2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.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/checksum.h>
105 #include <net/xfrm.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/module.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #include <linux/delay.h>
112 #include <net/iw_handler.h>
113 #include <asm/current.h>
114 #include <linux/audit.h>
115 #include <linux/dmaengine.h>
116 #include <linux/err.h>
117 #include <linux/ctype.h>
118 #include <linux/if_arp.h>
119 #include <linux/if_vlan.h>
120 #include <linux/ip.h>
122 #include <net/mpls.h>
123 #include <linux/ipv6.h>
124 #include <linux/in.h>
125 #include <linux/jhash.h>
126 #include <linux/random.h>
127 #include <trace/events/napi.h>
128 #include <trace/events/net.h>
129 #include <trace/events/skb.h>
130 #include <linux/pci.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>
141 #include "net-sysfs.h"
143 /* Instead of increasing this, you should create a hash table. */
144 #define MAX_GRO_SKBS 8
146 /* This should be increased if a protocol with a bigger head is added. */
147 #define GRO_MAX_HEAD (MAX_HEADER + 128)
149 static DEFINE_SPINLOCK(ptype_lock);
150 static DEFINE_SPINLOCK(offload_lock);
151 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
152 struct list_head ptype_all __read_mostly; /* Taps */
153 static struct list_head offload_base __read_mostly;
155 static int netif_rx_internal(struct sk_buff *skb);
156 static int call_netdevice_notifiers_info(unsigned long val,
157 struct net_device *dev,
158 struct netdev_notifier_info *info);
161 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
164 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
166 * Writers must hold the rtnl semaphore while they loop through the
167 * dev_base_head list, and hold dev_base_lock for writing when they do the
168 * actual updates. This allows pure readers to access the list even
169 * while a writer is preparing to update it.
171 * To put it another way, dev_base_lock is held for writing only to
172 * protect against pure readers; the rtnl semaphore provides the
173 * protection against other writers.
175 * See, for example usages, register_netdevice() and
176 * unregister_netdevice(), which must be called with the rtnl
179 DEFINE_RWLOCK(dev_base_lock);
180 EXPORT_SYMBOL(dev_base_lock);
182 /* protects napi_hash addition/deletion and napi_gen_id */
183 static DEFINE_SPINLOCK(napi_hash_lock);
185 static unsigned int napi_gen_id;
186 static DEFINE_HASHTABLE(napi_hash, 8);
188 static seqcount_t devnet_rename_seq;
189 static DEFINE_MUTEX(devnet_rename_mutex);
191 static inline void dev_base_seq_inc(struct net *net)
193 while (++net->dev_base_seq == 0);
196 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
198 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
200 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
203 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
205 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
208 static inline void rps_lock(struct softnet_data *sd)
211 raw_spin_lock(&sd->input_pkt_queue.raw_lock);
215 static inline void rps_unlock(struct softnet_data *sd)
218 raw_spin_unlock(&sd->input_pkt_queue.raw_lock);
222 /* Device list insertion */
223 static void list_netdevice(struct net_device *dev)
225 struct net *net = dev_net(dev);
229 write_lock_bh(&dev_base_lock);
230 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
231 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
232 hlist_add_head_rcu(&dev->index_hlist,
233 dev_index_hash(net, dev->ifindex));
234 write_unlock_bh(&dev_base_lock);
236 dev_base_seq_inc(net);
239 /* Device list removal
240 * caller must respect a RCU grace period before freeing/reusing dev
242 static void unlist_netdevice(struct net_device *dev)
246 /* Unlink dev from the device chain */
247 write_lock_bh(&dev_base_lock);
248 list_del_rcu(&dev->dev_list);
249 hlist_del_rcu(&dev->name_hlist);
250 hlist_del_rcu(&dev->index_hlist);
251 write_unlock_bh(&dev_base_lock);
253 dev_base_seq_inc(dev_net(dev));
260 static RAW_NOTIFIER_HEAD(netdev_chain);
263 * Device drivers call our routines to queue packets here. We empty the
264 * queue in the local softnet handler.
267 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
268 EXPORT_PER_CPU_SYMBOL(softnet_data);
270 #ifdef CONFIG_LOCKDEP
272 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
273 * according to dev->type
275 static const unsigned short netdev_lock_type[] =
276 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
277 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
278 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
279 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
280 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
281 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
282 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
283 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
284 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
285 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
286 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
287 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
288 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
289 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
290 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
292 static const char *const netdev_lock_name[] =
293 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
294 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
295 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
296 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
297 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
298 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
299 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
300 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
301 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
302 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
303 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
304 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
305 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
306 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
307 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
309 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
310 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
312 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
316 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
317 if (netdev_lock_type[i] == dev_type)
319 /* the last key is used by default */
320 return ARRAY_SIZE(netdev_lock_type) - 1;
323 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
324 unsigned short dev_type)
328 i = netdev_lock_pos(dev_type);
329 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
330 netdev_lock_name[i]);
333 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
337 i = netdev_lock_pos(dev->type);
338 lockdep_set_class_and_name(&dev->addr_list_lock,
339 &netdev_addr_lock_key[i],
340 netdev_lock_name[i]);
343 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
344 unsigned short dev_type)
347 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
352 /*******************************************************************************
354 Protocol management and registration routines
356 *******************************************************************************/
359 * Add a protocol ID to the list. Now that the input handler is
360 * smarter we can dispense with all the messy stuff that used to be
363 * BEWARE!!! Protocol handlers, mangling input packets,
364 * MUST BE last in hash buckets and checking protocol handlers
365 * MUST start from promiscuous ptype_all chain in net_bh.
366 * It is true now, do not change it.
367 * Explanation follows: if protocol handler, mangling packet, will
368 * be the first on list, it is not able to sense, that packet
369 * is cloned and should be copied-on-write, so that it will
370 * change it and subsequent readers will get broken packet.
374 static inline struct list_head *ptype_head(const struct packet_type *pt)
376 if (pt->type == htons(ETH_P_ALL))
377 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
379 return pt->dev ? &pt->dev->ptype_specific :
380 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
384 * dev_add_pack - add packet handler
385 * @pt: packet type declaration
387 * Add a protocol handler to the networking stack. The passed &packet_type
388 * is linked into kernel lists and may not be freed until it has been
389 * removed from the kernel lists.
391 * This call does not sleep therefore it can not
392 * guarantee all CPU's that are in middle of receiving packets
393 * will see the new packet type (until the next received packet).
396 void dev_add_pack(struct packet_type *pt)
398 struct list_head *head = ptype_head(pt);
400 spin_lock(&ptype_lock);
401 list_add_rcu(&pt->list, head);
402 spin_unlock(&ptype_lock);
404 EXPORT_SYMBOL(dev_add_pack);
407 * __dev_remove_pack - remove packet handler
408 * @pt: packet type declaration
410 * Remove a protocol handler that was previously added to the kernel
411 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
412 * from the kernel lists and can be freed or reused once this function
415 * The packet type might still be in use by receivers
416 * and must not be freed until after all the CPU's have gone
417 * through a quiescent state.
419 void __dev_remove_pack(struct packet_type *pt)
421 struct list_head *head = ptype_head(pt);
422 struct packet_type *pt1;
424 spin_lock(&ptype_lock);
426 list_for_each_entry(pt1, head, list) {
428 list_del_rcu(&pt->list);
433 pr_warn("dev_remove_pack: %p not found\n", pt);
435 spin_unlock(&ptype_lock);
437 EXPORT_SYMBOL(__dev_remove_pack);
440 * dev_remove_pack - remove packet handler
441 * @pt: packet type declaration
443 * Remove a protocol handler that was previously added to the kernel
444 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
445 * from the kernel lists and can be freed or reused once this function
448 * This call sleeps to guarantee that no CPU is looking at the packet
451 void dev_remove_pack(struct packet_type *pt)
453 __dev_remove_pack(pt);
457 EXPORT_SYMBOL(dev_remove_pack);
461 * dev_add_offload - register offload handlers
462 * @po: protocol offload declaration
464 * Add protocol offload handlers to the networking stack. The passed
465 * &proto_offload is linked into kernel lists and may not be freed until
466 * it has been removed from the kernel lists.
468 * This call does not sleep therefore it can not
469 * guarantee all CPU's that are in middle of receiving packets
470 * will see the new offload handlers (until the next received packet).
472 void dev_add_offload(struct packet_offload *po)
474 struct packet_offload *elem;
476 spin_lock(&offload_lock);
477 list_for_each_entry(elem, &offload_base, list) {
478 if (po->priority < elem->priority)
481 list_add_rcu(&po->list, elem->list.prev);
482 spin_unlock(&offload_lock);
484 EXPORT_SYMBOL(dev_add_offload);
487 * __dev_remove_offload - remove offload handler
488 * @po: packet offload declaration
490 * Remove a protocol offload handler that was previously added to the
491 * kernel offload handlers by dev_add_offload(). The passed &offload_type
492 * is removed from the kernel lists and can be freed or reused once this
495 * The packet type might still be in use by receivers
496 * and must not be freed until after all the CPU's have gone
497 * through a quiescent state.
499 static void __dev_remove_offload(struct packet_offload *po)
501 struct list_head *head = &offload_base;
502 struct packet_offload *po1;
504 spin_lock(&offload_lock);
506 list_for_each_entry(po1, head, list) {
508 list_del_rcu(&po->list);
513 pr_warn("dev_remove_offload: %p not found\n", po);
515 spin_unlock(&offload_lock);
519 * dev_remove_offload - remove packet offload handler
520 * @po: packet offload declaration
522 * Remove a packet offload handler that was previously added to the kernel
523 * offload handlers by dev_add_offload(). The passed &offload_type is
524 * removed from the kernel lists and can be freed or reused once this
527 * This call sleeps to guarantee that no CPU is looking at the packet
530 void dev_remove_offload(struct packet_offload *po)
532 __dev_remove_offload(po);
536 EXPORT_SYMBOL(dev_remove_offload);
538 /******************************************************************************
540 Device Boot-time Settings Routines
542 *******************************************************************************/
544 /* Boot time configuration table */
545 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
548 * netdev_boot_setup_add - add new setup entry
549 * @name: name of the device
550 * @map: configured settings for the device
552 * Adds new setup entry to the dev_boot_setup list. The function
553 * returns 0 on error and 1 on success. This is a generic routine to
556 static int netdev_boot_setup_add(char *name, struct ifmap *map)
558 struct netdev_boot_setup *s;
562 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
563 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
564 memset(s[i].name, 0, sizeof(s[i].name));
565 strlcpy(s[i].name, name, IFNAMSIZ);
566 memcpy(&s[i].map, map, sizeof(s[i].map));
571 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
575 * netdev_boot_setup_check - check boot time settings
576 * @dev: the netdevice
578 * Check boot time settings for the device.
579 * The found settings are set for the device to be used
580 * later in the device probing.
581 * Returns 0 if no settings found, 1 if they are.
583 int netdev_boot_setup_check(struct net_device *dev)
585 struct netdev_boot_setup *s = dev_boot_setup;
588 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
589 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
590 !strcmp(dev->name, s[i].name)) {
591 dev->irq = s[i].map.irq;
592 dev->base_addr = s[i].map.base_addr;
593 dev->mem_start = s[i].map.mem_start;
594 dev->mem_end = s[i].map.mem_end;
600 EXPORT_SYMBOL(netdev_boot_setup_check);
604 * netdev_boot_base - get address from boot time settings
605 * @prefix: prefix for network device
606 * @unit: id for network device
608 * Check boot time settings for the base address of device.
609 * The found settings are set for the device to be used
610 * later in the device probing.
611 * Returns 0 if no settings found.
613 unsigned long netdev_boot_base(const char *prefix, int unit)
615 const struct netdev_boot_setup *s = dev_boot_setup;
619 sprintf(name, "%s%d", prefix, unit);
622 * If device already registered then return base of 1
623 * to indicate not to probe for this interface
625 if (__dev_get_by_name(&init_net, name))
628 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
629 if (!strcmp(name, s[i].name))
630 return s[i].map.base_addr;
635 * Saves at boot time configured settings for any netdevice.
637 int __init netdev_boot_setup(char *str)
642 str = get_options(str, ARRAY_SIZE(ints), ints);
647 memset(&map, 0, sizeof(map));
651 map.base_addr = ints[2];
653 map.mem_start = ints[3];
655 map.mem_end = ints[4];
657 /* Add new entry to the list */
658 return netdev_boot_setup_add(str, &map);
661 __setup("netdev=", netdev_boot_setup);
663 /*******************************************************************************
665 Device Interface Subroutines
667 *******************************************************************************/
670 * dev_get_iflink - get 'iflink' value of a interface
671 * @dev: targeted interface
673 * Indicates the ifindex the interface is linked to.
674 * Physical interfaces have the same 'ifindex' and 'iflink' values.
677 int dev_get_iflink(const struct net_device *dev)
679 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
680 return dev->netdev_ops->ndo_get_iflink(dev);
684 EXPORT_SYMBOL(dev_get_iflink);
687 * dev_fill_metadata_dst - Retrieve tunnel egress information.
688 * @dev: targeted interface
691 * For better visibility of tunnel traffic OVS needs to retrieve
692 * egress tunnel information for a packet. Following API allows
693 * user to get this info.
695 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
697 struct ip_tunnel_info *info;
699 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
702 info = skb_tunnel_info_unclone(skb);
705 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
708 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
710 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
713 * __dev_get_by_name - find a device by its name
714 * @net: the applicable net namespace
715 * @name: name to find
717 * Find an interface by name. Must be called under RTNL semaphore
718 * or @dev_base_lock. If the name is found a pointer to the device
719 * is returned. If the name is not found then %NULL is returned. The
720 * reference counters are not incremented so the caller must be
721 * careful with locks.
724 struct net_device *__dev_get_by_name(struct net *net, const char *name)
726 struct net_device *dev;
727 struct hlist_head *head = dev_name_hash(net, name);
729 hlist_for_each_entry(dev, head, name_hlist)
730 if (!strncmp(dev->name, name, IFNAMSIZ))
735 EXPORT_SYMBOL(__dev_get_by_name);
738 * dev_get_by_name_rcu - find a device by its name
739 * @net: the applicable net namespace
740 * @name: name to find
742 * Find an interface by name.
743 * If the name is found a pointer to the device is returned.
