network_header(skb2);
2201 skb2->transport_header = skb2->network_header;
2202 skb2->pkt_type = PACKET_OUTGOING;
2206 if (ptype_list == &ptype_all) {
2207 ptype_list = &dev->ptype_all;
2212 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2213 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2219 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2222 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2223 * @dev: Network device
2224 * @txq: number of queues available
2226 * If real_num_tx_queues is changed the tc mappings may no longer be
2227 * valid. To resolve this verify the tc mapping remains valid and if
2228 * not NULL the mapping. With no priorities mapping to this
2229 * offset/count pair it will no longer be used. In the worst case TC0
2230 * is invalid nothing can be done so disable priority mappings. If is
2231 * expected that drivers will fix this mapping if they can before
2232 * calling netif_set_real_num_tx_queues.
2234 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2237 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2239 /* If TC0 is invalidated disable TC mapping */
2240 if (tc->offset + tc->count > txq) {
2241 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2246 /* Invalidated prio to tc mappings set to TC0 */
2247 for (i = 1; i < TC_BITMASK + 1; i++) {
2248 int q = netdev_get_prio_tc_map(dev, i);
2250 tc = &dev->tc_to_txq[q];
2251 if (tc->offset + tc->count > txq) {
2252 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2254 netdev_set_prio_tc_map(dev, i, 0);
2259 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2262 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2265 /* walk through the TCs and see if it falls into any of them */
2266 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2267 if ((txq - tc->offset) < tc->count)
2271 /* didn't find it, just return -1 to indicate no match */
2277 EXPORT_SYMBOL(netdev_txq_to_tc);
2280 struct static_key xps_needed __read_mostly;
2281 EXPORT_SYMBOL(xps_needed);
2282 struct static_key xps_rxqs_needed __read_mostly;
2283 EXPORT_SYMBOL(xps_rxqs_needed);
2284 static DEFINE_MUTEX(xps_map_mutex);
2285 #define xmap_dereference(P) \
2286 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2288 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2291 struct xps_map *map = NULL;
2295 map = xmap_dereference(dev_maps->attr_map[tci]);
2299 for (pos = map->len; pos--;) {
2300 if (map->queues[pos] != index)
2304 map->queues[pos] = map->queues[--map->len];
2308 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2309 kfree_rcu(map, rcu);
2316 static bool remove_xps_queue_cpu(struct net_device *dev,
2317 struct xps_dev_maps *dev_maps,
2318 int cpu, u16 offset, u16 count)
2320 int num_tc = dev->num_tc ? : 1;
2321 bool active = false;
2324 for (tci = cpu * num_tc; num_tc--; tci++) {
2327 for (i = count, j = offset; i--; j++) {
2328 if (!remove_xps_queue(dev_maps, tci, j))
2338 static void reset_xps_maps(struct net_device *dev,
2339 struct xps_dev_maps *dev_maps,
2343 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2344 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2346 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2348 static_key_slow_dec_cpuslocked(&xps_needed);
2349 kfree_rcu(dev_maps, rcu);
2352 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask,
2353 struct xps_dev_maps *dev_maps, unsigned int nr_ids,
2354 u16 offset, u16 count, bool is_rxqs_map)
2356 bool active = false;
2359 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids),
2361 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset,
2364 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2367 for (i = offset + (count - 1); count--; i--) {
2368 netdev_queue_numa_node_write(
2369 netdev_get_tx_queue(dev, i),
2375 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2378 const unsigned long *possible_mask = NULL;
2379 struct xps_dev_maps *dev_maps;
2380 unsigned int nr_ids;
2382 if (!static_key_false(&xps_needed))
2386 mutex_lock(&xps_map_mutex);
2388 if (static_key_false(&xps_rxqs_needed)) {
2389 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2391 nr_ids = dev->num_rx_queues;
2392 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids,
2393 offset, count, true);
2397 dev_maps = xmap_dereference(dev->xps_cpus_map);
2401 if (num_possible_cpus() > 1)
2402 possible_mask = cpumask_bits(cpu_possible_mask);
2403 nr_ids = nr_cpu_ids;
2404 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count,
2408 mutex_unlock(&xps_map_mutex);
2412 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2414 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2417 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2418 u16 index, bool is_rxqs_map)
2420 struct xps_map *new_map;
2421 int alloc_len = XPS_MIN_MAP_ALLOC;
2424 for (pos = 0; map && pos < map->len; pos++) {
2425 if (map->queues[pos] != index)
2430 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2432 if (pos < map->alloc_len)
2435 alloc_len = map->alloc_len * 2;
2438 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2442 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2444 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2445 cpu_to_node(attr_index));
2449 for (i = 0; i < pos; i++)
2450 new_map->queues[i] = map->queues[i];
2451 new_map->alloc_len = alloc_len;
2457 /* Must be called under cpus_read_lock */
2458 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2459 u16 index, bool is_rxqs_map)
2461 const unsigned long *online_mask = NULL, *possible_mask = NULL;
2462 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2463 int i, j, tci, numa_node_id = -2;
2464 int maps_sz, num_tc = 1, tc = 0;
2465 struct xps_map *map, *new_map;
2466 bool active = false;
2467 unsigned int nr_ids;
2470 /* Do not allow XPS on subordinate device directly */
2471 num_tc = dev->num_tc;
2475 /* If queue belongs to subordinate dev use its map */
2476 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2478 tc = netdev_txq_to_tc(dev, index);
2483 mutex_lock(&xps_map_mutex);
2485 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2486 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2487 nr_ids = dev->num_rx_queues;
2489 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2490 if (num_possible_cpus() > 1) {
2491 online_mask = cpumask_bits(cpu_online_mask);
2492 possible_mask = cpumask_bits(cpu_possible_mask);
2494 dev_maps = xmap_dereference(dev->xps_cpus_map);
2495 nr_ids = nr_cpu_ids;
2498 if (maps_sz < L1_CACHE_BYTES)
2499 maps_sz = L1_CACHE_BYTES;
2501 /* allocate memory for queue storage */
2502 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2505 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2506 if (!new_dev_maps) {
2507 mutex_unlock(&xps_map_mutex);
2511 tci = j * num_tc + tc;
2512 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) :
2515 map = expand_xps_map(map, j, index, is_rxqs_map);
2519 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2523 goto out_no_new_maps;
2526 /* Increment static keys at most once per type */
2527 static_key_slow_inc_cpuslocked(&xps_needed);
2529 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2532 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2534 /* copy maps belonging to foreign traffic classes */
2535 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) {
2536 /* fill in the new device map from the old device map */
2537 map = xmap_dereference(dev_maps->attr_map[tci]);
2538 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2541 /* We need to explicitly update tci as prevous loop
2542 * could break out early if dev_maps is NULL.
2544 tci = j * num_tc + tc;
2546 if (netif_attr_test_mask(j, mask, nr_ids) &&
2547 netif_attr_test_online(j, online_mask, nr_ids)) {
2548 /* add tx-queue to CPU/rx-queue maps */
2551 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2552 while ((pos < map->len) && (map->queues[pos] != index))
2555 if (pos == map->len)
2556 map->queues[map->len++] = index;
2559 if (numa_node_id == -2)
2560 numa_node_id = cpu_to_node(j);
2561 else if (numa_node_id != cpu_to_node(j))
2565 } else if (dev_maps) {
2566 /* fill in the new device map from the old device map */
2567 map = xmap_dereference(dev_maps->attr_map[tci]);
2568 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2571 /* copy maps belonging to foreign traffic classes */
2572 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2573 /* fill in the new device map from the old device map */
2574 map = xmap_dereference(dev_maps->attr_map[tci]);
2575 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2580 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps);
2582 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps);
2584 /* Cleanup old maps */
2586 goto out_no_old_maps;
2588 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2590 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2591 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2592 map = xmap_dereference(dev_maps->attr_map[tci]);
2593 if (map && map != new_map)
2594 kfree_rcu(map, rcu);
2598 kfree_rcu(dev_maps, rcu);
2601 dev_maps = new_dev_maps;
2606 /* update Tx queue numa node */
2607 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2608 (numa_node_id >= 0) ?
2609 numa_node_id : NUMA_NO_NODE);
2615 /* removes tx-queue from unused CPUs/rx-queues */
2616 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2618 for (i = tc, tci = j * num_tc; i--; tci++)
2619 active |= remove_xps_queue(dev_maps, tci, index);
2620 if (!netif_attr_test_mask(j, mask, nr_ids) ||
2621 !netif_attr_test_online(j, online_mask, nr_ids))
2622 active |= remove_xps_queue(dev_maps, tci, index);
2623 for (i = num_tc - tc, tci++; --i; tci++)
2624 active |= remove_xps_queue(dev_maps, tci, index);
2627 /* free map if not active */
2629 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2632 mutex_unlock(&xps_map_mutex);
2636 /* remove any maps that we added */
2637 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2639 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2640 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2642 xmap_dereference(dev_maps->attr_map[tci]) :
2644 if (new_map && new_map != map)
2649 mutex_unlock(&xps_map_mutex);
2651 kfree(new_dev_maps);
2654 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2656 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2662 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
2667 EXPORT_SYMBOL(netif_set_xps_queue);
2670 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2672 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2674 /* Unbind any subordinate channels */
2675 while (txq-- != &dev->_tx[0]) {
2677 netdev_unbind_sb_channel(dev, txq->sb_dev);
2681 void netdev_reset_tc(struct net_device *dev)
2684 netif_reset_xps_queues_gt(dev, 0);
2686 netdev_unbind_all_sb_channels(dev);
2688 /* Reset TC configuration of device */
2690 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2691 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2693 EXPORT_SYMBOL(netdev_reset_tc);
2695 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2697 if (tc >= dev->num_tc)
2701 netif_reset_xps_queues(dev, offset, count);
2703 dev->tc_to_txq[tc].count = count;
2704 dev->tc_to_txq[tc].offset = offset;
2707 EXPORT_SYMBOL(netdev_set_tc_queue);
2709 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2711 if (num_tc > TC_MAX_QUEUE)
2715 netif_reset_xps_queues_gt(dev, 0);
2717 netdev_unbind_all_sb_channels(dev);
2719 dev->num_tc = num_tc;
2722 EXPORT_SYMBOL(netdev_set_num_tc);
2724 void netdev_unbind_sb_channel(struct net_device *dev,
2725 struct net_device *sb_dev)
2727 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2730 netif_reset_xps_queues_gt(sb_dev, 0);
2732 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2733 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2735 while (txq-- != &dev->_tx[0]) {
2736 if (txq->sb_dev == sb_dev)
2740 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2742 int netdev_bind_sb_channel_queue(struct net_device *dev,
2743 struct net_device *sb_dev,
2744 u8 tc, u16 count, u16 offset)
2746 /* Make certain the sb_dev and dev are already configured */
2747 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2750 /* We cannot hand out queues we don't have */
2751 if ((offset + count) > dev->real_num_tx_queues)
2754 /* Record the mapping */
2755 sb_dev->tc_to_txq[tc].count = count;
2756 sb_dev->tc_to_txq[tc].offset = offset;
2758 /* Provide a way for Tx queue to find the tc_to_txq map or
2759 * XPS map for itself.
2762 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2766 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2768 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2770 /* Do not use a multiqueue device to represent a subordinate channel */
2771 if (netif_is_multiqueue(dev))
2774 /* We allow channels 1 - 32767 to be used for subordinate channels.
2775 * Channel 0 is meant to be "native" mode and used only to represent
2776 * the main root device. We allow writing 0 to reset the device back
2777 * to normal mode after being used as a subordinate channel.
2779 if (channel > S16_MAX)
2782 dev->num_tc = -channel;
2786 EXPORT_SYMBOL(netdev_set_sb_channel);
2789 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2790 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2792 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2797 disabling = txq < dev->real_num_tx_queues;
2799 if (txq < 1 || txq > dev->num_tx_queues)
2802 if (dev->reg_state == NETREG_REGISTERED ||
2803 dev->reg_state == NETREG_UNREGISTERING) {
2806 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2812 netif_setup_tc(dev, txq);
2814 dev->real_num_tx_queues = txq;
2818 qdisc_reset_all_tx_gt(dev, txq);
2820 netif_reset_xps_queues_gt(dev, txq);
2824 dev->real_num_tx_queues = txq;
2829 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2833 * netif_set_real_num_rx_queues - set actual number of RX queues used
2834 * @dev: Network device
2835 * @rxq: Actual number of RX queues
2837 * This must be called either with the rtnl_lock held or before
2838 * registration of the net device. Returns 0 on success, or a
2839 * negative error code. If called before registration, it always
2842 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2846 if (rxq < 1 || rxq > dev->num_rx_queues)
2849 if (dev->reg_state == NETREG_REGISTERED) {
2852 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2858 dev->real_num_rx_queues = rxq;
2861 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2865 * netif_get_num_default_rss_queues - default number of RSS queues
2867 * This routine should set an upper limit on the number of RSS queues
2868 * used by default by multiqueue devices.
2870 int netif_get_num_default_rss_queues(void)
2872 return is_kdump_kernel() ?
2873 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2875 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2877 static void __netif_reschedule(struct Qdisc *q)
2879 struct softnet_data *sd;
2880 unsigned long flags;
2882 local_irq_save(flags);
2883 sd = this_cpu_ptr(&softnet_data);
2884 q->next_sched = NULL;
2885 *sd->output_queue_tailp = q;
2886 sd->output_queue_tailp = &q->next_sched;
2887 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2888 local_irq_restore(flags);
2891 void __netif_schedule(struct Qdisc *q)
2893 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2894 __netif_reschedule(q);
2896 EXPORT_SYMBOL(__netif_schedule);
2898 struct dev_kfree_skb_cb {
2899 enum skb_free_reason reason;
2902 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2904 return (struct dev_kfree_skb_cb *)skb->cb;
2907 void netif_schedule_queue(struct netdev_queue *txq)
2910 if (!netif_xmit_stopped(txq)) {
2911 struct Qdisc *q = rcu_dereference(txq->qdisc);
2913 __netif_schedule(q);
2917 EXPORT_SYMBOL(netif_schedule_queue);
2919 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2921 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2925 q = rcu_dereference(dev_queue->qdisc);
2926 __netif_schedule(q);
2930 EXPORT_SYMBOL(netif_tx_wake_queue);
2932 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2934 unsigned long flags;
2939 if (likely(refcount_read(&skb->users) == 1)) {
2941 refcount_set(&skb->users, 0);
2942 } else if (likely(!refcount_dec_and_test(&skb->users))) {
2945 get_kfree_skb_cb(skb)->reason = reason;
2946 local_irq_save(flags);
2947 skb->next = __this_cpu_read(softnet_data.completion_queue);
2948 __this_cpu_write(softnet_data.completion_queue, skb);
2949 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2950 local_irq_restore(flags);
2952 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2954 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2956 if (in_irq() || irqs_disabled())
2957 __dev_kfree_skb_irq(skb, reason);
2961 EXPORT_SYMBOL(__dev_kfree_skb_any);
2965 * netif_device_detach - mark device as removed
2966 * @dev: network device
2968 * Mark device as removed from system and therefore no longer available.
2970 void netif_device_detach(struct net_device *dev)
2972 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2973 netif_running(dev)) {
2974 netif_tx_stop_all_queues(dev);
2977 EXPORT_SYMBOL(netif_device_detach);
2980 * netif_device_attach - mark device as attached
2981 * @dev: network device
2983 * Mark device as attached from system and restart if needed.
2985 void netif_device_attach(struct net_device *dev)
2987 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2988 netif_running(dev)) {
2989 netif_tx_wake_all_queues(dev);
2990 __netdev_watchdog_up(dev);
2993 EXPORT_SYMBOL(netif_device_attach);
2996 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2997 * to be used as a distribution range.
2999 static u16 skb_tx_hash(const struct net_device *dev,
3000 const struct net_device *sb_dev,
3001 struct sk_buff *skb)
3005 u16 qcount = dev->real_num_tx_queues;
3008 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3010 qoffset = sb_dev->tc_to_txq[tc].offset;
3011 qcount = sb_dev->tc_to_txq[tc].count;
3014 if (skb_rx_queue_recorded(skb)) {
3015 hash = skb_get_rx_queue(skb);
3016 while (unlikely(hash >= qcount))
3018 return hash + qoffset;
3021 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3024 static void skb_warn_bad_offload(const struct sk_buff *skb)
3026 static const netdev_features_t null_features;
3027 struct net_device *dev = skb->dev;
3028 const char *name = "";
3030 if (!net_ratelimit())
3034 if (dev->dev.parent)
3035 name = dev_driver_string(dev->dev.parent);
3037 name = netdev_name(dev);
3039 skb_dump(KERN_WARNING, skb, false);
3040 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3041 name, dev ? &dev->features : &null_features,
3042 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3046 * Invalidate hardware checksum when packet is to be mangled, and
3047 * complete checksum manually on outgoing path.
3049 int skb_checksum_help(struct sk_buff *skb)
3052 int ret = 0, offset;
3054 if (skb->ip_summed == CHECKSUM_COMPLETE)
3055 goto out_set_summed;
3057 if (unlikely(skb_shinfo(skb)->gso_size)) {
3058 skb_warn_bad_offload(skb);
3062 /* Before computing a checksum, we should make sure no frag could
3063 * be modified by an external entity : checksum could be wrong.
3065 if (skb_has_shared_frag(skb)) {
3066 ret = __skb_linearize(skb);
3071 offset = skb_checksum_start_offset(skb);
3072 BUG_ON(offset >= skb_headlen(skb));
3073 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3075 offset += skb->csum_offset;
3076 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
3078 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3082 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3084 skb->ip_summed = CHECKSUM_NONE;
3088 EXPORT_SYMBOL(skb_checksum_help);
3090 int skb_crc32c_csum_help(struct sk_buff *skb)
3093 int ret = 0, offset, start;
3095 if (skb->ip_summed != CHECKSUM_PARTIAL)
3098 if (unlikely(skb_is_gso(skb)))
3101 /* Before computing a checksum, we should make sure no frag could
3102 * be modified by an external entity : checksum could be wrong.
3104 if (unlikely(skb_has_shared_frag(skb))) {
3105 ret = __skb_linearize(skb);
3109 start = skb_checksum_start_offset(skb);
3110 offset = start + offsetof(struct sctphdr, checksum);
3111 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3116 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3120 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3121 skb->len - start, ~(__u32)0,
3123 *(__le32 *)(skb->data + offset) = crc32c_csum;
3124 skb->ip_summed = CHECKSUM_NONE;
3125 skb->csum_not_inet = 0;
3130 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3132 __be16 type = skb->protocol;
3134 /* Tunnel gso handlers can set protocol to ethernet. */
3135 if (type == htons(ETH_P_TEB)) {
3138 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3141 eth = (struct ethhdr *)skb->data;
3142 type = eth->h_proto;
3145 return __vlan_get_protocol(skb, type, depth);
3149 * skb_mac_gso_segment - mac layer segmentation handler.
3150 * @skb: buffer to segment
3151 * @features: features for the output path (see dev->features)
3153 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3154 netdev_features_t features)
3156 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3157 struct packet_offload *ptype;
3158 int vlan_depth = skb->mac_len;
3159 __be16 type = skb_network_protocol(skb, &vlan_depth);
3161 if (unlikely(!type))
3162 return ERR_PTR(-EINVAL);
3164 __skb_pull(skb, vlan_depth);
3167 list_for_each_entry_rcu(ptype, &offload_base, list) {
3168 if (ptype->type == type && ptype->callbacks.gso_segment) {
3169 segs = ptype->callbacks.gso_segment(skb, features);
3175 __skb_push(skb, skb->data - skb_mac_header(skb));
3179 EXPORT_SYMBOL(skb_mac_gso_segment);
3182 /* openvswitch calls this on rx path, so we need a different check.
3184 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3187 return skb->ip_summed != CHECKSUM_PARTIAL &&
3188 skb->ip_summed != CHECKSUM_UNNECESSARY;
3190 return skb->ip_summed == CHECKSUM_NONE;
3194 * __skb_gso_segment - Perform segmentation on skb.
3195 * @skb: buffer to segment
3196 * @features: features for the output path (see dev->features)
3197 * @tx_path: whether it is called in TX path
3199 * This function segments the given skb and returns a list of segments.
3201 * It may return NULL if the skb requires no segmentation. This is
3202 * only possible when GSO is used for verifying header integrity.
3204 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
3206 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3207 netdev_features_t features, bool tx_path)
3209 struct sk_buff *segs;
3211 if (unlikely(skb_needs_check(skb, tx_path))) {
3214 /* We're going to init ->check field in TCP or UDP header */
3215 err = skb_cow_head(skb, 0);
3217 return ERR_PTR(err);
3220 /* Only report GSO partial support if it will enable us to
3221 * support segmentation on this frame without needing additional
3224 if (features & NETIF_F_GSO_PARTIAL) {
3225 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3226 struct net_device *dev = skb->dev;
3228 partial_features |= dev->features & dev->gso_partial_features;
3229 if (!skb_gso_ok(skb, features | partial_features))
3230 features &= ~NETIF_F_GSO_PARTIAL;
3233 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
3234 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3236 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3237 SKB_GSO_CB(skb)->encap_level = 0;
3239 skb_reset_mac_header(skb);
3240 skb_reset_mac_len(skb);
3242 segs = skb_mac_gso_segment(skb, features);
3244 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3245 skb_warn_bad_offload(skb);
3249 EXPORT_SYMBOL(__skb_gso_segment);
3251 /* Take action when hardware reception checksum errors are detected. */
3253 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3255 if (net_ratelimit()) {
3256 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3257 skb_dump(KERN_ERR, skb, true);
3261 EXPORT_SYMBOL(netdev_rx_csum_fault);
3264 /* XXX: check that highmem exists at all on the given machine. */
3265 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3267 #ifdef CONFIG_HIGHMEM
3270 if (!(dev->features & NETIF_F_HIGHDMA)) {
3271 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3272 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3274 if (PageHighMem(skb_frag_page(frag)))
3282 /* If MPLS offload request, verify we are testing hardware MPLS features
3283 * instead of standard features for the netdev.
