1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * NET3 Protocol independent device support routines.
5 * Derived from the non IP parts of dev.c 1.0.19
7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
8 * Mark Evans, <evansmp@uhura.aston.ac.uk>
11 * Florian la Roche <rzsfl@rz.uni-sb.de>
12 * Alan Cox <gw4pts@gw4pts.ampr.org>
13 * David Hinds <dahinds@users.sourceforge.net>
14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
15 * Adam Sulmicki <adam@cfar.umd.edu>
16 * Pekka Riikonen <priikone@poesidon.pspt.fi>
19 * D.J. Barrow : Fixed bug where dev->refcnt gets set
20 * to 2 if register_netdev gets called
21 * before net_dev_init & also removed a
22 * few lines of code in the process.
23 * Alan Cox : device private ioctl copies fields back.
24 * Alan Cox : Transmit queue code does relevant
25 * stunts to keep the queue safe.
26 * Alan Cox : Fixed double lock.
27 * Alan Cox : Fixed promisc NULL pointer trap
28 * ???????? : Support the full private ioctl range
29 * Alan Cox : Moved ioctl permission check into
31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
32 * Alan Cox : 100 backlog just doesn't cut it when
33 * you start doing multicast video 8)
34 * Alan Cox : Rewrote net_bh and list manager.
35 * Alan Cox : Fix ETH_P_ALL echoback lengths.
36 * Alan Cox : Took out transmit every packet pass
37 * Saved a few bytes in the ioctl handler
38 * Alan Cox : Network driver sets packet type before
39 * calling netif_rx. Saves a function
41 * Alan Cox : Hashed net_bh()
42 * Richard Kooijman: Timestamp fixes.
43 * Alan Cox : Wrong field in SIOCGIFDSTADDR
44 * Alan Cox : Device lock protection.
45 * Alan Cox : Fixed nasty side effect of device close
47 * Rudi Cilibrasi : Pass the right thing to
49 * Dave Miller : 32bit quantity for the device lock to
50 * make it work out on a Sparc.
51 * Bjorn Ekwall : Added KERNELD hack.
52 * Alan Cox : Cleaned up the backlog initialise.
53 * Craig Metz : SIOCGIFCONF fix if space for under
55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
56 * is no device open function.
57 * Andi Kleen : Fix error reporting for SIOCGIFCONF
58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
59 * Cyrus Durgin : Cleaned for KMOD
60 * Adam Sulmicki : Bug Fix : Network Device Unload
61 * A network device unload needs to purge
63 * Paul Rusty Russell : SIOCSIFNAME
64 * Pekka Riikonen : Netdev boot-time settings code
65 * Andrew Morton : Make unregister_netdevice wait
66 * indefinitely on dev->refcnt
67 * J Hadi Salim : - Backlog queue sampling
68 * - netif_rx() feedback
71 #include <linux/uaccess.h>
72 #include <linux/bitops.h>
73 #include <linux/capability.h>
74 #include <linux/cpu.h>
75 #include <linux/types.h>
76 #include <linux/kernel.h>
77 #include <linux/hash.h>
78 #include <linux/slab.h>
79 #include <linux/sched.h>
80 #include <linux/sched/mm.h>
81 #include <linux/mutex.h>
82 #include <linux/rwsem.h>
83 #include <linux/string.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/ethtool.h>
93 #include <linux/skbuff.h>
94 #include <linux/bpf.h>
95 #include <linux/bpf_trace.h>
96 #include <net/net_namespace.h>
98 #include <net/busy_poll.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/dst_metadata.h>
103 #include <net/pkt_sched.h>
104 #include <net/pkt_cls.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/module.h>
110 #include <linux/netpoll.h>
111 #include <linux/rcupdate.h>
112 #include <linux/delay.h>
113 #include <net/iw_handler.h>
114 #include <asm/current.h>
115 #include <linux/audit.h>
116 #include <linux/dmaengine.h>
117 #include <linux/err.h>
118 #include <linux/ctype.h>
119 #include <linux/if_arp.h>
120 #include <linux/if_vlan.h>
121 #include <linux/ip.h>
123 #include <net/mpls.h>
124 #include <linux/ipv6.h>
125 #include <linux/in.h>
126 #include <linux/jhash.h>
127 #include <linux/random.h>
128 #include <trace/events/napi.h>
129 #include <trace/events/net.h>
130 #include <trace/events/skb.h>
131 #include <linux/inetdevice.h>
132 #include <linux/cpu_rmap.h>
133 #include <linux/static_key.h>
134 #include <linux/hashtable.h>
135 #include <linux/vmalloc.h>
136 #include <linux/if_macvlan.h>
137 #include <linux/errqueue.h>
138 #include <linux/hrtimer.h>
139 #include <linux/netfilter_ingress.h>
140 #include <linux/crash_dump.h>
141 #include <linux/sctp.h>
142 #include <net/udp_tunnel.h>
143 #include <linux/net_namespace.h>
144 #include <linux/indirect_call_wrapper.h>
145 #include <net/devlink.h>
147 #include "net-sysfs.h"
149 #define MAX_GRO_SKBS 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 DECLARE_RWSEM(devnet_rename_sem);
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 /* Device list insertion */
233 static void list_netdevice(struct net_device *dev)
235 struct net *net = dev_net(dev);
239 write_lock_bh(&dev_base_lock);
240 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
241 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
242 hlist_add_head_rcu(&dev->index_hlist,
243 dev_index_hash(net, dev->ifindex));
244 write_unlock_bh(&dev_base_lock);
246 dev_base_seq_inc(net);
249 /* Device list removal
250 * caller must respect a RCU grace period before freeing/reusing dev
252 static void unlist_netdevice(struct net_device *dev)
256 /* Unlink dev from the device chain */
257 write_lock_bh(&dev_base_lock);
258 list_del_rcu(&dev->dev_list);
259 hlist_del_rcu(&dev->name_hlist);
260 hlist_del_rcu(&dev->index_hlist);
261 write_unlock_bh(&dev_base_lock);
263 dev_base_seq_inc(dev_net(dev));
270 static RAW_NOTIFIER_HEAD(netdev_chain);
273 * Device drivers call our routines to queue packets here. We empty the
274 * queue in the local softnet handler.
277 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
278 EXPORT_PER_CPU_SYMBOL(softnet_data);
280 /*******************************************************************************
282 * Protocol management and registration routines
284 *******************************************************************************/
288 * Add a protocol ID to the list. Now that the input handler is
289 * smarter we can dispense with all the messy stuff that used to be
292 * BEWARE!!! Protocol handlers, mangling input packets,
293 * MUST BE last in hash buckets and checking protocol handlers
294 * MUST start from promiscuous ptype_all chain in net_bh.
295 * It is true now, do not change it.
296 * Explanation follows: if protocol handler, mangling packet, will
297 * be the first on list, it is not able to sense, that packet
298 * is cloned and should be copied-on-write, so that it will
299 * change it and subsequent readers will get broken packet.
303 static inline struct list_head *ptype_head(const struct packet_type *pt)
305 if (pt->type == htons(ETH_P_ALL))
306 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
308 return pt->dev ? &pt->dev->ptype_specific :
309 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
313 * dev_add_pack - add packet handler
314 * @pt: packet type declaration
316 * Add a protocol handler to the networking stack. The passed &packet_type
317 * is linked into kernel lists and may not be freed until it has been
318 * removed from the kernel lists.
320 * This call does not sleep therefore it can not
321 * guarantee all CPU's that are in middle of receiving packets
322 * will see the new packet type (until the next received packet).
325 void dev_add_pack(struct packet_type *pt)
327 struct list_head *head = ptype_head(pt);
329 spin_lock(&ptype_lock);
330 list_add_rcu(&pt->list, head);
331 spin_unlock(&ptype_lock);
333 EXPORT_SYMBOL(dev_add_pack);
336 * __dev_remove_pack - remove packet handler
337 * @pt: packet type declaration
339 * Remove a protocol handler that was previously added to the kernel
340 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
341 * from the kernel lists and can be freed or reused once this function
344 * The packet type might still be in use by receivers
345 * and must not be freed until after all the CPU's have gone
346 * through a quiescent state.
348 void __dev_remove_pack(struct packet_type *pt)
350 struct list_head *head = ptype_head(pt);
351 struct packet_type *pt1;
353 spin_lock(&ptype_lock);
355 list_for_each_entry(pt1, head, list) {
357 list_del_rcu(&pt->list);
362 pr_warn("dev_remove_pack: %p not found\n", pt);
364 spin_unlock(&ptype_lock);
366 EXPORT_SYMBOL(__dev_remove_pack);
369 * dev_remove_pack - remove packet handler
370 * @pt: packet type declaration
372 * Remove a protocol handler that was previously added to the kernel
373 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
374 * from the kernel lists and can be freed or reused once this function
377 * This call sleeps to guarantee that no CPU is looking at the packet
380 void dev_remove_pack(struct packet_type *pt)
382 __dev_remove_pack(pt);
386 EXPORT_SYMBOL(dev_remove_pack);
390 * dev_add_offload - register offload handlers
391 * @po: protocol offload declaration
393 * Add protocol offload handlers to the networking stack. The passed
394 * &proto_offload is linked into kernel lists and may not be freed until
395 * it has been removed from the kernel lists.
397 * This call does not sleep therefore it can not
398 * guarantee all CPU's that are in middle of receiving packets
399 * will see the new offload handlers (until the next received packet).
401 void dev_add_offload(struct packet_offload *po)
403 struct packet_offload *elem;
405 spin_lock(&offload_lock);
406 list_for_each_entry(elem, &offload_base, list) {
407 if (po->priority < elem->priority)
410 list_add_rcu(&po->list, elem->list.prev);
411 spin_unlock(&offload_lock);
413 EXPORT_SYMBOL(dev_add_offload);
416 * __dev_remove_offload - remove offload handler
417 * @po: packet offload declaration
419 * Remove a protocol offload handler that was previously added to the
420 * kernel offload handlers by dev_add_offload(). The passed &offload_type
421 * is removed from the kernel lists and can be freed or reused once this
424 * The packet type might still be in use by receivers
425 * and must not be freed until after all the CPU's have gone
426 * through a quiescent state.
428 static void __dev_remove_offload(struct packet_offload *po)
430 struct list_head *head = &offload_base;
431 struct packet_offload *po1;
433 spin_lock(&offload_lock);
435 list_for_each_entry(po1, head, list) {
437 list_del_rcu(&po->list);
442 pr_warn("dev_remove_offload: %p not found\n", po);
444 spin_unlock(&offload_lock);
448 * dev_remove_offload - remove packet offload handler
449 * @po: packet offload declaration
451 * Remove a packet offload handler that was previously added to the kernel
452 * offload handlers by dev_add_offload(). The passed &offload_type is
453 * removed from the kernel lists and can be freed or reused once this
456 * This call sleeps to guarantee that no CPU is looking at the packet
459 void dev_remove_offload(struct packet_offload *po)
461 __dev_remove_offload(po);
465 EXPORT_SYMBOL(dev_remove_offload);
467 /******************************************************************************
469 * Device Boot-time Settings Routines
471 ******************************************************************************/
473 /* Boot time configuration table */
474 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
477 * netdev_boot_setup_add - add new setup entry
478 * @name: name of the device
479 * @map: configured settings for the device
481 * Adds new setup entry to the dev_boot_setup list. The function
482 * returns 0 on error and 1 on success. This is a generic routine to
485 static int netdev_boot_setup_add(char *name, struct ifmap *map)
487 struct netdev_boot_setup *s;
491 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
492 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
493 memset(s[i].name, 0, sizeof(s[i].name));
494 strlcpy(s[i].name, name, IFNAMSIZ);
495 memcpy(&s[i].map, map, sizeof(s[i].map));
500 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
504 * netdev_boot_setup_check - check boot time settings
505 * @dev: the netdevice
507 * Check boot time settings for the device.
508 * The found settings are set for the device to be used
509 * later in the device probing.
510 * Returns 0 if no settings found, 1 if they are.
512 int netdev_boot_setup_check(struct net_device *dev)
514 struct netdev_boot_setup *s = dev_boot_setup;
517 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
518 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
519 !strcmp(dev->name, s[i].name)) {
520 dev->irq = s[i].map.irq;
521 dev->base_addr = s[i].map.base_addr;
522 dev->mem_start = s[i].map.mem_start;
523 dev->mem_end = s[i].map.mem_end;
529 EXPORT_SYMBOL(netdev_boot_setup_check);
533 * netdev_boot_base - get address from boot time settings
534 * @prefix: prefix for network device
535 * @unit: id for network device
537 * Check boot time settings for the base address of device.
538 * The found settings are set for the device to be used
539 * later in the device probing.
540 * Returns 0 if no settings found.
542 unsigned long netdev_boot_base(const char *prefix, int unit)
544 const struct netdev_boot_setup *s = dev_boot_setup;
548 sprintf(name, "%s%d", prefix, unit);
551 * If device already registered then return base of 1
552 * to indicate not to probe for this interface
554 if (__dev_get_by_name(&init_net, name))
557 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
558 if (!strcmp(name, s[i].name))
559 return s[i].map.base_addr;
564 * Saves at boot time configured settings for any netdevice.
566 int __init netdev_boot_setup(char *str)
571 str = get_options(str, ARRAY_SIZE(ints), ints);
576 memset(&map, 0, sizeof(map));
580 map.base_addr = ints[2];
582 map.mem_start = ints[3];
584 map.mem_end = ints[4];
586 /* Add new entry to the list */
587 return netdev_boot_setup_add(str, &map);
590 __setup("netdev=", netdev_boot_setup);
592 /*******************************************************************************
594 * Device Interface Subroutines
596 *******************************************************************************/
599 * dev_get_iflink - get 'iflink' value of a interface
600 * @dev: targeted interface
602 * Indicates the ifindex the interface is linked to.
603 * Physical interfaces have the same 'ifindex' and 'iflink' values.
606 int dev_get_iflink(const struct net_device *dev)
608 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
609 return dev->netdev_ops->ndo_get_iflink(dev);
613 EXPORT_SYMBOL(dev_get_iflink);
616 * dev_fill_metadata_dst - Retrieve tunnel egress information.
617 * @dev: targeted interface
620 * For better visibility of tunnel traffic OVS needs to retrieve
621 * egress tunnel information for a packet. Following API allows
622 * user to get this info.
624 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
626 struct ip_tunnel_info *info;
628 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
631 info = skb_tunnel_info_unclone(skb);
634 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
637 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
639 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
642 * __dev_get_by_name - find a device by its name
643 * @net: the applicable net namespace
644 * @name: name to find
646 * Find an interface by name. Must be called under RTNL semaphore
647 * or @dev_base_lock. If the name is found a pointer to the device
648 * is returned. If the name is not found then %NULL is returned. The
649 * reference counters are not incremented so the caller must be
650 * careful with locks.
653 struct net_device *__dev_get_by_name(struct net *net, const char *name)
655 struct net_device *dev;
656 struct hlist_head *head = dev_name_hash(net, name);
658 hlist_for_each_entry(dev, head, name_hlist)
659 if (!strncmp(dev->name, name, IFNAMSIZ))
664 EXPORT_SYMBOL(__dev_get_by_name);
667 * dev_get_by_name_rcu - find a device by its name
668 * @net: the applicable net namespace
669 * @name: name to find
671 * Find an interface by name.
672 * If the name is found a pointer to the device is returned.
673 * If the name is not found then %NULL is returned.
674 * The reference counters are not incremented so the caller must be
675 * careful with locks. The caller must hold RCU lock.
678 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
680 struct net_device *dev;
681 struct hlist_head *head = dev_name_hash(net, name);
683 hlist_for_each_entry_rcu(dev, head, name_hlist)
684 if (!strncmp(dev->name, name, IFNAMSIZ))
689 EXPORT_SYMBOL(dev_get_by_name_rcu);
692 * dev_get_by_name - find a device by its name
693 * @net: the applicable net namespace
694 * @name: name to find
696 * Find an interface by name. This can be called from any
697 * context and does its own locking. The returned handle has
698 * the usage count incremented and the caller must use dev_put() to
699 * release it when it is no longer needed. %NULL is returned if no
700 * matching device is found.
703 struct net_device *dev_get_by_name(struct net *net, const char *name)
705 struct net_device *dev;
708 dev = dev_get_by_name_rcu(net, name);
714 EXPORT_SYMBOL(dev_get_by_name);
717 * __dev_get_by_index - find a device by its ifindex
718 * @net: the applicable net namespace
719 * @ifindex: index of device
721 * Search for an interface by index. Returns %NULL if the device
722 * is not found or a pointer to the device. The device has not
723 * had its reference counter increased so the caller must be careful
724 * about locking. The caller must hold either the RTNL semaphore
728 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
730 struct net_device *dev;
731 struct hlist_head *head = dev_index_hash(net, ifindex);
733 hlist_for_each_entry(dev, head, index_hlist)
734 if (dev->ifindex == ifindex)
739 EXPORT_SYMBOL(__dev_get_by_index);
742 * dev_get_by_index_rcu - find a device by its ifindex
743 * @net: the applicable net namespace
744 * @ifindex: index of device
746 * Search for an interface by index. Returns %NULL if the device
747 * is not found or a pointer to the device. The device has not
748 * had its reference counter increased so the caller must be careful
749 * about locking. The caller must hold RCU lock.
752 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
754 struct net_device *dev;
755 struct hlist_head *head = dev_index_hash(net, ifindex);
757 hlist_for_each_entry_rcu(dev, head, index_hlist)
758 if (dev->ifindex == ifindex)
763 EXPORT_SYMBOL(dev_get_by_index_rcu);
767 * dev_get_by_index - find a device by its ifindex
768 * @net: the applicable net namespace
769 * @ifindex: index of device
771 * Search for an interface by index. Returns NULL if the device
772 * is not found or a pointer to the device. The device returned has
773 * had a reference added and the pointer is safe until the user calls
774 * dev_put to indicate they have finished with it.
777 struct net_device *dev_get_by_index(struct net *net, int ifindex)
779 struct net_device *dev;
782 dev = dev_get_by_index_rcu(net, ifindex);
788 EXPORT_SYMBOL(dev_get_by_index);
791 * dev_get_by_napi_id - find a device by napi_id
792 * @napi_id: ID of the NAPI struct
794 * Search for an interface by NAPI ID. Returns %NULL if the device
795 * is not found or a pointer to the device. The device has not had
796 * its reference counter increased so the caller must be careful
797 * about locking. The caller must hold RCU lock.
800 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
802 struct napi_struct *napi;
804 WARN_ON_ONCE(!rcu_read_lock_held());
806 if (napi_id < MIN_NAPI_ID)
809 napi = napi_by_id(napi_id);
811 return napi ? napi->dev : NULL;
813 EXPORT_SYMBOL(dev_get_by_napi_id);
816 * netdev_get_name - get a netdevice name, knowing its ifindex.
817 * @net: network namespace
818 * @name: a pointer to the buffer where the name will be stored.
819 * @ifindex: the ifindex of the interface to get the name from.
821 int netdev_get_name(struct net *net, char *name, int ifindex)
823 struct net_device *dev;
826 down_read(&devnet_rename_sem);
829 dev = dev_get_by_index_rcu(net, ifindex);
835 strcpy(name, dev->name);
840 up_read(&devnet_rename_sem);
845 * dev_getbyhwaddr_rcu - find a device by its hardware address
846 * @net: the applicable net namespace
847 * @type: media type of device
848 * @ha: hardware address
850 * Search for an interface by MAC address. Returns NULL if the device
851 * is not found or a pointer to the device.
852 * The caller must hold RCU or RTNL.
853 * The returned device has not had its ref count increased
854 * and the caller must therefore be careful about locking
858 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
861 struct net_device *dev;
863 for_each_netdev_rcu(net, dev)
864 if (dev->type == type &&
865 !memcmp(dev->dev_addr, ha, dev->addr_len))
870 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
872 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
874 struct net_device *dev;
877 for_each_netdev(net, dev)
878 if (dev->type == type)
883 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
885 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
887 struct net_device *dev, *ret = NULL;
890 for_each_netdev_rcu(net, dev)
891 if (dev->type == type) {
899 EXPORT_SYMBOL(dev_getfirstbyhwtype);
902 * __dev_get_by_flags - find any device with given flags
903 * @net: the applicable net namespace
904 * @if_flags: IFF_* values
905 * @mask: bitmask of bits in if_flags to check
907 * Search for any interface with the given flags. Returns NULL if a device
908 * is not found or a pointer to the device. Must be called inside
909 * rtnl_lock(), and result refcount is unchanged.
912 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
915 struct net_device *dev, *ret;
920 for_each_netdev(net, dev) {
921 if (((dev->flags ^ if_flags) & mask) == 0) {
928 EXPORT_SYMBOL(__dev_get_by_flags);
931 * dev_valid_name - check if name is okay for network device
934 * Network device names need to be valid file names to
935 * to allow sysfs to work. We also disallow any kind of
938 bool dev_valid_name(const char *name)
942 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
944 if (!strcmp(name, ".") || !strcmp(name, ".."))
948 if (*name == '/' || *name == ':' || isspace(*name))
954 EXPORT_SYMBOL(dev_valid_name);
957 * __dev_alloc_name - allocate a name for a device
958 * @net: network namespace to allocate the device name in
959 * @name: name format string
960 * @buf: scratch buffer and result name string
962 * Passed a format string - eg "lt%d" it will try and find a suitable
963 * id. It scans list of devices to build up a free map, then chooses
964 * the first empty slot. The caller must hold the dev_base or rtnl lock
965 * while allocating the name and adding the device in order to avoid
967 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
968 * Returns the number of the unit assigned or a negative errno code.
971 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
975 const int max_netdevices = 8*PAGE_SIZE;
976 unsigned long *inuse;
977 struct net_device *d;
979 if (!dev_valid_name(name))
982 p = strchr(name, '%');
985 * Verify the string as this thing may have come from
986 * the user. There must be either one "%d" and no other "%"
989 if (p[1] != 'd' || strchr(p + 2, '%'))
992 /* Use one page as a bit array of possible slots */
993 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
997 for_each_netdev(net, d) {
998 if (!sscanf(d->name, name, &i))
1000 if (i < 0 || i >= max_netdevices)
1003 /* avoid cases where sscanf is not exact inverse of printf */
1004 snprintf(buf, IFNAMSIZ, name, i);
1005 if (!strncmp(buf, d->name, IFNAMSIZ))
1009 i = find_first_zero_bit(inuse, max_netdevices);
1010 free_page((unsigned long) inuse);
1013 snprintf(buf, IFNAMSIZ, name, i);
1014 if (!__dev_get_by_name(net, buf))
1017 /* It is possible to run out of possible slots
1018 * when the name is long and there isn't enough space left
1019 * for the digits, or if all bits are used.
1024 static int dev_alloc_name_ns(struct net *net,
1025 struct net_device *dev,
1032 ret = __dev_alloc_name(net, name, buf);
1034 strlcpy(dev->name, buf, IFNAMSIZ);
1039 * dev_alloc_name - allocate a name for a device
1041 * @name: name format string
1043 * Passed a format string - eg "lt%d" it will try and find a suitable
1044 * id. It scans list of devices to build up a free map, then chooses
1045 * the first empty slot. The caller must hold the dev_base or rtnl lock
1046 * while allocating the name and adding the device in order to avoid
1048 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1049 * Returns the number of the unit assigned or a negative errno code.
1052 int dev_alloc_name(struct net_device *dev, const char *name)
1054 return dev_alloc_name_ns(dev_net(dev), dev, name);
1056 EXPORT_SYMBOL(dev_alloc_name);
1058 int dev_get_valid_name(struct net *net, struct net_device *dev,
1063 if (!dev_valid_name(name))
1066 if (strchr(name, '%'))
1067 return dev_alloc_name_ns(net, dev, name);
1068 else if (__dev_get_by_name(net, name))
1070 else if (dev->name != name)
1071 strlcpy(dev->name, name, IFNAMSIZ);
1075 EXPORT_SYMBOL(dev_get_valid_name);
1078 * dev_change_name - change name of a device
1080 * @newname: name (or format string) must be at least IFNAMSIZ
1082 * Change name of a device, can pass format strings "eth%d".
1085 int dev_change_name(struct net_device *dev, const char *newname)
1087 unsigned char old_assign_type;
1088 char oldname[IFNAMSIZ];
1094 BUG_ON(!dev_net(dev));
1098 /* Some auto-enslaved devices e.g. failover slaves are
1099 * special, as userspace might rename the device after
1100 * the interface had been brought up and running since
1101 * the point kernel initiated auto-enslavement. Allow
1102 * live name change even when these slave devices are
1105 * Typically, users of these auto-enslaving devices
1106 * don't actually care about slave name change, as
1107 * they are supposed to operate on master interface
1110 if (dev->flags & IFF_UP &&
1111 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1114 down_write(&devnet_rename_sem);
1116 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1117 up_write(&devnet_rename_sem);
1121 memcpy(oldname, dev->name, IFNAMSIZ);
1123 err = dev_get_valid_name(net, dev, newname);
1125 up_write(&devnet_rename_sem);
1129 if (oldname[0] && !strchr(oldname, '%'))
1130 netdev_info(dev, "renamed from %s\n", oldname);
1132 old_assign_type = dev->name_assign_type;
1133 dev->name_assign_type = NET_NAME_RENAMED;
1136 ret = device_rename(&dev->dev, dev->name);
1138 memcpy(dev->name, oldname, IFNAMSIZ);
1139 dev->name_assign_type = old_assign_type;
1140 up_write(&devnet_rename_sem);
1144 up_write(&devnet_rename_sem);
1146 netdev_adjacent_rename_links(dev, oldname);
1148 write_lock_bh(&dev_base_lock);
1149 hlist_del_rcu(&dev->name_hlist);
1150 write_unlock_bh(&dev_base_lock);
1154 write_lock_bh(&dev_base_lock);
1155 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1156 write_unlock_bh(&dev_base_lock);
1158 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1159 ret = notifier_to_errno(ret);
1162 /* err >= 0 after dev_alloc_name() or stores the first errno */
1165 down_write(&devnet_rename_sem);
1166 memcpy(dev->name, oldname, IFNAMSIZ);
1167 memcpy(oldname, newname, IFNAMSIZ);
1168 dev->name_assign_type = old_assign_type;
1169 old_assign_type = NET_NAME_RENAMED;
1172 pr_err("%s: name change rollback failed: %d\n",
1181 * dev_set_alias - change ifalias of a device
1183 * @alias: name up to IFALIASZ
1184 * @len: limit of bytes to copy from info
1186 * Set ifalias for a device,
1188 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1190 struct dev_ifalias *new_alias = NULL;
1192 if (len >= IFALIASZ)
1196 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1200 memcpy(new_alias->ifalias, alias, len);
1201 new_alias->ifalias[len] = 0;
1204 mutex_lock(&ifalias_mutex);
1205 rcu_swap_protected(dev->ifalias, new_alias,
1206 mutex_is_locked(&ifalias_mutex));
1207 mutex_unlock(&ifalias_mutex);
1210 kfree_rcu(new_alias, rcuhead);
1214 EXPORT_SYMBOL(dev_set_alias);
1217 * dev_get_alias - get ifalias of a device
1219 * @name: buffer to store name of ifalias
1220 * @len: size of buffer
1222 * get ifalias for a device. Caller must make sure dev cannot go
1223 * away, e.g. rcu read lock or own a reference count to device.
1225 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1227 const struct dev_ifalias *alias;
1231 alias = rcu_dereference(dev->ifalias);
1233 ret = snprintf(name, len, "%s", alias->ifalias);
1240 * netdev_features_change - device changes features
1241 * @dev: device to cause notification
1243 * Called to indicate a device has changed features.
1245 void netdev_features_change(struct net_device *dev)
1247 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1249 EXPORT_SYMBOL(netdev_features_change);
1252 * netdev_state_change - device changes state
1253 * @dev: device to cause notification
1255 * Called to indicate a device has changed state. This function calls
1256 * the notifier chains for netdev_chain and sends a NEWLINK message
1257 * to the routing socket.