744 * If the name is not found then %NULL is returned.
745 * The reference counters are not incremented so the caller must be
746 * careful with locks. The caller must hold RCU lock.
749 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
751 struct net_device *dev;
752 struct hlist_head *head = dev_name_hash(net, name);
754 hlist_for_each_entry_rcu(dev, head, name_hlist)
755 if (!strncmp(dev->name, name, IFNAMSIZ))
760 EXPORT_SYMBOL(dev_get_by_name_rcu);
763 * dev_get_by_name - find a device by its name
764 * @net: the applicable net namespace
765 * @name: name to find
767 * Find an interface by name. This can be called from any
768 * context and does its own locking. The returned handle has
769 * the usage count incremented and the caller must use dev_put() to
770 * release it when it is no longer needed. %NULL is returned if no
771 * matching device is found.
774 struct net_device *dev_get_by_name(struct net *net, const char *name)
776 struct net_device *dev;
779 dev = dev_get_by_name_rcu(net, name);
785 EXPORT_SYMBOL(dev_get_by_name);
788 * __dev_get_by_index - find a device by its ifindex
789 * @net: the applicable net namespace
790 * @ifindex: index of device
792 * Search for an interface by index. Returns %NULL if the device
793 * is not found or a pointer to the device. The device has not
794 * had its reference counter increased so the caller must be careful
795 * about locking. The caller must hold either the RTNL semaphore
799 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
801 struct net_device *dev;
802 struct hlist_head *head = dev_index_hash(net, ifindex);
804 hlist_for_each_entry(dev, head, index_hlist)
805 if (dev->ifindex == ifindex)
810 EXPORT_SYMBOL(__dev_get_by_index);
813 * dev_get_by_index_rcu - find a device by its ifindex
814 * @net: the applicable net namespace
815 * @ifindex: index of device
817 * Search for an interface by index. Returns %NULL if the device
818 * is not found or a pointer to the device. The device has not
819 * had its reference counter increased so the caller must be careful
820 * about locking. The caller must hold RCU lock.
823 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
825 struct net_device *dev;
826 struct hlist_head *head = dev_index_hash(net, ifindex);
828 hlist_for_each_entry_rcu(dev, head, index_hlist)
829 if (dev->ifindex == ifindex)
834 EXPORT_SYMBOL(dev_get_by_index_rcu);
838 * dev_get_by_index - find a device by its ifindex
839 * @net: the applicable net namespace
840 * @ifindex: index of device
842 * Search for an interface by index. Returns NULL if the device
843 * is not found or a pointer to the device. The device returned has
844 * had a reference added and the pointer is safe until the user calls
845 * dev_put to indicate they have finished with it.
848 struct net_device *dev_get_by_index(struct net *net, int ifindex)
850 struct net_device *dev;
853 dev = dev_get_by_index_rcu(net, ifindex);
859 EXPORT_SYMBOL(dev_get_by_index);
862 * netdev_get_name - get a netdevice name, knowing its ifindex.
863 * @net: network namespace
864 * @name: a pointer to the buffer where the name will be stored.
865 * @ifindex: the ifindex of the interface to get the name from.
867 * The use of raw_seqcount_begin() and cond_resched() before
868 * retrying is required as we want to give the writers a chance
869 * to complete when CONFIG_PREEMPT is not set.
871 int netdev_get_name(struct net *net, char *name, int ifindex)
873 struct net_device *dev;
877 seq = raw_seqcount_begin(&devnet_rename_seq);
879 dev = dev_get_by_index_rcu(net, ifindex);
885 strcpy(name, dev->name);
887 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
888 mutex_lock(&devnet_rename_mutex);
889 mutex_unlock(&devnet_rename_mutex);
897 * dev_getbyhwaddr_rcu - find a device by its hardware address
898 * @net: the applicable net namespace
899 * @type: media type of device
900 * @ha: hardware address
902 * Search for an interface by MAC address. Returns NULL if the device
903 * is not found or a pointer to the device.
904 * The caller must hold RCU or RTNL.
905 * The returned device has not had its ref count increased
906 * and the caller must therefore be careful about locking
910 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
913 struct net_device *dev;
915 for_each_netdev_rcu(net, dev)
916 if (dev->type == type &&
917 !memcmp(dev->dev_addr, ha, dev->addr_len))
922 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
924 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
926 struct net_device *dev;
929 for_each_netdev(net, dev)
930 if (dev->type == type)
935 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
937 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
939 struct net_device *dev, *ret = NULL;
942 for_each_netdev_rcu(net, dev)
943 if (dev->type == type) {
951 EXPORT_SYMBOL(dev_getfirstbyhwtype);
954 * __dev_get_by_flags - find any device with given flags
955 * @net: the applicable net namespace
956 * @if_flags: IFF_* values
957 * @mask: bitmask of bits in if_flags to check
959 * Search for any interface with the given flags. Returns NULL if a device
960 * is not found or a pointer to the device. Must be called inside
961 * rtnl_lock(), and result refcount is unchanged.
964 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
967 struct net_device *dev, *ret;
972 for_each_netdev(net, dev) {
973 if (((dev->flags ^ if_flags) & mask) == 0) {
980 EXPORT_SYMBOL(__dev_get_by_flags);
983 * dev_valid_name - check if name is okay for network device
986 * Network device names need to be valid file names to
987 * to allow sysfs to work. We also disallow any kind of
990 bool dev_valid_name(const char *name)
994 if (strlen(name) >= IFNAMSIZ)
996 if (!strcmp(name, ".") || !strcmp(name, ".."))
1000 if (*name == '/' || *name == ':' || isspace(*name))
1006 EXPORT_SYMBOL(dev_valid_name);
1009 * __dev_alloc_name - allocate a name for a device
1010 * @net: network namespace to allocate the device name in
1011 * @name: name format string
1012 * @buf: scratch buffer and result name string
1014 * Passed a format string - eg "lt%d" it will try and find a suitable
1015 * id. It scans list of devices to build up a free map, then chooses
1016 * the first empty slot. The caller must hold the dev_base or rtnl lock
1017 * while allocating the name and adding the device in order to avoid
1019 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1020 * Returns the number of the unit assigned or a negative errno code.
1023 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1027 const int max_netdevices = 8*PAGE_SIZE;
1028 unsigned long *inuse;
1029 struct net_device *d;
1031 p = strnchr(name, IFNAMSIZ-1, '%');
1034 * Verify the string as this thing may have come from
1035 * the user. There must be either one "%d" and no other "%"
1038 if (p[1] != 'd' || strchr(p + 2, '%'))
1041 /* Use one page as a bit array of possible slots */
1042 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1046 for_each_netdev(net, d) {
1047 if (!sscanf(d->name, name, &i))
1049 if (i < 0 || i >= max_netdevices)
1052 /* avoid cases where sscanf is not exact inverse of printf */
1053 snprintf(buf, IFNAMSIZ, name, i);
1054 if (!strncmp(buf, d->name, IFNAMSIZ))
1058 i = find_first_zero_bit(inuse, max_netdevices);
1059 free_page((unsigned long) inuse);
1063 snprintf(buf, IFNAMSIZ, name, i);
1064 if (!__dev_get_by_name(net, buf))
1067 /* It is possible to run out of possible slots
1068 * when the name is long and there isn't enough space left
1069 * for the digits, or if all bits are used.
1075 * dev_alloc_name - allocate a name for a device
1077 * @name: name format string
1079 * Passed a format string - eg "lt%d" it will try and find a suitable
1080 * id. It scans list of devices to build up a free map, then chooses
1081 * the first empty slot. The caller must hold the dev_base or rtnl lock
1082 * while allocating the name and adding the device in order to avoid
1084 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1085 * Returns the number of the unit assigned or a negative errno code.
1088 int dev_alloc_name(struct net_device *dev, const char *name)
1094 BUG_ON(!dev_net(dev));
1096 ret = __dev_alloc_name(net, name, buf);
1098 strlcpy(dev->name, buf, IFNAMSIZ);
1101 EXPORT_SYMBOL(dev_alloc_name);
1103 static int dev_alloc_name_ns(struct net *net,
1104 struct net_device *dev,
1110 ret = __dev_alloc_name(net, name, buf);
1112 strlcpy(dev->name, buf, IFNAMSIZ);
1116 static int dev_get_valid_name(struct net *net,
1117 struct net_device *dev,
1122 if (!dev_valid_name(name))
1125 if (strchr(name, '%'))
1126 return dev_alloc_name_ns(net, dev, name);
1127 else if (__dev_get_by_name(net, name))
1129 else if (dev->name != name)
1130 strlcpy(dev->name, name, IFNAMSIZ);
1136 * dev_change_name - change name of a device
1138 * @newname: name (or format string) must be at least IFNAMSIZ
1140 * Change name of a device, can pass format strings "eth%d".
1143 int dev_change_name(struct net_device *dev, const char *newname)
1145 unsigned char old_assign_type;
1146 char oldname[IFNAMSIZ];
1152 BUG_ON(!dev_net(dev));
1155 if (dev->flags & IFF_UP)
1158 mutex_lock(&devnet_rename_mutex);
1159 __raw_write_seqcount_begin(&devnet_rename_seq);
1161 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1164 memcpy(oldname, dev->name, IFNAMSIZ);
1166 err = dev_get_valid_name(net, dev, newname);
1170 if (oldname[0] && !strchr(oldname, '%'))
1171 netdev_info(dev, "renamed from %s\n", oldname);
1173 old_assign_type = dev->name_assign_type;
1174 dev->name_assign_type = NET_NAME_RENAMED;
1177 ret = device_rename(&dev->dev, dev->name);
1179 memcpy(dev->name, oldname, IFNAMSIZ);
1180 dev->name_assign_type = old_assign_type;
1185 __raw_write_seqcount_end(&devnet_rename_seq);
1186 mutex_unlock(&devnet_rename_mutex);
1188 netdev_adjacent_rename_links(dev, oldname);
1190 write_lock_bh(&dev_base_lock);
1191 hlist_del_rcu(&dev->name_hlist);
1192 write_unlock_bh(&dev_base_lock);
1196 write_lock_bh(&dev_base_lock);
1197 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1198 write_unlock_bh(&dev_base_lock);
1200 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1201 ret = notifier_to_errno(ret);
1204 /* err >= 0 after dev_alloc_name() or stores the first errno */
1207 mutex_lock(&devnet_rename_mutex);
1208 __raw_write_seqcount_begin(&devnet_rename_seq);
1209 memcpy(dev->name, oldname, IFNAMSIZ);
1210 memcpy(oldname, newname, IFNAMSIZ);
1211 dev->name_assign_type = old_assign_type;
1212 old_assign_type = NET_NAME_RENAMED;
1215 pr_err("%s: name change rollback failed: %d\n",
1223 __raw_write_seqcount_end(&devnet_rename_seq);
1224 mutex_unlock(&devnet_rename_mutex);
1229 * dev_set_alias - change ifalias of a device
1231 * @alias: name up to IFALIASZ
1232 * @len: limit of bytes to copy from info
1234 * Set ifalias for a device,
1236 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1242 if (len >= IFALIASZ)
1246 kfree(dev->ifalias);
1247 dev->ifalias = NULL;
1251 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1254 dev->ifalias = new_ifalias;
1256 strlcpy(dev->ifalias, alias, len+1);
1262 * netdev_features_change - device changes features
1263 * @dev: device to cause notification
1265 * Called to indicate a device has changed features.
1267 void netdev_features_change(struct net_device *dev)
1269 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1271 EXPORT_SYMBOL(netdev_features_change);
1274 * netdev_state_change - device changes state
1275 * @dev: device to cause notification
1277 * Called to indicate a device has changed state. This function calls
1278 * the notifier chains for netdev_chain and sends a NEWLINK message
1279 * to the routing socket.
1281 void netdev_state_change(struct net_device *dev)
1283 if (dev->flags & IFF_UP) {
1284 struct netdev_notifier_change_info change_info;
1286 change_info.flags_changed = 0;
1287 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1289 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1292 EXPORT_SYMBOL(netdev_state_change);
1295 * netdev_notify_peers - notify network peers about existence of @dev
1296 * @dev: network device
1298 * Generate traffic such that interested network peers are aware of
1299 * @dev, such as by generating a gratuitous ARP. This may be used when
1300 * a device wants to inform the rest of the network about some sort of
1301 * reconfiguration such as a failover event or virtual machine
1304 void netdev_notify_peers(struct net_device *dev)
1307 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1310 EXPORT_SYMBOL(netdev_notify_peers);
1312 static int __dev_open(struct net_device *dev)
1314 const struct net_device_ops *ops = dev->netdev_ops;
1319 if (!netif_device_present(dev))
1322 /* Block netpoll from trying to do any rx path servicing.
1323 * If we don't do this there is a chance ndo_poll_controller
1324 * or ndo_poll may be running while we open the device
1326 netpoll_poll_disable(dev);
1328 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1329 ret = notifier_to_errno(ret);
1333 set_bit(__LINK_STATE_START, &dev->state);
1335 if (ops->ndo_validate_addr)
1336 ret = ops->ndo_validate_addr(dev);
1338 if (!ret && ops->ndo_open)
1339 ret = ops->ndo_open(dev);
1341 netpoll_poll_enable(dev);
1344 clear_bit(__LINK_STATE_START, &dev->state);
1346 dev->flags |= IFF_UP;
1347 dev_set_rx_mode(dev);
1349 add_device_randomness(dev->dev_addr, dev->addr_len);
1356 * dev_open - prepare an interface for use.
1357 * @dev: device to open
1359 * Takes a device from down to up state. The device's private open
1360 * function is invoked and then the multicast lists are loaded. Finally
1361 * the device is moved into the up state and a %NETDEV_UP message is
1362 * sent to the netdev notifier chain.
1364 * Calling this function on an active interface is a nop. On a failure
1365 * a negative errno code is returned.
1367 int dev_open(struct net_device *dev)
1371 if (dev->flags & IFF_UP)
1374 ret = __dev_open(dev);
1378 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1379 call_netdevice_notifiers(NETDEV_UP, dev);
1383 EXPORT_SYMBOL(dev_open);
1385 static int __dev_close_many(struct list_head *head)
1387 struct net_device *dev;
1392 list_for_each_entry(dev, head, close_list) {
1393 /* Temporarily disable netpoll until the interface is down */
1394 netpoll_poll_disable(dev);
1396 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1398 clear_bit(__LINK_STATE_START, &dev->state);
1400 /* Synchronize to scheduled poll. We cannot touch poll list, it
1401 * can be even on different cpu. So just clear netif_running().