3285 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3286 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3287 netdev_features_t features,
3290 if (eth_p_mpls(type))
3291 features &= skb->dev->mpls_features;
3296 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3297 netdev_features_t features,
3304 static netdev_features_t harmonize_features(struct sk_buff *skb,
3305 netdev_features_t features)
3310 type = skb_network_protocol(skb, &tmp);
3311 features = net_mpls_features(skb, features, type);
3313 if (skb->ip_summed != CHECKSUM_NONE &&
3314 !can_checksum_protocol(features, type)) {
3315 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3317 if (illegal_highdma(skb->dev, skb))
3318 features &= ~NETIF_F_SG;
3323 netdev_features_t passthru_features_check(struct sk_buff *skb,
3324 struct net_device *dev,
3325 netdev_features_t features)
3329 EXPORT_SYMBOL(passthru_features_check);
3331 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3332 struct net_device *dev,
3333 netdev_features_t features)
3335 return vlan_features_check(skb, features);
3338 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3339 struct net_device *dev,
3340 netdev_features_t features)
3342 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3344 if (gso_segs > dev->gso_max_segs)
3345 return features & ~NETIF_F_GSO_MASK;
3347 /* Support for GSO partial features requires software
3348 * intervention before we can actually process the packets
3349 * so we need to strip support for any partial features now
3350 * and we can pull them back in after we have partially
3351 * segmented the frame.
3353 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3354 features &= ~dev->gso_partial_features;
3356 /* Make sure to clear the IPv4 ID mangling feature if the
3357 * IPv4 header has the potential to be fragmented.
3359 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3360 struct iphdr *iph = skb->encapsulation ?
3361 inner_ip_hdr(skb) : ip_hdr(skb);
3363 if (!(iph->frag_off & htons(IP_DF)))
3364 features &= ~NETIF_F_TSO_MANGLEID;
3370 netdev_features_t netif_skb_features(struct sk_buff *skb)
3372 struct net_device *dev = skb->dev;
3373 netdev_features_t features = dev->features;
3375 if (skb_is_gso(skb))
3376 features = gso_features_check(skb, dev, features);
3378 /* If encapsulation offload request, verify we are testing
3379 * hardware encapsulation features instead of standard
3380 * features for the netdev
3382 if (skb->encapsulation)
3383 features &= dev->hw_enc_features;
3385 if (skb_vlan_tagged(skb))
3386 features = netdev_intersect_features(features,
3387 dev->vlan_features |
3388 NETIF_F_HW_VLAN_CTAG_TX |
3389 NETIF_F_HW_VLAN_STAG_TX);
3391 if (dev->netdev_ops->ndo_features_check)
3392 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3395 features &= dflt_features_check(skb, dev, features);
3397 return harmonize_features(skb, features);
3399 EXPORT_SYMBOL(netif_skb_features);
3401 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3402 struct netdev_queue *txq, bool more)
3407 if (dev_nit_active(dev))
3408 dev_queue_xmit_nit(skb, dev);
3411 trace_net_dev_start_xmit(skb, dev);
3412 rc = netdev_start_xmit(skb, dev, txq, more);
3413 trace_net_dev_xmit(skb, rc, dev, len);
3418 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3419 struct netdev_queue *txq, int *ret)
3421 struct sk_buff *skb = first;
3422 int rc = NETDEV_TX_OK;
3425 struct sk_buff *next = skb->next;
3427 skb_mark_not_on_list(skb);
3428 rc = xmit_one(skb, dev, txq, next != NULL);
3429 if (unlikely(!dev_xmit_complete(rc))) {
3435 if (netif_tx_queue_stopped(txq) && skb) {
3436 rc = NETDEV_TX_BUSY;
3446 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3447 netdev_features_t features)
3449 if (skb_vlan_tag_present(skb) &&
3450 !vlan_hw_offload_capable(features, skb->vlan_proto))
3451 skb = __vlan_hwaccel_push_inside(skb);
3455 int skb_csum_hwoffload_help(struct sk_buff *skb,
3456 const netdev_features_t features)
3458 if (unlikely(skb->csum_not_inet))
3459 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3460 skb_crc32c_csum_help(skb);
3462 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3464 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3466 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3468 netdev_features_t features;
3470 features = netif_skb_features(skb);
3471 skb = validate_xmit_vlan(skb, features);
3475 skb = sk_validate_xmit_skb(skb, dev);
3479 if (netif_needs_gso(skb, features)) {
3480 struct sk_buff *segs;
3482 segs = skb_gso_segment(skb, features);
3490 if (skb_needs_linearize(skb, features) &&
3491 __skb_linearize(skb))
3494 /* If packet is not checksummed and device does not
3495 * support checksumming for this protocol, complete
3496 * checksumming here.
3498 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3499 if (skb->encapsulation)
3500 skb_set_inner_transport_header(skb,
3501 skb_checksum_start_offset(skb));
3503 skb_set_transport_header(skb,
3504 skb_checksum_start_offset(skb));
3505 if (skb_csum_hwoffload_help(skb, features))
3510 skb = validate_xmit_xfrm(skb, features, again);
3517 atomic_long_inc(&dev->tx_dropped);
3521 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3523 struct sk_buff *next, *head = NULL, *tail;
3525 for (; skb != NULL; skb = next) {
3527 skb_mark_not_on_list(skb);
3529 /* in case skb wont be segmented, point to itself */
3532 skb = validate_xmit_skb(skb, dev, again);
3540 /* If skb was segmented, skb->prev points to
3541 * the last segment. If not, it still contains skb.
3547 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3549 static void qdisc_pkt_len_init(struct sk_buff *skb)
3551 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3553 qdisc_skb_cb(skb)->pkt_len = skb->len;
3555 /* To get more precise estimation of bytes sent on wire,
3556 * we add to pkt_len the headers size of all segments
3558 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3559 unsigned int hdr_len;
3560 u16 gso_segs = shinfo->gso_segs;
3562 /* mac layer + network layer */
3563 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3565 /* + transport layer */
3566 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3567 const struct tcphdr *th;
3568 struct tcphdr _tcphdr;
3570 th = skb_header_pointer(skb, skb_transport_offset(skb),
3571 sizeof(_tcphdr), &_tcphdr);
3573 hdr_len += __tcp_hdrlen(th);
3575 struct udphdr _udphdr;
3577 if (skb_header_pointer(skb, skb_transport_offset(skb),
3578 sizeof(_udphdr), &_udphdr))
3579 hdr_len += sizeof(struct udphdr);
3582 if (shinfo->gso_type & SKB_GSO_DODGY)
3583 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3586 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3590 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3591 struct net_device *dev,
3592 struct netdev_queue *txq)
3594 spinlock_t *root_lock = qdisc_lock(q);
3595 struct sk_buff *to_free = NULL;
3599 qdisc_calculate_pkt_len(skb, q);
3601 if (q->flags & TCQ_F_NOLOCK) {
3602 if ((q->flags & TCQ_F_CAN_BYPASS) && READ_ONCE(q->empty) &&
3603 qdisc_run_begin(q)) {
3604 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
3606 __qdisc_drop(skb, &to_free);
3610 qdisc_bstats_cpu_update(q, skb);
3612 rc = NET_XMIT_SUCCESS;
3613 if (sch_direct_xmit(skb, q, dev, txq, NULL, true))
3619 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3623 if (unlikely(to_free))
3624 kfree_skb_list(to_free);
3629 * Heuristic to force contended enqueues to serialize on a
3630 * separate lock before trying to get qdisc main lock.
3631 * This permits qdisc->running owner to get the lock more
3632 * often and dequeue packets faster.
3634 contended = qdisc_is_running(q);
3635 if (unlikely(contended))
3636 spin_lock(&q->busylock);
3638 spin_lock(root_lock);
3639 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3640 __qdisc_drop(skb, &to_free);
3642 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3643 qdisc_run_begin(q)) {
3645 * This is a work-conserving queue; there are no old skbs
3646 * waiting to be sent out; and the qdisc is not running -
3647 * xmit the skb directly.
3650 qdisc_bstats_update(q, skb);
3652 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3653 if (unlikely(contended)) {
3654 spin_unlock(&q->busylock);
3661 rc = NET_XMIT_SUCCESS;
3663 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3664 if (qdisc_run_begin(q)) {
3665 if (unlikely(contended)) {
3666 spin_unlock(&q->busylock);
3673 spin_unlock(root_lock);
3674 if (unlikely(to_free))
3675 kfree_skb_list(to_free);
3676 if (unlikely(contended))
3677 spin_unlock(&q->busylock);
3681 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3682 static void skb_update_prio(struct sk_buff *skb)
3684 const struct netprio_map *map;
3685 const struct sock *sk;
3686 unsigned int prioidx;
3690 map = rcu_dereference_bh(skb->dev->priomap);
3693 sk = skb_to_full_sk(skb);
3697 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3699 if (prioidx < map->priomap_len)
3700 skb->priority = map->priomap[prioidx];
3703 #define skb_update_prio(skb)
3707 * dev_loopback_xmit - loop back @skb
3708 * @net: network namespace this loopback is happening in
3709 * @sk: sk needed to be a netfilter okfn
3710 * @skb: buffer to transmit
3712 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3714 skb_reset_mac_header(skb);
3715 __skb_pull(skb, skb_network_offset(skb));
3716 skb->pkt_type = PACKET_LOOPBACK;
3717 skb->ip_summed = CHECKSUM_UNNECESSARY;
3718 WARN_ON(!skb_dst(skb));
3723 EXPORT_SYMBOL(dev_loopback_xmit);
3725 #ifdef CONFIG_NET_EGRESS
3726 static struct sk_buff *
3727 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3729 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3730 struct tcf_result cl_res;
3735 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3736 mini_qdisc_bstats_cpu_update(miniq, skb);
3738 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3740 case TC_ACT_RECLASSIFY:
3741 skb->tc_index = TC_H_MIN(cl_res.classid);
3744 mini_qdisc_qstats_cpu_drop(miniq);
3745 *ret = NET_XMIT_DROP;
3751 *ret = NET_XMIT_SUCCESS;
3754 case TC_ACT_REDIRECT:
3755 /* No need to push/pop skb's mac_header here on egress! */
3756 skb_do_redirect(skb);
3757 *ret = NET_XMIT_SUCCESS;
3765 #endif /* CONFIG_NET_EGRESS */
3768 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3769 struct xps_dev_maps *dev_maps, unsigned int tci)
3771 struct xps_map *map;
3772 int queue_index = -1;
3776 tci += netdev_get_prio_tc_map(dev, skb->priority);
3779 map = rcu_dereference(dev_maps->attr_map[tci]);
3782 queue_index = map->queues[0];
3784 queue_index = map->queues[reciprocal_scale(
3785 skb_get_hash(skb), map->len)];
3786 if (unlikely(queue_index >= dev->real_num_tx_queues))
3793 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3794 struct sk_buff *skb)
3797 struct xps_dev_maps *dev_maps;
3798 struct sock *sk = skb->sk;
3799 int queue_index = -1;
3801 if (!static_key_false(&xps_needed))
3805 if (!static_key_false(&xps_rxqs_needed))
3808 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map);
3810 int tci = sk_rx_queue_get(sk);
3812 if (tci >= 0 && tci < dev->num_rx_queues)
3813 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3818 if (queue_index < 0) {
3819 dev_maps = rcu_dereference(sb_dev->xps_cpus_map);
3821 unsigned int tci = skb->sender_cpu - 1;
3823 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3835 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3836 struct net_device *sb_dev)
3840 EXPORT_SYMBOL(dev_pick_tx_zero);
3842 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
3843 struct net_device *sb_dev)
3845 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3847 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3849 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
3850 struct net_device *sb_dev)
3852 struct sock *sk = skb->sk;
3853 int queue_index = sk_tx_queue_get(sk);
3855 sb_dev = sb_dev ? : dev;
3857 if (queue_index < 0 || skb->ooo_okay ||
3858 queue_index >= dev->real_num_tx_queues) {
3859 int new_index = get_xps_queue(dev, sb_dev, skb);
3862 new_index = skb_tx_hash(dev, sb_dev, skb);
3864 if (queue_index != new_index && sk &&
3866 rcu_access_pointer(sk->sk_dst_cache))
3867 sk_tx_queue_set(sk, new_index);
3869 queue_index = new_index;
3874 EXPORT_SYMBOL(netdev_pick_tx);
3876 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
3877 struct sk_buff *skb,
3878 struct net_device *sb_dev)
3880 int queue_index = 0;
3883 u32 sender_cpu = skb->sender_cpu - 1;
3885 if (sender_cpu >= (u32)NR_CPUS)
3886 skb->sender_cpu = raw_smp_processor_id() + 1;
3889 if (dev->real_num_tx_queues != 1) {
3890 const struct net_device_ops *ops = dev->netdev_ops;
3892 if (ops->ndo_select_queue)
3893 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
3895 queue_index = netdev_pick_tx(dev, skb, sb_dev);
3897 queue_index = netdev_cap_txqueue(dev, queue_index);
3900 skb_set_queue_mapping(skb, queue_index);
3901 return netdev_get_tx_queue(dev, queue_index);
3905 * __dev_queue_xmit - transmit a buffer
3906 * @skb: buffer to transmit
3907 * @sb_dev: suboordinate device used for L2 forwarding offload
3909 * Queue a buffer for transmission to a network device. The caller must
3910 * have set the device and priority and built the buffer before calling
3911 * this function. The function can be called from an interrupt.
3913 * A negative errno code is returned on a failure. A success does not
3914 * guarantee the frame will be transmitted as it may be dropped due
3915 * to congestion or traffic shaping.
3917 * -----------------------------------------------------------------------------------
3918 * I notice this method can also return errors from the queue disciplines,
3919 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3922 * Regardless of the return value, the skb is consumed, so it is currently
3923 * difficult to retry a send to this method. (You can bump the ref count
3924 * before sending to hold a reference for retry if you are careful.)
3926 * When calling this method, interrupts MUST be enabled. This is because
3927 * the BH enable code must have IRQs enabled so that it will not deadlock.
3930 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
3932 struct net_device *dev = skb->dev;
3933 struct netdev_queue *txq;
3938 skb_reset_mac_header(skb);
3940 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3941 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3943 /* Disable soft irqs for various locks below. Also
3944 * stops preemption for RCU.
3948 skb_update_prio(skb);
3950 qdisc_pkt_len_init(skb);
3951 #ifdef CONFIG_NET_CLS_ACT
3952 skb->tc_at_ingress = 0;
3953 # ifdef CONFIG_NET_EGRESS
3954 if (static_branch_unlikely(&egress_needed_key)) {
3955 skb = sch_handle_egress(skb, &rc, dev);
3961 /* If device/qdisc don't need skb->dst, release it right now while
3962 * its hot in this cpu cache.
3964 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3969 txq = netdev_core_pick_tx(dev, skb, sb_dev);
3970 q = rcu_dereference_bh(txq->qdisc);
3972 trace_net_dev_queue(skb);
3974 rc = __dev_xmit_skb(skb, q, dev, txq);
3978 /* The device has no queue. Common case for software devices:
3979 * loopback, all the sorts of tunnels...
3981 * Really, it is unlikely that netif_tx_lock protection is necessary
3982 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
3984 * However, it is possible, that they rely on protection
3987 * Check this and shot the lock. It is not prone from deadlocks.
3988 *Either shot noqueue qdisc, it is even simpler 8)
3990 if (dev->flags & IFF_UP) {
3991 int cpu = smp_processor_id(); /* ok because BHs are off */
3993 if (txq->xmit_lock_owner != cpu) {
3994 if (dev_xmit_recursion())
3995 goto recursion_alert;
3997 skb = validate_xmit_skb(skb, dev, &again);
4001 HARD_TX_LOCK(dev, txq, cpu);
4003 if (!netif_xmit_stopped(txq)) {
4004 dev_xmit_recursion_inc();
4005 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4006 dev_xmit_recursion_dec();
4007 if (dev_xmit_complete(rc)) {
4008 HARD_TX_UNLOCK(dev, txq);
4012 HARD_TX_UNLOCK(dev, txq);
4013 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4016 /* Recursion is detected! It is possible,
4020 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4026 rcu_read_unlock_bh();
4028 atomic_long_inc(&dev->tx_dropped);
4029 kfree_skb_list(skb);
4032 rcu_read_unlock_bh();
4036 int dev_queue_xmit(struct sk_buff *skb)
4038 return __dev_queue_xmit(skb, NULL);
4040 EXPORT_SYMBOL(dev_queue_xmit);
4042 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
4044 return __dev_queue_xmit(skb, sb_dev);
4046 EXPORT_SYMBOL(dev_queue_xmit_accel);
4048 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4050 struct net_device *dev = skb->dev;
4051 struct sk_buff *orig_skb = skb;
4052 struct netdev_queue *txq;
4053 int ret = NETDEV_TX_BUSY;
4056 if (unlikely(!netif_running(dev) ||
4057 !netif_carrier_ok(dev)))
4060 skb = validate_xmit_skb_list(skb, dev, &again);
4061 if (skb != orig_skb)
4064 skb_set_queue_mapping(skb, queue_id);
4065 txq = skb_get_tx_queue(dev, skb);
4069 HARD_TX_LOCK(dev, txq, smp_processor_id());
4070 if (!netif_xmit_frozen_or_drv_stopped(txq))
4071 ret = netdev_start_xmit(skb, dev, txq, false);
4072 HARD_TX_UNLOCK(dev, txq);
4076 if (!dev_xmit_complete(ret))
4081 atomic_long_inc(&dev->tx_dropped);
4082 kfree_skb_list(skb);
4083 return NET_XMIT_DROP;
4085 EXPORT_SYMBOL(dev_direct_xmit);
4087 /*************************************************************************
4089 *************************************************************************/
4091 int netdev_max_backlog __read_mostly = 1000;
4092 EXPORT_SYMBOL(netdev_max_backlog);
4094 int netdev_tstamp_prequeue __read_mostly = 1;
4095 int netdev_budget __read_mostly = 300;
4096 unsigned int __read_mostly netdev_budget_usecs = 2000;
4097 int weight_p __read_mostly = 64; /* old backlog weight */
4098 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4099 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4100 int dev_rx_weight __read_mostly = 64;
4101 int dev_tx_weight __read_mostly = 64;
4102 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
4103 int gro_normal_batch __read_mostly = 8;
4105 /* Called with irq disabled */
4106 static inline void ____napi_schedule(struct softnet_data *sd,
4107 struct napi_struct *napi)
4109 list_add_tail(&napi->poll_list, &sd->poll_list);
4110 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4115 /* One global table that all flow-based protocols share. */
4116 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4117 EXPORT_SYMBOL(rps_sock_flow_table);
4118 u32 rps_cpu_mask __read_mostly;
4119 EXPORT_SYMBOL(rps_cpu_mask);
4121 struct static_key_false rps_needed __read_mostly;
4122 EXPORT_SYMBOL(rps_needed);
4123 struct static_key_false rfs_needed __read_mostly;
4124 EXPORT_SYMBOL(rfs_needed);
4126 static struct rps_dev_flow *
4127 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4128 struct rps_dev_flow *rflow, u16 next_cpu)
4130 if (next_cpu < nr_cpu_ids) {
4131 #ifdef CONFIG_RFS_ACCEL
4132 struct netdev_rx_queue *rxqueue;
4133 struct rps_dev_flow_table *flow_table;
4134 struct rps_dev_flow *old_rflow;
4139 /* Should we steer this flow to a different hardware queue? */
4140 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4141 !(dev->features & NETIF_F_NTUPLE))
4143 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4144 if (rxq_index == skb_get_rx_queue(skb))
4147 rxqueue = dev->_rx + rxq_index;
4148 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4151 flow_id = skb_get_hash(skb) & flow_table->mask;
4152 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4153 rxq_index, flow_id);
4157 rflow = &flow_table->flows[flow_id];
4159 if (old_rflow->filter == rflow->filter)
4160 old_rflow->filter = RPS_NO_FILTER;
4164 per_cpu(softnet_data, next_cpu).input_queue_head;
4167 rflow->cpu = next_cpu;
4172 * get_rps_cpu is called from netif_receive_skb and returns the target
4173 * CPU from the RPS map of the receiving queue for a given skb.
4174 * rcu_read_lock must be held on entry.
4176 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4177 struct rps_dev_flow **rflowp)
4179 const struct rps_sock_flow_table *sock_flow_table;
4180 struct netdev_rx_queue *rxqueue = dev->_rx;
4181 struct rps_dev_flow_table *flow_table;
4182 struct rps_map *map;
4187 if (skb_rx_queue_recorded(skb)) {
4188 u16 index = skb_get_rx_queue(skb);
4190 if (unlikely(index >= dev->real_num_rx_queues)) {
4191 WARN_ONCE(dev->real_num_rx_queues > 1,
4192 "%s received packet on queue %u, but number "
4193 "of RX queues is %u\n",
4194 dev->name, index, dev->real_num_rx_queues);
4200 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4202 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4203 map = rcu_dereference(rxqueue->rps_map);
4204 if (!flow_table && !map)
4207 skb_reset_network_header(skb);
4208 hash = skb_get_hash(skb);
4212 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4213 if (flow_table && sock_flow_table) {
4214 struct rps_dev_flow *rflow;
4218 /* First check into global flow table if there is a match */
4219 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4220 if ((ident ^ hash) & ~rps_cpu_mask)
4223 next_cpu = ident & rps_cpu_mask;
4225 /* OK, now we know there is a match,
4226 * we can look at the local (per receive queue) flow table
4228 rflow = &flow_table->flows[hash & flow_table->mask];
4232 * If the desired CPU (where last recvmsg was done) is
4233 * different from current CPU (one in the rx-queue flow
4234 * table entry), switch if one of the following holds:
4235 * - Current CPU is unset (>= nr_cpu_ids).
4236 * - Current CPU is offline.
4237 * - The current CPU's queue tail has advanced beyond the
4238 * last packet that was enqueued using this table entry.
4239 * This guarantees that all previous packets for the flow
4240 * have been dequeued, thus preserving in order delivery.
4242 if (unlikely(tcpu != next_cpu) &&
4243 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4244 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4245 rflow->last_qtail)) >= 0)) {
4247 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4250 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4260 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4261 if (cpu_online(tcpu)) {
4271 #ifdef CONFIG_RFS_ACCEL
4274 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4275 * @dev: Device on which the filter was set
4276 * @rxq_index: RX queue index
4277 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4278 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4280 * Drivers that implement ndo_rx_flow_steer() should periodically call
4281 * this function for each installed filter and remove the filters for
4282 * which it returns %true.