1259 void netdev_state_change(struct net_device *dev)
1261 if (dev->flags & IFF_UP) {
1262 struct netdev_notifier_change_info change_info = {
1266 call_netdevice_notifiers_info(NETDEV_CHANGE,
1268 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1271 EXPORT_SYMBOL(netdev_state_change);
1274 * netdev_notify_peers - notify network peers about existence of @dev
1275 * @dev: network device
1277 * Generate traffic such that interested network peers are aware of
1278 * @dev, such as by generating a gratuitous ARP. This may be used when
1279 * a device wants to inform the rest of the network about some sort of
1280 * reconfiguration such as a failover event or virtual machine
1283 void netdev_notify_peers(struct net_device *dev)
1286 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1287 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1290 EXPORT_SYMBOL(netdev_notify_peers);
1292 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1294 const struct net_device_ops *ops = dev->netdev_ops;
1299 if (!netif_device_present(dev))
1302 /* Block netpoll from trying to do any rx path servicing.
1303 * If we don't do this there is a chance ndo_poll_controller
1304 * or ndo_poll may be running while we open the device
1306 netpoll_poll_disable(dev);
1308 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1309 ret = notifier_to_errno(ret);
1313 set_bit(__LINK_STATE_START, &dev->state);
1315 if (ops->ndo_validate_addr)
1316 ret = ops->ndo_validate_addr(dev);
1318 if (!ret && ops->ndo_open)
1319 ret = ops->ndo_open(dev);
1321 netpoll_poll_enable(dev);
1324 clear_bit(__LINK_STATE_START, &dev->state);
1326 dev->flags |= IFF_UP;
1327 dev_set_rx_mode(dev);
1329 add_device_randomness(dev->dev_addr, dev->addr_len);
1336 * dev_open - prepare an interface for use.
1337 * @dev: device to open
1338 * @extack: netlink extended ack
1340 * Takes a device from down to up state. The device's private open
1341 * function is invoked and then the multicast lists are loaded. Finally
1342 * the device is moved into the up state and a %NETDEV_UP message is
1343 * sent to the netdev notifier chain.
1345 * Calling this function on an active interface is a nop. On a failure
1346 * a negative errno code is returned.
1348 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1352 if (dev->flags & IFF_UP)
1355 ret = __dev_open(dev, extack);
1359 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1360 call_netdevice_notifiers(NETDEV_UP, dev);
1364 EXPORT_SYMBOL(dev_open);
1366 static void __dev_close_many(struct list_head *head)
1368 struct net_device *dev;
1373 list_for_each_entry(dev, head, close_list) {
1374 /* Temporarily disable netpoll until the interface is down */
1375 netpoll_poll_disable(dev);
1377 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1379 clear_bit(__LINK_STATE_START, &dev->state);
1381 /* Synchronize to scheduled poll. We cannot touch poll list, it
1382 * can be even on different cpu. So just clear netif_running().
1384 * dev->stop() will invoke napi_disable() on all of it's
1385 * napi_struct instances on this device.
1387 smp_mb__after_atomic(); /* Commit netif_running(). */
1390 dev_deactivate_many(head);
1392 list_for_each_entry(dev, head, close_list) {
1393 const struct net_device_ops *ops = dev->netdev_ops;
1396 * Call the device specific close. This cannot fail.
1397 * Only if device is UP
1399 * We allow it to be called even after a DETACH hot-plug
1405 dev->flags &= ~IFF_UP;
1406 netpoll_poll_enable(dev);
1410 static void __dev_close(struct net_device *dev)
1414 list_add(&dev->close_list, &single);
1415 __dev_close_many(&single);
1419 void dev_close_many(struct list_head *head, bool unlink)
1421 struct net_device *dev, *tmp;
1423 /* Remove the devices that don't need to be closed */
1424 list_for_each_entry_safe(dev, tmp, head, close_list)
1425 if (!(dev->flags & IFF_UP))
1426 list_del_init(&dev->close_list);
1428 __dev_close_many(head);
1430 list_for_each_entry_safe(dev, tmp, head, close_list) {
1431 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1432 call_netdevice_notifiers(NETDEV_DOWN, dev);
1434 list_del_init(&dev->close_list);
1437 EXPORT_SYMBOL(dev_close_many);
1440 * dev_close - shutdown an interface.
1441 * @dev: device to shutdown
1443 * This function moves an active device into down state. A
1444 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1445 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1448 void dev_close(struct net_device *dev)
1450 if (dev->flags & IFF_UP) {
1453 list_add(&dev->close_list, &single);
1454 dev_close_many(&single, true);
1458 EXPORT_SYMBOL(dev_close);
1462 * dev_disable_lro - disable Large Receive Offload on a device
1465 * Disable Large Receive Offload (LRO) on a net device. Must be
1466 * called under RTNL. This is needed if received packets may be
1467 * forwarded to another interface.
1469 void dev_disable_lro(struct net_device *dev)
1471 struct net_device *lower_dev;
1472 struct list_head *iter;
1474 dev->wanted_features &= ~NETIF_F_LRO;
1475 netdev_update_features(dev);
1477 if (unlikely(dev->features & NETIF_F_LRO))
1478 netdev_WARN(dev, "failed to disable LRO!\n");
1480 netdev_for_each_lower_dev(dev, lower_dev, iter)
1481 dev_disable_lro(lower_dev);
1483 EXPORT_SYMBOL(dev_disable_lro);
1486 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1489 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1490 * called under RTNL. This is needed if Generic XDP is installed on
1493 static void dev_disable_gro_hw(struct net_device *dev)
1495 dev->wanted_features &= ~NETIF_F_GRO_HW;
1496 netdev_update_features(dev);
1498 if (unlikely(dev->features & NETIF_F_GRO_HW))
1499 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1502 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1505 case NETDEV_##val: \
1506 return "NETDEV_" __stringify(val);
1508 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1509 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1510 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1511 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1512 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1513 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1514 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1515 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1516 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1520 return "UNKNOWN_NETDEV_EVENT";
1522 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1524 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1525 struct net_device *dev)
1527 struct netdev_notifier_info info = {
1531 return nb->notifier_call(nb, val, &info);
1534 static int dev_boot_phase = 1;
1537 * register_netdevice_notifier - register a network notifier block
1540 * Register a notifier to be called when network device events occur.
1541 * The notifier passed is linked into the kernel structures and must
1542 * not be reused until it has been unregistered. A negative errno code
1543 * is returned on a failure.
1545 * When registered all registration and up events are replayed
1546 * to the new notifier to allow device to have a race free
1547 * view of the network device list.
1550 int register_netdevice_notifier(struct notifier_block *nb)
1552 struct net_device *dev;
1553 struct net_device *last;
1557 /* Close race with setup_net() and cleanup_net() */
1558 down_write(&pernet_ops_rwsem);
1560 err = raw_notifier_chain_register(&netdev_chain, nb);
1566 for_each_netdev(net, dev) {
1567 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1568 err = notifier_to_errno(err);
1572 if (!(dev->flags & IFF_UP))
1575 call_netdevice_notifier(nb, NETDEV_UP, dev);
1581 up_write(&pernet_ops_rwsem);
1587 for_each_netdev(net, dev) {
1591 if (dev->flags & IFF_UP) {
1592 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1594 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1596 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1601 raw_notifier_chain_unregister(&netdev_chain, nb);
1604 EXPORT_SYMBOL(register_netdevice_notifier);
1607 * unregister_netdevice_notifier - unregister a network notifier block
1610 * Unregister a notifier previously registered by
1611 * register_netdevice_notifier(). The notifier is unlinked into the
1612 * kernel structures and may then be reused. A negative errno code
1613 * is returned on a failure.
1615 * After unregistering unregister and down device events are synthesized
1616 * for all devices on the device list to the removed notifier to remove
1617 * the need for special case cleanup code.
1620 int unregister_netdevice_notifier(struct notifier_block *nb)
1622 struct net_device *dev;
1626 /* Close race with setup_net() and cleanup_net() */
1627 down_write(&pernet_ops_rwsem);
1629 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1634 for_each_netdev(net, dev) {
1635 if (dev->flags & IFF_UP) {
1636 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1638 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1640 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1645 up_write(&pernet_ops_rwsem);
1648 EXPORT_SYMBOL(unregister_netdevice_notifier);
1651 * call_netdevice_notifiers_info - call all network notifier blocks
1652 * @val: value passed unmodified to notifier function
1653 * @info: notifier information data
1655 * Call all network notifier blocks. Parameters and return value
1656 * are as for raw_notifier_call_chain().
1659 static int call_netdevice_notifiers_info(unsigned long val,
1660 struct netdev_notifier_info *info)
1663 return raw_notifier_call_chain(&netdev_chain, val, info);
1666 static int call_netdevice_notifiers_extack(unsigned long val,
1667 struct net_device *dev,
1668 struct netlink_ext_ack *extack)
1670 struct netdev_notifier_info info = {
1675 return call_netdevice_notifiers_info(val, &info);
1679 * call_netdevice_notifiers - call all network notifier blocks
1680 * @val: value passed unmodified to notifier function
1681 * @dev: net_device pointer passed unmodified to notifier function
1683 * Call all network notifier blocks. Parameters and return value
1684 * are as for raw_notifier_call_chain().
1687 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1689 return call_netdevice_notifiers_extack(val, dev, NULL);
1691 EXPORT_SYMBOL(call_netdevice_notifiers);
1694 * call_netdevice_notifiers_mtu - call all network notifier blocks
1695 * @val: value passed unmodified to notifier function
1696 * @dev: net_device pointer passed unmodified to notifier function
1697 * @arg: additional u32 argument passed to the notifier function
1699 * Call all network notifier blocks. Parameters and return value
1700 * are as for raw_notifier_call_chain().
1702 static int call_netdevice_notifiers_mtu(unsigned long val,
1703 struct net_device *dev, u32 arg)
1705 struct netdev_notifier_info_ext info = {
1710 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
1712 return call_netdevice_notifiers_info(val, &info.info);
1715 #ifdef CONFIG_NET_INGRESS
1716 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
1718 void net_inc_ingress_queue(void)
1720 static_branch_inc(&ingress_needed_key);
1722 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1724 void net_dec_ingress_queue(void)
1726 static_branch_dec(&ingress_needed_key);
1728 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1731 #ifdef CONFIG_NET_EGRESS
1732 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
1734 void net_inc_egress_queue(void)
1736 static_branch_inc(&egress_needed_key);
1738 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1740 void net_dec_egress_queue(void)
1742 static_branch_dec(&egress_needed_key);
1744 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1747 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
1748 #ifdef CONFIG_JUMP_LABEL
1749 static atomic_t netstamp_needed_deferred;
1750 static atomic_t netstamp_wanted;
1751 static void netstamp_clear(struct work_struct *work)
1753 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1756 wanted = atomic_add_return(deferred, &netstamp_wanted);
1758 static_branch_enable(&netstamp_needed_key);
1760 static_branch_disable(&netstamp_needed_key);
1762 static DECLARE_WORK(netstamp_work, netstamp_clear);
1765 void net_enable_timestamp(void)
1767 #ifdef CONFIG_JUMP_LABEL
1771 wanted = atomic_read(&netstamp_wanted);
1774 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1777 atomic_inc(&netstamp_needed_deferred);
1778 schedule_work(&netstamp_work);
1780 static_branch_inc(&netstamp_needed_key);
1783 EXPORT_SYMBOL(net_enable_timestamp);
1785 void net_disable_timestamp(void)
1787 #ifdef CONFIG_JUMP_LABEL
1791 wanted = atomic_read(&netstamp_wanted);
1794 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
1797 atomic_dec(&netstamp_needed_deferred);
1798 schedule_work(&netstamp_work);
1800 static_branch_dec(&netstamp_needed_key);
1803 EXPORT_SYMBOL(net_disable_timestamp);
1805 static inline void net_timestamp_set(struct sk_buff *skb)
1808 if (static_branch_unlikely(&netstamp_needed_key))
1809 __net_timestamp(skb);
1812 #define net_timestamp_check(COND, SKB) \
1813 if (static_branch_unlikely(&netstamp_needed_key)) { \
1814 if ((COND) && !(SKB)->tstamp) \
1815 __net_timestamp(SKB); \
1818 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1822 if (!(dev->flags & IFF_UP))
1825 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1826 if (skb->len <= len)
1829 /* if TSO is enabled, we don't care about the length as the packet
1830 * could be forwarded without being segmented before
1832 if (skb_is_gso(skb))
1837 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1839 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1841 int ret = ____dev_forward_skb(dev, skb);
1844 skb->protocol = eth_type_trans(skb, dev);
1845 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1850 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1853 * dev_forward_skb - loopback an skb to another netif
1855 * @dev: destination network device
1856 * @skb: buffer to forward
1859 * NET_RX_SUCCESS (no congestion)
1860 * NET_RX_DROP (packet was dropped, but freed)
1862 * dev_forward_skb can be used for injecting an skb from the
1863 * start_xmit function of one device into the receive queue
1864 * of another device.
1866 * The receiving device may be in another namespace, so
1867 * we have to clear all information in the skb that could
1868 * impact namespace isolation.
1870 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1872 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1874 EXPORT_SYMBOL_GPL(dev_forward_skb);
1876 static inline int deliver_skb(struct sk_buff *skb,
1877 struct packet_type *pt_prev,
1878 struct net_device *orig_dev)
1880 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1882 refcount_inc(&skb->users);
1883 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1886 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1887 struct packet_type **pt,
1888 struct net_device *orig_dev,
1890 struct list_head *ptype_list)
1892 struct packet_type *ptype, *pt_prev = *pt;
1894 list_for_each_entry_rcu(ptype, ptype_list, list) {
1895 if (ptype->type != type)
1898 deliver_skb(skb, pt_prev, orig_dev);
1904 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1906 if (!ptype->af_packet_priv || !skb->sk)
1909 if (ptype->id_match)
1910 return ptype->id_match(ptype, skb->sk);
1911 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1918 * dev_nit_active - return true if any network interface taps are in use
1920 * @dev: network device to check for the presence of taps
1922 bool dev_nit_active(struct net_device *dev)
1924 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
1926 EXPORT_SYMBOL_GPL(dev_nit_active);
1929 * Support routine. Sends outgoing frames to any network
1930 * taps currently in use.
1933 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1935 struct packet_type *ptype;
1936 struct sk_buff *skb2 = NULL;
1937 struct packet_type *pt_prev = NULL;
1938 struct list_head *ptype_list = &ptype_all;
1942 list_for_each_entry_rcu(ptype, ptype_list, list) {
1943 if (ptype->ignore_outgoing)
1946 /* Never send packets back to the socket
1947 * they originated from - MvS (miquels@drinkel.ow.org)
1949 if (skb_loop_sk(ptype, skb))
1953 deliver_skb(skb2, pt_prev, skb->dev);
1958 /* need to clone skb, done only once */
1959 skb2 = skb_clone(skb, GFP_ATOMIC);
1963 net_timestamp_set(skb2);
1965 /* skb->nh should be correctly
1966 * set by sender, so that the second statement is
1967 * just protection against buggy protocols.
1969 skb_reset_mac_header(skb2);
1971 if (skb_network_header(skb2) < skb2->data ||
1972 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1973 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1974 ntohs(skb2->protocol),
1976 skb_reset_network_header(skb2);
1979 skb2->transport_header = skb2->network_header;
1980 skb2->pkt_type = PACKET_OUTGOING;
1984 if (ptype_list == &ptype_all) {
1985 ptype_list = &dev->ptype_all;
1990 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
1991 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1997 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2000 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2001 * @dev: Network device
2002 * @txq: number of queues available
2004 * If real_num_tx_queues is changed the tc mappings may no longer be
2005 * valid. To resolve this verify the tc mapping remains valid and if
2006 * not NULL the mapping. With no priorities mapping to this
2007 * offset/count pair it will no longer be used. In the worst case TC0
2008 * is invalid nothing can be done so disable priority mappings. If is
2009 * expected that drivers will fix this mapping if they can before
2010 * calling netif_set_real_num_tx_queues.
2012 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2015 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2017 /* If TC0 is invalidated disable TC mapping */
2018 if (tc->offset + tc->count > txq) {
2019 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2024 /* Invalidated prio to tc mappings set to TC0 */
2025 for (i = 1; i < TC_BITMASK + 1; i++) {
2026 int q = netdev_get_prio_tc_map(dev, i);
2028 tc = &dev->tc_to_txq[q];
2029 if (tc->offset + tc->count > txq) {
2030 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2032 netdev_set_prio_tc_map(dev, i, 0);
2037 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2040 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2043 /* walk through the TCs and see if it falls into any of them */
2044 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2045 if ((txq - tc->offset) < tc->count)
2049 /* didn't find it, just return -1 to indicate no match */
2055 EXPORT_SYMBOL(netdev_txq_to_tc);
2058 struct static_key xps_needed __read_mostly;
2059 EXPORT_SYMBOL(xps_needed);
2060 struct static_key xps_rxqs_needed __read_mostly;
2061 EXPORT_SYMBOL(xps_rxqs_needed);
2062 static DEFINE_MUTEX(xps_map_mutex);
2063 #define xmap_dereference(P) \
2064 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2066 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2069 struct xps_map *map = NULL;
2073 map = xmap_dereference(dev_maps->attr_map[tci]);
2077 for (pos = map->len; pos--;) {
2078 if (map->queues[pos] != index)
2082 map->queues[pos] = map->queues[--map->len];
2086 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2087 kfree_rcu(map, rcu);
2094 static bool remove_xps_queue_cpu(struct net_device *dev,
2095 struct xps_dev_maps *dev_maps,
2096 int cpu, u16 offset, u16 count)
2098 int num_tc = dev->num_tc ? : 1;
2099 bool active = false;
2102 for (tci = cpu * num_tc; num_tc--; tci++) {
2105 for (i = count, j = offset; i--; j++) {
2106 if (!remove_xps_queue(dev_maps, tci, j))
2116 static void reset_xps_maps(struct net_device *dev,
2117 struct xps_dev_maps *dev_maps,
2121 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2122 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2124 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2126 static_key_slow_dec_cpuslocked(&xps_needed);
2127 kfree_rcu(dev_maps, rcu);
2130 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask,
2131 struct xps_dev_maps *dev_maps, unsigned int nr_ids,
2132 u16 offset, u16 count, bool is_rxqs_map)
2134 bool active = false;
2137 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids),
2139 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset,
2142 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2145 for (i = offset + (count - 1); count--; i--) {
2146 netdev_queue_numa_node_write(
2147 netdev_get_tx_queue(dev, i),
2153 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2156 const unsigned long *possible_mask = NULL;
2157 struct xps_dev_maps *dev_maps;
2158 unsigned int nr_ids;
2160 if (!static_key_false(&xps_needed))
2164 mutex_lock(&xps_map_mutex);
2166 if (static_key_false(&xps_rxqs_needed)) {
2167 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2169 nr_ids = dev->num_rx_queues;
2170 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids,
2171 offset, count, true);
2175 dev_maps = xmap_dereference(dev->xps_cpus_map);
2179 if (num_possible_cpus() > 1)
2180 possible_mask = cpumask_bits(cpu_possible_mask);
2181 nr_ids = nr_cpu_ids;
2182 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count,
2186 mutex_unlock(&xps_map_mutex);
2190 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2192 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2195 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2196 u16 index, bool is_rxqs_map)
2198 struct xps_map *new_map;
2199 int alloc_len = XPS_MIN_MAP_ALLOC;
2202 for (pos = 0; map && pos < map->len; pos++) {
2203 if (map->queues[pos] != index)
2208 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2210 if (pos < map->alloc_len)
2213 alloc_len = map->alloc_len * 2;
2216 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2220 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2222 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2223 cpu_to_node(attr_index));
2227 for (i = 0; i < pos; i++)
2228 new_map->queues[i] = map->queues[i];
2229 new_map->alloc_len = alloc_len;
2235 /* Must be called under cpus_read_lock */
2236 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2237 u16 index, bool is_rxqs_map)
2239 const unsigned long *online_mask = NULL, *possible_mask = NULL;
2240 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2241 int i, j, tci, numa_node_id = -2;
2242 int maps_sz, num_tc = 1, tc = 0;
2243 struct xps_map *map, *new_map;
2244 bool active = false;
2245 unsigned int nr_ids;
2247 WARN_ON_ONCE(index >= dev->num_tx_queues);
2250 /* Do not allow XPS on subordinate device directly */
2251 num_tc = dev->num_tc;
2255 /* If queue belongs to subordinate dev use its map */
2256 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2258 tc = netdev_txq_to_tc(dev, index);
2263 mutex_lock(&xps_map_mutex);
2265 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2266 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2267 nr_ids = dev->num_rx_queues;
2269 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2270 if (num_possible_cpus() > 1) {
2271 online_mask = cpumask_bits(cpu_online_mask);
2272 possible_mask = cpumask_bits(cpu_possible_mask);
2274 dev_maps = xmap_dereference(dev->xps_cpus_map);
2275 nr_ids = nr_cpu_ids;
2278 if (maps_sz < L1_CACHE_BYTES)
2279 maps_sz = L1_CACHE_BYTES;
2281 /* allocate memory for queue storage */
2282 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2285 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2286 if (!new_dev_maps) {
2287 mutex_unlock(&xps_map_mutex);
2291 tci = j * num_tc + tc;
2292 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) :
2295 map = expand_xps_map(map, j, index, is_rxqs_map);
2299 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2303 goto out_no_new_maps;
2306 /* Increment static keys at most once per type */
2307 static_key_slow_inc_cpuslocked(&xps_needed);
2309 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2312 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2314 /* copy maps belonging to foreign traffic classes */
2315 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) {
2316 /* fill in the new device map from the old device map */
2317 map = xmap_dereference(dev_maps->attr_map[tci]);
2318 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2321 /* We need to explicitly update tci as prevous loop
2322 * could break out early if dev_maps is NULL.
2324 tci = j * num_tc + tc;
2326 if (netif_attr_test_mask(j, mask, nr_ids) &&
2327 netif_attr_test_online(j, online_mask, nr_ids)) {
2328 /* add tx-queue to CPU/rx-queue maps */
2331 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2332 while ((pos < map->len) && (map->queues[pos] != index))
2335 if (pos == map->len)
2336 map->queues[map->len++] = index;
2339 if (numa_node_id == -2)
2340 numa_node_id = cpu_to_node(j);
2341 else if (numa_node_id != cpu_to_node(j))
2345 } else if (dev_maps) {
2346 /* fill in the new device map from the old device map */
2347 map = xmap_dereference(dev_maps->attr_map[tci]);
2348 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2351 /* copy maps belonging to foreign traffic classes */
2352 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2353 /* fill in the new device map from the old device map */
2354 map = xmap_dereference(dev_maps->attr_map[tci]);
2355 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2360 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps);
2362 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps);
2364 /* Cleanup old maps */
2366 goto out_no_old_maps;
2368 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2370 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2371 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2372 map = xmap_dereference(dev_maps->attr_map[tci]);
2373 if (map && map != new_map)
2374 kfree_rcu(map, rcu);
2378 kfree_rcu(dev_maps, rcu);
2381 dev_maps = new_dev_maps;
2386 /* update Tx queue numa node */
2387 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2388 (numa_node_id >= 0) ?
2389 numa_node_id : NUMA_NO_NODE);
2395 /* removes tx-queue from unused CPUs/rx-queues */
2396 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2398 for (i = tc, tci = j * num_tc; i--; tci++)
2399 active |= remove_xps_queue(dev_maps, tci, index);
2400 if (!netif_attr_test_mask(j, mask, nr_ids) ||
2401 !netif_attr_test_online(j, online_mask, nr_ids))
2402 active |= remove_xps_queue(dev_maps, tci, index);
2403 for (i = num_tc - tc, tci++; --i; tci++)
2404 active |= remove_xps_queue(dev_maps, tci, index);
2407 /* free map if not active */
2409 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2412 mutex_unlock(&xps_map_mutex);
2416 /* remove any maps that we added */
2417 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2419 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2420 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2422 xmap_dereference(dev_maps->attr_map[tci]) :
2424 if (new_map && new_map != map)
2429 mutex_unlock(&xps_map_mutex);
2431 kfree(new_dev_maps);
2434 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2436 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2442 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
2447 EXPORT_SYMBOL(netif_set_xps_queue);
2450 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2452 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2454 /* Unbind any subordinate channels */
2455 while (txq-- != &dev->_tx[0]) {
2457 netdev_unbind_sb_channel(dev, txq->sb_dev);
2461 void netdev_reset_tc(struct net_device *dev)
2464 netif_reset_xps_queues_gt(dev, 0);
2466 netdev_unbind_all_sb_channels(dev);
2468 /* Reset TC configuration of device */
2470 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2471 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2473 EXPORT_SYMBOL(netdev_reset_tc);
2475 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2477 if (tc >= dev->num_tc)
2481 netif_reset_xps_queues(dev, offset, count);
2483 dev->tc_to_txq[tc].count = count;
2484 dev->tc_to_txq[tc].offset = offset;
2487 EXPORT_SYMBOL(netdev_set_tc_queue);
2489 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2491 if (num_tc > TC_MAX_QUEUE)
2495 netif_reset_xps_queues_gt(dev, 0);
2497 netdev_unbind_all_sb_channels(dev);
2499 dev->num_tc = num_tc;
2502 EXPORT_SYMBOL(netdev_set_num_tc);
2504 void netdev_unbind_sb_channel(struct net_device *dev,
2505 struct net_device *sb_dev)
2507 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2510 netif_reset_xps_queues_gt(sb_dev, 0);
2512 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2513 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2515 while (txq-- != &dev->_tx[0]) {
2516 if (txq->sb_dev == sb_dev)
2520 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2522 int netdev_bind_sb_channel_queue(struct net_device *dev,
2523 struct net_device *sb_dev,
2524 u8 tc, u16 count, u16 offset)
2526 /* Make certain the sb_dev and dev are already configured */
2527 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2530 /* We cannot hand out queues we don't have */
2531 if ((offset + count) > dev->real_num_tx_queues)
2534 /* Record the mapping */
2535 sb_dev->tc_to_txq[tc].count = count;
2536 sb_dev->tc_to_txq[tc].offset = offset;
2538 /* Provide a way for Tx queue to find the tc_to_txq map or
2539 * XPS map for itself.
2542 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2546 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2548 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2550 /* Do not use a multiqueue device to represent a subordinate channel */
2551 if (netif_is_multiqueue(dev))
2554 /* We allow channels 1 - 32767 to be used for subordinate channels.
2555 * Channel 0 is meant to be "native" mode and used only to represent
2556 * the main root device. We allow writing 0 to reset the device back
2557 * to normal mode after being used as a subordinate channel.
2559 if (channel > S16_MAX)
2562 dev->num_tc = -channel;
2566 EXPORT_SYMBOL(netdev_set_sb_channel);
2569 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2570 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2572 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2577 disabling = txq < dev->real_num_tx_queues;
2579 if (txq < 1 || txq > dev->num_tx_queues)
2582 if (dev->reg_state == NETREG_REGISTERED ||
2583 dev->reg_state == NETREG_UNREGISTERING) {
2586 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2592 netif_setup_tc(dev, txq);
2594 dev_qdisc_change_real_num_tx(dev, txq);
2596 dev->real_num_tx_queues = txq;
2600 qdisc_reset_all_tx_gt(dev, txq);
2602 netif_reset_xps_queues_gt(dev, txq);
2606 dev->real_num_tx_queues = txq;
2611 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2615 * netif_set_real_num_rx_queues - set actual number of RX queues used
2616 * @dev: Network device
2617 * @rxq: Actual number of RX queues
2619 * This must be called either with the rtnl_lock held or before
2620 * registration of the net device. Returns 0 on success, or a
2621 * negative error code. If called before registration, it always
2624 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2628 if (rxq < 1 || rxq > dev->num_rx_queues)
2631 if (dev->reg_state == NETREG_REGISTERED) {
2634 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2640 dev->real_num_rx_queues = rxq;
2643 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2647 * netif_get_num_default_rss_queues - default number of RSS queues
2649 * This routine should set an upper limit on the number of RSS queues
2650 * used by default by multiqueue devices.