1403 * dev->stop() will invoke napi_disable() on all of it's
1404 * napi_struct instances on this device.
1406 smp_mb__after_atomic(); /* Commit netif_running(). */
1409 dev_deactivate_many(head);
1411 list_for_each_entry(dev, head, close_list) {
1412 const struct net_device_ops *ops = dev->netdev_ops;
1415 * Call the device specific close. This cannot fail.
1416 * Only if device is UP
1418 * We allow it to be called even after a DETACH hot-plug
1424 dev->flags &= ~IFF_UP;
1425 netpoll_poll_enable(dev);
1431 static int __dev_close(struct net_device *dev)
1436 list_add(&dev->close_list, &single);
1437 retval = __dev_close_many(&single);
1443 int dev_close_many(struct list_head *head, bool unlink)
1445 struct net_device *dev, *tmp;
1447 /* Remove the devices that don't need to be closed */
1448 list_for_each_entry_safe(dev, tmp, head, close_list)
1449 if (!(dev->flags & IFF_UP))
1450 list_del_init(&dev->close_list);
1452 __dev_close_many(head);
1454 list_for_each_entry_safe(dev, tmp, head, close_list) {
1455 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1456 call_netdevice_notifiers(NETDEV_DOWN, dev);
1458 list_del_init(&dev->close_list);
1463 EXPORT_SYMBOL(dev_close_many);
1466 * dev_close - shutdown an interface.
1467 * @dev: device to shutdown
1469 * This function moves an active device into down state. A
1470 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1471 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1474 int dev_close(struct net_device *dev)
1476 if (dev->flags & IFF_UP) {
1479 list_add(&dev->close_list, &single);
1480 dev_close_many(&single, true);
1485 EXPORT_SYMBOL(dev_close);
1489 * dev_disable_lro - disable Large Receive Offload on a device
1492 * Disable Large Receive Offload (LRO) on a net device. Must be
1493 * called under RTNL. This is needed if received packets may be
1494 * forwarded to another interface.
1496 void dev_disable_lro(struct net_device *dev)
1498 struct net_device *lower_dev;
1499 struct list_head *iter;
1501 dev->wanted_features &= ~NETIF_F_LRO;
1502 netdev_update_features(dev);
1504 if (unlikely(dev->features & NETIF_F_LRO))
1505 netdev_WARN(dev, "failed to disable LRO!\n");
1507 netdev_for_each_lower_dev(dev, lower_dev, iter)
1508 dev_disable_lro(lower_dev);
1510 EXPORT_SYMBOL(dev_disable_lro);
1512 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1513 struct net_device *dev)
1515 struct netdev_notifier_info info;
1517 netdev_notifier_info_init(&info, dev);
1518 return nb->notifier_call(nb, val, &info);
1521 static int dev_boot_phase = 1;
1524 * register_netdevice_notifier - register a network notifier block
1527 * Register a notifier to be called when network device events occur.
1528 * The notifier passed is linked into the kernel structures and must
1529 * not be reused until it has been unregistered. A negative errno code
1530 * is returned on a failure.
1532 * When registered all registration and up events are replayed
1533 * to the new notifier to allow device to have a race free
1534 * view of the network device list.
1537 int register_netdevice_notifier(struct notifier_block *nb)
1539 struct net_device *dev;
1540 struct net_device *last;
1545 err = raw_notifier_chain_register(&netdev_chain, nb);
1551 for_each_netdev(net, dev) {
1552 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1553 err = notifier_to_errno(err);
1557 if (!(dev->flags & IFF_UP))
1560 call_netdevice_notifier(nb, NETDEV_UP, dev);
1571 for_each_netdev(net, dev) {
1575 if (dev->flags & IFF_UP) {
1576 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1578 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1580 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1585 raw_notifier_chain_unregister(&netdev_chain, nb);
1588 EXPORT_SYMBOL(register_netdevice_notifier);
1591 * unregister_netdevice_notifier - unregister a network notifier block
1594 * Unregister a notifier previously registered by
1595 * register_netdevice_notifier(). The notifier is unlinked into the
1596 * kernel structures and may then be reused. A negative errno code
1597 * is returned on a failure.
1599 * After unregistering unregister and down device events are synthesized
1600 * for all devices on the device list to the removed notifier to remove
1601 * the need for special case cleanup code.
1604 int unregister_netdevice_notifier(struct notifier_block *nb)
1606 struct net_device *dev;
1611 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1616 for_each_netdev(net, dev) {
1617 if (dev->flags & IFF_UP) {
1618 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1620 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1622 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1629 EXPORT_SYMBOL(unregister_netdevice_notifier);
1632 * call_netdevice_notifiers_info - call all network notifier blocks
1633 * @val: value passed unmodified to notifier function
1634 * @dev: net_device pointer passed unmodified to notifier function
1635 * @info: notifier information data
1637 * Call all network notifier blocks. Parameters and return value
1638 * are as for raw_notifier_call_chain().
1641 static int call_netdevice_notifiers_info(unsigned long val,
1642 struct net_device *dev,
1643 struct netdev_notifier_info *info)
1646 netdev_notifier_info_init(info, dev);
1647 return raw_notifier_call_chain(&netdev_chain, val, info);
1651 * call_netdevice_notifiers - call all network notifier blocks
1652 * @val: value passed unmodified to notifier function
1653 * @dev: net_device pointer passed unmodified to notifier function
1655 * Call all network notifier blocks. Parameters and return value
1656 * are as for raw_notifier_call_chain().
1659 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1661 struct netdev_notifier_info info;
1663 return call_netdevice_notifiers_info(val, dev, &info);
1665 EXPORT_SYMBOL(call_netdevice_notifiers);
1667 #ifdef CONFIG_NET_INGRESS
1668 static struct static_key ingress_needed __read_mostly;
1670 void net_inc_ingress_queue(void)
1672 static_key_slow_inc(&ingress_needed);
1674 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1676 void net_dec_ingress_queue(void)
1678 static_key_slow_dec(&ingress_needed);
1680 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1683 static struct static_key netstamp_needed __read_mostly;
1684 #ifdef HAVE_JUMP_LABEL
1685 /* We are not allowed to call static_key_slow_dec() from irq context
1686 * If net_disable_timestamp() is called from irq context, defer the
1687 * static_key_slow_dec() calls.
1689 static atomic_t netstamp_needed_deferred;
1692 void net_enable_timestamp(void)
1694 #ifdef HAVE_JUMP_LABEL
1695 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1699 static_key_slow_dec(&netstamp_needed);
1703 static_key_slow_inc(&netstamp_needed);
1705 EXPORT_SYMBOL(net_enable_timestamp);
1707 void net_disable_timestamp(void)
1709 #ifdef HAVE_JUMP_LABEL
1710 if (in_interrupt()) {
1711 atomic_inc(&netstamp_needed_deferred);
1715 static_key_slow_dec(&netstamp_needed);
1717 EXPORT_SYMBOL(net_disable_timestamp);
1719 static inline void net_timestamp_set(struct sk_buff *skb)
1721 skb->tstamp.tv64 = 0;
1722 if (static_key_false(&netstamp_needed))
1723 __net_timestamp(skb);
1726 #define net_timestamp_check(COND, SKB) \
1727 if (static_key_false(&netstamp_needed)) { \
1728 if ((COND) && !(SKB)->tstamp.tv64) \
1729 __net_timestamp(SKB); \
1732 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1736 if (!(dev->flags & IFF_UP))
1739 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1740 if (skb->len <= len)
1743 /* if TSO is enabled, we don't care about the length as the packet
1744 * could be forwarded without being segmented before
1746 if (skb_is_gso(skb))
1751 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1753 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1755 if (skb_orphan_frags(skb, GFP_ATOMIC) ||
1756 unlikely(!is_skb_forwardable(dev, skb))) {
1757 atomic_long_inc(&dev->rx_dropped);
1762 skb_scrub_packet(skb, true);
1764 skb->protocol = eth_type_trans(skb, dev);
1765 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1769 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1772 * dev_forward_skb - loopback an skb to another netif
1774 * @dev: destination network device
1775 * @skb: buffer to forward
1778 * NET_RX_SUCCESS (no congestion)
1779 * NET_RX_DROP (packet was dropped, but freed)
1781 * dev_forward_skb can be used for injecting an skb from the
1782 * start_xmit function of one device into the receive queue
1783 * of another device.
1785 * The receiving device may be in another namespace, so
1786 * we have to clear all information in the skb that could
1787 * impact namespace isolation.
1789 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1791 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1793 EXPORT_SYMBOL_GPL(dev_forward_skb);
1795 static inline int deliver_skb(struct sk_buff *skb,
1796 struct packet_type *pt_prev,
1797 struct net_device *orig_dev)
1799 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1801 atomic_inc(&skb->users);
1802 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1805 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1806 struct packet_type **pt,
1807 struct net_device *orig_dev,
1809 struct list_head *ptype_list)
1811 struct packet_type *ptype, *pt_prev = *pt;
1813 list_for_each_entry_rcu(ptype, ptype_list, list) {
1814 if (ptype->type != type)
1817 deliver_skb(skb, pt_prev, orig_dev);
1823 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1825 if (!ptype->af_packet_priv || !skb->sk)
1828 if (ptype->id_match)
1829 return ptype->id_match(ptype, skb->sk);
1830 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1837 * Support routine. Sends outgoing frames to any network
1838 * taps currently in use.
1841 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1843 struct packet_type *ptype;
1844 struct sk_buff *skb2 = NULL;
1845 struct packet_type *pt_prev = NULL;
1846 struct list_head *ptype_list = &ptype_all;
1850 list_for_each_entry_rcu(ptype, ptype_list, list) {
1851 /* Never send packets back to the socket
1852 * they originated from - MvS (miquels@drinkel.ow.org)
1854 if (skb_loop_sk(ptype, skb))
1858 deliver_skb(skb2, pt_prev, skb->dev);
1863 /* need to clone skb, done only once */
1864 skb2 = skb_clone(skb, GFP_ATOMIC);
1868 net_timestamp_set(skb2);
1870 /* skb->nh should be correctly
1871 * set by sender, so that the second statement is
1872 * just protection against buggy protocols.
1874 skb_reset_mac_header(skb2);
1876 if (skb_network_header(skb2) < skb2->data ||
1877 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1878 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1879 ntohs(skb2->protocol),
1881 skb_reset_network_header(skb2);
1884 skb2->transport_header = skb2->network_header;
1885 skb2->pkt_type = PACKET_OUTGOING;
1889 if (ptype_list == &ptype_all) {
1890 ptype_list = &dev->ptype_all;
1895 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1900 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1901 * @dev: Network device
1902 * @txq: number of queues available
1904 * If real_num_tx_queues is changed the tc mappings may no longer be
1905 * valid. To resolve this verify the tc mapping remains valid and if
1906 * not NULL the mapping. With no priorities mapping to this
1907 * offset/count pair it will no longer be used. In the worst case TC0
1908 * is invalid nothing can be done so disable priority mappings. If is
1909 * expected that drivers will fix this mapping if they can before
1910 * calling netif_set_real_num_tx_queues.
1912 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1915 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1917 /* If TC0 is invalidated disable TC mapping */
1918 if (tc->offset + tc->count > txq) {
1919 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1924 /* Invalidated prio to tc mappings set to TC0 */
1925 for (i = 1; i < TC_BITMASK + 1; i++) {
1926 int q = netdev_get_prio_tc_map(dev, i);
1928 tc = &dev->tc_to_txq[q];
1929 if (tc->offset + tc->count > txq) {
1930 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1932 netdev_set_prio_tc_map(dev, i, 0);
1938 static DEFINE_MUTEX(xps_map_mutex);
1939 #define xmap_dereference(P) \
1940 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1942 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1945 struct xps_map *map = NULL;
1949 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1951 for (pos = 0; map && pos < map->len; pos++) {
1952 if (map->queues[pos] == index) {
1954 map->queues[pos] = map->queues[--map->len];
1956 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1957 kfree_rcu(map, rcu);
1967 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1969 struct xps_dev_maps *dev_maps;
1971 bool active = false;
1973 mutex_lock(&xps_map_mutex);
1974 dev_maps = xmap_dereference(dev->xps_maps);
1979 for_each_possible_cpu(cpu) {
1980 for (i = index; i < dev->num_tx_queues; i++) {
1981 if (!remove_xps_queue(dev_maps, cpu, i))
1984 if (i == dev->num_tx_queues)
1989 RCU_INIT_POINTER(dev->xps_maps, NULL);
1990 kfree_rcu(dev_maps, rcu);
1993 for (i = index; i < dev->num_tx_queues; i++)
1994 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1998 mutex_unlock(&xps_map_mutex);
2001 static struct xps_map *expand_xps_map(struct xps_map *map,
2004 struct xps_map *new_map;
2005 int alloc_len = XPS_MIN_MAP_ALLOC;
2008 for (pos = 0; map && pos < map->len; pos++) {
2009 if (map->queues[pos] != index)
2014 /* Need to add queue to this CPU's existing map */
2016 if (pos < map->alloc_len)
2019 alloc_len = map->alloc_len * 2;
2022 /* Need to allocate new map to store queue on this CPU's map */
2023 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2028 for (i = 0; i < pos; i++)
2029 new_map->queues[i] = map->queues[i];
2030 new_map->alloc_len = alloc_len;
2036 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2039 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2040 struct xps_map *map, *new_map;
2041 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2042 int cpu, numa_node_id = -2;
2043 bool active = false;
2045 mutex_lock(&xps_map_mutex);
2047 dev_maps = xmap_dereference(dev->xps_maps);
2049 /* allocate memory for queue storage */
2050 for_each_online_cpu(cpu) {
2051 if (!cpumask_test_cpu(cpu, mask))
2055 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2056 if (!new_dev_maps) {
2057 mutex_unlock(&xps_map_mutex);
2061 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2064 map = expand_xps_map(map, cpu, index);
2068 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2072 goto out_no_new_maps;
2074 for_each_possible_cpu(cpu) {
2075 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2076 /* add queue to CPU maps */
2079 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2080 while ((pos < map->len) && (map->queues[pos] != index))
2083 if (pos == map->len)
2084 map->queues[map->len++] = index;
2086 if (numa_node_id == -2)
2087 numa_node_id = cpu_to_node(cpu);
2088 else if (numa_node_id != cpu_to_node(cpu))
2091 } else if (dev_maps) {
2092 /* fill in the new device map from the old device map */
2093 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2094 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2099 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2101 /* Cleanup old maps */
2103 for_each_possible_cpu(cpu) {
2104 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2105 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2106 if (map && map != new_map)
2107 kfree_rcu(map, rcu);
2110 kfree_rcu(dev_maps, rcu);
2113 dev_maps = new_dev_maps;
2117 /* update Tx queue numa node */
2118 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2119 (numa_node_id >= 0) ? numa_node_id :
2125 /* removes queue from unused CPUs */
2126 for_each_possible_cpu(cpu) {
2127 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2130 if (remove_xps_queue(dev_maps, cpu, index))
2134 /* free map if not active */
2136 RCU_INIT_POINTER(dev->xps_maps, NULL);
2137 kfree_rcu(dev_maps, rcu);
2141 mutex_unlock(&xps_map_mutex);
2145 /* remove any maps that we added */
2146 for_each_possible_cpu(cpu) {
2147 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2148 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2150 if (new_map && new_map != map)
2154 mutex_unlock(&xps_map_mutex);
2156 kfree(new_dev_maps);
2159 EXPORT_SYMBOL(netif_set_xps_queue);
2163 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2164 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2166 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2170 if (txq < 1 || txq > dev->num_tx_queues)
2173 if (dev->reg_state == NETREG_REGISTERED ||
2174 dev->reg_state == NETREG_UNREGISTERING) {
2177 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2183 netif_setup_tc(dev, txq);
2185 if (txq < dev->real_num_tx_queues) {
2186 qdisc_reset_all_tx_gt(dev, txq);
2188 netif_reset_xps_queues_gt(dev, txq);
2193 dev->real_num_tx_queues = txq;
2196 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2200 * netif_set_real_num_rx_queues - set actual number of RX queues used
2201 * @dev: Network device
2202 * @rxq: Actual number of RX queues
2204 * This must be called either with the rtnl_lock held or before
2205 * registration of the net device. Returns 0 on success, or a
2206 * negative error code. If called before registration, it always
2209 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2213 if (rxq < 1 || rxq > dev->num_rx_queues)
2216 if (dev->reg_state == NETREG_REGISTERED) {
2219 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2225 dev->real_num_rx_queues = rxq;
2228 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2232 * netif_get_num_default_rss_queues - default number of RSS queues
2234 * This routine should set an upper limit on the number of RSS queues
2235 * used by default by multiqueue devices.