4284 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4285 u32 flow_id, u16 filter_id)
4287 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4288 struct rps_dev_flow_table *flow_table;
4289 struct rps_dev_flow *rflow;
4294 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4295 if (flow_table && flow_id <= flow_table->mask) {
4296 rflow = &flow_table->flows[flow_id];
4297 cpu = READ_ONCE(rflow->cpu);
4298 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4299 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4300 rflow->last_qtail) <
4301 (int)(10 * flow_table->mask)))
4307 EXPORT_SYMBOL(rps_may_expire_flow);
4309 #endif /* CONFIG_RFS_ACCEL */
4311 /* Called from hardirq (IPI) context */
4312 static void rps_trigger_softirq(void *data)
4314 struct softnet_data *sd = data;
4316 ____napi_schedule(sd, &sd->backlog);
4320 #endif /* CONFIG_RPS */
4323 * Check if this softnet_data structure is another cpu one
4324 * If yes, queue it to our IPI list and return 1
4327 static int rps_ipi_queued(struct softnet_data *sd)
4330 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4333 sd->rps_ipi_next = mysd->rps_ipi_list;
4334 mysd->rps_ipi_list = sd;
4336 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4339 #endif /* CONFIG_RPS */
4343 #ifdef CONFIG_NET_FLOW_LIMIT
4344 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4347 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4349 #ifdef CONFIG_NET_FLOW_LIMIT
4350 struct sd_flow_limit *fl;
4351 struct softnet_data *sd;
4352 unsigned int old_flow, new_flow;
4354 if (qlen < (netdev_max_backlog >> 1))
4357 sd = this_cpu_ptr(&softnet_data);
4360 fl = rcu_dereference(sd->flow_limit);
4362 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4363 old_flow = fl->history[fl->history_head];
4364 fl->history[fl->history_head] = new_flow;
4367 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4369 if (likely(fl->buckets[old_flow]))
4370 fl->buckets[old_flow]--;
4372 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4384 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4385 * queue (may be a remote CPU queue).
4387 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4388 unsigned int *qtail)
4390 struct softnet_data *sd;
4391 unsigned long flags;
4394 sd = &per_cpu(softnet_data, cpu);
4396 local_irq_save(flags);
4399 if (!netif_running(skb->dev))
4401 qlen = skb_queue_len(&sd->input_pkt_queue);
4402 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4405 __skb_queue_tail(&sd->input_pkt_queue, skb);
4406 input_queue_tail_incr_save(sd, qtail);
4408 local_irq_restore(flags);
4409 return NET_RX_SUCCESS;
4412 /* Schedule NAPI for backlog device
4413 * We can use non atomic operation since we own the queue lock
4415 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4416 if (!rps_ipi_queued(sd))
4417 ____napi_schedule(sd, &sd->backlog);
4426 local_irq_restore(flags);
4428 atomic_long_inc(&skb->dev->rx_dropped);
4433 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4435 struct net_device *dev = skb->dev;
4436 struct netdev_rx_queue *rxqueue;
4440 if (skb_rx_queue_recorded(skb)) {
4441 u16 index = skb_get_rx_queue(skb);
4443 if (unlikely(index >= dev->real_num_rx_queues)) {
4444 WARN_ONCE(dev->real_num_rx_queues > 1,
4445 "%s received packet on queue %u, but number "
4446 "of RX queues is %u\n",
4447 dev->name, index, dev->real_num_rx_queues);
4449 return rxqueue; /* Return first rxqueue */
4456 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4457 struct xdp_buff *xdp,
4458 struct bpf_prog *xdp_prog)
4460 struct netdev_rx_queue *rxqueue;
4461 void *orig_data, *orig_data_end;
4462 u32 metalen, act = XDP_DROP;
4463 __be16 orig_eth_type;
4469 /* Reinjected packets coming from act_mirred or similar should
4470 * not get XDP generic processing.
4472 if (skb_cloned(skb) || skb_is_tc_redirected(skb))
4475 /* XDP packets must be linear and must have sufficient headroom
4476 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4477 * native XDP provides, thus we need to do it here as well.
4479 if (skb_is_nonlinear(skb) ||
4480 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4481 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4482 int troom = skb->tail + skb->data_len - skb->end;
4484 /* In case we have to go down the path and also linearize,
4485 * then lets do the pskb_expand_head() work just once here.
4487 if (pskb_expand_head(skb,
4488 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4489 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4491 if (skb_linearize(skb))
4495 /* The XDP program wants to see the packet starting at the MAC
4498 mac_len = skb->data - skb_mac_header(skb);
4499 hlen = skb_headlen(skb) + mac_len;
4500 xdp->data = skb->data - mac_len;
4501 xdp->data_meta = xdp->data;
4502 xdp->data_end = xdp->data + hlen;
4503 xdp->data_hard_start = skb->data - skb_headroom(skb);
4504 orig_data_end = xdp->data_end;
4505 orig_data = xdp->data;
4506 eth = (struct ethhdr *)xdp->data;
4507 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4508 orig_eth_type = eth->h_proto;
4510 rxqueue = netif_get_rxqueue(skb);
4511 xdp->rxq = &rxqueue->xdp_rxq;
4513 act = bpf_prog_run_xdp(xdp_prog, xdp);
4515 /* check if bpf_xdp_adjust_head was used */
4516 off = xdp->data - orig_data;
4519 __skb_pull(skb, off);
4521 __skb_push(skb, -off);
4523 skb->mac_header += off;
4524 skb_reset_network_header(skb);
4527 /* check if bpf_xdp_adjust_tail was used. it can only "shrink"
4530 off = orig_data_end - xdp->data_end;
4532 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4537 /* check if XDP changed eth hdr such SKB needs update */
4538 eth = (struct ethhdr *)xdp->data;
4539 if ((orig_eth_type != eth->h_proto) ||
4540 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4541 __skb_push(skb, ETH_HLEN);
4542 skb->protocol = eth_type_trans(skb, skb->dev);
4548 __skb_push(skb, mac_len);
4551 metalen = xdp->data - xdp->data_meta;
4553 skb_metadata_set(skb, metalen);
4556 bpf_warn_invalid_xdp_action(act);
4559 trace_xdp_exception(skb->dev, xdp_prog, act);
4570 /* When doing generic XDP we have to bypass the qdisc layer and the
4571 * network taps in order to match in-driver-XDP behavior.
4573 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4575 struct net_device *dev = skb->dev;
4576 struct netdev_queue *txq;
4577 bool free_skb = true;
4580 txq = netdev_core_pick_tx(dev, skb, NULL);
4581 cpu = smp_processor_id();
4582 HARD_TX_LOCK(dev, txq, cpu);
4583 if (!netif_xmit_stopped(txq)) {
4584 rc = netdev_start_xmit(skb, dev, txq, 0);
4585 if (dev_xmit_complete(rc))
4588 HARD_TX_UNLOCK(dev, txq);
4590 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4594 EXPORT_SYMBOL_GPL(generic_xdp_tx);
4596 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4598 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4601 struct xdp_buff xdp;
4605 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4606 if (act != XDP_PASS) {
4609 err = xdp_do_generic_redirect(skb->dev, skb,
4615 generic_xdp_tx(skb, xdp_prog);
4626 EXPORT_SYMBOL_GPL(do_xdp_generic);
4628 static int netif_rx_internal(struct sk_buff *skb)
4632 net_timestamp_check(netdev_tstamp_prequeue, skb);
4634 trace_netif_rx(skb);
4637 if (static_branch_unlikely(&rps_needed)) {
4638 struct rps_dev_flow voidflow, *rflow = &voidflow;
4644 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4646 cpu = smp_processor_id();
4648 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4657 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4664 * netif_rx - post buffer to the network code
4665 * @skb: buffer to post
4667 * This function receives a packet from a device driver and queues it for
4668 * the upper (protocol) levels to process. It always succeeds. The buffer
4669 * may be dropped during processing for congestion control or by the
4673 * NET_RX_SUCCESS (no congestion)
4674 * NET_RX_DROP (packet was dropped)
4678 int netif_rx(struct sk_buff *skb)
4682 trace_netif_rx_entry(skb);
4684 ret = netif_rx_internal(skb);
4685 trace_netif_rx_exit(ret);
4689 EXPORT_SYMBOL(netif_rx);
4691 int netif_rx_ni(struct sk_buff *skb)
4695 trace_netif_rx_ni_entry(skb);
4698 err = netif_rx_internal(skb);
4699 if (local_softirq_pending())
4702 trace_netif_rx_ni_exit(err);
4706 EXPORT_SYMBOL(netif_rx_ni);
4708 static __latent_entropy void net_tx_action(struct softirq_action *h)
4710 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4712 if (sd->completion_queue) {
4713 struct sk_buff *clist;
4715 local_irq_disable();
4716 clist = sd->completion_queue;
4717 sd->completion_queue = NULL;
4721 struct sk_buff *skb = clist;
4723 clist = clist->next;
4725 WARN_ON(refcount_read(&skb->users));
4726 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4727 trace_consume_skb(skb);
4729 trace_kfree_skb(skb, net_tx_action);
4731 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4734 __kfree_skb_defer(skb);
4737 __kfree_skb_flush();
4740 if (sd->output_queue) {
4743 local_irq_disable();
4744 head = sd->output_queue;
4745 sd->output_queue = NULL;
4746 sd->output_queue_tailp = &sd->output_queue;
4750 struct Qdisc *q = head;
4751 spinlock_t *root_lock = NULL;
4753 head = head->next_sched;
4755 if (!(q->flags & TCQ_F_NOLOCK)) {
4756 root_lock = qdisc_lock(q);
4757 spin_lock(root_lock);
4759 /* We need to make sure head->next_sched is read
4760 * before clearing __QDISC_STATE_SCHED
4762 smp_mb__before_atomic();
4763 clear_bit(__QDISC_STATE_SCHED, &q->state);
4766 spin_unlock(root_lock);
4770 xfrm_dev_backlog(sd);
4773 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4774 /* This hook is defined here for ATM LANE */
4775 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4776 unsigned char *addr) __read_mostly;
4777 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4780 static inline struct sk_buff *
4781 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4782 struct net_device *orig_dev)
4784 #ifdef CONFIG_NET_CLS_ACT
4785 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4786 struct tcf_result cl_res;
4788 /* If there's at least one ingress present somewhere (so
4789 * we get here via enabled static key), remaining devices
4790 * that are not configured with an ingress qdisc will bail
4797 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4801 qdisc_skb_cb(skb)->pkt_len = skb->len;
4802 skb->tc_at_ingress = 1;
4803 mini_qdisc_bstats_cpu_update(miniq, skb);
4805 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
4807 case TC_ACT_RECLASSIFY:
4808 skb->tc_index = TC_H_MIN(cl_res.classid);
4811 mini_qdisc_qstats_cpu_drop(miniq);
4819 case TC_ACT_REDIRECT:
4820 /* skb_mac_header check was done by cls/act_bpf, so
4821 * we can safely push the L2 header back before
4822 * redirecting to another netdev
4824 __skb_push(skb, skb->mac_len);
4825 skb_do_redirect(skb);
4827 case TC_ACT_CONSUMED:
4832 #endif /* CONFIG_NET_CLS_ACT */
4837 * netdev_is_rx_handler_busy - check if receive handler is registered
4838 * @dev: device to check
4840 * Check if a receive handler is already registered for a given device.
4841 * Return true if there one.
4843 * The caller must hold the rtnl_mutex.
4845 bool netdev_is_rx_handler_busy(struct net_device *dev)
4848 return dev && rtnl_dereference(dev->rx_handler);
4850 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4853 * netdev_rx_handler_register - register receive handler
4854 * @dev: device to register a handler for
4855 * @rx_handler: receive handler to register
4856 * @rx_handler_data: data pointer that is used by rx handler
4858 * Register a receive handler for a device. This handler will then be
4859 * called from __netif_receive_skb. A negative errno code is returned
4862 * The caller must hold the rtnl_mutex.
4864 * For a general description of rx_handler, see enum rx_handler_result.
4866 int netdev_rx_handler_register(struct net_device *dev,
4867 rx_handler_func_t *rx_handler,
4868 void *rx_handler_data)
4870 if (netdev_is_rx_handler_busy(dev))
4873 if (dev->priv_flags & IFF_NO_RX_HANDLER)
4876 /* Note: rx_handler_data must be set before rx_handler */
4877 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4878 rcu_assign_pointer(dev->rx_handler, rx_handler);
4882 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4885 * netdev_rx_handler_unregister - unregister receive handler
4886 * @dev: device to unregister a handler from
4888 * Unregister a receive handler from a device.
4890 * The caller must hold the rtnl_mutex.
4892 void netdev_rx_handler_unregister(struct net_device *dev)
4896 RCU_INIT_POINTER(dev->rx_handler, NULL);
4897 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4898 * section has a guarantee to see a non NULL rx_handler_data
4902 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4904 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4907 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4908 * the special handling of PFMEMALLOC skbs.
4910 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4912 switch (skb->protocol) {
4913 case htons(ETH_P_ARP):
4914 case htons(ETH_P_IP):
4915 case htons(ETH_P_IPV6):
4916 case htons(ETH_P_8021Q):
4917 case htons(ETH_P_8021AD):
4924 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4925 int *ret, struct net_device *orig_dev)
4927 if (nf_hook_ingress_active(skb)) {
4931 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4936 ingress_retval = nf_hook_ingress(skb);
4938 return ingress_retval;
4943 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
4944 struct packet_type **ppt_prev)
4946 struct packet_type *ptype, *pt_prev;
4947 rx_handler_func_t *rx_handler;
4948 struct net_device *orig_dev;
4949 bool deliver_exact = false;
4950 int ret = NET_RX_DROP;
4953 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4955 trace_netif_receive_skb(skb);
4957 orig_dev = skb->dev;
4959 skb_reset_network_header(skb);
4960 if (!skb_transport_header_was_set(skb))
4961 skb_reset_transport_header(skb);
4962 skb_reset_mac_len(skb);
4967 skb->skb_iif = skb->dev->ifindex;
4969 __this_cpu_inc(softnet_data.processed);
4971 if (static_branch_unlikely(&generic_xdp_needed_key)) {
4975 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
4978 if (ret2 != XDP_PASS)
4980 skb_reset_mac_len(skb);
4983 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4984 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4985 skb = skb_vlan_untag(skb);
4990 if (skb_skip_tc_classify(skb))
4996 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4998 ret = deliver_skb(skb, pt_prev, orig_dev);
5002 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5004 ret = deliver_skb(skb, pt_prev, orig_dev);
5009 #ifdef CONFIG_NET_INGRESS
5010 if (static_branch_unlikely(&ingress_needed_key)) {
5011 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
5015 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5021 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5024 if (skb_vlan_tag_present(skb)) {
5026 ret = deliver_skb(skb, pt_prev, orig_dev);
5029 if (vlan_do_receive(&skb))
5031 else if (unlikely(!skb))
5035 rx_handler = rcu_dereference(skb->dev->rx_handler);
5038 ret = deliver_skb(skb, pt_prev, orig_dev);
5041 switch (rx_handler(&skb)) {
5042 case RX_HANDLER_CONSUMED:
5043 ret = NET_RX_SUCCESS;
5045 case RX_HANDLER_ANOTHER:
5047 case RX_HANDLER_EXACT:
5048 deliver_exact = true;
5049 case RX_HANDLER_PASS:
5056 if (unlikely(skb_vlan_tag_present(skb))) {
5058 if (skb_vlan_tag_get_id(skb)) {
5059 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5062 skb->pkt_type = PACKET_OTHERHOST;
5063 } else if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
5064 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
5065 /* Outer header is 802.1P with vlan 0, inner header is
5066 * 802.1Q or 802.1AD and vlan_do_receive() above could
5067 * not find vlan dev for vlan id 0.
5069 __vlan_hwaccel_clear_tag(skb);
5070 skb = skb_vlan_untag(skb);
5073 if (vlan_do_receive(&skb))
5074 /* After stripping off 802.1P header with vlan 0
5075 * vlan dev is found for inner header.
5078 else if (unlikely(!skb))
5081 /* We have stripped outer 802.1P vlan 0 header.
5082 * But could not find vlan dev.
5083 * check again for vlan id to set OTHERHOST.
5087 /* Note: we might in the future use prio bits
5088 * and set skb->priority like in vlan_do_receive()
5089 * For the time being, just ignore Priority Code Point
5091 __vlan_hwaccel_clear_tag(skb);
5094 type = skb->protocol;
5096 /* deliver only exact match when indicated */
5097 if (likely(!deliver_exact)) {
5098 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5099 &ptype_base[ntohs(type) &
5103 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5104 &orig_dev->ptype_specific);
5106 if (unlikely(skb->dev != orig_dev)) {
5107 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5108 &skb->dev->ptype_specific);
5112 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5114 *ppt_prev = pt_prev;
5118 atomic_long_inc(&skb->dev->rx_dropped);
5120 atomic_long_inc(&skb->dev->rx_nohandler);
5122 /* Jamal, now you will not able to escape explaining
5123 * me how you were going to use this. :-)
5132 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5134 struct net_device *orig_dev = skb->dev;
5135 struct packet_type *pt_prev = NULL;
5138 ret = __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
5140 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5141 skb->dev, pt_prev, orig_dev);
5146 * netif_receive_skb_core - special purpose version of netif_receive_skb
5147 * @skb: buffer to process
5149 * More direct receive version of netif_receive_skb(). It should
5150 * only be used by callers that have a need to skip RPS and Generic XDP.
5151 * Caller must also take care of handling if (page_is_)pfmemalloc.
5153 * This function may only be called from softirq context and interrupts
5154 * should be enabled.
5156 * Return values (usually ignored):
5157 * NET_RX_SUCCESS: no congestion
5158 * NET_RX_DROP: packet was dropped
5160 int netif_receive_skb_core(struct sk_buff *skb)
5165 ret = __netif_receive_skb_one_core(skb, false);
5170 EXPORT_SYMBOL(netif_receive_skb_core);
5172 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5173 struct packet_type *pt_prev,
5174 struct net_device *orig_dev)
5176 struct sk_buff *skb, *next;
5180 if (list_empty(head))
5182 if (pt_prev->list_func != NULL)
5183 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5184 ip_list_rcv, head, pt_prev, orig_dev);
5186 list_for_each_entry_safe(skb, next, head, list) {
5187 skb_list_del_init(skb);
5188 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5192 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5194 /* Fast-path assumptions:
5195 * - There is no RX handler.
5196 * - Only one packet_type matches.
5197 * If either of these fails, we will end up doing some per-packet
5198 * processing in-line, then handling the 'last ptype' for the whole
5199 * sublist. This can't cause out-of-order delivery to any single ptype,
5200 * because the 'last ptype' must be constant across the sublist, and all
5201 * other ptypes are handled per-packet.
5203 /* Current (common) ptype of sublist */
5204 struct packet_type *pt_curr = NULL;
5205 /* Current (common) orig_dev of sublist */
5206 struct net_device *od_curr = NULL;
5207 struct list_head sublist;
5208 struct sk_buff *skb, *next;
5210 INIT_LIST_HEAD(&sublist);
5211 list_for_each_entry_safe(skb, next, head, list) {
5212 struct net_device *orig_dev = skb->dev;
5213 struct packet_type *pt_prev = NULL;
5215 skb_list_del_init(skb);
5216 __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
5219 if (pt_curr != pt_prev || od_curr != orig_dev) {
5220 /* dispatch old sublist */
5221 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5222 /* start new sublist */
5223 INIT_LIST_HEAD(&sublist);
5227 list_add_tail(&skb->list, &sublist);
5230 /* dispatch final sublist */
5231 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5234 static int __netif_receive_skb(struct sk_buff *skb)
5238 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5239 unsigned int noreclaim_flag;
5242 * PFMEMALLOC skbs are special, they should
5243 * - be delivered to SOCK_MEMALLOC sockets only
5244 * - stay away from userspace
5245 * - have bounded memory usage
5247 * Use PF_MEMALLOC as this saves us from propagating the allocation
5248 * context down to all allocation sites.
5250 noreclaim_flag = memalloc_noreclaim_save();
5251 ret = __netif_receive_skb_one_core(skb, true);
5252 memalloc_noreclaim_restore(noreclaim_flag);
5254 ret = __netif_receive_skb_one_core(skb, false);
5259 static void __netif_receive_skb_list(struct list_head *head)
5261 unsigned long noreclaim_flag = 0;
5262 struct sk_buff *skb, *next;
5263 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5265 list_for_each_entry_safe(skb, next, head, list) {
5266 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5267 struct list_head sublist;
5269 /* Handle the previous sublist */
5270 list_cut_before(&sublist, head, &skb->list);
5271 if (!list_empty(&sublist))
5272 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5273 pfmemalloc = !pfmemalloc;
5274 /* See comments in __netif_receive_skb */
5276 noreclaim_flag = memalloc_noreclaim_save();
5278 memalloc_noreclaim_restore(noreclaim_flag);
5281 /* Handle the remaining sublist */
5282 if (!list_empty(head))
5283 __netif_receive_skb_list_core(head, pfmemalloc);
5284 /* Restore pflags */
5286 memalloc_noreclaim_restore(noreclaim_flag);
5289 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5291 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5292 struct bpf_prog *new = xdp->prog;
5295 switch (xdp->command) {
5296 case XDP_SETUP_PROG:
5297 rcu_assign_pointer(dev->xdp_prog, new);
5302 static_branch_dec(&generic_xdp_needed_key);
5303 } else if (new && !old) {
5304 static_branch_inc(&generic_xdp_needed_key);
5305 dev_disable_lro(dev);
5306 dev_disable_gro_hw(dev);
5310 case XDP_QUERY_PROG:
5311 xdp->prog_id = old ? old->aux->id : 0;
5322 static int netif_receive_skb_internal(struct sk_buff *skb)
5326 net_timestamp_check(netdev_tstamp_prequeue, skb);
5328 if (skb_defer_rx_timestamp(skb))
5329 return NET_RX_SUCCESS;
5333 if (static_branch_unlikely(&rps_needed)) {
5334 struct rps_dev_flow voidflow, *rflow = &voidflow;
5335 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5338 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5344 ret = __netif_receive_skb(skb);
5349 static void netif_receive_skb_list_internal(struct list_head *head)
5351 struct sk_buff *skb, *next;
5352 struct list_head sublist;
5354 INIT_LIST_HEAD(&sublist);
5355 list_for_each_entry_safe(skb, next, head, list) {
5356 net_timestamp_check(netdev_tstamp_prequeue, skb);
5357 skb_list_del_init(skb);
5358 if (!skb_defer_rx_timestamp(skb))
5359 list_add_tail(&skb->list, &sublist);
5361 list_splice_init(&sublist, head);
5365 if (static_branch_unlikely(&rps_needed)) {
5366 list_for_each_entry_safe(skb, next, head, list) {
5367 struct rps_dev_flow voidflow, *rflow = &voidflow;
5368 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5371 /* Will be handled, remove from list */
5372 skb_list_del_init(skb);
5373 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5378 __netif_receive_skb_list(head);
5383 * netif_receive_skb - process receive buffer from network
5384 * @skb: buffer to process
5386 * netif_receive_skb() is the main receive data processing function.