2652 int netif_get_num_default_rss_queues(void)
2654 return is_kdump_kernel() ?
2655 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2657 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2659 static void __netif_reschedule(struct Qdisc *q)
2661 struct softnet_data *sd;
2662 unsigned long flags;
2664 local_irq_save(flags);
2665 sd = this_cpu_ptr(&softnet_data);
2666 q->next_sched = NULL;
2667 *sd->output_queue_tailp = q;
2668 sd->output_queue_tailp = &q->next_sched;
2669 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2670 local_irq_restore(flags);
2673 void __netif_schedule(struct Qdisc *q)
2675 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2676 __netif_reschedule(q);
2678 EXPORT_SYMBOL(__netif_schedule);
2680 struct dev_kfree_skb_cb {
2681 enum skb_free_reason reason;
2684 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2686 return (struct dev_kfree_skb_cb *)skb->cb;
2689 void netif_schedule_queue(struct netdev_queue *txq)
2692 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2693 struct Qdisc *q = rcu_dereference(txq->qdisc);
2695 __netif_schedule(q);
2699 EXPORT_SYMBOL(netif_schedule_queue);
2701 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2703 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2707 q = rcu_dereference(dev_queue->qdisc);
2708 __netif_schedule(q);
2712 EXPORT_SYMBOL(netif_tx_wake_queue);
2714 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2716 unsigned long flags;
2721 if (likely(refcount_read(&skb->users) == 1)) {
2723 refcount_set(&skb->users, 0);
2724 } else if (likely(!refcount_dec_and_test(&skb->users))) {
2727 get_kfree_skb_cb(skb)->reason = reason;
2728 local_irq_save(flags);
2729 skb->next = __this_cpu_read(softnet_data.completion_queue);
2730 __this_cpu_write(softnet_data.completion_queue, skb);
2731 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2732 local_irq_restore(flags);
2734 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2736 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2738 if (in_irq() || irqs_disabled())
2739 __dev_kfree_skb_irq(skb, reason);
2740 else if (unlikely(reason == SKB_REASON_DROPPED))
2745 EXPORT_SYMBOL(__dev_kfree_skb_any);
2749 * netif_device_detach - mark device as removed
2750 * @dev: network device
2752 * Mark device as removed from system and therefore no longer available.
2754 void netif_device_detach(struct net_device *dev)
2756 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2757 netif_running(dev)) {
2758 netif_tx_stop_all_queues(dev);
2761 EXPORT_SYMBOL(netif_device_detach);
2764 * netif_device_attach - mark device as attached
2765 * @dev: network device
2767 * Mark device as attached from system and restart if needed.
2769 void netif_device_attach(struct net_device *dev)
2771 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2772 netif_running(dev)) {
2773 netif_tx_wake_all_queues(dev);
2774 __netdev_watchdog_up(dev);
2777 EXPORT_SYMBOL(netif_device_attach);
2780 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2781 * to be used as a distribution range.
2783 static u16 skb_tx_hash(const struct net_device *dev,
2784 const struct net_device *sb_dev,
2785 struct sk_buff *skb)
2789 u16 qcount = dev->real_num_tx_queues;
2792 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2794 qoffset = sb_dev->tc_to_txq[tc].offset;
2795 qcount = sb_dev->tc_to_txq[tc].count;
2796 if (unlikely(!qcount)) {
2797 net_warn_ratelimited("%s: invalid qcount, qoffset %u for tc %u\n",
2798 sb_dev->name, qoffset, tc);
2800 qcount = dev->real_num_tx_queues;
2804 if (skb_rx_queue_recorded(skb)) {
2805 hash = skb_get_rx_queue(skb);
2806 if (hash >= qoffset)
2808 while (unlikely(hash >= qcount))
2810 return hash + qoffset;
2813 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2816 static void skb_warn_bad_offload(const struct sk_buff *skb)
2818 static const netdev_features_t null_features;
2819 struct net_device *dev = skb->dev;
2820 const char *name = "";
2822 if (!net_ratelimit())
2826 if (dev->dev.parent)
2827 name = dev_driver_string(dev->dev.parent);
2829 name = netdev_name(dev);
2831 skb_dump(KERN_WARNING, skb, false);
2832 WARN(1, "%s: caps=(%pNF, %pNF)\n",
2833 name, dev ? &dev->features : &null_features,
2834 skb->sk ? &skb->sk->sk_route_caps : &null_features);
2838 * Invalidate hardware checksum when packet is to be mangled, and
2839 * complete checksum manually on outgoing path.
2841 int skb_checksum_help(struct sk_buff *skb)
2844 int ret = 0, offset;
2846 if (skb->ip_summed == CHECKSUM_COMPLETE)
2847 goto out_set_summed;
2849 if (unlikely(skb_shinfo(skb)->gso_size)) {
2850 skb_warn_bad_offload(skb);
2854 /* Before computing a checksum, we should make sure no frag could
2855 * be modified by an external entity : checksum could be wrong.
2857 if (skb_has_shared_frag(skb)) {
2858 ret = __skb_linearize(skb);
2863 offset = skb_checksum_start_offset(skb);
2864 BUG_ON(offset >= skb_headlen(skb));
2865 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2867 offset += skb->csum_offset;
2868 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2870 if (skb_cloned(skb) &&
2871 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2872 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2877 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2879 skb->ip_summed = CHECKSUM_NONE;
2883 EXPORT_SYMBOL(skb_checksum_help);
2885 int skb_crc32c_csum_help(struct sk_buff *skb)
2888 int ret = 0, offset, start;
2890 if (skb->ip_summed != CHECKSUM_PARTIAL)
2893 if (unlikely(skb_is_gso(skb)))
2896 /* Before computing a checksum, we should make sure no frag could
2897 * be modified by an external entity : checksum could be wrong.
2899 if (unlikely(skb_has_shared_frag(skb))) {
2900 ret = __skb_linearize(skb);
2904 start = skb_checksum_start_offset(skb);
2905 offset = start + offsetof(struct sctphdr, checksum);
2906 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
2910 if (skb_cloned(skb) &&
2911 !skb_clone_writable(skb, offset + sizeof(__le32))) {
2912 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2916 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
2917 skb->len - start, ~(__u32)0,
2919 *(__le32 *)(skb->data + offset) = crc32c_csum;
2920 skb->ip_summed = CHECKSUM_NONE;
2921 skb->csum_not_inet = 0;
2926 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2928 __be16 type = skb->protocol;
2930 /* Tunnel gso handlers can set protocol to ethernet. */
2931 if (type == htons(ETH_P_TEB)) {
2934 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2937 eth = (struct ethhdr *)skb->data;
2938 type = eth->h_proto;
2941 return vlan_get_protocol_and_depth(skb, type, depth);
2945 * skb_mac_gso_segment - mac layer segmentation handler.
2946 * @skb: buffer to segment
2947 * @features: features for the output path (see dev->features)
2949 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2950 netdev_features_t features)
2952 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2953 struct packet_offload *ptype;
2954 int vlan_depth = skb->mac_len;
2955 __be16 type = skb_network_protocol(skb, &vlan_depth);
2957 if (unlikely(!type))
2958 return ERR_PTR(-EINVAL);
2960 __skb_pull(skb, vlan_depth);
2963 list_for_each_entry_rcu(ptype, &offload_base, list) {
2964 if (ptype->type == type && ptype->callbacks.gso_segment) {
2965 segs = ptype->callbacks.gso_segment(skb, features);
2971 __skb_push(skb, skb->data - skb_mac_header(skb));
2975 EXPORT_SYMBOL(skb_mac_gso_segment);
2978 /* openvswitch calls this on rx path, so we need a different check.
2980 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2983 return skb->ip_summed != CHECKSUM_PARTIAL &&
2984 skb->ip_summed != CHECKSUM_UNNECESSARY;
2986 return skb->ip_summed == CHECKSUM_NONE;
2990 * __skb_gso_segment - Perform segmentation on skb.
2991 * @skb: buffer to segment
2992 * @features: features for the output path (see dev->features)
2993 * @tx_path: whether it is called in TX path
2995 * This function segments the given skb and returns a list of segments.
2997 * It may return NULL if the skb requires no segmentation. This is
2998 * only possible when GSO is used for verifying header integrity.
3000 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
3002 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3003 netdev_features_t features, bool tx_path)
3005 struct sk_buff *segs;
3007 if (unlikely(skb_needs_check(skb, tx_path))) {
3010 /* We're going to init ->check field in TCP or UDP header */
3011 err = skb_cow_head(skb, 0);
3013 return ERR_PTR(err);
3016 /* Only report GSO partial support if it will enable us to
3017 * support segmentation on this frame without needing additional
3020 if (features & NETIF_F_GSO_PARTIAL) {
3021 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3022 struct net_device *dev = skb->dev;
3024 partial_features |= dev->features & dev->gso_partial_features;
3025 if (!skb_gso_ok(skb, features | partial_features))
3026 features &= ~NETIF_F_GSO_PARTIAL;
3029 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
3030 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3032 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3033 SKB_GSO_CB(skb)->encap_level = 0;
3035 skb_reset_mac_header(skb);
3036 skb_reset_mac_len(skb);
3038 segs = skb_mac_gso_segment(skb, features);
3040 if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3041 skb_warn_bad_offload(skb);
3045 EXPORT_SYMBOL(__skb_gso_segment);
3047 /* Take action when hardware reception checksum errors are detected. */
3049 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3051 if (net_ratelimit()) {
3052 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3053 skb_dump(KERN_ERR, skb, true);
3057 EXPORT_SYMBOL(netdev_rx_csum_fault);
3060 /* XXX: check that highmem exists at all on the given machine. */
3061 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3063 #ifdef CONFIG_HIGHMEM
3066 if (!(dev->features & NETIF_F_HIGHDMA)) {
3067 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3068 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3070 if (PageHighMem(skb_frag_page(frag)))
3078 /* If MPLS offload request, verify we are testing hardware MPLS features
3079 * instead of standard features for the netdev.
3081 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3082 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3083 netdev_features_t features,
3086 if (eth_p_mpls(type))
3087 features &= skb->dev->mpls_features;
3092 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3093 netdev_features_t features,
3100 static netdev_features_t harmonize_features(struct sk_buff *skb,
3101 netdev_features_t features)
3106 type = skb_network_protocol(skb, &tmp);
3107 features = net_mpls_features(skb, features, type);
3109 if (skb->ip_summed != CHECKSUM_NONE &&
3110 !can_checksum_protocol(features, type)) {
3111 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3113 if (illegal_highdma(skb->dev, skb))
3114 features &= ~NETIF_F_SG;
3119 netdev_features_t passthru_features_check(struct sk_buff *skb,
3120 struct net_device *dev,
3121 netdev_features_t features)
3125 EXPORT_SYMBOL(passthru_features_check);
3127 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3128 struct net_device *dev,
3129 netdev_features_t features)
3131 return vlan_features_check(skb, features);
3134 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3135 struct net_device *dev,
3136 netdev_features_t features)
3138 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3140 if (gso_segs > dev->gso_max_segs)
3141 return features & ~NETIF_F_GSO_MASK;
3143 /* Support for GSO partial features requires software
3144 * intervention before we can actually process the packets
3145 * so we need to strip support for any partial features now
3146 * and we can pull them back in after we have partially
3147 * segmented the frame.
3149 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3150 features &= ~dev->gso_partial_features;
3152 /* Make sure to clear the IPv4 ID mangling feature if the
3153 * IPv4 header has the potential to be fragmented.
3155 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3156 struct iphdr *iph = skb->encapsulation ?
3157 inner_ip_hdr(skb) : ip_hdr(skb);
3159 if (!(iph->frag_off & htons(IP_DF)))
3160 features &= ~NETIF_F_TSO_MANGLEID;
3166 netdev_features_t netif_skb_features(struct sk_buff *skb)
3168 struct net_device *dev = skb->dev;
3169 netdev_features_t features = dev->features;
3171 if (skb_is_gso(skb))
3172 features = gso_features_check(skb, dev, features);
3174 /* If encapsulation offload request, verify we are testing
3175 * hardware encapsulation features instead of standard
3176 * features for the netdev
3178 if (skb->encapsulation)
3179 features &= dev->hw_enc_features;
3181 if (skb_vlan_tagged(skb))
3182 features = netdev_intersect_features(features,
3183 dev->vlan_features |
3184 NETIF_F_HW_VLAN_CTAG_TX |
3185 NETIF_F_HW_VLAN_STAG_TX);
3187 if (dev->netdev_ops->ndo_features_check)
3188 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3191 features &= dflt_features_check(skb, dev, features);
3193 return harmonize_features(skb, features);
3195 EXPORT_SYMBOL(netif_skb_features);
3197 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3198 struct netdev_queue *txq, bool more)
3203 if (dev_nit_active(dev))
3204 dev_queue_xmit_nit(skb, dev);
3207 trace_net_dev_start_xmit(skb, dev);
3208 rc = netdev_start_xmit(skb, dev, txq, more);
3209 trace_net_dev_xmit(skb, rc, dev, len);
3214 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3215 struct netdev_queue *txq, int *ret)
3217 struct sk_buff *skb = first;
3218 int rc = NETDEV_TX_OK;
3221 struct sk_buff *next = skb->next;
3223 skb_mark_not_on_list(skb);
3224 rc = xmit_one(skb, dev, txq, next != NULL);
3225 if (unlikely(!dev_xmit_complete(rc))) {
3231 if (netif_tx_queue_stopped(txq) && skb) {
3232 rc = NETDEV_TX_BUSY;
3242 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3243 netdev_features_t features)
3245 if (skb_vlan_tag_present(skb) &&
3246 !vlan_hw_offload_capable(features, skb->vlan_proto))
3247 skb = __vlan_hwaccel_push_inside(skb);
3251 int skb_csum_hwoffload_help(struct sk_buff *skb,
3252 const netdev_features_t features)
3254 if (unlikely(skb->csum_not_inet))
3255 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3256 skb_crc32c_csum_help(skb);
3258 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3260 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3262 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3264 netdev_features_t features;
3266 features = netif_skb_features(skb);
3267 skb = validate_xmit_vlan(skb, features);
3271 skb = sk_validate_xmit_skb(skb, dev);
3275 if (netif_needs_gso(skb, features)) {
3276 struct sk_buff *segs;
3278 segs = skb_gso_segment(skb, features);
3286 if (skb_needs_linearize(skb, features) &&
3287 __skb_linearize(skb))
3290 /* If packet is not checksummed and device does not
3291 * support checksumming for this protocol, complete
3292 * checksumming here.
3294 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3295 if (skb->encapsulation)
3296 skb_set_inner_transport_header(skb,
3297 skb_checksum_start_offset(skb));
3299 skb_set_transport_header(skb,
3300 skb_checksum_start_offset(skb));
3301 if (skb_csum_hwoffload_help(skb, features))
3306 skb = validate_xmit_xfrm(skb, features, again);
3313 atomic_long_inc(&dev->tx_dropped);
3317 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3319 struct sk_buff *next, *head = NULL, *tail;
3321 for (; skb != NULL; skb = next) {
3323 skb_mark_not_on_list(skb);
3325 /* in case skb wont be segmented, point to itself */
3328 skb = validate_xmit_skb(skb, dev, again);
3336 /* If skb was segmented, skb->prev points to
3337 * the last segment. If not, it still contains skb.
3343 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3345 static void qdisc_pkt_len_init(struct sk_buff *skb)
3347 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3349 qdisc_skb_cb(skb)->pkt_len = skb->len;
3351 /* To get more precise estimation of bytes sent on wire,
3352 * we add to pkt_len the headers size of all segments
3354 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3355 unsigned int hdr_len;
3356 u16 gso_segs = shinfo->gso_segs;
3358 /* mac layer + network layer */
3359 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3361 /* + transport layer */
3362 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3363 const struct tcphdr *th;
3364 struct tcphdr _tcphdr;
3366 th = skb_header_pointer(skb, skb_transport_offset(skb),
3367 sizeof(_tcphdr), &_tcphdr);
3369 hdr_len += __tcp_hdrlen(th);
3371 struct udphdr _udphdr;
3373 if (skb_header_pointer(skb, skb_transport_offset(skb),
3374 sizeof(_udphdr), &_udphdr))
3375 hdr_len += sizeof(struct udphdr);
3378 if (shinfo->gso_type & SKB_GSO_DODGY)
3379 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3382 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3386 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3387 struct net_device *dev,
3388 struct netdev_queue *txq)
3390 spinlock_t *root_lock = qdisc_lock(q);
3391 struct sk_buff *to_free = NULL;
3395 qdisc_calculate_pkt_len(skb, q);
3397 if (q->flags & TCQ_F_NOLOCK) {
3398 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3399 if (likely(!netif_xmit_frozen_or_stopped(txq)))
3402 if (unlikely(to_free))
3403 kfree_skb_list(to_free);
3408 * Heuristic to force contended enqueues to serialize on a
3409 * separate lock before trying to get qdisc main lock.
3410 * This permits qdisc->running owner to get the lock more
3411 * often and dequeue packets faster.
3413 contended = qdisc_is_running(q);
3414 if (unlikely(contended))
3415 spin_lock(&q->busylock);
3417 spin_lock(root_lock);
3418 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3419 __qdisc_drop(skb, &to_free);
3421 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3422 qdisc_run_begin(q)) {
3424 * This is a work-conserving queue; there are no old skbs
3425 * waiting to be sent out; and the qdisc is not running -
3426 * xmit the skb directly.
3429 qdisc_bstats_update(q, skb);
3431 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3432 if (unlikely(contended)) {
3433 spin_unlock(&q->busylock);
3440 rc = NET_XMIT_SUCCESS;
3442 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3443 if (qdisc_run_begin(q)) {
3444 if (unlikely(contended)) {
3445 spin_unlock(&q->busylock);
3452 spin_unlock(root_lock);
3453 if (unlikely(to_free))
3454 kfree_skb_list(to_free);
3455 if (unlikely(contended))
3456 spin_unlock(&q->busylock);
3460 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3461 static void skb_update_prio(struct sk_buff *skb)
3463 const struct netprio_map *map;
3464 const struct sock *sk;
3465 unsigned int prioidx;
3469 map = rcu_dereference_bh(skb->dev->priomap);
3472 sk = skb_to_full_sk(skb);
3476 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3478 if (prioidx < map->priomap_len)
3479 skb->priority = map->priomap[prioidx];
3482 #define skb_update_prio(skb)
3486 * dev_loopback_xmit - loop back @skb
3487 * @net: network namespace this loopback is happening in
3488 * @sk: sk needed to be a netfilter okfn
3489 * @skb: buffer to transmit
3491 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3493 skb_reset_mac_header(skb);
3494 __skb_pull(skb, skb_network_offset(skb));
3495 skb->pkt_type = PACKET_LOOPBACK;
3496 if (skb->ip_summed == CHECKSUM_NONE)
3497 skb->ip_summed = CHECKSUM_UNNECESSARY;
3498 WARN_ON(!skb_dst(skb));
3503 EXPORT_SYMBOL(dev_loopback_xmit);
3505 #ifdef CONFIG_NET_EGRESS
3506 static struct sk_buff *
3507 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3509 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3510 struct tcf_result cl_res;
3515 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3516 mini_qdisc_bstats_cpu_update(miniq, skb);
3518 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3520 case TC_ACT_RECLASSIFY:
3521 skb->tc_index = TC_H_MIN(cl_res.classid);
3524 mini_qdisc_qstats_cpu_drop(miniq);
3525 *ret = NET_XMIT_DROP;
3531 *ret = NET_XMIT_SUCCESS;
3534 case TC_ACT_REDIRECT:
3535 /* No need to push/pop skb's mac_header here on egress! */
3536 skb_do_redirect(skb);
3537 *ret = NET_XMIT_SUCCESS;
3545 #endif /* CONFIG_NET_EGRESS */
3548 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3549 struct xps_dev_maps *dev_maps, unsigned int tci)
3551 struct xps_map *map;
3552 int queue_index = -1;
3556 tci += netdev_get_prio_tc_map(dev, skb->priority);
3559 map = rcu_dereference(dev_maps->attr_map[tci]);
3562 queue_index = map->queues[0];
3564 queue_index = map->queues[reciprocal_scale(
3565 skb_get_hash(skb), map->len)];
3566 if (unlikely(queue_index >= dev->real_num_tx_queues))
3573 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3574 struct sk_buff *skb)
3577 struct xps_dev_maps *dev_maps;
3578 struct sock *sk = skb->sk;
3579 int queue_index = -1;
3581 if (!static_key_false(&xps_needed))
3585 if (!static_key_false(&xps_rxqs_needed))
3588 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map);
3590 int tci = sk_rx_queue_get(sk);
3592 if (tci >= 0 && tci < dev->num_rx_queues)
3593 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3598 if (queue_index < 0) {
3599 dev_maps = rcu_dereference(sb_dev->xps_cpus_map);
3601 unsigned int tci = skb->sender_cpu - 1;
3603 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3615 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3616 struct net_device *sb_dev)
3620 EXPORT_SYMBOL(dev_pick_tx_zero);
3622 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
3623 struct net_device *sb_dev)
3625 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3627 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3629 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
3630 struct net_device *sb_dev)
3632 struct sock *sk = skb->sk;
3633 int queue_index = sk_tx_queue_get(sk);
3635 sb_dev = sb_dev ? : dev;
3637 if (queue_index < 0 || skb->ooo_okay ||
3638 queue_index >= dev->real_num_tx_queues) {
3639 int new_index = get_xps_queue(dev, sb_dev, skb);
3642 new_index = skb_tx_hash(dev, sb_dev, skb);
3644 if (queue_index != new_index && sk &&
3646 rcu_access_pointer(sk->sk_dst_cache))
3647 sk_tx_queue_set(sk, new_index);
3649 queue_index = new_index;
3654 EXPORT_SYMBOL(netdev_pick_tx);
3656 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
3657 struct sk_buff *skb,
3658 struct net_device *sb_dev)
3660 int queue_index = 0;
3663 u32 sender_cpu = skb->sender_cpu - 1;
3665 if (sender_cpu >= (u32)NR_CPUS)
3666 skb->sender_cpu = raw_smp_processor_id() + 1;
3669 if (dev->real_num_tx_queues != 1) {
3670 const struct net_device_ops *ops = dev->netdev_ops;
3672 if (ops->ndo_select_queue)
3673 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
3675 queue_index = netdev_pick_tx(dev, skb, sb_dev);
3677 queue_index = netdev_cap_txqueue(dev, queue_index);
3680 skb_set_queue_mapping(skb, queue_index);
3681 return netdev_get_tx_queue(dev, queue_index);
3685 * __dev_queue_xmit - transmit a buffer
3686 * @skb: buffer to transmit
3687 * @sb_dev: suboordinate device used for L2 forwarding offload
3689 * Queue a buffer for transmission to a network device. The caller must
3690 * have set the device and priority and built the buffer before calling
3691 * this function. The function can be called from an interrupt.
3693 * A negative errno code is returned on a failure. A success does not
3694 * guarantee the frame will be transmitted as it may be dropped due
3695 * to congestion or traffic shaping.
3697 * -----------------------------------------------------------------------------------
3698 * I notice this method can also return errors from the queue disciplines,
3699 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3702 * Regardless of the return value, the skb is consumed, so it is currently
3703 * difficult to retry a send to this method. (You can bump the ref count
3704 * before sending to hold a reference for retry if you are careful.)
3706 * When calling this method, interrupts MUST be enabled. This is because
3707 * the BH enable code must have IRQs enabled so that it will not deadlock.
3710 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
3712 struct net_device *dev = skb->dev;
3713 struct netdev_queue *txq;
3718 skb_reset_mac_header(skb);
3719 skb_assert_len(skb);
3721 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3722 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3724 /* Disable soft irqs for various locks below. Also
3725 * stops preemption for RCU.
3729 skb_update_prio(skb);
3731 qdisc_pkt_len_init(skb);
3732 #ifdef CONFIG_NET_CLS_ACT
3733 skb->tc_at_ingress = 0;
3734 # ifdef CONFIG_NET_EGRESS
3735 if (static_branch_unlikely(&egress_needed_key)) {
3736 skb = sch_handle_egress(skb, &rc, dev);
3742 /* If device/qdisc don't need skb->dst, release it right now while
3743 * its hot in this cpu cache.
3745 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3750 txq = netdev_core_pick_tx(dev, skb, sb_dev);
3751 q = rcu_dereference_bh(txq->qdisc);
3753 trace_net_dev_queue(skb);
3755 rc = __dev_xmit_skb(skb, q, dev, txq);
3759 /* The device has no queue. Common case for software devices:
3760 * loopback, all the sorts of tunnels...
3762 * Really, it is unlikely that netif_tx_lock protection is necessary
3763 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
3765 * However, it is possible, that they rely on protection
3768 * Check this and shot the lock. It is not prone from deadlocks.
3769 *Either shot noqueue qdisc, it is even simpler 8)
3771 if (dev->flags & IFF_UP) {
3772 int cpu = smp_processor_id(); /* ok because BHs are off */
3774 /* Other cpus might concurrently change txq->xmit_lock_owner
3775 * to -1 or to their cpu id, but not to our id.
3777 if (READ_ONCE(txq->xmit_lock_owner) != cpu) {
3778 if (dev_xmit_recursion())
3779 goto recursion_alert;
3781 skb = validate_xmit_skb(skb, dev, &again);
3785 HARD_TX_LOCK(dev, txq, cpu);
3787 if (!netif_xmit_stopped(txq)) {
3788 dev_xmit_recursion_inc();
3789 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3790 dev_xmit_recursion_dec();
3791 if (dev_xmit_complete(rc)) {
3792 HARD_TX_UNLOCK(dev, txq);
3796 HARD_TX_UNLOCK(dev, txq);
3797 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3800 /* Recursion is detected! It is possible,
3804 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3810 rcu_read_unlock_bh();
3812 atomic_long_inc(&dev->tx_dropped);
3813 kfree_skb_list(skb);
3816 rcu_read_unlock_bh();
3820 int dev_queue_xmit(struct sk_buff *skb)
3822 return __dev_queue_xmit(skb, NULL);
3824 EXPORT_SYMBOL(dev_queue_xmit);
3826 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
3828 return __dev_queue_xmit(skb, sb_dev);
3830 EXPORT_SYMBOL(dev_queue_xmit_accel);
3832 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
3834 struct net_device *dev = skb->dev;
3835 struct sk_buff *orig_skb = skb;
3836 struct netdev_queue *txq;
3837 int ret = NETDEV_TX_BUSY;
3840 if (unlikely(!netif_running(dev) ||
3841 !netif_carrier_ok(dev)))
3844 skb = validate_xmit_skb_list(skb, dev, &again);
3845 if (skb != orig_skb)
3848 skb_set_queue_mapping(skb, queue_id);
3849 txq = skb_get_tx_queue(dev, skb);
3853 dev_xmit_recursion_inc();
3854 HARD_TX_LOCK(dev, txq, smp_processor_id());
3855 if (!netif_xmit_frozen_or_drv_stopped(txq))
3856 ret = netdev_start_xmit(skb, dev, txq, false);
3857 HARD_TX_UNLOCK(dev, txq);
3858 dev_xmit_recursion_dec();
3862 if (!dev_xmit_complete(ret))
3867 atomic_long_inc(&dev->tx_dropped);
3868 kfree_skb_list(skb);
3869 return NET_XMIT_DROP;
3871 EXPORT_SYMBOL(dev_direct_xmit);
3873 /*************************************************************************
3875 *************************************************************************/
3877 int netdev_max_backlog __read_mostly = 1000;
3878 EXPORT_SYMBOL(netdev_max_backlog);
3880 int netdev_tstamp_prequeue __read_mostly = 1;
3881 int netdev_budget __read_mostly = 300;
3882 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
3883 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
3884 int weight_p __read_mostly = 64; /* old backlog weight */
3885 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
3886 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
3887 int dev_rx_weight __read_mostly = 64;
3888 int dev_tx_weight __read_mostly = 64;
3889 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
3890 int gro_normal_batch __read_mostly = 8;
3892 /* Called with irq disabled */
3893 static inline void ____napi_schedule(struct softnet_data *sd,
3894 struct napi_struct *napi)
3896 list_add_tail(&napi->poll_list, &sd->poll_list);
3897 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3902 /* One global table that all flow-based protocols share. */
3903 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3904 EXPORT_SYMBOL(rps_sock_flow_table);
3905 u32 rps_cpu_mask __read_mostly;
3906 EXPORT_SYMBOL(rps_cpu_mask);
3908 struct static_key_false rps_needed __read_mostly;
3909 EXPORT_SYMBOL(rps_needed);
3910 struct static_key_false rfs_needed __read_mostly;
3911 EXPORT_SYMBOL(rfs_needed);
3913 static struct rps_dev_flow *
3914 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3915 struct rps_dev_flow *rflow, u16 next_cpu)
3917 if (next_cpu < nr_cpu_ids) {
3918 #ifdef CONFIG_RFS_ACCEL
3919 struct netdev_rx_queue *rxqueue;
3920 struct rps_dev_flow_table *flow_table;
3921 struct rps_dev_flow *old_rflow;
3926 /* Should we steer this flow to a different hardware queue? */
3927 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3928 !(dev->features & NETIF_F_NTUPLE))
3930 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3931 if (rxq_index == skb_get_rx_queue(skb))
3934 rxqueue = dev->_rx + rxq_index;
3935 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3938 flow_id = skb_get_hash(skb) & flow_table->mask;
3939 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3940 rxq_index, flow_id);
3944 rflow = &flow_table->flows[flow_id];
3946 if (old_rflow->filter == rflow->filter)
3947 old_rflow->filter = RPS_NO_FILTER;
3951 per_cpu(softnet_data, next_cpu).input_queue_head;
3954 rflow->cpu = next_cpu;
3959 * get_rps_cpu is called from netif_receive_skb and returns the target
3960 * CPU from the RPS map of the receiving queue for a given skb.