2237 int netif_get_num_default_rss_queues(void)
2239 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2241 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2243 static inline void __netif_reschedule(struct Qdisc *q)
2245 struct softnet_data *sd;
2246 unsigned long flags;
2248 local_irq_save(flags);
2249 sd = this_cpu_ptr(&softnet_data);
2250 q->next_sched = NULL;
2251 *sd->output_queue_tailp = q;
2252 sd->output_queue_tailp = &q->next_sched;
2253 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2254 local_irq_restore(flags);
2255 preempt_check_resched_rt();
2258 void __netif_schedule(struct Qdisc *q)
2260 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2261 __netif_reschedule(q);
2263 EXPORT_SYMBOL(__netif_schedule);
2265 struct dev_kfree_skb_cb {
2266 enum skb_free_reason reason;
2269 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2271 return (struct dev_kfree_skb_cb *)skb->cb;
2274 void netif_schedule_queue(struct netdev_queue *txq)
2277 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2278 struct Qdisc *q = rcu_dereference(txq->qdisc);
2280 __netif_schedule(q);
2284 EXPORT_SYMBOL(netif_schedule_queue);
2287 * netif_wake_subqueue - allow sending packets on subqueue
2288 * @dev: network device
2289 * @queue_index: sub queue index
2291 * Resume individual transmit queue of a device with multiple transmit queues.
2293 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2295 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2297 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2301 q = rcu_dereference(txq->qdisc);
2302 __netif_schedule(q);
2306 EXPORT_SYMBOL(netif_wake_subqueue);
2308 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2310 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2314 q = rcu_dereference(dev_queue->qdisc);
2315 __netif_schedule(q);
2319 EXPORT_SYMBOL(netif_tx_wake_queue);
2321 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2323 unsigned long flags;
2325 if (likely(atomic_read(&skb->users) == 1)) {
2327 atomic_set(&skb->users, 0);
2328 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2331 get_kfree_skb_cb(skb)->reason = reason;
2332 local_irq_save(flags);
2333 skb->next = __this_cpu_read(softnet_data.completion_queue);
2334 __this_cpu_write(softnet_data.completion_queue, skb);
2335 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2336 local_irq_restore(flags);
2337 preempt_check_resched_rt();
2339 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2341 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2343 if (in_irq() || irqs_disabled())
2344 __dev_kfree_skb_irq(skb, reason);
2348 EXPORT_SYMBOL(__dev_kfree_skb_any);
2352 * netif_device_detach - mark device as removed
2353 * @dev: network device
2355 * Mark device as removed from system and therefore no longer available.
2357 void netif_device_detach(struct net_device *dev)
2359 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2360 netif_running(dev)) {
2361 netif_tx_stop_all_queues(dev);
2364 EXPORT_SYMBOL(netif_device_detach);
2367 * netif_device_attach - mark device as attached
2368 * @dev: network device
2370 * Mark device as attached from system and restart if needed.
2372 void netif_device_attach(struct net_device *dev)
2374 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2375 netif_running(dev)) {
2376 netif_tx_wake_all_queues(dev);
2377 __netdev_watchdog_up(dev);
2380 EXPORT_SYMBOL(netif_device_attach);
2383 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2384 * to be used as a distribution range.
2386 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2387 unsigned int num_tx_queues)
2391 u16 qcount = num_tx_queues;
2393 if (skb_rx_queue_recorded(skb)) {
2394 hash = skb_get_rx_queue(skb);
2395 while (unlikely(hash >= num_tx_queues))
2396 hash -= num_tx_queues;
2401 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2402 qoffset = dev->tc_to_txq[tc].offset;
2403 qcount = dev->tc_to_txq[tc].count;
2406 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2408 EXPORT_SYMBOL(__skb_tx_hash);
2410 static void skb_warn_bad_offload(const struct sk_buff *skb)
2412 static const netdev_features_t null_features = 0;
2413 struct net_device *dev = skb->dev;
2414 const char *name = "";
2416 if (!net_ratelimit())
2420 if (dev->dev.parent)
2421 name = dev_driver_string(dev->dev.parent);
2423 name = netdev_name(dev);
2425 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2426 "gso_type=%d ip_summed=%d\n",
2427 name, dev ? &dev->features : &null_features,
2428 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2429 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2430 skb_shinfo(skb)->gso_type, skb->ip_summed);
2434 * Invalidate hardware checksum when packet is to be mangled, and
2435 * complete checksum manually on outgoing path.
2437 int skb_checksum_help(struct sk_buff *skb)
2440 int ret = 0, offset;
2442 if (skb->ip_summed == CHECKSUM_COMPLETE)
2443 goto out_set_summed;
2445 if (unlikely(skb_shinfo(skb)->gso_size)) {
2446 skb_warn_bad_offload(skb);
2450 /* Before computing a checksum, we should make sure no frag could
2451 * be modified by an external entity : checksum could be wrong.
2453 if (skb_has_shared_frag(skb)) {
2454 ret = __skb_linearize(skb);
2459 offset = skb_checksum_start_offset(skb);
2460 BUG_ON(offset >= skb_headlen(skb));
2461 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2463 offset += skb->csum_offset;
2464 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2466 if (skb_cloned(skb) &&
2467 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2468 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2473 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2475 skb->ip_summed = CHECKSUM_NONE;
2479 EXPORT_SYMBOL(skb_checksum_help);
2481 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2483 __be16 type = skb->protocol;
2485 /* Tunnel gso handlers can set protocol to ethernet. */
2486 if (type == htons(ETH_P_TEB)) {
2489 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2492 eth = (struct ethhdr *)skb_mac_header(skb);
2493 type = eth->h_proto;
2496 return __vlan_get_protocol(skb, type, depth);
2500 * skb_mac_gso_segment - mac layer segmentation handler.
2501 * @skb: buffer to segment
2502 * @features: features for the output path (see dev->features)
2504 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2505 netdev_features_t features)
2507 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2508 struct packet_offload *ptype;
2509 int vlan_depth = skb->mac_len;
2510 __be16 type = skb_network_protocol(skb, &vlan_depth);
2512 if (unlikely(!type))
2513 return ERR_PTR(-EINVAL);
2515 __skb_pull(skb, vlan_depth);
2518 list_for_each_entry_rcu(ptype, &offload_base, list) {
2519 if (ptype->type == type && ptype->callbacks.gso_segment) {
2520 segs = ptype->callbacks.gso_segment(skb, features);
2526 __skb_push(skb, skb->data - skb_mac_header(skb));
2530 EXPORT_SYMBOL(skb_mac_gso_segment);
2533 /* openvswitch calls this on rx path, so we need a different check.
2535 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2538 return skb->ip_summed != CHECKSUM_PARTIAL;
2540 return skb->ip_summed == CHECKSUM_NONE;
2544 * __skb_gso_segment - Perform segmentation on skb.
2545 * @skb: buffer to segment
2546 * @features: features for the output path (see dev->features)
2547 * @tx_path: whether it is called in TX path
2549 * This function segments the given skb and returns a list of segments.
2551 * It may return NULL if the skb requires no segmentation. This is
2552 * only possible when GSO is used for verifying header integrity.
2554 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2556 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2557 netdev_features_t features, bool tx_path)
2559 if (unlikely(skb_needs_check(skb, tx_path))) {
2562 skb_warn_bad_offload(skb);
2564 err = skb_cow_head(skb, 0);
2566 return ERR_PTR(err);
2569 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2570 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2572 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2573 SKB_GSO_CB(skb)->encap_level = 0;
2575 skb_reset_mac_header(skb);
2576 skb_reset_mac_len(skb);
2578 return skb_mac_gso_segment(skb, features);
2580 EXPORT_SYMBOL(__skb_gso_segment);
2582 /* Take action when hardware reception checksum errors are detected. */
2584 void netdev_rx_csum_fault(struct net_device *dev)
2586 if (net_ratelimit()) {
2587 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2591 EXPORT_SYMBOL(netdev_rx_csum_fault);
2594 /* Actually, we should eliminate this check as soon as we know, that:
2595 * 1. IOMMU is present and allows to map all the memory.
2596 * 2. No high memory really exists on this machine.
2599 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2601 #ifdef CONFIG_HIGHMEM
2603 if (!(dev->features & NETIF_F_HIGHDMA)) {
2604 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2605 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2606 if (PageHighMem(skb_frag_page(frag)))
2611 if (PCI_DMA_BUS_IS_PHYS) {
2612 struct device *pdev = dev->dev.parent;
2616 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2617 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2618 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2619 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2627 /* If MPLS offload request, verify we are testing hardware MPLS features
2628 * instead of standard features for the netdev.
2630 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2631 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2632 netdev_features_t features,
2635 if (eth_p_mpls(type))
2636 features &= skb->dev->mpls_features;
2641 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2642 netdev_features_t features,
2649 static netdev_features_t harmonize_features(struct sk_buff *skb,
2650 netdev_features_t features)
2655 type = skb_network_protocol(skb, &tmp);
2656 features = net_mpls_features(skb, features, type);
2658 if (skb->ip_summed != CHECKSUM_NONE &&
2659 !can_checksum_protocol(features, type)) {
2660 features &= ~NETIF_F_ALL_CSUM;
2661 } else if (illegal_highdma(skb->dev, skb)) {
2662 features &= ~NETIF_F_SG;
2668 netdev_features_t passthru_features_check(struct sk_buff *skb,
2669 struct net_device *dev,
2670 netdev_features_t features)
2674 EXPORT_SYMBOL(passthru_features_check);
2676 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2677 struct net_device *dev,
2678 netdev_features_t features)
2680 return vlan_features_check(skb, features);
2683 netdev_features_t netif_skb_features(struct sk_buff *skb)
2685 struct net_device *dev = skb->dev;
2686 netdev_features_t features = dev->features;
2687 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2689 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2690 features &= ~NETIF_F_GSO_MASK;
2692 /* If encapsulation offload request, verify we are testing
2693 * hardware encapsulation features instead of standard
2694 * features for the netdev
2696 if (skb->encapsulation)
2697 features &= dev->hw_enc_features;
2699 if (skb_vlan_tagged(skb))
2700 features = netdev_intersect_features(features,
2701 dev->vlan_features |
2702 NETIF_F_HW_VLAN_CTAG_TX |
2703 NETIF_F_HW_VLAN_STAG_TX);
2705 if (dev->netdev_ops->ndo_features_check)
2706 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2709 features &= dflt_features_check(skb, dev, features);
2711 return harmonize_features(skb, features);
2713 EXPORT_SYMBOL(netif_skb_features);
2715 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2716 struct netdev_queue *txq, bool more)
2721 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2722 dev_queue_xmit_nit(skb, dev);
2725 trace_net_dev_start_xmit(skb, dev);
2726 rc = netdev_start_xmit(skb, dev, txq, more);
2727 trace_net_dev_xmit(skb, rc, dev, len);
2732 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2733 struct netdev_queue *txq, int *ret)
2735 struct sk_buff *skb = first;
2736 int rc = NETDEV_TX_OK;
2739 struct sk_buff *next = skb->next;
2742 rc = xmit_one(skb, dev, txq, next != NULL);
2743 if (unlikely(!dev_xmit_complete(rc))) {
2749 if (netif_xmit_stopped(txq) && skb) {
2750 rc = NETDEV_TX_BUSY;
2760 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2761 netdev_features_t features)
2763 if (skb_vlan_tag_present(skb) &&
2764 !vlan_hw_offload_capable(features, skb->vlan_proto))
2765 skb = __vlan_hwaccel_push_inside(skb);
2769 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2771 netdev_features_t features;
2776 features = netif_skb_features(skb);
2777 skb = validate_xmit_vlan(skb, features);
2781 if (netif_needs_gso(skb, features)) {
2782 struct sk_buff *segs;
2784 segs = skb_gso_segment(skb, features);
2792 if (skb_needs_linearize(skb, features) &&
2793 __skb_linearize(skb))
2796 /* If packet is not checksummed and device does not
2797 * support checksumming for this protocol, complete
2798 * checksumming here.
