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)
1767 struct net_device *dev;
1771 /* Close race with setup_net() and cleanup_net() */
1772 down_write(&pernet_ops_rwsem);
1774 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1779 for_each_netdev(net, dev) {
1780 if (dev->flags & IFF_UP) {
1781 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1783 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1785 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1790 up_write(&pernet_ops_rwsem);
1793 EXPORT_SYMBOL(unregister_netdevice_notifier);
1796 * register_netdevice_notifier_net - register a per-netns network notifier block
1797 * @net: network namespace
1800 * Register a notifier to be called when network device events occur.
1801 * The notifier passed is linked into the kernel structures and must
1802 * not be reused until it has been unregistered. A negative errno code
1803 * is returned on a failure.
1805 * When registered all registration and up events are replayed
1806 * to the new notifier to allow device to have a race free
1807 * view of the network device list.
1810 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1815 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1821 err = call_netdevice_register_net_notifiers(nb, net);
1823 goto chain_unregister;
1830 raw_notifier_chain_unregister(&netdev_chain, nb);
1833 EXPORT_SYMBOL(register_netdevice_notifier_net);
1836 * unregister_netdevice_notifier_net - unregister a per-netns
1837 * network notifier block
1838 * @net: network namespace
1841 * Unregister a notifier previously registered by
1842 * register_netdevice_notifier(). The notifier is unlinked into the
1843 * kernel structures and may then be reused. A negative errno code
1844 * is returned on a failure.
1846 * After unregistering unregister and down device events are synthesized
1847 * for all devices on the device list to the removed notifier to remove
1848 * the need for special case cleanup code.
1851 int unregister_netdevice_notifier_net(struct net *net,
1852 struct notifier_block *nb)
1857 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1861 call_netdevice_unregister_net_notifiers(nb, net);
1867 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1870 * call_netdevice_notifiers_info - call all network notifier blocks
1871 * @val: value passed unmodified to notifier function
1872 * @info: notifier information data
1874 * Call all network notifier blocks. Parameters and return value
1875 * are as for raw_notifier_call_chain().
1878 static int call_netdevice_notifiers_info(unsigned long val,
1879 struct netdev_notifier_info *info)
1881 struct net *net = dev_net(info->dev);
1886 /* Run per-netns notifier block chain first, then run the global one.
1887 * Hopefully, one day, the global one is going to be removed after
1888 * all notifier block registrators get converted to be per-netns.
1890 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
1891 if (ret & NOTIFY_STOP_MASK)
1893 return raw_notifier_call_chain(&netdev_chain, val, info);
1896 static int call_netdevice_notifiers_extack(unsigned long val,
1897 struct net_device *dev,
1898 struct netlink_ext_ack *extack)
1900 struct netdev_notifier_info info = {
1905 return call_netdevice_notifiers_info(val, &info);
1909 * call_netdevice_notifiers - call all network notifier blocks
1910 * @val: value passed unmodified to notifier function
1911 * @dev: net_device pointer passed unmodified to notifier function
1913 * Call all network notifier blocks. Parameters and return value
1914 * are as for raw_notifier_call_chain().
1917 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1919 return call_netdevice_notifiers_extack(val, dev, NULL);
1921 EXPORT_SYMBOL(call_netdevice_notifiers);
1924 * call_netdevice_notifiers_mtu - call all network notifier blocks
1925 * @val: value passed unmodified to notifier function
1926 * @dev: net_device pointer passed unmodified to notifier function
1927 * @arg: additional u32 argument passed to the notifier function
1929 * Call all network notifier blocks. Parameters and return value
1930 * are as for raw_notifier_call_chain().
1932 static int call_netdevice_notifiers_mtu(unsigned long val,
1933 struct net_device *dev, u32 arg)
1935 struct netdev_notifier_info_ext info = {
1940 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
1942 return call_netdevice_notifiers_info(val, &info.info);
1945 #ifdef CONFIG_NET_INGRESS
1946 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
1948 void net_inc_ingress_queue(void)
1950 static_branch_inc(&ingress_needed_key);
1952 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1954 void net_dec_ingress_queue(void)
1956 static_branch_dec(&ingress_needed_key);
1958 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1961 #ifdef CONFIG_NET_EGRESS
1962 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
1964 void net_inc_egress_queue(void)
1966 static_branch_inc(&egress_needed_key);
1968 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1970 void net_dec_egress_queue(void)
1972 static_branch_dec(&egress_needed_key);
1974 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1977 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
1978 #ifdef CONFIG_JUMP_LABEL
1979 static atomic_t netstamp_needed_deferred;
1980 static atomic_t netstamp_wanted;
1981 static void netstamp_clear(struct work_struct *work)
1983 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1986 wanted = atomic_add_return(deferred, &netstamp_wanted);
1988 static_branch_enable(&netstamp_needed_key);
1990 static_branch_disable(&netstamp_needed_key);
1992 static DECLARE_WORK(netstamp_work, netstamp_clear);
1995 void net_enable_timestamp(void)
1997 #ifdef CONFIG_JUMP_LABEL
2001 wanted = atomic_read(&netstamp_wanted);
2004 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
2007 atomic_inc(&netstamp_needed_deferred);
2008 schedule_work(&netstamp_work);
2010 static_branch_inc(&netstamp_needed_key);
2013 EXPORT_SYMBOL(net_enable_timestamp);
2015 void net_disable_timestamp(void)
2017 #ifdef CONFIG_JUMP_LABEL
2021 wanted = atomic_read(&netstamp_wanted);
2024 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2027 atomic_dec(&netstamp_needed_deferred);
2028 schedule_work(&netstamp_work);
2030 static_branch_dec(&netstamp_needed_key);
2033 EXPORT_SYMBOL(net_disable_timestamp);
2035 static inline void net_timestamp_set(struct sk_buff *skb)
2038 if (static_branch_unlikely(&netstamp_needed_key))
2039 __net_timestamp(skb);
2042 #define net_timestamp_check(COND, SKB) \
2043 if (static_branch_unlikely(&netstamp_needed_key)) { \
2044 if ((COND) && !(SKB)->tstamp) \
2045 __net_timestamp(SKB); \
2048 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2052 if (!(dev->flags & IFF_UP))
2055 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
2056 if (skb->len <= len)
2059 /* if TSO is enabled, we don't care about the length as the packet
2060 * could be forwarded without being segmented before
2062 if (skb_is_gso(skb))
2067 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2069 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2071 int ret = ____dev_forward_skb(dev, skb);
2074 skb->protocol = eth_type_trans(skb, dev);
2075 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2080 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2083 * dev_forward_skb - loopback an skb to another netif
2085 * @dev: destination network device
2086 * @skb: buffer to forward
2089 * NET_RX_SUCCESS (no congestion)
2090 * NET_RX_DROP (packet was dropped, but freed)
2092 * dev_forward_skb can be used for injecting an skb from the
2093 * start_xmit function of one device into the receive queue
2094 * of another device.
2096 * The receiving device may be in another namespace, so
2097 * we have to clear all information in the skb that could
2098 * impact namespace isolation.
2100 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2102 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2104 EXPORT_SYMBOL_GPL(dev_forward_skb);
2106 static inline int deliver_skb(struct sk_buff *skb,
2107 struct packet_type *pt_prev,
2108 struct net_device *orig_dev)
2110 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2112 refcount_inc(&skb->users);
2113 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2116 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2117 struct packet_type **pt,
2118 struct net_device *orig_dev,
2120 struct list_head *ptype_list)
2122 struct packet_type *ptype, *pt_prev = *pt;
2124 list_for_each_entry_rcu(ptype, ptype_list, list) {
2125 if (ptype->type != type)
2128 deliver_skb(skb, pt_prev, orig_dev);
2134 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2136 if (!ptype->af_packet_priv || !skb->sk)
2139 if (ptype->id_match)
2140 return ptype->id_match(ptype, skb->sk);
2141 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2148 * dev_nit_active - return true if any network interface taps are in use
2150 * @dev: network device to check for the presence of taps
2152 bool dev_nit_active(struct net_device *dev)
2154 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2156 EXPORT_SYMBOL_GPL(dev_nit_active);
2159 * Support routine. Sends outgoing frames to any network
2160 * taps currently in use.
2163 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2165 struct packet_type *ptype;
2166 struct sk_buff *skb2 = NULL;
2167 struct packet_type *pt_prev = NULL;
2168 struct list_head *ptype_list = &ptype_all;
2172 list_for_each_entry_rcu(ptype, ptype_list, list) {
2173 if (ptype->ignore_outgoing)
2176 /* Never send packets back to the socket
2177 * they originated from - MvS (miquels@drinkel.ow.org)
2179 if (skb_loop_sk(ptype, skb))
2183 deliver_skb(skb2, pt_prev, skb->dev);
2188 /* need to clone skb, done only once */
2189 skb2 = skb_clone(skb, GFP_ATOMIC);
2193 net_timestamp_set(skb2);
2195 /* skb->nh should be correctly
2196 * set by sender, so that the second statement is
2197 * just protection against buggy protocols.
2199 skb_reset_mac_header(skb2);
2201 if (skb_network_header(skb2) < skb2->data ||
2202 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2203 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2204 ntohs(skb2->protocol),
2206 skb_reset_network_header(skb2);
2209 skb2->transport_header = skb2->network_header;
2210 skb2->pkt_type = PACKET_OUTGOING;
2214 if (ptype_list == &ptype_all) {
2215 ptype_list = &dev->ptype_all;
2220 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2221 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2227 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2230 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2231 * @dev: Network device
2232 * @txq: number of queues available
2234 * If real_num_tx_queues is changed the tc mappings may no longer be
2235 * valid. To resolve this verify the tc mapping remains valid and if
2236 * not NULL the mapping. With no priorities mapping to this
2237 * offset/count pair it will no longer be used. In the worst case TC0
2238 * is invalid nothing can be done so disable priority mappings. If is
2239 * expected that drivers will fix this mapping if they can before
2240 * calling netif_set_real_num_tx_queues.
2242 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2245 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2247 /* If TC0 is invalidated disable TC mapping */
2248 if (tc->offset + tc->count > txq) {
2249 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2254 /* Invalidated prio to tc mappings set to TC0 */
2255 for (i = 1; i < TC_BITMASK + 1; i++) {
2256 int q = netdev_get_prio_tc_map(dev, i);
2258 tc = &dev->tc_to_txq[q];
2259 if (tc->offset + tc->count > txq) {
2260 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2262 netdev_set_prio_tc_map(dev, i, 0);
2267 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2270 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2273 /* walk through the TCs and see if it falls into any of them */
2274 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2275 if ((txq - tc->offset) < tc->count)
2279 /* didn't find it, just return -1 to indicate no match */
2285 EXPORT_SYMBOL(netdev_txq_to_tc);
2288 struct static_key xps_needed __read_mostly;
2289 EXPORT_SYMBOL(xps_needed);
2290 struct static_key xps_rxqs_needed __read_mostly;
2291 EXPORT_SYMBOL(xps_rxqs_needed);
2292 static DEFINE_MUTEX(xps_map_mutex);
2293 #define xmap_dereference(P) \
2294 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2296 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2299 struct xps_map *map = NULL;
2303 map = xmap_dereference(dev_maps->attr_map[tci]);
2307 for (pos = map->len; pos--;) {
2308 if (map->queues[pos] != index)
2312 map->queues[pos] = map->queues[--map->len];
2316 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2317 kfree_rcu(map, rcu);
2324 static bool remove_xps_queue_cpu(struct net_device *dev,
2325 struct xps_dev_maps *dev_maps,
2326 int cpu, u16 offset, u16 count)
2328 int num_tc = dev->num_tc ? : 1;
2329 bool active = false;
2332 for (tci = cpu * num_tc; num_tc--; tci++) {
2335 for (i = count, j = offset; i--; j++) {
2336 if (!remove_xps_queue(dev_maps, tci, j))
2346 static void reset_xps_maps(struct net_device *dev,
2347 struct xps_dev_maps *dev_maps,
2351 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2352 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2354 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2356 static_key_slow_dec_cpuslocked(&xps_needed);
2357 kfree_rcu(dev_maps, rcu);
2360 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask,
2361 struct xps_dev_maps *dev_maps, unsigned int nr_ids,
2362 u16 offset, u16 count, bool is_rxqs_map)
2364 bool active = false;
2367 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids),
2369 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset,
2372 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2375 for (i = offset + (count - 1); count--; i--) {
2376 netdev_queue_numa_node_write(
2377 netdev_get_tx_queue(dev, i),
2383 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2386 const unsigned long *possible_mask = NULL;
2387 struct xps_dev_maps *dev_maps;
2388 unsigned int nr_ids;
2390 if (!static_key_false(&xps_needed))
2394 mutex_lock(&xps_map_mutex);
2396 if (static_key_false(&xps_rxqs_needed)) {
2397 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2399 nr_ids = dev->num_rx_queues;
2400 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids,
2401 offset, count, true);
2405 dev_maps = xmap_dereference(dev->xps_cpus_map);
2409 if (num_possible_cpus() > 1)
2410 possible_mask = cpumask_bits(cpu_possible_mask);
2411 nr_ids = nr_cpu_ids;
2412 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count,
2416 mutex_unlock(&xps_map_mutex);
2420 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2422 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2425 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2426 u16 index, bool is_rxqs_map)
2428 struct xps_map *new_map;
2429 int alloc_len = XPS_MIN_MAP_ALLOC;
2432 for (pos = 0; map && pos < map->len; pos++) {
2433 if (map->queues[pos] != index)
2438 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2440 if (pos < map->alloc_len)
2443 alloc_len = map->alloc_len * 2;
2446 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2450 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2452 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2453 cpu_to_node(attr_index));
2457 for (i = 0; i < pos; i++)
2458 new_map->queues[i] = map->queues[i];
2459 new_map->alloc_len = alloc_len;
2465 /* Must be called under cpus_read_lock */
2466 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2467 u16 index, bool is_rxqs_map)
2469 const unsigned long *online_mask = NULL, *possible_mask = NULL;
2470 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2471 int i, j, tci, numa_node_id = -2;
2472 int maps_sz, num_tc = 1, tc = 0;
2473 struct xps_map *map, *new_map;
2474 bool active = false;
2475 unsigned int nr_ids;
2478 /* Do not allow XPS on subordinate device directly */
2479 num_tc = dev->num_tc;
2483 /* If queue belongs to subordinate dev use its map */
2484 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2486 tc = netdev_txq_to_tc(dev, index);
2491 mutex_lock(&xps_map_mutex);
2493 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2494 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2495 nr_ids = dev->num_rx_queues;
2497 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2498 if (num_possible_cpus() > 1) {
2499 online_mask = cpumask_bits(cpu_online_mask);
2500 possible_mask = cpumask_bits(cpu_possible_mask);
2502 dev_maps = xmap_dereference(dev->xps_cpus_map);
2503 nr_ids = nr_cpu_ids;
2506 if (maps_sz < L1_CACHE_BYTES)
2507 maps_sz = L1_CACHE_BYTES;
2509 /* allocate memory for queue storage */
2510 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2513 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2514 if (!new_dev_maps) {
2515 mutex_unlock(&xps_map_mutex);
2519 tci = j * num_tc + tc;
2520 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) :
2523 map = expand_xps_map(map, j, index, is_rxqs_map);
2527 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2531 goto out_no_new_maps;
2534 /* Increment static keys at most once per type */
2535 static_key_slow_inc_cpuslocked(&xps_needed);
2537 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2540 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2542 /* copy maps belonging to foreign traffic classes */
2543 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) {
2544 /* fill in the new device map from the old device map */
2545 map = xmap_dereference(dev_maps->attr_map[tci]);
2546 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2549 /* We need to explicitly update tci as prevous loop
2550 * could break out early if dev_maps is NULL.
2552 tci = j * num_tc + tc;
2554 if (netif_attr_test_mask(j, mask, nr_ids) &&
2555 netif_attr_test_online(j, online_mask, nr_ids)) {
2556 /* add tx-queue to CPU/rx-queue maps */
2559 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2560 while ((pos < map->len) && (map->queues[pos] != index))
2563 if (pos == map->len)
2564 map->queues[map->len++] = index;
2567 if (numa_node_id == -2)
2568 numa_node_id = cpu_to_node(j);
2569 else if (numa_node_id != cpu_to_node(j))
2573 } else if (dev_maps) {
2574 /* fill in the new device map from the old device map */
2575 map = xmap_dereference(dev_maps->attr_map[tci]);
2576 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2579 /* copy maps belonging to foreign traffic classes */
2580 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2581 /* fill in the new device map from the old device map */
2582 map = xmap_dereference(dev_maps->attr_map[tci]);
2583 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2588 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps);
2590 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps);
2592 /* Cleanup old maps */
2594 goto out_no_old_maps;
2596 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2598 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2599 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2600 map = xmap_dereference(dev_maps->attr_map[tci]);
2601 if (map && map != new_map)
2602 kfree_rcu(map, rcu);
2606 kfree_rcu(dev_maps, rcu);
2609 dev_maps = new_dev_maps;
2614 /* update Tx queue numa node */
2615 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2616 (numa_node_id >= 0) ?
2617 numa_node_id : NUMA_NO_NODE);
2623 /* removes tx-queue from unused CPUs/rx-queues */
2624 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2626 for (i = tc, tci = j * num_tc; i--; tci++)
2627 active |= remove_xps_queue(dev_maps, tci, index);
2628 if (!netif_attr_test_mask(j, mask, nr_ids) ||
2629 !netif_attr_test_online(j, online_mask, nr_ids))
2630 active |= remove_xps_queue(dev_maps, tci, index);
2631 for (i = num_tc - tc, tci++; --i; tci++)
2632 active |= remove_xps_queue(dev_maps, tci, index);
2635 /* free map if not active */
2637 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2640 mutex_unlock(&xps_map_mutex);
2644 /* remove any maps that we added */
2645 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2647 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2648 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2650 xmap_dereference(dev_maps->attr_map[tci]) :
2652 if (new_map && new_map != map)
2657 mutex_unlock(&xps_map_mutex);
2659 kfree(new_dev_maps);
2662 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2664 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2670 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
2675 EXPORT_SYMBOL(netif_set_xps_queue);
2678 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2680 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2682 /* Unbind any subordinate channels */
2683 while (txq-- != &dev->_tx[0]) {
2685 netdev_unbind_sb_channel(dev, txq->sb_dev);
2689 void netdev_reset_tc(struct net_device *dev)
2692 netif_reset_xps_queues_gt(dev, 0);
2694 netdev_unbind_all_sb_channels(dev);
2696 /* Reset TC configuration of device */
2698 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2699 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2701 EXPORT_SYMBOL(netdev_reset_tc);
2703 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2705 if (tc >= dev->num_tc)
2709 netif_reset_xps_queues(dev, offset, count);
2711 dev->tc_to_txq[tc].count = count;
2712 dev->tc_to_txq[tc].offset = offset;
2715 EXPORT_SYMBOL(netdev_set_tc_queue);
2717 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2719 if (num_tc > TC_MAX_QUEUE)
2723 netif_reset_xps_queues_gt(dev, 0);
2725 netdev_unbind_all_sb_channels(dev);
2727 dev->num_tc = num_tc;
2730 EXPORT_SYMBOL(netdev_set_num_tc);
2732 void netdev_unbind_sb_channel(struct net_device *dev,
2733 struct net_device *sb_dev)
2735 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2738 netif_reset_xps_queues_gt(sb_dev, 0);
2740 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2741 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2743 while (txq-- != &dev->_tx[0]) {
2744 if (txq->sb_dev == sb_dev)
2748 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2750 int netdev_bind_sb_channel_queue(struct net_device *dev,
2751 struct net_device *sb_dev,
2752 u8 tc, u16 count, u16 offset)
2754 /* Make certain the sb_dev and dev are already configured */
2755 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2758 /* We cannot hand out queues we don't have */
2759 if ((offset + count) > dev->real_num_tx_queues)
2762 /* Record the mapping */
2763 sb_dev->tc_to_txq[tc].count = count;
2764 sb_dev->tc_to_txq[tc].offset = offset;
2766 /* Provide a way for Tx queue to find the tc_to_txq map or
2767 * XPS map for itself.
2770 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2774 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2776 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2778 /* Do not use a multiqueue device to represent a subordinate channel */
2779 if (netif_is_multiqueue(dev))
2782 /* We allow channels 1 - 32767 to be used for subordinate channels.
2783 * Channel 0 is meant to be "native" mode and used only to represent
2784 * the main root device. We allow writing 0 to reset the device back
2785 * to normal mode after being used as a subordinate channel.
2787 if (channel > S16_MAX)
2790 dev->num_tc = -channel;
2794 EXPORT_SYMBOL(netdev_set_sb_channel);
2797 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2798 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2800 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2805 disabling = txq < dev->real_num_tx_queues;
2807 if (txq < 1 || txq > dev->num_tx_queues)
2810 if (dev->reg_state == NETREG_REGISTERED ||
2811 dev->reg_state == NETREG_UNREGISTERING) {
2814 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2820 netif_setup_tc(dev, txq);
2822 dev->real_num_tx_queues = txq;
2826 qdisc_reset_all_tx_gt(dev, txq);
2828 netif_reset_xps_queues_gt(dev, txq);
2832 dev->real_num_tx_queues = txq;
2837 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2841 * netif_set_real_num_rx_queues - set actual number of RX queues used
2842 * @dev: Network device
2843 * @rxq: Actual number of RX queues
2845 * This must be called either with the rtnl_lock held or before
2846 * registration of the net device. Returns 0 on success, or a
2847 * negative error code. If called before registration, it always
2850 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2854 if (rxq < 1 || rxq > dev->num_rx_queues)
2857 if (dev->reg_state == NETREG_REGISTERED) {
2860 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2866 dev->real_num_rx_queues = rxq;
2869 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2873 * netif_get_num_default_rss_queues - default number of RSS queues
2875 * This routine should set an upper limit on the number of RSS queues
2876 * used by default by multiqueue devices.
2878 int netif_get_num_default_rss_queues(void)
2880 return is_kdump_kernel() ?
2881 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2883 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2885 static void __netif_reschedule(struct Qdisc *q)
2887 struct softnet_data *sd;
2888 unsigned long flags;
2890 local_irq_save(flags);
2891 sd = this_cpu_ptr(&softnet_data);
2892 q->next_sched = NULL;
2893 *sd->output_queue_tailp = q;
2894 sd->output_queue_tailp = &q->next_sched;
2895 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2896 local_irq_restore(flags);
2899 void __netif_schedule(struct Qdisc *q)
2901 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2902 __netif_reschedule(q);
2904 EXPORT_SYMBOL(__netif_schedule);
2906 struct dev_kfree_skb_cb {
2907 enum skb_free_reason reason;
2910 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2912 return (struct dev_kfree_skb_cb *)skb->cb;
2915 void netif_schedule_queue(struct netdev_queue *txq)
2918 if (!netif_xmit_stopped(txq)) {
2919 struct Qdisc *q = rcu_dereference(txq->qdisc);
2921 __netif_schedule(q);
2925 EXPORT_SYMBOL(netif_schedule_queue);
2927 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2929 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2933 q = rcu_dereference(dev_queue->qdisc);
2934 __netif_schedule(q);
2938 EXPORT_SYMBOL(netif_tx_wake_queue);
2940 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2942 unsigned long flags;
2947 if (likely(refcount_read(&skb->users) == 1)) {
2949 refcount_set(&skb->users, 0);
2950 } else if (likely(!refcount_dec_and_test(&skb->users))) {
2953 get_kfree_skb_cb(skb)->reason = reason;
2954 local_irq_save(flags);
2955 skb->next = __this_cpu_read(softnet_data.completion_queue);
2956 __this_cpu_write(softnet_data.completion_queue, skb);
2957 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2958 local_irq_restore(flags);
2960 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2962 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2964 if (in_irq() || irqs_disabled())
2965 __dev_kfree_skb_irq(skb, reason);
2969 EXPORT_SYMBOL(__dev_kfree_skb_any);
2973 * netif_device_detach - mark device as removed
2974 * @dev: network device
2976 * Mark device as removed from system and therefore no longer available.