5387 * It always succeeds. The buffer may be dropped during processing
5388 * for congestion control or by the protocol layers.
5390 * This function may only be called from softirq context and interrupts
5391 * should be enabled.
5393 * Return values (usually ignored):
5394 * NET_RX_SUCCESS: no congestion
5395 * NET_RX_DROP: packet was dropped
5397 int netif_receive_skb(struct sk_buff *skb)
5401 trace_netif_receive_skb_entry(skb);
5403 ret = netif_receive_skb_internal(skb);
5404 trace_netif_receive_skb_exit(ret);
5408 EXPORT_SYMBOL(netif_receive_skb);
5411 * netif_receive_skb_list - process many receive buffers from network
5412 * @head: list of skbs to process.
5414 * Since return value of netif_receive_skb() is normally ignored, and
5415 * wouldn't be meaningful for a list, this function returns void.
5417 * This function may only be called from softirq context and interrupts
5418 * should be enabled.
5420 void netif_receive_skb_list(struct list_head *head)
5422 struct sk_buff *skb;
5424 if (list_empty(head))
5426 if (trace_netif_receive_skb_list_entry_enabled()) {
5427 list_for_each_entry(skb, head, list)
5428 trace_netif_receive_skb_list_entry(skb);
5430 netif_receive_skb_list_internal(head);
5431 trace_netif_receive_skb_list_exit(0);
5433 EXPORT_SYMBOL(netif_receive_skb_list);
5435 DEFINE_PER_CPU(struct work_struct, flush_works);
5437 /* Network device is going away, flush any packets still pending */
5438 static void flush_backlog(struct work_struct *work)
5440 struct sk_buff *skb, *tmp;
5441 struct softnet_data *sd;
5444 sd = this_cpu_ptr(&softnet_data);
5446 local_irq_disable();
5448 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5449 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5450 __skb_unlink(skb, &sd->input_pkt_queue);
5452 input_queue_head_incr(sd);
5458 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5459 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5460 __skb_unlink(skb, &sd->process_queue);
5462 input_queue_head_incr(sd);
5468 static void flush_all_backlogs(void)
5474 for_each_online_cpu(cpu)
5475 queue_work_on(cpu, system_highpri_wq,
5476 per_cpu_ptr(&flush_works, cpu));
5478 for_each_online_cpu(cpu)
5479 flush_work(per_cpu_ptr(&flush_works, cpu));
5484 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */
5485 static void gro_normal_list(struct napi_struct *napi)
5487 if (!napi->rx_count)
5489 netif_receive_skb_list_internal(&napi->rx_list);
5490 INIT_LIST_HEAD(&napi->rx_list);
5494 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded,
5495 * pass the whole batch up to the stack.
5497 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb)
5499 list_add_tail(&skb->list, &napi->rx_list);
5500 if (++napi->rx_count >= gro_normal_batch)
5501 gro_normal_list(napi);
5504 INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int));
5505 INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int));
5506 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb)
5508 struct packet_offload *ptype;
5509 __be16 type = skb->protocol;
5510 struct list_head *head = &offload_base;
5513 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5515 if (NAPI_GRO_CB(skb)->count == 1) {
5516 skb_shinfo(skb)->gso_size = 0;
5521 list_for_each_entry_rcu(ptype, head, list) {
5522 if (ptype->type != type || !ptype->callbacks.gro_complete)
5525 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5526 ipv6_gro_complete, inet_gro_complete,
5533 WARN_ON(&ptype->list == head);
5535 return NET_RX_SUCCESS;
5539 gro_normal_one(napi, skb);
5540 return NET_RX_SUCCESS;
5543 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5546 struct list_head *head = &napi->gro_hash[index].list;
5547 struct sk_buff *skb, *p;
5549 list_for_each_entry_safe_reverse(skb, p, head, list) {
5550 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5552 skb_list_del_init(skb);
5553 napi_gro_complete(napi, skb);
5554 napi->gro_hash[index].count--;
5557 if (!napi->gro_hash[index].count)
5558 __clear_bit(index, &napi->gro_bitmask);
5561 /* napi->gro_hash[].list contains packets ordered by age.
5562 * youngest packets at the head of it.
5563 * Complete skbs in reverse order to reduce latencies.
5565 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5567 unsigned long bitmask = napi->gro_bitmask;
5568 unsigned int i, base = ~0U;
5570 while ((i = ffs(bitmask)) != 0) {
5573 __napi_gro_flush_chain(napi, base, flush_old);
5576 EXPORT_SYMBOL(napi_gro_flush);
5578 static struct list_head *gro_list_prepare(struct napi_struct *napi,
5579 struct sk_buff *skb)
5581 unsigned int maclen = skb->dev->hard_header_len;
5582 u32 hash = skb_get_hash_raw(skb);
5583 struct list_head *head;
5586 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list;
5587 list_for_each_entry(p, head, list) {
5588 unsigned long diffs;
5590 NAPI_GRO_CB(p)->flush = 0;
5592 if (hash != skb_get_hash_raw(p)) {
5593 NAPI_GRO_CB(p)->same_flow = 0;
5597 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5598 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5599 if (skb_vlan_tag_present(p))
5600 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb);
5601 diffs |= skb_metadata_dst_cmp(p, skb);
5602 diffs |= skb_metadata_differs(p, skb);
5603 if (maclen == ETH_HLEN)
5604 diffs |= compare_ether_header(skb_mac_header(p),
5605 skb_mac_header(skb));
5607 diffs = memcmp(skb_mac_header(p),
5608 skb_mac_header(skb),
5610 NAPI_GRO_CB(p)->same_flow = !diffs;
5616 static void skb_gro_reset_offset(struct sk_buff *skb)
5618 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5619 const skb_frag_t *frag0 = &pinfo->frags[0];
5621 NAPI_GRO_CB(skb)->data_offset = 0;
5622 NAPI_GRO_CB(skb)->frag0 = NULL;
5623 NAPI_GRO_CB(skb)->frag0_len = 0;
5625 if (!skb_headlen(skb) && pinfo->nr_frags &&
5626 !PageHighMem(skb_frag_page(frag0))) {
5627 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5628 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5629 skb_frag_size(frag0),
5630 skb->end - skb->tail);
5634 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5636 struct skb_shared_info *pinfo = skb_shinfo(skb);
5638 BUG_ON(skb->end - skb->tail < grow);
5640 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5642 skb->data_len -= grow;
5645 skb_frag_off_add(&pinfo->frags[0], grow);
5646 skb_frag_size_sub(&pinfo->frags[0], grow);
5648 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5649 skb_frag_unref(skb, 0);
5650 memmove(pinfo->frags, pinfo->frags + 1,
5651 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
5655 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head)
5657 struct sk_buff *oldest;
5659 oldest = list_last_entry(head, struct sk_buff, list);
5661 /* We are called with head length >= MAX_GRO_SKBS, so this is
5664 if (WARN_ON_ONCE(!oldest))
5667 /* Do not adjust napi->gro_hash[].count, caller is adding a new
5670 skb_list_del_init(oldest);
5671 napi_gro_complete(napi, oldest);
5674 INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *,
5676 INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *,
5678 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5680 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
5681 struct list_head *head = &offload_base;
5682 struct packet_offload *ptype;
5683 __be16 type = skb->protocol;
5684 struct list_head *gro_head;
5685 struct sk_buff *pp = NULL;
5686 enum gro_result ret;
5690 if (netif_elide_gro(skb->dev))
5693 gro_head = gro_list_prepare(napi, skb);
5696 list_for_each_entry_rcu(ptype, head, list) {
5697 if (ptype->type != type || !ptype->callbacks.gro_receive)
5700 skb_set_network_header(skb, skb_gro_offset(skb));
5701 skb_reset_mac_len(skb);
5702 NAPI_GRO_CB(skb)->same_flow = 0;
5703 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
5704 NAPI_GRO_CB(skb)->free = 0;
5705 NAPI_GRO_CB(skb)->encap_mark = 0;
5706 NAPI_GRO_CB(skb)->recursion_counter = 0;
5707 NAPI_GRO_CB(skb)->is_fou = 0;
5708 NAPI_GRO_CB(skb)->is_atomic = 1;
5709 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
5711 /* Setup for GRO checksum validation */
5712 switch (skb->ip_summed) {
5713 case CHECKSUM_COMPLETE:
5714 NAPI_GRO_CB(skb)->csum = skb->csum;
5715 NAPI_GRO_CB(skb)->csum_valid = 1;
5716 NAPI_GRO_CB(skb)->csum_cnt = 0;
5718 case CHECKSUM_UNNECESSARY:
5719 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
5720 NAPI_GRO_CB(skb)->csum_valid = 0;
5723 NAPI_GRO_CB(skb)->csum_cnt = 0;
5724 NAPI_GRO_CB(skb)->csum_valid = 0;
5727 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
5728 ipv6_gro_receive, inet_gro_receive,
5734 if (&ptype->list == head)
5737 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
5742 same_flow = NAPI_GRO_CB(skb)->same_flow;
5743 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5746 skb_list_del_init(pp);
5747 napi_gro_complete(napi, pp);
5748 napi->gro_hash[hash].count--;
5754 if (NAPI_GRO_CB(skb)->flush)
5757 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
5758 gro_flush_oldest(napi, gro_head);
5760 napi->gro_hash[hash].count++;
5762 NAPI_GRO_CB(skb)->count = 1;
5763 NAPI_GRO_CB(skb)->age = jiffies;
5764 NAPI_GRO_CB(skb)->last = skb;
5765 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
5766 list_add(&skb->list, gro_head);
5770 grow = skb_gro_offset(skb) - skb_headlen(skb);
5772 gro_pull_from_frag0(skb, grow);
5774 if (napi->gro_hash[hash].count) {
5775 if (!test_bit(hash, &napi->gro_bitmask))
5776 __set_bit(hash, &napi->gro_bitmask);
5777 } else if (test_bit(hash, &napi->gro_bitmask)) {
5778 __clear_bit(hash, &napi->gro_bitmask);
5788 struct packet_offload *gro_find_receive_by_type(__be16 type)
5790 struct list_head *offload_head = &offload_base;
5791 struct packet_offload *ptype;
5793 list_for_each_entry_rcu(ptype, offload_head, list) {
5794 if (ptype->type != type || !ptype->callbacks.gro_receive)
5800 EXPORT_SYMBOL(gro_find_receive_by_type);
5802 struct packet_offload *gro_find_complete_by_type(__be16 type)
5804 struct list_head *offload_head = &offload_base;
5805 struct packet_offload *ptype;
5807 list_for_each_entry_rcu(ptype, offload_head, list) {
5808 if (ptype->type != type || !ptype->callbacks.gro_complete)
5814 EXPORT_SYMBOL(gro_find_complete_by_type);
5816 static void napi_skb_free_stolen_head(struct sk_buff *skb)
5820 kmem_cache_free(skbuff_head_cache, skb);
5823 static gro_result_t napi_skb_finish(struct napi_struct *napi,
5824 struct sk_buff *skb,
5829 gro_normal_one(napi, skb);
5836 case GRO_MERGED_FREE:
5837 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5838 napi_skb_free_stolen_head(skb);
5852 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5856 skb_mark_napi_id(skb, napi);
5857 trace_napi_gro_receive_entry(skb);
5859 skb_gro_reset_offset(skb);
5861 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
5862 trace_napi_gro_receive_exit(ret);
5866 EXPORT_SYMBOL(napi_gro_receive);
5868 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
5870 if (unlikely(skb->pfmemalloc)) {
5874 __skb_pull(skb, skb_headlen(skb));
5875 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
5876 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
5877 __vlan_hwaccel_clear_tag(skb);
5878 skb->dev = napi->dev;
5881 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
5882 skb->pkt_type = PACKET_HOST;
5884 skb->encapsulation = 0;
5885 skb_shinfo(skb)->gso_type = 0;
5886 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5892 struct sk_buff *napi_get_frags(struct napi_struct *napi)
5894 struct sk_buff *skb = napi->skb;
5897 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
5900 skb_mark_napi_id(skb, napi);
5905 EXPORT_SYMBOL(napi_get_frags);
5907 static gro_result_t napi_frags_finish(struct napi_struct *napi,
5908 struct sk_buff *skb,
5914 __skb_push(skb, ETH_HLEN);
5915 skb->protocol = eth_type_trans(skb, skb->dev);
5916 if (ret == GRO_NORMAL)
5917 gro_normal_one(napi, skb);
5921 napi_reuse_skb(napi, skb);
5924 case GRO_MERGED_FREE:
5925 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5926 napi_skb_free_stolen_head(skb);
5928 napi_reuse_skb(napi, skb);
5939 /* Upper GRO stack assumes network header starts at gro_offset=0
5940 * Drivers could call both napi_gro_frags() and napi_gro_receive()
5941 * We copy ethernet header into skb->data to have a common layout.
5943 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
5945 struct sk_buff *skb = napi->skb;
5946 const struct ethhdr *eth;
5947 unsigned int hlen = sizeof(*eth);
5951 skb_reset_mac_header(skb);
5952 skb_gro_reset_offset(skb);
5954 if (unlikely(skb_gro_header_hard(skb, hlen))) {
5955 eth = skb_gro_header_slow(skb, hlen, 0);
5956 if (unlikely(!eth)) {
5957 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
5958 __func__, napi->dev->name);
5959 napi_reuse_skb(napi, skb);
5963 eth = (const struct ethhdr *)skb->data;
5964 gro_pull_from_frag0(skb, hlen);
5965 NAPI_GRO_CB(skb)->frag0 += hlen;
5966 NAPI_GRO_CB(skb)->frag0_len -= hlen;
5968 __skb_pull(skb, hlen);
5971 * This works because the only protocols we care about don't require
5973 * We'll fix it up properly in napi_frags_finish()
5975 skb->protocol = eth->h_proto;
5980 gro_result_t napi_gro_frags(struct napi_struct *napi)
5983 struct sk_buff *skb = napi_frags_skb(napi);
5988 trace_napi_gro_frags_entry(skb);
5990 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
5991 trace_napi_gro_frags_exit(ret);
5995 EXPORT_SYMBOL(napi_gro_frags);
5997 /* Compute the checksum from gro_offset and return the folded value
5998 * after adding in any pseudo checksum.
6000 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
6005 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
6007 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
6008 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
6009 /* See comments in __skb_checksum_complete(). */
6011 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
6012 !skb->csum_complete_sw)
6013 netdev_rx_csum_fault(skb->dev, skb);
6016 NAPI_GRO_CB(skb)->csum = wsum;
6017 NAPI_GRO_CB(skb)->csum_valid = 1;
6021 EXPORT_SYMBOL(__skb_gro_checksum_complete);
6023 static void net_rps_send_ipi(struct softnet_data *remsd)
6027 struct softnet_data *next = remsd->rps_ipi_next;
6029 if (cpu_online(remsd->cpu))
6030 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6037 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6038 * Note: called with local irq disabled, but exits with local irq enabled.
6040 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6043 struct softnet_data *remsd = sd->rps_ipi_list;
6046 sd->rps_ipi_list = NULL;
6050 /* Send pending IPI's to kick RPS processing on remote cpus. */
6051 net_rps_send_ipi(remsd);
6057 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6060 return sd->rps_ipi_list != NULL;
6066 static int process_backlog(struct napi_struct *napi, int quota)
6068 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6072 /* Check if we have pending ipi, its better to send them now,
6073 * not waiting net_rx_action() end.
6075 if (sd_has_rps_ipi_waiting(sd)) {
6076 local_irq_disable();
6077 net_rps_action_and_irq_enable(sd);
6080 napi->weight = dev_rx_weight;
6082 struct sk_buff *skb;
6084 while ((skb = __skb_dequeue(&sd->process_queue))) {
6086 __netif_receive_skb(skb);
6088 input_queue_head_incr(sd);
6089 if (++work >= quota)
6094 local_irq_disable();
6096 if (skb_queue_empty(&sd->input_pkt_queue)) {
6098 * Inline a custom version of __napi_complete().
6099 * only current cpu owns and manipulates this napi,
6100 * and NAPI_STATE_SCHED is the only possible flag set
6102 * We can use a plain write instead of clear_bit(),
6103 * and we dont need an smp_mb() memory barrier.
6108 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6109 &sd->process_queue);
6119 * __napi_schedule - schedule for receive
6120 * @n: entry to schedule
6122 * The entry's receive function will be scheduled to run.
6123 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6125 void __napi_schedule(struct napi_struct *n)
6127 unsigned long flags;
6129 local_irq_save(flags);
6130 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6131 local_irq_restore(flags);
6133 EXPORT_SYMBOL(__napi_schedule);
6136 * napi_schedule_prep - check if napi can be scheduled
6139 * Test if NAPI routine is already running, and if not mark
6140 * it as running. This is used as a condition variable
6141 * insure only one NAPI poll instance runs. We also make
6142 * sure there is no pending NAPI disable.
6144 bool napi_schedule_prep(struct napi_struct *n)
6146 unsigned long val, new;
6149 val = READ_ONCE(n->state);
6150 if (unlikely(val & NAPIF_STATE_DISABLE))
6152 new = val | NAPIF_STATE_SCHED;
6154 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6155 * This was suggested by Alexander Duyck, as compiler
6156 * emits better code than :
6157 * if (val & NAPIF_STATE_SCHED)
6158 * new |= NAPIF_STATE_MISSED;
6160 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6162 } while (cmpxchg(&n->state, val, new) != val);
6164 return !(val & NAPIF_STATE_SCHED);
6166 EXPORT_SYMBOL(napi_schedule_prep);
6169 * __napi_schedule_irqoff - schedule for receive
6170 * @n: entry to schedule
6172 * Variant of __napi_schedule() assuming hard irqs are masked
6174 void __napi_schedule_irqoff(struct napi_struct *n)
6176 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6178 EXPORT_SYMBOL(__napi_schedule_irqoff);
6180 bool napi_complete_done(struct napi_struct *n, int work_done)
6182 unsigned long flags, val, new;
6185 * 1) Don't let napi dequeue from the cpu poll list
6186 * just in case its running on a different cpu.
6187 * 2) If we are busy polling, do nothing here, we have
6188 * the guarantee we will be called later.
6190 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6191 NAPIF_STATE_IN_BUSY_POLL)))
6194 if (n->gro_bitmask) {
6195 unsigned long timeout = 0;
6198 timeout = n->dev->gro_flush_timeout;
6200 /* When the NAPI instance uses a timeout and keeps postponing
6201 * it, we need to bound somehow the time packets are kept in
6204 napi_gro_flush(n, !!timeout);
6206 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6207 HRTIMER_MODE_REL_PINNED);
6212 if (unlikely(!list_empty(&n->poll_list))) {
6213 /* If n->poll_list is not empty, we need to mask irqs */
6214 local_irq_save(flags);
6215 list_del_init(&n->poll_list);
6216 local_irq_restore(flags);
6220 val = READ_ONCE(n->state);
6222 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6224 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
6226 /* If STATE_MISSED was set, leave STATE_SCHED set,
6227 * because we will call napi->poll() one more time.
6228 * This C code was suggested by Alexander Duyck to help gcc.
6230 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6232 } while (cmpxchg(&n->state, val, new) != val);
6234 if (unlikely(val & NAPIF_STATE_MISSED)) {
6241 EXPORT_SYMBOL(napi_complete_done);
6243 /* must be called under rcu_read_lock(), as we dont take a reference */
6244 static struct napi_struct *napi_by_id(unsigned int napi_id)
6246 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6247 struct napi_struct *napi;
6249 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6250 if (napi->napi_id == napi_id)
6256 #if defined(CONFIG_NET_RX_BUSY_POLL)
6258 #define BUSY_POLL_BUDGET 8
6260 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
6264 /* Busy polling means there is a high chance device driver hard irq
6265 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6266 * set in napi_schedule_prep().
6267 * Since we are about to call napi->poll() once more, we can safely
6268 * clear NAPI_STATE_MISSED.
6270 * Note: x86 could use a single "lock and ..." instruction
6271 * to perform these two clear_bit()
6273 clear_bit(NAPI_STATE_MISSED, &napi->state);
6274 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6278 /* All we really want here is to re-enable device interrupts.
6279 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6281 rc = napi->poll(napi, BUSY_POLL_BUDGET);
6282 /* We can't gro_normal_list() here, because napi->poll() might have
6283 * rearmed the napi (napi_complete_done()) in which case it could
6284 * already be running on another CPU.
6286 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
6287 netpoll_poll_unlock(have_poll_lock);
6288 if (rc == BUSY_POLL_BUDGET) {
6289 /* As the whole budget was spent, we still own the napi so can
6290 * safely handle the rx_list.
6292 gro_normal_list(napi);
6293 __napi_schedule(napi);
6298 void napi_busy_loop(unsigned int napi_id,
6299 bool (*loop_end)(void *, unsigned long),
6302 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6303 int (*napi_poll)(struct napi_struct *napi, int budget);
6304 void *have_poll_lock = NULL;
6305 struct napi_struct *napi;
6312 napi = napi_by_id(napi_id);
6322 unsigned long val = READ_ONCE(napi->state);
6324 /* If multiple threads are competing for this napi,
6325 * we avoid dirtying napi->state as much as we can.