2800 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2801 if (skb->encapsulation)
2802 skb_set_inner_transport_header(skb,
2803 skb_checksum_start_offset(skb));
2805 skb_set_transport_header(skb,
2806 skb_checksum_start_offset(skb));
2807 if (!(features & NETIF_F_ALL_CSUM) &&
2808 skb_checksum_help(skb))
2821 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2823 struct sk_buff *next, *head = NULL, *tail;
2825 for (; skb != NULL; skb = next) {
2829 /* in case skb wont be segmented, point to itself */
2832 skb = validate_xmit_skb(skb, dev);
2840 /* If skb was segmented, skb->prev points to
2841 * the last segment. If not, it still contains skb.
2848 static void qdisc_pkt_len_init(struct sk_buff *skb)
2850 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2852 qdisc_skb_cb(skb)->pkt_len = skb->len;
2854 /* To get more precise estimation of bytes sent on wire,
2855 * we add to pkt_len the headers size of all segments
2857 if (shinfo->gso_size) {
2858 unsigned int hdr_len;
2859 u16 gso_segs = shinfo->gso_segs;
2861 /* mac layer + network layer */
2862 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2864 /* + transport layer */
2865 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2866 hdr_len += tcp_hdrlen(skb);
2868 hdr_len += sizeof(struct udphdr);
2870 if (shinfo->gso_type & SKB_GSO_DODGY)
2871 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2874 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2878 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2879 struct net_device *dev,
2880 struct netdev_queue *txq)
2882 spinlock_t *root_lock = qdisc_lock(q);
2886 qdisc_pkt_len_init(skb);
2887 qdisc_calculate_pkt_len(skb, q);
2889 * Heuristic to force contended enqueues to serialize on a
2890 * separate lock before trying to get qdisc main lock.
2891 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2892 * often and dequeue packets faster.
2894 #ifdef CONFIG_PREEMPT_RT_FULL
2897 contended = qdisc_is_running(q);
2899 if (unlikely(contended))
2900 spin_lock(&q->busylock);
2902 spin_lock(root_lock);
2903 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2906 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2907 qdisc_run_begin(q)) {
2909 * This is a work-conserving queue; there are no old skbs
2910 * waiting to be sent out; and the qdisc is not running -
2911 * xmit the skb directly.
2914 qdisc_bstats_update(q, skb);
2916 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2917 if (unlikely(contended)) {
2918 spin_unlock(&q->busylock);
2925 rc = NET_XMIT_SUCCESS;
2927 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2928 if (qdisc_run_begin(q)) {
2929 if (unlikely(contended)) {
2930 spin_unlock(&q->busylock);
2936 spin_unlock(root_lock);
2937 if (unlikely(contended))
2938 spin_unlock(&q->busylock);
2942 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2943 static void skb_update_prio(struct sk_buff *skb)
2945 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2947 if (!skb->priority && skb->sk && map) {
2948 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2950 if (prioidx < map->priomap_len)
2951 skb->priority = map->priomap[prioidx];
2955 #define skb_update_prio(skb)
2958 #ifdef CONFIG_PREEMPT_RT_FULL
2960 static inline int xmit_rec_read(void)
2962 return current->xmit_recursion;
2965 static inline void xmit_rec_inc(void)
2967 current->xmit_recursion++;
2970 static inline void xmit_rec_dec(void)
2972 current->xmit_recursion--;
2977 DEFINE_PER_CPU(int, xmit_recursion);
2978 EXPORT_SYMBOL(xmit_recursion);
2980 static inline int xmit_rec_read(void)
2982 return __this_cpu_read(xmit_recursion);
2985 static inline void xmit_rec_inc(void)
2987 __this_cpu_inc(xmit_recursion);
2990 static inline void xmit_rec_dec(void)
2992 __this_cpu_dec(xmit_recursion);
2996 #define RECURSION_LIMIT 10
2999 * dev_loopback_xmit - loop back @skb
3000 * @net: network namespace this loopback is happening in
3001 * @sk: sk needed to be a netfilter okfn
3002 * @skb: buffer to transmit
3004 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3006 skb_reset_mac_header(skb);
3007 __skb_pull(skb, skb_network_offset(skb));
3008 skb->pkt_type = PACKET_LOOPBACK;
3009 skb->ip_summed = CHECKSUM_UNNECESSARY;
3010 WARN_ON(!skb_dst(skb));
3015 EXPORT_SYMBOL(dev_loopback_xmit);
3017 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3020 struct xps_dev_maps *dev_maps;
3021 struct xps_map *map;
3022 int queue_index = -1;
3025 dev_maps = rcu_dereference(dev->xps_maps);
3027 map = rcu_dereference(
3028 dev_maps->cpu_map[skb->sender_cpu - 1]);
3031 queue_index = map->queues[0];
3033 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3035 if (unlikely(queue_index >= dev->real_num_tx_queues))
3047 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3049 struct sock *sk = skb->sk;
3050 int queue_index = sk_tx_queue_get(sk);
3052 if (queue_index < 0 || skb->ooo_okay ||
3053 queue_index >= dev->real_num_tx_queues) {
3054 int new_index = get_xps_queue(dev, skb);
3056 new_index = skb_tx_hash(dev, skb);
3058 if (queue_index != new_index && sk &&
3060 rcu_access_pointer(sk->sk_dst_cache))
3061 sk_tx_queue_set(sk, new_index);
3063 queue_index = new_index;
3069 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3070 struct sk_buff *skb,
3073 int queue_index = 0;
3076 if (skb->sender_cpu == 0)
3077 skb->sender_cpu = raw_smp_processor_id() + 1;
3080 if (dev->real_num_tx_queues != 1) {
3081 const struct net_device_ops *ops = dev->netdev_ops;
3082 if (ops->ndo_select_queue)
3083 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3086 queue_index = __netdev_pick_tx(dev, skb);
3089 queue_index = netdev_cap_txqueue(dev, queue_index);
3092 skb_set_queue_mapping(skb, queue_index);
3093 return netdev_get_tx_queue(dev, queue_index);
3097 * __dev_queue_xmit - transmit a buffer
3098 * @skb: buffer to transmit
3099 * @accel_priv: private data used for L2 forwarding offload
3101 * Queue a buffer for transmission to a network device. The caller must
3102 * have set the device and priority and built the buffer before calling
3103 * this function. The function can be called from an interrupt.
3105 * A negative errno code is returned on a failure. A success does not
3106 * guarantee the frame will be transmitted as it may be dropped due
3107 * to congestion or traffic shaping.
3109 * -----------------------------------------------------------------------------------
3110 * I notice this method can also return errors from the queue disciplines,
3111 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3114 * Regardless of the return value, the skb is consumed, so it is currently
3115 * difficult to retry a send to this method. (You can bump the ref count
3116 * before sending to hold a reference for retry if you are careful.)
3118 * When calling this method, interrupts MUST be enabled. This is because
3119 * the BH enable code must have IRQs enabled so that it will not deadlock.
3122 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3124 struct net_device *dev = skb->dev;
3125 struct netdev_queue *txq;
3129 skb_reset_mac_header(skb);
3131 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3132 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3134 /* Disable soft irqs for various locks below. Also
3135 * stops preemption for RCU.
3139 skb_update_prio(skb);
3141 /* If device/qdisc don't need skb->dst, release it right now while
3142 * its hot in this cpu cache.
3144 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3149 #ifdef CONFIG_NET_SWITCHDEV
3150 /* Don't forward if offload device already forwarded */
3151 if (skb->offload_fwd_mark &&
3152 skb->offload_fwd_mark == dev->offload_fwd_mark) {
3154 rc = NET_XMIT_SUCCESS;
3159 txq = netdev_pick_tx(dev, skb, accel_priv);
3160 q = rcu_dereference_bh(txq->qdisc);
3162 #ifdef CONFIG_NET_CLS_ACT
3163 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3165 trace_net_dev_queue(skb);
3167 rc = __dev_xmit_skb(skb, q, dev, txq);
3171 /* The device has no queue. Common case for software devices:
3172 loopback, all the sorts of tunnels...
3174 Really, it is unlikely that netif_tx_lock protection is necessary
3175 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3177 However, it is possible, that they rely on protection
3180 Check this and shot the lock. It is not prone from deadlocks.
3181 Either shot noqueue qdisc, it is even simpler 8)
3183 if (dev->flags & IFF_UP) {
3184 int cpu = smp_processor_id(); /* ok because BHs are off */
3186 if (txq->xmit_lock_owner != cpu) {
3188 if (xmit_rec_read() > RECURSION_LIMIT)
3189 goto recursion_alert;
3191 skb = validate_xmit_skb(skb, dev);
3195 HARD_TX_LOCK(dev, txq, cpu);
3197 if (!netif_xmit_stopped(txq)) {
3199 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3201 if (dev_xmit_complete(rc)) {
3202 HARD_TX_UNLOCK(dev, txq);
3206 HARD_TX_UNLOCK(dev, txq);
3207 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3210 /* Recursion is detected! It is possible,
3214 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3221 rcu_read_unlock_bh();
3223 atomic_long_inc(&dev->tx_dropped);
3224 kfree_skb_list(skb);
3227 rcu_read_unlock_bh();
3231 int dev_queue_xmit(struct sk_buff *skb)
3233 return __dev_queue_xmit(skb, NULL);
3235 EXPORT_SYMBOL(dev_queue_xmit);
3237 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3239 return __dev_queue_xmit(skb, accel_priv);
3241 EXPORT_SYMBOL(dev_queue_xmit_accel);
3244 /*=======================================================================
3246 =======================================================================*/
3248 int netdev_max_backlog __read_mostly = 1000;
3249 EXPORT_SYMBOL(netdev_max_backlog);
3251 int netdev_tstamp_prequeue __read_mostly = 1;
3252 int netdev_budget __read_mostly = 300;
3253 int weight_p __read_mostly = 64; /* old backlog weight */
3255 /* Called with irq disabled */
3256 static inline void ____napi_schedule(struct softnet_data *sd,
3257 struct napi_struct *napi)
3259 list_add_tail(&napi->poll_list, &sd->poll_list);
3260 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3265 /* One global table that all flow-based protocols share. */
3266 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3267 EXPORT_SYMBOL(rps_sock_flow_table);
3268 u32 rps_cpu_mask __read_mostly;
3269 EXPORT_SYMBOL(rps_cpu_mask);
3271 struct static_key rps_needed __read_mostly;
3273 static struct rps_dev_flow *
3274 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3275 struct rps_dev_flow *rflow, u16 next_cpu)
3277 if (next_cpu < nr_cpu_ids) {
3278 #ifdef CONFIG_RFS_ACCEL
3279 struct netdev_rx_queue *rxqueue;
3280 struct rps_dev_flow_table *flow_table;
3281 struct rps_dev_flow *old_rflow;
3286 /* Should we steer this flow to a different hardware queue? */
3287 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3288 !(dev->features & NETIF_F_NTUPLE))
3290 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3291 if (rxq_index == skb_get_rx_queue(skb))
3294 rxqueue = dev->_rx + rxq_index;
3295 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3298 flow_id = skb_get_hash(skb) & flow_table->mask;
3299 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3300 rxq_index, flow_id);
3304 rflow = &flow_table->flows[flow_id];
3306 if (old_rflow->filter == rflow->filter)
3307 old_rflow->filter = RPS_NO_FILTER;
3311 per_cpu(softnet_data, next_cpu).input_queue_head;
3314 rflow->cpu = next_cpu;
3319 * get_rps_cpu is called from netif_receive_skb and returns the target
3320 * CPU from the RPS map of the receiving queue for a given skb.
3321 * rcu_read_lock must be held on entry.
3323 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3324 struct rps_dev_flow **rflowp)
3326 const struct rps_sock_flow_table *sock_flow_table;
3327 struct netdev_rx_queue *rxqueue = dev->_rx;
3328 struct rps_dev_flow_table *flow_table;
3329 struct rps_map *map;
3334 if (skb_rx_queue_recorded(skb)) {
3335 u16 index = skb_get_rx_queue(skb);
3337 if (unlikely(index >= dev->real_num_rx_queues)) {
3338 WARN_ONCE(dev->real_num_rx_queues > 1,
3339 "%s received packet on queue %u, but number "
3340 "of RX queues is %u\n",
3341 dev->name, index, dev->real_num_rx_queues);
3347 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3349 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3350 map = rcu_dereference(rxqueue->rps_map);
3351 if (!flow_table && !map)
3354 skb_reset_network_header(skb);
3355 hash = skb_get_hash(skb);
3359 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3360 if (flow_table && sock_flow_table) {
3361 struct rps_dev_flow *rflow;
3365 /* First check into global flow table if there is a match */
3366 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3367 if ((ident ^ hash) & ~rps_cpu_mask)
3370 next_cpu = ident & rps_cpu_mask;
3372 /* OK, now we know there is a match,
3373 * we can look at the local (per receive queue) flow table
3375 rflow = &flow_table->flows[hash & flow_table->mask];
3379 * If the desired CPU (where last recvmsg was done) is
3380 * different from current CPU (one in the rx-queue flow
3381 * table entry), switch if one of the following holds:
3382 * - Current CPU is unset (>= nr_cpu_ids).
3383 * - Current CPU is offline.
3384 * - The current CPU's queue tail has advanced beyond the
3385 * last packet that was enqueued using this table entry.
3386 * This guarantees that all previous packets for the flow
3387 * have been dequeued, thus preserving in order delivery.
3389 if (unlikely(tcpu != next_cpu) &&
3390 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3391 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3392 rflow->last_qtail)) >= 0)) {
3394 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3397 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3407 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3408 if (cpu_online(tcpu)) {
3418 #ifdef CONFIG_RFS_ACCEL
3421 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3422 * @dev: Device on which the filter was set
3423 * @rxq_index: RX queue index
3424 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3425 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3427 * Drivers that implement ndo_rx_flow_steer() should periodically call
3428 * this function for each installed filter and remove the filters for
3429 * which it returns %true.