2978 void netif_device_detach(struct net_device *dev)
2980 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2981 netif_running(dev)) {
2982 netif_tx_stop_all_queues(dev);
2985 EXPORT_SYMBOL(netif_device_detach);
2988 * netif_device_attach - mark device as attached
2989 * @dev: network device
2991 * Mark device as attached from system and restart if needed.
2993 void netif_device_attach(struct net_device *dev)
2995 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2996 netif_running(dev)) {
2997 netif_tx_wake_all_queues(dev);
2998 __netdev_watchdog_up(dev);
3001 EXPORT_SYMBOL(netif_device_attach);
3004 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3005 * to be used as a distribution range.
3007 static u16 skb_tx_hash(const struct net_device *dev,
3008 const struct net_device *sb_dev,
3009 struct sk_buff *skb)
3013 u16 qcount = dev->real_num_tx_queues;
3016 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3018 qoffset = sb_dev->tc_to_txq[tc].offset;
3019 qcount = sb_dev->tc_to_txq[tc].count;
3022 if (skb_rx_queue_recorded(skb)) {
3023 hash = skb_get_rx_queue(skb);
3024 while (unlikely(hash >= qcount))
3026 return hash + qoffset;
3029 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3032 static void skb_warn_bad_offload(const struct sk_buff *skb)
3034 static const netdev_features_t null_features;
3035 struct net_device *dev = skb->dev;
3036 const char *name = "";
3038 if (!net_ratelimit())
3042 if (dev->dev.parent)
3043 name = dev_driver_string(dev->dev.parent);
3045 name = netdev_name(dev);
3047 skb_dump(KERN_WARNING, skb, false);
3048 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3049 name, dev ? &dev->features : &null_features,
3050 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3054 * Invalidate hardware checksum when packet is to be mangled, and
3055 * complete checksum manually on outgoing path.
3057 int skb_checksum_help(struct sk_buff *skb)
3060 int ret = 0, offset;
3062 if (skb->ip_summed == CHECKSUM_COMPLETE)
3063 goto out_set_summed;
3065 if (unlikely(skb_shinfo(skb)->gso_size)) {
3066 skb_warn_bad_offload(skb);
3070 /* Before computing a checksum, we should make sure no frag could
3071 * be modified by an external entity : checksum could be wrong.
3073 if (skb_has_shared_frag(skb)) {
3074 ret = __skb_linearize(skb);
3079 offset = skb_checksum_start_offset(skb);
3080 BUG_ON(offset >= skb_headlen(skb));
3081 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3083 offset += skb->csum_offset;
3084 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
3086 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3090 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3092 skb->ip_summed = CHECKSUM_NONE;
3096 EXPORT_SYMBOL(skb_checksum_help);
3098 int skb_crc32c_csum_help(struct sk_buff *skb)
3101 int ret = 0, offset, start;
3103 if (skb->ip_summed != CHECKSUM_PARTIAL)
3106 if (unlikely(skb_is_gso(skb)))
3109 /* Before computing a checksum, we should make sure no frag could
3110 * be modified by an external entity : checksum could be wrong.
3112 if (unlikely(skb_has_shared_frag(skb))) {
3113 ret = __skb_linearize(skb);
3117 start = skb_checksum_start_offset(skb);
3118 offset = start + offsetof(struct sctphdr, checksum);
3119 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3124 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3128 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3129 skb->len - start, ~(__u32)0,
3131 *(__le32 *)(skb->data + offset) = crc32c_csum;
3132 skb->ip_summed = CHECKSUM_NONE;
3133 skb->csum_not_inet = 0;
3138 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3140 __be16 type = skb->protocol;
3142 /* Tunnel gso handlers can set protocol to ethernet. */
3143 if (type == htons(ETH_P_TEB)) {
3146 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3149 eth = (struct ethhdr *)skb->data;
3150 type = eth->h_proto;
3153 return __vlan_get_protocol(skb, type, depth);
3157 * skb_mac_gso_segment - mac layer segmentation handler.
3158 * @skb: buffer to segment
3159 * @features: features for the output path (see dev->features)
3161 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3162 netdev_features_t features)
3164 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3165 struct packet_offload *ptype;
3166 int vlan_depth = skb->mac_len;
3167 __be16 type = skb_network_protocol(skb, &vlan_depth);
3169 if (unlikely(!type))
3170 return ERR_PTR(-EINVAL);
3172 __skb_pull(skb, vlan_depth);
3175 list_for_each_entry_rcu(ptype, &offload_base, list) {
3176 if (ptype->type == type && ptype->callbacks.gso_segment) {
3177 segs = ptype->callbacks.gso_segment(skb, features);
3183 __skb_push(skb, skb->data - skb_mac_header(skb));
3187 EXPORT_SYMBOL(skb_mac_gso_segment);
3190 /* openvswitch calls this on rx path, so we need a different check.
3192 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3195 return skb->ip_summed != CHECKSUM_PARTIAL &&
3196 skb->ip_summed != CHECKSUM_UNNECESSARY;
3198 return skb->ip_summed == CHECKSUM_NONE;
3202 * __skb_gso_segment - Perform segmentation on skb.
3203 * @skb: buffer to segment
3204 * @features: features for the output path (see dev->features)
3205 * @tx_path: whether it is called in TX path
3207 * This function segments the given skb and returns a list of segments.
3209 * It may return NULL if the skb requires no segmentation. This is
3210 * only possible when GSO is used for verifying header integrity.
3212 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
3214 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3215 netdev_features_t features, bool tx_path)
3217 struct sk_buff *segs;
3219 if (unlikely(skb_needs_check(skb, tx_path))) {
3222 /* We're going to init ->check field in TCP or UDP header */
3223 err = skb_cow_head(skb, 0);
3225 return ERR_PTR(err);
3228 /* Only report GSO partial support if it will enable us to
3229 * support segmentation on this frame without needing additional
3232 if (features & NETIF_F_GSO_PARTIAL) {
3233 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3234 struct net_device *dev = skb->dev;
3236 partial_features |= dev->features & dev->gso_partial_features;
3237 if (!skb_gso_ok(skb, features | partial_features))
3238 features &= ~NETIF_F_GSO_PARTIAL;
3241 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
3242 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3244 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3245 SKB_GSO_CB(skb)->encap_level = 0;
3247 skb_reset_mac_header(skb);
3248 skb_reset_mac_len(skb);
3250 segs = skb_mac_gso_segment(skb, features);
3252 if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3253 skb_warn_bad_offload(skb);
3257 EXPORT_SYMBOL(__skb_gso_segment);
3259 /* Take action when hardware reception checksum errors are detected. */
3261 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3263 if (net_ratelimit()) {
3264 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3265 skb_dump(KERN_ERR, skb, true);
3269 EXPORT_SYMBOL(netdev_rx_csum_fault);
3272 /* XXX: check that highmem exists at all on the given machine. */
3273 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3275 #ifdef CONFIG_HIGHMEM
3278 if (!(dev->features & NETIF_F_HIGHDMA)) {
3279 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3280 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3282 if (PageHighMem(skb_frag_page(frag)))
3290 /* If MPLS offload request, verify we are testing hardware MPLS features
3291 * instead of standard features for the netdev.
3293 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3294 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3295 netdev_features_t features,
3298 if (eth_p_mpls(type))
3299 features &= skb->dev->mpls_features;
3304 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3305 netdev_features_t features,
3312 static netdev_features_t harmonize_features(struct sk_buff *skb,
3313 netdev_features_t features)
3318 type = skb_network_protocol(skb, &tmp);
3319 features = net_mpls_features(skb, features, type);
3321 if (skb->ip_summed != CHECKSUM_NONE &&
3322 !can_checksum_protocol(features, type)) {
3323 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3325 if (illegal_highdma(skb->dev, skb))
3326 features &= ~NETIF_F_SG;
3331 netdev_features_t passthru_features_check(struct sk_buff *skb,
3332 struct net_device *dev,
3333 netdev_features_t features)
3337 EXPORT_SYMBOL(passthru_features_check);
3339 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3340 struct net_device *dev,
3341 netdev_features_t features)
3343 return vlan_features_check(skb, features);
3346 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3347 struct net_device *dev,
3348 netdev_features_t features)
3350 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3352 if (gso_segs > dev->gso_max_segs)
3353 return features & ~NETIF_F_GSO_MASK;
3355 /* Support for GSO partial features requires software
3356 * intervention before we can actually process the packets
3357 * so we need to strip support for any partial features now
3358 * and we can pull them back in after we have partially
3359 * segmented the frame.
3361 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3362 features &= ~dev->gso_partial_features;
3364 /* Make sure to clear the IPv4 ID mangling feature if the
3365 * IPv4 header has the potential to be fragmented.
3367 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3368 struct iphdr *iph = skb->encapsulation ?
3369 inner_ip_hdr(skb) : ip_hdr(skb);
3371 if (!(iph->frag_off & htons(IP_DF)))
3372 features &= ~NETIF_F_TSO_MANGLEID;
3378 netdev_features_t netif_skb_features(struct sk_buff *skb)
3380 struct net_device *dev = skb->dev;
3381 netdev_features_t features = dev->features;
3383 if (skb_is_gso(skb))
3384 features = gso_features_check(skb, dev, features);
3386 /* If encapsulation offload request, verify we are testing
3387 * hardware encapsulation features instead of standard
3388 * features for the netdev
3390 if (skb->encapsulation)
3391 features &= dev->hw_enc_features;
3393 if (skb_vlan_tagged(skb))
3394 features = netdev_intersect_features(features,
3395 dev->vlan_features |
3396 NETIF_F_HW_VLAN_CTAG_TX |
3397 NETIF_F_HW_VLAN_STAG_TX);
3399 if (dev->netdev_ops->ndo_features_check)
3400 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3403 features &= dflt_features_check(skb, dev, features);
3405 return harmonize_features(skb, features);
3407 EXPORT_SYMBOL(netif_skb_features);
3409 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3410 struct netdev_queue *txq, bool more)
3415 if (dev_nit_active(dev))
3416 dev_queue_xmit_nit(skb, dev);
3419 trace_net_dev_start_xmit(skb, dev);
3420 rc = netdev_start_xmit(skb, dev, txq, more);
3421 trace_net_dev_xmit(skb, rc, dev, len);
3426 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3427 struct netdev_queue *txq, int *ret)
3429 struct sk_buff *skb = first;
3430 int rc = NETDEV_TX_OK;
3433 struct sk_buff *next = skb->next;
3435 skb_mark_not_on_list(skb);
3436 rc = xmit_one(skb, dev, txq, next != NULL);
3437 if (unlikely(!dev_xmit_complete(rc))) {
3443 if (netif_tx_queue_stopped(txq) && skb) {
3444 rc = NETDEV_TX_BUSY;
3454 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3455 netdev_features_t features)
3457 if (skb_vlan_tag_present(skb) &&
3458 !vlan_hw_offload_capable(features, skb->vlan_proto))
3459 skb = __vlan_hwaccel_push_inside(skb);
3463 int skb_csum_hwoffload_help(struct sk_buff *skb,
3464 const netdev_features_t features)
3466 if (unlikely(skb->csum_not_inet))
3467 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3468 skb_crc32c_csum_help(skb);
3470 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3472 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3474 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3476 netdev_features_t features;
3478 features = netif_skb_features(skb);
3479 skb = validate_xmit_vlan(skb, features);
3483 skb = sk_validate_xmit_skb(skb, dev);
3487 if (netif_needs_gso(skb, features)) {
3488 struct sk_buff *segs;
3490 segs = skb_gso_segment(skb, features);
3498 if (skb_needs_linearize(skb, features) &&
3499 __skb_linearize(skb))
3502 /* If packet is not checksummed and device does not
3503 * support checksumming for this protocol, complete
3504 * checksumming here.
3506 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3507 if (skb->encapsulation)
3508 skb_set_inner_transport_header(skb,
3509 skb_checksum_start_offset(skb));
3511 skb_set_transport_header(skb,
3512 skb_checksum_start_offset(skb));
3513 if (skb_csum_hwoffload_help(skb, features))
3518 skb = validate_xmit_xfrm(skb, features, again);
3525 atomic_long_inc(&dev->tx_dropped);
3529 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3531 struct sk_buff *next, *head = NULL, *tail;
3533 for (; skb != NULL; skb = next) {
3535 skb_mark_not_on_list(skb);
3537 /* in case skb wont be segmented, point to itself */
3540 skb = validate_xmit_skb(skb, dev, again);
3548 /* If skb was segmented, skb->prev points to
3549 * the last segment. If not, it still contains skb.
3555 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3557 static void qdisc_pkt_len_init(struct sk_buff *skb)
3559 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3561 qdisc_skb_cb(skb)->pkt_len = skb->len;
3563 /* To get more precise estimation of bytes sent on wire,
3564 * we add to pkt_len the headers size of all segments
3566 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3567 unsigned int hdr_len;
3568 u16 gso_segs = shinfo->gso_segs;
3570 /* mac layer + network layer */
3571 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3573 /* + transport layer */
3574 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3575 const struct tcphdr *th;
3576 struct tcphdr _tcphdr;
3578 th = skb_header_pointer(skb, skb_transport_offset(skb),
3579 sizeof(_tcphdr), &_tcphdr);
3581 hdr_len += __tcp_hdrlen(th);
3583 struct udphdr _udphdr;
3585 if (skb_header_pointer(skb, skb_transport_offset(skb),
3586 sizeof(_udphdr), &_udphdr))
3587 hdr_len += sizeof(struct udphdr);
3590 if (shinfo->gso_type & SKB_GSO_DODGY)
3591 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3594 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3598 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3599 struct net_device *dev,
3600 struct netdev_queue *txq)
3602 spinlock_t *root_lock = qdisc_lock(q);
3603 struct sk_buff *to_free = NULL;
3607 qdisc_calculate_pkt_len(skb, q);
3609 if (q->flags & TCQ_F_NOLOCK) {
3610 if ((q->flags & TCQ_F_CAN_BYPASS) && READ_ONCE(q->empty) &&
3611 qdisc_run_begin(q)) {
3612 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
3614 __qdisc_drop(skb, &to_free);
3618 qdisc_bstats_cpu_update(q, skb);
3620 rc = NET_XMIT_SUCCESS;
3621 if (sch_direct_xmit(skb, q, dev, txq, NULL, true))
3627 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3631 if (unlikely(to_free))
3632 kfree_skb_list(to_free);
3637 * Heuristic to force contended enqueues to serialize on a
3638 * separate lock before trying to get qdisc main lock.
3639 * This permits qdisc->running owner to get the lock more
3640 * often and dequeue packets faster.
3642 contended = qdisc_is_running(q);
3643 if (unlikely(contended))
3644 spin_lock(&q->busylock);
3646 spin_lock(root_lock);
3647 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3648 __qdisc_drop(skb, &to_free);
3650 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3651 qdisc_run_begin(q)) {
3653 * This is a work-conserving queue; there are no old skbs
3654 * waiting to be sent out; and the qdisc is not running -
3655 * xmit the skb directly.
3658 qdisc_bstats_update(q, skb);
3660 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3661 if (unlikely(contended)) {
3662 spin_unlock(&q->busylock);
3669 rc = NET_XMIT_SUCCESS;
3671 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3672 if (qdisc_run_begin(q)) {
3673 if (unlikely(contended)) {
3674 spin_unlock(&q->busylock);
3681 spin_unlock(root_lock);
3682 if (unlikely(to_free))
3683 kfree_skb_list(to_free);
3684 if (unlikely(contended))
3685 spin_unlock(&q->busylock);
3689 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3690 static void skb_update_prio(struct sk_buff *skb)
3692 const struct netprio_map *map;
3693 const struct sock *sk;
3694 unsigned int prioidx;
3698 map = rcu_dereference_bh(skb->dev->priomap);
3701 sk = skb_to_full_sk(skb);
3705 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3707 if (prioidx < map->priomap_len)
3708 skb->priority = map->priomap[prioidx];
3711 #define skb_update_prio(skb)
3715 * dev_loopback_xmit - loop back @skb
3716 * @net: network namespace this loopback is happening in
3717 * @sk: sk needed to be a netfilter okfn
3718 * @skb: buffer to transmit
3720 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3722 skb_reset_mac_header(skb);
3723 __skb_pull(skb, skb_network_offset(skb));
3724 skb->pkt_type = PACKET_LOOPBACK;
3725 skb->ip_summed = CHECKSUM_UNNECESSARY;
3726 WARN_ON(!skb_dst(skb));
3731 EXPORT_SYMBOL(dev_loopback_xmit);
3733 #ifdef CONFIG_NET_EGRESS
3734 static struct sk_buff *
3735 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3737 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3738 struct tcf_result cl_res;
3743 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3744 mini_qdisc_bstats_cpu_update(miniq, skb);
3746 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3748 case TC_ACT_RECLASSIFY:
3749 skb->tc_index = TC_H_MIN(cl_res.classid);
3752 mini_qdisc_qstats_cpu_drop(miniq);
3753 *ret = NET_XMIT_DROP;
3759 *ret = NET_XMIT_SUCCESS;
3762 case TC_ACT_REDIRECT:
3763 /* No need to push/pop skb's mac_header here on egress! */
3764 skb_do_redirect(skb);
3765 *ret = NET_XMIT_SUCCESS;
3773 #endif /* CONFIG_NET_EGRESS */
3776 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3777 struct xps_dev_maps *dev_maps, unsigned int tci)
3779 struct xps_map *map;
3780 int queue_index = -1;
3784 tci += netdev_get_prio_tc_map(dev, skb->priority);
3787 map = rcu_dereference(dev_maps->attr_map[tci]);
3790 queue_index = map->queues[0];
3792 queue_index = map->queues[reciprocal_scale(
3793 skb_get_hash(skb), map->len)];
3794 if (unlikely(queue_index >= dev->real_num_tx_queues))
3801 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3802 struct sk_buff *skb)
3805 struct xps_dev_maps *dev_maps;
3806 struct sock *sk = skb->sk;
3807 int queue_index = -1;
3809 if (!static_key_false(&xps_needed))
3813 if (!static_key_false(&xps_rxqs_needed))
3816 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map);
3818 int tci = sk_rx_queue_get(sk);
3820 if (tci >= 0 && tci < dev->num_rx_queues)
3821 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3826 if (queue_index < 0) {
3827 dev_maps = rcu_dereference(sb_dev->xps_cpus_map);
3829 unsigned int tci = skb->sender_cpu - 1;
3831 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3843 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3844 struct net_device *sb_dev)
3848 EXPORT_SYMBOL(dev_pick_tx_zero);
3850 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
3851 struct net_device *sb_dev)
3853 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3855 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3857 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
3858 struct net_device *sb_dev)
3860 struct sock *sk = skb->sk;
3861 int queue_index = sk_tx_queue_get(sk);
3863 sb_dev = sb_dev ? : dev;
3865 if (queue_index < 0 || skb->ooo_okay ||
3866 queue_index >= dev->real_num_tx_queues) {
3867 int new_index = get_xps_queue(dev, sb_dev, skb);
3870 new_index = skb_tx_hash(dev, sb_dev, skb);
3872 if (queue_index != new_index && sk &&
3874 rcu_access_pointer(sk->sk_dst_cache))
3875 sk_tx_queue_set(sk, new_index);
3877 queue_index = new_index;
3882 EXPORT_SYMBOL(netdev_pick_tx);
3884 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
3885 struct sk_buff *skb,
3886 struct net_device *sb_dev)
3888 int queue_index = 0;
3891 u32 sender_cpu = skb->sender_cpu - 1;
3893 if (sender_cpu >= (u32)NR_CPUS)
3894 skb->sender_cpu = raw_smp_processor_id() + 1;
3897 if (dev->real_num_tx_queues != 1) {
3898 const struct net_device_ops *ops = dev->netdev_ops;
3900 if (ops->ndo_select_queue)
3901 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
3903 queue_index = netdev_pick_tx(dev, skb, sb_dev);
3905 queue_index = netdev_cap_txqueue(dev, queue_index);
3908 skb_set_queue_mapping(skb, queue_index);
3909 return netdev_get_tx_queue(dev, queue_index);
3913 * __dev_queue_xmit - transmit a buffer
3914 * @skb: buffer to transmit
3915 * @sb_dev: suboordinate device used for L2 forwarding offload
3917 * Queue a buffer for transmission to a network device. The caller must
3918 * have set the device and priority and built the buffer before calling
3919 * this function. The function can be called from an interrupt.
3921 * A negative errno code is returned on a failure. A success does not
3922 * guarantee the frame will be transmitted as it may be dropped due
3923 * to congestion or traffic shaping.
3925 * -----------------------------------------------------------------------------------
3926 * I notice this method can also return errors from the queue disciplines,
3927 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3930 * Regardless of the return value, the skb is consumed, so it is currently
3931 * difficult to retry a send to this method. (You can bump the ref count
3932 * before sending to hold a reference for retry if you are careful.)
3934 * When calling this method, interrupts MUST be enabled. This is because
3935 * the BH enable code must have IRQs enabled so that it will not deadlock.
3938 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
3940 struct net_device *dev = skb->dev;
3941 struct netdev_queue *txq;
3946 skb_reset_mac_header(skb);
3948 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3949 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3951 /* Disable soft irqs for various locks below. Also
3952 * stops preemption for RCU.
3956 skb_update_prio(skb);
3958 qdisc_pkt_len_init(skb);
3959 #ifdef CONFIG_NET_CLS_ACT
3960 skb->tc_at_ingress = 0;
3961 # ifdef CONFIG_NET_EGRESS
3962 if (static_branch_unlikely(&egress_needed_key)) {
3963 skb = sch_handle_egress(skb, &rc, dev);
3969 /* If device/qdisc don't need skb->dst, release it right now while
3970 * its hot in this cpu cache.
3972 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3977 txq = netdev_core_pick_tx(dev, skb, sb_dev);
3978 q = rcu_dereference_bh(txq->qdisc);
3980 trace_net_dev_queue(skb);
3982 rc = __dev_xmit_skb(skb, q, dev, txq);
3986 /* The device has no queue. Common case for software devices:
3987 * loopback, all the sorts of tunnels...
3989 * Really, it is unlikely that netif_tx_lock protection is necessary
3990 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
3992 * However, it is possible, that they rely on protection
3995 * Check this and shot the lock. It is not prone from deadlocks.