3961 * rcu_read_lock must be held on entry.
3963 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3964 struct rps_dev_flow **rflowp)
3966 const struct rps_sock_flow_table *sock_flow_table;
3967 struct netdev_rx_queue *rxqueue = dev->_rx;
3968 struct rps_dev_flow_table *flow_table;
3969 struct rps_map *map;
3974 if (skb_rx_queue_recorded(skb)) {
3975 u16 index = skb_get_rx_queue(skb);
3977 if (unlikely(index >= dev->real_num_rx_queues)) {
3978 WARN_ONCE(dev->real_num_rx_queues > 1,
3979 "%s received packet on queue %u, but number "
3980 "of RX queues is %u\n",
3981 dev->name, index, dev->real_num_rx_queues);
3987 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3989 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3990 map = rcu_dereference(rxqueue->rps_map);
3991 if (!flow_table && !map)
3994 skb_reset_network_header(skb);
3995 hash = skb_get_hash(skb);
3999 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4000 if (flow_table && sock_flow_table) {
4001 struct rps_dev_flow *rflow;
4005 /* First check into global flow table if there is a match.
4006 * This READ_ONCE() pairs with WRITE_ONCE() from rps_record_sock_flow().
4008 ident = READ_ONCE(sock_flow_table->ents[hash & sock_flow_table->mask]);
4009 if ((ident ^ hash) & ~rps_cpu_mask)
4012 next_cpu = ident & rps_cpu_mask;
4014 /* OK, now we know there is a match,
4015 * we can look at the local (per receive queue) flow table
4017 rflow = &flow_table->flows[hash & flow_table->mask];
4021 * If the desired CPU (where last recvmsg was done) is
4022 * different from current CPU (one in the rx-queue flow
4023 * table entry), switch if one of the following holds:
4024 * - Current CPU is unset (>= nr_cpu_ids).
4025 * - Current CPU is offline.
4026 * - The current CPU's queue tail has advanced beyond the
4027 * last packet that was enqueued using this table entry.
4028 * This guarantees that all previous packets for the flow
4029 * have been dequeued, thus preserving in order delivery.
4031 if (unlikely(tcpu != next_cpu) &&
4032 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4033 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4034 rflow->last_qtail)) >= 0)) {
4036 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4039 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4049 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4050 if (cpu_online(tcpu)) {
4060 #ifdef CONFIG_RFS_ACCEL
4063 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4064 * @dev: Device on which the filter was set
4065 * @rxq_index: RX queue index
4066 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4067 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4069 * Drivers that implement ndo_rx_flow_steer() should periodically call
4070 * this function for each installed filter and remove the filters for
4071 * which it returns %true.
4073 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4074 u32 flow_id, u16 filter_id)
4076 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4077 struct rps_dev_flow_table *flow_table;
4078 struct rps_dev_flow *rflow;
4083 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4084 if (flow_table && flow_id <= flow_table->mask) {
4085 rflow = &flow_table->flows[flow_id];
4086 cpu = READ_ONCE(rflow->cpu);
4087 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4088 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4089 rflow->last_qtail) <
4090 (int)(10 * flow_table->mask)))
4096 EXPORT_SYMBOL(rps_may_expire_flow);
4098 #endif /* CONFIG_RFS_ACCEL */
4100 /* Called from hardirq (IPI) context */
4101 static void rps_trigger_softirq(void *data)
4103 struct softnet_data *sd = data;
4105 ____napi_schedule(sd, &sd->backlog);
4109 #endif /* CONFIG_RPS */
4112 * Check if this softnet_data structure is another cpu one
4113 * If yes, queue it to our IPI list and return 1
4116 static int rps_ipi_queued(struct softnet_data *sd)
4119 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4122 sd->rps_ipi_next = mysd->rps_ipi_list;
4123 mysd->rps_ipi_list = sd;
4125 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4128 #endif /* CONFIG_RPS */
4132 #ifdef CONFIG_NET_FLOW_LIMIT
4133 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4136 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4138 #ifdef CONFIG_NET_FLOW_LIMIT
4139 struct sd_flow_limit *fl;
4140 struct softnet_data *sd;
4141 unsigned int old_flow, new_flow;
4143 if (qlen < (netdev_max_backlog >> 1))
4146 sd = this_cpu_ptr(&softnet_data);
4149 fl = rcu_dereference(sd->flow_limit);
4151 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4152 old_flow = fl->history[fl->history_head];
4153 fl->history[fl->history_head] = new_flow;
4156 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4158 if (likely(fl->buckets[old_flow]))
4159 fl->buckets[old_flow]--;
4161 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4173 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4174 * queue (may be a remote CPU queue).
4176 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4177 unsigned int *qtail)
4179 struct softnet_data *sd;
4180 unsigned long flags;
4183 sd = &per_cpu(softnet_data, cpu);
4185 local_irq_save(flags);
4188 if (!netif_running(skb->dev))
4190 qlen = skb_queue_len(&sd->input_pkt_queue);
4191 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4194 __skb_queue_tail(&sd->input_pkt_queue, skb);
4195 input_queue_tail_incr_save(sd, qtail);
4197 local_irq_restore(flags);
4198 return NET_RX_SUCCESS;
4201 /* Schedule NAPI for backlog device
4202 * We can use non atomic operation since we own the queue lock
4204 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4205 if (!rps_ipi_queued(sd))
4206 ____napi_schedule(sd, &sd->backlog);
4215 local_irq_restore(flags);
4217 atomic_long_inc(&skb->dev->rx_dropped);
4222 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4224 struct net_device *dev = skb->dev;
4225 struct netdev_rx_queue *rxqueue;
4229 if (skb_rx_queue_recorded(skb)) {
4230 u16 index = skb_get_rx_queue(skb);
4232 if (unlikely(index >= dev->real_num_rx_queues)) {
4233 WARN_ONCE(dev->real_num_rx_queues > 1,
4234 "%s received packet on queue %u, but number "
4235 "of RX queues is %u\n",
4236 dev->name, index, dev->real_num_rx_queues);
4238 return rxqueue; /* Return first rxqueue */
4245 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4246 struct xdp_buff *xdp,
4247 struct bpf_prog *xdp_prog)
4249 struct netdev_rx_queue *rxqueue;
4250 void *orig_data, *orig_data_end;
4251 u32 metalen, act = XDP_DROP;
4252 __be16 orig_eth_type;
4258 /* Reinjected packets coming from act_mirred or similar should
4259 * not get XDP generic processing.
4261 if (skb_is_redirected(skb))
4264 /* XDP packets must be linear and must have sufficient headroom
4265 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4266 * native XDP provides, thus we need to do it here as well.
4268 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4269 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4270 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4271 int troom = skb->tail + skb->data_len - skb->end;
4273 /* In case we have to go down the path and also linearize,
4274 * then lets do the pskb_expand_head() work just once here.
4276 if (pskb_expand_head(skb,
4277 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4278 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4280 if (skb_linearize(skb))
4284 /* The XDP program wants to see the packet starting at the MAC
4287 mac_len = skb->data - skb_mac_header(skb);
4288 hlen = skb_headlen(skb) + mac_len;
4289 xdp->data = skb->data - mac_len;
4290 xdp->data_meta = xdp->data;
4291 xdp->data_end = xdp->data + hlen;
4292 xdp->data_hard_start = skb->data - skb_headroom(skb);
4293 orig_data_end = xdp->data_end;
4294 orig_data = xdp->data;
4295 eth = (struct ethhdr *)xdp->data;
4296 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4297 orig_eth_type = eth->h_proto;
4299 rxqueue = netif_get_rxqueue(skb);
4300 xdp->rxq = &rxqueue->xdp_rxq;
4302 act = bpf_prog_run_xdp(xdp_prog, xdp);
4304 /* check if bpf_xdp_adjust_head was used */
4305 off = xdp->data - orig_data;
4308 __skb_pull(skb, off);
4310 __skb_push(skb, -off);
4312 skb->mac_header += off;
4313 skb_reset_network_header(skb);
4316 /* check if bpf_xdp_adjust_tail was used. it can only "shrink"
4319 off = orig_data_end - xdp->data_end;
4321 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4326 /* check if XDP changed eth hdr such SKB needs update */
4327 eth = (struct ethhdr *)xdp->data;
4328 if ((orig_eth_type != eth->h_proto) ||
4329 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4330 __skb_push(skb, ETH_HLEN);
4331 skb->protocol = eth_type_trans(skb, skb->dev);
4337 __skb_push(skb, mac_len);
4340 metalen = xdp->data - xdp->data_meta;
4342 skb_metadata_set(skb, metalen);
4345 bpf_warn_invalid_xdp_action(act);
4348 trace_xdp_exception(skb->dev, xdp_prog, act);
4359 /* When doing generic XDP we have to bypass the qdisc layer and the
4360 * network taps in order to match in-driver-XDP behavior.
4362 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4364 struct net_device *dev = skb->dev;
4365 struct netdev_queue *txq;
4366 bool free_skb = true;
4369 txq = netdev_core_pick_tx(dev, skb, NULL);
4370 cpu = smp_processor_id();
4371 HARD_TX_LOCK(dev, txq, cpu);
4372 if (!netif_xmit_stopped(txq)) {
4373 rc = netdev_start_xmit(skb, dev, txq, 0);
4374 if (dev_xmit_complete(rc))
4377 HARD_TX_UNLOCK(dev, txq);
4379 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4383 EXPORT_SYMBOL_GPL(generic_xdp_tx);
4385 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4387 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4390 struct xdp_buff xdp;
4394 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4395 if (act != XDP_PASS) {
4398 err = xdp_do_generic_redirect(skb->dev, skb,
4404 generic_xdp_tx(skb, xdp_prog);
4415 EXPORT_SYMBOL_GPL(do_xdp_generic);
4417 static int netif_rx_internal(struct sk_buff *skb)
4421 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
4423 trace_netif_rx(skb);
4426 if (static_branch_unlikely(&rps_needed)) {
4427 struct rps_dev_flow voidflow, *rflow = &voidflow;
4433 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4435 cpu = smp_processor_id();
4437 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4446 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4453 * netif_rx - post buffer to the network code
4454 * @skb: buffer to post
4456 * This function receives a packet from a device driver and queues it for
4457 * the upper (protocol) levels to process. It always succeeds. The buffer
4458 * may be dropped during processing for congestion control or by the
4462 * NET_RX_SUCCESS (no congestion)
4463 * NET_RX_DROP (packet was dropped)
4467 int netif_rx(struct sk_buff *skb)
4471 trace_netif_rx_entry(skb);
4473 ret = netif_rx_internal(skb);
4474 trace_netif_rx_exit(ret);
4478 EXPORT_SYMBOL(netif_rx);
4480 int netif_rx_ni(struct sk_buff *skb)
4484 trace_netif_rx_ni_entry(skb);
4487 err = netif_rx_internal(skb);
4488 if (local_softirq_pending())
4491 trace_netif_rx_ni_exit(err);
4495 EXPORT_SYMBOL(netif_rx_ni);
4497 static __latent_entropy void net_tx_action(struct softirq_action *h)
4499 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4501 if (sd->completion_queue) {
4502 struct sk_buff *clist;
4504 local_irq_disable();
4505 clist = sd->completion_queue;
4506 sd->completion_queue = NULL;
4510 struct sk_buff *skb = clist;
4512 clist = clist->next;
4514 WARN_ON(refcount_read(&skb->users));
4515 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4516 trace_consume_skb(skb);
4518 trace_kfree_skb(skb, net_tx_action);
4520 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4523 __kfree_skb_defer(skb);
4526 __kfree_skb_flush();
4529 if (sd->output_queue) {
4532 local_irq_disable();
4533 head = sd->output_queue;
4534 sd->output_queue = NULL;
4535 sd->output_queue_tailp = &sd->output_queue;
4541 struct Qdisc *q = head;
4542 spinlock_t *root_lock = NULL;
4544 head = head->next_sched;
4546 /* We need to make sure head->next_sched is read
4547 * before clearing __QDISC_STATE_SCHED
4549 smp_mb__before_atomic();
4551 if (!(q->flags & TCQ_F_NOLOCK)) {
4552 root_lock = qdisc_lock(q);
4553 spin_lock(root_lock);
4554 } else if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED,
4556 /* There is a synchronize_net() between
4557 * STATE_DEACTIVATED flag being set and
4558 * qdisc_reset()/some_qdisc_is_busy() in
4559 * dev_deactivate(), so we can safely bail out
4560 * early here to avoid data race between
4561 * qdisc_deactivate() and some_qdisc_is_busy()
4562 * for lockless qdisc.
4564 clear_bit(__QDISC_STATE_SCHED, &q->state);
4568 clear_bit(__QDISC_STATE_SCHED, &q->state);
4571 spin_unlock(root_lock);
4577 xfrm_dev_backlog(sd);
4580 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4581 /* This hook is defined here for ATM LANE */
4582 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4583 unsigned char *addr) __read_mostly;
4584 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4587 static inline struct sk_buff *
4588 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4589 struct net_device *orig_dev)
4591 #ifdef CONFIG_NET_CLS_ACT
4592 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4593 struct tcf_result cl_res;
4595 /* If there's at least one ingress present somewhere (so
4596 * we get here via enabled static key), remaining devices
4597 * that are not configured with an ingress qdisc will bail
4604 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4608 qdisc_skb_cb(skb)->pkt_len = skb->len;
4609 skb->tc_at_ingress = 1;
4610 mini_qdisc_bstats_cpu_update(miniq, skb);
4612 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
4614 case TC_ACT_RECLASSIFY:
4615 skb->tc_index = TC_H_MIN(cl_res.classid);
4618 mini_qdisc_qstats_cpu_drop(miniq);
4626 case TC_ACT_REDIRECT:
4627 /* skb_mac_header check was done by cls/act_bpf, so
4628 * we can safely push the L2 header back before
4629 * redirecting to another netdev
4631 __skb_push(skb, skb->mac_len);
4632 skb_do_redirect(skb);
4634 case TC_ACT_CONSUMED:
4639 #endif /* CONFIG_NET_CLS_ACT */
4644 * netdev_is_rx_handler_busy - check if receive handler is registered
4645 * @dev: device to check
4647 * Check if a receive handler is already registered for a given device.
4648 * Return true if there one.
4650 * The caller must hold the rtnl_mutex.
4652 bool netdev_is_rx_handler_busy(struct net_device *dev)
4655 return dev && rtnl_dereference(dev->rx_handler);
4657 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4660 * netdev_rx_handler_register - register receive handler
4661 * @dev: device to register a handler for
4662 * @rx_handler: receive handler to register
4663 * @rx_handler_data: data pointer that is used by rx handler
4665 * Register a receive handler for a device. This handler will then be
4666 * called from __netif_receive_skb. A negative errno code is returned
4669 * The caller must hold the rtnl_mutex.
4671 * For a general description of rx_handler, see enum rx_handler_result.
4673 int netdev_rx_handler_register(struct net_device *dev,
4674 rx_handler_func_t *rx_handler,
4675 void *rx_handler_data)
4677 if (netdev_is_rx_handler_busy(dev))
4680 if (dev->priv_flags & IFF_NO_RX_HANDLER)
4683 /* Note: rx_handler_data must be set before rx_handler */
4684 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4685 rcu_assign_pointer(dev->rx_handler, rx_handler);
4689 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4692 * netdev_rx_handler_unregister - unregister receive handler
4693 * @dev: device to unregister a handler from
4695 * Unregister a receive handler from a device.
4697 * The caller must hold the rtnl_mutex.
4699 void netdev_rx_handler_unregister(struct net_device *dev)
4703 RCU_INIT_POINTER(dev->rx_handler, NULL);
4704 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4705 * section has a guarantee to see a non NULL rx_handler_data
4709 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4711 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4714 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4715 * the special handling of PFMEMALLOC skbs.
4717 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4719 switch (skb->protocol) {
4720 case htons(ETH_P_ARP):
4721 case htons(ETH_P_IP):
4722 case htons(ETH_P_IPV6):
4723 case htons(ETH_P_8021Q):
4724 case htons(ETH_P_8021AD):
4731 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4732 int *ret, struct net_device *orig_dev)
4734 #ifdef CONFIG_NETFILTER_INGRESS
4735 if (nf_hook_ingress_active(skb)) {
4739 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4744 ingress_retval = nf_hook_ingress(skb);
4746 return ingress_retval;
4748 #endif /* CONFIG_NETFILTER_INGRESS */
4752 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
4753 struct packet_type **ppt_prev)
4755 struct packet_type *ptype, *pt_prev;
4756 rx_handler_func_t *rx_handler;
4757 struct sk_buff *skb = *pskb;
4758 struct net_device *orig_dev;
4759 bool deliver_exact = false;
4760 int ret = NET_RX_DROP;
4763 net_timestamp_check(!READ_ONCE(netdev_tstamp_prequeue), skb);
4765 trace_netif_receive_skb(skb);
4767 orig_dev = skb->dev;
4769 skb_reset_network_header(skb);
4770 if (!skb_transport_header_was_set(skb))
4771 skb_reset_transport_header(skb);
4772 skb_reset_mac_len(skb);
4777 skb->skb_iif = skb->dev->ifindex;
4779 __this_cpu_inc(softnet_data.processed);
4781 if (static_branch_unlikely(&generic_xdp_needed_key)) {
4785 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
4788 if (ret2 != XDP_PASS) {
4792 skb_reset_mac_len(skb);
4795 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4796 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4797 skb = skb_vlan_untag(skb);
4802 if (skb_skip_tc_classify(skb))
4808 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4810 ret = deliver_skb(skb, pt_prev, orig_dev);
4814 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4816 ret = deliver_skb(skb, pt_prev, orig_dev);
4821 #ifdef CONFIG_NET_INGRESS
4822 if (static_branch_unlikely(&ingress_needed_key)) {
4823 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4827 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4831 skb_reset_redirect(skb);
4833 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4836 if (skb_vlan_tag_present(skb)) {
4838 ret = deliver_skb(skb, pt_prev, orig_dev);
4841 if (vlan_do_receive(&skb))
4843 else if (unlikely(!skb))
4847 rx_handler = rcu_dereference(skb->dev->rx_handler);
4850 ret = deliver_skb(skb, pt_prev, orig_dev);
4853 switch (rx_handler(&skb)) {
4854 case RX_HANDLER_CONSUMED:
4855 ret = NET_RX_SUCCESS;
4857 case RX_HANDLER_ANOTHER:
4859 case RX_HANDLER_EXACT:
4860 deliver_exact = true;
4861 case RX_HANDLER_PASS:
4868 if (unlikely(skb_vlan_tag_present(skb))) {
4870 if (skb_vlan_tag_get_id(skb)) {
4871 /* Vlan id is non 0 and vlan_do_receive() above couldn't
4874 skb->pkt_type = PACKET_OTHERHOST;
4875 } else if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4876 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4877 /* Outer header is 802.1P with vlan 0, inner header is
4878 * 802.1Q or 802.1AD and vlan_do_receive() above could
4879 * not find vlan dev for vlan id 0.
4881 __vlan_hwaccel_clear_tag(skb);
4882 skb = skb_vlan_untag(skb);
4885 if (vlan_do_receive(&skb))
4886 /* After stripping off 802.1P header with vlan 0
4887 * vlan dev is found for inner header.
4890 else if (unlikely(!skb))
4893 /* We have stripped outer 802.1P vlan 0 header.
4894 * But could not find vlan dev.
4895 * check again for vlan id to set OTHERHOST.
4899 /* Note: we might in the future use prio bits
4900 * and set skb->priority like in vlan_do_receive()
4901 * For the time being, just ignore Priority Code Point
4903 __vlan_hwaccel_clear_tag(skb);
4906 type = skb->protocol;
4908 /* deliver only exact match when indicated */
4909 if (likely(!deliver_exact)) {
4910 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4911 &ptype_base[ntohs(type) &
4915 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4916 &orig_dev->ptype_specific);
4918 if (unlikely(skb->dev != orig_dev)) {
4919 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4920 &skb->dev->ptype_specific);
4924 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
4926 *ppt_prev = pt_prev;
4930 atomic_long_inc(&skb->dev->rx_dropped);
4932 atomic_long_inc(&skb->dev->rx_nohandler);
4934 /* Jamal, now you will not able to escape explaining
4935 * me how you were going to use this. :-)
4941 /* The invariant here is that if *ppt_prev is not NULL
4942 * then skb should also be non-NULL.
4944 * Apparently *ppt_prev assignment above holds this invariant due to
4945 * skb dereferencing near it.
4951 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
4953 struct net_device *orig_dev = skb->dev;
4954 struct packet_type *pt_prev = NULL;
4957 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
4959 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
4960 skb->dev, pt_prev, orig_dev);
4965 * netif_receive_skb_core - special purpose version of netif_receive_skb
4966 * @skb: buffer to process
4968 * More direct receive version of netif_receive_skb(). It should
4969 * only be used by callers that have a need to skip RPS and Generic XDP.
4970 * Caller must also take care of handling if (page_is_)pfmemalloc.
4972 * This function may only be called from softirq context and interrupts
4973 * should be enabled.
4975 * Return values (usually ignored):
4976 * NET_RX_SUCCESS: no congestion
4977 * NET_RX_DROP: packet was dropped
4979 int netif_receive_skb_core(struct sk_buff *skb)
4984 ret = __netif_receive_skb_one_core(skb, false);
4989 EXPORT_SYMBOL(netif_receive_skb_core);
4991 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
4992 struct packet_type *pt_prev,
4993 struct net_device *orig_dev)
4995 struct sk_buff *skb, *next;
4999 if (list_empty(head))
5001 if (pt_prev->list_func != NULL)
5002 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5003 ip_list_rcv, head, pt_prev, orig_dev);
5005 list_for_each_entry_safe(skb, next, head, list) {
5006 skb_list_del_init(skb);
5007 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5011 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5013 /* Fast-path assumptions:
5014 * - There is no RX handler.
5015 * - Only one packet_type matches.
5016 * If either of these fails, we will end up doing some per-packet
5017 * processing in-line, then handling the 'last ptype' for the whole
5018 * sublist. This can't cause out-of-order delivery to any single ptype,
5019 * because the 'last ptype' must be constant across the sublist, and all
5020 * other ptypes are handled per-packet.
5022 /* Current (common) ptype of sublist */
5023 struct packet_type *pt_curr = NULL;
5024 /* Current (common) orig_dev of sublist */
5025 struct net_device *od_curr = NULL;
5026 struct list_head sublist;
5027 struct sk_buff *skb, *next;
5029 INIT_LIST_HEAD(&sublist);
5030 list_for_each_entry_safe(skb, next, head, list) {
5031 struct net_device *orig_dev = skb->dev;
5032 struct packet_type *pt_prev = NULL;
5034 skb_list_del_init(skb);
5035 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5038 if (pt_curr != pt_prev || od_curr != orig_dev) {
5039 /* dispatch old sublist */
5040 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5041 /* start new sublist */
5042 INIT_LIST_HEAD(&sublist);
5046 list_add_tail(&skb->list, &sublist);
5049 /* dispatch final sublist */
5050 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5053 static int __netif_receive_skb(struct sk_buff *skb)
5057 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5058 unsigned int noreclaim_flag;
5061 * PFMEMALLOC skbs are special, they should
5062 * - be delivered to SOCK_MEMALLOC sockets only
5063 * - stay away from userspace
5064 * - have bounded memory usage
5066 * Use PF_MEMALLOC as this saves us from propagating the allocation
5067 * context down to all allocation sites.
5069 noreclaim_flag = memalloc_noreclaim_save();
5070 ret = __netif_receive_skb_one_core(skb, true);
5071 memalloc_noreclaim_restore(noreclaim_flag);
5073 ret = __netif_receive_skb_one_core(skb, false);
5078 static void __netif_receive_skb_list(struct list_head *head)
5080 unsigned long noreclaim_flag = 0;
5081 struct sk_buff *skb, *next;
5082 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5084 list_for_each_entry_safe(skb, next, head, list) {
5085 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5086 struct list_head sublist;
5088 /* Handle the previous sublist */
5089 list_cut_before(&sublist, head, &skb->list);
5090 if (!list_empty(&sublist))
5091 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5092 pfmemalloc = !pfmemalloc;
5093 /* See comments in __netif_receive_skb */
5095 noreclaim_flag = memalloc_noreclaim_save();
5097 memalloc_noreclaim_restore(noreclaim_flag);
5100 /* Handle the remaining sublist */
5101 if (!list_empty(head))
5102 __netif_receive_skb_list_core(head, pfmemalloc);
5103 /* Restore pflags */
5105 memalloc_noreclaim_restore(noreclaim_flag);
5108 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5110 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5111 struct bpf_prog *new = xdp->prog;
5114 switch (xdp->command) {
5115 case XDP_SETUP_PROG:
5116 rcu_assign_pointer(dev->xdp_prog, new);
5121 static_branch_dec(&generic_xdp_needed_key);
5122 } else if (new && !old) {
5123 static_branch_inc(&generic_xdp_needed_key);
5124 dev_disable_lro(dev);
5125 dev_disable_gro_hw(dev);
5129 case XDP_QUERY_PROG:
5130 xdp->prog_id = old ? old->aux->id : 0;
5141 static int netif_receive_skb_internal(struct sk_buff *skb)
5145 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5147 if (skb_defer_rx_timestamp(skb))
5148 return NET_RX_SUCCESS;
5152 if (static_branch_unlikely(&rps_needed)) {
5153 struct rps_dev_flow voidflow, *rflow = &voidflow;
5154 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5157 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5163 ret = __netif_receive_skb(skb);
5168 static void netif_receive_skb_list_internal(struct list_head *head)
5170 struct sk_buff *skb, *next;
5171 struct list_head sublist;
5173 INIT_LIST_HEAD(&sublist);
5174 list_for_each_entry_safe(skb, next, head, list) {
5175 net_timestamp_check(READ_ONCE(netdev_tstamp_prequeue), skb);
5176 skb_list_del_init(skb);
5177 if (!skb_defer_rx_timestamp(skb))
5178 list_add_tail(&skb->list, &sublist);
5180 list_splice_init(&sublist, head);
5184 if (static_branch_unlikely(&rps_needed)) {
5185 list_for_each_entry_safe(skb, next, head, list) {
5186 struct rps_dev_flow voidflow, *rflow = &voidflow;
5187 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5190 /* Will be handled, remove from list */
5191 skb_list_del_init(skb);
5192 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5197 __netif_receive_skb_list(head);
5202 * netif_receive_skb - process receive buffer from network
5203 * @skb: buffer to process
5205 * netif_receive_skb() is the main receive data processing function.