6327 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6328 NAPIF_STATE_IN_BUSY_POLL))
6330 if (cmpxchg(&napi->state, val,
6331 val | NAPIF_STATE_IN_BUSY_POLL |
6332 NAPIF_STATE_SCHED) != val)
6334 have_poll_lock = netpoll_poll_lock(napi);
6335 napi_poll = napi->poll;
6337 work = napi_poll(napi, BUSY_POLL_BUDGET);
6338 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
6339 gro_normal_list(napi);
6342 __NET_ADD_STATS(dev_net(napi->dev),
6343 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6346 if (!loop_end || loop_end(loop_end_arg, start_time))
6349 if (unlikely(need_resched())) {
6351 busy_poll_stop(napi, have_poll_lock);
6355 if (loop_end(loop_end_arg, start_time))
6362 busy_poll_stop(napi, have_poll_lock);
6367 EXPORT_SYMBOL(napi_busy_loop);
6369 #endif /* CONFIG_NET_RX_BUSY_POLL */
6371 static void napi_hash_add(struct napi_struct *napi)
6373 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
6374 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
6377 spin_lock(&napi_hash_lock);
6379 /* 0..NR_CPUS range is reserved for sender_cpu use */
6381 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6382 napi_gen_id = MIN_NAPI_ID;
6383 } while (napi_by_id(napi_gen_id));
6384 napi->napi_id = napi_gen_id;
6386 hlist_add_head_rcu(&napi->napi_hash_node,
6387 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6389 spin_unlock(&napi_hash_lock);
6392 /* Warning : caller is responsible to make sure rcu grace period
6393 * is respected before freeing memory containing @napi
6395 bool napi_hash_del(struct napi_struct *napi)
6397 bool rcu_sync_needed = false;
6399 spin_lock(&napi_hash_lock);
6401 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
6402 rcu_sync_needed = true;
6403 hlist_del_rcu(&napi->napi_hash_node);
6405 spin_unlock(&napi_hash_lock);
6406 return rcu_sync_needed;
6408 EXPORT_SYMBOL_GPL(napi_hash_del);
6410 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6412 struct napi_struct *napi;
6414 napi = container_of(timer, struct napi_struct, timer);
6416 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6417 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6419 if (napi->gro_bitmask && !napi_disable_pending(napi) &&
6420 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
6421 __napi_schedule_irqoff(napi);
6423 return HRTIMER_NORESTART;
6426 static void init_gro_hash(struct napi_struct *napi)
6430 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6431 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6432 napi->gro_hash[i].count = 0;
6434 napi->gro_bitmask = 0;
6437 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6438 int (*poll)(struct napi_struct *, int), int weight)
6440 INIT_LIST_HEAD(&napi->poll_list);
6441 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6442 napi->timer.function = napi_watchdog;
6443 init_gro_hash(napi);
6445 INIT_LIST_HEAD(&napi->rx_list);
6448 if (weight > NAPI_POLL_WEIGHT)
6449 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6451 napi->weight = weight;
6452 list_add(&napi->dev_list, &dev->napi_list);
6454 #ifdef CONFIG_NETPOLL
6455 napi->poll_owner = -1;
6457 set_bit(NAPI_STATE_SCHED, &napi->state);
6458 napi_hash_add(napi);
6460 EXPORT_SYMBOL(netif_napi_add);
6462 void napi_disable(struct napi_struct *n)
6465 set_bit(NAPI_STATE_DISABLE, &n->state);
6467 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6469 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6472 hrtimer_cancel(&n->timer);
6474 clear_bit(NAPI_STATE_DISABLE, &n->state);
6476 EXPORT_SYMBOL(napi_disable);
6478 static void flush_gro_hash(struct napi_struct *napi)
6482 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6483 struct sk_buff *skb, *n;
6485 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6487 napi->gro_hash[i].count = 0;
6491 /* Must be called in process context */
6492 void netif_napi_del(struct napi_struct *napi)
6495 if (napi_hash_del(napi))
6497 list_del_init(&napi->dev_list);
6498 napi_free_frags(napi);
6500 flush_gro_hash(napi);
6501 napi->gro_bitmask = 0;
6503 EXPORT_SYMBOL(netif_napi_del);
6505 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6510 list_del_init(&n->poll_list);
6512 have = netpoll_poll_lock(n);
6516 /* This NAPI_STATE_SCHED test is for avoiding a race
6517 * with netpoll's poll_napi(). Only the entity which
6518 * obtains the lock and sees NAPI_STATE_SCHED set will
6519 * actually make the ->poll() call. Therefore we avoid
6520 * accidentally calling ->poll() when NAPI is not scheduled.
6523 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6524 work = n->poll(n, weight);
6525 trace_napi_poll(n, work, weight);
6528 WARN_ON_ONCE(work > weight);
6530 if (likely(work < weight))
6533 /* Drivers must not modify the NAPI state if they
6534 * consume the entire weight. In such cases this code
6535 * still "owns" the NAPI instance and therefore can
6536 * move the instance around on the list at-will.
6538 if (unlikely(napi_disable_pending(n))) {
6543 if (n->gro_bitmask) {
6544 /* flush too old packets
6545 * If HZ < 1000, flush all packets.
6547 napi_gro_flush(n, HZ >= 1000);
6552 /* Some drivers may have called napi_schedule
6553 * prior to exhausting their budget.
6555 if (unlikely(!list_empty(&n->poll_list))) {
6556 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6557 n->dev ? n->dev->name : "backlog");
6561 list_add_tail(&n->poll_list, repoll);
6564 netpoll_poll_unlock(have);
6569 static __latent_entropy void net_rx_action(struct softirq_action *h)
6571 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6572 unsigned long time_limit = jiffies +
6573 usecs_to_jiffies(netdev_budget_usecs);
6574 int budget = netdev_budget;
6578 local_irq_disable();
6579 list_splice_init(&sd->poll_list, &list);
6583 struct napi_struct *n;
6585 if (list_empty(&list)) {
6586 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6591 n = list_first_entry(&list, struct napi_struct, poll_list);
6592 budget -= napi_poll(n, &repoll);
6594 /* If softirq window is exhausted then punt.
6595 * Allow this to run for 2 jiffies since which will allow
6596 * an average latency of 1.5/HZ.
6598 if (unlikely(budget <= 0 ||
6599 time_after_eq(jiffies, time_limit))) {
6605 local_irq_disable();
6607 list_splice_tail_init(&sd->poll_list, &list);
6608 list_splice_tail(&repoll, &list);
6609 list_splice(&list, &sd->poll_list);
6610 if (!list_empty(&sd->poll_list))
6611 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6613 net_rps_action_and_irq_enable(sd);
6615 __kfree_skb_flush();
6618 struct netdev_adjacent {
6619 struct net_device *dev;
6621 /* upper master flag, there can only be one master device per list */
6624 /* lookup ignore flag */
6627 /* counter for the number of times this device was added to us */
6630 /* private field for the users */
6633 struct list_head list;
6634 struct rcu_head rcu;
6637 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6638 struct list_head *adj_list)
6640 struct netdev_adjacent *adj;
6642 list_for_each_entry(adj, adj_list, list) {
6643 if (adj->dev == adj_dev)
6649 static int ____netdev_has_upper_dev(struct net_device *upper_dev, void *data)
6651 struct net_device *dev = data;
6653 return upper_dev == dev;
6657 * netdev_has_upper_dev - Check if device is linked to an upper device
6659 * @upper_dev: upper device to check
6661 * Find out if a device is linked to specified upper device and return true
6662 * in case it is. Note that this checks only immediate upper device,
6663 * not through a complete stack of devices. The caller must hold the RTNL lock.
6665 bool netdev_has_upper_dev(struct net_device *dev,
6666 struct net_device *upper_dev)
6670 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6673 EXPORT_SYMBOL(netdev_has_upper_dev);
6676 * netdev_has_upper_dev_all - Check if device is linked to an upper device
6678 * @upper_dev: upper device to check
6680 * Find out if a device is linked to specified upper device and return true
6681 * in case it is. Note that this checks the entire upper device chain.
6682 * The caller must hold rcu lock.
6685 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6686 struct net_device *upper_dev)
6688 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6691 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6694 * netdev_has_any_upper_dev - Check if device is linked to some device
6697 * Find out if a device is linked to an upper device and return true in case
6698 * it is. The caller must hold the RTNL lock.
6700 bool netdev_has_any_upper_dev(struct net_device *dev)
6704 return !list_empty(&dev->adj_list.upper);
6706 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6709 * netdev_master_upper_dev_get - Get master upper device
6712 * Find a master upper device and return pointer to it or NULL in case
6713 * it's not there. The caller must hold the RTNL lock.
6715 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6717 struct netdev_adjacent *upper;
6721 if (list_empty(&dev->adj_list.upper))
6724 upper = list_first_entry(&dev->adj_list.upper,
6725 struct netdev_adjacent, list);
6726 if (likely(upper->master))
6730 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6732 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6734 struct netdev_adjacent *upper;
6738 if (list_empty(&dev->adj_list.upper))
6741 upper = list_first_entry(&dev->adj_list.upper,
6742 struct netdev_adjacent, list);
6743 if (likely(upper->master) && !upper->ignore)
6749 * netdev_has_any_lower_dev - Check if device is linked to some device
6752 * Find out if a device is linked to a lower device and return true in case
6753 * it is. The caller must hold the RTNL lock.
6755 static bool netdev_has_any_lower_dev(struct net_device *dev)
6759 return !list_empty(&dev->adj_list.lower);
6762 void *netdev_adjacent_get_private(struct list_head *adj_list)
6764 struct netdev_adjacent *adj;
6766 adj = list_entry(adj_list, struct netdev_adjacent, list);
6768 return adj->private;
6770 EXPORT_SYMBOL(netdev_adjacent_get_private);
6773 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6775 * @iter: list_head ** of the current position
6777 * Gets the next device from the dev's upper list, starting from iter
6778 * position. The caller must hold RCU read lock.
6780 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6781 struct list_head **iter)
6783 struct netdev_adjacent *upper;
6785 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6787 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6789 if (&upper->list == &dev->adj_list.upper)
6792 *iter = &upper->list;
6796 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6798 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6799 struct list_head **iter,
6802 struct netdev_adjacent *upper;
6804 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6806 if (&upper->list == &dev->adj_list.upper)
6809 *iter = &upper->list;
6810 *ignore = upper->ignore;
6815 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6816 struct list_head **iter)
6818 struct netdev_adjacent *upper;
6820 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6822 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6824 if (&upper->list == &dev->adj_list.upper)
6827 *iter = &upper->list;
6832 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6833 int (*fn)(struct net_device *dev,
6837 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6838 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6843 iter = &dev->adj_list.upper;
6847 ret = fn(now, data);
6854 udev = __netdev_next_upper_dev(now, &iter, &ignore);
6861 niter = &udev->adj_list.upper;
6862 dev_stack[cur] = now;
6863 iter_stack[cur++] = iter;
6870 next = dev_stack[--cur];
6871 niter = iter_stack[cur];
6881 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6882 int (*fn)(struct net_device *dev,
6886 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6887 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6891 iter = &dev->adj_list.upper;
6895 ret = fn(now, data);
6902 udev = netdev_next_upper_dev_rcu(now, &iter);
6907 niter = &udev->adj_list.upper;
6908 dev_stack[cur] = now;
6909 iter_stack[cur++] = iter;
6916 next = dev_stack[--cur];
6917 niter = iter_stack[cur];
6926 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6928 static bool __netdev_has_upper_dev(struct net_device *dev,
6929 struct net_device *upper_dev)
6933 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
6938 * netdev_lower_get_next_private - Get the next ->private from the
6939 * lower neighbour list
6941 * @iter: list_head ** of the current position
6943 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6944 * list, starting from iter position. The caller must hold either hold the
6945 * RTNL lock or its own locking that guarantees that the neighbour lower
6946 * list will remain unchanged.
6948 void *netdev_lower_get_next_private(struct net_device *dev,
6949 struct list_head **iter)
6951 struct netdev_adjacent *lower;
6953 lower = list_entry(*iter, struct netdev_adjacent, list);
6955 if (&lower->list == &dev->adj_list.lower)
6958 *iter = lower->list.next;
6960 return lower->private;
6962 EXPORT_SYMBOL(netdev_lower_get_next_private);
6965 * netdev_lower_get_next_private_rcu - Get the next ->private from the
6966 * lower neighbour list, RCU
6969 * @iter: list_head ** of the current position
6971 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6972 * list, starting from iter position. The caller must hold RCU read lock.
6974 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
6975 struct list_head **iter)
6977 struct netdev_adjacent *lower;
6979 WARN_ON_ONCE(!rcu_read_lock_held());
6981 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6983 if (&lower->list == &dev->adj_list.lower)
6986 *iter = &lower->list;
6988 return lower->private;
6990 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
6993 * netdev_lower_get_next - Get the next device from the lower neighbour
6996 * @iter: list_head ** of the current position
6998 * Gets the next netdev_adjacent from the dev's lower neighbour
6999 * list, starting from iter position. The caller must hold RTNL lock or
7000 * its own locking that guarantees that the neighbour lower
7001 * list will remain unchanged.
7003 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7005 struct netdev_adjacent *lower;
7007 lower = list_entry(*iter, struct netdev_adjacent, list);
7009 if (&lower->list == &dev->adj_list.lower)
7012 *iter = lower->list.next;
7016 EXPORT_SYMBOL(netdev_lower_get_next);
7018 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7019 struct list_head **iter)
7021 struct netdev_adjacent *lower;
7023 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7025 if (&lower->list == &dev->adj_list.lower)
7028 *iter = &lower->list;
7033 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7034 struct list_head **iter,
7037 struct netdev_adjacent *lower;
7039 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7041 if (&lower->list == &dev->adj_list.lower)
7044 *iter = &lower->list;
7045 *ignore = lower->ignore;
7050 int netdev_walk_all_lower_dev(struct net_device *dev,
7051 int (*fn)(struct net_device *dev,
7055 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7056 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7060 iter = &dev->adj_list.lower;
7064 ret = fn(now, data);
7071 ldev = netdev_next_lower_dev(now, &iter);
7076 niter = &ldev->adj_list.lower;
7077 dev_stack[cur] = now;
7078 iter_stack[cur++] = iter;
7085 next = dev_stack[--cur];
7086 niter = iter_stack[cur];
7095 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7097 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7098 int (*fn)(struct net_device *dev,
7102 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7103 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7108 iter = &dev->adj_list.lower;
7112 ret = fn(now, data);
7119 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7126 niter = &ldev->adj_list.lower;
7127 dev_stack[cur] = now;
7128 iter_stack[cur++] = iter;
7135 next = dev_stack[--cur];
7136 niter = iter_stack[cur];
7146 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7147 struct list_head **iter)
7149 struct netdev_adjacent *lower;
7151 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7152 if (&lower->list == &dev->adj_list.lower)
7155 *iter = &lower->list;
7160 static u8 __netdev_upper_depth(struct net_device *dev)
7162 struct net_device *udev;
7163 struct list_head *iter;
7167 for (iter = &dev->adj_list.upper,
7168 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7170 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7173 if (max_depth < udev->upper_level)
7174 max_depth = udev->upper_level;
7180 static u8 __netdev_lower_depth(struct net_device *dev)
7182 struct net_device *ldev;
7183 struct list_head *iter;
7187 for (iter = &dev->adj_list.lower,
7188 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7190 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7193 if (max_depth < ldev->lower_level)
7194 max_depth = ldev->lower_level;
7200 static int __netdev_update_upper_level(struct net_device *dev, void *data)
7202 dev->upper_level = __netdev_upper_depth(dev) + 1;
7206 static int __netdev_update_lower_level(struct net_device *dev, void *data)
7208 dev->lower_level = __netdev_lower_depth(dev) + 1;
7212 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7213 int (*fn)(struct net_device *dev,
7217 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7218 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7222 iter = &dev->adj_list.lower;
7226 ret = fn(now, data);
7233 ldev = netdev_next_lower_dev_rcu(now, &iter);
7238 niter = &ldev->adj_list.lower;
7239 dev_stack[cur] = now;
7240 iter_stack[cur++] = iter;
7247 next = dev_stack[--cur];
7248 niter = iter_stack[cur];
7257 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7260 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7261 * lower neighbour list, RCU
7265 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7266 * list. The caller must hold RCU read lock.
7268 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7270 struct netdev_adjacent *lower;
7272 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7273 struct netdev_adjacent, list);
7275 return lower->private;
7278 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7281 * netdev_master_upper_dev_get_rcu - Get master upper device
7284 * Find a master upper device and return pointer to it or NULL in case
7285 * it's not there. The caller must hold the RCU read lock.
7287 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7289 struct netdev_adjacent *upper;
7291 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7292 struct netdev_adjacent, list);
7293 if (upper && likely(upper->master))
7297 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7299 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7300 struct net_device *adj_dev,
7301 struct list_head *dev_list)
7303 char linkname[IFNAMSIZ+7];
7305 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7306 "upper_%s" : "lower_%s", adj_dev->name);
7307 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7310 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7312 struct list_head *dev_list)
7314 char linkname[IFNAMSIZ+7];
7316 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7317 "upper_%s" : "lower_%s", name);
7318 sysfs_remove_link(&(dev->dev.kobj), linkname);
7321 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7322 struct net_device *adj_dev,
7323 struct list_head *dev_list)
7325 return (dev_list == &dev->adj_list.upper ||
7326 dev_list == &dev->adj_list.lower) &&
7327 net_eq(dev_net(dev), dev_net(adj_dev));
7330 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7331 struct net_device *adj_dev,
7332 struct list_head *dev_list,
7333 void *private, bool master)
7335 struct netdev_adjacent *adj;
7338 adj = __netdev_find_adj(adj_dev, dev_list);
7342 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7343 dev->name, adj_dev->name, adj->ref_nr);
7348 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7353 adj->master = master;
7355 adj->private = private;
7356 adj->ignore = false;
7359 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7360 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7362 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7363 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7368 /* Ensure that master link is always the first item in list. */
7370 ret = sysfs_create_link(&(dev->dev.kobj),
7371 &(adj_dev->dev.kobj), "master");
7373 goto remove_symlinks;
7375 list_add_rcu(&adj->list, dev_list);
7377 list_add_tail_rcu(&adj->list, dev_list);
7383 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7384 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7392 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7393 struct net_device *adj_dev,
7395 struct list_head *dev_list)
7397 struct netdev_adjacent *adj;
7399 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7400 dev->name, adj_dev->name, ref_nr);
7402 adj = __netdev_find_adj(adj_dev, dev_list);
7405 pr_err("Adjacency does not exist for device %s from %s\n",
7406 dev->name, adj_dev->name);
7411 if (adj->ref_nr > ref_nr) {
7412 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7413 dev->name, adj_dev->name, ref_nr,
7414 adj->ref_nr - ref_nr);
7415 adj->ref_nr -= ref_nr;
7420 sysfs_remove_link(&(dev->dev.kobj), "master");
7422 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7423 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7425 list_del_rcu(&adj->list);
7426 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7427 adj_dev->name, dev->name, adj_dev->name);
7429 kfree_rcu(adj, rcu);
7432 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7433 struct net_device *upper_dev,
7434 struct list_head *up_list,
7435 struct list_head *down_list,
7436 void *private, bool master)
7440 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7445 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7448 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7455 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7456 struct net_device *upper_dev,
7458 struct list_head *up_list,
7459 struct list_head *down_list)
7461 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7462 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7465 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7466 struct net_device *upper_dev,
7467 void *private, bool master)
7469 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7470 &dev->adj_list.upper,
7471 &upper_dev->adj_list.lower,
7475 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7476 struct net_device *upper_dev)
7478 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7479 &dev->adj_list.upper,
7480 &upper_dev->adj_list.lower);
7483 static int __netdev_upper_dev_link(struct net_device *dev,
7484 struct net_device *upper_dev, bool master,
7485 void *upper_priv, void *upper_info,
7486 struct netlink_ext_ack *extack)
7488 struct netdev_notifier_changeupper_info changeupper_info = {
7493 .upper_dev = upper_dev,
7496 .upper_info = upper_info,
7498 struct net_device *master_dev;
7503 if (dev == upper_dev)
7506 /* To prevent loops, check if dev is not upper device to upper_dev. */
7507 if (__netdev_has_upper_dev(upper_dev, dev))
7510 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7514 if (__netdev_has_upper_dev(dev, upper_dev))
7517 master_dev = __netdev_master_upper_dev_get(dev);
7519 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7522 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7523 &changeupper_info.info);
7524 ret = notifier_to_errno(ret);
7528 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7533 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7534 &changeupper_info.info);
7535 ret = notifier_to_errno(ret);
7539 __netdev_update_upper_level(dev, NULL);
7540 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7542 __netdev_update_lower_level(upper_dev, NULL);
7543 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7549 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7555 * netdev_upper_dev_link - Add a link to the upper device
7557 * @upper_dev: new upper device
7558 * @extack: netlink extended ack
7560 * Adds a link to device which is upper to this one. The caller must hold
7561 * the RTNL lock. On a failure a negative errno code is returned.
7562 * On success the reference counts are adjusted and the function
7565 int netdev_upper_dev_link(struct net_device *dev,
7566 struct net_device *upper_dev,
7567 struct netlink_ext_ack *extack)
7569 return __netdev_upper_dev_link(dev, upper_dev, false,
7570 NULL, NULL, extack);
7572 EXPORT_SYMBOL(netdev_upper_dev_link);
7575 * netdev_master_upper_dev_link - Add a master link to the upper device
7577 * @upper_dev: new upper device
7578 * @upper_priv: upper device private
7579 * @upper_info: upper info to be passed down via notifier
7580 * @extack: netlink extended ack
7582 * Adds a link to device which is upper to this one. In this case, only
7583 * one master upper device can be linked, although other non-master devices
7584 * might be linked as well. The caller must hold the RTNL lock.
7585 * On a failure a negative errno code is returned. On success the reference
7586 * counts are adjusted and the function returns zero.