3996 *Either shot noqueue qdisc, it is even simpler 8)
3998 if (dev->flags & IFF_UP) {
3999 int cpu = smp_processor_id(); /* ok because BHs are off */
4001 if (txq->xmit_lock_owner != cpu) {
4002 if (dev_xmit_recursion())
4003 goto recursion_alert;
4005 skb = validate_xmit_skb(skb, dev, &again);
4009 HARD_TX_LOCK(dev, txq, cpu);
4011 if (!netif_xmit_stopped(txq)) {
4012 dev_xmit_recursion_inc();
4013 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4014 dev_xmit_recursion_dec();
4015 if (dev_xmit_complete(rc)) {
4016 HARD_TX_UNLOCK(dev, txq);
4020 HARD_TX_UNLOCK(dev, txq);
4021 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4024 /* Recursion is detected! It is possible,
4028 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4034 rcu_read_unlock_bh();
4036 atomic_long_inc(&dev->tx_dropped);
4037 kfree_skb_list(skb);
4040 rcu_read_unlock_bh();
4044 int dev_queue_xmit(struct sk_buff *skb)
4046 return __dev_queue_xmit(skb, NULL);
4048 EXPORT_SYMBOL(dev_queue_xmit);
4050 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
4052 return __dev_queue_xmit(skb, sb_dev);
4054 EXPORT_SYMBOL(dev_queue_xmit_accel);
4056 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4058 struct net_device *dev = skb->dev;
4059 struct sk_buff *orig_skb = skb;
4060 struct netdev_queue *txq;
4061 int ret = NETDEV_TX_BUSY;
4064 if (unlikely(!netif_running(dev) ||
4065 !netif_carrier_ok(dev)))
4068 skb = validate_xmit_skb_list(skb, dev, &again);
4069 if (skb != orig_skb)
4072 skb_set_queue_mapping(skb, queue_id);
4073 txq = skb_get_tx_queue(dev, skb);
4077 HARD_TX_LOCK(dev, txq, smp_processor_id());
4078 if (!netif_xmit_frozen_or_drv_stopped(txq))
4079 ret = netdev_start_xmit(skb, dev, txq, false);
4080 HARD_TX_UNLOCK(dev, txq);
4084 if (!dev_xmit_complete(ret))
4089 atomic_long_inc(&dev->tx_dropped);
4090 kfree_skb_list(skb);
4091 return NET_XMIT_DROP;
4093 EXPORT_SYMBOL(dev_direct_xmit);
4095 /*************************************************************************
4097 *************************************************************************/
4099 int netdev_max_backlog __read_mostly = 1000;
4100 EXPORT_SYMBOL(netdev_max_backlog);
4102 int netdev_tstamp_prequeue __read_mostly = 1;
4103 int netdev_budget __read_mostly = 300;
4104 unsigned int __read_mostly netdev_budget_usecs = 2000;
4105 int weight_p __read_mostly = 64; /* old backlog weight */
4106 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4107 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4108 int dev_rx_weight __read_mostly = 64;
4109 int dev_tx_weight __read_mostly = 64;
4110 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
4111 int gro_normal_batch __read_mostly = 8;
4113 /* Called with irq disabled */
4114 static inline void ____napi_schedule(struct softnet_data *sd,
4115 struct napi_struct *napi)
4117 list_add_tail(&napi->poll_list, &sd->poll_list);
4118 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4123 /* One global table that all flow-based protocols share. */
4124 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4125 EXPORT_SYMBOL(rps_sock_flow_table);
4126 u32 rps_cpu_mask __read_mostly;
4127 EXPORT_SYMBOL(rps_cpu_mask);
4129 struct static_key_false rps_needed __read_mostly;
4130 EXPORT_SYMBOL(rps_needed);
4131 struct static_key_false rfs_needed __read_mostly;
4132 EXPORT_SYMBOL(rfs_needed);
4134 static struct rps_dev_flow *
4135 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4136 struct rps_dev_flow *rflow, u16 next_cpu)
4138 if (next_cpu < nr_cpu_ids) {
4139 #ifdef CONFIG_RFS_ACCEL
4140 struct netdev_rx_queue *rxqueue;
4141 struct rps_dev_flow_table *flow_table;
4142 struct rps_dev_flow *old_rflow;
4147 /* Should we steer this flow to a different hardware queue? */
4148 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4149 !(dev->features & NETIF_F_NTUPLE))
4151 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4152 if (rxq_index == skb_get_rx_queue(skb))
4155 rxqueue = dev->_rx + rxq_index;
4156 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4159 flow_id = skb_get_hash(skb) & flow_table->mask;
4160 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4161 rxq_index, flow_id);
4165 rflow = &flow_table->flows[flow_id];
4167 if (old_rflow->filter == rflow->filter)
4168 old_rflow->filter = RPS_NO_FILTER;
4172 per_cpu(softnet_data, next_cpu).input_queue_head;
4175 rflow->cpu = next_cpu;
4180 * get_rps_cpu is called from netif_receive_skb and returns the target
4181 * CPU from the RPS map of the receiving queue for a given skb.
4182 * rcu_read_lock must be held on entry.
4184 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4185 struct rps_dev_flow **rflowp)
4187 const struct rps_sock_flow_table *sock_flow_table;
4188 struct netdev_rx_queue *rxqueue = dev->_rx;
4189 struct rps_dev_flow_table *flow_table;
4190 struct rps_map *map;
4195 if (skb_rx_queue_recorded(skb)) {
4196 u16 index = skb_get_rx_queue(skb);
4198 if (unlikely(index >= dev->real_num_rx_queues)) {
4199 WARN_ONCE(dev->real_num_rx_queues > 1,
4200 "%s received packet on queue %u, but number "
4201 "of RX queues is %u\n",
4202 dev->name, index, dev->real_num_rx_queues);
4208 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4210 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4211 map = rcu_dereference(rxqueue->rps_map);
4212 if (!flow_table && !map)
4215 skb_reset_network_header(skb);
4216 hash = skb_get_hash(skb);
4220 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4221 if (flow_table && sock_flow_table) {
4222 struct rps_dev_flow *rflow;
4226 /* First check into global flow table if there is a match */
4227 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4228 if ((ident ^ hash) & ~rps_cpu_mask)
4231 next_cpu = ident & rps_cpu_mask;
4233 /* OK, now we know there is a match,
4234 * we can look at the local (per receive queue) flow table
4236 rflow = &flow_table->flows[hash & flow_table->mask];
4240 * If the desired CPU (where last recvmsg was done) is
4241 * different from current CPU (one in the rx-queue flow
4242 * table entry), switch if one of the following holds:
4243 * - Current CPU is unset (>= nr_cpu_ids).
4244 * - Current CPU is offline.
4245 * - The current CPU's queue tail has advanced beyond the
4246 * last packet that was enqueued using this table entry.
4247 * This guarantees that all previous packets for the flow
4248 * have been dequeued, thus preserving in order delivery.
4250 if (unlikely(tcpu != next_cpu) &&
4251 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4252 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4253 rflow->last_qtail)) >= 0)) {
4255 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4258 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4268 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4269 if (cpu_online(tcpu)) {
4279 #ifdef CONFIG_RFS_ACCEL
4282 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4283 * @dev: Device on which the filter was set
4284 * @rxq_index: RX queue index
4285 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4286 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4288 * Drivers that implement ndo_rx_flow_steer() should periodically call
4289 * this function for each installed filter and remove the filters for
4290 * which it returns %true.
4292 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4293 u32 flow_id, u16 filter_id)
4295 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4296 struct rps_dev_flow_table *flow_table;
4297 struct rps_dev_flow *rflow;
4302 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4303 if (flow_table && flow_id <= flow_table->mask) {
4304 rflow = &flow_table->flows[flow_id];
4305 cpu = READ_ONCE(rflow->cpu);
4306 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4307 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4308 rflow->last_qtail) <
4309 (int)(10 * flow_table->mask)))
4315 EXPORT_SYMBOL(rps_may_expire_flow);
4317 #endif /* CONFIG_RFS_ACCEL */
4319 /* Called from hardirq (IPI) context */
4320 static void rps_trigger_softirq(void *data)
4322 struct softnet_data *sd = data;
4324 ____napi_schedule(sd, &sd->backlog);
4328 #endif /* CONFIG_RPS */
4331 * Check if this softnet_data structure is another cpu one
4332 * If yes, queue it to our IPI list and return 1
4335 static int rps_ipi_queued(struct softnet_data *sd)
4338 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4341 sd->rps_ipi_next = mysd->rps_ipi_list;
4342 mysd->rps_ipi_list = sd;
4344 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4347 #endif /* CONFIG_RPS */
4351 #ifdef CONFIG_NET_FLOW_LIMIT
4352 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4355 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4357 #ifdef CONFIG_NET_FLOW_LIMIT
4358 struct sd_flow_limit *fl;
4359 struct softnet_data *sd;
4360 unsigned int old_flow, new_flow;
4362 if (qlen < (netdev_max_backlog >> 1))
4365 sd = this_cpu_ptr(&softnet_data);
4368 fl = rcu_dereference(sd->flow_limit);
4370 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4371 old_flow = fl->history[fl->history_head];
4372 fl->history[fl->history_head] = new_flow;
4375 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4377 if (likely(fl->buckets[old_flow]))
4378 fl->buckets[old_flow]--;
4380 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4392 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4393 * queue (may be a remote CPU queue).
4395 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4396 unsigned int *qtail)
4398 struct softnet_data *sd;
4399 unsigned long flags;
4402 sd = &per_cpu(softnet_data, cpu);
4404 local_irq_save(flags);
4407 if (!netif_running(skb->dev))
4409 qlen = skb_queue_len(&sd->input_pkt_queue);
4410 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4413 __skb_queue_tail(&sd->input_pkt_queue, skb);
4414 input_queue_tail_incr_save(sd, qtail);
4416 local_irq_restore(flags);
4417 return NET_RX_SUCCESS;
4420 /* Schedule NAPI for backlog device
4421 * We can use non atomic operation since we own the queue lock
4423 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4424 if (!rps_ipi_queued(sd))
4425 ____napi_schedule(sd, &sd->backlog);
4434 local_irq_restore(flags);
4436 atomic_long_inc(&skb->dev->rx_dropped);
4441 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4443 struct net_device *dev = skb->dev;
4444 struct netdev_rx_queue *rxqueue;
4448 if (skb_rx_queue_recorded(skb)) {
4449 u16 index = skb_get_rx_queue(skb);
4451 if (unlikely(index >= dev->real_num_rx_queues)) {
4452 WARN_ONCE(dev->real_num_rx_queues > 1,
4453 "%s received packet on queue %u, but number "
4454 "of RX queues is %u\n",
4455 dev->name, index, dev->real_num_rx_queues);
4457 return rxqueue; /* Return first rxqueue */
4464 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4465 struct xdp_buff *xdp,
4466 struct bpf_prog *xdp_prog)
4468 struct netdev_rx_queue *rxqueue;
4469 void *orig_data, *orig_data_end;
4470 u32 metalen, act = XDP_DROP;
4471 __be16 orig_eth_type;
4477 /* Reinjected packets coming from act_mirred or similar should
4478 * not get XDP generic processing.
4480 if (skb_cloned(skb) || skb_is_tc_redirected(skb))
4483 /* XDP packets must be linear and must have sufficient headroom
4484 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4485 * native XDP provides, thus we need to do it here as well.
4487 if (skb_is_nonlinear(skb) ||
4488 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4489 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4490 int troom = skb->tail + skb->data_len - skb->end;
4492 /* In case we have to go down the path and also linearize,
4493 * then lets do the pskb_expand_head() work just once here.
4495 if (pskb_expand_head(skb,
4496 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4497 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4499 if (skb_linearize(skb))
4503 /* The XDP program wants to see the packet starting at the MAC
4506 mac_len = skb->data - skb_mac_header(skb);
4507 hlen = skb_headlen(skb) + mac_len;
4508 xdp->data = skb->data - mac_len;
4509 xdp->data_meta = xdp->data;
4510 xdp->data_end = xdp->data + hlen;
4511 xdp->data_hard_start = skb->data - skb_headroom(skb);
4512 orig_data_end = xdp->data_end;
4513 orig_data = xdp->data;
4514 eth = (struct ethhdr *)xdp->data;
4515 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4516 orig_eth_type = eth->h_proto;
4518 rxqueue = netif_get_rxqueue(skb);
4519 xdp->rxq = &rxqueue->xdp_rxq;
4521 act = bpf_prog_run_xdp(xdp_prog, xdp);
4523 /* check if bpf_xdp_adjust_head was used */
4524 off = xdp->data - orig_data;
4527 __skb_pull(skb, off);
4529 __skb_push(skb, -off);
4531 skb->mac_header += off;
4532 skb_reset_network_header(skb);
4535 /* check if bpf_xdp_adjust_tail was used. it can only "shrink"
4538 off = orig_data_end - xdp->data_end;
4540 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4545 /* check if XDP changed eth hdr such SKB needs update */
4546 eth = (struct ethhdr *)xdp->data;
4547 if ((orig_eth_type != eth->h_proto) ||
4548 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4549 __skb_push(skb, ETH_HLEN);
4550 skb->protocol = eth_type_trans(skb, skb->dev);
4556 __skb_push(skb, mac_len);
4559 metalen = xdp->data - xdp->data_meta;
4561 skb_metadata_set(skb, metalen);
4564 bpf_warn_invalid_xdp_action(act);
4567 trace_xdp_exception(skb->dev, xdp_prog, act);
4578 /* When doing generic XDP we have to bypass the qdisc layer and the
4579 * network taps in order to match in-driver-XDP behavior.
4581 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4583 struct net_device *dev = skb->dev;
4584 struct netdev_queue *txq;
4585 bool free_skb = true;
4588 txq = netdev_core_pick_tx(dev, skb, NULL);
4589 cpu = smp_processor_id();
4590 HARD_TX_LOCK(dev, txq, cpu);
4591 if (!netif_xmit_stopped(txq)) {
4592 rc = netdev_start_xmit(skb, dev, txq, 0);
4593 if (dev_xmit_complete(rc))
4596 HARD_TX_UNLOCK(dev, txq);
4598 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4602 EXPORT_SYMBOL_GPL(generic_xdp_tx);
4604 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4606 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4609 struct xdp_buff xdp;
4613 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4614 if (act != XDP_PASS) {
4617 err = xdp_do_generic_redirect(skb->dev, skb,
4623 generic_xdp_tx(skb, xdp_prog);
4634 EXPORT_SYMBOL_GPL(do_xdp_generic);
4636 static int netif_rx_internal(struct sk_buff *skb)
4640 net_timestamp_check(netdev_tstamp_prequeue, skb);
4642 trace_netif_rx(skb);
4645 if (static_branch_unlikely(&rps_needed)) {
4646 struct rps_dev_flow voidflow, *rflow = &voidflow;
4652 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4654 cpu = smp_processor_id();
4656 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4665 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4672 * netif_rx - post buffer to the network code
4673 * @skb: buffer to post
4675 * This function receives a packet from a device driver and queues it for
4676 * the upper (protocol) levels to process. It always succeeds. The buffer
4677 * may be dropped during processing for congestion control or by the
4681 * NET_RX_SUCCESS (no congestion)
4682 * NET_RX_DROP (packet was dropped)
4686 int netif_rx(struct sk_buff *skb)
4690 trace_netif_rx_entry(skb);
4692 ret = netif_rx_internal(skb);
4693 trace_netif_rx_exit(ret);
4697 EXPORT_SYMBOL(netif_rx);
4699 int netif_rx_ni(struct sk_buff *skb)
4703 trace_netif_rx_ni_entry(skb);
4706 err = netif_rx_internal(skb);
4707 if (local_softirq_pending())
4710 trace_netif_rx_ni_exit(err);
4714 EXPORT_SYMBOL(netif_rx_ni);
4716 static __latent_entropy void net_tx_action(struct softirq_action *h)
4718 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4720 if (sd->completion_queue) {
4721 struct sk_buff *clist;
4723 local_irq_disable();
4724 clist = sd->completion_queue;
4725 sd->completion_queue = NULL;
4729 struct sk_buff *skb = clist;
4731 clist = clist->next;
4733 WARN_ON(refcount_read(&skb->users));
4734 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4735 trace_consume_skb(skb);
4737 trace_kfree_skb(skb, net_tx_action);
4739 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4742 __kfree_skb_defer(skb);
4745 __kfree_skb_flush();
4748 if (sd->output_queue) {
4751 local_irq_disable();
4752 head = sd->output_queue;
4753 sd->output_queue = NULL;
4754 sd->output_queue_tailp = &sd->output_queue;
4758 struct Qdisc *q = head;
4759 spinlock_t *root_lock = NULL;
4761 head = head->next_sched;
4763 if (!(q->flags & TCQ_F_NOLOCK)) {
4764 root_lock = qdisc_lock(q);
4765 spin_lock(root_lock);
4767 /* We need to make sure head->next_sched is read
4768 * before clearing __QDISC_STATE_SCHED
4770 smp_mb__before_atomic();
4771 clear_bit(__QDISC_STATE_SCHED, &q->state);
4774 spin_unlock(root_lock);
4778 xfrm_dev_backlog(sd);
4781 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4782 /* This hook is defined here for ATM LANE */
4783 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4784 unsigned char *addr) __read_mostly;
4785 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4788 static inline struct sk_buff *
4789 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4790 struct net_device *orig_dev)
4792 #ifdef CONFIG_NET_CLS_ACT
4793 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4794 struct tcf_result cl_res;
4796 /* If there's at least one ingress present somewhere (so
4797 * we get here via enabled static key), remaining devices
4798 * that are not configured with an ingress qdisc will bail
4805 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4809 qdisc_skb_cb(skb)->pkt_len = skb->len;
4810 skb->tc_at_ingress = 1;
4811 mini_qdisc_bstats_cpu_update(miniq, skb);
4813 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
4815 case TC_ACT_RECLASSIFY:
4816 skb->tc_index = TC_H_MIN(cl_res.classid);
4819 mini_qdisc_qstats_cpu_drop(miniq);
4827 case TC_ACT_REDIRECT:
4828 /* skb_mac_header check was done by cls/act_bpf, so
4829 * we can safely push the L2 header back before
4830 * redirecting to another netdev
4832 __skb_push(skb, skb->mac_len);
4833 skb_do_redirect(skb);
4835 case TC_ACT_CONSUMED:
4840 #endif /* CONFIG_NET_CLS_ACT */
4845 * netdev_is_rx_handler_busy - check if receive handler is registered
4846 * @dev: device to check
4848 * Check if a receive handler is already registered for a given device.
4849 * Return true if there one.
4851 * The caller must hold the rtnl_mutex.
4853 bool netdev_is_rx_handler_busy(struct net_device *dev)
4856 return dev && rtnl_dereference(dev->rx_handler);
4858 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4861 * netdev_rx_handler_register - register receive handler
4862 * @dev: device to register a handler for
4863 * @rx_handler: receive handler to register
4864 * @rx_handler_data: data pointer that is used by rx handler
4866 * Register a receive handler for a device. This handler will then be
4867 * called from __netif_receive_skb. A negative errno code is returned
4870 * The caller must hold the rtnl_mutex.
4872 * For a general description of rx_handler, see enum rx_handler_result.
4874 int netdev_rx_handler_register(struct net_device *dev,
4875 rx_handler_func_t *rx_handler,
4876 void *rx_handler_data)
4878 if (netdev_is_rx_handler_busy(dev))
4881 if (dev->priv_flags & IFF_NO_RX_HANDLER)
4884 /* Note: rx_handler_data must be set before rx_handler */
4885 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4886 rcu_assign_pointer(dev->rx_handler, rx_handler);
4890 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4893 * netdev_rx_handler_unregister - unregister receive handler
4894 * @dev: device to unregister a handler from
4896 * Unregister a receive handler from a device.
4898 * The caller must hold the rtnl_mutex.
4900 void netdev_rx_handler_unregister(struct net_device *dev)
4904 RCU_INIT_POINTER(dev->rx_handler, NULL);
4905 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4906 * section has a guarantee to see a non NULL rx_handler_data
4910 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4912 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4915 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4916 * the special handling of PFMEMALLOC skbs.
4918 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4920 switch (skb->protocol) {
4921 case htons(ETH_P_ARP):
4922 case htons(ETH_P_IP):
4923 case htons(ETH_P_IPV6):
4924 case htons(ETH_P_8021Q):
4925 case htons(ETH_P_8021AD):
4932 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4933 int *ret, struct net_device *orig_dev)
4935 if (nf_hook_ingress_active(skb)) {
4939 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4944 ingress_retval = nf_hook_ingress(skb);
4946 return ingress_retval;
4951 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
4952 struct packet_type **ppt_prev)
4954 struct packet_type *ptype, *pt_prev;
4955 rx_handler_func_t *rx_handler;
4956 struct net_device *orig_dev;
4957 bool deliver_exact = false;
4958 int ret = NET_RX_DROP;
4961 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4963 trace_netif_receive_skb(skb);
4965 orig_dev = skb->dev;
4967 skb_reset_network_header(skb);
4968 if (!skb_transport_header_was_set(skb))
4969 skb_reset_transport_header(skb);
4970 skb_reset_mac_len(skb);
4975 skb->skb_iif = skb->dev->ifindex;
4977 __this_cpu_inc(softnet_data.processed);
4979 if (static_branch_unlikely(&generic_xdp_needed_key)) {
4983 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
4986 if (ret2 != XDP_PASS)
4988 skb_reset_mac_len(skb);
4991 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4992 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4993 skb = skb_vlan_untag(skb);
4998 if (skb_skip_tc_classify(skb))
5004 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5006 ret = deliver_skb(skb, pt_prev, orig_dev);
5010 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5012 ret = deliver_skb(skb, pt_prev, orig_dev);
5017 #ifdef CONFIG_NET_INGRESS
5018 if (static_branch_unlikely(&ingress_needed_key)) {
5019 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
5023 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5029 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5032 if (skb_vlan_tag_present(skb)) {
5034 ret = deliver_skb(skb, pt_prev, orig_dev);
5037 if (vlan_do_receive(&skb))
5039 else if (unlikely(!skb))
5043 rx_handler = rcu_dereference(skb->dev->rx_handler);
5046 ret = deliver_skb(skb, pt_prev, orig_dev);
5049 switch (rx_handler(&skb)) {
5050 case RX_HANDLER_CONSUMED:
5051 ret = NET_RX_SUCCESS;
5053 case RX_HANDLER_ANOTHER:
5055 case RX_HANDLER_EXACT:
5056 deliver_exact = true;
5057 case RX_HANDLER_PASS:
5064 if (unlikely(skb_vlan_tag_present(skb))) {
5066 if (skb_vlan_tag_get_id(skb)) {
5067 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5070 skb->pkt_type = PACKET_OTHERHOST;
5071 } else if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
5072 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
5073 /* Outer header is 802.1P with vlan 0, inner header is
5074 * 802.1Q or 802.1AD and vlan_do_receive() above could
5075 * not find vlan dev for vlan id 0.
5077 __vlan_hwaccel_clear_tag(skb);
5078 skb = skb_vlan_untag(skb);
5081 if (vlan_do_receive(&skb))
5082 /* After stripping off 802.1P header with vlan 0
5083 * vlan dev is found for inner header.
5086 else if (unlikely(!skb))
5089 /* We have stripped outer 802.1P vlan 0 header.
5090 * But could not find vlan dev.
5091 * check again for vlan id to set OTHERHOST.
5095 /* Note: we might in the future use prio bits
5096 * and set skb->priority like in vlan_do_receive()
5097 * For the time being, just ignore Priority Code Point
5099 __vlan_hwaccel_clear_tag(skb);
5102 type = skb->protocol;
5104 /* deliver only exact match when indicated */
5105 if (likely(!deliver_exact)) {
5106 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5107 &ptype_base[ntohs(type) &
5111 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5112 &orig_dev->ptype_specific);
5114 if (unlikely(skb->dev != orig_dev)) {
5115 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5116 &skb->dev->ptype_specific);
5120 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5122 *ppt_prev = pt_prev;
5126 atomic_long_inc(&skb->dev->rx_dropped);
5128 atomic_long_inc(&skb->dev->rx_nohandler);
5130 /* Jamal, now you will not able to escape explaining
5131 * me how you were going to use this. :-)
5140 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5142 struct net_device *orig_dev = skb->dev;
5143 struct packet_type *pt_prev = NULL;
5146 ret = __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
5148 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5149 skb->dev, pt_prev, orig_dev);
5154 * netif_receive_skb_core - special purpose version of netif_receive_skb
5155 * @skb: buffer to process
5157 * More direct receive version of netif_receive_skb(). It should
5158 * only be used by callers that have a need to skip RPS and Generic XDP.