5206 * It always succeeds. The buffer may be dropped during processing
5207 * for congestion control or by the protocol layers.
5209 * This function may only be called from softirq context and interrupts
5210 * should be enabled.
5212 * Return values (usually ignored):
5213 * NET_RX_SUCCESS: no congestion
5214 * NET_RX_DROP: packet was dropped
5216 int netif_receive_skb(struct sk_buff *skb)
5220 trace_netif_receive_skb_entry(skb);
5222 ret = netif_receive_skb_internal(skb);
5223 trace_netif_receive_skb_exit(ret);
5227 EXPORT_SYMBOL(netif_receive_skb);
5230 * netif_receive_skb_list - process many receive buffers from network
5231 * @head: list of skbs to process.
5233 * Since return value of netif_receive_skb() is normally ignored, and
5234 * wouldn't be meaningful for a list, this function returns void.
5236 * This function may only be called from softirq context and interrupts
5237 * should be enabled.
5239 void netif_receive_skb_list(struct list_head *head)
5241 struct sk_buff *skb;
5243 if (list_empty(head))
5245 if (trace_netif_receive_skb_list_entry_enabled()) {
5246 list_for_each_entry(skb, head, list)
5247 trace_netif_receive_skb_list_entry(skb);
5249 netif_receive_skb_list_internal(head);
5250 trace_netif_receive_skb_list_exit(0);
5252 EXPORT_SYMBOL(netif_receive_skb_list);
5254 DEFINE_PER_CPU(struct work_struct, flush_works);
5256 /* Network device is going away, flush any packets still pending */
5257 static void flush_backlog(struct work_struct *work)
5259 struct sk_buff *skb, *tmp;
5260 struct softnet_data *sd;
5263 sd = this_cpu_ptr(&softnet_data);
5265 local_irq_disable();
5267 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5268 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5269 __skb_unlink(skb, &sd->input_pkt_queue);
5270 dev_kfree_skb_irq(skb);
5271 input_queue_head_incr(sd);
5277 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5278 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5279 __skb_unlink(skb, &sd->process_queue);
5281 input_queue_head_incr(sd);
5287 static void flush_all_backlogs(void)
5293 for_each_online_cpu(cpu)
5294 queue_work_on(cpu, system_highpri_wq,
5295 per_cpu_ptr(&flush_works, cpu));
5297 for_each_online_cpu(cpu)
5298 flush_work(per_cpu_ptr(&flush_works, cpu));
5303 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */
5304 static void gro_normal_list(struct napi_struct *napi)
5306 if (!napi->rx_count)
5308 netif_receive_skb_list_internal(&napi->rx_list);
5309 INIT_LIST_HEAD(&napi->rx_list);
5313 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded,
5314 * pass the whole batch up to the stack.
5316 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs)
5318 list_add_tail(&skb->list, &napi->rx_list);
5319 napi->rx_count += segs;
5320 if (napi->rx_count >= gro_normal_batch)
5321 gro_normal_list(napi);
5324 INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int));
5325 INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int));
5326 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb)
5328 struct packet_offload *ptype;
5329 __be16 type = skb->protocol;
5330 struct list_head *head = &offload_base;
5333 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5335 if (NAPI_GRO_CB(skb)->count == 1) {
5336 skb_shinfo(skb)->gso_size = 0;
5341 list_for_each_entry_rcu(ptype, head, list) {
5342 if (ptype->type != type || !ptype->callbacks.gro_complete)
5345 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5346 ipv6_gro_complete, inet_gro_complete,
5353 WARN_ON(&ptype->list == head);
5355 return NET_RX_SUCCESS;
5359 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count);
5360 return NET_RX_SUCCESS;
5363 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5366 struct list_head *head = &napi->gro_hash[index].list;
5367 struct sk_buff *skb, *p;
5369 list_for_each_entry_safe_reverse(skb, p, head, list) {
5370 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5372 skb_list_del_init(skb);
5373 napi_gro_complete(napi, skb);
5374 napi->gro_hash[index].count--;
5377 if (!napi->gro_hash[index].count)
5378 __clear_bit(index, &napi->gro_bitmask);
5381 /* napi->gro_hash[].list contains packets ordered by age.
5382 * youngest packets at the head of it.
5383 * Complete skbs in reverse order to reduce latencies.
5385 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5387 unsigned long bitmask = napi->gro_bitmask;
5388 unsigned int i, base = ~0U;
5390 while ((i = ffs(bitmask)) != 0) {
5393 __napi_gro_flush_chain(napi, base, flush_old);
5396 EXPORT_SYMBOL(napi_gro_flush);
5398 static struct list_head *gro_list_prepare(struct napi_struct *napi,
5399 struct sk_buff *skb)
5401 unsigned int maclen = skb->dev->hard_header_len;
5402 u32 hash = skb_get_hash_raw(skb);
5403 struct list_head *head;
5406 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list;
5407 list_for_each_entry(p, head, list) {
5408 unsigned long diffs;
5410 NAPI_GRO_CB(p)->flush = 0;
5412 if (hash != skb_get_hash_raw(p)) {
5413 NAPI_GRO_CB(p)->same_flow = 0;
5417 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5418 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5419 if (skb_vlan_tag_present(p))
5420 diffs |= p->vlan_tci ^ skb->vlan_tci;
5421 diffs |= skb_metadata_dst_cmp(p, skb);
5422 diffs |= skb_metadata_differs(p, skb);
5423 if (maclen == ETH_HLEN)
5424 diffs |= compare_ether_header(skb_mac_header(p),
5425 skb_mac_header(skb));
5427 diffs = memcmp(skb_mac_header(p),
5428 skb_mac_header(skb),
5430 NAPI_GRO_CB(p)->same_flow = !diffs;
5436 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff)
5438 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5439 const skb_frag_t *frag0 = &pinfo->frags[0];
5441 NAPI_GRO_CB(skb)->data_offset = 0;
5442 NAPI_GRO_CB(skb)->frag0 = NULL;
5443 NAPI_GRO_CB(skb)->frag0_len = 0;
5445 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
5447 !PageHighMem(skb_frag_page(frag0)) &&
5448 (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) {
5449 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5450 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5451 skb_frag_size(frag0),
5452 skb->end - skb->tail);
5456 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5458 struct skb_shared_info *pinfo = skb_shinfo(skb);
5460 BUG_ON(skb->end - skb->tail < grow);
5462 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5464 skb->data_len -= grow;
5467 skb_frag_off_add(&pinfo->frags[0], grow);
5468 skb_frag_size_sub(&pinfo->frags[0], grow);
5470 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5471 skb_frag_unref(skb, 0);
5472 memmove(pinfo->frags, pinfo->frags + 1,
5473 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
5477 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head)
5479 struct sk_buff *oldest;
5481 oldest = list_last_entry(head, struct sk_buff, list);
5483 /* We are called with head length >= MAX_GRO_SKBS, so this is
5486 if (WARN_ON_ONCE(!oldest))
5489 /* Do not adjust napi->gro_hash[].count, caller is adding a new
5492 skb_list_del_init(oldest);
5493 napi_gro_complete(napi, oldest);
5496 INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *,
5498 INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *,
5500 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5502 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
5503 struct list_head *head = &offload_base;
5504 struct packet_offload *ptype;
5505 __be16 type = skb->protocol;
5506 struct list_head *gro_head;
5507 struct sk_buff *pp = NULL;
5508 enum gro_result ret;
5512 if (netif_elide_gro(skb->dev))
5515 gro_head = gro_list_prepare(napi, skb);
5518 list_for_each_entry_rcu(ptype, head, list) {
5519 if (ptype->type != type || !ptype->callbacks.gro_receive)
5522 skb_set_network_header(skb, skb_gro_offset(skb));
5523 skb_reset_mac_len(skb);
5524 NAPI_GRO_CB(skb)->same_flow = 0;
5525 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
5526 NAPI_GRO_CB(skb)->free = 0;
5527 NAPI_GRO_CB(skb)->encap_mark = 0;
5528 NAPI_GRO_CB(skb)->recursion_counter = 0;
5529 NAPI_GRO_CB(skb)->is_fou = 0;
5530 NAPI_GRO_CB(skb)->is_atomic = 1;
5531 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
5533 /* Setup for GRO checksum validation */
5534 switch (skb->ip_summed) {
5535 case CHECKSUM_COMPLETE:
5536 NAPI_GRO_CB(skb)->csum = skb->csum;
5537 NAPI_GRO_CB(skb)->csum_valid = 1;
5538 NAPI_GRO_CB(skb)->csum_cnt = 0;
5540 case CHECKSUM_UNNECESSARY:
5541 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
5542 NAPI_GRO_CB(skb)->csum_valid = 0;
5545 NAPI_GRO_CB(skb)->csum_cnt = 0;
5546 NAPI_GRO_CB(skb)->csum_valid = 0;
5549 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
5550 ipv6_gro_receive, inet_gro_receive,
5556 if (&ptype->list == head)
5559 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
5564 same_flow = NAPI_GRO_CB(skb)->same_flow;
5565 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5568 skb_list_del_init(pp);
5569 napi_gro_complete(napi, pp);
5570 napi->gro_hash[hash].count--;
5576 if (NAPI_GRO_CB(skb)->flush)
5579 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
5580 gro_flush_oldest(napi, gro_head);
5582 napi->gro_hash[hash].count++;
5584 NAPI_GRO_CB(skb)->count = 1;
5585 NAPI_GRO_CB(skb)->age = jiffies;
5586 NAPI_GRO_CB(skb)->last = skb;
5587 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
5588 list_add(&skb->list, gro_head);
5592 grow = skb_gro_offset(skb) - skb_headlen(skb);
5594 gro_pull_from_frag0(skb, grow);
5596 if (napi->gro_hash[hash].count) {
5597 if (!test_bit(hash, &napi->gro_bitmask))
5598 __set_bit(hash, &napi->gro_bitmask);
5599 } else if (test_bit(hash, &napi->gro_bitmask)) {
5600 __clear_bit(hash, &napi->gro_bitmask);
5610 struct packet_offload *gro_find_receive_by_type(__be16 type)
5612 struct list_head *offload_head = &offload_base;
5613 struct packet_offload *ptype;
5615 list_for_each_entry_rcu(ptype, offload_head, list) {
5616 if (ptype->type != type || !ptype->callbacks.gro_receive)
5622 EXPORT_SYMBOL(gro_find_receive_by_type);
5624 struct packet_offload *gro_find_complete_by_type(__be16 type)
5626 struct list_head *offload_head = &offload_base;
5627 struct packet_offload *ptype;
5629 list_for_each_entry_rcu(ptype, offload_head, list) {
5630 if (ptype->type != type || !ptype->callbacks.gro_complete)
5636 EXPORT_SYMBOL(gro_find_complete_by_type);
5638 static void napi_skb_free_stolen_head(struct sk_buff *skb)
5642 kmem_cache_free(skbuff_head_cache, skb);
5645 static gro_result_t napi_skb_finish(struct napi_struct *napi,
5646 struct sk_buff *skb,
5651 gro_normal_one(napi, skb, 1);
5658 case GRO_MERGED_FREE:
5659 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5660 napi_skb_free_stolen_head(skb);
5674 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5678 skb_mark_napi_id(skb, napi);
5679 trace_napi_gro_receive_entry(skb);
5681 skb_gro_reset_offset(skb, 0);
5683 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
5684 trace_napi_gro_receive_exit(ret);
5688 EXPORT_SYMBOL(napi_gro_receive);
5690 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
5692 if (unlikely(skb->pfmemalloc)) {
5696 __skb_pull(skb, skb_headlen(skb));
5697 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
5698 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
5699 __vlan_hwaccel_clear_tag(skb);
5700 skb->dev = napi->dev;
5703 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
5704 skb->pkt_type = PACKET_HOST;
5706 skb->encapsulation = 0;
5707 skb_shinfo(skb)->gso_type = 0;
5708 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5714 struct sk_buff *napi_get_frags(struct napi_struct *napi)
5716 struct sk_buff *skb = napi->skb;
5719 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
5722 skb_mark_napi_id(skb, napi);
5727 EXPORT_SYMBOL(napi_get_frags);
5729 static gro_result_t napi_frags_finish(struct napi_struct *napi,
5730 struct sk_buff *skb,
5736 __skb_push(skb, ETH_HLEN);
5737 skb->protocol = eth_type_trans(skb, skb->dev);
5738 if (ret == GRO_NORMAL)
5739 gro_normal_one(napi, skb, 1);
5743 napi_reuse_skb(napi, skb);
5746 case GRO_MERGED_FREE:
5747 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5748 napi_skb_free_stolen_head(skb);
5750 napi_reuse_skb(napi, skb);
5761 /* Upper GRO stack assumes network header starts at gro_offset=0
5762 * Drivers could call both napi_gro_frags() and napi_gro_receive()
5763 * We copy ethernet header into skb->data to have a common layout.
5765 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
5767 struct sk_buff *skb = napi->skb;
5768 const struct ethhdr *eth;
5769 unsigned int hlen = sizeof(*eth);
5773 skb_reset_mac_header(skb);
5774 skb_gro_reset_offset(skb, hlen);
5776 if (unlikely(skb_gro_header_hard(skb, hlen))) {
5777 eth = skb_gro_header_slow(skb, hlen, 0);
5778 if (unlikely(!eth)) {
5779 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
5780 __func__, napi->dev->name);
5781 napi_reuse_skb(napi, skb);
5785 eth = (const struct ethhdr *)skb->data;
5786 gro_pull_from_frag0(skb, hlen);
5787 NAPI_GRO_CB(skb)->frag0 += hlen;
5788 NAPI_GRO_CB(skb)->frag0_len -= hlen;
5790 __skb_pull(skb, hlen);
5793 * This works because the only protocols we care about don't require
5795 * We'll fix it up properly in napi_frags_finish()
5797 skb->protocol = eth->h_proto;
5802 gro_result_t napi_gro_frags(struct napi_struct *napi)
5805 struct sk_buff *skb = napi_frags_skb(napi);
5810 trace_napi_gro_frags_entry(skb);
5812 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
5813 trace_napi_gro_frags_exit(ret);
5817 EXPORT_SYMBOL(napi_gro_frags);
5819 /* Compute the checksum from gro_offset and return the folded value
5820 * after adding in any pseudo checksum.
5822 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
5827 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
5829 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
5830 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
5831 /* See comments in __skb_checksum_complete(). */
5833 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
5834 !skb->csum_complete_sw)
5835 netdev_rx_csum_fault(skb->dev, skb);
5838 NAPI_GRO_CB(skb)->csum = wsum;
5839 NAPI_GRO_CB(skb)->csum_valid = 1;
5843 EXPORT_SYMBOL(__skb_gro_checksum_complete);
5845 static void net_rps_send_ipi(struct softnet_data *remsd)
5849 struct softnet_data *next = remsd->rps_ipi_next;
5851 if (cpu_online(remsd->cpu))
5852 smp_call_function_single_async(remsd->cpu, &remsd->csd);
5859 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5860 * Note: called with local irq disabled, but exits with local irq enabled.
5862 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5865 struct softnet_data *remsd = sd->rps_ipi_list;
5868 sd->rps_ipi_list = NULL;
5872 /* Send pending IPI's to kick RPS processing on remote cpus. */
5873 net_rps_send_ipi(remsd);
5879 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5882 return sd->rps_ipi_list != NULL;
5888 static int process_backlog(struct napi_struct *napi, int quota)
5890 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5894 /* Check if we have pending ipi, its better to send them now,
5895 * not waiting net_rx_action() end.
5897 if (sd_has_rps_ipi_waiting(sd)) {
5898 local_irq_disable();
5899 net_rps_action_and_irq_enable(sd);
5902 napi->weight = READ_ONCE(dev_rx_weight);
5904 struct sk_buff *skb;
5906 while ((skb = __skb_dequeue(&sd->process_queue))) {
5908 __netif_receive_skb(skb);
5910 input_queue_head_incr(sd);
5911 if (++work >= quota)
5916 local_irq_disable();
5918 if (skb_queue_empty(&sd->input_pkt_queue)) {
5920 * Inline a custom version of __napi_complete().
5921 * only current cpu owns and manipulates this napi,
5922 * and NAPI_STATE_SCHED is the only possible flag set
5924 * We can use a plain write instead of clear_bit(),
5925 * and we dont need an smp_mb() memory barrier.
5930 skb_queue_splice_tail_init(&sd->input_pkt_queue,
5931 &sd->process_queue);
5941 * __napi_schedule - schedule for receive
5942 * @n: entry to schedule
5944 * The entry's receive function will be scheduled to run.
5945 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5947 void __napi_schedule(struct napi_struct *n)
5949 unsigned long flags;
5951 local_irq_save(flags);
5952 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5953 local_irq_restore(flags);
5955 EXPORT_SYMBOL(__napi_schedule);
5958 * napi_schedule_prep - check if napi can be scheduled
5961 * Test if NAPI routine is already running, and if not mark
5962 * it as running. This is used as a condition variable
5963 * insure only one NAPI poll instance runs. We also make
5964 * sure there is no pending NAPI disable.
5966 bool napi_schedule_prep(struct napi_struct *n)
5968 unsigned long val, new;
5971 val = READ_ONCE(n->state);
5972 if (unlikely(val & NAPIF_STATE_DISABLE))
5974 new = val | NAPIF_STATE_SCHED;
5976 /* Sets STATE_MISSED bit if STATE_SCHED was already set
5977 * This was suggested by Alexander Duyck, as compiler
5978 * emits better code than :
5979 * if (val & NAPIF_STATE_SCHED)
5980 * new |= NAPIF_STATE_MISSED;
5982 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5984 } while (cmpxchg(&n->state, val, new) != val);
5986 return !(val & NAPIF_STATE_SCHED);
5988 EXPORT_SYMBOL(napi_schedule_prep);
5991 * __napi_schedule_irqoff - schedule for receive
5992 * @n: entry to schedule
5994 * Variant of __napi_schedule() assuming hard irqs are masked.
5996 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
5997 * because the interrupt disabled assumption might not be true
5998 * due to force-threaded interrupts and spinlock substitution.
6000 void __napi_schedule_irqoff(struct napi_struct *n)
6002 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
6003 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6007 EXPORT_SYMBOL(__napi_schedule_irqoff);
6009 bool napi_complete_done(struct napi_struct *n, int work_done)
6011 unsigned long flags, val, new;
6014 * 1) Don't let napi dequeue from the cpu poll list
6015 * just in case its running on a different cpu.
6016 * 2) If we are busy polling, do nothing here, we have
6017 * the guarantee we will be called later.
6019 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6020 NAPIF_STATE_IN_BUSY_POLL)))
6023 if (n->gro_bitmask) {
6024 unsigned long timeout = 0;
6027 timeout = n->dev->gro_flush_timeout;
6029 /* When the NAPI instance uses a timeout and keeps postponing
6030 * it, we need to bound somehow the time packets are kept in
6033 napi_gro_flush(n, !!timeout);
6035 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6036 HRTIMER_MODE_REL_PINNED);
6041 if (unlikely(!list_empty(&n->poll_list))) {
6042 /* If n->poll_list is not empty, we need to mask irqs */
6043 local_irq_save(flags);
6044 list_del_init(&n->poll_list);
6045 local_irq_restore(flags);
6049 val = READ_ONCE(n->state);
6051 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6053 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
6055 /* If STATE_MISSED was set, leave STATE_SCHED set,
6056 * because we will call napi->poll() one more time.
6057 * This C code was suggested by Alexander Duyck to help gcc.
6059 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6061 } while (cmpxchg(&n->state, val, new) != val);
6063 if (unlikely(val & NAPIF_STATE_MISSED)) {
6070 EXPORT_SYMBOL(napi_complete_done);
6072 /* must be called under rcu_read_lock(), as we dont take a reference */
6073 static struct napi_struct *napi_by_id(unsigned int napi_id)
6075 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6076 struct napi_struct *napi;
6078 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6079 if (napi->napi_id == napi_id)
6085 #if defined(CONFIG_NET_RX_BUSY_POLL)
6087 #define BUSY_POLL_BUDGET 8
6089 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
6093 /* Busy polling means there is a high chance device driver hard irq
6094 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6095 * set in napi_schedule_prep().
6096 * Since we are about to call napi->poll() once more, we can safely
6097 * clear NAPI_STATE_MISSED.
6099 * Note: x86 could use a single "lock and ..." instruction
6100 * to perform these two clear_bit()
6102 clear_bit(NAPI_STATE_MISSED, &napi->state);
6103 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6107 /* All we really want here is to re-enable device interrupts.
6108 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6110 rc = napi->poll(napi, BUSY_POLL_BUDGET);
6111 /* We can't gro_normal_list() here, because napi->poll() might have
6112 * rearmed the napi (napi_complete_done()) in which case it could
6113 * already be running on another CPU.
6115 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
6116 netpoll_poll_unlock(have_poll_lock);
6117 if (rc == BUSY_POLL_BUDGET) {
6118 /* As the whole budget was spent, we still own the napi so can
6119 * safely handle the rx_list.
6121 gro_normal_list(napi);
6122 __napi_schedule(napi);
6127 void napi_busy_loop(unsigned int napi_id,
6128 bool (*loop_end)(void *, unsigned long),
6131 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6132 int (*napi_poll)(struct napi_struct *napi, int budget);
6133 void *have_poll_lock = NULL;
6134 struct napi_struct *napi;
6141 napi = napi_by_id(napi_id);
6151 unsigned long val = READ_ONCE(napi->state);
6153 /* If multiple threads are competing for this napi,
6154 * we avoid dirtying napi->state as much as we can.
6156 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6157 NAPIF_STATE_IN_BUSY_POLL))
6159 if (cmpxchg(&napi->state, val,
6160 val | NAPIF_STATE_IN_BUSY_POLL |
6161 NAPIF_STATE_SCHED) != val)
6163 have_poll_lock = netpoll_poll_lock(napi);
6164 napi_poll = napi->poll;
6166 work = napi_poll(napi, BUSY_POLL_BUDGET);
6167 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
6168 gro_normal_list(napi);
6171 __NET_ADD_STATS(dev_net(napi->dev),
6172 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6175 if (!loop_end || loop_end(loop_end_arg, start_time))
6178 if (unlikely(need_resched())) {
6180 busy_poll_stop(napi, have_poll_lock);
6184 if (loop_end(loop_end_arg, start_time))
6191 busy_poll_stop(napi, have_poll_lock);
6196 EXPORT_SYMBOL(napi_busy_loop);
6198 #endif /* CONFIG_NET_RX_BUSY_POLL */
6200 static void napi_hash_add(struct napi_struct *napi)
6202 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
6203 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
6206 spin_lock(&napi_hash_lock);
6208 /* 0..NR_CPUS range is reserved for sender_cpu use */
6210 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6211 napi_gen_id = MIN_NAPI_ID;
6212 } while (napi_by_id(napi_gen_id));
6213 napi->napi_id = napi_gen_id;
6215 hlist_add_head_rcu(&napi->napi_hash_node,
6216 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6218 spin_unlock(&napi_hash_lock);
6221 /* Warning : caller is responsible to make sure rcu grace period
6222 * is respected before freeing memory containing @napi
6224 bool napi_hash_del(struct napi_struct *napi)
6226 bool rcu_sync_needed = false;
6228 spin_lock(&napi_hash_lock);
6230 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
6231 rcu_sync_needed = true;
6232 hlist_del_rcu(&napi->napi_hash_node);
6234 spin_unlock(&napi_hash_lock);
6235 return rcu_sync_needed;
6237 EXPORT_SYMBOL_GPL(napi_hash_del);
6239 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6241 struct napi_struct *napi;
6243 napi = container_of(timer, struct napi_struct, timer);
6245 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6246 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6248 if (napi->gro_bitmask && !napi_disable_pending(napi) &&
6249 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
6250 __napi_schedule_irqoff(napi);
6252 return HRTIMER_NORESTART;
6255 static void init_gro_hash(struct napi_struct *napi)
6259 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6260 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6261 napi->gro_hash[i].count = 0;
6263 napi->gro_bitmask = 0;
6266 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6267 int (*poll)(struct napi_struct *, int), int weight)
6269 INIT_LIST_HEAD(&napi->poll_list);
6270 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6271 napi->timer.function = napi_watchdog;
6272 init_gro_hash(napi);
6274 INIT_LIST_HEAD(&napi->rx_list);
6277 if (weight > NAPI_POLL_WEIGHT)
6278 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6280 napi->weight = weight;
6282 #ifdef CONFIG_NETPOLL
6283 napi->poll_owner = -1;
6285 set_bit(NAPI_STATE_SCHED, &napi->state);
6286 set_bit(NAPI_STATE_NPSVC, &napi->state);
6287 list_add_rcu(&napi->dev_list, &dev->napi_list);
6288 napi_hash_add(napi);
6290 EXPORT_SYMBOL(netif_napi_add);
6292 void napi_disable(struct napi_struct *n)
6295 set_bit(NAPI_STATE_DISABLE, &n->state);
6297 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6299 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6302 hrtimer_cancel(&n->timer);
6304 clear_bit(NAPI_STATE_DISABLE, &n->state);
6306 EXPORT_SYMBOL(napi_disable);
6308 static void flush_gro_hash(struct napi_struct *napi)
6312 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6313 struct sk_buff *skb, *n;
6315 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6317 napi->gro_hash[i].count = 0;
6321 /* Must be called in process context */
6322 void netif_napi_del(struct napi_struct *napi)
6325 if (napi_hash_del(napi))
6327 list_del_init(&napi->dev_list);
6328 napi_free_frags(napi);
6330 flush_gro_hash(napi);
6331 napi->gro_bitmask = 0;
6333 EXPORT_SYMBOL(netif_napi_del);
6335 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6340 list_del_init(&n->poll_list);
6342 have = netpoll_poll_lock(n);
6346 /* This NAPI_STATE_SCHED test is for avoiding a race
6347 * with netpoll's poll_napi(). Only the entity which
6348 * obtains the lock and sees NAPI_STATE_SCHED set will
6349 * actually make the ->poll() call. Therefore we avoid
6350 * accidentally calling ->poll() when NAPI is not scheduled.
6353 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6354 work = n->poll(n, weight);
6355 trace_napi_poll(n, work, weight);
6358 WARN_ON_ONCE(work > weight);
6360 if (likely(work < weight))
6363 /* Drivers must not modify the NAPI state if they
6364 * consume the entire weight. In such cases this code
6365 * still "owns" the NAPI instance and therefore can
6366 * move the instance around on the list at-will.
6368 if (unlikely(napi_disable_pending(n))) {
6373 if (n->gro_bitmask) {
6374 /* flush too old packets
6375 * If HZ < 1000, flush all packets.
6377 napi_gro_flush(n, HZ >= 1000);
6382 /* Some drivers may have called napi_schedule
6383 * prior to exhausting their budget.
6385 if (unlikely(!list_empty(&n->poll_list))) {
6386 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6387 n->dev ? n->dev->name : "backlog");
6391 list_add_tail(&n->poll_list, repoll);
6394 netpoll_poll_unlock(have);
6399 static __latent_entropy void net_rx_action(struct softirq_action *h)
6401 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6402 unsigned long time_limit = jiffies +
6403 usecs_to_jiffies(READ_ONCE(netdev_budget_usecs));
6404 int budget = READ_ONCE(netdev_budget);
6408 local_irq_disable();
6409 list_splice_init(&sd->poll_list, &list);
6413 struct napi_struct *n;
6415 if (list_empty(&list)) {
6416 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6421 n = list_first_entry(&list, struct napi_struct, poll_list);
6422 budget -= napi_poll(n, &repoll);
6424 /* If softirq window is exhausted then punt.
6425 * Allow this to run for 2 jiffies since which will allow
6426 * an average latency of 1.5/HZ.