7588 int netdev_master_upper_dev_link(struct net_device *dev,
7589 struct net_device *upper_dev,
7590 void *upper_priv, void *upper_info,
7591 struct netlink_ext_ack *extack)
7593 return __netdev_upper_dev_link(dev, upper_dev, true,
7594 upper_priv, upper_info, extack);
7596 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7599 * netdev_upper_dev_unlink - Removes a link to upper device
7601 * @upper_dev: new upper device
7603 * Removes a link to device which is upper to this one. The caller must hold
7606 void netdev_upper_dev_unlink(struct net_device *dev,
7607 struct net_device *upper_dev)
7609 struct netdev_notifier_changeupper_info changeupper_info = {
7613 .upper_dev = upper_dev,
7619 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7621 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7622 &changeupper_info.info);
7624 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7626 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7627 &changeupper_info.info);
7629 __netdev_update_upper_level(dev, NULL);
7630 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7632 __netdev_update_lower_level(upper_dev, NULL);
7633 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7636 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7638 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7639 struct net_device *lower_dev,
7642 struct netdev_adjacent *adj;
7644 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7648 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7653 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7654 struct net_device *lower_dev)
7656 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7659 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7660 struct net_device *lower_dev)
7662 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7665 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7666 struct net_device *new_dev,
7667 struct net_device *dev,
7668 struct netlink_ext_ack *extack)
7675 if (old_dev && new_dev != old_dev)
7676 netdev_adjacent_dev_disable(dev, old_dev);
7678 err = netdev_upper_dev_link(new_dev, dev, extack);
7680 if (old_dev && new_dev != old_dev)
7681 netdev_adjacent_dev_enable(dev, old_dev);
7687 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7689 void netdev_adjacent_change_commit(struct net_device *old_dev,
7690 struct net_device *new_dev,
7691 struct net_device *dev)
7693 if (!new_dev || !old_dev)
7696 if (new_dev == old_dev)
7699 netdev_adjacent_dev_enable(dev, old_dev);
7700 netdev_upper_dev_unlink(old_dev, dev);
7702 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7704 void netdev_adjacent_change_abort(struct net_device *old_dev,
7705 struct net_device *new_dev,
7706 struct net_device *dev)
7711 if (old_dev && new_dev != old_dev)
7712 netdev_adjacent_dev_enable(dev, old_dev);
7714 netdev_upper_dev_unlink(new_dev, dev);
7716 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7719 * netdev_bonding_info_change - Dispatch event about slave change
7721 * @bonding_info: info to dispatch
7723 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7724 * The caller must hold the RTNL lock.
7726 void netdev_bonding_info_change(struct net_device *dev,
7727 struct netdev_bonding_info *bonding_info)
7729 struct netdev_notifier_bonding_info info = {
7733 memcpy(&info.bonding_info, bonding_info,
7734 sizeof(struct netdev_bonding_info));
7735 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7738 EXPORT_SYMBOL(netdev_bonding_info_change);
7740 static void netdev_adjacent_add_links(struct net_device *dev)
7742 struct netdev_adjacent *iter;
7744 struct net *net = dev_net(dev);
7746 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7747 if (!net_eq(net, dev_net(iter->dev)))
7749 netdev_adjacent_sysfs_add(iter->dev, dev,
7750 &iter->dev->adj_list.lower);
7751 netdev_adjacent_sysfs_add(dev, iter->dev,
7752 &dev->adj_list.upper);
7755 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7756 if (!net_eq(net, dev_net(iter->dev)))
7758 netdev_adjacent_sysfs_add(iter->dev, dev,
7759 &iter->dev->adj_list.upper);
7760 netdev_adjacent_sysfs_add(dev, iter->dev,
7761 &dev->adj_list.lower);
7765 static void netdev_adjacent_del_links(struct net_device *dev)
7767 struct netdev_adjacent *iter;
7769 struct net *net = dev_net(dev);
7771 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7772 if (!net_eq(net, dev_net(iter->dev)))
7774 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7775 &iter->dev->adj_list.lower);
7776 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7777 &dev->adj_list.upper);
7780 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7781 if (!net_eq(net, dev_net(iter->dev)))
7783 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7784 &iter->dev->adj_list.upper);
7785 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7786 &dev->adj_list.lower);
7790 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
7792 struct netdev_adjacent *iter;
7794 struct net *net = dev_net(dev);
7796 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7797 if (!net_eq(net, dev_net(iter->dev)))
7799 netdev_adjacent_sysfs_del(iter->dev, oldname,
7800 &iter->dev->adj_list.lower);
7801 netdev_adjacent_sysfs_add(iter->dev, dev,
7802 &iter->dev->adj_list.lower);
7805 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7806 if (!net_eq(net, dev_net(iter->dev)))
7808 netdev_adjacent_sysfs_del(iter->dev, oldname,
7809 &iter->dev->adj_list.upper);
7810 netdev_adjacent_sysfs_add(iter->dev, dev,
7811 &iter->dev->adj_list.upper);
7815 void *netdev_lower_dev_get_private(struct net_device *dev,
7816 struct net_device *lower_dev)
7818 struct netdev_adjacent *lower;
7822 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
7826 return lower->private;
7828 EXPORT_SYMBOL(netdev_lower_dev_get_private);
7832 * netdev_lower_change - Dispatch event about lower device state change
7833 * @lower_dev: device
7834 * @lower_state_info: state to dispatch
7836 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
7837 * The caller must hold the RTNL lock.
7839 void netdev_lower_state_changed(struct net_device *lower_dev,
7840 void *lower_state_info)
7842 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
7843 .info.dev = lower_dev,
7847 changelowerstate_info.lower_state_info = lower_state_info;
7848 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
7849 &changelowerstate_info.info);
7851 EXPORT_SYMBOL(netdev_lower_state_changed);
7853 static void dev_change_rx_flags(struct net_device *dev, int flags)
7855 const struct net_device_ops *ops = dev->netdev_ops;
7857 if (ops->ndo_change_rx_flags)
7858 ops->ndo_change_rx_flags(dev, flags);
7861 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
7863 unsigned int old_flags = dev->flags;
7869 dev->flags |= IFF_PROMISC;
7870 dev->promiscuity += inc;
7871 if (dev->promiscuity == 0) {
7874 * If inc causes overflow, untouch promisc and return error.
7877 dev->flags &= ~IFF_PROMISC;
7879 dev->promiscuity -= inc;
7880 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
7885 if (dev->flags != old_flags) {
7886 pr_info("device %s %s promiscuous mode\n",
7888 dev->flags & IFF_PROMISC ? "entered" : "left");
7889 if (audit_enabled) {
7890 current_uid_gid(&uid, &gid);
7891 audit_log(audit_context(), GFP_ATOMIC,
7892 AUDIT_ANOM_PROMISCUOUS,
7893 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
7894 dev->name, (dev->flags & IFF_PROMISC),
7895 (old_flags & IFF_PROMISC),
7896 from_kuid(&init_user_ns, audit_get_loginuid(current)),
7897 from_kuid(&init_user_ns, uid),
7898 from_kgid(&init_user_ns, gid),
7899 audit_get_sessionid(current));
7902 dev_change_rx_flags(dev, IFF_PROMISC);
7905 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
7910 * dev_set_promiscuity - update promiscuity count on a device
7914 * Add or remove promiscuity from a device. While the count in the device
7915 * remains above zero the interface remains promiscuous. Once it hits zero
7916 * the device reverts back to normal filtering operation. A negative inc
7917 * value is used to drop promiscuity on the device.
7918 * Return 0 if successful or a negative errno code on error.
7920 int dev_set_promiscuity(struct net_device *dev, int inc)
7922 unsigned int old_flags = dev->flags;
7925 err = __dev_set_promiscuity(dev, inc, true);
7928 if (dev->flags != old_flags)
7929 dev_set_rx_mode(dev);
7932 EXPORT_SYMBOL(dev_set_promiscuity);
7934 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
7936 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
7940 dev->flags |= IFF_ALLMULTI;
7941 dev->allmulti += inc;
7942 if (dev->allmulti == 0) {
7945 * If inc causes overflow, untouch allmulti and return error.
7948 dev->flags &= ~IFF_ALLMULTI;
7950 dev->allmulti -= inc;
7951 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
7956 if (dev->flags ^ old_flags) {
7957 dev_change_rx_flags(dev, IFF_ALLMULTI);
7958 dev_set_rx_mode(dev);
7960 __dev_notify_flags(dev, old_flags,
7961 dev->gflags ^ old_gflags);
7967 * dev_set_allmulti - update allmulti count on a device
7971 * Add or remove reception of all multicast frames to a device. While the
7972 * count in the device remains above zero the interface remains listening
7973 * to all interfaces. Once it hits zero the device reverts back to normal
7974 * filtering operation. A negative @inc value is used to drop the counter
7975 * when releasing a resource needing all multicasts.
7976 * Return 0 if successful or a negative errno code on error.
7979 int dev_set_allmulti(struct net_device *dev, int inc)
7981 return __dev_set_allmulti(dev, inc, true);
7983 EXPORT_SYMBOL(dev_set_allmulti);
7986 * Upload unicast and multicast address lists to device and
7987 * configure RX filtering. When the device doesn't support unicast
7988 * filtering it is put in promiscuous mode while unicast addresses
7991 void __dev_set_rx_mode(struct net_device *dev)
7993 const struct net_device_ops *ops = dev->netdev_ops;
7995 /* dev_open will call this function so the list will stay sane. */
7996 if (!(dev->flags&IFF_UP))
7999 if (!netif_device_present(dev))
8002 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8003 /* Unicast addresses changes may only happen under the rtnl,
8004 * therefore calling __dev_set_promiscuity here is safe.
8006 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8007 __dev_set_promiscuity(dev, 1, false);
8008 dev->uc_promisc = true;
8009 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8010 __dev_set_promiscuity(dev, -1, false);
8011 dev->uc_promisc = false;
8015 if (ops->ndo_set_rx_mode)
8016 ops->ndo_set_rx_mode(dev);
8019 void dev_set_rx_mode(struct net_device *dev)
8021 netif_addr_lock_bh(dev);
8022 __dev_set_rx_mode(dev);
8023 netif_addr_unlock_bh(dev);
8027 * dev_get_flags - get flags reported to userspace
8030 * Get the combination of flag bits exported through APIs to userspace.
8032 unsigned int dev_get_flags(const struct net_device *dev)
8036 flags = (dev->flags & ~(IFF_PROMISC |
8041 (dev->gflags & (IFF_PROMISC |
8044 if (netif_running(dev)) {
8045 if (netif_oper_up(dev))
8046 flags |= IFF_RUNNING;
8047 if (netif_carrier_ok(dev))
8048 flags |= IFF_LOWER_UP;
8049 if (netif_dormant(dev))
8050 flags |= IFF_DORMANT;
8055 EXPORT_SYMBOL(dev_get_flags);
8057 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8058 struct netlink_ext_ack *extack)
8060 unsigned int old_flags = dev->flags;
8066 * Set the flags on our device.
8069 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8070 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8072 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8076 * Load in the correct multicast list now the flags have changed.
8079 if ((old_flags ^ flags) & IFF_MULTICAST)
8080 dev_change_rx_flags(dev, IFF_MULTICAST);
8082 dev_set_rx_mode(dev);
8085 * Have we downed the interface. We handle IFF_UP ourselves
8086 * according to user attempts to set it, rather than blindly
8091 if ((old_flags ^ flags) & IFF_UP) {
8092 if (old_flags & IFF_UP)
8095 ret = __dev_open(dev, extack);
8098 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8099 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8100 unsigned int old_flags = dev->flags;
8102 dev->gflags ^= IFF_PROMISC;
8104 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8105 if (dev->flags != old_flags)
8106 dev_set_rx_mode(dev);
8109 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8110 * is important. Some (broken) drivers set IFF_PROMISC, when
8111 * IFF_ALLMULTI is requested not asking us and not reporting.
8113 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8114 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8116 dev->gflags ^= IFF_ALLMULTI;
8117 __dev_set_allmulti(dev, inc, false);
8123 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8124 unsigned int gchanges)
8126 unsigned int changes = dev->flags ^ old_flags;
8129 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8131 if (changes & IFF_UP) {
8132 if (dev->flags & IFF_UP)
8133 call_netdevice_notifiers(NETDEV_UP, dev);
8135 call_netdevice_notifiers(NETDEV_DOWN, dev);
8138 if (dev->flags & IFF_UP &&
8139 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8140 struct netdev_notifier_change_info change_info = {
8144 .flags_changed = changes,
8147 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8152 * dev_change_flags - change device settings
8154 * @flags: device state flags
8155 * @extack: netlink extended ack
8157 * Change settings on device based state flags. The flags are
8158 * in the userspace exported format.
8160 int dev_change_flags(struct net_device *dev, unsigned int flags,
8161 struct netlink_ext_ack *extack)
8164 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8166 ret = __dev_change_flags(dev, flags, extack);
8170 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8171 __dev_notify_flags(dev, old_flags, changes);
8174 EXPORT_SYMBOL(dev_change_flags);
8176 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8178 const struct net_device_ops *ops = dev->netdev_ops;
8180 if (ops->ndo_change_mtu)
8181 return ops->ndo_change_mtu(dev, new_mtu);
8183 /* Pairs with all the lockless reads of dev->mtu in the stack */
8184 WRITE_ONCE(dev->mtu, new_mtu);
8187 EXPORT_SYMBOL(__dev_set_mtu);
8189 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8190 struct netlink_ext_ack *extack)
8192 /* MTU must be positive, and in range */
8193 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8194 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8198 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8199 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8206 * dev_set_mtu_ext - Change maximum transfer unit
8208 * @new_mtu: new transfer unit
8209 * @extack: netlink extended ack
8211 * Change the maximum transfer size of the network device.
8213 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8214 struct netlink_ext_ack *extack)
8218 if (new_mtu == dev->mtu)
8221 err = dev_validate_mtu(dev, new_mtu, extack);
8225 if (!netif_device_present(dev))
8228 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8229 err = notifier_to_errno(err);
8233 orig_mtu = dev->mtu;
8234 err = __dev_set_mtu(dev, new_mtu);
8237 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8239 err = notifier_to_errno(err);
8241 /* setting mtu back and notifying everyone again,
8242 * so that they have a chance to revert changes.
8244 __dev_set_mtu(dev, orig_mtu);
8245 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8252 int dev_set_mtu(struct net_device *dev, int new_mtu)
8254 struct netlink_ext_ack extack;
8257 memset(&extack, 0, sizeof(extack));
8258 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8259 if (err && extack._msg)
8260 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8263 EXPORT_SYMBOL(dev_set_mtu);
8266 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8268 * @new_len: new tx queue length
8270 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8272 unsigned int orig_len = dev->tx_queue_len;
8275 if (new_len != (unsigned int)new_len)
8278 if (new_len != orig_len) {
8279 dev->tx_queue_len = new_len;
8280 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8281 res = notifier_to_errno(res);
8284 res = dev_qdisc_change_tx_queue_len(dev);
8292 netdev_err(dev, "refused to change device tx_queue_len\n");
8293 dev->tx_queue_len = orig_len;
8298 * dev_set_group - Change group this device belongs to
8300 * @new_group: group this device should belong to
8302 void dev_set_group(struct net_device *dev, int new_group)
8304 dev->group = new_group;
8306 EXPORT_SYMBOL(dev_set_group);
8309 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8311 * @addr: new address
8312 * @extack: netlink extended ack
8314 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8315 struct netlink_ext_ack *extack)
8317 struct netdev_notifier_pre_changeaddr_info info = {
8319 .info.extack = extack,
8324 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8325 return notifier_to_errno(rc);
8327 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8330 * dev_set_mac_address - Change Media Access Control Address
8333 * @extack: netlink extended ack
8335 * Change the hardware (MAC) address of the device
8337 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8338 struct netlink_ext_ack *extack)
8340 const struct net_device_ops *ops = dev->netdev_ops;
8343 if (!ops->ndo_set_mac_address)
8345 if (sa->sa_family != dev->type)
8347 if (!netif_device_present(dev))
8349 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8352 err = ops->ndo_set_mac_address(dev, sa);
8355 dev->addr_assign_type = NET_ADDR_SET;
8356 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8357 add_device_randomness(dev->dev_addr, dev->addr_len);
8360 EXPORT_SYMBOL(dev_set_mac_address);
8363 * dev_change_carrier - Change device carrier
8365 * @new_carrier: new value
8367 * Change device carrier
8369 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8371 const struct net_device_ops *ops = dev->netdev_ops;
8373 if (!ops->ndo_change_carrier)
8375 if (!netif_device_present(dev))
8377 return ops->ndo_change_carrier(dev, new_carrier);
8379 EXPORT_SYMBOL(dev_change_carrier);
8382 * dev_get_phys_port_id - Get device physical port ID
8386 * Get device physical port ID
8388 int dev_get_phys_port_id(struct net_device *dev,
8389 struct netdev_phys_item_id *ppid)
8391 const struct net_device_ops *ops = dev->netdev_ops;
8393 if (!ops->ndo_get_phys_port_id)
8395 return ops->ndo_get_phys_port_id(dev, ppid);
8397 EXPORT_SYMBOL(dev_get_phys_port_id);
8400 * dev_get_phys_port_name - Get device physical port name
8403 * @len: limit of bytes to copy to name
8405 * Get device physical port name
8407 int dev_get_phys_port_name(struct net_device *dev,
8408 char *name, size_t len)
8410 const struct net_device_ops *ops = dev->netdev_ops;
8413 if (ops->ndo_get_phys_port_name) {
8414 err = ops->ndo_get_phys_port_name(dev, name, len);
8415 if (err != -EOPNOTSUPP)
8418 return devlink_compat_phys_port_name_get(dev, name, len);
8420 EXPORT_SYMBOL(dev_get_phys_port_name);
8423 * dev_get_port_parent_id - Get the device's port parent identifier
8424 * @dev: network device
8425 * @ppid: pointer to a storage for the port's parent identifier
8426 * @recurse: allow/disallow recursion to lower devices
8428 * Get the devices's port parent identifier
8430 int dev_get_port_parent_id(struct net_device *dev,
8431 struct netdev_phys_item_id *ppid,
8434 const struct net_device_ops *ops = dev->netdev_ops;
8435 struct netdev_phys_item_id first = { };
8436 struct net_device *lower_dev;
8437 struct list_head *iter;
8440 if (ops->ndo_get_port_parent_id) {
8441 err = ops->ndo_get_port_parent_id(dev, ppid);
8442 if (err != -EOPNOTSUPP)
8446 err = devlink_compat_switch_id_get(dev, ppid);
8447 if (!err || err != -EOPNOTSUPP)
8453 netdev_for_each_lower_dev(dev, lower_dev, iter) {
8454 err = dev_get_port_parent_id(lower_dev, ppid, recurse);
8459 else if (memcmp(&first, ppid, sizeof(*ppid)))
8465 EXPORT_SYMBOL(dev_get_port_parent_id);
8468 * netdev_port_same_parent_id - Indicate if two network devices have
8469 * the same port parent identifier
8470 * @a: first network device
8471 * @b: second network device
8473 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8475 struct netdev_phys_item_id a_id = { };
8476 struct netdev_phys_item_id b_id = { };
8478 if (dev_get_port_parent_id(a, &a_id, true) ||
8479 dev_get_port_parent_id(b, &b_id, true))
8482 return netdev_phys_item_id_same(&a_id, &b_id);
8484 EXPORT_SYMBOL(netdev_port_same_parent_id);
8487 * dev_change_proto_down - update protocol port state information
8489 * @proto_down: new value
8491 * This info can be used by switch drivers to set the phys state of the
8494 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8496 const struct net_device_ops *ops = dev->netdev_ops;
8498 if (!ops->ndo_change_proto_down)
8500 if (!netif_device_present(dev))
8502 return ops->ndo_change_proto_down(dev, proto_down);
8504 EXPORT_SYMBOL(dev_change_proto_down);
8507 * dev_change_proto_down_generic - generic implementation for
8508 * ndo_change_proto_down that sets carrier according to
8512 * @proto_down: new value
8514 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
8517 netif_carrier_off(dev);
8519 netif_carrier_on(dev);
8520 dev->proto_down = proto_down;
8523 EXPORT_SYMBOL(dev_change_proto_down_generic);
8525 u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op,
8526 enum bpf_netdev_command cmd)
8528 struct netdev_bpf xdp;
8533 memset(&xdp, 0, sizeof(xdp));
8536 /* Query must always succeed. */
8537 WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG);
8542 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op,
8543 struct netlink_ext_ack *extack, u32 flags,
8544 struct bpf_prog *prog)
8546 bool non_hw = !(flags & XDP_FLAGS_HW_MODE);
8547 struct bpf_prog *prev_prog = NULL;
8548 struct netdev_bpf xdp;
8552 prev_prog = bpf_prog_by_id(__dev_xdp_query(dev, bpf_op,
8554 if (IS_ERR(prev_prog))
8558 memset(&xdp, 0, sizeof(xdp));
8559 if (flags & XDP_FLAGS_HW_MODE)
8560 xdp.command = XDP_SETUP_PROG_HW;
8562 xdp.command = XDP_SETUP_PROG;
8563 xdp.extack = extack;
8567 err = bpf_op(dev, &xdp);
8569 bpf_prog_change_xdp(prev_prog, prog);
8572 bpf_prog_put(prev_prog);
8577 static void dev_xdp_uninstall(struct net_device *dev)
8579 struct netdev_bpf xdp;
8582 /* Remove generic XDP */
8583 WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL));
8585 /* Remove from the driver */
8586 ndo_bpf = dev->netdev_ops->ndo_bpf;
8590 memset(&xdp, 0, sizeof(xdp));
8591 xdp.command = XDP_QUERY_PROG;
8592 WARN_ON(ndo_bpf(dev, &xdp));
8594 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8597 /* Remove HW offload */
8598 memset(&xdp, 0, sizeof(xdp));
8599 xdp.command = XDP_QUERY_PROG_HW;
8600 if (!ndo_bpf(dev, &xdp) && xdp.prog_id)
8601 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8606 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
8608 * @extack: netlink extended ack
8609 * @fd: new program fd or negative value to clear
8610 * @flags: xdp-related flags
8612 * Set or clear a bpf program for a device
8614 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
8617 const struct net_device_ops *ops = dev->netdev_ops;
8618 enum bpf_netdev_command query;
8619 struct bpf_prog *prog = NULL;
8620 bpf_op_t bpf_op, bpf_chk;
8626 offload = flags & XDP_FLAGS_HW_MODE;
8627 query = offload ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG;
8629 bpf_op = bpf_chk = ops->ndo_bpf;
8630 if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) {
8631 NL_SET_ERR_MSG(extack, "underlying driver does not support XDP in native mode");
8634 if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE))
8635 bpf_op = generic_xdp_install;
8636 if (bpf_op == bpf_chk)
8637 bpf_chk = generic_xdp_install;
8642 if (!offload && __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG)) {
8643 NL_SET_ERR_MSG(extack, "native and generic XDP can't be active at the same time");
8647 prog_id = __dev_xdp_query(dev, bpf_op, query);
8648 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && prog_id) {
8649 NL_SET_ERR_MSG(extack, "XDP program already attached");
8653 prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
8654 bpf_op == ops->ndo_bpf);
8656 return PTR_ERR(prog);
8658 if (!offload && bpf_prog_is_dev_bound(prog->aux)) {
8659 NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported");
8664 /* prog->aux->id may be 0 for orphaned device-bound progs */
8665 if (prog->aux->id && prog->aux->id == prog_id) {
8670 if (!__dev_xdp_query(dev, bpf_op, query))
8674 err = dev_xdp_install(dev, bpf_op, extack, flags, prog);
8675 if (err < 0 && prog)
8682 * dev_new_index - allocate an ifindex
8683 * @net: the applicable net namespace
8685 * Returns a suitable unique value for a new device interface
8686 * number. The caller must hold the rtnl semaphore or the
8687 * dev_base_lock to be sure it remains unique.