5159 * Caller must also take care of handling if (page_is_)pfmemalloc.
5161 * This function may only be called from softirq context and interrupts
5162 * should be enabled.
5164 * Return values (usually ignored):
5165 * NET_RX_SUCCESS: no congestion
5166 * NET_RX_DROP: packet was dropped
5168 int netif_receive_skb_core(struct sk_buff *skb)
5173 ret = __netif_receive_skb_one_core(skb, false);
5178 EXPORT_SYMBOL(netif_receive_skb_core);
5180 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5181 struct packet_type *pt_prev,
5182 struct net_device *orig_dev)
5184 struct sk_buff *skb, *next;
5188 if (list_empty(head))
5190 if (pt_prev->list_func != NULL)
5191 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5192 ip_list_rcv, head, pt_prev, orig_dev);
5194 list_for_each_entry_safe(skb, next, head, list) {
5195 skb_list_del_init(skb);
5196 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5200 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5202 /* Fast-path assumptions:
5203 * - There is no RX handler.
5204 * - Only one packet_type matches.
5205 * If either of these fails, we will end up doing some per-packet
5206 * processing in-line, then handling the 'last ptype' for the whole
5207 * sublist. This can't cause out-of-order delivery to any single ptype,
5208 * because the 'last ptype' must be constant across the sublist, and all
5209 * other ptypes are handled per-packet.
5211 /* Current (common) ptype of sublist */
5212 struct packet_type *pt_curr = NULL;
5213 /* Current (common) orig_dev of sublist */
5214 struct net_device *od_curr = NULL;
5215 struct list_head sublist;
5216 struct sk_buff *skb, *next;
5218 INIT_LIST_HEAD(&sublist);
5219 list_for_each_entry_safe(skb, next, head, list) {
5220 struct net_device *orig_dev = skb->dev;
5221 struct packet_type *pt_prev = NULL;
5223 skb_list_del_init(skb);
5224 __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
5227 if (pt_curr != pt_prev || od_curr != orig_dev) {
5228 /* dispatch old sublist */
5229 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5230 /* start new sublist */
5231 INIT_LIST_HEAD(&sublist);
5235 list_add_tail(&skb->list, &sublist);
5238 /* dispatch final sublist */
5239 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5242 static int __netif_receive_skb(struct sk_buff *skb)
5246 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5247 unsigned int noreclaim_flag;
5250 * PFMEMALLOC skbs are special, they should
5251 * - be delivered to SOCK_MEMALLOC sockets only
5252 * - stay away from userspace
5253 * - have bounded memory usage
5255 * Use PF_MEMALLOC as this saves us from propagating the allocation
5256 * context down to all allocation sites.
5258 noreclaim_flag = memalloc_noreclaim_save();
5259 ret = __netif_receive_skb_one_core(skb, true);
5260 memalloc_noreclaim_restore(noreclaim_flag);
5262 ret = __netif_receive_skb_one_core(skb, false);
5267 static void __netif_receive_skb_list(struct list_head *head)
5269 unsigned long noreclaim_flag = 0;
5270 struct sk_buff *skb, *next;
5271 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5273 list_for_each_entry_safe(skb, next, head, list) {
5274 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5275 struct list_head sublist;
5277 /* Handle the previous sublist */
5278 list_cut_before(&sublist, head, &skb->list);
5279 if (!list_empty(&sublist))
5280 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5281 pfmemalloc = !pfmemalloc;
5282 /* See comments in __netif_receive_skb */
5284 noreclaim_flag = memalloc_noreclaim_save();
5286 memalloc_noreclaim_restore(noreclaim_flag);
5289 /* Handle the remaining sublist */
5290 if (!list_empty(head))
5291 __netif_receive_skb_list_core(head, pfmemalloc);
5292 /* Restore pflags */
5294 memalloc_noreclaim_restore(noreclaim_flag);
5297 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5299 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5300 struct bpf_prog *new = xdp->prog;
5303 switch (xdp->command) {
5304 case XDP_SETUP_PROG:
5305 rcu_assign_pointer(dev->xdp_prog, new);
5310 static_branch_dec(&generic_xdp_needed_key);
5311 } else if (new && !old) {
5312 static_branch_inc(&generic_xdp_needed_key);
5313 dev_disable_lro(dev);
5314 dev_disable_gro_hw(dev);
5318 case XDP_QUERY_PROG:
5319 xdp->prog_id = old ? old->aux->id : 0;
5330 static int netif_receive_skb_internal(struct sk_buff *skb)
5334 net_timestamp_check(netdev_tstamp_prequeue, skb);
5336 if (skb_defer_rx_timestamp(skb))
5337 return NET_RX_SUCCESS;
5341 if (static_branch_unlikely(&rps_needed)) {
5342 struct rps_dev_flow voidflow, *rflow = &voidflow;
5343 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5346 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5352 ret = __netif_receive_skb(skb);
5357 static void netif_receive_skb_list_internal(struct list_head *head)
5359 struct sk_buff *skb, *next;
5360 struct list_head sublist;
5362 INIT_LIST_HEAD(&sublist);
5363 list_for_each_entry_safe(skb, next, head, list) {
5364 net_timestamp_check(netdev_tstamp_prequeue, skb);
5365 skb_list_del_init(skb);
5366 if (!skb_defer_rx_timestamp(skb))
5367 list_add_tail(&skb->list, &sublist);
5369 list_splice_init(&sublist, head);
5373 if (static_branch_unlikely(&rps_needed)) {
5374 list_for_each_entry_safe(skb, next, head, list) {
5375 struct rps_dev_flow voidflow, *rflow = &voidflow;
5376 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5379 /* Will be handled, remove from list */
5380 skb_list_del_init(skb);
5381 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5386 __netif_receive_skb_list(head);
5391 * netif_receive_skb - process receive buffer from network
5392 * @skb: buffer to process
5394 * netif_receive_skb() is the main receive data processing function.
5395 * It always succeeds. The buffer may be dropped during processing
5396 * for congestion control or by the protocol layers.
5398 * This function may only be called from softirq context and interrupts
5399 * should be enabled.
5401 * Return values (usually ignored):
5402 * NET_RX_SUCCESS: no congestion
5403 * NET_RX_DROP: packet was dropped
5405 int netif_receive_skb(struct sk_buff *skb)
5409 trace_netif_receive_skb_entry(skb);
5411 ret = netif_receive_skb_internal(skb);
5412 trace_netif_receive_skb_exit(ret);
5416 EXPORT_SYMBOL(netif_receive_skb);
5419 * netif_receive_skb_list - process many receive buffers from network
5420 * @head: list of skbs to process.
5422 * Since return value of netif_receive_skb() is normally ignored, and
5423 * wouldn't be meaningful for a list, this function returns void.
5425 * This function may only be called from softirq context and interrupts
5426 * should be enabled.
5428 void netif_receive_skb_list(struct list_head *head)
5430 struct sk_buff *skb;
5432 if (list_empty(head))
5434 if (trace_netif_receive_skb_list_entry_enabled()) {
5435 list_for_each_entry(skb, head, list)
5436 trace_netif_receive_skb_list_entry(skb);
5438 netif_receive_skb_list_internal(head);
5439 trace_netif_receive_skb_list_exit(0);
5441 EXPORT_SYMBOL(netif_receive_skb_list);
5443 DEFINE_PER_CPU(struct work_struct, flush_works);
5445 /* Network device is going away, flush any packets still pending */
5446 static void flush_backlog(struct work_struct *work)
5448 struct sk_buff *skb, *tmp;
5449 struct softnet_data *sd;
5452 sd = this_cpu_ptr(&softnet_data);
5454 local_irq_disable();
5456 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5457 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5458 __skb_unlink(skb, &sd->input_pkt_queue);
5460 input_queue_head_incr(sd);
5466 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5467 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5468 __skb_unlink(skb, &sd->process_queue);
5470 input_queue_head_incr(sd);
5476 static void flush_all_backlogs(void)
5482 for_each_online_cpu(cpu)
5483 queue_work_on(cpu, system_highpri_wq,
5484 per_cpu_ptr(&flush_works, cpu));
5486 for_each_online_cpu(cpu)
5487 flush_work(per_cpu_ptr(&flush_works, cpu));
5492 INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int));
5493 INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int));
5494 static int napi_gro_complete(struct sk_buff *skb)
5496 struct packet_offload *ptype;
5497 __be16 type = skb->protocol;
5498 struct list_head *head = &offload_base;
5501 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5503 if (NAPI_GRO_CB(skb)->count == 1) {
5504 skb_shinfo(skb)->gso_size = 0;
5509 list_for_each_entry_rcu(ptype, head, list) {
5510 if (ptype->type != type || !ptype->callbacks.gro_complete)
5513 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5514 ipv6_gro_complete, inet_gro_complete,
5521 WARN_ON(&ptype->list == head);
5523 return NET_RX_SUCCESS;
5527 return netif_receive_skb_internal(skb);
5530 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5533 struct list_head *head = &napi->gro_hash[index].list;
5534 struct sk_buff *skb, *p;
5536 list_for_each_entry_safe_reverse(skb, p, head, list) {
5537 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5539 skb_list_del_init(skb);
5540 napi_gro_complete(skb);
5541 napi->gro_hash[index].count--;
5544 if (!napi->gro_hash[index].count)
5545 __clear_bit(index, &napi->gro_bitmask);
5548 /* napi->gro_hash[].list contains packets ordered by age.
5549 * youngest packets at the head of it.
5550 * Complete skbs in reverse order to reduce latencies.
5552 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5554 unsigned long bitmask = napi->gro_bitmask;
5555 unsigned int i, base = ~0U;
5557 while ((i = ffs(bitmask)) != 0) {
5560 __napi_gro_flush_chain(napi, base, flush_old);
5563 EXPORT_SYMBOL(napi_gro_flush);
5565 static struct list_head *gro_list_prepare(struct napi_struct *napi,
5566 struct sk_buff *skb)
5568 unsigned int maclen = skb->dev->hard_header_len;
5569 u32 hash = skb_get_hash_raw(skb);
5570 struct list_head *head;
5573 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list;
5574 list_for_each_entry(p, head, list) {
5575 unsigned long diffs;
5577 NAPI_GRO_CB(p)->flush = 0;
5579 if (hash != skb_get_hash_raw(p)) {
5580 NAPI_GRO_CB(p)->same_flow = 0;
5584 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5585 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5586 if (skb_vlan_tag_present(p))
5587 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb);
5588 diffs |= skb_metadata_dst_cmp(p, skb);
5589 diffs |= skb_metadata_differs(p, skb);
5590 if (maclen == ETH_HLEN)
5591 diffs |= compare_ether_header(skb_mac_header(p),
5592 skb_mac_header(skb));
5594 diffs = memcmp(skb_mac_header(p),
5595 skb_mac_header(skb),
5597 NAPI_GRO_CB(p)->same_flow = !diffs;
5603 static void skb_gro_reset_offset(struct sk_buff *skb)
5605 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5606 const skb_frag_t *frag0 = &pinfo->frags[0];
5608 NAPI_GRO_CB(skb)->data_offset = 0;
5609 NAPI_GRO_CB(skb)->frag0 = NULL;
5610 NAPI_GRO_CB(skb)->frag0_len = 0;
5612 if (!skb_headlen(skb) && pinfo->nr_frags &&
5613 !PageHighMem(skb_frag_page(frag0))) {
5614 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5615 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5616 skb_frag_size(frag0),
5617 skb->end - skb->tail);
5621 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5623 struct skb_shared_info *pinfo = skb_shinfo(skb);
5625 BUG_ON(skb->end - skb->tail < grow);
5627 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5629 skb->data_len -= grow;
5632 skb_frag_off_add(&pinfo->frags[0], grow);
5633 skb_frag_size_sub(&pinfo->frags[0], grow);
5635 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5636 skb_frag_unref(skb, 0);
5637 memmove(pinfo->frags, pinfo->frags + 1,
5638 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
5642 static void gro_flush_oldest(struct list_head *head)
5644 struct sk_buff *oldest;
5646 oldest = list_last_entry(head, struct sk_buff, list);
5648 /* We are called with head length >= MAX_GRO_SKBS, so this is
5651 if (WARN_ON_ONCE(!oldest))
5654 /* Do not adjust napi->gro_hash[].count, caller is adding a new
5657 skb_list_del_init(oldest);
5658 napi_gro_complete(oldest);
5661 INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *,
5663 INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *,
5665 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5667 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
5668 struct list_head *head = &offload_base;
5669 struct packet_offload *ptype;
5670 __be16 type = skb->protocol;
5671 struct list_head *gro_head;
5672 struct sk_buff *pp = NULL;
5673 enum gro_result ret;
5677 if (netif_elide_gro(skb->dev))
5680 gro_head = gro_list_prepare(napi, skb);
5683 list_for_each_entry_rcu(ptype, head, list) {
5684 if (ptype->type != type || !ptype->callbacks.gro_receive)
5687 skb_set_network_header(skb, skb_gro_offset(skb));
5688 skb_reset_mac_len(skb);
5689 NAPI_GRO_CB(skb)->same_flow = 0;
5690 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
5691 NAPI_GRO_CB(skb)->free = 0;
5692 NAPI_GRO_CB(skb)->encap_mark = 0;
5693 NAPI_GRO_CB(skb)->recursion_counter = 0;
5694 NAPI_GRO_CB(skb)->is_fou = 0;
5695 NAPI_GRO_CB(skb)->is_atomic = 1;
5696 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
5698 /* Setup for GRO checksum validation */
5699 switch (skb->ip_summed) {
5700 case CHECKSUM_COMPLETE:
5701 NAPI_GRO_CB(skb)->csum = skb->csum;
5702 NAPI_GRO_CB(skb)->csum_valid = 1;
5703 NAPI_GRO_CB(skb)->csum_cnt = 0;
5705 case CHECKSUM_UNNECESSARY:
5706 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
5707 NAPI_GRO_CB(skb)->csum_valid = 0;
5710 NAPI_GRO_CB(skb)->csum_cnt = 0;
5711 NAPI_GRO_CB(skb)->csum_valid = 0;
5714 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
5715 ipv6_gro_receive, inet_gro_receive,
5721 if (&ptype->list == head)
5724 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
5729 same_flow = NAPI_GRO_CB(skb)->same_flow;
5730 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5733 skb_list_del_init(pp);
5734 napi_gro_complete(pp);
5735 napi->gro_hash[hash].count--;
5741 if (NAPI_GRO_CB(skb)->flush)
5744 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
5745 gro_flush_oldest(gro_head);
5747 napi->gro_hash[hash].count++;
5749 NAPI_GRO_CB(skb)->count = 1;
5750 NAPI_GRO_CB(skb)->age = jiffies;
5751 NAPI_GRO_CB(skb)->last = skb;
5752 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
5753 list_add(&skb->list, gro_head);
5757 grow = skb_gro_offset(skb) - skb_headlen(skb);
5759 gro_pull_from_frag0(skb, grow);
5761 if (napi->gro_hash[hash].count) {
5762 if (!test_bit(hash, &napi->gro_bitmask))
5763 __set_bit(hash, &napi->gro_bitmask);
5764 } else if (test_bit(hash, &napi->gro_bitmask)) {
5765 __clear_bit(hash, &napi->gro_bitmask);
5775 struct packet_offload *gro_find_receive_by_type(__be16 type)
5777 struct list_head *offload_head = &offload_base;
5778 struct packet_offload *ptype;
5780 list_for_each_entry_rcu(ptype, offload_head, list) {
5781 if (ptype->type != type || !ptype->callbacks.gro_receive)
5787 EXPORT_SYMBOL(gro_find_receive_by_type);
5789 struct packet_offload *gro_find_complete_by_type(__be16 type)
5791 struct list_head *offload_head = &offload_base;
5792 struct packet_offload *ptype;
5794 list_for_each_entry_rcu(ptype, offload_head, list) {
5795 if (ptype->type != type || !ptype->callbacks.gro_complete)
5801 EXPORT_SYMBOL(gro_find_complete_by_type);
5803 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */
5804 static void gro_normal_list(struct napi_struct *napi)
5806 if (!napi->rx_count)
5808 netif_receive_skb_list_internal(&napi->rx_list);
5809 INIT_LIST_HEAD(&napi->rx_list);
5813 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded,
5814 * pass the whole batch up to the stack.
5816 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb)
5818 list_add_tail(&skb->list, &napi->rx_list);
5819 if (++napi->rx_count >= gro_normal_batch)
5820 gro_normal_list(napi);
5823 static void napi_skb_free_stolen_head(struct sk_buff *skb)
5827 kmem_cache_free(skbuff_head_cache, skb);
5830 static gro_result_t napi_skb_finish(struct napi_struct *napi,
5831 struct sk_buff *skb,
5836 gro_normal_one(napi, skb);
5843 case GRO_MERGED_FREE:
5844 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5845 napi_skb_free_stolen_head(skb);
5859 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5863 skb_mark_napi_id(skb, napi);
5864 trace_napi_gro_receive_entry(skb);
5866 skb_gro_reset_offset(skb);
5868 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
5869 trace_napi_gro_receive_exit(ret);
5873 EXPORT_SYMBOL(napi_gro_receive);
5875 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
5877 if (unlikely(skb->pfmemalloc)) {
5881 __skb_pull(skb, skb_headlen(skb));
5882 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
5883 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
5884 __vlan_hwaccel_clear_tag(skb);
5885 skb->dev = napi->dev;
5888 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
5889 skb->pkt_type = PACKET_HOST;
5891 skb->encapsulation = 0;
5892 skb_shinfo(skb)->gso_type = 0;
5893 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5899 struct sk_buff *napi_get_frags(struct napi_struct *napi)
5901 struct sk_buff *skb = napi->skb;
5904 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
5907 skb_mark_napi_id(skb, napi);
5912 EXPORT_SYMBOL(napi_get_frags);
5914 static gro_result_t napi_frags_finish(struct napi_struct *napi,
5915 struct sk_buff *skb,
5921 __skb_push(skb, ETH_HLEN);
5922 skb->protocol = eth_type_trans(skb, skb->dev);
5923 if (ret == GRO_NORMAL)
5924 gro_normal_one(napi, skb);
5928 napi_reuse_skb(napi, skb);
5931 case GRO_MERGED_FREE:
5932 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5933 napi_skb_free_stolen_head(skb);
5935 napi_reuse_skb(napi, skb);
5946 /* Upper GRO stack assumes network header starts at gro_offset=0
5947 * Drivers could call both napi_gro_frags() and napi_gro_receive()
5948 * We copy ethernet header into skb->data to have a common layout.
5950 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
5952 struct sk_buff *skb = napi->skb;
5953 const struct ethhdr *eth;
5954 unsigned int hlen = sizeof(*eth);
5958 skb_reset_mac_header(skb);
5959 skb_gro_reset_offset(skb);
5961 if (unlikely(skb_gro_header_hard(skb, hlen))) {
5962 eth = skb_gro_header_slow(skb, hlen, 0);
5963 if (unlikely(!eth)) {
5964 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
5965 __func__, napi->dev->name);
5966 napi_reuse_skb(napi, skb);
5970 eth = (const struct ethhdr *)skb->data;
5971 gro_pull_from_frag0(skb, hlen);
5972 NAPI_GRO_CB(skb)->frag0 += hlen;
5973 NAPI_GRO_CB(skb)->frag0_len -= hlen;
5975 __skb_pull(skb, hlen);
5978 * This works because the only protocols we care about don't require
5980 * We'll fix it up properly in napi_frags_finish()
5982 skb->protocol = eth->h_proto;
5987 gro_result_t napi_gro_frags(struct napi_struct *napi)
5990 struct sk_buff *skb = napi_frags_skb(napi);
5995 trace_napi_gro_frags_entry(skb);
5997 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
5998 trace_napi_gro_frags_exit(ret);
6002 EXPORT_SYMBOL(napi_gro_frags);
6004 /* Compute the checksum from gro_offset and return the folded value
6005 * after adding in any pseudo checksum.
6007 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
6012 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
6014 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
6015 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
6016 /* See comments in __skb_checksum_complete(). */
6018 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
6019 !skb->csum_complete_sw)
6020 netdev_rx_csum_fault(skb->dev, skb);
6023 NAPI_GRO_CB(skb)->csum = wsum;
6024 NAPI_GRO_CB(skb)->csum_valid = 1;
6028 EXPORT_SYMBOL(__skb_gro_checksum_complete);
6030 static void net_rps_send_ipi(struct softnet_data *remsd)
6034 struct softnet_data *next = remsd->rps_ipi_next;
6036 if (cpu_online(remsd->cpu))
6037 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6044 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6045 * Note: called with local irq disabled, but exits with local irq enabled.
6047 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6050 struct softnet_data *remsd = sd->rps_ipi_list;
6053 sd->rps_ipi_list = NULL;
6057 /* Send pending IPI's to kick RPS processing on remote cpus. */
6058 net_rps_send_ipi(remsd);
6064 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6067 return sd->rps_ipi_list != NULL;
6073 static int process_backlog(struct napi_struct *napi, int quota)
6075 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6079 /* Check if we have pending ipi, its better to send them now,
6080 * not waiting net_rx_action() end.
6082 if (sd_has_rps_ipi_waiting(sd)) {
6083 local_irq_disable();
6084 net_rps_action_and_irq_enable(sd);
6087 napi->weight = dev_rx_weight;
6089 struct sk_buff *skb;
6091 while ((skb = __skb_dequeue(&sd->process_queue))) {
6093 __netif_receive_skb(skb);
6095 input_queue_head_incr(sd);
6096 if (++work >= quota)
6101 local_irq_disable();
6103 if (skb_queue_empty(&sd->input_pkt_queue)) {
6105 * Inline a custom version of __napi_complete().
6106 * only current cpu owns and manipulates this napi,
6107 * and NAPI_STATE_SCHED is the only possible flag set
6109 * We can use a plain write instead of clear_bit(),
6110 * and we dont need an smp_mb() memory barrier.
6115 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6116 &sd->process_queue);
6126 * __napi_schedule - schedule for receive
6127 * @n: entry to schedule
6129 * The entry's receive function will be scheduled to run.
6130 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6132 void __napi_schedule(struct napi_struct *n)
6134 unsigned long flags;
6136 local_irq_save(flags);
6137 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6138 local_irq_restore(flags);
6140 EXPORT_SYMBOL(__napi_schedule);
6143 * napi_schedule_prep - check if napi can be scheduled
6146 * Test if NAPI routine is already running, and if not mark
6147 * it as running. This is used as a condition variable
6148 * insure only one NAPI poll instance runs. We also make
6149 * sure there is no pending NAPI disable.