6428 if (unlikely(budget <= 0 ||
6429 time_after_eq(jiffies, time_limit))) {
6435 local_irq_disable();
6437 list_splice_tail_init(&sd->poll_list, &list);
6438 list_splice_tail(&repoll, &list);
6439 list_splice(&list, &sd->poll_list);
6440 if (!list_empty(&sd->poll_list))
6441 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6443 net_rps_action_and_irq_enable(sd);
6445 __kfree_skb_flush();
6448 struct netdev_adjacent {
6449 struct net_device *dev;
6451 /* upper master flag, there can only be one master device per list */
6454 /* lookup ignore flag */
6457 /* counter for the number of times this device was added to us */
6460 /* private field for the users */
6463 struct list_head list;
6464 struct rcu_head rcu;
6467 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6468 struct list_head *adj_list)
6470 struct netdev_adjacent *adj;
6472 list_for_each_entry(adj, adj_list, list) {
6473 if (adj->dev == adj_dev)
6479 static int ____netdev_has_upper_dev(struct net_device *upper_dev, void *data)
6481 struct net_device *dev = data;
6483 return upper_dev == dev;
6487 * netdev_has_upper_dev - Check if device is linked to an upper device
6489 * @upper_dev: upper device to check
6491 * Find out if a device is linked to specified upper device and return true
6492 * in case it is. Note that this checks only immediate upper device,
6493 * not through a complete stack of devices. The caller must hold the RTNL lock.
6495 bool netdev_has_upper_dev(struct net_device *dev,
6496 struct net_device *upper_dev)
6500 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6503 EXPORT_SYMBOL(netdev_has_upper_dev);
6506 * netdev_has_upper_dev_all - Check if device is linked to an upper device
6508 * @upper_dev: upper device to check
6510 * Find out if a device is linked to specified upper device and return true
6511 * in case it is. Note that this checks the entire upper device chain.
6512 * The caller must hold rcu lock.
6515 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6516 struct net_device *upper_dev)
6518 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6521 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6524 * netdev_has_any_upper_dev - Check if device is linked to some device
6527 * Find out if a device is linked to an upper device and return true in case
6528 * it is. The caller must hold the RTNL lock.
6530 bool netdev_has_any_upper_dev(struct net_device *dev)
6534 return !list_empty(&dev->adj_list.upper);
6536 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6539 * netdev_master_upper_dev_get - Get master upper device
6542 * Find a master upper device and return pointer to it or NULL in case
6543 * it's not there. The caller must hold the RTNL lock.
6545 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6547 struct netdev_adjacent *upper;
6551 if (list_empty(&dev->adj_list.upper))
6554 upper = list_first_entry(&dev->adj_list.upper,
6555 struct netdev_adjacent, list);
6556 if (likely(upper->master))
6560 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6562 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6564 struct netdev_adjacent *upper;
6568 if (list_empty(&dev->adj_list.upper))
6571 upper = list_first_entry(&dev->adj_list.upper,
6572 struct netdev_adjacent, list);
6573 if (likely(upper->master) && !upper->ignore)
6579 * netdev_has_any_lower_dev - Check if device is linked to some device
6582 * Find out if a device is linked to a lower device and return true in case
6583 * it is. The caller must hold the RTNL lock.
6585 static bool netdev_has_any_lower_dev(struct net_device *dev)
6589 return !list_empty(&dev->adj_list.lower);
6592 void *netdev_adjacent_get_private(struct list_head *adj_list)
6594 struct netdev_adjacent *adj;
6596 adj = list_entry(adj_list, struct netdev_adjacent, list);
6598 return adj->private;
6600 EXPORT_SYMBOL(netdev_adjacent_get_private);
6603 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6605 * @iter: list_head ** of the current position
6607 * Gets the next device from the dev's upper list, starting from iter
6608 * position. The caller must hold RCU read lock.
6610 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6611 struct list_head **iter)
6613 struct netdev_adjacent *upper;
6615 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6617 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6619 if (&upper->list == &dev->adj_list.upper)
6622 *iter = &upper->list;
6626 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6628 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6629 struct list_head **iter,
6632 struct netdev_adjacent *upper;
6634 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6636 if (&upper->list == &dev->adj_list.upper)
6639 *iter = &upper->list;
6640 *ignore = upper->ignore;
6645 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6646 struct list_head **iter)
6648 struct netdev_adjacent *upper;
6650 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6652 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6654 if (&upper->list == &dev->adj_list.upper)
6657 *iter = &upper->list;
6662 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6663 int (*fn)(struct net_device *dev,
6667 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6668 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6673 iter = &dev->adj_list.upper;
6677 ret = fn(now, data);
6684 udev = __netdev_next_upper_dev(now, &iter, &ignore);
6691 niter = &udev->adj_list.upper;
6692 dev_stack[cur] = now;
6693 iter_stack[cur++] = iter;
6700 next = dev_stack[--cur];
6701 niter = iter_stack[cur];
6711 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6712 int (*fn)(struct net_device *dev,
6716 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6717 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6721 iter = &dev->adj_list.upper;
6725 ret = fn(now, data);
6732 udev = netdev_next_upper_dev_rcu(now, &iter);
6737 niter = &udev->adj_list.upper;
6738 dev_stack[cur] = now;
6739 iter_stack[cur++] = iter;
6746 next = dev_stack[--cur];
6747 niter = iter_stack[cur];
6756 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6758 static bool __netdev_has_upper_dev(struct net_device *dev,
6759 struct net_device *upper_dev)
6763 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
6768 * netdev_lower_get_next_private - Get the next ->private from the
6769 * lower neighbour list
6771 * @iter: list_head ** of the current position
6773 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6774 * list, starting from iter position. The caller must hold either hold the
6775 * RTNL lock or its own locking that guarantees that the neighbour lower
6776 * list will remain unchanged.
6778 void *netdev_lower_get_next_private(struct net_device *dev,
6779 struct list_head **iter)
6781 struct netdev_adjacent *lower;
6783 lower = list_entry(*iter, struct netdev_adjacent, list);
6785 if (&lower->list == &dev->adj_list.lower)
6788 *iter = lower->list.next;
6790 return lower->private;
6792 EXPORT_SYMBOL(netdev_lower_get_next_private);
6795 * netdev_lower_get_next_private_rcu - Get the next ->private from the
6796 * lower neighbour list, RCU
6799 * @iter: list_head ** of the current position
6801 * Gets the next netdev_adjacent->private from the dev's lower neighbour
6802 * list, starting from iter position. The caller must hold RCU read lock.
6804 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
6805 struct list_head **iter)
6807 struct netdev_adjacent *lower;
6809 WARN_ON_ONCE(!rcu_read_lock_held());
6811 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6813 if (&lower->list == &dev->adj_list.lower)
6816 *iter = &lower->list;
6818 return lower->private;
6820 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
6823 * netdev_lower_get_next - Get the next device from the lower neighbour
6826 * @iter: list_head ** of the current position
6828 * Gets the next netdev_adjacent from the dev's lower neighbour
6829 * list, starting from iter position. The caller must hold RTNL lock or
6830 * its own locking that guarantees that the neighbour lower
6831 * list will remain unchanged.
6833 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
6835 struct netdev_adjacent *lower;
6837 lower = list_entry(*iter, struct netdev_adjacent, list);
6839 if (&lower->list == &dev->adj_list.lower)
6842 *iter = lower->list.next;
6846 EXPORT_SYMBOL(netdev_lower_get_next);
6848 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
6849 struct list_head **iter)
6851 struct netdev_adjacent *lower;
6853 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
6855 if (&lower->list == &dev->adj_list.lower)
6858 *iter = &lower->list;
6863 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
6864 struct list_head **iter,
6867 struct netdev_adjacent *lower;
6869 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
6871 if (&lower->list == &dev->adj_list.lower)
6874 *iter = &lower->list;
6875 *ignore = lower->ignore;
6880 int netdev_walk_all_lower_dev(struct net_device *dev,
6881 int (*fn)(struct net_device *dev,
6885 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6886 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6890 iter = &dev->adj_list.lower;
6894 ret = fn(now, data);
6901 ldev = netdev_next_lower_dev(now, &iter);
6906 niter = &ldev->adj_list.lower;
6907 dev_stack[cur] = now;
6908 iter_stack[cur++] = iter;
6915 next = dev_stack[--cur];
6916 niter = iter_stack[cur];
6925 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
6927 static int __netdev_walk_all_lower_dev(struct net_device *dev,
6928 int (*fn)(struct net_device *dev,
6932 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6933 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
6938 iter = &dev->adj_list.lower;
6942 ret = fn(now, data);
6949 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
6956 niter = &ldev->adj_list.lower;
6957 dev_stack[cur] = now;
6958 iter_stack[cur++] = iter;
6965 next = dev_stack[--cur];
6966 niter = iter_stack[cur];
6976 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
6977 struct list_head **iter)
6979 struct netdev_adjacent *lower;
6981 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6982 if (&lower->list == &dev->adj_list.lower)
6985 *iter = &lower->list;
6989 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
6991 static u8 __netdev_upper_depth(struct net_device *dev)
6993 struct net_device *udev;
6994 struct list_head *iter;
6998 for (iter = &dev->adj_list.upper,
6999 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7001 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7004 if (max_depth < udev->upper_level)
7005 max_depth = udev->upper_level;
7011 static u8 __netdev_lower_depth(struct net_device *dev)
7013 struct net_device *ldev;
7014 struct list_head *iter;
7018 for (iter = &dev->adj_list.lower,
7019 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7021 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7024 if (max_depth < ldev->lower_level)
7025 max_depth = ldev->lower_level;
7031 static int __netdev_update_upper_level(struct net_device *dev, void *data)
7033 dev->upper_level = __netdev_upper_depth(dev) + 1;
7037 static int __netdev_update_lower_level(struct net_device *dev, void *data)
7039 dev->lower_level = __netdev_lower_depth(dev) + 1;
7043 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7044 int (*fn)(struct net_device *dev,
7048 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7049 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7053 iter = &dev->adj_list.lower;
7057 ret = fn(now, data);
7064 ldev = netdev_next_lower_dev_rcu(now, &iter);
7069 niter = &ldev->adj_list.lower;
7070 dev_stack[cur] = now;
7071 iter_stack[cur++] = iter;
7078 next = dev_stack[--cur];
7079 niter = iter_stack[cur];
7088 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7091 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7092 * lower neighbour list, RCU
7096 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7097 * list. The caller must hold RCU read lock.
7099 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7101 struct netdev_adjacent *lower;
7103 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7104 struct netdev_adjacent, list);
7106 return lower->private;
7109 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7112 * netdev_master_upper_dev_get_rcu - Get master upper device
7115 * Find a master upper device and return pointer to it or NULL in case
7116 * it's not there. The caller must hold the RCU read lock.
7118 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7120 struct netdev_adjacent *upper;
7122 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7123 struct netdev_adjacent, list);
7124 if (upper && likely(upper->master))
7128 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7130 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7131 struct net_device *adj_dev,
7132 struct list_head *dev_list)
7134 char linkname[IFNAMSIZ+7];
7136 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7137 "upper_%s" : "lower_%s", adj_dev->name);
7138 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7141 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7143 struct list_head *dev_list)
7145 char linkname[IFNAMSIZ+7];
7147 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7148 "upper_%s" : "lower_%s", name);
7149 sysfs_remove_link(&(dev->dev.kobj), linkname);
7152 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7153 struct net_device *adj_dev,
7154 struct list_head *dev_list)
7156 return (dev_list == &dev->adj_list.upper ||
7157 dev_list == &dev->adj_list.lower) &&
7158 net_eq(dev_net(dev), dev_net(adj_dev));
7161 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7162 struct net_device *adj_dev,
7163 struct list_head *dev_list,
7164 void *private, bool master)
7166 struct netdev_adjacent *adj;
7169 adj = __netdev_find_adj(adj_dev, dev_list);
7173 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7174 dev->name, adj_dev->name, adj->ref_nr);
7179 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7184 adj->master = master;
7186 adj->private = private;
7187 adj->ignore = false;
7190 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7191 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7193 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7194 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7199 /* Ensure that master link is always the first item in list. */
7201 ret = sysfs_create_link(&(dev->dev.kobj),
7202 &(adj_dev->dev.kobj), "master");
7204 goto remove_symlinks;
7206 list_add_rcu(&adj->list, dev_list);
7208 list_add_tail_rcu(&adj->list, dev_list);
7214 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7215 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7223 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7224 struct net_device *adj_dev,
7226 struct list_head *dev_list)
7228 struct netdev_adjacent *adj;
7230 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7231 dev->name, adj_dev->name, ref_nr);
7233 adj = __netdev_find_adj(adj_dev, dev_list);
7236 pr_err("Adjacency does not exist for device %s from %s\n",
7237 dev->name, adj_dev->name);
7242 if (adj->ref_nr > ref_nr) {
7243 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7244 dev->name, adj_dev->name, ref_nr,
7245 adj->ref_nr - ref_nr);
7246 adj->ref_nr -= ref_nr;
7251 sysfs_remove_link(&(dev->dev.kobj), "master");
7253 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7254 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7256 list_del_rcu(&adj->list);
7257 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7258 adj_dev->name, dev->name, adj_dev->name);
7260 kfree_rcu(adj, rcu);
7263 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7264 struct net_device *upper_dev,
7265 struct list_head *up_list,
7266 struct list_head *down_list,
7267 void *private, bool master)
7271 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7276 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7279 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7286 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7287 struct net_device *upper_dev,
7289 struct list_head *up_list,
7290 struct list_head *down_list)
7292 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7293 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7296 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7297 struct net_device *upper_dev,
7298 void *private, bool master)
7300 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7301 &dev->adj_list.upper,
7302 &upper_dev->adj_list.lower,
7306 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7307 struct net_device *upper_dev)
7309 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7310 &dev->adj_list.upper,
7311 &upper_dev->adj_list.lower);
7314 static int __netdev_upper_dev_link(struct net_device *dev,
7315 struct net_device *upper_dev, bool master,
7316 void *upper_priv, void *upper_info,
7317 struct netlink_ext_ack *extack)
7319 struct netdev_notifier_changeupper_info changeupper_info = {
7324 .upper_dev = upper_dev,
7327 .upper_info = upper_info,
7329 struct net_device *master_dev;
7334 if (dev == upper_dev)
7337 /* To prevent loops, check if dev is not upper device to upper_dev. */
7338 if (__netdev_has_upper_dev(upper_dev, dev))
7341 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7345 if (__netdev_has_upper_dev(dev, upper_dev))
7348 master_dev = __netdev_master_upper_dev_get(dev);
7350 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7353 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7354 &changeupper_info.info);
7355 ret = notifier_to_errno(ret);
7359 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7364 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7365 &changeupper_info.info);
7366 ret = notifier_to_errno(ret);
7370 __netdev_update_upper_level(dev, NULL);
7371 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7373 __netdev_update_lower_level(upper_dev, NULL);
7374 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7380 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7386 * netdev_upper_dev_link - Add a link to the upper device
7388 * @upper_dev: new upper device
7389 * @extack: netlink extended ack
7391 * Adds a link to device which is upper to this one. The caller must hold
7392 * the RTNL lock. On a failure a negative errno code is returned.
7393 * On success the reference counts are adjusted and the function
7396 int netdev_upper_dev_link(struct net_device *dev,
7397 struct net_device *upper_dev,
7398 struct netlink_ext_ack *extack)
7400 return __netdev_upper_dev_link(dev, upper_dev, false,
7401 NULL, NULL, extack);
7403 EXPORT_SYMBOL(netdev_upper_dev_link);
7406 * netdev_master_upper_dev_link - Add a master link to the upper device
7408 * @upper_dev: new upper device
7409 * @upper_priv: upper device private
7410 * @upper_info: upper info to be passed down via notifier
7411 * @extack: netlink extended ack
7413 * Adds a link to device which is upper to this one. In this case, only
7414 * one master upper device can be linked, although other non-master devices
7415 * might be linked as well. The caller must hold the RTNL lock.
7416 * On a failure a negative errno code is returned. On success the reference
7417 * counts are adjusted and the function returns zero.
7419 int netdev_master_upper_dev_link(struct net_device *dev,
7420 struct net_device *upper_dev,
7421 void *upper_priv, void *upper_info,
7422 struct netlink_ext_ack *extack)
7424 return __netdev_upper_dev_link(dev, upper_dev, true,
7425 upper_priv, upper_info, extack);
7427 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7430 * netdev_upper_dev_unlink - Removes a link to upper device
7432 * @upper_dev: new upper device
7434 * Removes a link to device which is upper to this one. The caller must hold
7437 void netdev_upper_dev_unlink(struct net_device *dev,
7438 struct net_device *upper_dev)
7440 struct netdev_notifier_changeupper_info changeupper_info = {
7444 .upper_dev = upper_dev,
7450 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7452 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7453 &changeupper_info.info);
7455 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7457 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7458 &changeupper_info.info);
7460 __netdev_update_upper_level(dev, NULL);
7461 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7463 __netdev_update_lower_level(upper_dev, NULL);
7464 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7467 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7469 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7470 struct net_device *lower_dev,
7473 struct netdev_adjacent *adj;
7475 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7479 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7484 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7485 struct net_device *lower_dev)
7487 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7490 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7491 struct net_device *lower_dev)
7493 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7496 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7497 struct net_device *new_dev,
7498 struct net_device *dev,
7499 struct netlink_ext_ack *extack)
7506 if (old_dev && new_dev != old_dev)
7507 netdev_adjacent_dev_disable(dev, old_dev);
7509 err = netdev_upper_dev_link(new_dev, dev, extack);
7511 if (old_dev && new_dev != old_dev)
7512 netdev_adjacent_dev_enable(dev, old_dev);
7518 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7520 void netdev_adjacent_change_commit(struct net_device *old_dev,
7521 struct net_device *new_dev,
7522 struct net_device *dev)
7524 if (!new_dev || !old_dev)
7527 if (new_dev == old_dev)
7530 netdev_adjacent_dev_enable(dev, old_dev);
7531 netdev_upper_dev_unlink(old_dev, dev);
7533 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7535 void netdev_adjacent_change_abort(struct net_device *old_dev,
7536 struct net_device *new_dev,
7537 struct net_device *dev)
7542 if (old_dev && new_dev != old_dev)
7543 netdev_adjacent_dev_enable(dev, old_dev);
7545 netdev_upper_dev_unlink(new_dev, dev);
7547 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7550 * netdev_bonding_info_change - Dispatch event about slave change
7552 * @bonding_info: info to dispatch
7554 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7555 * The caller must hold the RTNL lock.
7557 void netdev_bonding_info_change(struct net_device *dev,
7558 struct netdev_bonding_info *bonding_info)
7560 struct netdev_notifier_bonding_info info = {
7564 memcpy(&info.bonding_info, bonding_info,
7565 sizeof(struct netdev_bonding_info));
7566 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7569 EXPORT_SYMBOL(netdev_bonding_info_change);
7571 static void netdev_adjacent_add_links(struct net_device *dev)
7573 struct netdev_adjacent *iter;
7575 struct net *net = dev_net(dev);
7577 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7578 if (!net_eq(net, dev_net(iter->dev)))
7580 netdev_adjacent_sysfs_add(iter->dev, dev,
7581 &iter->dev->adj_list.lower);
7582 netdev_adjacent_sysfs_add(dev, iter->dev,
7583 &dev->adj_list.upper);
7586 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7587 if (!net_eq(net, dev_net(iter->dev)))
7589 netdev_adjacent_sysfs_add(iter->dev, dev,
7590 &iter->dev->adj_list.upper);
7591 netdev_adjacent_sysfs_add(dev, iter->dev,
7592 &dev->adj_list.lower);
7596 static void netdev_adjacent_del_links(struct net_device *dev)
7598 struct netdev_adjacent *iter;
7600 struct net *net = dev_net(dev);
7602 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7603 if (!net_eq(net, dev_net(iter->dev)))
7605 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7606 &iter->dev->adj_list.lower);
7607 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7608 &dev->adj_list.upper);
7611 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7612 if (!net_eq(net, dev_net(iter->dev)))
7614 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7615 &iter->dev->adj_list.upper);
7616 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7617 &dev->adj_list.lower);
7621 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
7623 struct netdev_adjacent *iter;
7625 struct net *net = dev_net(dev);
7627 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7628 if (!net_eq(net, dev_net(iter->dev)))
7630 netdev_adjacent_sysfs_del(iter->dev, oldname,
7631 &iter->dev->adj_list.lower);
7632 netdev_adjacent_sysfs_add(iter->dev, dev,
7633 &iter->dev->adj_list.lower);
7636 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7637 if (!net_eq(net, dev_net(iter->dev)))
7639 netdev_adjacent_sysfs_del(iter->dev, oldname,
7640 &iter->dev->adj_list.upper);
7641 netdev_adjacent_sysfs_add(iter->dev, dev,
7642 &iter->dev->adj_list.upper);
7646 void *netdev_lower_dev_get_private(struct net_device *dev,
7647 struct net_device *lower_dev)
7649 struct netdev_adjacent *lower;
7653 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
7657 return lower->private;
7659 EXPORT_SYMBOL(netdev_lower_dev_get_private);
7663 * netdev_lower_change - Dispatch event about lower device state change
7664 * @lower_dev: device
7665 * @lower_state_info: state to dispatch
7667 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
7668 * The caller must hold the RTNL lock.
7670 void netdev_lower_state_changed(struct net_device *lower_dev,
7671 void *lower_state_info)
7673 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
7674 .info.dev = lower_dev,
7678 changelowerstate_info.lower_state_info = lower_state_info;
7679 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
7680 &changelowerstate_info.info);
7682 EXPORT_SYMBOL(netdev_lower_state_changed);
7684 static void dev_change_rx_flags(struct net_device *dev, int flags)
7686 const struct net_device_ops *ops = dev->netdev_ops;
7688 if (ops->ndo_change_rx_flags)
7689 ops->ndo_change_rx_flags(dev, flags);
7692 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
7694 unsigned int old_flags = dev->flags;
7700 dev->flags |= IFF_PROMISC;
7701 dev->promiscuity += inc;
7702 if (dev->promiscuity == 0) {
7705 * If inc causes overflow, untouch promisc and return error.
7708 dev->flags &= ~IFF_PROMISC;
7710 dev->promiscuity -= inc;
7711 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
7716 if (dev->flags != old_flags) {
7717 pr_info("device %s %s promiscuous mode\n",
7719 dev->flags & IFF_PROMISC ? "entered" : "left");
7720 if (audit_enabled) {
7721 current_uid_gid(&uid, &gid);
7722 audit_log(audit_context(), GFP_ATOMIC,
7723 AUDIT_ANOM_PROMISCUOUS,
7724 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
7725 dev->name, (dev->flags & IFF_PROMISC),
7726 (old_flags & IFF_PROMISC),
7727 from_kuid(&init_user_ns, audit_get_loginuid(current)),
7728 from_kuid(&init_user_ns, uid),
7729 from_kgid(&init_user_ns, gid),
7730 audit_get_sessionid(current));
7733 dev_change_rx_flags(dev, IFF_PROMISC);
7736 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
7741 * dev_set_promiscuity - update promiscuity count on a device
7745 * Add or remove promiscuity from a device. While the count in the device
7746 * remains above zero the interface remains promiscuous. Once it hits zero
7747 * the device reverts back to normal filtering operation. A negative inc
7748 * value is used to drop promiscuity on the device.
7749 * Return 0 if successful or a negative errno code on error.
7751 int dev_set_promiscuity(struct net_device *dev, int inc)
7753 unsigned int old_flags = dev->flags;
7756 err = __dev_set_promiscuity(dev, inc, true);
7759 if (dev->flags != old_flags)
7760 dev_set_rx_mode(dev);
7763 EXPORT_SYMBOL(dev_set_promiscuity);
7765 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
7767 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
7771 dev->flags |= IFF_ALLMULTI;
7772 dev->allmulti += inc;
7773 if (dev->allmulti == 0) {
7776 * If inc causes overflow, untouch allmulti and return error.
7779 dev->flags &= ~IFF_ALLMULTI;
7781 dev->allmulti -= inc;
7782 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
7787 if (dev->flags ^ old_flags) {
7788 dev_change_rx_flags(dev, IFF_ALLMULTI);
7789 dev_set_rx_mode(dev);
7791 __dev_notify_flags(dev, old_flags,
7792 dev->gflags ^ old_gflags);
7798 * dev_set_allmulti - update allmulti count on a device
7802 * Add or remove reception of all multicast frames to a device. While the
7803 * count in the device remains above zero the interface remains listening
7804 * to all interfaces. Once it hits zero the device reverts back to normal
7805 * filtering operation. A negative @inc value is used to drop the counter
7806 * when releasing a resource needing all multicasts.
7807 * Return 0 if successful or a negative errno code on error.
7810 int dev_set_allmulti(struct net_device *dev, int inc)
7812 return __dev_set_allmulti(dev, inc, true);
7814 EXPORT_SYMBOL(dev_set_allmulti);
7817 * Upload unicast and multicast address lists to device and
7818 * configure RX filtering. When the device doesn't support unicast
7819 * filtering it is put in promiscuous mode while unicast addresses
7822 void __dev_set_rx_mode(struct net_device *dev)
7824 const struct net_device_ops *ops = dev->netdev_ops;
7826 /* dev_open will call this function so the list will stay sane. */
7827 if (!(dev->flags&IFF_UP))
7830 if (!netif_device_present(dev))
7833 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
7834 /* Unicast addresses changes may only happen under the rtnl,
7835 * therefore calling __dev_set_promiscuity here is safe.
7837 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
7838 __dev_set_promiscuity(dev, 1, false);
7839 dev->uc_promisc = true;
7840 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
7841 __dev_set_promiscuity(dev, -1, false);
7842 dev->uc_promisc = false;
7846 if (ops->ndo_set_rx_mode)
7847 ops->ndo_set_rx_mode(dev);
7850 void dev_set_rx_mode(struct net_device *dev)
7852 netif_addr_lock_bh(dev);
7853 __dev_set_rx_mode(dev);
7854 netif_addr_unlock_bh(dev);
7858 * dev_get_flags - get flags reported to userspace
7861 * Get the combination of flag bits exported through APIs to userspace.
7863 unsigned int dev_get_flags(const struct net_device *dev)
7867 flags = (dev->flags & ~(IFF_PROMISC |
7872 (dev->gflags & (IFF_PROMISC |
7875 if (netif_running(dev)) {
7876 if (netif_oper_up(dev))
7877 flags |= IFF_RUNNING;
7878 if (netif_carrier_ok(dev))
7879 flags |= IFF_LOWER_UP;
7880 if (netif_dormant(dev))
7881 flags |= IFF_DORMANT;
7886 EXPORT_SYMBOL(dev_get_flags);
7888 int __dev_change_flags(struct net_device *dev, unsigned int flags,
7889 struct netlink_ext_ack *extack)
7891 unsigned int old_flags = dev->flags;
7897 * Set the flags on our device.
7900 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
7901 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
7903 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
7907 * Load in the correct multicast list now the flags have changed.
7910 if ((old_flags ^ flags) & IFF_MULTICAST)
7911 dev_change_rx_flags(dev, IFF_MULTICAST);
7913 dev_set_rx_mode(dev);
7916 * Have we downed the interface. We handle IFF_UP ourselves
7917 * according to user attempts to set it, rather than blindly
7922 if ((old_flags ^ flags) & IFF_UP) {
7923 if (old_flags & IFF_UP)
7926 ret = __dev_open(dev, extack);
7929 if ((flags ^ dev->gflags) & IFF_PROMISC) {
7930 int inc = (flags & IFF_PROMISC) ? 1 : -1;
7931 unsigned int old_flags = dev->flags;
7933 dev->gflags ^= IFF_PROMISC;
7935 if (__dev_set_promiscuity(dev, inc, false) >= 0)
7936 if (dev->flags != old_flags)
7937 dev_set_rx_mode(dev);
7940 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
7941 * is important. Some (broken) drivers set IFF_PROMISC, when
7942 * IFF_ALLMULTI is requested not asking us and not reporting.