8689 static int dev_new_index(struct net *net)
8691 int ifindex = net->ifindex;
8696 if (!__dev_get_by_index(net, ifindex))
8697 return net->ifindex = ifindex;
8701 /* Delayed registration/unregisteration */
8702 static LIST_HEAD(net_todo_list);
8703 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
8705 static void net_set_todo(struct net_device *dev)
8707 list_add_tail(&dev->todo_list, &net_todo_list);
8708 dev_net(dev)->dev_unreg_count++;
8711 static void rollback_registered_many(struct list_head *head)
8713 struct net_device *dev, *tmp;
8714 LIST_HEAD(close_head);
8716 BUG_ON(dev_boot_phase);
8719 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
8720 /* Some devices call without registering
8721 * for initialization unwind. Remove those
8722 * devices and proceed with the remaining.
8724 if (dev->reg_state == NETREG_UNINITIALIZED) {
8725 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
8729 list_del(&dev->unreg_list);
8732 dev->dismantle = true;
8733 BUG_ON(dev->reg_state != NETREG_REGISTERED);
8736 /* If device is running, close it first. */
8737 list_for_each_entry(dev, head, unreg_list)
8738 list_add_tail(&dev->close_list, &close_head);
8739 dev_close_many(&close_head, true);
8741 list_for_each_entry(dev, head, unreg_list) {
8742 /* And unlink it from device chain. */
8743 unlist_netdevice(dev);
8745 dev->reg_state = NETREG_UNREGISTERING;
8747 flush_all_backlogs();
8751 list_for_each_entry(dev, head, unreg_list) {
8752 struct sk_buff *skb = NULL;
8754 /* Shutdown queueing discipline. */
8757 dev_xdp_uninstall(dev);
8759 /* Notify protocols, that we are about to destroy
8760 * this device. They should clean all the things.
8762 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8764 if (!dev->rtnl_link_ops ||
8765 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8766 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
8767 GFP_KERNEL, NULL, 0);
8770 * Flush the unicast and multicast chains
8775 netdev_name_node_alt_flush(dev);
8776 netdev_name_node_free(dev->name_node);
8778 if (dev->netdev_ops->ndo_uninit)
8779 dev->netdev_ops->ndo_uninit(dev);
8782 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
8784 /* Notifier chain MUST detach us all upper devices. */
8785 WARN_ON(netdev_has_any_upper_dev(dev));
8786 WARN_ON(netdev_has_any_lower_dev(dev));
8788 /* Remove entries from kobject tree */
8789 netdev_unregister_kobject(dev);
8791 /* Remove XPS queueing entries */
8792 netif_reset_xps_queues_gt(dev, 0);
8798 list_for_each_entry(dev, head, unreg_list)
8802 static void rollback_registered(struct net_device *dev)
8806 list_add(&dev->unreg_list, &single);
8807 rollback_registered_many(&single);
8811 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
8812 struct net_device *upper, netdev_features_t features)
8814 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8815 netdev_features_t feature;
8818 for_each_netdev_feature(upper_disables, feature_bit) {
8819 feature = __NETIF_F_BIT(feature_bit);
8820 if (!(upper->wanted_features & feature)
8821 && (features & feature)) {
8822 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
8823 &feature, upper->name);
8824 features &= ~feature;
8831 static void netdev_sync_lower_features(struct net_device *upper,
8832 struct net_device *lower, netdev_features_t features)
8834 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8835 netdev_features_t feature;
8838 for_each_netdev_feature(upper_disables, feature_bit) {
8839 feature = __NETIF_F_BIT(feature_bit);
8840 if (!(features & feature) && (lower->features & feature)) {
8841 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
8842 &feature, lower->name);
8843 lower->wanted_features &= ~feature;
8844 netdev_update_features(lower);
8846 if (unlikely(lower->features & feature))
8847 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
8848 &feature, lower->name);
8853 static netdev_features_t netdev_fix_features(struct net_device *dev,
8854 netdev_features_t features)
8856 /* Fix illegal checksum combinations */
8857 if ((features & NETIF_F_HW_CSUM) &&
8858 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
8859 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
8860 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
8863 /* TSO requires that SG is present as well. */
8864 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
8865 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
8866 features &= ~NETIF_F_ALL_TSO;
8869 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
8870 !(features & NETIF_F_IP_CSUM)) {
8871 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
8872 features &= ~NETIF_F_TSO;
8873 features &= ~NETIF_F_TSO_ECN;
8876 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
8877 !(features & NETIF_F_IPV6_CSUM)) {
8878 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
8879 features &= ~NETIF_F_TSO6;
8882 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
8883 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
8884 features &= ~NETIF_F_TSO_MANGLEID;
8886 /* TSO ECN requires that TSO is present as well. */
8887 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
8888 features &= ~NETIF_F_TSO_ECN;
8890 /* Software GSO depends on SG. */
8891 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
8892 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
8893 features &= ~NETIF_F_GSO;
8896 /* GSO partial features require GSO partial be set */
8897 if ((features & dev->gso_partial_features) &&
8898 !(features & NETIF_F_GSO_PARTIAL)) {
8900 "Dropping partially supported GSO features since no GSO partial.\n");
8901 features &= ~dev->gso_partial_features;
8904 if (!(features & NETIF_F_RXCSUM)) {
8905 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
8906 * successfully merged by hardware must also have the
8907 * checksum verified by hardware. If the user does not
8908 * want to enable RXCSUM, logically, we should disable GRO_HW.
8910 if (features & NETIF_F_GRO_HW) {
8911 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
8912 features &= ~NETIF_F_GRO_HW;
8916 /* LRO/HW-GRO features cannot be combined with RX-FCS */
8917 if (features & NETIF_F_RXFCS) {
8918 if (features & NETIF_F_LRO) {
8919 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
8920 features &= ~NETIF_F_LRO;
8923 if (features & NETIF_F_GRO_HW) {
8924 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
8925 features &= ~NETIF_F_GRO_HW;
8932 int __netdev_update_features(struct net_device *dev)
8934 struct net_device *upper, *lower;
8935 netdev_features_t features;
8936 struct list_head *iter;
8941 features = netdev_get_wanted_features(dev);
8943 if (dev->netdev_ops->ndo_fix_features)
8944 features = dev->netdev_ops->ndo_fix_features(dev, features);
8946 /* driver might be less strict about feature dependencies */
8947 features = netdev_fix_features(dev, features);
8949 /* some features can't be enabled if they're off an an upper device */
8950 netdev_for_each_upper_dev_rcu(dev, upper, iter)
8951 features = netdev_sync_upper_features(dev, upper, features);
8953 if (dev->features == features)
8956 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
8957 &dev->features, &features);
8959 if (dev->netdev_ops->ndo_set_features)
8960 err = dev->netdev_ops->ndo_set_features(dev, features);
8964 if (unlikely(err < 0)) {
8966 "set_features() failed (%d); wanted %pNF, left %pNF\n",
8967 err, &features, &dev->features);
8968 /* return non-0 since some features might have changed and
8969 * it's better to fire a spurious notification than miss it
8975 /* some features must be disabled on lower devices when disabled
8976 * on an upper device (think: bonding master or bridge)
8978 netdev_for_each_lower_dev(dev, lower, iter)
8979 netdev_sync_lower_features(dev, lower, features);
8982 netdev_features_t diff = features ^ dev->features;
8984 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
8985 /* udp_tunnel_{get,drop}_rx_info both need
8986 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
8987 * device, or they won't do anything.
8988 * Thus we need to update dev->features
8989 * *before* calling udp_tunnel_get_rx_info,
8990 * but *after* calling udp_tunnel_drop_rx_info.
8992 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
8993 dev->features = features;
8994 udp_tunnel_get_rx_info(dev);
8996 udp_tunnel_drop_rx_info(dev);
9000 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9001 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9002 dev->features = features;
9003 err |= vlan_get_rx_ctag_filter_info(dev);
9005 vlan_drop_rx_ctag_filter_info(dev);
9009 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9010 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9011 dev->features = features;
9012 err |= vlan_get_rx_stag_filter_info(dev);
9014 vlan_drop_rx_stag_filter_info(dev);
9018 dev->features = features;
9021 return err < 0 ? 0 : 1;
9025 * netdev_update_features - recalculate device features
9026 * @dev: the device to check
9028 * Recalculate dev->features set and send notifications if it
9029 * has changed. Should be called after driver or hardware dependent
9030 * conditions might have changed that influence the features.
9032 void netdev_update_features(struct net_device *dev)
9034 if (__netdev_update_features(dev))
9035 netdev_features_change(dev);
9037 EXPORT_SYMBOL(netdev_update_features);
9040 * netdev_change_features - recalculate device features
9041 * @dev: the device to check
9043 * Recalculate dev->features set and send notifications even
9044 * if they have not changed. Should be called instead of
9045 * netdev_update_features() if also dev->vlan_features might
9046 * have changed to allow the changes to be propagated to stacked
9049 void netdev_change_features(struct net_device *dev)
9051 __netdev_update_features(dev);
9052 netdev_features_change(dev);
9054 EXPORT_SYMBOL(netdev_change_features);
9057 * netif_stacked_transfer_operstate - transfer operstate
9058 * @rootdev: the root or lower level device to transfer state from
9059 * @dev: the device to transfer operstate to
9061 * Transfer operational state from root to device. This is normally
9062 * called when a stacking relationship exists between the root
9063 * device and the device(a leaf device).
9065 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9066 struct net_device *dev)
9068 if (rootdev->operstate == IF_OPER_DORMANT)
9069 netif_dormant_on(dev);
9071 netif_dormant_off(dev);
9073 if (netif_carrier_ok(rootdev))
9074 netif_carrier_on(dev);
9076 netif_carrier_off(dev);
9078 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9080 static int netif_alloc_rx_queues(struct net_device *dev)
9082 unsigned int i, count = dev->num_rx_queues;
9083 struct netdev_rx_queue *rx;
9084 size_t sz = count * sizeof(*rx);
9089 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9095 for (i = 0; i < count; i++) {
9098 /* XDP RX-queue setup */
9099 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i);
9106 /* Rollback successful reg's and free other resources */
9108 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9114 static void netif_free_rx_queues(struct net_device *dev)
9116 unsigned int i, count = dev->num_rx_queues;
9118 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9122 for (i = 0; i < count; i++)
9123 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9128 static void netdev_init_one_queue(struct net_device *dev,
9129 struct netdev_queue *queue, void *_unused)
9131 /* Initialize queue lock */
9132 spin_lock_init(&queue->_xmit_lock);
9133 lockdep_set_class(&queue->_xmit_lock, &dev->qdisc_xmit_lock_key);
9134 queue->xmit_lock_owner = -1;
9135 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9138 dql_init(&queue->dql, HZ);
9142 static void netif_free_tx_queues(struct net_device *dev)
9147 static int netif_alloc_netdev_queues(struct net_device *dev)
9149 unsigned int count = dev->num_tx_queues;
9150 struct netdev_queue *tx;
9151 size_t sz = count * sizeof(*tx);
9153 if (count < 1 || count > 0xffff)
9156 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9162 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9163 spin_lock_init(&dev->tx_global_lock);
9168 void netif_tx_stop_all_queues(struct net_device *dev)
9172 for (i = 0; i < dev->num_tx_queues; i++) {
9173 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9175 netif_tx_stop_queue(txq);
9178 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9180 static void netdev_register_lockdep_key(struct net_device *dev)
9182 lockdep_register_key(&dev->qdisc_tx_busylock_key);
9183 lockdep_register_key(&dev->qdisc_running_key);
9184 lockdep_register_key(&dev->qdisc_xmit_lock_key);
9185 lockdep_register_key(&dev->addr_list_lock_key);
9188 static void netdev_unregister_lockdep_key(struct net_device *dev)
9190 lockdep_unregister_key(&dev->qdisc_tx_busylock_key);
9191 lockdep_unregister_key(&dev->qdisc_running_key);
9192 lockdep_unregister_key(&dev->qdisc_xmit_lock_key);
9193 lockdep_unregister_key(&dev->addr_list_lock_key);
9196 void netdev_update_lockdep_key(struct net_device *dev)
9198 lockdep_unregister_key(&dev->addr_list_lock_key);
9199 lockdep_register_key(&dev->addr_list_lock_key);
9201 lockdep_set_class(&dev->addr_list_lock, &dev->addr_list_lock_key);
9203 EXPORT_SYMBOL(netdev_update_lockdep_key);
9206 * register_netdevice - register a network device
9207 * @dev: device to register
9209 * Take a completed network device structure and add it to the kernel
9210 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9211 * chain. 0 is returned on success. A negative errno code is returned
9212 * on a failure to set up the device, or if the name is a duplicate.
9214 * Callers must hold the rtnl semaphore. You may want
9215 * register_netdev() instead of this.
9218 * The locking appears insufficient to guarantee two parallel registers
9219 * will not get the same name.
9222 int register_netdevice(struct net_device *dev)
9225 struct net *net = dev_net(dev);
9227 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9228 NETDEV_FEATURE_COUNT);
9229 BUG_ON(dev_boot_phase);
9234 /* When net_device's are persistent, this will be fatal. */
9235 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9238 spin_lock_init(&dev->addr_list_lock);
9239 lockdep_set_class(&dev->addr_list_lock, &dev->addr_list_lock_key);
9241 ret = dev_get_valid_name(net, dev, dev->name);
9246 dev->name_node = netdev_name_node_head_alloc(dev);
9247 if (!dev->name_node)
9250 /* Init, if this function is available */
9251 if (dev->netdev_ops->ndo_init) {
9252 ret = dev->netdev_ops->ndo_init(dev);
9260 if (((dev->hw_features | dev->features) &
9261 NETIF_F_HW_VLAN_CTAG_FILTER) &&
9262 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
9263 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
9264 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
9271 dev->ifindex = dev_new_index(net);
9272 else if (__dev_get_by_index(net, dev->ifindex))
9275 /* Transfer changeable features to wanted_features and enable
9276 * software offloads (GSO and GRO).
9278 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
9279 dev->features |= NETIF_F_SOFT_FEATURES;
9281 if (dev->netdev_ops->ndo_udp_tunnel_add) {
9282 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9283 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9286 dev->wanted_features = dev->features & dev->hw_features;
9288 if (!(dev->flags & IFF_LOOPBACK))
9289 dev->hw_features |= NETIF_F_NOCACHE_COPY;
9291 /* If IPv4 TCP segmentation offload is supported we should also
9292 * allow the device to enable segmenting the frame with the option
9293 * of ignoring a static IP ID value. This doesn't enable the
9294 * feature itself but allows the user to enable it later.
9296 if (dev->hw_features & NETIF_F_TSO)
9297 dev->hw_features |= NETIF_F_TSO_MANGLEID;
9298 if (dev->vlan_features & NETIF_F_TSO)
9299 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
9300 if (dev->mpls_features & NETIF_F_TSO)
9301 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
9302 if (dev->hw_enc_features & NETIF_F_TSO)
9303 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
9305 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
9307 dev->vlan_features |= NETIF_F_HIGHDMA;
9309 /* Make NETIF_F_SG inheritable to tunnel devices.
9311 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
9313 /* Make NETIF_F_SG inheritable to MPLS.
9315 dev->mpls_features |= NETIF_F_SG;
9317 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
9318 ret = notifier_to_errno(ret);
9322 ret = netdev_register_kobject(dev);
9324 dev->reg_state = NETREG_UNREGISTERED;
9327 dev->reg_state = NETREG_REGISTERED;
9329 __netdev_update_features(dev);
9332 * Default initial state at registry is that the
9333 * device is present.
9336 set_bit(__LINK_STATE_PRESENT, &dev->state);
9338 linkwatch_init_dev(dev);
9340 dev_init_scheduler(dev);
9342 list_netdevice(dev);
9343 add_device_randomness(dev->dev_addr, dev->addr_len);
9345 /* If the device has permanent device address, driver should
9346 * set dev_addr and also addr_assign_type should be set to
9347 * NET_ADDR_PERM (default value).
9349 if (dev->addr_assign_type == NET_ADDR_PERM)
9350 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
9352 /* Notify protocols, that a new device appeared. */
9353 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
9354 ret = notifier_to_errno(ret);
9356 rollback_registered(dev);
9359 dev->reg_state = NETREG_UNREGISTERED;
9362 * Prevent userspace races by waiting until the network
9363 * device is fully setup before sending notifications.
9365 if (!dev->rtnl_link_ops ||
9366 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
9367 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
9373 if (dev->netdev_ops->ndo_uninit)
9374 dev->netdev_ops->ndo_uninit(dev);
9375 if (dev->priv_destructor)
9376 dev->priv_destructor(dev);
9378 netdev_name_node_free(dev->name_node);
9381 EXPORT_SYMBOL(register_netdevice);
9384 * init_dummy_netdev - init a dummy network device for NAPI
9385 * @dev: device to init
9387 * This takes a network device structure and initialize the minimum
9388 * amount of fields so it can be used to schedule NAPI polls without
9389 * registering a full blown interface. This is to be used by drivers
9390 * that need to tie several hardware interfaces to a single NAPI
9391 * poll scheduler due to HW limitations.
9393 int init_dummy_netdev(struct net_device *dev)
9395 /* Clear everything. Note we don't initialize spinlocks
9396 * are they aren't supposed to be taken by any of the
9397 * NAPI code and this dummy netdev is supposed to be
9398 * only ever used for NAPI polls
9400 memset(dev, 0, sizeof(struct net_device));
9402 /* make sure we BUG if trying to hit standard
9403 * register/unregister code path
9405 dev->reg_state = NETREG_DUMMY;
9407 /* NAPI wants this */
9408 INIT_LIST_HEAD(&dev->napi_list);
9410 /* a dummy interface is started by default */
9411 set_bit(__LINK_STATE_PRESENT, &dev->state);
9412 set_bit(__LINK_STATE_START, &dev->state);
9414 /* napi_busy_loop stats accounting wants this */
9415 dev_net_set(dev, &init_net);
9417 /* Note : We dont allocate pcpu_refcnt for dummy devices,
9418 * because users of this 'device' dont need to change
9424 EXPORT_SYMBOL_GPL(init_dummy_netdev);
9428 * register_netdev - register a network device
9429 * @dev: device to register
9431 * Take a completed network device structure and add it to the kernel
9432 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9433 * chain. 0 is returned on success. A negative errno code is returned
9434 * on a failure to set up the device, or if the name is a duplicate.
9436 * This is a wrapper around register_netdevice that takes the rtnl semaphore
9437 * and expands the device name if you passed a format string to
9440 int register_netdev(struct net_device *dev)
9444 if (rtnl_lock_killable())
9446 err = register_netdevice(dev);
9450 EXPORT_SYMBOL(register_netdev);
9452 int netdev_refcnt_read(const struct net_device *dev)
9456 for_each_possible_cpu(i)
9457 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
9460 EXPORT_SYMBOL(netdev_refcnt_read);
9463 * netdev_wait_allrefs - wait until all references are gone.
9464 * @dev: target net_device
9466 * This is called when unregistering network devices.
9468 * Any protocol or device that holds a reference should register
9469 * for netdevice notification, and cleanup and put back the
9470 * reference if they receive an UNREGISTER event.
9471 * We can get stuck here if buggy protocols don't correctly
9474 static void netdev_wait_allrefs(struct net_device *dev)
9476 unsigned long rebroadcast_time, warning_time;
9479 linkwatch_forget_dev(dev);
9481 rebroadcast_time = warning_time = jiffies;
9482 refcnt = netdev_refcnt_read(dev);
9484 while (refcnt != 0) {
9485 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
9488 /* Rebroadcast unregister notification */
9489 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
9495 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
9497 /* We must not have linkwatch events
9498 * pending on unregister. If this
9499 * happens, we simply run the queue
9500 * unscheduled, resulting in a noop
9503 linkwatch_run_queue();
9508 rebroadcast_time = jiffies;
9513 refcnt = netdev_refcnt_read(dev);
9515 if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) {
9516 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
9518 warning_time = jiffies;
9527 * register_netdevice(x1);
9528 * register_netdevice(x2);
9530 * unregister_netdevice(y1);
9531 * unregister_netdevice(y2);
9537 * We are invoked by rtnl_unlock().
9538 * This allows us to deal with problems:
9539 * 1) We can delete sysfs objects which invoke hotplug
9540 * without deadlocking with linkwatch via keventd.
9541 * 2) Since we run with the RTNL semaphore not held, we can sleep
9542 * safely in order to wait for the netdev refcnt to drop to zero.
9544 * We must not return until all unregister events added during
9545 * the interval the lock was held have been completed.
9547 void netdev_run_todo(void)
9549 struct list_head list;
9551 /* Snapshot list, allow later requests */
9552 list_replace_init(&net_todo_list, &list);
9557 /* Wait for rcu callbacks to finish before next phase */
9558 if (!list_empty(&list))
9561 while (!list_empty(&list)) {
9562 struct net_device *dev
9563 = list_first_entry(&list, struct net_device, todo_list);
9564 list_del(&dev->todo_list);
9566 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
9567 pr_err("network todo '%s' but state %d\n",
9568 dev->name, dev->reg_state);
9573 dev->reg_state = NETREG_UNREGISTERED;
9575 netdev_wait_allrefs(dev);
9578 BUG_ON(netdev_refcnt_read(dev));
9579 BUG_ON(!list_empty(&dev->ptype_all));
9580 BUG_ON(!list_empty(&dev->ptype_specific));
9581 WARN_ON(rcu_access_pointer(dev->ip_ptr));
9582 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
9583 #if IS_ENABLED(CONFIG_DECNET)
9584 WARN_ON(dev->dn_ptr);
9586 if (dev->priv_destructor)
9587 dev->priv_destructor(dev);
9588 if (dev->needs_free_netdev)
9591 /* Report a network device has been unregistered */
9593 dev_net(dev)->dev_unreg_count--;
9595 wake_up(&netdev_unregistering_wq);
9597 /* Free network device */
9598 kobject_put(&dev->dev.kobj);
9602 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
9603 * all the same fields in the same order as net_device_stats, with only
9604 * the type differing, but rtnl_link_stats64 may have additional fields
9605 * at the end for newer counters.