3431 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3432 u32 flow_id, u16 filter_id)
3434 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3435 struct rps_dev_flow_table *flow_table;
3436 struct rps_dev_flow *rflow;
3441 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3442 if (flow_table && flow_id <= flow_table->mask) {
3443 rflow = &flow_table->flows[flow_id];
3444 cpu = ACCESS_ONCE(rflow->cpu);
3445 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3446 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3447 rflow->last_qtail) <
3448 (int)(10 * flow_table->mask)))
3454 EXPORT_SYMBOL(rps_may_expire_flow);
3456 #endif /* CONFIG_RFS_ACCEL */
3458 /* Called from hardirq (IPI) context */
3459 static void rps_trigger_softirq(void *data)
3461 struct softnet_data *sd = data;
3463 ____napi_schedule(sd, &sd->backlog);
3467 #endif /* CONFIG_RPS */
3470 * Check if this softnet_data structure is another cpu one
3471 * If yes, queue it to our IPI list and return 1
3474 static int rps_ipi_queued(struct softnet_data *sd)
3477 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3480 sd->rps_ipi_next = mysd->rps_ipi_list;
3481 mysd->rps_ipi_list = sd;
3483 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3486 #endif /* CONFIG_RPS */
3490 #ifdef CONFIG_NET_FLOW_LIMIT
3491 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3494 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3496 #ifdef CONFIG_NET_FLOW_LIMIT
3497 struct sd_flow_limit *fl;
3498 struct softnet_data *sd;
3499 unsigned int old_flow, new_flow;
3501 if (qlen < (netdev_max_backlog >> 1))
3504 sd = this_cpu_ptr(&softnet_data);
3507 fl = rcu_dereference(sd->flow_limit);
3509 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3510 old_flow = fl->history[fl->history_head];
3511 fl->history[fl->history_head] = new_flow;
3514 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3516 if (likely(fl->buckets[old_flow]))
3517 fl->buckets[old_flow]--;
3519 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3531 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3532 * queue (may be a remote CPU queue).
3534 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3535 unsigned int *qtail)
3537 struct softnet_data *sd;
3538 unsigned long flags;
3541 sd = &per_cpu(softnet_data, cpu);
3543 local_irq_save(flags);
3546 if (!netif_running(skb->dev))
3548 qlen = skb_queue_len(&sd->input_pkt_queue);
3549 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3552 __skb_queue_tail(&sd->input_pkt_queue, skb);
3553 input_queue_tail_incr_save(sd, qtail);
3555 local_irq_restore(flags);
3556 return NET_RX_SUCCESS;
3559 /* Schedule NAPI for backlog device
3560 * We can use non atomic operation since we own the queue lock
3562 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3563 if (!rps_ipi_queued(sd))
3564 ____napi_schedule(sd, &sd->backlog);
3573 local_irq_restore(flags);
3574 preempt_check_resched_rt();
3576 atomic_long_inc(&skb->dev->rx_dropped);
3581 static int netif_rx_internal(struct sk_buff *skb)
3585 net_timestamp_check(netdev_tstamp_prequeue, skb);
3587 trace_netif_rx(skb);
3589 if (static_key_false(&rps_needed)) {
3590 struct rps_dev_flow voidflow, *rflow = &voidflow;
3596 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3598 cpu = smp_processor_id();
3600 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3608 ret = enqueue_to_backlog(skb, get_cpu_light(), &qtail);
3615 * netif_rx - post buffer to the network code
3616 * @skb: buffer to post
3618 * This function receives a packet from a device driver and queues it for
3619 * the upper (protocol) levels to process. It always succeeds. The buffer
3620 * may be dropped during processing for congestion control or by the
3624 * NET_RX_SUCCESS (no congestion)
3625 * NET_RX_DROP (packet was dropped)
3629 int netif_rx(struct sk_buff *skb)
3631 trace_netif_rx_entry(skb);
3633 return netif_rx_internal(skb);
3635 EXPORT_SYMBOL(netif_rx);
3637 int netif_rx_ni(struct sk_buff *skb)
3641 trace_netif_rx_ni_entry(skb);
3644 err = netif_rx_internal(skb);
3649 EXPORT_SYMBOL(netif_rx_ni);
3651 #ifdef CONFIG_PREEMPT_RT_FULL
3653 * RT runs ksoftirqd as a real time thread and the root_lock is a
3654 * "sleeping spinlock". If the trylock fails then we can go into an
3655 * infinite loop when ksoftirqd preempted the task which actually
3656 * holds the lock, because we requeue q and raise NET_TX softirq
3657 * causing ksoftirqd to loop forever.
3659 * It's safe to use spin_lock on RT here as softirqs run in thread
3660 * context and cannot deadlock against the thread which is holding
3663 * On !RT the trylock might fail, but there we bail out from the
3664 * softirq loop after 10 attempts which we can't do on RT. And the
3665 * task holding root_lock cannot be preempted, so the only downside of
3666 * that trylock is that we need 10 loops to decide that we should have
3667 * given up in the first one :)
3669 static inline int take_root_lock(spinlock_t *lock)
3675 static inline int take_root_lock(spinlock_t *lock)
3677 return spin_trylock(lock);
3681 static void net_tx_action(struct softirq_action *h)
3683 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3685 if (sd->completion_queue) {
3686 struct sk_buff *clist;
3688 local_irq_disable();
3689 clist = sd->completion_queue;
3690 sd->completion_queue = NULL;
3694 struct sk_buff *skb = clist;
3695 clist = clist->next;
3697 WARN_ON(atomic_read(&skb->users));
3698 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3699 trace_consume_skb(skb);
3701 trace_kfree_skb(skb, net_tx_action);
3706 if (sd->output_queue) {
3709 local_irq_disable();
3710 head = sd->output_queue;
3711 sd->output_queue = NULL;
3712 sd->output_queue_tailp = &sd->output_queue;
3716 struct Qdisc *q = head;
3717 spinlock_t *root_lock;
3719 head = head->next_sched;
3721 root_lock = qdisc_lock(q);
3722 if (take_root_lock(root_lock)) {
3723 smp_mb__before_atomic();
3724 clear_bit(__QDISC_STATE_SCHED,
3727 spin_unlock(root_lock);
3729 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3731 __netif_reschedule(q);
3733 smp_mb__before_atomic();
3734 clear_bit(__QDISC_STATE_SCHED,
3742 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3743 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3744 /* This hook is defined here for ATM LANE */
3745 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3746 unsigned char *addr) __read_mostly;
3747 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3750 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3751 struct packet_type **pt_prev,
3752 int *ret, struct net_device *orig_dev)
3754 #ifdef CONFIG_NET_CLS_ACT
3755 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3756 struct tcf_result cl_res;
3758 /* If there's at least one ingress present somewhere (so
3759 * we get here via enabled static key), remaining devices
3760 * that are not configured with an ingress qdisc will bail
3766 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3770 qdisc_skb_cb(skb)->pkt_len = skb->len;
3771 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3772 qdisc_bstats_cpu_update(cl->q, skb);
3774 switch (tc_classify(skb, cl, &cl_res, false)) {
3776 case TC_ACT_RECLASSIFY:
3777 skb->tc_index = TC_H_MIN(cl_res.classid);
3780 qdisc_qstats_cpu_drop(cl->q);
3785 case TC_ACT_REDIRECT:
3786 /* skb_mac_header check was done by cls/act_bpf, so
3787 * we can safely push the L2 header back before
3788 * redirecting to another netdev
3790 __skb_push(skb, skb->mac_len);
3791 skb_do_redirect(skb);
3796 #endif /* CONFIG_NET_CLS_ACT */
3801 * netdev_rx_handler_register - register receive handler
3802 * @dev: device to register a handler for
3803 * @rx_handler: receive handler to register
3804 * @rx_handler_data: data pointer that is used by rx handler
3806 * Register a receive handler for a device. This handler will then be
3807 * called from __netif_receive_skb. A negative errno code is returned
3810 * The caller must hold the rtnl_mutex.
3812 * For a general description of rx_handler, see enum rx_handler_result.
3814 int netdev_rx_handler_register(struct net_device *dev,
3815 rx_handler_func_t *rx_handler,
3816 void *rx_handler_data)
3820 if (dev->rx_handler)
3823 /* Note: rx_handler_data must be set before rx_handler */
3824 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3825 rcu_assign_pointer(dev->rx_handler, rx_handler);
3829 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3832 * netdev_rx_handler_unregister - unregister receive handler
3833 * @dev: device to unregister a handler from
3835 * Unregister a receive handler from a device.
3837 * The caller must hold the rtnl_mutex.
3839 void netdev_rx_handler_unregister(struct net_device *dev)
3843 RCU_INIT_POINTER(dev->rx_handler, NULL);
3844 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3845 * section has a guarantee to see a non NULL rx_handler_data
3849 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3851 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3854 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3855 * the special handling of PFMEMALLOC skbs.
3857 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3859 switch (skb->protocol) {
3860 case htons(ETH_P_ARP):
3861 case htons(ETH_P_IP):
3862 case htons(ETH_P_IPV6):
3863 case htons(ETH_P_8021Q):
3864 case htons(ETH_P_8021AD):
3871 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
3872 int *ret, struct net_device *orig_dev)
3874 #ifdef CONFIG_NETFILTER_INGRESS
3875 if (nf_hook_ingress_active(skb)) {
3877 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3881 return nf_hook_ingress(skb);
3883 #endif /* CONFIG_NETFILTER_INGRESS */
3887 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3889 struct packet_type *ptype, *pt_prev;
3890 rx_handler_func_t *rx_handler;
3891 struct net_device *orig_dev;
3892 bool deliver_exact = false;
3893 int ret = NET_RX_DROP;
3896 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3898 trace_netif_receive_skb(skb);
3900 orig_dev = skb->dev;
3902 skb_reset_network_header(skb);
3903 if (!skb_transport_header_was_set(skb))
3904 skb_reset_transport_header(skb);
3905 skb_reset_mac_len(skb);
3910 skb->skb_iif = skb->dev->ifindex;
3912 __this_cpu_inc(softnet_data.processed);
3914 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3915 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3916 skb = skb_vlan_untag(skb);
3921 #ifdef CONFIG_NET_CLS_ACT
3922 if (skb->tc_verd & TC_NCLS) {
3923 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3931 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3933 ret = deliver_skb(skb, pt_prev, orig_dev);
3937 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3939 ret = deliver_skb(skb, pt_prev, orig_dev);
3944 #ifdef CONFIG_NET_INGRESS
3945 if (static_key_false(&ingress_needed)) {
3946 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3950 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
3954 #ifdef CONFIG_NET_CLS_ACT
3958 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3961 if (skb_vlan_tag_present(skb)) {
3963 ret = deliver_skb(skb, pt_prev, orig_dev);
3966 if (vlan_do_receive(&skb))
3968 else if (unlikely(!skb))
3972 rx_handler = rcu_dereference(skb->dev->rx_handler);
3975 ret = deliver_skb(skb, pt_prev, orig_dev);
3978 switch (rx_handler(&skb)) {
3979 case RX_HANDLER_CONSUMED:
3980 ret = NET_RX_SUCCESS;
3982 case RX_HANDLER_ANOTHER:
3984 case RX_HANDLER_EXACT:
3985 deliver_exact = true;
3986 case RX_HANDLER_PASS:
3993 if (unlikely(skb_vlan_tag_present(skb))) {
3994 if (skb_vlan_tag_get_id(skb))
3995 skb->pkt_type = PACKET_OTHERHOST;
3996 /* Note: we might in the future use prio bits
3997 * and set skb->priority like in vlan_do_receive()
3998 * For the time being, just ignore Priority Code Point
4003 type = skb->protocol;
4005 /* deliver only exact match when indicated */
4006 if (likely(!deliver_exact)) {
4007 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4008 &ptype_base[ntohs(type) &
4012 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4013 &orig_dev->ptype_specific);
4015 if (unlikely(skb->dev != orig_dev)) {
4016 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4017 &skb->dev->ptype_specific);
4021 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
4024 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4027 atomic_long_inc(&skb->dev->rx_dropped);
4029 /* Jamal, now you will not able to escape explaining
4030 * me how you were going to use this. :-)
4039 static int __netif_receive_skb(struct sk_buff *skb)
4043 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4044 unsigned long pflags = current->flags;
4047 * PFMEMALLOC skbs are special, they should
4048 * - be delivered to SOCK_MEMALLOC sockets only
4049 * - stay away from userspace
4050 * - have bounded memory usage
4052 * Use PF_MEMALLOC as this saves us from propagating the allocation
4053 * context down to all allocation sites.
4055 current->flags |= PF_MEMALLOC;
4056 ret = __netif_receive_skb_core(skb, true);
4057 tsk_restore_flags(current, pflags, PF_MEMALLOC);
4059 ret = __netif_receive_skb_core(skb, false);
4064 static int netif_receive_skb_internal(struct sk_buff *skb)
4068 net_timestamp_check(netdev_tstamp_prequeue, skb);
4070 if (skb_defer_rx_timestamp(skb))
4071 return NET_RX_SUCCESS;
4076 if (static_key_false(&rps_needed)) {
4077 struct rps_dev_flow voidflow, *rflow = &voidflow;
4078 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4081 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4087 ret = __netif_receive_skb(skb);
4093 * netif_receive_skb - process receive buffer from network
4094 * @skb: buffer to process
4096 * netif_receive_skb() is the main receive data processing function.
4097 * It always succeeds. The buffer may be dropped during processing
4098 * for congestion control or by the protocol layers.
4100 * This function may only be called from softirq context and interrupts
4101 * should be enabled.
4103 * Return values (usually ignored):
4104 * NET_RX_SUCCESS: no congestion
4105 * NET_RX_DROP: packet was dropped
4107 int netif_receive_skb(struct sk_buff *skb)
4109 trace_netif_receive_skb_entry(skb);
4111 return netif_receive_skb_internal(skb);
4113 EXPORT_SYMBOL(netif_receive_skb);
4115 /* Network device is going away, flush any packets still pending
4116 * Called with irqs disabled.