6151 bool napi_schedule_prep(struct napi_struct *n)
6153 unsigned long val, new;
6156 val = READ_ONCE(n->state);
6157 if (unlikely(val & NAPIF_STATE_DISABLE))
6159 new = val | NAPIF_STATE_SCHED;
6161 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6162 * This was suggested by Alexander Duyck, as compiler
6163 * emits better code than :
6164 * if (val & NAPIF_STATE_SCHED)
6165 * new |= NAPIF_STATE_MISSED;
6167 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6169 } while (cmpxchg(&n->state, val, new) != val);
6171 return !(val & NAPIF_STATE_SCHED);
6173 EXPORT_SYMBOL(napi_schedule_prep);
6176 * __napi_schedule_irqoff - schedule for receive
6177 * @n: entry to schedule
6179 * Variant of __napi_schedule() assuming hard irqs are masked
6181 void __napi_schedule_irqoff(struct napi_struct *n)
6183 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6185 EXPORT_SYMBOL(__napi_schedule_irqoff);
6187 bool napi_complete_done(struct napi_struct *n, int work_done)
6189 unsigned long flags, val, new;
6192 * 1) Don't let napi dequeue from the cpu poll list
6193 * just in case its running on a different cpu.
6194 * 2) If we are busy polling, do nothing here, we have
6195 * the guarantee we will be called later.
6197 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6198 NAPIF_STATE_IN_BUSY_POLL)))
6203 if (n->gro_bitmask) {
6204 unsigned long timeout = 0;
6207 timeout = n->dev->gro_flush_timeout;
6209 /* When the NAPI instance uses a timeout and keeps postponing
6210 * it, we need to bound somehow the time packets are kept in
6213 napi_gro_flush(n, !!timeout);
6215 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6216 HRTIMER_MODE_REL_PINNED);
6218 if (unlikely(!list_empty(&n->poll_list))) {
6219 /* If n->poll_list is not empty, we need to mask irqs */
6220 local_irq_save(flags);
6221 list_del_init(&n->poll_list);
6222 local_irq_restore(flags);
6226 val = READ_ONCE(n->state);
6228 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6230 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
6232 /* If STATE_MISSED was set, leave STATE_SCHED set,
6233 * because we will call napi->poll() one more time.
6234 * This C code was suggested by Alexander Duyck to help gcc.
6236 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6238 } while (cmpxchg(&n->state, val, new) != val);
6240 if (unlikely(val & NAPIF_STATE_MISSED)) {
6247 EXPORT_SYMBOL(napi_complete_done);
6249 /* must be called under rcu_read_lock(), as we dont take a reference */
6250 static struct napi_struct *napi_by_id(unsigned int napi_id)
6252 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6253 struct napi_struct *napi;
6255 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6256 if (napi->napi_id == napi_id)
6262 #if defined(CONFIG_NET_RX_BUSY_POLL)
6264 #define BUSY_POLL_BUDGET 8
6266 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
6270 /* Busy polling means there is a high chance device driver hard irq
6271 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6272 * set in napi_schedule_prep().
6273 * Since we are about to call napi->poll() once more, we can safely
6274 * clear NAPI_STATE_MISSED.
6276 * Note: x86 could use a single "lock and ..." instruction
6277 * to perform these two clear_bit()
6279 clear_bit(NAPI_STATE_MISSED, &napi->state);
6280 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6284 /* All we really want here is to re-enable device interrupts.
6285 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6287 rc = napi->poll(napi, BUSY_POLL_BUDGET);
6288 /* We can't gro_normal_list() here, because napi->poll() might have
6289 * rearmed the napi (napi_complete_done()) in which case it could
6290 * already be running on another CPU.
6292 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
6293 netpoll_poll_unlock(have_poll_lock);
6294 if (rc == BUSY_POLL_BUDGET) {
6295 /* As the whole budget was spent, we still own the napi so can
6296 * safely handle the rx_list.
6298 gro_normal_list(napi);
6299 __napi_schedule(napi);
6304 void napi_busy_loop(unsigned int napi_id,
6305 bool (*loop_end)(void *, unsigned long),
6308 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6309 int (*napi_poll)(struct napi_struct *napi, int budget);
6310 void *have_poll_lock = NULL;
6311 struct napi_struct *napi;
6318 napi = napi_by_id(napi_id);
6328 unsigned long val = READ_ONCE(napi->state);
6330 /* If multiple threads are competing for this napi,
6331 * we avoid dirtying napi->state as much as we can.
6333 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6334 NAPIF_STATE_IN_BUSY_POLL))
6336 if (cmpxchg(&napi->state, val,
6337 val | NAPIF_STATE_IN_BUSY_POLL |
6338 NAPIF_STATE_SCHED) != val)
6340 have_poll_lock = netpoll_poll_lock(napi);
6341 napi_poll = napi->poll;
6343 work = napi_poll(napi, BUSY_POLL_BUDGET);
6344 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
6345 gro_normal_list(napi);
6348 __NET_ADD_STATS(dev_net(napi->dev),
6349 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6352 if (!loop_end || loop_end(loop_end_arg, start_time))
6355 if (unlikely(need_resched())) {
6357 busy_poll_stop(napi, have_poll_lock);
6361 if (loop_end(loop_end_arg, start_time))
6368 busy_poll_stop(napi, have_poll_lock);
6373 EXPORT_SYMBOL(napi_busy_loop);
6375 #endif /* CONFIG_NET_RX_BUSY_POLL */
6377 static void napi_hash_add(struct napi_struct *napi)
6379 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
6380 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
6383 spin_lock(&napi_hash_lock);
6385 /* 0..NR_CPUS range is reserved for sender_cpu use */
6387 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6388 napi_gen_id = MIN_NAPI_ID;
6389 } while (napi_by_id(napi_gen_id));
6390 napi->napi_id = napi_gen_id;
6392 hlist_add_head_rcu(&napi->napi_hash_node,
6393 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6395 spin_unlock(&napi_hash_lock);
6398 /* Warning : caller is responsible to make sure rcu grace period
6399 * is respected before freeing memory containing @napi
6401 bool napi_hash_del(struct napi_struct *napi)
6403 bool rcu_sync_needed = false;
6405 spin_lock(&napi_hash_lock);
6407 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
6408 rcu_sync_needed = true;
6409 hlist_del_rcu(&napi->napi_hash_node);
6411 spin_unlock(&napi_hash_lock);
6412 return rcu_sync_needed;
6414 EXPORT_SYMBOL_GPL(napi_hash_del);
6416 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6418 struct napi_struct *napi;
6420 napi = container_of(timer, struct napi_struct, timer);
6422 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6423 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6425 if (napi->gro_bitmask && !napi_disable_pending(napi) &&
6426 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
6427 __napi_schedule_irqoff(napi);
6429 return HRTIMER_NORESTART;
6432 static void init_gro_hash(struct napi_struct *napi)
6436 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6437 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6438 napi->gro_hash[i].count = 0;
6440 napi->gro_bitmask = 0;
6443 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6444 int (*poll)(struct napi_struct *, int), int weight)
6446 INIT_LIST_HEAD(&napi->poll_list);
6447 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6448 napi->timer.function = napi_watchdog;
6449 init_gro_hash(napi);
6451 INIT_LIST_HEAD(&napi->rx_list);
6454 if (weight > NAPI_POLL_WEIGHT)
6455 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6457 napi->weight = weight;
6458 list_add(&napi->dev_list, &dev->napi_list);
6460 #ifdef CONFIG_NETPOLL
6461 napi->poll_owner = -1;
6463 set_bit(NAPI_STATE_SCHED, &napi->state);
6464 napi_hash_add(napi);
6466 EXPORT_SYMBOL(netif_napi_add);
6468 void napi_disable(struct napi_struct *n)
6471 set_bit(NAPI_STATE_DISABLE, &n->state);
6473 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6475 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6478 hrtimer_cancel(&n->timer);
6480 clear_bit(NAPI_STATE_DISABLE, &n->state);
6482 EXPORT_SYMBOL(napi_disable);
6484 static void flush_gro_hash(struct napi_struct *napi)
6488 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6489 struct sk_buff *skb, *n;
6491 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6493 napi->gro_hash[i].count = 0;
6497 /* Must be called in process context */
6498 void netif_napi_del(struct napi_struct *napi)
6501 if (napi_hash_del(napi))
6503 list_del_init(&napi->dev_list);
6504 napi_free_frags(napi);
6506 flush_gro_hash(napi);
6507 napi->gro_bitmask = 0;
6509 EXPORT_SYMBOL(netif_napi_del);
6511 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6516 list_del_init(&n->poll_list);
6518 have = netpoll_poll_lock(n);
6522 /* This NAPI_STATE_SCHED test is for avoiding a race
6523 * with netpoll's poll_napi(). Only the entity which
6524 * obtains the lock and sees NAPI_STATE_SCHED set will
6525 * actually make the ->poll() call. Therefore we avoid
6526 * accidentally calling ->poll() when NAPI is not scheduled.
6529 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6530 work = n->poll(n, weight);
6531 trace_napi_poll(n, work, weight);
6534 WARN_ON_ONCE(work > weight);
6536 if (likely(work < weight))
6539 /* Drivers must not modify the NAPI state if they
6540 * consume the entire weight. In such cases this code
6541 * still "owns" the NAPI instance and therefore can
6542 * move the instance around on the list at-will.
6544 if (unlikely(napi_disable_pending(n))) {
6551 if (n->gro_bitmask) {
6552 /* flush too old packets
6553 * If HZ < 1000, flush all packets.
6555 napi_gro_flush(n, HZ >= 1000);
6558 /* Some drivers may have called napi_schedule
6559 * prior to exhausting their budget.
6561 if (unlikely(!list_empty(&n->poll_list))) {
6562 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6563 n->dev ? n->dev->name : "backlog");
6567 list_add_tail(&n->poll_list, repoll);
6570 netpoll_poll_unlock(have);
6575 static __latent_entropy void net_rx_action(struct softirq_action *h)
6577 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6578 unsigned long time_limit = jiffies +
6579 usecs_to_jiffies(netdev_budget_usecs);
6580 int budget = netdev_budget;
6584 local_irq_disable();
6585 list_splice_init(&sd->poll_list, &list);
6589 struct napi_struct *n;
6591 if (list_empty(&list)) {
6592 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6597 n = list_first_entry(&list, struct napi_struct, poll_list);
6598 budget -= napi_poll(n, &repoll);
6600 /* If softirq window is exhausted then punt.
6601 * Allow this to run for 2 jiffies since which will allow
6602 * an average latency of 1.5/HZ.
6604 if (unlikely(budget <= 0 ||
6605 time_after_eq(jiffies, time_limit))) {
6611 local_irq_disable();
6613 list_splice_tail_init(&sd->poll_list, &list);
6614 list_splice_tail(&repoll, &list);
6615 list_splice(&list, &sd->poll_list);
6616 if (!list_empty(&sd->poll_list))
6617 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6619 net_rps_action_and_irq_enable(sd);
6621 __kfree_skb_flush();
6624 struct netdev_adjacent {
6625 struct net_device *dev;
6627 /* upper master flag, there can only be one master device per list */
6630 /* lookup ignore flag */
6633 /* counter for the number of times this device was added to us */
6636 /* private field for the users */
6639 struct list_head list;
6640 struct rcu_head rcu;
6643 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6644 struct list_head *adj_list)
6646 struct netdev_adjacent *adj;
6648 list_for_each_entry(adj, adj_list, list) {
6649 if (adj->dev == adj_dev)
6655 static int ____netdev_has_upper_dev(struct net_device *upper_dev, void *data)
6657 struct net_device *dev = data;
6659 return upper_dev == dev;
6663 * netdev_has_upper_dev - Check if device is linked to an upper device
6665 * @upper_dev: upper device to check
6667 * Find out if a device is linked to specified upper device and return true
6668 * in case it is. Note that this checks only immediate upper device,
6669 * not through a complete stack of devices. The caller must hold the RTNL lock.
6671 bool netdev_has_upper_dev(struct net_device *dev,
6672 struct net_device *upper_dev)
6676 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6679 EXPORT_SYMBOL(netdev_has_upper_dev);
6682 * netdev_has_upper_dev_all - Check if device is linked to an upper device
6684 * @upper_dev: upper device to check
6686 * Find out if a device is linked to specified upper device and return true
6687 * in case it is. Note that this checks the entire upper device chain.
6688 * The caller must hold rcu lock.
6691 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6692 struct net_device *upper_dev)
6694 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6697 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6700 * netdev_has_any_upper_dev - Check if device is linked to some device
6703 * Find out if a device is linked to an upper device and return true in case
6704 * it is. The caller must hold the RTNL lock.
6706 bool netdev_has_any_upper_dev(struct net_device *dev)
6710 return !list_empty(&dev->adj_list.upper);
6712 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6715 * netdev_master_upper_dev_get - Get master upper device
6718 * Find a master upper device and return pointer to it or NULL in case
6719 * it's not there. The caller must hold the RTNL lock.
6721 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6723 struct netdev_adjacent *upper;
6727 if (list_empty(&dev->adj_list.upper))
6730 upper = list_first_entry(&dev->adj_list.upper,
6731 struct netdev_adjacent, list);
6732 if (likely(upper->master))
6736 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6738 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6740 struct netdev_adjacent *upper;
6744 if (list_empty(&dev->adj_list.upper))
6747 upper = list_first_entry(&dev->adj_list.upper,
6748 struct netdev_adjacent, list);
6749 if (likely(upper->master) && !upper->ignore)
6755 * netdev_has_any_lower_dev - Check if device is linked to some device
6758 * Find out if a device is linked to a lower device and return true in case
6759 * it is. The caller must hold the RTNL lock.
6761 static bool netdev_has_any_lower_dev(struct net_device *dev)
6765 return !list_empty(&dev->adj_list.lower);
6768 void *netdev_adjacent_get_private(struct list_head *adj_list)
6770 struct netdev_adjacent *adj;
6772 adj = list_entry(adj_list, struct netdev_adjacent, list);
6774 return adj->private;
6776 EXPORT_SYMBOL(netdev_adjacent_get_private);
6779 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6781 * @iter: list_head ** of the current position
6783 * Gets the next device from the dev's upper list, starting from iter
6784 * position. The caller must hold RCU read lock.
6786 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6787 struct list_head **iter)
6789 struct netdev_adjacent *upper;
6791 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6793 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6795 if (&upper->list == &dev->adj_list.upper)
6798 *iter = &upper->list;
6802 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6804 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6805 struct list_head **iter,
6808 struct netdev_adjacent *upper;
6810 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6812 if (&upper->list == &dev->adj_list.upper)
6815 *iter = &upper->list;
6816 *ignore = upper->ignore;
6821 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6822 struct list_head **iter)
6824 struct netdev_adjacent *upper;
6826 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6828 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6830 if (&upper->list == &dev->adj_list.upper)
6833 *iter = &upper->list;
6838 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6839 int (*fn)(struct net_device *dev,
6843 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6844 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6849 iter = &dev->adj_list.upper;
6853 ret = fn(now, data);
6860 udev = __netdev_next_upper_dev(now, &iter, &ignore);
6867 niter = &udev->adj_list.upper;
6868 dev_stack[cur] = now;
6869 iter_stack[cur++] = iter;
6876 next = dev_stack[--cur];
6877 niter = iter_stack[cur];
6887 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6888 int (*fn)(struct net_device *dev,
6892 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6893 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6897 iter = &dev->adj_list.upper;
6901 ret = fn(now, data);
6908 udev = netdev_next_upper_dev_rcu(now, &iter);
6913 niter = &udev->adj_list.upper;
6914 dev_stack[cur] = now;
6915 iter_stack[cur++] = iter;
6922 next = dev_stack[--cur];
6923 niter = iter_stack[cur];
6932 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6934 static bool __netdev_has_upper_dev(struct net_device *dev,
6935 struct net_device *upper_dev)
6939 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
6944 * netdev_lower_get_next_private - Get the next ->private from the
6945 * lower neighbour list
6947 * @iter: list_head ** of the current position
6949 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6950 * list, starting from iter position. The caller must hold either hold the
6951 * RTNL lock or its own locking that guarantees that the neighbour lower
6952 * list will remain unchanged.
6954 void *netdev_lower_get_next_private(struct net_device *dev,
6955 struct list_head **iter)
6957 struct netdev_adjacent *lower;
6959 lower = list_entry(*iter, struct netdev_adjacent, list);
6961 if (&lower->list == &dev->adj_list.lower)
6964 *iter = lower->list.next;
6966 return lower->private;
6968 EXPORT_SYMBOL(netdev_lower_get_next_private);
6971 * netdev_lower_get_next_private_rcu - Get the next ->private from the
6972 * lower neighbour list, RCU
6975 * @iter: list_head ** of the current position
6977 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6978 * list, starting from iter position. The caller must hold RCU read lock.
6980 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
6981 struct list_head **iter)
6983 struct netdev_adjacent *lower;
6985 WARN_ON_ONCE(!rcu_read_lock_held());
6987 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6989 if (&lower->list == &dev->adj_list.lower)
6992 *iter = &lower->list;
6994 return lower->private;
6996 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
6999 * netdev_lower_get_next - Get the next device from the lower neighbour
7002 * @iter: list_head ** of the current position
7004 * Gets the next netdev_adjacent from the dev's lower neighbour
7005 * list, starting from iter position. The caller must hold RTNL lock or
7006 * its own locking that guarantees that the neighbour lower
7007 * list will remain unchanged.
7009 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7011 struct netdev_adjacent *lower;
7013 lower = list_entry(*iter, struct netdev_adjacent, list);
7015 if (&lower->list == &dev->adj_list.lower)
7018 *iter = lower->list.next;
7022 EXPORT_SYMBOL(netdev_lower_get_next);
7024 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7025 struct list_head **iter)
7027 struct netdev_adjacent *lower;
7029 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7031 if (&lower->list == &dev->adj_list.lower)
7034 *iter = &lower->list;
7039 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7040 struct list_head **iter,
7043 struct netdev_adjacent *lower;
7045 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7047 if (&lower->list == &dev->adj_list.lower)
7050 *iter = &lower->list;
7051 *ignore = lower->ignore;
7056 int netdev_walk_all_lower_dev(struct net_device *dev,
7057 int (*fn)(struct net_device *dev,
7061 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7062 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7066 iter = &dev->adj_list.lower;
7070 ret = fn(now, data);
7077 ldev = netdev_next_lower_dev(now, &iter);
7082 niter = &ldev->adj_list.lower;
7083 dev_stack[cur] = now;
7084 iter_stack[cur++] = iter;
7091 next = dev_stack[--cur];
7092 niter = iter_stack[cur];
7101 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7103 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7104 int (*fn)(struct net_device *dev,
7108 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7109 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7114 iter = &dev->adj_list.lower;
7118 ret = fn(now, data);
7125 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7132 niter = &ldev->adj_list.lower;
7133 dev_stack[cur] = now;
7134 iter_stack[cur++] = iter;
7141 next = dev_stack[--cur];
7142 niter = iter_stack[cur];
7152 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7153 struct list_head **iter)
7155 struct netdev_adjacent *lower;
7157 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7158 if (&lower->list == &dev->adj_list.lower)
7161 *iter = &lower->list;
7166 static u8 __netdev_upper_depth(struct net_device *dev)
7168 struct net_device *udev;
7169 struct list_head *iter;
7173 for (iter = &dev->adj_list.upper,
7174 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7176 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7179 if (max_depth < udev->upper_level)
7180 max_depth = udev->upper_level;
7186 static u8 __netdev_lower_depth(struct net_device *dev)
7188 struct net_device *ldev;
7189 struct list_head *iter;
7193 for (iter = &dev->adj_list.lower,
7194 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7196 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7199 if (max_depth < ldev->lower_level)
7200 max_depth = ldev->lower_level;
7206 static int __netdev_update_upper_level(struct net_device *dev, void *data)
7208 dev->upper_level = __netdev_upper_depth(dev) + 1;
7212 static int __netdev_update_lower_level(struct net_device *dev, void *data)
7214 dev->lower_level = __netdev_lower_depth(dev) + 1;
7218 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7219 int (*fn)(struct net_device *dev,
7223 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7224 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7228 iter = &dev->adj_list.lower;
7232 ret = fn(now, data);
7239 ldev = netdev_next_lower_dev_rcu(now, &iter);
7244 niter = &ldev->adj_list.lower;
7245 dev_stack[cur] = now;
7246 iter_stack[cur++] = iter;
7253 next = dev_stack[--cur];
7254 niter = iter_stack[cur];
7263 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7266 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7267 * lower neighbour list, RCU
7271 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7272 * list. The caller must hold RCU read lock.
7274 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7276 struct netdev_adjacent *lower;
7278 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7279 struct netdev_adjacent, list);
7281 return lower->private;
7284 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7287 * netdev_master_upper_dev_get_rcu - Get master upper device
7290 * Find a master upper device and return pointer to it or NULL in case
7291 * it's not there. The caller must hold the RCU read lock.
7293 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7295 struct netdev_adjacent *upper;
7297 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7298 struct netdev_adjacent, list);
7299 if (upper && likely(upper->master))
7303 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7305 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7306 struct net_device *adj_dev,
7307 struct list_head *dev_list)
7309 char linkname[IFNAMSIZ+7];
7311 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7312 "upper_%s" : "lower_%s", adj_dev->name);
7313 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7316 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7318 struct list_head *dev_list)
7320 char linkname[IFNAMSIZ+7];
7322 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7323 "upper_%s" : "lower_%s", name);
7324 sysfs_remove_link(&(dev->dev.kobj), linkname);
7327 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7328 struct net_device *adj_dev,
7329 struct list_head *dev_list)
7331 return (dev_list == &dev->adj_list.upper ||
7332 dev_list == &dev->adj_list.lower) &&
7333 net_eq(dev_net(dev), dev_net(adj_dev));
7336 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7337 struct net_device *adj_dev,
7338 struct list_head *dev_list,
7339 void *private, bool master)
7341 struct netdev_adjacent *adj;
7344 adj = __netdev_find_adj(adj_dev, dev_list);
7348 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7349 dev->name, adj_dev->name, adj->ref_nr);
7354 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7359 adj->master = master;
7361 adj->private = private;
7362 adj->ignore = false;
7365 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7366 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7368 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7369 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7374 /* Ensure that master link is always the first item in list. */
7376 ret = sysfs_create_link(&(dev->dev.kobj),
7377 &(adj_dev->dev.kobj), "master");
7379 goto remove_symlinks;
7381 list_add_rcu(&adj->list, dev_list);
7383 list_add_tail_rcu(&adj->list, dev_list);
7389 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7390 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7398 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7399 struct net_device *adj_dev,
7401 struct list_head *dev_list)
7403 struct netdev_adjacent *adj;
7405 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7406 dev->name, adj_dev->name, ref_nr);
7408 adj = __netdev_find_adj(adj_dev, dev_list);
7411 pr_err("Adjacency does not exist for device %s from %s\n",
7412 dev->name, adj_dev->name);
7417 if (adj->ref_nr > ref_nr) {
7418 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7419 dev->name, adj_dev->name, ref_nr,
7420 adj->ref_nr - ref_nr);
7421 adj->ref_nr -= ref_nr;
7426 sysfs_remove_link(&(dev->dev.kobj), "master");
7428 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7429 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7431 list_del_rcu(&adj->list);
7432 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7433 adj_dev->name, dev->name, adj_dev->name);
7435 kfree_rcu(adj, rcu);
7438 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7439 struct net_device *upper_dev,
7440 struct list_head *up_list,
7441 struct list_head *down_list,
7442 void *private, bool master)
7446 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7451 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7454 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7461 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7462 struct net_device *upper_dev,
7464 struct list_head *up_list,
7465 struct list_head *down_list)
7467 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7468 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7471 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7472 struct net_device *upper_dev,
7473 void *private, bool master)
7475 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7476 &dev->adj_list.upper,
7477 &upper_dev->adj_list.lower,
7481 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7482 struct net_device *upper_dev)
7484 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7485 &dev->adj_list.upper,
7486 &upper_dev->adj_list.lower);
7489 static int __netdev_upper_dev_link(struct net_device *dev,
7490 struct net_device *upper_dev, bool master,
7491 void *upper_priv, void *upper_info,
7492 struct netlink_ext_ack *extack)
7494 struct netdev_notifier_changeupper_info changeupper_info = {
7499 .upper_dev = upper_dev,
7502 .upper_info = upper_info,
7504 struct net_device *master_dev;
7509 if (dev == upper_dev)
7512 /* To prevent loops, check if dev is not upper device to upper_dev. */
7513 if (__netdev_has_upper_dev(upper_dev, dev))
7516 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7520 if (__netdev_has_upper_dev(dev, upper_dev))
7523 master_dev = __netdev_master_upper_dev_get(dev);
7525 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7528 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7529 &changeupper_info.info);
7530 ret = notifier_to_errno(ret);
7534 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7539 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7540 &changeupper_info.info);
7541 ret = notifier_to_errno(ret);
7545 __netdev_update_upper_level(dev, NULL);
7546 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7548 __netdev_update_lower_level(upper_dev, NULL);
7549 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7555 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7561 * netdev_upper_dev_link - Add a link to the upper device
7563 * @upper_dev: new upper device
7564 * @extack: netlink extended ack
7566 * Adds a link to device which is upper to this one. The caller must hold
7567 * the RTNL lock. On a failure a negative errno code is returned.