9607 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
9608 const struct net_device_stats *netdev_stats)
9610 #if BITS_PER_LONG == 64
9611 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
9612 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
9613 /* zero out counters that only exist in rtnl_link_stats64 */
9614 memset((char *)stats64 + sizeof(*netdev_stats), 0,
9615 sizeof(*stats64) - sizeof(*netdev_stats));
9617 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
9618 const unsigned long *src = (const unsigned long *)netdev_stats;
9619 u64 *dst = (u64 *)stats64;
9621 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
9622 for (i = 0; i < n; i++)
9624 /* zero out counters that only exist in rtnl_link_stats64 */
9625 memset((char *)stats64 + n * sizeof(u64), 0,
9626 sizeof(*stats64) - n * sizeof(u64));
9629 EXPORT_SYMBOL(netdev_stats_to_stats64);
9632 * dev_get_stats - get network device statistics
9633 * @dev: device to get statistics from
9634 * @storage: place to store stats
9636 * Get network statistics from device. Return @storage.
9637 * The device driver may provide its own method by setting
9638 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
9639 * otherwise the internal statistics structure is used.
9641 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
9642 struct rtnl_link_stats64 *storage)
9644 const struct net_device_ops *ops = dev->netdev_ops;
9646 if (ops->ndo_get_stats64) {
9647 memset(storage, 0, sizeof(*storage));
9648 ops->ndo_get_stats64(dev, storage);
9649 } else if (ops->ndo_get_stats) {
9650 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
9652 netdev_stats_to_stats64(storage, &dev->stats);
9654 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
9655 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
9656 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
9659 EXPORT_SYMBOL(dev_get_stats);
9661 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
9663 struct netdev_queue *queue = dev_ingress_queue(dev);
9665 #ifdef CONFIG_NET_CLS_ACT
9668 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
9671 netdev_init_one_queue(dev, queue, NULL);
9672 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
9673 queue->qdisc_sleeping = &noop_qdisc;
9674 rcu_assign_pointer(dev->ingress_queue, queue);
9679 static const struct ethtool_ops default_ethtool_ops;
9681 void netdev_set_default_ethtool_ops(struct net_device *dev,
9682 const struct ethtool_ops *ops)
9684 if (dev->ethtool_ops == &default_ethtool_ops)
9685 dev->ethtool_ops = ops;
9687 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
9689 void netdev_freemem(struct net_device *dev)
9691 char *addr = (char *)dev - dev->padded;
9697 * alloc_netdev_mqs - allocate network device
9698 * @sizeof_priv: size of private data to allocate space for
9699 * @name: device name format string
9700 * @name_assign_type: origin of device name
9701 * @setup: callback to initialize device
9702 * @txqs: the number of TX subqueues to allocate
9703 * @rxqs: the number of RX subqueues to allocate
9705 * Allocates a struct net_device with private data area for driver use
9706 * and performs basic initialization. Also allocates subqueue structs
9707 * for each queue on the device.
9709 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
9710 unsigned char name_assign_type,
9711 void (*setup)(struct net_device *),
9712 unsigned int txqs, unsigned int rxqs)
9714 struct net_device *dev;
9715 unsigned int alloc_size;
9716 struct net_device *p;
9718 BUG_ON(strlen(name) >= sizeof(dev->name));
9721 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
9726 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
9730 alloc_size = sizeof(struct net_device);
9732 /* ensure 32-byte alignment of private area */
9733 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
9734 alloc_size += sizeof_priv;
9736 /* ensure 32-byte alignment of whole construct */
9737 alloc_size += NETDEV_ALIGN - 1;
9739 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9743 dev = PTR_ALIGN(p, NETDEV_ALIGN);
9744 dev->padded = (char *)dev - (char *)p;
9746 dev->pcpu_refcnt = alloc_percpu(int);
9747 if (!dev->pcpu_refcnt)
9750 if (dev_addr_init(dev))
9756 dev_net_set(dev, &init_net);
9758 netdev_register_lockdep_key(dev);
9760 dev->gso_max_size = GSO_MAX_SIZE;
9761 dev->gso_max_segs = GSO_MAX_SEGS;
9762 dev->upper_level = 1;
9763 dev->lower_level = 1;
9765 INIT_LIST_HEAD(&dev->napi_list);
9766 INIT_LIST_HEAD(&dev->unreg_list);
9767 INIT_LIST_HEAD(&dev->close_list);
9768 INIT_LIST_HEAD(&dev->link_watch_list);
9769 INIT_LIST_HEAD(&dev->adj_list.upper);
9770 INIT_LIST_HEAD(&dev->adj_list.lower);
9771 INIT_LIST_HEAD(&dev->ptype_all);
9772 INIT_LIST_HEAD(&dev->ptype_specific);
9773 #ifdef CONFIG_NET_SCHED
9774 hash_init(dev->qdisc_hash);
9776 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
9779 if (!dev->tx_queue_len) {
9780 dev->priv_flags |= IFF_NO_QUEUE;
9781 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
9784 dev->num_tx_queues = txqs;
9785 dev->real_num_tx_queues = txqs;
9786 if (netif_alloc_netdev_queues(dev))
9789 dev->num_rx_queues = rxqs;
9790 dev->real_num_rx_queues = rxqs;
9791 if (netif_alloc_rx_queues(dev))
9794 strcpy(dev->name, name);
9795 dev->name_assign_type = name_assign_type;
9796 dev->group = INIT_NETDEV_GROUP;
9797 if (!dev->ethtool_ops)
9798 dev->ethtool_ops = &default_ethtool_ops;
9800 nf_hook_ingress_init(dev);
9809 free_percpu(dev->pcpu_refcnt);
9811 netdev_freemem(dev);
9814 EXPORT_SYMBOL(alloc_netdev_mqs);
9817 * free_netdev - free network device
9820 * This function does the last stage of destroying an allocated device
9821 * interface. The reference to the device object is released. If this
9822 * is the last reference then it will be freed.Must be called in process
9825 void free_netdev(struct net_device *dev)
9827 struct napi_struct *p, *n;
9830 netif_free_tx_queues(dev);
9831 netif_free_rx_queues(dev);
9833 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
9835 /* Flush device addresses */
9836 dev_addr_flush(dev);
9838 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
9841 free_percpu(dev->pcpu_refcnt);
9842 dev->pcpu_refcnt = NULL;
9843 free_percpu(dev->xdp_bulkq);
9844 dev->xdp_bulkq = NULL;
9846 netdev_unregister_lockdep_key(dev);
9848 /* Compatibility with error handling in drivers */
9849 if (dev->reg_state == NETREG_UNINITIALIZED) {
9850 netdev_freemem(dev);
9854 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
9855 dev->reg_state = NETREG_RELEASED;
9857 /* will free via device release */
9858 put_device(&dev->dev);
9860 EXPORT_SYMBOL(free_netdev);
9863 * synchronize_net - Synchronize with packet receive processing
9865 * Wait for packets currently being received to be done.
9866 * Does not block later packets from starting.
9868 void synchronize_net(void)
9871 if (rtnl_is_locked())
9872 synchronize_rcu_expedited();
9876 EXPORT_SYMBOL(synchronize_net);
9879 * unregister_netdevice_queue - remove device from the kernel
9883 * This function shuts down a device interface and removes it
9884 * from the kernel tables.
9885 * If head not NULL, device is queued to be unregistered later.
9887 * Callers must hold the rtnl semaphore. You may want
9888 * unregister_netdev() instead of this.
9891 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
9896 list_move_tail(&dev->unreg_list, head);
9898 rollback_registered(dev);
9899 /* Finish processing unregister after unlock */
9903 EXPORT_SYMBOL(unregister_netdevice_queue);
9906 * unregister_netdevice_many - unregister many devices
9907 * @head: list of devices
9909 * Note: As most callers use a stack allocated list_head,
9910 * we force a list_del() to make sure stack wont be corrupted later.
9912 void unregister_netdevice_many(struct list_head *head)
9914 struct net_device *dev;
9916 if (!list_empty(head)) {
9917 rollback_registered_many(head);
9918 list_for_each_entry(dev, head, unreg_list)
9923 EXPORT_SYMBOL(unregister_netdevice_many);
9926 * unregister_netdev - remove device from the kernel
9929 * This function shuts down a device interface and removes it
9930 * from the kernel tables.
9932 * This is just a wrapper for unregister_netdevice that takes
9933 * the rtnl semaphore. In general you want to use this and not
9934 * unregister_netdevice.
9936 void unregister_netdev(struct net_device *dev)
9939 unregister_netdevice(dev);
9942 EXPORT_SYMBOL(unregister_netdev);
9945 * dev_change_net_namespace - move device to different nethost namespace
9947 * @net: network namespace
9948 * @pat: If not NULL name pattern to try if the current device name
9949 * is already taken in the destination network namespace.
9951 * This function shuts down a device interface and moves it
9952 * to a new network namespace. On success 0 is returned, on
9953 * a failure a netagive errno code is returned.
9955 * Callers must hold the rtnl semaphore.
9958 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
9960 int err, new_nsid, new_ifindex;
9964 /* Don't allow namespace local devices to be moved. */
9966 if (dev->features & NETIF_F_NETNS_LOCAL)
9969 /* Ensure the device has been registrered */
9970 if (dev->reg_state != NETREG_REGISTERED)
9973 /* Get out if there is nothing todo */
9975 if (net_eq(dev_net(dev), net))
9978 /* Pick the destination device name, and ensure
9979 * we can use it in the destination network namespace.
9982 if (__dev_get_by_name(net, dev->name)) {
9983 /* We get here if we can't use the current device name */
9986 err = dev_get_valid_name(net, dev, pat);
9992 * And now a mini version of register_netdevice unregister_netdevice.
9995 /* If device is running close it first. */
9998 /* And unlink it from device chain */
9999 unlist_netdevice(dev);
10003 /* Shutdown queueing discipline. */
10006 /* Notify protocols, that we are about to destroy
10007 * this device. They should clean all the things.
10009 * Note that dev->reg_state stays at NETREG_REGISTERED.
10010 * This is wanted because this way 8021q and macvlan know
10011 * the device is just moving and can keep their slaves up.
10013 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10016 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
10017 /* If there is an ifindex conflict assign a new one */
10018 if (__dev_get_by_index(net, dev->ifindex))
10019 new_ifindex = dev_new_index(net);
10021 new_ifindex = dev->ifindex;
10023 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
10027 * Flush the unicast and multicast chains
10032 /* Send a netdev-removed uevent to the old namespace */
10033 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
10034 netdev_adjacent_del_links(dev);
10036 /* Actually switch the network namespace */
10037 dev_net_set(dev, net);
10038 dev->ifindex = new_ifindex;
10040 /* Send a netdev-add uevent to the new namespace */
10041 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
10042 netdev_adjacent_add_links(dev);
10044 /* Fixup kobjects */
10045 err = device_rename(&dev->dev, dev->name);
10048 /* Add the device back in the hashes */
10049 list_netdevice(dev);
10051 /* Notify protocols, that a new device appeared. */
10052 call_netdevice_notifiers(NETDEV_REGISTER, dev);
10055 * Prevent userspace races by waiting until the network
10056 * device is fully setup before sending notifications.
10058 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10065 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
10067 static int dev_cpu_dead(unsigned int oldcpu)
10069 struct sk_buff **list_skb;
10070 struct sk_buff *skb;
10072 struct softnet_data *sd, *oldsd, *remsd = NULL;
10074 local_irq_disable();
10075 cpu = smp_processor_id();
10076 sd = &per_cpu(softnet_data, cpu);
10077 oldsd = &per_cpu(softnet_data, oldcpu);
10079 /* Find end of our completion_queue. */
10080 list_skb = &sd->completion_queue;
10082 list_skb = &(*list_skb)->next;
10083 /* Append completion queue from offline CPU. */
10084 *list_skb = oldsd->completion_queue;
10085 oldsd->completion_queue = NULL;
10087 /* Append output queue from offline CPU. */
10088 if (oldsd->output_queue) {
10089 *sd->output_queue_tailp = oldsd->output_queue;
10090 sd->output_queue_tailp = oldsd->output_queue_tailp;
10091 oldsd->output_queue = NULL;
10092 oldsd->output_queue_tailp = &oldsd->output_queue;
10094 /* Append NAPI poll list from offline CPU, with one exception :
10095 * process_backlog() must be called by cpu owning percpu backlog.
10096 * We properly handle process_queue & input_pkt_queue later.
10098 while (!list_empty(&oldsd->poll_list)) {
10099 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
10100 struct napi_struct,
10103 list_del_init(&napi->poll_list);
10104 if (napi->poll == process_backlog)
10107 ____napi_schedule(sd, napi);
10110 raise_softirq_irqoff(NET_TX_SOFTIRQ);
10111 local_irq_enable();
10114 remsd = oldsd->rps_ipi_list;
10115 oldsd->rps_ipi_list = NULL;
10117 /* send out pending IPI's on offline CPU */
10118 net_rps_send_ipi(remsd);
10120 /* Process offline CPU's input_pkt_queue */
10121 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
10123 input_queue_head_incr(oldsd);
10125 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
10127 input_queue_head_incr(oldsd);
10134 * netdev_increment_features - increment feature set by one
10135 * @all: current feature set
10136 * @one: new feature set
10137 * @mask: mask feature set
10139 * Computes a new feature set after adding a device with feature set
10140 * @one to the master device with current feature set @all. Will not
10141 * enable anything that is off in @mask. Returns the new feature set.
10143 netdev_features_t netdev_increment_features(netdev_features_t all,
10144 netdev_features_t one, netdev_features_t mask)
10146 if (mask & NETIF_F_HW_CSUM)
10147 mask |= NETIF_F_CSUM_MASK;
10148 mask |= NETIF_F_VLAN_CHALLENGED;
10150 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
10151 all &= one | ~NETIF_F_ALL_FOR_ALL;
10153 /* If one device supports hw checksumming, set for all. */
10154 if (all & NETIF_F_HW_CSUM)
10155 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
10159 EXPORT_SYMBOL(netdev_increment_features);
10161 static struct hlist_head * __net_init netdev_create_hash(void)
10164 struct hlist_head *hash;
10166 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
10168 for (i = 0; i < NETDEV_HASHENTRIES; i++)
10169 INIT_HLIST_HEAD(&hash[i]);
10174 /* Initialize per network namespace state */
10175 static int __net_init netdev_init(struct net *net)
10177 BUILD_BUG_ON(GRO_HASH_BUCKETS >
10178 8 * sizeof_field(struct napi_struct, gro_bitmask));
10180 if (net != &init_net)
10181 INIT_LIST_HEAD(&net->dev_base_head);
10183 net->dev_name_head = netdev_create_hash();
10184 if (net->dev_name_head == NULL)
10187 net->dev_index_head = netdev_create_hash();
10188 if (net->dev_index_head == NULL)
10191 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
10196 kfree(net->dev_name_head);
10202 * netdev_drivername - network driver for the device
10203 * @dev: network device
10205 * Determine network driver for device.
10207 const char *netdev_drivername(const struct net_device *dev)
10209 const struct device_driver *driver;
10210 const struct device *parent;
10211 const char *empty = "";
10213 parent = dev->dev.parent;
10217 driver = parent->driver;
10218 if (driver && driver->name)
10219 return driver->name;
10223 static void __netdev_printk(const char *level, const struct net_device *dev,
10224 struct va_format *vaf)
10226 if (dev && dev->dev.parent) {
10227 dev_printk_emit(level[1] - '0',
10230 dev_driver_string(dev->dev.parent),
10231 dev_name(dev->dev.parent),
10232 netdev_name(dev), netdev_reg_state(dev),
10235 printk("%s%s%s: %pV",
10236 level, netdev_name(dev), netdev_reg_state(dev), vaf);
10238 printk("%s(NULL net_device): %pV", level, vaf);
10242 void netdev_printk(const char *level, const struct net_device *dev,
10243 const char *format, ...)
10245 struct va_format vaf;
10248 va_start(args, format);
10253 __netdev_printk(level, dev, &vaf);
10257 EXPORT_SYMBOL(netdev_printk);
10259 #define define_netdev_printk_level(func, level) \
10260 void func(const struct net_device *dev, const char *fmt, ...) \
10262 struct va_format vaf; \
10265 va_start(args, fmt); \
10270 __netdev_printk(level, dev, &vaf); \
10274 EXPORT_SYMBOL(func);
10276 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
10277 define_netdev_printk_level(netdev_alert, KERN_ALERT);
10278 define_netdev_printk_level(netdev_crit, KERN_CRIT);
10279 define_netdev_printk_level(netdev_err, KERN_ERR);
10280 define_netdev_printk_level(netdev_warn, KERN_WARNING);
10281 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
10282 define_netdev_printk_level(netdev_info, KERN_INFO);
10284 static void __net_exit netdev_exit(struct net *net)
10286 kfree(net->dev_name_head);
10287 kfree(net->dev_index_head);
10288 if (net != &init_net)
10289 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
10292 static struct pernet_operations __net_initdata netdev_net_ops = {
10293 .init = netdev_init,
10294 .exit = netdev_exit,
10297 static void __net_exit default_device_exit(struct net *net)
10299 struct net_device *dev, *aux;
10301 * Push all migratable network devices back to the
10302 * initial network namespace
10305 for_each_netdev_safe(net, dev, aux) {
10307 char fb_name[IFNAMSIZ];
10309 /* Ignore unmoveable devices (i.e. loopback) */
10310 if (dev->features & NETIF_F_NETNS_LOCAL)
10313 /* Leave virtual devices for the generic cleanup */
10314 if (dev->rtnl_link_ops)
10317 /* Push remaining network devices to init_net */
10318 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
10319 if (__dev_get_by_name(&init_net, fb_name))
10320 snprintf(fb_name, IFNAMSIZ, "dev%%d");
10321 err = dev_change_net_namespace(dev, &init_net, fb_name);
10323 pr_emerg("%s: failed to move %s to init_net: %d\n",
10324 __func__, dev->name, err);
10331 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
10333 /* Return with the rtnl_lock held when there are no network
10334 * devices unregistering in any network namespace in net_list.
10337 bool unregistering;
10338 DEFINE_WAIT_FUNC(wait, woken_wake_function);
10340 add_wait_queue(&netdev_unregistering_wq, &wait);
10342 unregistering = false;
10344 list_for_each_entry(net, net_list, exit_list) {
10345 if (net->dev_unreg_count > 0) {
10346 unregistering = true;
10350 if (!unregistering)
10354 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
10356 remove_wait_queue(&netdev_unregistering_wq, &wait);
10359 static void __net_exit default_device_exit_batch(struct list_head *net_list)
10361 /* At exit all network devices most be removed from a network
10362 * namespace. Do this in the reverse order of registration.
10363 * Do this across as many network namespaces as possible to
10364 * improve batching efficiency.
10366 struct net_device *dev;
10368 LIST_HEAD(dev_kill_list);
10370 /* To prevent network device cleanup code from dereferencing
10371 * loopback devices or network devices that have been freed
10372 * wait here for all pending unregistrations to complete,
10373 * before unregistring the loopback device and allowing the
10374 * network namespace be freed.
10376 * The netdev todo list containing all network devices
10377 * unregistrations that happen in default_device_exit_batch
10378 * will run in the rtnl_unlock() at the end of
10379 * default_device_exit_batch.
10381 rtnl_lock_unregistering(net_list);
10382 list_for_each_entry(net, net_list, exit_list) {
10383 for_each_netdev_reverse(net, dev) {
10384 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
10385 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
10387 unregister_netdevice_queue(dev, &dev_kill_list);
10390 unregister_netdevice_many(&dev_kill_list);
10394 static struct pernet_operations __net_initdata default_device_ops = {
10395 .exit = default_device_exit,
10396 .exit_batch = default_device_exit_batch,
10400 * Initialize the DEV module. At boot time this walks the device list and
10401 * unhooks any devices that fail to initialise (normally hardware not
10402 * present) and leaves us with a valid list of present and active devices.
10407 * This is called single threaded during boot, so no need
10408 * to take the rtnl semaphore.
10410 static int __init net_dev_init(void)
10412 int i, rc = -ENOMEM;
10414 BUG_ON(!dev_boot_phase);
10416 if (dev_proc_init())
10419 if (netdev_kobject_init())
10422 INIT_LIST_HEAD(&ptype_all);
10423 for (i = 0; i < PTYPE_HASH_SIZE; i++)
10424 INIT_LIST_HEAD(&ptype_base[i]);
10426 INIT_LIST_HEAD(&offload_base);
10428 if (register_pernet_subsys(&netdev_net_ops))
10432 * Initialise the packet receive queues.
10435 for_each_possible_cpu(i) {
10436 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
10437 struct softnet_data *sd = &per_cpu(softnet_data, i);
10439 INIT_WORK(flush, flush_backlog);
10441 skb_queue_head_init(&sd->input_pkt_queue);
10442 skb_queue_head_init(&sd->process_queue);
10443 #ifdef CONFIG_XFRM_OFFLOAD
10444 skb_queue_head_init(&sd->xfrm_backlog);
10446 INIT_LIST_HEAD(&sd->poll_list);
10447 sd->output_queue_tailp = &sd->output_queue;
10449 sd->csd.func = rps_trigger_softirq;
10454 init_gro_hash(&sd->backlog);
10455 sd->backlog.poll = process_backlog;
10456 sd->backlog.weight = weight_p;
10459 dev_boot_phase = 0;
10461 /* The loopback device is special if any other network devices
10462 * is present in a network namespace the loopback device must
10463 * be present. Since we now dynamically allocate and free the
10464 * loopback device ensure this invariant is maintained by
10465 * keeping the loopback device as the first device on the
10466 * list of network devices. Ensuring the loopback devices
10467 * is the first device that appears and the last network device
10470 if (register_pernet_device(&loopback_net_ops))
10473 if (register_pernet_device(&default_device_ops))
10476 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
10477 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
10479 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
10480 NULL, dev_cpu_dead);
10487 subsys_initcall(net_dev_init);