4118 static void flush_backlog(void *arg)
4120 struct net_device *dev = arg;
4121 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4122 struct sk_buff *skb, *tmp;
4125 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4126 if (skb->dev == dev) {
4127 __skb_unlink(skb, &sd->input_pkt_queue);
4128 __skb_queue_tail(&sd->tofree_queue, skb);
4129 input_queue_head_incr(sd);
4134 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4135 if (skb->dev == dev) {
4136 __skb_unlink(skb, &sd->process_queue);
4137 __skb_queue_tail(&sd->tofree_queue, skb);
4138 input_queue_head_incr(sd);
4142 if (!skb_queue_empty(&sd->tofree_queue))
4143 raise_softirq_irqoff(NET_RX_SOFTIRQ);
4146 static int napi_gro_complete(struct sk_buff *skb)
4148 struct packet_offload *ptype;
4149 __be16 type = skb->protocol;
4150 struct list_head *head = &offload_base;
4153 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4155 if (NAPI_GRO_CB(skb)->count == 1) {
4156 skb_shinfo(skb)->gso_size = 0;
4161 list_for_each_entry_rcu(ptype, head, list) {
4162 if (ptype->type != type || !ptype->callbacks.gro_complete)
4165 err = ptype->callbacks.gro_complete(skb, 0);
4171 WARN_ON(&ptype->list == head);
4173 return NET_RX_SUCCESS;
4177 return netif_receive_skb_internal(skb);
4180 /* napi->gro_list contains packets ordered by age.
4181 * youngest packets at the head of it.
4182 * Complete skbs in reverse order to reduce latencies.
4184 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4186 struct sk_buff *skb, *prev = NULL;
4188 /* scan list and build reverse chain */
4189 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4194 for (skb = prev; skb; skb = prev) {
4197 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4201 napi_gro_complete(skb);
4205 napi->gro_list = NULL;
4207 EXPORT_SYMBOL(napi_gro_flush);
4209 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4212 unsigned int maclen = skb->dev->hard_header_len;
4213 u32 hash = skb_get_hash_raw(skb);
4215 for (p = napi->gro_list; p; p = p->next) {
4216 unsigned long diffs;
4218 NAPI_GRO_CB(p)->flush = 0;
4220 if (hash != skb_get_hash_raw(p)) {
4221 NAPI_GRO_CB(p)->same_flow = 0;
4225 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4226 diffs |= p->vlan_tci ^ skb->vlan_tci;
4227 diffs |= skb_metadata_dst_cmp(p, skb);
4228 if (maclen == ETH_HLEN)
4229 diffs |= compare_ether_header(skb_mac_header(p),
4230 skb_mac_header(skb));
4232 diffs = memcmp(skb_mac_header(p),
4233 skb_mac_header(skb),
4235 NAPI_GRO_CB(p)->same_flow = !diffs;
4239 static void skb_gro_reset_offset(struct sk_buff *skb)
4241 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4242 const skb_frag_t *frag0 = &pinfo->frags[0];
4244 NAPI_GRO_CB(skb)->data_offset = 0;
4245 NAPI_GRO_CB(skb)->frag0 = NULL;
4246 NAPI_GRO_CB(skb)->frag0_len = 0;
4248 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4250 !PageHighMem(skb_frag_page(frag0))) {
4251 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4252 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4256 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4258 struct skb_shared_info *pinfo = skb_shinfo(skb);
4260 BUG_ON(skb->end - skb->tail < grow);
4262 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4264 skb->data_len -= grow;
4267 pinfo->frags[0].page_offset += grow;
4268 skb_frag_size_sub(&pinfo->frags[0], grow);
4270 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4271 skb_frag_unref(skb, 0);
4272 memmove(pinfo->frags, pinfo->frags + 1,
4273 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4277 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4279 struct sk_buff **pp = NULL;
4280 struct packet_offload *ptype;
4281 __be16 type = skb->protocol;
4282 struct list_head *head = &offload_base;
4284 enum gro_result ret;
4287 if (!(skb->dev->features & NETIF_F_GRO))
4290 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4293 gro_list_prepare(napi, skb);
4296 list_for_each_entry_rcu(ptype, head, list) {
4297 if (ptype->type != type || !ptype->callbacks.gro_receive)
4300 skb_set_network_header(skb, skb_gro_offset(skb));
4301 skb_reset_mac_len(skb);
4302 NAPI_GRO_CB(skb)->same_flow = 0;
4303 NAPI_GRO_CB(skb)->flush = 0;
4304 NAPI_GRO_CB(skb)->free = 0;
4305 NAPI_GRO_CB(skb)->udp_mark = 0;
4306 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4308 /* Setup for GRO checksum validation */
4309 switch (skb->ip_summed) {
4310 case CHECKSUM_COMPLETE:
4311 NAPI_GRO_CB(skb)->csum = skb->csum;
4312 NAPI_GRO_CB(skb)->csum_valid = 1;
4313 NAPI_GRO_CB(skb)->csum_cnt = 0;
4315 case CHECKSUM_UNNECESSARY:
4316 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4317 NAPI_GRO_CB(skb)->csum_valid = 0;
4320 NAPI_GRO_CB(skb)->csum_cnt = 0;
4321 NAPI_GRO_CB(skb)->csum_valid = 0;
4324 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4329 if (&ptype->list == head)
4332 same_flow = NAPI_GRO_CB(skb)->same_flow;
4333 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4336 struct sk_buff *nskb = *pp;
4340 napi_gro_complete(nskb);
4347 if (NAPI_GRO_CB(skb)->flush)
4350 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4351 struct sk_buff *nskb = napi->gro_list;
4353 /* locate the end of the list to select the 'oldest' flow */
4354 while (nskb->next) {
4360 napi_gro_complete(nskb);
4364 NAPI_GRO_CB(skb)->count = 1;
4365 NAPI_GRO_CB(skb)->age = jiffies;
4366 NAPI_GRO_CB(skb)->last = skb;
4367 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4368 skb->next = napi->gro_list;
4369 napi->gro_list = skb;
4373 grow = skb_gro_offset(skb) - skb_headlen(skb);
4375 gro_pull_from_frag0(skb, grow);
4384 struct packet_offload *gro_find_receive_by_type(__be16 type)
4386 struct list_head *offload_head = &offload_base;
4387 struct packet_offload *ptype;
4389 list_for_each_entry_rcu(ptype, offload_head, list) {
4390 if (ptype->type != type || !ptype->callbacks.gro_receive)
4396 EXPORT_SYMBOL(gro_find_receive_by_type);
4398 struct packet_offload *gro_find_complete_by_type(__be16 type)
4400 struct list_head *offload_head = &offload_base;
4401 struct packet_offload *ptype;
4403 list_for_each_entry_rcu(ptype, offload_head, list) {
4404 if (ptype->type != type || !ptype->callbacks.gro_complete)
4410 EXPORT_SYMBOL(gro_find_complete_by_type);
4412 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4416 if (netif_receive_skb_internal(skb))
4424 case GRO_MERGED_FREE:
4425 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) {
4427 kmem_cache_free(skbuff_head_cache, skb);
4441 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4443 trace_napi_gro_receive_entry(skb);
4445 skb_gro_reset_offset(skb);
4447 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4449 EXPORT_SYMBOL(napi_gro_receive);
4451 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4453 if (unlikely(skb->pfmemalloc)) {
4457 __skb_pull(skb, skb_headlen(skb));
4458 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4459 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4461 skb->dev = napi->dev;
4463 skb->encapsulation = 0;
4464 skb_shinfo(skb)->gso_type = 0;
4465 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4470 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4472 struct sk_buff *skb = napi->skb;
4475 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4480 EXPORT_SYMBOL(napi_get_frags);
4482 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4483 struct sk_buff *skb,
4489 __skb_push(skb, ETH_HLEN);
4490 skb->protocol = eth_type_trans(skb, skb->dev);
4491 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4496 case GRO_MERGED_FREE:
4497 napi_reuse_skb(napi, skb);
4507 /* Upper GRO stack assumes network header starts at gro_offset=0
4508 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4509 * We copy ethernet header into skb->data to have a common layout.
4511 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4513 struct sk_buff *skb = napi->skb;
4514 const struct ethhdr *eth;
4515 unsigned int hlen = sizeof(*eth);
4519 skb_reset_mac_header(skb);
4520 skb_gro_reset_offset(skb);
4522 eth = skb_gro_header_fast(skb, 0);
4523 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4524 eth = skb_gro_header_slow(skb, hlen, 0);
4525 if (unlikely(!eth)) {
4526 napi_reuse_skb(napi, skb);
4530 gro_pull_from_frag0(skb, hlen);
4531 NAPI_GRO_CB(skb)->frag0 += hlen;
4532 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4534 __skb_pull(skb, hlen);
4537 * This works because the only protocols we care about don't require
4539 * We'll fix it up properly in napi_frags_finish()
4541 skb->protocol = eth->h_proto;
4546 gro_result_t napi_gro_frags(struct napi_struct *napi)
4548 struct sk_buff *skb = napi_frags_skb(napi);
4553 trace_napi_gro_frags_entry(skb);
4555 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4557 EXPORT_SYMBOL(napi_gro_frags);
4559 /* Compute the checksum from gro_offset and return the folded value
4560 * after adding in any pseudo checksum.
4562 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4567 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4569 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4570 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4572 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4573 !skb->csum_complete_sw)
4574 netdev_rx_csum_fault(skb->dev);
4577 NAPI_GRO_CB(skb)->csum = wsum;
4578 NAPI_GRO_CB(skb)->csum_valid = 1;
4582 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4585 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4586 * Note: called with local irq disabled, but exits with local irq enabled.
4588 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4591 struct softnet_data *remsd = sd->rps_ipi_list;
4594 sd->rps_ipi_list = NULL;
4597 preempt_check_resched_rt();
4599 /* Send pending IPI's to kick RPS processing on remote cpus. */
4601 struct softnet_data *next = remsd->rps_ipi_next;
4603 if (cpu_online(remsd->cpu))
4604 smp_call_function_single_async(remsd->cpu,
4611 preempt_check_resched_rt();
4614 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4617 return sd->rps_ipi_list != NULL;
4623 static int process_backlog(struct napi_struct *napi, int quota)
4626 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4628 /* Check if we have pending ipi, its better to send them now,
4629 * not waiting net_rx_action() end.
4631 if (sd_has_rps_ipi_waiting(sd)) {
4632 local_irq_disable();
4633 net_rps_action_and_irq_enable(sd);
4636 napi->weight = weight_p;
4637 local_irq_disable();
4639 struct sk_buff *skb;
4641 while ((skb = __skb_dequeue(&sd->process_queue))) {
4644 __netif_receive_skb(skb);
4646 local_irq_disable();
4647 input_queue_head_incr(sd);
4648 if (++work >= quota) {
4655 if (skb_queue_empty(&sd->input_pkt_queue)) {
4657 * Inline a custom version of __napi_complete().
4658 * only current cpu owns and manipulates this napi,
4659 * and NAPI_STATE_SCHED is the only possible flag set
4661 * We can use a plain write instead of clear_bit(),
4662 * and we dont need an smp_mb() memory barrier.
4670 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4671 &sd->process_queue);
4680 * __napi_schedule - schedule for receive
4681 * @n: entry to schedule
4683 * The entry's receive function will be scheduled to run.
4684 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4686 void __napi_schedule(struct napi_struct *n)
4688 unsigned long flags;
4690 local_irq_save(flags);
4691 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4692 local_irq_restore(flags);
4693 preempt_check_resched_rt();
4695 EXPORT_SYMBOL(__napi_schedule);
4698 * __napi_schedule_irqoff - schedule for receive
4699 * @n: entry to schedule
4701 * Variant of __napi_schedule() assuming hard irqs are masked
4703 void __napi_schedule_irqoff(struct napi_struct *n)
4705 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4707 EXPORT_SYMBOL(__napi_schedule_irqoff);
4709 void __napi_complete(struct napi_struct *n)
4711 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4713 list_del_init(&n->poll_list);
4714 smp_mb__before_atomic();
4715 clear_bit(NAPI_STATE_SCHED, &n->state);
4717 EXPORT_SYMBOL(__napi_complete);
4719 void napi_complete_done(struct napi_struct *n, int work_done)
4721 unsigned long flags;
4724 * don't let napi dequeue from the cpu poll list
4725 * just in case its running on a different cpu
4727 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4731 unsigned long timeout = 0;
4734 timeout = n->dev->gro_flush_timeout;
4737 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4738 HRTIMER_MODE_REL_PINNED);
4740 napi_gro_flush(n, false);
4742 if (likely(list_empty(&n->poll_list))) {
4743 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4745 /* If n->poll_list is not empty, we need to mask irqs */
4746 local_irq_save(flags);
4748 local_irq_restore(flags);
4751 EXPORT_SYMBOL(napi_complete_done);
4753 /* must be called under rcu_read_lock(), as we dont take a reference */
4754 struct napi_struct *napi_by_id(unsigned int napi_id)
4756 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4757 struct napi_struct *napi;
4759 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4760 if (napi->napi_id == napi_id)
4765 EXPORT_SYMBOL_GPL(napi_by_id);
4767 void napi_hash_add(struct napi_struct *napi)
4769 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4771 spin_lock(&napi_hash_lock);
4773 /* 0 is not a valid id, we also skip an id that is taken
4774 * we expect both events to be extremely rare
4777 while (!napi->napi_id) {
4778 napi->napi_id = ++napi_gen_id;
4779 if (napi_by_id(napi->napi_id))
4783 hlist_add_head_rcu(&napi->napi_hash_node,
4784 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4786 spin_unlock(&napi_hash_lock);
4789 EXPORT_SYMBOL_GPL(napi_hash_add);
4791 /* Warning : caller is responsible to make sure rcu grace period
4792 * is respected before freeing memory containing @napi
4794 void napi_hash_del(struct napi_struct *napi)
4796 spin_lock(&napi_hash_lock);
4798 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4799 hlist_del_rcu(&napi->napi_hash_node);
4801 spin_unlock(&napi_hash_lock);
4803 EXPORT_SYMBOL_GPL(napi_hash_del);
4805 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4807 struct napi_struct *napi;
4809 napi = container_of(timer, struct napi_struct, timer);
4811 napi_schedule(napi);
4813 return HRTIMER_NORESTART;
4816 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4817 int (*poll)(struct napi_struct *, int), int weight)
4819 INIT_LIST_HEAD(&napi->poll_list);
4820 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4821 napi->timer.function = napi_watchdog;
4822 napi->gro_count = 0;
4823 napi->gro_list = NULL;
4826 if (weight > NAPI_POLL_WEIGHT)
4827 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4829 napi->weight = weight;
4830 list_add(&napi->dev_list, &dev->napi_list);
4832 #ifdef CONFIG_NETPOLL
4833 spin_lock_init(&napi->poll_lock);
4834 napi->poll_owner = -1;
4836 set_bit(NAPI_STATE_SCHED, &napi->state);
4838 EXPORT_SYMBOL(netif_napi_add);
4840 void napi_disable(struct napi_struct *n)
4843 set_bit(NAPI_STATE_DISABLE, &n->state);
4845 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4847 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
4850 hrtimer_cancel(&n->timer);
4852 clear_bit(NAPI_STATE_DISABLE, &n->state);
4854 EXPORT_SYMBOL(napi_disable);
4856 void netif_napi_del(struct napi_struct *napi)
4858 list_del_init(&napi->dev_list);
4859 napi_free_frags(napi);
4861 kfree_skb_list(napi->gro_list);
4862 napi->gro_list = NULL;
4863 napi->gro_count = 0;
4865 EXPORT_SYMBOL(netif_napi_del);
4867 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4872 list_del_init(&n->poll_list);
4874 have = netpoll_poll_lock(n);
4878 /* This NAPI_STATE_SCHED test is for avoiding a race
4879 * with netpoll's poll_napi(). Only the entity which
4880 * obtains the lock and sees NAPI_STATE_SCHED set will
4881 * actually make the ->poll() call. Therefore we avoid
4882 * accidentally calling ->poll() when NAPI is not scheduled.