7568 * On success the reference counts are adjusted and the function
7571 int netdev_upper_dev_link(struct net_device *dev,
7572 struct net_device *upper_dev,
7573 struct netlink_ext_ack *extack)
7575 return __netdev_upper_dev_link(dev, upper_dev, false,
7576 NULL, NULL, extack);
7578 EXPORT_SYMBOL(netdev_upper_dev_link);
7581 * netdev_master_upper_dev_link - Add a master link to the upper device
7583 * @upper_dev: new upper device
7584 * @upper_priv: upper device private
7585 * @upper_info: upper info to be passed down via notifier
7586 * @extack: netlink extended ack
7588 * Adds a link to device which is upper to this one. In this case, only
7589 * one master upper device can be linked, although other non-master devices
7590 * might be linked as well. The caller must hold the RTNL lock.
7591 * On a failure a negative errno code is returned. On success the reference
7592 * counts are adjusted and the function returns zero.
7594 int netdev_master_upper_dev_link(struct net_device *dev,
7595 struct net_device *upper_dev,
7596 void *upper_priv, void *upper_info,
7597 struct netlink_ext_ack *extack)
7599 return __netdev_upper_dev_link(dev, upper_dev, true,
7600 upper_priv, upper_info, extack);
7602 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7605 * netdev_upper_dev_unlink - Removes a link to upper device
7607 * @upper_dev: new upper device
7609 * Removes a link to device which is upper to this one. The caller must hold
7612 void netdev_upper_dev_unlink(struct net_device *dev,
7613 struct net_device *upper_dev)
7615 struct netdev_notifier_changeupper_info changeupper_info = {
7619 .upper_dev = upper_dev,
7625 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7627 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7628 &changeupper_info.info);
7630 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7632 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7633 &changeupper_info.info);
7635 __netdev_update_upper_level(dev, NULL);
7636 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7638 __netdev_update_lower_level(upper_dev, NULL);
7639 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7642 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7644 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7645 struct net_device *lower_dev,
7648 struct netdev_adjacent *adj;
7650 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7654 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7659 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7660 struct net_device *lower_dev)
7662 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7665 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7666 struct net_device *lower_dev)
7668 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7671 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7672 struct net_device *new_dev,
7673 struct net_device *dev,
7674 struct netlink_ext_ack *extack)
7681 if (old_dev && new_dev != old_dev)
7682 netdev_adjacent_dev_disable(dev, old_dev);
7684 err = netdev_upper_dev_link(new_dev, dev, extack);
7686 if (old_dev && new_dev != old_dev)
7687 netdev_adjacent_dev_enable(dev, old_dev);
7693 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7695 void netdev_adjacent_change_commit(struct net_device *old_dev,
7696 struct net_device *new_dev,
7697 struct net_device *dev)
7699 if (!new_dev || !old_dev)
7702 if (new_dev == old_dev)
7705 netdev_adjacent_dev_enable(dev, old_dev);
7706 netdev_upper_dev_unlink(old_dev, dev);
7708 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7710 void netdev_adjacent_change_abort(struct net_device *old_dev,
7711 struct net_device *new_dev,
7712 struct net_device *dev)
7717 if (old_dev && new_dev != old_dev)
7718 netdev_adjacent_dev_enable(dev, old_dev);
7720 netdev_upper_dev_unlink(new_dev, dev);
7722 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7725 * netdev_bonding_info_change - Dispatch event about slave change
7727 * @bonding_info: info to dispatch
7729 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7730 * The caller must hold the RTNL lock.
7732 void netdev_bonding_info_change(struct net_device *dev,
7733 struct netdev_bonding_info *bonding_info)
7735 struct netdev_notifier_bonding_info info = {
7739 memcpy(&info.bonding_info, bonding_info,
7740 sizeof(struct netdev_bonding_info));
7741 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7744 EXPORT_SYMBOL(netdev_bonding_info_change);
7746 static void netdev_adjacent_add_links(struct net_device *dev)
7748 struct netdev_adjacent *iter;
7750 struct net *net = dev_net(dev);
7752 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7753 if (!net_eq(net, dev_net(iter->dev)))
7755 netdev_adjacent_sysfs_add(iter->dev, dev,
7756 &iter->dev->adj_list.lower);
7757 netdev_adjacent_sysfs_add(dev, iter->dev,
7758 &dev->adj_list.upper);
7761 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7762 if (!net_eq(net, dev_net(iter->dev)))
7764 netdev_adjacent_sysfs_add(iter->dev, dev,
7765 &iter->dev->adj_list.upper);
7766 netdev_adjacent_sysfs_add(dev, iter->dev,
7767 &dev->adj_list.lower);
7771 static void netdev_adjacent_del_links(struct net_device *dev)
7773 struct netdev_adjacent *iter;
7775 struct net *net = dev_net(dev);
7777 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7778 if (!net_eq(net, dev_net(iter->dev)))
7780 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7781 &iter->dev->adj_list.lower);
7782 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7783 &dev->adj_list.upper);
7786 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7787 if (!net_eq(net, dev_net(iter->dev)))
7789 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7790 &iter->dev->adj_list.upper);
7791 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7792 &dev->adj_list.lower);
7796 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
7798 struct netdev_adjacent *iter;
7800 struct net *net = dev_net(dev);
7802 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7803 if (!net_eq(net, dev_net(iter->dev)))
7805 netdev_adjacent_sysfs_del(iter->dev, oldname,
7806 &iter->dev->adj_list.lower);
7807 netdev_adjacent_sysfs_add(iter->dev, dev,
7808 &iter->dev->adj_list.lower);
7811 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7812 if (!net_eq(net, dev_net(iter->dev)))
7814 netdev_adjacent_sysfs_del(iter->dev, oldname,
7815 &iter->dev->adj_list.upper);
7816 netdev_adjacent_sysfs_add(iter->dev, dev,
7817 &iter->dev->adj_list.upper);
7821 void *netdev_lower_dev_get_private(struct net_device *dev,
7822 struct net_device *lower_dev)
7824 struct netdev_adjacent *lower;
7828 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
7832 return lower->private;
7834 EXPORT_SYMBOL(netdev_lower_dev_get_private);
7838 * netdev_lower_change - Dispatch event about lower device state change
7839 * @lower_dev: device
7840 * @lower_state_info: state to dispatch
7842 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
7843 * The caller must hold the RTNL lock.
7845 void netdev_lower_state_changed(struct net_device *lower_dev,
7846 void *lower_state_info)
7848 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
7849 .info.dev = lower_dev,
7853 changelowerstate_info.lower_state_info = lower_state_info;
7854 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
7855 &changelowerstate_info.info);
7857 EXPORT_SYMBOL(netdev_lower_state_changed);
7859 static void dev_change_rx_flags(struct net_device *dev, int flags)
7861 const struct net_device_ops *ops = dev->netdev_ops;
7863 if (ops->ndo_change_rx_flags)
7864 ops->ndo_change_rx_flags(dev, flags);
7867 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
7869 unsigned int old_flags = dev->flags;
7875 dev->flags |= IFF_PROMISC;
7876 dev->promiscuity += inc;
7877 if (dev->promiscuity == 0) {
7880 * If inc causes overflow, untouch promisc and return error.
7883 dev->flags &= ~IFF_PROMISC;
7885 dev->promiscuity -= inc;
7886 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
7891 if (dev->flags != old_flags) {
7892 pr_info("device %s %s promiscuous mode\n",
7894 dev->flags & IFF_PROMISC ? "entered" : "left");
7895 if (audit_enabled) {
7896 current_uid_gid(&uid, &gid);
7897 audit_log(audit_context(), GFP_ATOMIC,
7898 AUDIT_ANOM_PROMISCUOUS,
7899 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
7900 dev->name, (dev->flags & IFF_PROMISC),
7901 (old_flags & IFF_PROMISC),
7902 from_kuid(&init_user_ns, audit_get_loginuid(current)),
7903 from_kuid(&init_user_ns, uid),
7904 from_kgid(&init_user_ns, gid),
7905 audit_get_sessionid(current));
7908 dev_change_rx_flags(dev, IFF_PROMISC);
7911 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
7916 * dev_set_promiscuity - update promiscuity count on a device
7920 * Add or remove promiscuity from a device. While the count in the device
7921 * remains above zero the interface remains promiscuous. Once it hits zero
7922 * the device reverts back to normal filtering operation. A negative inc
7923 * value is used to drop promiscuity on the device.
7924 * Return 0 if successful or a negative errno code on error.
7926 int dev_set_promiscuity(struct net_device *dev, int inc)
7928 unsigned int old_flags = dev->flags;
7931 err = __dev_set_promiscuity(dev, inc, true);
7934 if (dev->flags != old_flags)
7935 dev_set_rx_mode(dev);
7938 EXPORT_SYMBOL(dev_set_promiscuity);
7940 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
7942 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
7946 dev->flags |= IFF_ALLMULTI;
7947 dev->allmulti += inc;
7948 if (dev->allmulti == 0) {
7951 * If inc causes overflow, untouch allmulti and return error.
7954 dev->flags &= ~IFF_ALLMULTI;
7956 dev->allmulti -= inc;
7957 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
7962 if (dev->flags ^ old_flags) {
7963 dev_change_rx_flags(dev, IFF_ALLMULTI);
7964 dev_set_rx_mode(dev);
7966 __dev_notify_flags(dev, old_flags,
7967 dev->gflags ^ old_gflags);
7973 * dev_set_allmulti - update allmulti count on a device
7977 * Add or remove reception of all multicast frames to a device. While the
7978 * count in the device remains above zero the interface remains listening
7979 * to all interfaces. Once it hits zero the device reverts back to normal
7980 * filtering operation. A negative @inc value is used to drop the counter
7981 * when releasing a resource needing all multicasts.
7982 * Return 0 if successful or a negative errno code on error.
7985 int dev_set_allmulti(struct net_device *dev, int inc)
7987 return __dev_set_allmulti(dev, inc, true);
7989 EXPORT_SYMBOL(dev_set_allmulti);
7992 * Upload unicast and multicast address lists to device and
7993 * configure RX filtering. When the device doesn't support unicast
7994 * filtering it is put in promiscuous mode while unicast addresses
7997 void __dev_set_rx_mode(struct net_device *dev)
7999 const struct net_device_ops *ops = dev->netdev_ops;
8001 /* dev_open will call this function so the list will stay sane. */
8002 if (!(dev->flags&IFF_UP))
8005 if (!netif_device_present(dev))
8008 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8009 /* Unicast addresses changes may only happen under the rtnl,
8010 * therefore calling __dev_set_promiscuity here is safe.
8012 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8013 __dev_set_promiscuity(dev, 1, false);
8014 dev->uc_promisc = true;
8015 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8016 __dev_set_promiscuity(dev, -1, false);
8017 dev->uc_promisc = false;
8021 if (ops->ndo_set_rx_mode)
8022 ops->ndo_set_rx_mode(dev);
8025 void dev_set_rx_mode(struct net_device *dev)
8027 netif_addr_lock_bh(dev);
8028 __dev_set_rx_mode(dev);
8029 netif_addr_unlock_bh(dev);
8033 * dev_get_flags - get flags reported to userspace
8036 * Get the combination of flag bits exported through APIs to userspace.
8038 unsigned int dev_get_flags(const struct net_device *dev)
8042 flags = (dev->flags & ~(IFF_PROMISC |
8047 (dev->gflags & (IFF_PROMISC |
8050 if (netif_running(dev)) {
8051 if (netif_oper_up(dev))
8052 flags |= IFF_RUNNING;
8053 if (netif_carrier_ok(dev))
8054 flags |= IFF_LOWER_UP;
8055 if (netif_dormant(dev))
8056 flags |= IFF_DORMANT;
8061 EXPORT_SYMBOL(dev_get_flags);
8063 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8064 struct netlink_ext_ack *extack)
8066 unsigned int old_flags = dev->flags;
8072 * Set the flags on our device.
8075 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8076 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8078 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8082 * Load in the correct multicast list now the flags have changed.
8085 if ((old_flags ^ flags) & IFF_MULTICAST)
8086 dev_change_rx_flags(dev, IFF_MULTICAST);
8088 dev_set_rx_mode(dev);
8091 * Have we downed the interface. We handle IFF_UP ourselves
8092 * according to user attempts to set it, rather than blindly
8097 if ((old_flags ^ flags) & IFF_UP) {
8098 if (old_flags & IFF_UP)
8101 ret = __dev_open(dev, extack);
8104 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8105 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8106 unsigned int old_flags = dev->flags;
8108 dev->gflags ^= IFF_PROMISC;
8110 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8111 if (dev->flags != old_flags)
8112 dev_set_rx_mode(dev);
8115 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8116 * is important. Some (broken) drivers set IFF_PROMISC, when
8117 * IFF_ALLMULTI is requested not asking us and not reporting.
8119 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8120 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8122 dev->gflags ^= IFF_ALLMULTI;
8123 __dev_set_allmulti(dev, inc, false);
8129 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8130 unsigned int gchanges)
8132 unsigned int changes = dev->flags ^ old_flags;
8135 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8137 if (changes & IFF_UP) {
8138 if (dev->flags & IFF_UP)
8139 call_netdevice_notifiers(NETDEV_UP, dev);
8141 call_netdevice_notifiers(NETDEV_DOWN, dev);
8144 if (dev->flags & IFF_UP &&
8145 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8146 struct netdev_notifier_change_info change_info = {
8150 .flags_changed = changes,
8153 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8158 * dev_change_flags - change device settings
8160 * @flags: device state flags
8161 * @extack: netlink extended ack
8163 * Change settings on device based state flags. The flags are
8164 * in the userspace exported format.
8166 int dev_change_flags(struct net_device *dev, unsigned int flags,
8167 struct netlink_ext_ack *extack)
8170 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8172 ret = __dev_change_flags(dev, flags, extack);
8176 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8177 __dev_notify_flags(dev, old_flags, changes);
8180 EXPORT_SYMBOL(dev_change_flags);
8182 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8184 const struct net_device_ops *ops = dev->netdev_ops;
8186 if (ops->ndo_change_mtu)
8187 return ops->ndo_change_mtu(dev, new_mtu);
8189 /* Pairs with all the lockless reads of dev->mtu in the stack */
8190 WRITE_ONCE(dev->mtu, new_mtu);
8193 EXPORT_SYMBOL(__dev_set_mtu);
8196 * dev_set_mtu_ext - Change maximum transfer unit
8198 * @new_mtu: new transfer unit
8199 * @extack: netlink extended ack
8201 * Change the maximum transfer size of the network device.
8203 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8204 struct netlink_ext_ack *extack)
8208 if (new_mtu == dev->mtu)
8211 /* MTU must be positive, and in range */
8212 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8213 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8217 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8218 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8222 if (!netif_device_present(dev))
8225 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8226 err = notifier_to_errno(err);
8230 orig_mtu = dev->mtu;
8231 err = __dev_set_mtu(dev, new_mtu);
8234 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8236 err = notifier_to_errno(err);
8238 /* setting mtu back and notifying everyone again,
8239 * so that they have a chance to revert changes.
8241 __dev_set_mtu(dev, orig_mtu);
8242 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8249 int dev_set_mtu(struct net_device *dev, int new_mtu)
8251 struct netlink_ext_ack extack;
8254 memset(&extack, 0, sizeof(extack));
8255 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8256 if (err && extack._msg)
8257 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8260 EXPORT_SYMBOL(dev_set_mtu);
8263 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8265 * @new_len: new tx queue length
8267 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8269 unsigned int orig_len = dev->tx_queue_len;
8272 if (new_len != (unsigned int)new_len)
8275 if (new_len != orig_len) {
8276 dev->tx_queue_len = new_len;
8277 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8278 res = notifier_to_errno(res);
8281 res = dev_qdisc_change_tx_queue_len(dev);
8289 netdev_err(dev, "refused to change device tx_queue_len\n");
8290 dev->tx_queue_len = orig_len;
8295 * dev_set_group - Change group this device belongs to
8297 * @new_group: group this device should belong to
8299 void dev_set_group(struct net_device *dev, int new_group)
8301 dev->group = new_group;
8303 EXPORT_SYMBOL(dev_set_group);
8306 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8308 * @addr: new address
8309 * @extack: netlink extended ack
8311 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8312 struct netlink_ext_ack *extack)
8314 struct netdev_notifier_pre_changeaddr_info info = {
8316 .info.extack = extack,
8321 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8322 return notifier_to_errno(rc);
8324 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8327 * dev_set_mac_address - Change Media Access Control Address
8330 * @extack: netlink extended ack
8332 * Change the hardware (MAC) address of the device
8334 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8335 struct netlink_ext_ack *extack)
8337 const struct net_device_ops *ops = dev->netdev_ops;
8340 if (!ops->ndo_set_mac_address)
8342 if (sa->sa_family != dev->type)
8344 if (!netif_device_present(dev))
8346 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8349 err = ops->ndo_set_mac_address(dev, sa);
8352 dev->addr_assign_type = NET_ADDR_SET;
8353 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8354 add_device_randomness(dev->dev_addr, dev->addr_len);
8357 EXPORT_SYMBOL(dev_set_mac_address);
8360 * dev_change_carrier - Change device carrier
8362 * @new_carrier: new value
8364 * Change device carrier
8366 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8368 const struct net_device_ops *ops = dev->netdev_ops;
8370 if (!ops->ndo_change_carrier)
8372 if (!netif_device_present(dev))
8374 return ops->ndo_change_carrier(dev, new_carrier);
8376 EXPORT_SYMBOL(dev_change_carrier);
8379 * dev_get_phys_port_id - Get device physical port ID
8383 * Get device physical port ID
8385 int dev_get_phys_port_id(struct net_device *dev,
8386 struct netdev_phys_item_id *ppid)
8388 const struct net_device_ops *ops = dev->netdev_ops;
8390 if (!ops->ndo_get_phys_port_id)
8392 return ops->ndo_get_phys_port_id(dev, ppid);
8394 EXPORT_SYMBOL(dev_get_phys_port_id);
8397 * dev_get_phys_port_name - Get device physical port name
8400 * @len: limit of bytes to copy to name
8402 * Get device physical port name
8404 int dev_get_phys_port_name(struct net_device *dev,
8405 char *name, size_t len)
8407 const struct net_device_ops *ops = dev->netdev_ops;
8410 if (ops->ndo_get_phys_port_name) {
8411 err = ops->ndo_get_phys_port_name(dev, name, len);
8412 if (err != -EOPNOTSUPP)
8415 return devlink_compat_phys_port_name_get(dev, name, len);
8417 EXPORT_SYMBOL(dev_get_phys_port_name);
8420 * dev_get_port_parent_id - Get the device's port parent identifier
8421 * @dev: network device
8422 * @ppid: pointer to a storage for the port's parent identifier
8423 * @recurse: allow/disallow recursion to lower devices
8425 * Get the devices's port parent identifier
8427 int dev_get_port_parent_id(struct net_device *dev,
8428 struct netdev_phys_item_id *ppid,
8431 const struct net_device_ops *ops = dev->netdev_ops;
8432 struct netdev_phys_item_id first = { };
8433 struct net_device *lower_dev;
8434 struct list_head *iter;
8437 if (ops->ndo_get_port_parent_id) {
8438 err = ops->ndo_get_port_parent_id(dev, ppid);
8439 if (err != -EOPNOTSUPP)
8443 err = devlink_compat_switch_id_get(dev, ppid);
8444 if (!err || err != -EOPNOTSUPP)
8450 netdev_for_each_lower_dev(dev, lower_dev, iter) {
8451 err = dev_get_port_parent_id(lower_dev, ppid, recurse);
8456 else if (memcmp(&first, ppid, sizeof(*ppid)))
8462 EXPORT_SYMBOL(dev_get_port_parent_id);
8465 * netdev_port_same_parent_id - Indicate if two network devices have
8466 * the same port parent identifier
8467 * @a: first network device
8468 * @b: second network device
8470 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8472 struct netdev_phys_item_id a_id = { };
8473 struct netdev_phys_item_id b_id = { };
8475 if (dev_get_port_parent_id(a, &a_id, true) ||
8476 dev_get_port_parent_id(b, &b_id, true))
8479 return netdev_phys_item_id_same(&a_id, &b_id);
8481 EXPORT_SYMBOL(netdev_port_same_parent_id);
8484 * dev_change_proto_down - update protocol port state information
8486 * @proto_down: new value
8488 * This info can be used by switch drivers to set the phys state of the
8491 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8493 const struct net_device_ops *ops = dev->netdev_ops;
8495 if (!ops->ndo_change_proto_down)
8497 if (!netif_device_present(dev))
8499 return ops->ndo_change_proto_down(dev, proto_down);
8501 EXPORT_SYMBOL(dev_change_proto_down);
8504 * dev_change_proto_down_generic - generic implementation for
8505 * ndo_change_proto_down that sets carrier according to
8509 * @proto_down: new value
8511 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
8514 netif_carrier_off(dev);
8516 netif_carrier_on(dev);
8517 dev->proto_down = proto_down;
8520 EXPORT_SYMBOL(dev_change_proto_down_generic);
8522 u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op,
8523 enum bpf_netdev_command cmd)
8525 struct netdev_bpf xdp;
8530 memset(&xdp, 0, sizeof(xdp));
8533 /* Query must always succeed. */
8534 WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG);
8539 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op,
8540 struct netlink_ext_ack *extack, u32 flags,
8541 struct bpf_prog *prog)
8543 bool non_hw = !(flags & XDP_FLAGS_HW_MODE);
8544 struct bpf_prog *prev_prog = NULL;
8545 struct netdev_bpf xdp;
8549 prev_prog = bpf_prog_by_id(__dev_xdp_query(dev, bpf_op,
8551 if (IS_ERR(prev_prog))
8555 memset(&xdp, 0, sizeof(xdp));
8556 if (flags & XDP_FLAGS_HW_MODE)
8557 xdp.command = XDP_SETUP_PROG_HW;
8559 xdp.command = XDP_SETUP_PROG;
8560 xdp.extack = extack;
8564 err = bpf_op(dev, &xdp);
8566 bpf_prog_change_xdp(prev_prog, prog);
8569 bpf_prog_put(prev_prog);
8574 static void dev_xdp_uninstall(struct net_device *dev)
8576 struct netdev_bpf xdp;
8579 /* Remove generic XDP */
8580 WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL));
8582 /* Remove from the driver */
8583 ndo_bpf = dev->netdev_ops->ndo_bpf;
8587 memset(&xdp, 0, sizeof(xdp));
8588 xdp.command = XDP_QUERY_PROG;
8589 WARN_ON(ndo_bpf(dev, &xdp));
8591 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8594 /* Remove HW offload */
8595 memset(&xdp, 0, sizeof(xdp));
8596 xdp.command = XDP_QUERY_PROG_HW;
8597 if (!ndo_bpf(dev, &xdp) && xdp.prog_id)
8598 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8603 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
8605 * @extack: netlink extended ack
8606 * @fd: new program fd or negative value to clear
8607 * @flags: xdp-related flags
8609 * Set or clear a bpf program for a device
8611 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
8614 const struct net_device_ops *ops = dev->netdev_ops;
8615 enum bpf_netdev_command query;
8616 struct bpf_prog *prog = NULL;
8617 bpf_op_t bpf_op, bpf_chk;
8623 offload = flags & XDP_FLAGS_HW_MODE;
8624 query = offload ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG;
8626 bpf_op = bpf_chk = ops->ndo_bpf;
8627 if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) {
8628 NL_SET_ERR_MSG(extack, "underlying driver does not support XDP in native mode");
8631 if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE))
8632 bpf_op = generic_xdp_install;
8633 if (bpf_op == bpf_chk)
8634 bpf_chk = generic_xdp_install;
8639 if (!offload && __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG)) {
8640 NL_SET_ERR_MSG(extack, "native and generic XDP can't be active at the same time");
8644 prog_id = __dev_xdp_query(dev, bpf_op, query);
8645 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && prog_id) {
8646 NL_SET_ERR_MSG(extack, "XDP program already attached");
8650 prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
8651 bpf_op == ops->ndo_bpf);
8653 return PTR_ERR(prog);
8655 if (!offload && bpf_prog_is_dev_bound(prog->aux)) {
8656 NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported");
8661 /* prog->aux->id may be 0 for orphaned device-bound progs */
8662 if (prog->aux->id && prog->aux->id == prog_id) {
8667 if (!__dev_xdp_query(dev, bpf_op, query))
8671 err = dev_xdp_install(dev, bpf_op, extack, flags, prog);
8672 if (err < 0 && prog)
8679 * dev_new_index - allocate an ifindex
8680 * @net: the applicable net namespace
8682 * Returns a suitable unique value for a new device interface
8683 * number. The caller must hold the rtnl semaphore or the
8684 * dev_base_lock to be sure it remains unique.