7944 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
7945 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
7947 dev->gflags ^= IFF_ALLMULTI;
7948 __dev_set_allmulti(dev, inc, false);
7954 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
7955 unsigned int gchanges)
7957 unsigned int changes = dev->flags ^ old_flags;
7960 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
7962 if (changes & IFF_UP) {
7963 if (dev->flags & IFF_UP)
7964 call_netdevice_notifiers(NETDEV_UP, dev);
7966 call_netdevice_notifiers(NETDEV_DOWN, dev);
7969 if (dev->flags & IFF_UP &&
7970 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
7971 struct netdev_notifier_change_info change_info = {
7975 .flags_changed = changes,
7978 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
7983 * dev_change_flags - change device settings
7985 * @flags: device state flags
7986 * @extack: netlink extended ack
7988 * Change settings on device based state flags. The flags are
7989 * in the userspace exported format.
7991 int dev_change_flags(struct net_device *dev, unsigned int flags,
7992 struct netlink_ext_ack *extack)
7995 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
7997 ret = __dev_change_flags(dev, flags, extack);
8001 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8002 __dev_notify_flags(dev, old_flags, changes);
8005 EXPORT_SYMBOL(dev_change_flags);
8007 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8009 const struct net_device_ops *ops = dev->netdev_ops;
8011 if (ops->ndo_change_mtu)
8012 return ops->ndo_change_mtu(dev, new_mtu);
8014 /* Pairs with all the lockless reads of dev->mtu in the stack */
8015 WRITE_ONCE(dev->mtu, new_mtu);
8018 EXPORT_SYMBOL(__dev_set_mtu);
8020 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8021 struct netlink_ext_ack *extack)
8023 /* MTU must be positive, and in range */
8024 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8025 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8029 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8030 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8037 * dev_set_mtu_ext - Change maximum transfer unit
8039 * @new_mtu: new transfer unit
8040 * @extack: netlink extended ack
8042 * Change the maximum transfer size of the network device.
8044 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8045 struct netlink_ext_ack *extack)
8049 if (new_mtu == dev->mtu)
8052 err = dev_validate_mtu(dev, new_mtu, extack);
8056 if (!netif_device_present(dev))
8059 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8060 err = notifier_to_errno(err);
8064 orig_mtu = dev->mtu;
8065 err = __dev_set_mtu(dev, new_mtu);
8068 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8070 err = notifier_to_errno(err);
8072 /* setting mtu back and notifying everyone again,
8073 * so that they have a chance to revert changes.
8075 __dev_set_mtu(dev, orig_mtu);
8076 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8083 int dev_set_mtu(struct net_device *dev, int new_mtu)
8085 struct netlink_ext_ack extack;
8088 memset(&extack, 0, sizeof(extack));
8089 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8090 if (err && extack._msg)
8091 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8094 EXPORT_SYMBOL(dev_set_mtu);
8097 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8099 * @new_len: new tx queue length
8101 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8103 unsigned int orig_len = dev->tx_queue_len;
8106 if (new_len != (unsigned int)new_len)
8109 if (new_len != orig_len) {
8110 dev->tx_queue_len = new_len;
8111 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8112 res = notifier_to_errno(res);
8115 res = dev_qdisc_change_tx_queue_len(dev);
8123 netdev_err(dev, "refused to change device tx_queue_len\n");
8124 dev->tx_queue_len = orig_len;
8129 * dev_set_group - Change group this device belongs to
8131 * @new_group: group this device should belong to
8133 void dev_set_group(struct net_device *dev, int new_group)
8135 dev->group = new_group;
8137 EXPORT_SYMBOL(dev_set_group);
8140 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8142 * @addr: new address
8143 * @extack: netlink extended ack
8145 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8146 struct netlink_ext_ack *extack)
8148 struct netdev_notifier_pre_changeaddr_info info = {
8150 .info.extack = extack,
8155 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8156 return notifier_to_errno(rc);
8158 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8161 * dev_set_mac_address - Change Media Access Control Address
8164 * @extack: netlink extended ack
8166 * Change the hardware (MAC) address of the device
8168 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8169 struct netlink_ext_ack *extack)
8171 const struct net_device_ops *ops = dev->netdev_ops;
8174 if (!ops->ndo_set_mac_address)
8176 if (sa->sa_family != dev->type)
8178 if (!netif_device_present(dev))
8180 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8183 err = ops->ndo_set_mac_address(dev, sa);
8186 dev->addr_assign_type = NET_ADDR_SET;
8187 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8188 add_device_randomness(dev->dev_addr, dev->addr_len);
8191 EXPORT_SYMBOL(dev_set_mac_address);
8193 static DECLARE_RWSEM(dev_addr_sem);
8195 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
8196 struct netlink_ext_ack *extack)
8200 down_write(&dev_addr_sem);
8201 ret = dev_set_mac_address(dev, sa, extack);
8202 up_write(&dev_addr_sem);
8205 EXPORT_SYMBOL(dev_set_mac_address_user);
8207 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
8209 size_t size = sizeof(sa->sa_data);
8210 struct net_device *dev;
8213 down_read(&dev_addr_sem);
8216 dev = dev_get_by_name_rcu(net, dev_name);
8222 memset(sa->sa_data, 0, size);
8224 memcpy(sa->sa_data, dev->dev_addr,
8225 min_t(size_t, size, dev->addr_len));
8226 sa->sa_family = dev->type;
8230 up_read(&dev_addr_sem);
8233 EXPORT_SYMBOL(dev_get_mac_address);
8236 * dev_change_carrier - Change device carrier
8238 * @new_carrier: new value
8240 * Change device carrier
8242 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8244 const struct net_device_ops *ops = dev->netdev_ops;
8246 if (!ops->ndo_change_carrier)
8248 if (!netif_device_present(dev))
8250 return ops->ndo_change_carrier(dev, new_carrier);
8252 EXPORT_SYMBOL(dev_change_carrier);
8255 * dev_get_phys_port_id - Get device physical port ID
8259 * Get device physical port ID
8261 int dev_get_phys_port_id(struct net_device *dev,
8262 struct netdev_phys_item_id *ppid)
8264 const struct net_device_ops *ops = dev->netdev_ops;
8266 if (!ops->ndo_get_phys_port_id)
8268 return ops->ndo_get_phys_port_id(dev, ppid);
8270 EXPORT_SYMBOL(dev_get_phys_port_id);
8273 * dev_get_phys_port_name - Get device physical port name
8276 * @len: limit of bytes to copy to name
8278 * Get device physical port name
8280 int dev_get_phys_port_name(struct net_device *dev,
8281 char *name, size_t len)
8283 const struct net_device_ops *ops = dev->netdev_ops;
8286 if (ops->ndo_get_phys_port_name) {
8287 err = ops->ndo_get_phys_port_name(dev, name, len);
8288 if (err != -EOPNOTSUPP)
8291 return devlink_compat_phys_port_name_get(dev, name, len);
8293 EXPORT_SYMBOL(dev_get_phys_port_name);
8296 * dev_get_port_parent_id - Get the device's port parent identifier
8297 * @dev: network device
8298 * @ppid: pointer to a storage for the port's parent identifier
8299 * @recurse: allow/disallow recursion to lower devices
8301 * Get the devices's port parent identifier
8303 int dev_get_port_parent_id(struct net_device *dev,
8304 struct netdev_phys_item_id *ppid,
8307 const struct net_device_ops *ops = dev->netdev_ops;
8308 struct netdev_phys_item_id first = { };
8309 struct net_device *lower_dev;
8310 struct list_head *iter;
8313 if (ops->ndo_get_port_parent_id) {
8314 err = ops->ndo_get_port_parent_id(dev, ppid);
8315 if (err != -EOPNOTSUPP)
8319 err = devlink_compat_switch_id_get(dev, ppid);
8320 if (!err || err != -EOPNOTSUPP)
8326 netdev_for_each_lower_dev(dev, lower_dev, iter) {
8327 err = dev_get_port_parent_id(lower_dev, ppid, recurse);
8332 else if (memcmp(&first, ppid, sizeof(*ppid)))
8338 EXPORT_SYMBOL(dev_get_port_parent_id);
8341 * netdev_port_same_parent_id - Indicate if two network devices have
8342 * the same port parent identifier
8343 * @a: first network device
8344 * @b: second network device
8346 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8348 struct netdev_phys_item_id a_id = { };
8349 struct netdev_phys_item_id b_id = { };
8351 if (dev_get_port_parent_id(a, &a_id, true) ||
8352 dev_get_port_parent_id(b, &b_id, true))
8355 return netdev_phys_item_id_same(&a_id, &b_id);
8357 EXPORT_SYMBOL(netdev_port_same_parent_id);
8360 * dev_change_proto_down - update protocol port state information
8362 * @proto_down: new value
8364 * This info can be used by switch drivers to set the phys state of the
8367 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8369 const struct net_device_ops *ops = dev->netdev_ops;
8371 if (!ops->ndo_change_proto_down)
8373 if (!netif_device_present(dev))
8375 return ops->ndo_change_proto_down(dev, proto_down);
8377 EXPORT_SYMBOL(dev_change_proto_down);
8380 * dev_change_proto_down_generic - generic implementation for
8381 * ndo_change_proto_down that sets carrier according to
8385 * @proto_down: new value
8387 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
8390 netif_carrier_off(dev);
8392 netif_carrier_on(dev);
8393 dev->proto_down = proto_down;
8396 EXPORT_SYMBOL(dev_change_proto_down_generic);
8398 u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op,
8399 enum bpf_netdev_command cmd)
8401 struct netdev_bpf xdp;
8406 memset(&xdp, 0, sizeof(xdp));
8409 /* Query must always succeed. */
8410 WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG);
8415 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op,
8416 struct netlink_ext_ack *extack, u32 flags,
8417 struct bpf_prog *prog)
8419 struct netdev_bpf xdp;
8421 memset(&xdp, 0, sizeof(xdp));
8422 if (flags & XDP_FLAGS_HW_MODE)
8423 xdp.command = XDP_SETUP_PROG_HW;
8425 xdp.command = XDP_SETUP_PROG;
8426 xdp.extack = extack;
8430 return bpf_op(dev, &xdp);
8433 static void dev_xdp_uninstall(struct net_device *dev)
8435 struct netdev_bpf xdp;
8438 /* Remove generic XDP */
8439 WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL));
8441 /* Remove from the driver */
8442 ndo_bpf = dev->netdev_ops->ndo_bpf;
8446 memset(&xdp, 0, sizeof(xdp));
8447 xdp.command = XDP_QUERY_PROG;
8448 WARN_ON(ndo_bpf(dev, &xdp));
8450 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8453 /* Remove HW offload */
8454 memset(&xdp, 0, sizeof(xdp));
8455 xdp.command = XDP_QUERY_PROG_HW;
8456 if (!ndo_bpf(dev, &xdp) && xdp.prog_id)
8457 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8462 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
8464 * @extack: netlink extended ack
8465 * @fd: new program fd or negative value to clear
8466 * @flags: xdp-related flags
8468 * Set or clear a bpf program for a device
8470 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
8473 const struct net_device_ops *ops = dev->netdev_ops;
8474 enum bpf_netdev_command query;
8475 struct bpf_prog *prog = NULL;
8476 bpf_op_t bpf_op, bpf_chk;
8482 offload = flags & XDP_FLAGS_HW_MODE;
8483 query = offload ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG;
8485 bpf_op = bpf_chk = ops->ndo_bpf;
8486 if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) {
8487 NL_SET_ERR_MSG(extack, "underlying driver does not support XDP in native mode");
8490 if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE))
8491 bpf_op = generic_xdp_install;
8492 if (bpf_op == bpf_chk)
8493 bpf_chk = generic_xdp_install;
8498 if (!offload && __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG)) {
8499 NL_SET_ERR_MSG(extack, "native and generic XDP can't be active at the same time");
8503 prog_id = __dev_xdp_query(dev, bpf_op, query);
8504 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && prog_id) {
8505 NL_SET_ERR_MSG(extack, "XDP program already attached");
8509 prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
8510 bpf_op == ops->ndo_bpf);
8512 return PTR_ERR(prog);
8514 if (!offload && bpf_prog_is_dev_bound(prog->aux)) {
8515 NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported");
8520 /* prog->aux->id may be 0 for orphaned device-bound progs */
8521 if (prog->aux->id && prog->aux->id == prog_id) {
8526 if (!__dev_xdp_query(dev, bpf_op, query))
8530 err = dev_xdp_install(dev, bpf_op, extack, flags, prog);
8531 if (err < 0 && prog)
8538 * dev_new_index - allocate an ifindex
8539 * @net: the applicable net namespace
8541 * Returns a suitable unique value for a new device interface
8542 * number. The caller must hold the rtnl semaphore or the
8543 * dev_base_lock to be sure it remains unique.
8545 static int dev_new_index(struct net *net)
8547 int ifindex = net->ifindex;
8552 if (!__dev_get_by_index(net, ifindex))
8553 return net->ifindex = ifindex;
8557 /* Delayed registration/unregisteration */
8558 static LIST_HEAD(net_todo_list);
8559 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
8561 static void net_set_todo(struct net_device *dev)
8563 list_add_tail(&dev->todo_list, &net_todo_list);
8564 dev_net(dev)->dev_unreg_count++;
8567 static void rollback_registered_many(struct list_head *head)
8569 struct net_device *dev, *tmp;
8570 LIST_HEAD(close_head);
8572 BUG_ON(dev_boot_phase);
8575 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
8576 /* Some devices call without registering
8577 * for initialization unwind. Remove those
8578 * devices and proceed with the remaining.
8580 if (dev->reg_state == NETREG_UNINITIALIZED) {
8581 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
8585 list_del(&dev->unreg_list);
8588 dev->dismantle = true;
8589 BUG_ON(dev->reg_state != NETREG_REGISTERED);
8592 /* If device is running, close it first. */
8593 list_for_each_entry(dev, head, unreg_list)
8594 list_add_tail(&dev->close_list, &close_head);
8595 dev_close_many(&close_head, true);
8597 list_for_each_entry(dev, head, unreg_list) {
8598 /* And unlink it from device chain. */
8599 unlist_netdevice(dev);
8601 dev->reg_state = NETREG_UNREGISTERING;
8603 flush_all_backlogs();
8607 list_for_each_entry(dev, head, unreg_list) {
8608 struct sk_buff *skb = NULL;
8610 /* Shutdown queueing discipline. */
8613 dev_xdp_uninstall(dev);
8615 /* Notify protocols, that we are about to destroy
8616 * this device. They should clean all the things.
8618 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8620 if (!dev->rtnl_link_ops ||
8621 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8622 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
8623 GFP_KERNEL, NULL, 0);
8626 * Flush the unicast and multicast chains
8631 if (dev->netdev_ops->ndo_uninit)
8632 dev->netdev_ops->ndo_uninit(dev);
8635 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
8637 /* Notifier chain MUST detach us all upper devices. */
8638 WARN_ON(netdev_has_any_upper_dev(dev));
8639 WARN_ON(netdev_has_any_lower_dev(dev));
8641 /* Remove entries from kobject tree */
8642 netdev_unregister_kobject(dev);
8644 /* Remove XPS queueing entries */
8645 netif_reset_xps_queues_gt(dev, 0);
8651 list_for_each_entry(dev, head, unreg_list)
8655 static void rollback_registered(struct net_device *dev)
8659 list_add(&dev->unreg_list, &single);
8660 rollback_registered_many(&single);
8664 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
8665 struct net_device *upper, netdev_features_t features)
8667 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8668 netdev_features_t feature;
8671 for_each_netdev_feature(upper_disables, feature_bit) {
8672 feature = __NETIF_F_BIT(feature_bit);
8673 if (!(upper->wanted_features & feature)
8674 && (features & feature)) {
8675 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
8676 &feature, upper->name);
8677 features &= ~feature;
8684 static void netdev_sync_lower_features(struct net_device *upper,
8685 struct net_device *lower, netdev_features_t features)
8687 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
8688 netdev_features_t feature;
8691 for_each_netdev_feature(upper_disables, feature_bit) {
8692 feature = __NETIF_F_BIT(feature_bit);
8693 if (!(features & feature) && (lower->features & feature)) {
8694 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
8695 &feature, lower->name);
8696 lower->wanted_features &= ~feature;
8697 __netdev_update_features(lower);
8699 if (unlikely(lower->features & feature))
8700 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
8701 &feature, lower->name);
8703 netdev_features_change(lower);
8708 static netdev_features_t netdev_fix_features(struct net_device *dev,
8709 netdev_features_t features)
8711 /* Fix illegal checksum combinations */
8712 if ((features & NETIF_F_HW_CSUM) &&
8713 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
8714 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
8715 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
8718 /* TSO requires that SG is present as well. */
8719 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
8720 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
8721 features &= ~NETIF_F_ALL_TSO;
8724 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
8725 !(features & NETIF_F_IP_CSUM)) {
8726 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
8727 features &= ~NETIF_F_TSO;
8728 features &= ~NETIF_F_TSO_ECN;
8731 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
8732 !(features & NETIF_F_IPV6_CSUM)) {
8733 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
8734 features &= ~NETIF_F_TSO6;
8737 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
8738 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
8739 features &= ~NETIF_F_TSO_MANGLEID;
8741 /* TSO ECN requires that TSO is present as well. */
8742 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
8743 features &= ~NETIF_F_TSO_ECN;
8745 /* Software GSO depends on SG. */
8746 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
8747 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
8748 features &= ~NETIF_F_GSO;
8751 /* GSO partial features require GSO partial be set */
8752 if ((features & dev->gso_partial_features) &&
8753 !(features & NETIF_F_GSO_PARTIAL)) {
8755 "Dropping partially supported GSO features since no GSO partial.\n");
8756 features &= ~dev->gso_partial_features;
8759 if (!(features & NETIF_F_RXCSUM)) {
8760 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
8761 * successfully merged by hardware must also have the
8762 * checksum verified by hardware. If the user does not
8763 * want to enable RXCSUM, logically, we should disable GRO_HW.
8765 if (features & NETIF_F_GRO_HW) {
8766 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
8767 features &= ~NETIF_F_GRO_HW;
8771 /* LRO/HW-GRO features cannot be combined with RX-FCS */
8772 if (features & NETIF_F_RXFCS) {
8773 if (features & NETIF_F_LRO) {
8774 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
8775 features &= ~NETIF_F_LRO;
8778 if (features & NETIF_F_GRO_HW) {
8779 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
8780 features &= ~NETIF_F_GRO_HW;
8784 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
8785 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
8786 features &= ~NETIF_F_HW_TLS_RX;
8792 int __netdev_update_features(struct net_device *dev)
8794 struct net_device *upper, *lower;
8795 netdev_features_t features;
8796 struct list_head *iter;
8801 features = netdev_get_wanted_features(dev);
8803 if (dev->netdev_ops->ndo_fix_features)
8804 features = dev->netdev_ops->ndo_fix_features(dev, features);
8806 /* driver might be less strict about feature dependencies */
8807 features = netdev_fix_features(dev, features);
8809 /* some features can't be enabled if they're off an an upper device */
8810 netdev_for_each_upper_dev_rcu(dev, upper, iter)
8811 features = netdev_sync_upper_features(dev, upper, features);
8813 if (dev->features == features)
8816 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
8817 &dev->features, &features);
8819 if (dev->netdev_ops->ndo_set_features)
8820 err = dev->netdev_ops->ndo_set_features(dev, features);
8824 if (unlikely(err < 0)) {
8826 "set_features() failed (%d); wanted %pNF, left %pNF\n",
8827 err, &features, &dev->features);
8828 /* return non-0 since some features might have changed and
8829 * it's better to fire a spurious notification than miss it
8835 /* some features must be disabled on lower devices when disabled
8836 * on an upper device (think: bonding master or bridge)
8838 netdev_for_each_lower_dev(dev, lower, iter)
8839 netdev_sync_lower_features(dev, lower, features);
8842 netdev_features_t diff = features ^ dev->features;
8844 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
8845 /* udp_tunnel_{get,drop}_rx_info both need
8846 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
8847 * device, or they won't do anything.
8848 * Thus we need to update dev->features
8849 * *before* calling udp_tunnel_get_rx_info,
8850 * but *after* calling udp_tunnel_drop_rx_info.
8852 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
8853 dev->features = features;
8854 udp_tunnel_get_rx_info(dev);
8856 udp_tunnel_drop_rx_info(dev);
8860 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
8861 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
8862 dev->features = features;
8863 err |= vlan_get_rx_ctag_filter_info(dev);
8865 vlan_drop_rx_ctag_filter_info(dev);
8869 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
8870 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
8871 dev->features = features;
8872 err |= vlan_get_rx_stag_filter_info(dev);
8874 vlan_drop_rx_stag_filter_info(dev);
8878 dev->features = features;
8881 return err < 0 ? 0 : 1;
8885 * netdev_update_features - recalculate device features
8886 * @dev: the device to check
8888 * Recalculate dev->features set and send notifications if it
8889 * has changed. Should be called after driver or hardware dependent
8890 * conditions might have changed that influence the features.
8892 void netdev_update_features(struct net_device *dev)
8894 if (__netdev_update_features(dev))
8895 netdev_features_change(dev);
8897 EXPORT_SYMBOL(netdev_update_features);
8900 * netdev_change_features - recalculate device features
8901 * @dev: the device to check
8903 * Recalculate dev->features set and send notifications even
8904 * if they have not changed. Should be called instead of
8905 * netdev_update_features() if also dev->vlan_features might
8906 * have changed to allow the changes to be propagated to stacked
8909 void netdev_change_features(struct net_device *dev)
8911 __netdev_update_features(dev);
8912 netdev_features_change(dev);
8914 EXPORT_SYMBOL(netdev_change_features);
8917 * netif_stacked_transfer_operstate - transfer operstate
8918 * @rootdev: the root or lower level device to transfer state from
8919 * @dev: the device to transfer operstate to
8921 * Transfer operational state from root to device. This is normally
8922 * called when a stacking relationship exists between the root
8923 * device and the device(a leaf device).
8925 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
8926 struct net_device *dev)
8928 if (rootdev->operstate == IF_OPER_DORMANT)
8929 netif_dormant_on(dev);
8931 netif_dormant_off(dev);
8933 if (netif_carrier_ok(rootdev))
8934 netif_carrier_on(dev);
8936 netif_carrier_off(dev);
8938 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
8940 static int netif_alloc_rx_queues(struct net_device *dev)
8942 unsigned int i, count = dev->num_rx_queues;
8943 struct netdev_rx_queue *rx;
8944 size_t sz = count * sizeof(*rx);
8949 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
8955 for (i = 0; i < count; i++) {
8958 /* XDP RX-queue setup */
8959 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i);
8966 /* Rollback successful reg's and free other resources */
8968 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
8974 static void netif_free_rx_queues(struct net_device *dev)
8976 unsigned int i, count = dev->num_rx_queues;
8978 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
8982 for (i = 0; i < count; i++)
8983 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
8988 static void netdev_init_one_queue(struct net_device *dev,
8989 struct netdev_queue *queue, void *_unused)
8991 /* Initialize queue lock */
8992 spin_lock_init(&queue->_xmit_lock);
8993 lockdep_set_class(&queue->_xmit_lock, &dev->qdisc_xmit_lock_key);
8994 queue->xmit_lock_owner = -1;
8995 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
8998 dql_init(&queue->dql, HZ);
9002 static void netif_free_tx_queues(struct net_device *dev)
9007 static int netif_alloc_netdev_queues(struct net_device *dev)
9009 unsigned int count = dev->num_tx_queues;
9010 struct netdev_queue *tx;
9011 size_t sz = count * sizeof(*tx);
9013 if (count < 1 || count > 0xffff)
9016 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9022 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9023 spin_lock_init(&dev->tx_global_lock);
9028 void netif_tx_stop_all_queues(struct net_device *dev)
9032 for (i = 0; i < dev->num_tx_queues; i++) {
9033 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9035 netif_tx_stop_queue(txq);
9038 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9040 static void netdev_register_lockdep_key(struct net_device *dev)
9042 lockdep_register_key(&dev->qdisc_tx_busylock_key);
9043 lockdep_register_key(&dev->qdisc_running_key);
9044 lockdep_register_key(&dev->qdisc_xmit_lock_key);
9045 lockdep_register_key(&dev->addr_list_lock_key);
9048 static void netdev_unregister_lockdep_key(struct net_device *dev)
9050 lockdep_unregister_key(&dev->qdisc_tx_busylock_key);
9051 lockdep_unregister_key(&dev->qdisc_running_key);
9052 lockdep_unregister_key(&dev->qdisc_xmit_lock_key);
9053 lockdep_unregister_key(&dev->addr_list_lock_key);
9056 void netdev_update_lockdep_key(struct net_device *dev)
9058 lockdep_unregister_key(&dev->addr_list_lock_key);
9059 lockdep_register_key(&dev->addr_list_lock_key);
9061 lockdep_set_class(&dev->addr_list_lock, &dev->addr_list_lock_key);
9063 EXPORT_SYMBOL(netdev_update_lockdep_key);
9066 * register_netdevice - register a network device
9067 * @dev: device to register
9069 * Take a completed network device structure and add it to the kernel
9070 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9071 * chain. 0 is returned on success. A negative errno code is returned
9072 * on a failure to set up the device, or if the name is a duplicate.
9074 * Callers must hold the rtnl semaphore. You may want
9075 * register_netdev() instead of this.
9078 * The locking appears insufficient to guarantee two parallel registers
9079 * will not get the same name.
9082 int register_netdevice(struct net_device *dev)
9085 struct net *net = dev_net(dev);
9087 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9088 NETDEV_FEATURE_COUNT);
9089 BUG_ON(dev_boot_phase);
9094 /* When net_device's are persistent, this will be fatal. */
9095 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9098 spin_lock_init(&dev->addr_list_lock);
9099 lockdep_set_class(&dev->addr_list_lock, &dev->addr_list_lock_key);
9101 ret = dev_get_valid_name(net, dev, dev->name);
9105 /* Init, if this function is available */
9106 if (dev->netdev_ops->ndo_init) {
9107 ret = dev->netdev_ops->ndo_init(dev);
9115 if (((dev->hw_features | dev->features) &
9116 NETIF_F_HW_VLAN_CTAG_FILTER) &&
9117 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
9118 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
9119 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
9126 dev->ifindex = dev_new_index(net);
9127 else if (__dev_get_by_index(net, dev->ifindex))
9130 /* Transfer changeable features to wanted_features and enable
9131 * software offloads (GSO and GRO).
9133 dev->hw_features |= NETIF_F_SOFT_FEATURES;
9134 dev->features |= NETIF_F_SOFT_FEATURES;
9136 if (dev->netdev_ops->ndo_udp_tunnel_add) {
9137 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9138 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9141 dev->wanted_features = dev->features & dev->hw_features;
9143 if (!(dev->flags & IFF_LOOPBACK))
9144 dev->hw_features |= NETIF_F_NOCACHE_COPY;
9146 /* If IPv4 TCP segmentation offload is supported we should also
9147 * allow the device to enable segmenting the frame with the option
9148 * of ignoring a static IP ID value. This doesn't enable the
9149 * feature itself but allows the user to enable it later.
9151 if (dev->hw_features & NETIF_F_TSO)
9152 dev->hw_features |= NETIF_F_TSO_MANGLEID;
9153 if (dev->vlan_features & NETIF_F_TSO)
9154 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
9155 if (dev->mpls_features & NETIF_F_TSO)
9156 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
9157 if (dev->hw_enc_features & NETIF_F_TSO)
9158 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
9160 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
9162 dev->vlan_features |= NETIF_F_HIGHDMA;
9164 /* Make NETIF_F_SG inheritable to tunnel devices.
9166 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
9168 /* Make NETIF_F_SG inheritable to MPLS.
9170 dev->mpls_features |= NETIF_F_SG;
9172 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
9173 ret = notifier_to_errno(ret);
9177 ret = netdev_register_kobject(dev);
9179 dev->reg_state = NETREG_UNREGISTERED;
9182 dev->reg_state = NETREG_REGISTERED;
9184 __netdev_update_features(dev);
9187 * Default initial state at registry is that the
9188 * device is present.