8686 static int dev_new_index(struct net *net)
8688 int ifindex = net->ifindex;
8693 if (!__dev_get_by_index(net, ifindex))
8694 return net->ifindex = ifindex;
8698 /* Delayed registration/unregisteration */
8699 static LIST_HEAD(net_todo_list);
8700 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
8702 static void net_set_todo(struct net_device *dev)
8704 list_add_tail(&dev->todo_list, &net_todo_list);
8705 dev_net(dev)->dev_unreg_count++;
8708 static void rollback_registered_many(struct list_head *head)
8710 struct net_device *dev, *tmp;
8711 LIST_HEAD(close_head);
8713 BUG_ON(dev_boot_phase);
8716 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
8717 /* Some devices call without registering
8718 * for initialization unwind. Remove those
8719 * devices and proceed with the remaining.
8721 if (dev->reg_state == NETREG_UNINITIALIZED) {
8722 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
8726 list_del(&dev->unreg_list);
8729 dev->dismantle = true;
8730 BUG_ON(dev->reg_state != NETREG_REGISTERED);
8733 /* If device is running, close it first. */
8734 list_for_each_entry(dev, head, unreg_list)
8735 list_add_tail(&dev->close_list, &close_head);
8736 dev_close_many(&close_head, true);
8738 list_for_each_entry(dev, head, unreg_list) {
8739 /* And unlink it from device chain. */
8740 unlist_netdevice(dev);
8742 dev->reg_state = NETREG_UNREGISTERING;
8744 flush_all_backlogs();
8748 list_for_each_entry(dev, head, unreg_list) {
8749 struct sk_buff *skb = NULL;
8751 /* Shutdown queueing discipline. */
8754 dev_xdp_uninstall(dev);
8756 /* Notify protocols, that we are about to destroy
8757 * this device. They should clean all the things.
8759 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8761 if (!dev->rtnl_link_ops ||
8762 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8763 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
8764 GFP_KERNEL, NULL, 0);
8767 * Flush the unicast and multicast chains
8772 netdev_name_node_alt_flush(dev);
8773 netdev_name_node_free(dev->name_node);
8775 if (dev->netdev_ops->ndo_uninit)
8776 dev->netdev_ops->ndo_uninit(dev);
8779 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
8781 /* Notifier chain MUST detach us all upper devices. */
8782 WARN_ON(netdev_has_any_upper_dev(dev));
8783 WARN_ON(netdev_has_any_lower_dev(dev));
8785 /* Remove entries from kobject tree */
8786 netdev_unregister_kobject(dev);
8788 /* Remove XPS queueing entries */
8789 netif_reset_xps_queues_gt(dev, 0);
8795 list_for_each_entry(dev, head, unreg_list)
8799 static void rollback_registered(struct net_device *dev)
8803 list_add(&dev->unreg_list, &single);
8804 rollback_registered_many(&single);
8808 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
8809 struct net_device *upper, netdev_features_t features)
8811 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8812 netdev_features_t feature;
8815 for_each_netdev_feature(upper_disables, feature_bit) {
8816 feature = __NETIF_F_BIT(feature_bit);
8817 if (!(upper->wanted_features & feature)
8818 && (features & feature)) {
8819 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
8820 &feature, upper->name);
8821 features &= ~feature;
8828 static void netdev_sync_lower_features(struct net_device *upper,
8829 struct net_device *lower, netdev_features_t features)
8831 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8832 netdev_features_t feature;
8835 for_each_netdev_feature(upper_disables, feature_bit) {
8836 feature = __NETIF_F_BIT(feature_bit);
8837 if (!(features & feature) && (lower->features & feature)) {
8838 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
8839 &feature, lower->name);
8840 lower->wanted_features &= ~feature;
8841 netdev_update_features(lower);
8843 if (unlikely(lower->features & feature))
8844 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
8845 &feature, lower->name);
8850 static netdev_features_t netdev_fix_features(struct net_device *dev,
8851 netdev_features_t features)
8853 /* Fix illegal checksum combinations */
8854 if ((features & NETIF_F_HW_CSUM) &&
8855 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
8856 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
8857 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
8860 /* TSO requires that SG is present as well. */
8861 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
8862 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
8863 features &= ~NETIF_F_ALL_TSO;
8866 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
8867 !(features & NETIF_F_IP_CSUM)) {
8868 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
8869 features &= ~NETIF_F_TSO;
8870 features &= ~NETIF_F_TSO_ECN;
8873 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
8874 !(features & NETIF_F_IPV6_CSUM)) {
8875 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
8876 features &= ~NETIF_F_TSO6;
8879 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
8880 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
8881 features &= ~NETIF_F_TSO_MANGLEID;
8883 /* TSO ECN requires that TSO is present as well. */
8884 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
8885 features &= ~NETIF_F_TSO_ECN;
8887 /* Software GSO depends on SG. */
8888 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
8889 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
8890 features &= ~NETIF_F_GSO;
8893 /* GSO partial features require GSO partial be set */
8894 if ((features & dev->gso_partial_features) &&
8895 !(features & NETIF_F_GSO_PARTIAL)) {
8897 "Dropping partially supported GSO features since no GSO partial.\n");
8898 features &= ~dev->gso_partial_features;
8901 if (!(features & NETIF_F_RXCSUM)) {
8902 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
8903 * successfully merged by hardware must also have the
8904 * checksum verified by hardware. If the user does not
8905 * want to enable RXCSUM, logically, we should disable GRO_HW.
8907 if (features & NETIF_F_GRO_HW) {
8908 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
8909 features &= ~NETIF_F_GRO_HW;
8913 /* LRO/HW-GRO features cannot be combined with RX-FCS */
8914 if (features & NETIF_F_RXFCS) {
8915 if (features & NETIF_F_LRO) {
8916 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
8917 features &= ~NETIF_F_LRO;
8920 if (features & NETIF_F_GRO_HW) {
8921 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
8922 features &= ~NETIF_F_GRO_HW;
8929 int __netdev_update_features(struct net_device *dev)
8931 struct net_device *upper, *lower;
8932 netdev_features_t features;
8933 struct list_head *iter;
8938 features = netdev_get_wanted_features(dev);
8940 if (dev->netdev_ops->ndo_fix_features)
8941 features = dev->netdev_ops->ndo_fix_features(dev, features);
8943 /* driver might be less strict about feature dependencies */
8944 features = netdev_fix_features(dev, features);
8946 /* some features can't be enabled if they're off an an upper device */
8947 netdev_for_each_upper_dev_rcu(dev, upper, iter)
8948 features = netdev_sync_upper_features(dev, upper, features);
8950 if (dev->features == features)
8953 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
8954 &dev->features, &features);
8956 if (dev->netdev_ops->ndo_set_features)
8957 err = dev->netdev_ops->ndo_set_features(dev, features);
8961 if (unlikely(err < 0)) {
8963 "set_features() failed (%d); wanted %pNF, left %pNF\n",
8964 err, &features, &dev->features);
8965 /* return non-0 since some features might have changed and
8966 * it's better to fire a spurious notification than miss it
8972 /* some features must be disabled on lower devices when disabled
8973 * on an upper device (think: bonding master or bridge)
8975 netdev_for_each_lower_dev(dev, lower, iter)
8976 netdev_sync_lower_features(dev, lower, features);
8979 netdev_features_t diff = features ^ dev->features;
8981 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
8982 /* udp_tunnel_{get,drop}_rx_info both need
8983 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
8984 * device, or they won't do anything.
8985 * Thus we need to update dev->features
8986 * *before* calling udp_tunnel_get_rx_info,
8987 * but *after* calling udp_tunnel_drop_rx_info.
8989 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
8990 dev->features = features;
8991 udp_tunnel_get_rx_info(dev);
8993 udp_tunnel_drop_rx_info(dev);
8997 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
8998 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
8999 dev->features = features;
9000 err |= vlan_get_rx_ctag_filter_info(dev);
9002 vlan_drop_rx_ctag_filter_info(dev);
9006 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9007 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9008 dev->features = features;
9009 err |= vlan_get_rx_stag_filter_info(dev);
9011 vlan_drop_rx_stag_filter_info(dev);
9015 dev->features = features;
9018 return err < 0 ? 0 : 1;
9022 * netdev_update_features - recalculate device features
9023 * @dev: the device to check
9025 * Recalculate dev->features set and send notifications if it
9026 * has changed. Should be called after driver or hardware dependent
9027 * conditions might have changed that influence the features.
9029 void netdev_update_features(struct net_device *dev)
9031 if (__netdev_update_features(dev))
9032 netdev_features_change(dev);
9034 EXPORT_SYMBOL(netdev_update_features);
9037 * netdev_change_features - recalculate device features
9038 * @dev: the device to check
9040 * Recalculate dev->features set and send notifications even
9041 * if they have not changed. Should be called instead of
9042 * netdev_update_features() if also dev->vlan_features might
9043 * have changed to allow the changes to be propagated to stacked
9046 void netdev_change_features(struct net_device *dev)
9048 __netdev_update_features(dev);
9049 netdev_features_change(dev);
9051 EXPORT_SYMBOL(netdev_change_features);
9054 * netif_stacked_transfer_operstate - transfer operstate
9055 * @rootdev: the root or lower level device to transfer state from
9056 * @dev: the device to transfer operstate to
9058 * Transfer operational state from root to device. This is normally
9059 * called when a stacking relationship exists between the root
9060 * device and the device(a leaf device).
9062 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9063 struct net_device *dev)
9065 if (rootdev->operstate == IF_OPER_DORMANT)
9066 netif_dormant_on(dev);
9068 netif_dormant_off(dev);
9070 if (netif_carrier_ok(rootdev))
9071 netif_carrier_on(dev);
9073 netif_carrier_off(dev);
9075 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9077 static int netif_alloc_rx_queues(struct net_device *dev)
9079 unsigned int i, count = dev->num_rx_queues;
9080 struct netdev_rx_queue *rx;
9081 size_t sz = count * sizeof(*rx);
9086 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9092 for (i = 0; i < count; i++) {
9095 /* XDP RX-queue setup */
9096 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i);
9103 /* Rollback successful reg's and free other resources */
9105 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9111 static void netif_free_rx_queues(struct net_device *dev)
9113 unsigned int i, count = dev->num_rx_queues;
9115 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9119 for (i = 0; i < count; i++)
9120 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9125 static void netdev_init_one_queue(struct net_device *dev,
9126 struct netdev_queue *queue, void *_unused)
9128 /* Initialize queue lock */
9129 spin_lock_init(&queue->_xmit_lock);
9130 lockdep_set_class(&queue->_xmit_lock, &dev->qdisc_xmit_lock_key);
9131 queue->xmit_lock_owner = -1;
9132 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9135 dql_init(&queue->dql, HZ);
9139 static void netif_free_tx_queues(struct net_device *dev)
9144 static int netif_alloc_netdev_queues(struct net_device *dev)
9146 unsigned int count = dev->num_tx_queues;
9147 struct netdev_queue *tx;
9148 size_t sz = count * sizeof(*tx);
9150 if (count < 1 || count > 0xffff)
9153 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9159 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9160 spin_lock_init(&dev->tx_global_lock);
9165 void netif_tx_stop_all_queues(struct net_device *dev)
9169 for (i = 0; i < dev->num_tx_queues; i++) {
9170 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9172 netif_tx_stop_queue(txq);
9175 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9177 static void netdev_register_lockdep_key(struct net_device *dev)
9179 lockdep_register_key(&dev->qdisc_tx_busylock_key);
9180 lockdep_register_key(&dev->qdisc_running_key);
9181 lockdep_register_key(&dev->qdisc_xmit_lock_key);
9182 lockdep_register_key(&dev->addr_list_lock_key);
9185 static void netdev_unregister_lockdep_key(struct net_device *dev)
9187 lockdep_unregister_key(&dev->qdisc_tx_busylock_key);
9188 lockdep_unregister_key(&dev->qdisc_running_key);
9189 lockdep_unregister_key(&dev->qdisc_xmit_lock_key);
9190 lockdep_unregister_key(&dev->addr_list_lock_key);
9193 void netdev_update_lockdep_key(struct net_device *dev)
9195 struct netdev_queue *queue;
9198 lockdep_unregister_key(&dev->qdisc_xmit_lock_key);
9199 lockdep_unregister_key(&dev->addr_list_lock_key);
9201 lockdep_register_key(&dev->qdisc_xmit_lock_key);
9202 lockdep_register_key(&dev->addr_list_lock_key);
9204 lockdep_set_class(&dev->addr_list_lock, &dev->addr_list_lock_key);
9205 for (i = 0; i < dev->num_tx_queues; i++) {
9206 queue = netdev_get_tx_queue(dev, i);
9208 lockdep_set_class(&queue->_xmit_lock,
9209 &dev->qdisc_xmit_lock_key);
9212 EXPORT_SYMBOL(netdev_update_lockdep_key);
9215 * register_netdevice - register a network device
9216 * @dev: device to register
9218 * Take a completed network device structure and add it to the kernel
9219 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9220 * chain. 0 is returned on success. A negative errno code is returned
9221 * on a failure to set up the device, or if the name is a duplicate.
9223 * Callers must hold the rtnl semaphore. You may want
9224 * register_netdev() instead of this.
9227 * The locking appears insufficient to guarantee two parallel registers
9228 * will not get the same name.
9231 int register_netdevice(struct net_device *dev)
9234 struct net *net = dev_net(dev);
9236 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9237 NETDEV_FEATURE_COUNT);
9238 BUG_ON(dev_boot_phase);
9243 /* When net_device's are persistent, this will be fatal. */
9244 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9247 spin_lock_init(&dev->addr_list_lock);
9248 lockdep_set_class(&dev->addr_list_lock, &dev->addr_list_lock_key);
9250 ret = dev_get_valid_name(net, dev, dev->name);
9255 dev->name_node = netdev_name_node_head_alloc(dev);
9256 if (!dev->name_node)
9259 /* Init, if this function is available */
9260 if (dev->netdev_ops->ndo_init) {
9261 ret = dev->netdev_ops->ndo_init(dev);
9269 if (((dev->hw_features | dev->features) &
9270 NETIF_F_HW_VLAN_CTAG_FILTER) &&
9271 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
9272 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
9273 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
9280 dev->ifindex = dev_new_index(net);
9281 else if (__dev_get_by_index(net, dev->ifindex))
9284 /* Transfer changeable features to wanted_features and enable
9285 * software offloads (GSO and GRO).
9287 dev->hw_features |= NETIF_F_SOFT_FEATURES;
9288 dev->features |= NETIF_F_SOFT_FEATURES;
9290 if (dev->netdev_ops->ndo_udp_tunnel_add) {
9291 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9292 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9295 dev->wanted_features = dev->features & dev->hw_features;
9297 if (!(dev->flags & IFF_LOOPBACK))
9298 dev->hw_features |= NETIF_F_NOCACHE_COPY;
9300 /* If IPv4 TCP segmentation offload is supported we should also
9301 * allow the device to enable segmenting the frame with the option
9302 * of ignoring a static IP ID value. This doesn't enable the
9303 * feature itself but allows the user to enable it later.
9305 if (dev->hw_features & NETIF_F_TSO)
9306 dev->hw_features |= NETIF_F_TSO_MANGLEID;
9307 if (dev->vlan_features & NETIF_F_TSO)
9308 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
9309 if (dev->mpls_features & NETIF_F_TSO)
9310 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
9311 if (dev->hw_enc_features & NETIF_F_TSO)
9312 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
9314 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
9316 dev->vlan_features |= NETIF_F_HIGHDMA;
9318 /* Make NETIF_F_SG inheritable to tunnel devices.
9320 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
9322 /* Make NETIF_F_SG inheritable to MPLS.
9324 dev->mpls_features |= NETIF_F_SG;
9326 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
9327 ret = notifier_to_errno(ret);
9331 ret = netdev_register_kobject(dev);
9334 dev->reg_state = NETREG_REGISTERED;
9336 __netdev_update_features(dev);
9339 * Default initial state at registry is that the
9340 * device is present.
9343 set_bit(__LINK_STATE_PRESENT, &dev->state);
9345 linkwatch_init_dev(dev);
9347 dev_init_scheduler(dev);
9349 list_netdevice(dev);
9350 add_device_randomness(dev->dev_addr, dev->addr_len);
9352 /* If the device has permanent device address, driver should
9353 * set dev_addr and also addr_assign_type should be set to
9354 * NET_ADDR_PERM (default value).
9356 if (dev->addr_assign_type == NET_ADDR_PERM)
9357 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
9359 /* Notify protocols, that a new device appeared. */
9360 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
9361 ret = notifier_to_errno(ret);
9363 rollback_registered(dev);
9366 dev->reg_state = NETREG_UNREGISTERED;
9369 * Prevent userspace races by waiting until the network
9370 * device is fully setup before sending notifications.
9372 if (!dev->rtnl_link_ops ||
9373 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
9374 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
9380 if (dev->netdev_ops->ndo_uninit)
9381 dev->netdev_ops->ndo_uninit(dev);
9382 if (dev->priv_destructor)
9383 dev->priv_destructor(dev);
9385 netdev_name_node_free(dev->name_node);
9388 EXPORT_SYMBOL(register_netdevice);
9391 * init_dummy_netdev - init a dummy network device for NAPI
9392 * @dev: device to init
9394 * This takes a network device structure and initialize the minimum
9395 * amount of fields so it can be used to schedule NAPI polls without
9396 * registering a full blown interface. This is to be used by drivers
9397 * that need to tie several hardware interfaces to a single NAPI
9398 * poll scheduler due to HW limitations.
9400 int init_dummy_netdev(struct net_device *dev)
9402 /* Clear everything. Note we don't initialize spinlocks
9403 * are they aren't supposed to be taken by any of the
9404 * NAPI code and this dummy netdev is supposed to be
9405 * only ever used for NAPI polls
9407 memset(dev, 0, sizeof(struct net_device));
9409 /* make sure we BUG if trying to hit standard
9410 * register/unregister code path
9412 dev->reg_state = NETREG_DUMMY;
9414 /* NAPI wants this */
9415 INIT_LIST_HEAD(&dev->napi_list);
9417 /* a dummy interface is started by default */
9418 set_bit(__LINK_STATE_PRESENT, &dev->state);
9419 set_bit(__LINK_STATE_START, &dev->state);
9421 /* napi_busy_loop stats accounting wants this */
9422 dev_net_set(dev, &init_net);
9424 /* Note : We dont allocate pcpu_refcnt for dummy devices,
9425 * because users of this 'device' dont need to change
9431 EXPORT_SYMBOL_GPL(init_dummy_netdev);
9435 * register_netdev - register a network device
9436 * @dev: device to register
9438 * Take a completed network device structure and add it to the kernel
9439 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9440 * chain. 0 is returned on success. A negative errno code is returned
9441 * on a failure to set up the device, or if the name is a duplicate.
9443 * This is a wrapper around register_netdevice that takes the rtnl semaphore
9444 * and expands the device name if you passed a format string to
9447 int register_netdev(struct net_device *dev)
9451 if (rtnl_lock_killable())
9453 err = register_netdevice(dev);
9457 EXPORT_SYMBOL(register_netdev);
9459 int netdev_refcnt_read(const struct net_device *dev)
9463 for_each_possible_cpu(i)
9464 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
9467 EXPORT_SYMBOL(netdev_refcnt_read);
9470 * netdev_wait_allrefs - wait until all references are gone.
9471 * @dev: target net_device
9473 * This is called when unregistering network devices.
9475 * Any protocol or device that holds a reference should register
9476 * for netdevice notification, and cleanup and put back the
9477 * reference if they receive an UNREGISTER event.
9478 * We can get stuck here if buggy protocols don't correctly
9481 static void netdev_wait_allrefs(struct net_device *dev)
9483 unsigned long rebroadcast_time, warning_time;
9486 linkwatch_forget_dev(dev);
9488 rebroadcast_time = warning_time = jiffies;
9489 refcnt = netdev_refcnt_read(dev);
9491 while (refcnt != 0) {
9492 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
9495 /* Rebroadcast unregister notification */
9496 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
9502 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
9504 /* We must not have linkwatch events
9505 * pending on unregister. If this
9506 * happens, we simply run the queue
9507 * unscheduled, resulting in a noop
9510 linkwatch_run_queue();
9515 rebroadcast_time = jiffies;
9520 refcnt = netdev_refcnt_read(dev);
9522 if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) {
9523 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
9525 warning_time = jiffies;
9534 * register_netdevice(x1);
9535 * register_netdevice(x2);
9537 * unregister_netdevice(y1);
9538 * unregister_netdevice(y2);
9544 * We are invoked by rtnl_unlock().
9545 * This allows us to deal with problems:
9546 * 1) We can delete sysfs objects which invoke hotplug
9547 * without deadlocking with linkwatch via keventd.
9548 * 2) Since we run with the RTNL semaphore not held, we can sleep
9549 * safely in order to wait for the netdev refcnt to drop to zero.
9551 * We must not return until all unregister events added during
9552 * the interval the lock was held have been completed.
9554 void netdev_run_todo(void)
9556 struct list_head list;
9558 /* Snapshot list, allow later requests */
9559 list_replace_init(&net_todo_list, &list);
9564 /* Wait for rcu callbacks to finish before next phase */
9565 if (!list_empty(&list))
9568 while (!list_empty(&list)) {
9569 struct net_device *dev
9570 = list_first_entry(&list, struct net_device, todo_list);
9571 list_del(&dev->todo_list);
9573 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
9574 pr_err("network todo '%s' but state %d\n",
9575 dev->name, dev->reg_state);
9580 dev->reg_state = NETREG_UNREGISTERED;
9582 netdev_wait_allrefs(dev);
9585 BUG_ON(netdev_refcnt_read(dev));
9586 BUG_ON(!list_empty(&dev->ptype_all));
9587 BUG_ON(!list_empty(&dev->ptype_specific));
9588 WARN_ON(rcu_access_pointer(dev->ip_ptr));
9589 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
9590 #if IS_ENABLED(CONFIG_DECNET)
9591 WARN_ON(dev->dn_ptr);
9593 if (dev->priv_destructor)
9594 dev->priv_destructor(dev);
9595 if (dev->needs_free_netdev)
9598 /* Report a network device has been unregistered */
9600 dev_net(dev)->dev_unreg_count--;
9602 wake_up(&netdev_unregistering_wq);
9604 /* Free network device */
9605 kobject_put(&dev->dev.kobj);
9609 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
9610 * all the same fields in the same order as net_device_stats, with only
9611 * the type differing, but rtnl_link_stats64 may have additional fields
9612 * at the end for newer counters.
9614 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
9615 const struct net_device_stats *netdev_stats)
9617 #if BITS_PER_LONG == 64
9618 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
9619 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
9620 /* zero out counters that only exist in rtnl_link_stats64 */
9621 memset((char *)stats64 + sizeof(*netdev_stats), 0,
9622 sizeof(*stats64) - sizeof(*netdev_stats));
9624 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
9625 const unsigned long *src = (const unsigned long *)netdev_stats;
9626 u64 *dst = (u64 *)stats64;
9628 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
9629 for (i = 0; i < n; i++)
9631 /* zero out counters that only exist in rtnl_link_stats64 */
9632 memset((char *)stats64 + n * sizeof(u64), 0,
9633 sizeof(*stats64) - n * sizeof(u64));
9636 EXPORT_SYMBOL(netdev_stats_to_stats64);
9639 * dev_get_stats - get network device statistics
9640 * @dev: device to get statistics from
9641 * @storage: place to store stats
9643 * Get network statistics from device. Return @storage.
9644 * The device driver may provide its own method by setting
9645 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
9646 * otherwise the internal statistics structure is used.
9648 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
9649 struct rtnl_link_stats64 *storage)
9651 const struct net_device_ops *ops = dev->netdev_ops;
9653 if (ops->ndo_get_stats64) {
9654 memset(storage, 0, sizeof(*storage));
9655 ops->ndo_get_stats64(dev, storage);
9656 } else if (ops->ndo_get_stats) {
9657 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
9659 netdev_stats_to_stats64(storage, &dev->stats);
9661 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
9662 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
9663 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
9666 EXPORT_SYMBOL(dev_get_stats);
9668 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
9670 struct netdev_queue *queue = dev_ingress_queue(dev);
9672 #ifdef CONFIG_NET_CLS_ACT
9675 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
9678 netdev_init_one_queue(dev, queue, NULL);
9679 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
9680 queue->qdisc_sleeping = &noop_qdisc;
9681 rcu_assign_pointer(dev->ingress_queue, queue);
9686 static const struct ethtool_ops default_ethtool_ops;
9688 void netdev_set_default_ethtool_ops(struct net_device *dev,
9689 const struct ethtool_ops *ops)
9691 if (dev->ethtool_ops == &default_ethtool_ops)
9692 dev->ethtool_ops = ops;
9694 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
9696 void netdev_freemem(struct net_device *dev)
9698 char *addr = (char *)dev - dev->padded;
9704 * alloc_netdev_mqs - allocate network device
9705 * @sizeof_priv: size of private data to allocate space for
9706 * @name: device name format string
9707 * @name_assign_type: origin of device name
9708 * @setup: callback to initialize device
9709 * @txqs: the number of TX subqueues to allocate
9710 * @rxqs: the number of RX subqueues to allocate
9712 * Allocates a struct net_device with private data area for driver use
9713 * and performs basic initialization. Also allocates subqueue structs
9714 * for each queue on the device.
9716 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
9717 unsigned char name_assign_type,
9718 void (*setup)(struct net_device *),
9719 unsigned int txqs, unsigned int rxqs)
9721 struct net_device *dev;
9722 unsigned int alloc_size;
9723 struct net_device *p;
9725 BUG_ON(strlen(name) >= sizeof(dev->name));
9728 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
9733 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
9737 alloc_size = sizeof(struct net_device);
9739 /* ensure 32-byte alignment of private area */
9740 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
9741 alloc_size += sizeof_priv;
9743 /* ensure 32-byte alignment of whole construct */
9744 alloc_size += NETDEV_ALIGN - 1;
9746 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9750 dev = PTR_ALIGN(p, NETDEV_ALIGN);
9751 dev->padded = (char *)dev - (char *)p;
9753 dev->pcpu_refcnt = alloc_percpu(int);
9754 if (!dev->pcpu_refcnt)
9757 if (dev_addr_init(dev))
9763 dev_net_set(dev, &init_net);
9765 netdev_register_lockdep_key(dev);
9767 dev->gso_max_size = GSO_MAX_SIZE;
9768 dev->gso_max_segs = GSO_MAX_SEGS;
9769 dev->upper_level = 1;
9770 dev->lower_level = 1;
9772 INIT_LIST_HEAD(&dev->napi_list);
9773 INIT_LIST_HEAD(&dev->unreg_list);
9774 INIT_LIST_HEAD(&dev->close_list);
9775 INIT_LIST_HEAD(&dev->link_watch_list);
9776 INIT_LIST_HEAD(&dev->adj_list.upper);
9777 INIT_LIST_HEAD(&dev->adj_list.lower);
9778 INIT_LIST_HEAD(&dev->ptype_all);
9779 INIT_LIST_HEAD(&dev->ptype_specific);
9780 #ifdef CONFIG_NET_SCHED
9781 hash_init(dev->qdisc_hash);
9783 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
9786 if (!dev->tx_queue_len) {
9787 dev->priv_flags |= IFF_NO_QUEUE;
9788 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
9791 dev->num_tx_queues = txqs;
9792 dev->real_num_tx_queues = txqs;
9793 if (netif_alloc_netdev_queues(dev))
9796 dev->num_rx_queues = rxqs;
9797 dev->real_num_rx_queues = rxqs;
9798 if (netif_alloc_rx_queues(dev))
9801 strcpy(dev->name, name);
9802 dev->name_assign_type = name_assign_type;
9803 dev->group = INIT_NETDEV_GROUP;
9804 if (!dev->ethtool_ops)
9805 dev->ethtool_ops = &default_ethtool_ops;
9807 nf_hook_ingress_init(dev);
9816 free_percpu(dev->pcpu_refcnt);
9818 netdev_freemem(dev);
9821 EXPORT_SYMBOL(alloc_netdev_mqs);
9824 * free_netdev - free network device
9827 * This function does the last stage of destroying an allocated device
9828 * interface. The reference to the device object is released. If this
9829 * is the last reference then it will be freed.Must be called in process
9832 void free_netdev(struct net_device *dev)
9834 struct napi_struct *p, *n;
9837 netif_free_tx_queues(dev);
9838 netif_free_rx_queues(dev);
9840 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
9842 /* Flush device addresses */
9843 dev_addr_flush(dev);
9845 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
9848 free_percpu(dev->pcpu_refcnt);
9849 dev->pcpu_refcnt = NULL;
9851 netdev_unregister_lockdep_key(dev);
9853 /* Compatibility with error handling in drivers */
9854 if (dev->reg_state == NETREG_UNINITIALIZED) {
9855 netdev_freemem(dev);
9859 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
9860 dev->reg_state = NETREG_RELEASED;
9862 /* will free via device release */
9863 put_device(&dev->dev);
9865 EXPORT_SYMBOL(free_netdev);
9868 * synchronize_net - Synchronize with packet receive processing
9870 * Wait for packets currently being received to be done.
9871 * Does not block later packets from starting.
9873 void synchronize_net(void)
9876 if (rtnl_is_locked())
9877 synchronize_rcu_expedited();
9881 EXPORT_SYMBOL(synchronize_net);
9884 * unregister_netdevice_queue - remove device from the kernel
9888 * This function shuts down a device interface and removes it
9889 * from the kernel tables.
9890 * If head not NULL, device is queued to be unregistered later.
9892 * Callers must hold the rtnl semaphore. You may want
9893 * unregister_netdev() instead of this.
9896 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
9901 list_move_tail(&dev->unreg_list, head);
9903 rollback_registered(dev);
9904 /* Finish processing unregister after unlock */
9908 EXPORT_SYMBOL(unregister_netdevice_queue);
9911 * unregister_netdevice_many - unregister many devices
9912 * @head: list of devices
9914 * Note: As most callers use a stack allocated list_head,
9915 * we force a list_del() to make sure stack wont be corrupted later.
9917 void unregister_netdevice_many(struct list_head *head)
9919 struct net_device *dev;
9921 if (!list_empty(head)) {
9922 rollback_registered_many(head);
9923 list_for_each_entry(dev, head, unreg_list)
9928 EXPORT_SYMBOL(unregister_netdevice_many);
9931 * unregister_netdev - remove device from the kernel
9934 * This function shuts down a device interface and removes it
9935 * from the kernel tables.
9937 * This is just a wrapper for unregister_netdevice that takes
9938 * the rtnl semaphore. In general you want to use this and not
9939 * unregister_netdevice.
9941 void unregister_netdev(struct net_device *dev)
9944 unregister_netdevice(dev);
9947 EXPORT_SYMBOL(unregister_netdev);
9950 * dev_change_net_namespace - move device to different nethost namespace
9952 * @net: network namespace
9953 * @pat: If not NULL name pattern to try if the current device name
9954 * is already taken in the destination network namespace.
9956 * This function shuts down a device interface and moves it
9957 * to a new network namespace. On success 0 is returned, on
9958 * a failure a netagive errno code is returned.
9960 * Callers must hold the rtnl semaphore.
9963 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
9965 int err, new_nsid, new_ifindex;
9969 /* Don't allow namespace local devices to be moved. */
9971 if (dev->features & NETIF_F_NETNS_LOCAL)
9974 /* Ensure the device has been registrered */
9975 if (dev->reg_state != NETREG_REGISTERED)
9978 /* Get out if there is nothing todo */
9980 if (net_eq(dev_net(dev), net))
9983 /* Pick the destination device name, and ensure
9984 * we can use it in the destination network namespace.
9987 if (__dev_get_by_name(net, dev->name)) {
9988 /* We get here if we can't use the current device name */
9991 err = dev_get_valid_name(net, dev, pat);
9997 * And now a mini version of register_netdevice unregister_netdevice.
10000 /* If device is running close it first. */
10003 /* And unlink it from device chain */
10004 unlist_netdevice(dev);
10008 /* Shutdown queueing discipline. */
10011 /* Notify protocols, that we are about to destroy
10012 * this device. They should clean all the things.
10014 * Note that dev->reg_state stays at NETREG_REGISTERED.
10015 * This is wanted because this way 8021q and macvlan know
10016 * the device is just moving and can keep their slaves up.
10018 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10021 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
10022 /* If there is an ifindex conflict assign a new one */
10023 if (__dev_get_by_index(net, dev->ifindex))
10024 new_ifindex = dev_new_index(net);
10026 new_ifindex = dev->ifindex;
10028 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
10032 * Flush the unicast and multicast chains
10037 /* Send a netdev-removed uevent to the old namespace */
10038 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
10039 netdev_adjacent_del_links(dev);
10041 /* Actually switch the network namespace */
10042 dev_net_set(dev, net);
10043 dev->ifindex = new_ifindex;
10045 /* Send a netdev-add uevent to the new namespace */
10046 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
10047 netdev_adjacent_add_links(dev);
10049 /* Fixup kobjects */
10050 err = device_rename(&dev->dev, dev->name);
10053 /* Add the device back in the hashes */
10054 list_netdevice(dev);
10056 /* Notify protocols, that a new device appeared. */
10057 call_netdevice_notifiers(NETDEV_REGISTER, dev);
10060 * Prevent userspace races by waiting until the network
10061 * device is fully setup before sending notifications.
10063 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10070 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
10072 static int dev_cpu_dead(unsigned int oldcpu)
10074 struct sk_buff **list_skb;
10075 struct sk_buff *skb;
10077 struct softnet_data *sd, *oldsd, *remsd = NULL;
10079 local_irq_disable();
10080 cpu = smp_processor_id();
10081 sd = &per_cpu(softnet_data, cpu);
10082 oldsd = &per_cpu(softnet_data, oldcpu);
10084 /* Find end of our completion_queue. */
10085 list_skb = &sd->completion_queue;
10087 list_skb = &(*list_skb)->next;
10088 /* Append completion queue from offline CPU. */
10089 *list_skb = oldsd->completion_queue;
10090 oldsd->completion_queue = NULL;
10092 /* Append output queue from offline CPU. */
10093 if (oldsd->output_queue) {
10094 *sd->output_queue_tailp = oldsd->output_queue;
10095 sd->output_queue_tailp = oldsd->output_queue_tailp;
10096 oldsd->output_queue = NULL;
10097 oldsd->output_queue_tailp = &oldsd->output_queue;
10099 /* Append NAPI poll list from offline CPU, with one exception :
10100 * process_backlog() must be called by cpu owning percpu backlog.
10101 * We properly handle process_queue & input_pkt_queue later.
10103 while (!list_empty(&oldsd->poll_list)) {
10104 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
10105 struct napi_struct,
10108 list_del_init(&napi->poll_list);
10109 if (napi->poll == process_backlog)
10112 ____napi_schedule(sd, napi);
10115 raise_softirq_irqoff(NET_TX_SOFTIRQ);
10116 local_irq_enable();
10119 remsd = oldsd->rps_ipi_list;
10120 oldsd->rps_ipi_list = NULL;
10122 /* send out pending IPI's on offline CPU */
10123 net_rps_send_ipi(remsd);
10125 /* Process offline CPU's input_pkt_queue */
10126 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
10128 input_queue_head_incr(oldsd);
10130 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
10132 input_queue_head_incr(oldsd);
10139 * netdev_increment_features - increment feature set by one
10140 * @all: current feature set
10141 * @one: new feature set
10142 * @mask: mask feature set
10144 * Computes a new feature set after adding a device with feature set
10145 * @one to the master device with current feature set @all. Will not
10146 * enable anything that is off in @mask. Returns the new feature set.
10148 netdev_features_t netdev_increment_features(netdev_features_t all,
10149 netdev_features_t one, netdev_features_t mask)
10151 if (mask & NETIF_F_HW_CSUM)
10152 mask |= NETIF_F_CSUM_MASK;
10153 mask |= NETIF_F_VLAN_CHALLENGED;
10155 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
10156 all &= one | ~NETIF_F_ALL_FOR_ALL;
10158 /* If one device supports hw checksumming, set for all. */
10159 if (all & NETIF_F_HW_CSUM)
10160 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
10164 EXPORT_SYMBOL(netdev_increment_features);
10166 static struct hlist_head * __net_init netdev_create_hash(void)
10169 struct hlist_head *hash;
10171 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
10173 for (i = 0; i < NETDEV_HASHENTRIES; i++)
10174 INIT_HLIST_HEAD(&hash[i]);
10179 /* Initialize per network namespace state */
10180 static int __net_init netdev_init(struct net *net)
10182 BUILD_BUG_ON(GRO_HASH_BUCKETS >
10183 8 * sizeof_field(struct napi_struct, gro_bitmask));
10185 if (net != &init_net)
10186 INIT_LIST_HEAD(&net->dev_base_head);
10188 net->dev_name_head = netdev_create_hash();
10189 if (net->dev_name_head == NULL)
10192 net->dev_index_head = netdev_create_hash();
10193 if (net->dev_index_head == NULL)
10196 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
10201 kfree(net->dev_name_head);
10207 * netdev_drivername - network driver for the device
10208 * @dev: network device
10210 * Determine network driver for device.
10212 const char *netdev_drivername(const struct net_device *dev)
10214 const struct device_driver *driver;
10215 const struct device *parent;
10216 const char *empty = "";
10218 parent = dev->dev.parent;
10222 driver = parent->driver;
10223 if (driver && driver->name)
10224 return driver->name;
10228 static void __netdev_printk(const char *level, const struct net_device *dev,
10229 struct va_format *vaf)
10231 if (dev && dev->dev.parent) {
10232 dev_printk_emit(level[1] - '0',
10235 dev_driver_string(dev->dev.parent),
10236 dev_name(dev->dev.parent),
10237 netdev_name(dev), netdev_reg_state(dev),
10240 printk("%s%s%s: %pV",
10241 level, netdev_name(dev), netdev_reg_state(dev), vaf);
10243 printk("%s(NULL net_device): %pV", level, vaf);
10247 void netdev_printk(const char *level, const struct net_device *dev,
10248 const char *format, ...)
10250 struct va_format vaf;
10253 va_start(args, format);
10258 __netdev_printk(level, dev, &vaf);
10262 EXPORT_SYMBOL(netdev_printk);
10264 #define define_netdev_printk_level(func, level) \
10265 void func(const struct net_device *dev, const char *fmt, ...) \
10267 struct va_format vaf; \
10270 va_start(args, fmt); \
10275 __netdev_printk(level, dev, &vaf); \
10279 EXPORT_SYMBOL(func);
10281 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
10282 define_netdev_printk_level(netdev_alert, KERN_ALERT);
10283 define_netdev_printk_level(netdev_crit, KERN_CRIT);
10284 define_netdev_printk_level(netdev_err, KERN_ERR);
10285 define_netdev_printk_level(netdev_warn, KERN_WARNING);
10286 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
10287 define_netdev_printk_level(netdev_info, KERN_INFO);
10289 static void __net_exit netdev_exit(struct net *net)
10291 kfree(net->dev_name_head);
10292 kfree(net->dev_index_head);
10293 if (net != &init_net)
10294 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
10297 static struct pernet_operations __net_initdata netdev_net_ops = {
10298 .init = netdev_init,
10299 .exit = netdev_exit,
10302 static void __net_exit default_device_exit(struct net *net)
10304 struct net_device *dev, *aux;
10306 * Push all migratable network devices back to the
10307 * initial network namespace
10310 for_each_netdev_safe(net, dev, aux) {
10312 char fb_name[IFNAMSIZ];
10314 /* Ignore unmoveable devices (i.e. loopback) */
10315 if (dev->features & NETIF_F_NETNS_LOCAL)
10318 /* Leave virtual devices for the generic cleanup */
10319 if (dev->rtnl_link_ops)
10322 /* Push remaining network devices to init_net */
10323 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
10324 if (__dev_get_by_name(&init_net, fb_name))
10325 snprintf(fb_name, IFNAMSIZ, "dev%%d");
10326 err = dev_change_net_namespace(dev, &init_net, fb_name);
10328 pr_emerg("%s: failed to move %s to init_net: %d\n",
10329 __func__, dev->name, err);
10336 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
10338 /* Return with the rtnl_lock held when there are no network
10339 * devices unregistering in any network namespace in net_list.
10342 bool unregistering;
10343 DEFINE_WAIT_FUNC(wait, woken_wake_function);
10345 add_wait_queue(&netdev_unregistering_wq, &wait);
10347 unregistering = false;
10349 list_for_each_entry(net, net_list, exit_list) {
10350 if (net->dev_unreg_count > 0) {
10351 unregistering = true;
10355 if (!unregistering)
10359 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
10361 remove_wait_queue(&netdev_unregistering_wq, &wait);
10364 static void __net_exit default_device_exit_batch(struct list_head *net_list)
10366 /* At exit all network devices most be removed from a network
10367 * namespace. Do this in the reverse order of registration.
10368 * Do this across as many network namespaces as possible to
10369 * improve batching efficiency.
10371 struct net_device *dev;
10373 LIST_HEAD(dev_kill_list);
10375 /* To prevent network device cleanup code from dereferencing
10376 * loopback devices or network devices that have been freed
10377 * wait here for all pending unregistrations to complete,
10378 * before unregistring the loopback device and allowing the
10379 * network namespace be freed.
10381 * The netdev todo list containing all network devices
10382 * unregistrations that happen in default_device_exit_batch
10383 * will run in the rtnl_unlock() at the end of
10384 * default_device_exit_batch.
10386 rtnl_lock_unregistering(net_list);
10387 list_for_each_entry(net, net_list, exit_list) {
10388 for_each_netdev_reverse(net, dev) {
10389 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
10390 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
10392 unregister_netdevice_queue(dev, &dev_kill_list);
10395 unregister_netdevice_many(&dev_kill_list);
10399 static struct pernet_operations __net_initdata default_device_ops = {
10400 .exit = default_device_exit,
10401 .exit_batch = default_device_exit_batch,
10405 * Initialize the DEV module. At boot time this walks the device list and
10406 * unhooks any devices that fail to initialise (normally hardware not
10407 * present) and leaves us with a valid list of present and active devices.
10412 * This is called single threaded during boot, so no need
10413 * to take the rtnl semaphore.
10415 static int __init net_dev_init(void)
10417 int i, rc = -ENOMEM;
10419 BUG_ON(!dev_boot_phase);
10421 if (dev_proc_init())
10424 if (netdev_kobject_init())
10427 INIT_LIST_HEAD(&ptype_all);
10428 for (i = 0; i < PTYPE_HASH_SIZE; i++)
10429 INIT_LIST_HEAD(&ptype_base[i]);
10431 INIT_LIST_HEAD(&offload_base);
10433 if (register_pernet_subsys(&netdev_net_ops))
10437 * Initialise the packet receive queues.
10440 for_each_possible_cpu(i) {
10441 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
10442 struct softnet_data *sd = &per_cpu(softnet_data, i);
10444 INIT_WORK(flush, flush_backlog);
10446 skb_queue_head_init(&sd->input_pkt_queue);
10447 skb_queue_head_init(&sd->process_queue);
10448 #ifdef CONFIG_XFRM_OFFLOAD
10449 skb_queue_head_init(&sd->xfrm_backlog);
10451 INIT_LIST_HEAD(&sd->poll_list);
10452 sd->output_queue_tailp = &sd->output_queue;
10454 sd->csd.func = rps_trigger_softirq;
10459 init_gro_hash(&sd->backlog);
10460 sd->backlog.poll = process_backlog;
10461 sd->backlog.weight = weight_p;
10464 dev_boot_phase = 0;
10466 /* The loopback device is special if any other network devices
10467 * is present in a network namespace the loopback device must
10468 * be present. Since we now dynamically allocate and free the
10469 * loopback device ensure this invariant is maintained by
10470 * keeping the loopback device as the first device on the
10471 * list of network devices. Ensuring the loopback devices
10472 * is the first device that appears and the last network device
10475 if (register_pernet_device(&loopback_net_ops))
10478 if (register_pernet_device(&default_device_ops))
10481 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
10482 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
10484 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
10485 NULL, dev_cpu_dead);
10492 subsys_initcall(net_dev_init);