9191 set_bit(__LINK_STATE_PRESENT, &dev->state);
9193 linkwatch_init_dev(dev);
9195 dev_init_scheduler(dev);
9197 list_netdevice(dev);
9198 add_device_randomness(dev->dev_addr, dev->addr_len);
9200 /* If the device has permanent device address, driver should
9201 * set dev_addr and also addr_assign_type should be set to
9202 * NET_ADDR_PERM (default value).
9204 if (dev->addr_assign_type == NET_ADDR_PERM)
9205 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
9207 /* Notify protocols, that a new device appeared. */
9208 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
9209 ret = notifier_to_errno(ret);
9211 rollback_registered(dev);
9214 dev->reg_state = NETREG_UNREGISTERED;
9215 /* We should put the kobject that hold in
9216 * netdev_unregister_kobject(), otherwise
9217 * the net device cannot be freed when
9218 * driver calls free_netdev(), because the
9219 * kobject is being hold.
9221 kobject_put(&dev->dev.kobj);
9224 * Prevent userspace races by waiting until the network
9225 * device is fully setup before sending notifications.
9227 if (!dev->rtnl_link_ops ||
9228 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
9229 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
9235 if (dev->netdev_ops->ndo_uninit)
9236 dev->netdev_ops->ndo_uninit(dev);
9237 if (dev->priv_destructor)
9238 dev->priv_destructor(dev);
9241 EXPORT_SYMBOL(register_netdevice);
9244 * init_dummy_netdev - init a dummy network device for NAPI
9245 * @dev: device to init
9247 * This takes a network device structure and initialize the minimum
9248 * amount of fields so it can be used to schedule NAPI polls without
9249 * registering a full blown interface. This is to be used by drivers
9250 * that need to tie several hardware interfaces to a single NAPI
9251 * poll scheduler due to HW limitations.
9253 int init_dummy_netdev(struct net_device *dev)
9255 /* Clear everything. Note we don't initialize spinlocks
9256 * are they aren't supposed to be taken by any of the
9257 * NAPI code and this dummy netdev is supposed to be
9258 * only ever used for NAPI polls
9260 memset(dev, 0, sizeof(struct net_device));
9262 /* make sure we BUG if trying to hit standard
9263 * register/unregister code path
9265 dev->reg_state = NETREG_DUMMY;
9267 /* NAPI wants this */
9268 INIT_LIST_HEAD(&dev->napi_list);
9270 /* a dummy interface is started by default */
9271 set_bit(__LINK_STATE_PRESENT, &dev->state);
9272 set_bit(__LINK_STATE_START, &dev->state);
9274 /* napi_busy_loop stats accounting wants this */
9275 dev_net_set(dev, &init_net);
9277 /* Note : We dont allocate pcpu_refcnt for dummy devices,
9278 * because users of this 'device' dont need to change
9284 EXPORT_SYMBOL_GPL(init_dummy_netdev);
9288 * register_netdev - register a network device
9289 * @dev: device to register
9291 * Take a completed network device structure and add it to the kernel
9292 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9293 * chain. 0 is returned on success. A negative errno code is returned
9294 * on a failure to set up the device, or if the name is a duplicate.
9296 * This is a wrapper around register_netdevice that takes the rtnl semaphore
9297 * and expands the device name if you passed a format string to
9300 int register_netdev(struct net_device *dev)
9304 if (rtnl_lock_killable())
9306 err = register_netdevice(dev);
9310 EXPORT_SYMBOL(register_netdev);
9312 int netdev_refcnt_read(const struct net_device *dev)
9316 for_each_possible_cpu(i)
9317 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
9320 EXPORT_SYMBOL(netdev_refcnt_read);
9323 * netdev_wait_allrefs - wait until all references are gone.
9324 * @dev: target net_device
9326 * This is called when unregistering network devices.
9328 * Any protocol or device that holds a reference should register
9329 * for netdevice notification, and cleanup and put back the
9330 * reference if they receive an UNREGISTER event.
9331 * We can get stuck here if buggy protocols don't correctly
9334 static void netdev_wait_allrefs(struct net_device *dev)
9336 unsigned long rebroadcast_time, warning_time;
9339 linkwatch_forget_dev(dev);
9341 rebroadcast_time = warning_time = jiffies;
9342 refcnt = netdev_refcnt_read(dev);
9344 while (refcnt != 0) {
9345 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
9348 /* Rebroadcast unregister notification */
9349 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
9355 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
9357 /* We must not have linkwatch events
9358 * pending on unregister. If this
9359 * happens, we simply run the queue
9360 * unscheduled, resulting in a noop
9363 linkwatch_run_queue();
9368 rebroadcast_time = jiffies;
9373 refcnt = netdev_refcnt_read(dev);
9375 if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) {
9376 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
9378 warning_time = jiffies;
9387 * register_netdevice(x1);
9388 * register_netdevice(x2);
9390 * unregister_netdevice(y1);
9391 * unregister_netdevice(y2);
9397 * We are invoked by rtnl_unlock().
9398 * This allows us to deal with problems:
9399 * 1) We can delete sysfs objects which invoke hotplug
9400 * without deadlocking with linkwatch via keventd.
9401 * 2) Since we run with the RTNL semaphore not held, we can sleep
9402 * safely in order to wait for the netdev refcnt to drop to zero.
9404 * We must not return until all unregister events added during
9405 * the interval the lock was held have been completed.
9407 void netdev_run_todo(void)
9409 struct list_head list;
9411 /* Snapshot list, allow later requests */
9412 list_replace_init(&net_todo_list, &list);
9417 /* Wait for rcu callbacks to finish before next phase */
9418 if (!list_empty(&list))
9421 while (!list_empty(&list)) {
9422 struct net_device *dev
9423 = list_first_entry(&list, struct net_device, todo_list);
9424 list_del(&dev->todo_list);
9426 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
9427 pr_err("network todo '%s' but state %d\n",
9428 dev->name, dev->reg_state);
9433 dev->reg_state = NETREG_UNREGISTERED;
9435 netdev_wait_allrefs(dev);
9438 BUG_ON(netdev_refcnt_read(dev));
9439 BUG_ON(!list_empty(&dev->ptype_all));
9440 BUG_ON(!list_empty(&dev->ptype_specific));
9441 WARN_ON(rcu_access_pointer(dev->ip_ptr));
9442 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
9444 if (dev->priv_destructor)
9445 dev->priv_destructor(dev);
9446 if (dev->needs_free_netdev)
9449 /* Report a network device has been unregistered */
9451 dev_net(dev)->dev_unreg_count--;
9453 wake_up(&netdev_unregistering_wq);
9455 /* Free network device */
9456 kobject_put(&dev->dev.kobj);
9460 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
9461 * all the same fields in the same order as net_device_stats, with only
9462 * the type differing, but rtnl_link_stats64 may have additional fields
9463 * at the end for newer counters.
9465 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
9466 const struct net_device_stats *netdev_stats)
9468 size_t i, n = sizeof(*netdev_stats) / sizeof(atomic_long_t);
9469 const atomic_long_t *src = (atomic_long_t *)netdev_stats;
9470 u64 *dst = (u64 *)stats64;
9472 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
9473 for (i = 0; i < n; i++)
9474 dst[i] = (unsigned long)atomic_long_read(&src[i]);
9475 /* zero out counters that only exist in rtnl_link_stats64 */
9476 memset((char *)stats64 + n * sizeof(u64), 0,
9477 sizeof(*stats64) - n * sizeof(u64));
9479 EXPORT_SYMBOL(netdev_stats_to_stats64);
9482 * dev_get_stats - get network device statistics
9483 * @dev: device to get statistics from
9484 * @storage: place to store stats
9486 * Get network statistics from device. Return @storage.
9487 * The device driver may provide its own method by setting
9488 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
9489 * otherwise the internal statistics structure is used.
9491 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
9492 struct rtnl_link_stats64 *storage)
9494 const struct net_device_ops *ops = dev->netdev_ops;
9496 if (ops->ndo_get_stats64) {
9497 memset(storage, 0, sizeof(*storage));
9498 ops->ndo_get_stats64(dev, storage);
9499 } else if (ops->ndo_get_stats) {
9500 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
9502 netdev_stats_to_stats64(storage, &dev->stats);
9504 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
9505 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
9506 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
9509 EXPORT_SYMBOL(dev_get_stats);
9511 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
9513 struct netdev_queue *queue = dev_ingress_queue(dev);
9515 #ifdef CONFIG_NET_CLS_ACT
9518 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
9521 netdev_init_one_queue(dev, queue, NULL);
9522 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
9523 queue->qdisc_sleeping = &noop_qdisc;
9524 rcu_assign_pointer(dev->ingress_queue, queue);
9529 static const struct ethtool_ops default_ethtool_ops;
9531 void netdev_set_default_ethtool_ops(struct net_device *dev,
9532 const struct ethtool_ops *ops)
9534 if (dev->ethtool_ops == &default_ethtool_ops)
9535 dev->ethtool_ops = ops;
9537 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
9539 void netdev_freemem(struct net_device *dev)
9541 char *addr = (char *)dev - dev->padded;
9547 * alloc_netdev_mqs - allocate network device
9548 * @sizeof_priv: size of private data to allocate space for
9549 * @name: device name format string
9550 * @name_assign_type: origin of device name
9551 * @setup: callback to initialize device
9552 * @txqs: the number of TX subqueues to allocate
9553 * @rxqs: the number of RX subqueues to allocate
9555 * Allocates a struct net_device with private data area for driver use
9556 * and performs basic initialization. Also allocates subqueue structs
9557 * for each queue on the device.
9559 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
9560 unsigned char name_assign_type,
9561 void (*setup)(struct net_device *),
9562 unsigned int txqs, unsigned int rxqs)
9564 struct net_device *dev;
9565 unsigned int alloc_size;
9566 struct net_device *p;
9568 BUG_ON(strlen(name) >= sizeof(dev->name));
9571 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
9576 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
9580 alloc_size = sizeof(struct net_device);
9582 /* ensure 32-byte alignment of private area */
9583 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
9584 alloc_size += sizeof_priv;
9586 /* ensure 32-byte alignment of whole construct */
9587 alloc_size += NETDEV_ALIGN - 1;
9589 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9593 dev = PTR_ALIGN(p, NETDEV_ALIGN);
9594 dev->padded = (char *)dev - (char *)p;
9596 dev->pcpu_refcnt = alloc_percpu(int);
9597 if (!dev->pcpu_refcnt)
9600 if (dev_addr_init(dev))
9606 dev_net_set(dev, &init_net);
9608 netdev_register_lockdep_key(dev);
9610 dev->gso_max_size = GSO_MAX_SIZE;
9611 dev->gso_max_segs = GSO_MAX_SEGS;
9612 dev->upper_level = 1;
9613 dev->lower_level = 1;
9615 INIT_LIST_HEAD(&dev->napi_list);
9616 INIT_LIST_HEAD(&dev->unreg_list);
9617 INIT_LIST_HEAD(&dev->close_list);
9618 INIT_LIST_HEAD(&dev->link_watch_list);
9619 INIT_LIST_HEAD(&dev->adj_list.upper);
9620 INIT_LIST_HEAD(&dev->adj_list.lower);
9621 INIT_LIST_HEAD(&dev->ptype_all);
9622 INIT_LIST_HEAD(&dev->ptype_specific);
9623 #ifdef CONFIG_NET_SCHED
9624 hash_init(dev->qdisc_hash);
9626 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
9629 if (!dev->tx_queue_len) {
9630 dev->priv_flags |= IFF_NO_QUEUE;
9631 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
9634 dev->num_tx_queues = txqs;
9635 dev->real_num_tx_queues = txqs;
9636 if (netif_alloc_netdev_queues(dev))
9639 dev->num_rx_queues = rxqs;
9640 dev->real_num_rx_queues = rxqs;
9641 if (netif_alloc_rx_queues(dev))
9644 strcpy(dev->name, name);
9645 dev->name_assign_type = name_assign_type;
9646 dev->group = INIT_NETDEV_GROUP;
9647 if (!dev->ethtool_ops)
9648 dev->ethtool_ops = &default_ethtool_ops;
9650 nf_hook_ingress_init(dev);
9659 free_percpu(dev->pcpu_refcnt);
9661 netdev_freemem(dev);
9664 EXPORT_SYMBOL(alloc_netdev_mqs);
9667 * free_netdev - free network device
9670 * This function does the last stage of destroying an allocated device
9671 * interface. The reference to the device object is released. If this
9672 * is the last reference then it will be freed.Must be called in process
9675 void free_netdev(struct net_device *dev)
9677 struct napi_struct *p, *n;
9680 netif_free_tx_queues(dev);
9681 netif_free_rx_queues(dev);
9683 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
9685 /* Flush device addresses */
9686 dev_addr_flush(dev);
9688 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
9691 free_percpu(dev->pcpu_refcnt);
9692 dev->pcpu_refcnt = NULL;
9694 netdev_unregister_lockdep_key(dev);
9696 /* Compatibility with error handling in drivers */
9697 if (dev->reg_state == NETREG_UNINITIALIZED) {
9698 netdev_freemem(dev);
9702 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
9703 dev->reg_state = NETREG_RELEASED;
9705 /* will free via device release */
9706 put_device(&dev->dev);
9708 EXPORT_SYMBOL(free_netdev);
9711 * synchronize_net - Synchronize with packet receive processing
9713 * Wait for packets currently being received to be done.
9714 * Does not block later packets from starting.
9716 void synchronize_net(void)
9719 if (rtnl_is_locked())
9720 synchronize_rcu_expedited();
9724 EXPORT_SYMBOL(synchronize_net);
9727 * unregister_netdevice_queue - remove device from the kernel
9731 * This function shuts down a device interface and removes it
9732 * from the kernel tables.
9733 * If head not NULL, device is queued to be unregistered later.
9735 * Callers must hold the rtnl semaphore. You may want
9736 * unregister_netdev() instead of this.
9739 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
9744 list_move_tail(&dev->unreg_list, head);
9746 rollback_registered(dev);
9747 /* Finish processing unregister after unlock */
9751 EXPORT_SYMBOL(unregister_netdevice_queue);
9754 * unregister_netdevice_many - unregister many devices
9755 * @head: list of devices
9757 * Note: As most callers use a stack allocated list_head,
9758 * we force a list_del() to make sure stack wont be corrupted later.
9760 void unregister_netdevice_many(struct list_head *head)
9762 struct net_device *dev;
9764 if (!list_empty(head)) {
9765 rollback_registered_many(head);
9766 list_for_each_entry(dev, head, unreg_list)
9771 EXPORT_SYMBOL(unregister_netdevice_many);
9774 * unregister_netdev - remove device from the kernel
9777 * This function shuts down a device interface and removes it
9778 * from the kernel tables.
9780 * This is just a wrapper for unregister_netdevice that takes
9781 * the rtnl semaphore. In general you want to use this and not
9782 * unregister_netdevice.
9784 void unregister_netdev(struct net_device *dev)
9787 unregister_netdevice(dev);
9790 EXPORT_SYMBOL(unregister_netdev);
9793 * dev_change_net_namespace - move device to different nethost namespace
9795 * @net: network namespace
9796 * @pat: If not NULL name pattern to try if the current device name
9797 * is already taken in the destination network namespace.
9799 * This function shuts down a device interface and moves it
9800 * to a new network namespace. On success 0 is returned, on
9801 * a failure a netagive errno code is returned.
9803 * Callers must hold the rtnl semaphore.
9806 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
9808 int err, new_nsid, new_ifindex;
9812 /* Don't allow namespace local devices to be moved. */
9814 if (dev->features & NETIF_F_NETNS_LOCAL)
9817 /* Ensure the device has been registrered */
9818 if (dev->reg_state != NETREG_REGISTERED)
9821 /* Get out if there is nothing todo */
9823 if (net_eq(dev_net(dev), net))
9826 /* Pick the destination device name, and ensure
9827 * we can use it in the destination network namespace.
9830 if (__dev_get_by_name(net, dev->name)) {
9831 /* We get here if we can't use the current device name */
9834 err = dev_get_valid_name(net, dev, pat);
9840 * And now a mini version of register_netdevice unregister_netdevice.
9843 /* If device is running close it first. */
9846 /* And unlink it from device chain */
9847 unlist_netdevice(dev);
9851 /* Shutdown queueing discipline. */
9854 /* Notify protocols, that we are about to destroy
9855 * this device. They should clean all the things.
9857 * Note that dev->reg_state stays at NETREG_REGISTERED.
9858 * This is wanted because this way 8021q and macvlan know
9859 * the device is just moving and can keep their slaves up.
9861 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
9864 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
9865 /* If there is an ifindex conflict assign a new one */
9866 if (__dev_get_by_index(net, dev->ifindex))
9867 new_ifindex = dev_new_index(net);
9869 new_ifindex = dev->ifindex;
9871 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
9875 * Flush the unicast and multicast chains
9880 /* Send a netdev-removed uevent to the old namespace */
9881 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
9882 netdev_adjacent_del_links(dev);
9884 /* Actually switch the network namespace */
9885 dev_net_set(dev, net);
9886 dev->ifindex = new_ifindex;
9888 /* Send a netdev-add uevent to the new namespace */
9889 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
9890 netdev_adjacent_add_links(dev);
9892 /* Fixup kobjects */
9893 err = device_rename(&dev->dev, dev->name);
9896 /* Add the device back in the hashes */
9897 list_netdevice(dev);
9899 /* Notify protocols, that a new device appeared. */
9900 call_netdevice_notifiers(NETDEV_REGISTER, dev);
9903 * Prevent userspace races by waiting until the network
9904 * device is fully setup before sending notifications.
9906 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
9913 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
9915 static int dev_cpu_dead(unsigned int oldcpu)
9917 struct sk_buff **list_skb;
9918 struct sk_buff *skb;
9920 struct softnet_data *sd, *oldsd, *remsd = NULL;
9922 local_irq_disable();
9923 cpu = smp_processor_id();
9924 sd = &per_cpu(softnet_data, cpu);
9925 oldsd = &per_cpu(softnet_data, oldcpu);
9927 /* Find end of our completion_queue. */
9928 list_skb = &sd->completion_queue;
9930 list_skb = &(*list_skb)->next;
9931 /* Append completion queue from offline CPU. */
9932 *list_skb = oldsd->completion_queue;
9933 oldsd->completion_queue = NULL;
9935 /* Append output queue from offline CPU. */
9936 if (oldsd->output_queue) {
9937 *sd->output_queue_tailp = oldsd->output_queue;
9938 sd->output_queue_tailp = oldsd->output_queue_tailp;
9939 oldsd->output_queue = NULL;
9940 oldsd->output_queue_tailp = &oldsd->output_queue;
9942 /* Append NAPI poll list from offline CPU, with one exception :
9943 * process_backlog() must be called by cpu owning percpu backlog.
9944 * We properly handle process_queue & input_pkt_queue later.
9946 while (!list_empty(&oldsd->poll_list)) {
9947 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
9951 list_del_init(&napi->poll_list);
9952 if (napi->poll == process_backlog)
9955 ____napi_schedule(sd, napi);
9958 raise_softirq_irqoff(NET_TX_SOFTIRQ);
9962 remsd = oldsd->rps_ipi_list;
9963 oldsd->rps_ipi_list = NULL;
9965 /* send out pending IPI's on offline CPU */
9966 net_rps_send_ipi(remsd);
9968 /* Process offline CPU's input_pkt_queue */
9969 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
9971 input_queue_head_incr(oldsd);
9973 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
9975 input_queue_head_incr(oldsd);
9982 * netdev_increment_features - increment feature set by one
9983 * @all: current feature set
9984 * @one: new feature set
9985 * @mask: mask feature set
9987 * Computes a new feature set after adding a device with feature set
9988 * @one to the master device with current feature set @all. Will not
9989 * enable anything that is off in @mask. Returns the new feature set.
9991 netdev_features_t netdev_increment_features(netdev_features_t all,
9992 netdev_features_t one, netdev_features_t mask)
9994 if (mask & NETIF_F_HW_CSUM)
9995 mask |= NETIF_F_CSUM_MASK;
9996 mask |= NETIF_F_VLAN_CHALLENGED;
9998 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
9999 all &= one | ~NETIF_F_ALL_FOR_ALL;
10001 /* If one device supports hw checksumming, set for all. */
10002 if (all & NETIF_F_HW_CSUM)
10003 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
10007 EXPORT_SYMBOL(netdev_increment_features);
10009 static struct hlist_head * __net_init netdev_create_hash(void)
10012 struct hlist_head *hash;
10014 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
10016 for (i = 0; i < NETDEV_HASHENTRIES; i++)
10017 INIT_HLIST_HEAD(&hash[i]);
10022 /* Initialize per network namespace state */
10023 static int __net_init netdev_init(struct net *net)
10025 BUILD_BUG_ON(GRO_HASH_BUCKETS >
10026 8 * FIELD_SIZEOF(struct napi_struct, gro_bitmask));
10028 if (net != &init_net)
10029 INIT_LIST_HEAD(&net->dev_base_head);
10031 net->dev_name_head = netdev_create_hash();
10032 if (net->dev_name_head == NULL)
10035 net->dev_index_head = netdev_create_hash();
10036 if (net->dev_index_head == NULL)
10042 kfree(net->dev_name_head);
10048 * netdev_drivername - network driver for the device
10049 * @dev: network device
10051 * Determine network driver for device.
10053 const char *netdev_drivername(const struct net_device *dev)
10055 const struct device_driver *driver;
10056 const struct device *parent;
10057 const char *empty = "";
10059 parent = dev->dev.parent;
10063 driver = parent->driver;
10064 if (driver && driver->name)
10065 return driver->name;
10069 static void __netdev_printk(const char *level, const struct net_device *dev,
10070 struct va_format *vaf)
10072 if (dev && dev->dev.parent) {
10073 dev_printk_emit(level[1] - '0',
10076 dev_driver_string(dev->dev.parent),
10077 dev_name(dev->dev.parent),
10078 netdev_name(dev), netdev_reg_state(dev),
10081 printk("%s%s%s: %pV",
10082 level, netdev_name(dev), netdev_reg_state(dev), vaf);
10084 printk("%s(NULL net_device): %pV", level, vaf);
10088 void netdev_printk(const char *level, const struct net_device *dev,
10089 const char *format, ...)
10091 struct va_format vaf;
10094 va_start(args, format);
10099 __netdev_printk(level, dev, &vaf);
10103 EXPORT_SYMBOL(netdev_printk);
10105 #define define_netdev_printk_level(func, level) \
10106 void func(const struct net_device *dev, const char *fmt, ...) \
10108 struct va_format vaf; \
10111 va_start(args, fmt); \
10116 __netdev_printk(level, dev, &vaf); \
10120 EXPORT_SYMBOL(func);
10122 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
10123 define_netdev_printk_level(netdev_alert, KERN_ALERT);
10124 define_netdev_printk_level(netdev_crit, KERN_CRIT);
10125 define_netdev_printk_level(netdev_err, KERN_ERR);
10126 define_netdev_printk_level(netdev_warn, KERN_WARNING);
10127 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
10128 define_netdev_printk_level(netdev_info, KERN_INFO);
10130 static void __net_exit netdev_exit(struct net *net)
10132 kfree(net->dev_name_head);
10133 kfree(net->dev_index_head);
10134 if (net != &init_net)
10135 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
10138 static struct pernet_operations __net_initdata netdev_net_ops = {
10139 .init = netdev_init,
10140 .exit = netdev_exit,
10143 static void __net_exit default_device_exit(struct net *net)
10145 struct net_device *dev, *aux;
10147 * Push all migratable network devices back to the
10148 * initial network namespace
10151 for_each_netdev_safe(net, dev, aux) {
10153 char fb_name[IFNAMSIZ];
10155 /* Ignore unmoveable devices (i.e. loopback) */
10156 if (dev->features & NETIF_F_NETNS_LOCAL)
10159 /* Leave virtual devices for the generic cleanup */
10160 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
10163 /* Push remaining network devices to init_net */
10164 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
10165 if (__dev_get_by_name(&init_net, fb_name))
10166 snprintf(fb_name, IFNAMSIZ, "dev%%d");
10167 err = dev_change_net_namespace(dev, &init_net, fb_name);
10169 pr_emerg("%s: failed to move %s to init_net: %d\n",
10170 __func__, dev->name, err);
10177 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
10179 /* Return with the rtnl_lock held when there are no network
10180 * devices unregistering in any network namespace in net_list.
10183 bool unregistering;
10184 DEFINE_WAIT_FUNC(wait, woken_wake_function);
10186 add_wait_queue(&netdev_unregistering_wq, &wait);
10188 unregistering = false;
10190 list_for_each_entry(net, net_list, exit_list) {
10191 if (net->dev_unreg_count > 0) {
10192 unregistering = true;
10196 if (!unregistering)
10200 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
10202 remove_wait_queue(&netdev_unregistering_wq, &wait);
10205 static void __net_exit default_device_exit_batch(struct list_head *net_list)
10207 /* At exit all network devices most be removed from a network
10208 * namespace. Do this in the reverse order of registration.
10209 * Do this across as many network namespaces as possible to
10210 * improve batching efficiency.
10212 struct net_device *dev;
10214 LIST_HEAD(dev_kill_list);
10216 /* To prevent network device cleanup code from dereferencing
10217 * loopback devices or network devices that have been freed
10218 * wait here for all pending unregistrations to complete,
10219 * before unregistring the loopback device and allowing the
10220 * network namespace be freed.
10222 * The netdev todo list containing all network devices
10223 * unregistrations that happen in default_device_exit_batch
10224 * will run in the rtnl_unlock() at the end of
10225 * default_device_exit_batch.
10227 rtnl_lock_unregistering(net_list);
10228 list_for_each_entry(net, net_list, exit_list) {
10229 for_each_netdev_reverse(net, dev) {
10230 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
10231 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
10233 unregister_netdevice_queue(dev, &dev_kill_list);
10236 unregister_netdevice_many(&dev_kill_list);
10240 static struct pernet_operations __net_initdata default_device_ops = {
10241 .exit = default_device_exit,
10242 .exit_batch = default_device_exit_batch,
10246 * Initialize the DEV module. At boot time this walks the device list and
10247 * unhooks any devices that fail to initialise (normally hardware not
10248 * present) and leaves us with a valid list of present and active devices.
10253 * This is called single threaded during boot, so no need
10254 * to take the rtnl semaphore.
10256 static int __init net_dev_init(void)
10258 int i, rc = -ENOMEM;
10260 BUG_ON(!dev_boot_phase);
10262 if (dev_proc_init())
10265 if (netdev_kobject_init())
10268 INIT_LIST_HEAD(&ptype_all);
10269 for (i = 0; i < PTYPE_HASH_SIZE; i++)
10270 INIT_LIST_HEAD(&ptype_base[i]);
10272 INIT_LIST_HEAD(&offload_base);
10274 if (register_pernet_subsys(&netdev_net_ops))
10278 * Initialise the packet receive queues.
10281 for_each_possible_cpu(i) {
10282 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
10283 struct softnet_data *sd = &per_cpu(softnet_data, i);
10285 INIT_WORK(flush, flush_backlog);
10287 skb_queue_head_init(&sd->input_pkt_queue);
10288 skb_queue_head_init(&sd->process_queue);
10289 #ifdef CONFIG_XFRM_OFFLOAD
10290 skb_queue_head_init(&sd->xfrm_backlog);
10292 INIT_LIST_HEAD(&sd->poll_list);
10293 sd->output_queue_tailp = &sd->output_queue;
10295 sd->csd.func = rps_trigger_softirq;
10300 init_gro_hash(&sd->backlog);
10301 sd->backlog.poll = process_backlog;
10302 sd->backlog.weight = weight_p;
10305 dev_boot_phase = 0;
10307 /* The loopback device is special if any other network devices
10308 * is present in a network namespace the loopback device must
10309 * be present. Since we now dynamically allocate and free the
10310 * loopback device ensure this invariant is maintained by
10311 * keeping the loopback device as the first device on the
10312 * list of network devices. Ensuring the loopback devices
10313 * is the first device that appears and the last network device
10316 if (register_pernet_device(&loopback_net_ops))
10319 if (register_pernet_device(&default_device_ops))
10322 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
10323 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
10325 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
10326 NULL, dev_cpu_dead);
10333 subsys_initcall(net_dev_init);