4885 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4886 work = n->poll(n, weight);
4890 WARN_ON_ONCE(work > weight);
4892 if (likely(work < weight))
4895 /* Drivers must not modify the NAPI state if they
4896 * consume the entire weight. In such cases this code
4897 * still "owns" the NAPI instance and therefore can
4898 * move the instance around on the list at-will.
4900 if (unlikely(napi_disable_pending(n))) {
4906 /* flush too old packets
4907 * If HZ < 1000, flush all packets.
4909 napi_gro_flush(n, HZ >= 1000);
4912 /* Some drivers may have called napi_schedule
4913 * prior to exhausting their budget.
4915 if (unlikely(!list_empty(&n->poll_list))) {
4916 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4917 n->dev ? n->dev->name : "backlog");
4921 list_add_tail(&n->poll_list, repoll);
4924 netpoll_poll_unlock(have);
4929 static void net_rx_action(struct softirq_action *h)
4931 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4932 unsigned long time_limit = jiffies + 2;
4933 int budget = netdev_budget;
4937 local_irq_disable();
4938 list_splice_init(&sd->poll_list, &list);
4942 struct napi_struct *n;
4944 if (list_empty(&list)) {
4945 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4950 n = list_first_entry(&list, struct napi_struct, poll_list);
4951 budget -= napi_poll(n, &repoll);
4953 /* If softirq window is exhausted then punt.
4954 * Allow this to run for 2 jiffies since which will allow
4955 * an average latency of 1.5/HZ.
4957 if (unlikely(budget <= 0 ||
4958 time_after_eq(jiffies, time_limit))) {
4964 local_irq_disable();
4966 list_splice_tail_init(&sd->poll_list, &list);
4967 list_splice_tail(&repoll, &list);
4968 list_splice(&list, &sd->poll_list);
4969 if (!list_empty(&sd->poll_list))
4970 __raise_softirq_irqoff_ksoft(NET_RX_SOFTIRQ);
4972 net_rps_action_and_irq_enable(sd);
4975 struct netdev_adjacent {
4976 struct net_device *dev;
4978 /* upper master flag, there can only be one master device per list */
4981 /* counter for the number of times this device was added to us */
4984 /* private field for the users */
4987 struct list_head list;
4988 struct rcu_head rcu;
4991 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
4992 struct list_head *adj_list)
4994 struct netdev_adjacent *adj;
4996 list_for_each_entry(adj, adj_list, list) {
4997 if (adj->dev == adj_dev)
5004 * netdev_has_upper_dev - Check if device is linked to an upper device
5006 * @upper_dev: upper device to check
5008 * Find out if a device is linked to specified upper device and return true
5009 * in case it is. Note that this checks only immediate upper device,
5010 * not through a complete stack of devices. The caller must hold the RTNL lock.
5012 bool netdev_has_upper_dev(struct net_device *dev,
5013 struct net_device *upper_dev)
5017 return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper);
5019 EXPORT_SYMBOL(netdev_has_upper_dev);
5022 * netdev_has_any_upper_dev - Check if device is linked to some device
5025 * Find out if a device is linked to an upper device and return true in case
5026 * it is. The caller must hold the RTNL lock.
5028 static bool netdev_has_any_upper_dev(struct net_device *dev)
5032 return !list_empty(&dev->all_adj_list.upper);
5036 * netdev_master_upper_dev_get - Get master upper device
5039 * Find a master upper device and return pointer to it or NULL in case
5040 * it's not there. The caller must hold the RTNL lock.
5042 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5044 struct netdev_adjacent *upper;
5048 if (list_empty(&dev->adj_list.upper))
5051 upper = list_first_entry(&dev->adj_list.upper,
5052 struct netdev_adjacent, list);
5053 if (likely(upper->master))
5057 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5059 void *netdev_adjacent_get_private(struct list_head *adj_list)
5061 struct netdev_adjacent *adj;
5063 adj = list_entry(adj_list, struct netdev_adjacent, list);
5065 return adj->private;
5067 EXPORT_SYMBOL(netdev_adjacent_get_private);
5070 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5072 * @iter: list_head ** of the current position
5074 * Gets the next device from the dev's upper list, starting from iter
5075 * position. The caller must hold RCU read lock.
5077 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5078 struct list_head **iter)
5080 struct netdev_adjacent *upper;
5082 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5084 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5086 if (&upper->list == &dev->adj_list.upper)
5089 *iter = &upper->list;
5093 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5096 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
5098 * @iter: list_head ** of the current position
5100 * Gets the next device from the dev's upper list, starting from iter
5101 * position. The caller must hold RCU read lock.
5103 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
5104 struct list_head **iter)
5106 struct netdev_adjacent *upper;
5108 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5110 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5112 if (&upper->list == &dev->all_adj_list.upper)
5115 *iter = &upper->list;
5119 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
5122 * netdev_lower_get_next_private - Get the next ->private from the
5123 * lower neighbour list
5125 * @iter: list_head ** of the current position
5127 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5128 * list, starting from iter position. The caller must hold either hold the
5129 * RTNL lock or its own locking that guarantees that the neighbour lower
5130 * list will remain unchanged.
5132 void *netdev_lower_get_next_private(struct net_device *dev,
5133 struct list_head **iter)
5135 struct netdev_adjacent *lower;
5137 lower = list_entry(*iter, struct netdev_adjacent, list);
5139 if (&lower->list == &dev->adj_list.lower)
5142 *iter = lower->list.next;
5144 return lower->private;
5146 EXPORT_SYMBOL(netdev_lower_get_next_private);
5149 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5150 * lower neighbour list, RCU
5153 * @iter: list_head ** of the current position
5155 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5156 * list, starting from iter position. The caller must hold RCU read lock.
5158 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5159 struct list_head **iter)
5161 struct netdev_adjacent *lower;
5163 WARN_ON_ONCE(!rcu_read_lock_held());
5165 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5167 if (&lower->list == &dev->adj_list.lower)
5170 *iter = &lower->list;
5172 return lower->private;
5174 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5177 * netdev_lower_get_next - Get the next device from the lower neighbour
5180 * @iter: list_head ** of the current position
5182 * Gets the next netdev_adjacent from the dev's lower neighbour
5183 * list, starting from iter position. The caller must hold RTNL lock or
5184 * its own locking that guarantees that the neighbour lower
5185 * list will remain unchanged.
5187 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5189 struct netdev_adjacent *lower;
5191 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5193 if (&lower->list == &dev->adj_list.lower)
5196 *iter = &lower->list;
5200 EXPORT_SYMBOL(netdev_lower_get_next);
5203 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5204 * lower neighbour list, RCU
5208 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5209 * list. The caller must hold RCU read lock.
5211 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5213 struct netdev_adjacent *lower;
5215 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5216 struct netdev_adjacent, list);
5218 return lower->private;
5221 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5224 * netdev_master_upper_dev_get_rcu - Get master upper device
5227 * Find a master upper device and return pointer to it or NULL in case
5228 * it's not there. The caller must hold the RCU read lock.
5230 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5232 struct netdev_adjacent *upper;
5234 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5235 struct netdev_adjacent, list);
5236 if (upper && likely(upper->master))
5240 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5242 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5243 struct net_device *adj_dev,
5244 struct list_head *dev_list)
5246 char linkname[IFNAMSIZ+7];
5247 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5248 "upper_%s" : "lower_%s", adj_dev->name);
5249 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5252 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5254 struct list_head *dev_list)
5256 char linkname[IFNAMSIZ+7];
5257 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5258 "upper_%s" : "lower_%s", name);
5259 sysfs_remove_link(&(dev->dev.kobj), linkname);
5262 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5263 struct net_device *adj_dev,
5264 struct list_head *dev_list)
5266 return (dev_list == &dev->adj_list.upper ||
5267 dev_list == &dev->adj_list.lower) &&
5268 net_eq(dev_net(dev), dev_net(adj_dev));
5271 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5272 struct net_device *adj_dev,
5273 struct list_head *dev_list,
5274 void *private, bool master)
5276 struct netdev_adjacent *adj;
5279 adj = __netdev_find_adj(adj_dev, dev_list);
5286 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5291 adj->master = master;
5293 adj->private = private;
5296 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5297 adj_dev->name, dev->name, adj_dev->name);
5299 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5300 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5305 /* Ensure that master link is always the first item in list. */
5307 ret = sysfs_create_link(&(dev->dev.kobj),
5308 &(adj_dev->dev.kobj), "master");
5310 goto remove_symlinks;
5312 list_add_rcu(&adj->list, dev_list);
5314 list_add_tail_rcu(&adj->list, dev_list);
5320 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5321 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5329 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5330 struct net_device *adj_dev,
5331 struct list_head *dev_list)
5333 struct netdev_adjacent *adj;
5335 adj = __netdev_find_adj(adj_dev, dev_list);
5338 pr_err("tried to remove device %s from %s\n",
5339 dev->name, adj_dev->name);
5343 if (adj->ref_nr > 1) {
5344 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5351 sysfs_remove_link(&(dev->dev.kobj), "master");
5353 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5354 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5356 list_del_rcu(&adj->list);
5357 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5358 adj_dev->name, dev->name, adj_dev->name);
5360 kfree_rcu(adj, rcu);
5363 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5364 struct net_device *upper_dev,
5365 struct list_head *up_list,
5366 struct list_head *down_list,
5367 void *private, bool master)
5371 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5376 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5379 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5386 static int __netdev_adjacent_dev_link(struct net_device *dev,
5387 struct net_device *upper_dev)
5389 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5390 &dev->all_adj_list.upper,
5391 &upper_dev->all_adj_list.lower,
5395 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5396 struct net_device *upper_dev,
5397 struct list_head *up_list,
5398 struct list_head *down_list)
5400 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5401 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5404 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5405 struct net_device *upper_dev)
5407 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5408 &dev->all_adj_list.upper,
5409 &upper_dev->all_adj_list.lower);
5412 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5413 struct net_device *upper_dev,
5414 void *private, bool master)
5416 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5421 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5422 &dev->adj_list.upper,
5423 &upper_dev->adj_list.lower,
5426 __netdev_adjacent_dev_unlink(dev, upper_dev);
5433 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5434 struct net_device *upper_dev)
5436 __netdev_adjacent_dev_unlink(dev, upper_dev);
5437 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5438 &dev->adj_list.upper,
5439 &upper_dev->adj_list.lower);
5442 static int __netdev_upper_dev_link(struct net_device *dev,
5443 struct net_device *upper_dev, bool master,
5446 struct netdev_notifier_changeupper_info changeupper_info;
5447 struct netdev_adjacent *i, *j, *to_i, *to_j;
5452 if (dev == upper_dev)
5455 /* To prevent loops, check if dev is not upper device to upper_dev. */
5456 if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper))
5459 if (__netdev_find_adj(upper_dev, &dev->adj_list.upper))
5462 if (master && netdev_master_upper_dev_get(dev))
5465 changeupper_info.upper_dev = upper_dev;
5466 changeupper_info.master = master;
5467 changeupper_info.linking = true;
5469 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5470 &changeupper_info.info);
5471 ret = notifier_to_errno(ret);
5475 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5480 /* Now that we linked these devs, make all the upper_dev's
5481 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5482 * versa, and don't forget the devices itself. All of these
5483 * links are non-neighbours.
5485 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5486 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5487 pr_debug("Interlinking %s with %s, non-neighbour\n",
5488 i->dev->name, j->dev->name);
5489 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5495 /* add dev to every upper_dev's upper device */
5496 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5497 pr_debug("linking %s's upper device %s with %s\n",
5498 upper_dev->name, i->dev->name, dev->name);
5499 ret = __netdev_adjacent_dev_link(dev, i->dev);
5501 goto rollback_upper_mesh;
5504 /* add upper_dev to every dev's lower device */
5505 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5506 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5507 i->dev->name, upper_dev->name);
5508 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5510 goto rollback_lower_mesh;
5513 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5514 &changeupper_info.info);
5517 rollback_lower_mesh:
5519 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5522 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5527 rollback_upper_mesh:
5529 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5532 __netdev_adjacent_dev_unlink(dev, i->dev);
5540 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5541 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5542 if (i == to_i && j == to_j)
5544 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5550 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5556 * netdev_upper_dev_link - Add a link to the upper device
5558 * @upper_dev: new upper device
5560 * Adds a link to device which is upper to this one. The caller must hold
5561 * the RTNL lock. On a failure a negative errno code is returned.
5562 * On success the reference counts are adjusted and the function
5565 int netdev_upper_dev_link(struct net_device *dev,
5566 struct net_device *upper_dev)
5568 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5570 EXPORT_SYMBOL(netdev_upper_dev_link);
5573 * netdev_master_upper_dev_link - Add a master link to the upper device
5575 * @upper_dev: new upper device
5577 * Adds a link to device which is upper to this one. In this case, only
5578 * one master upper device can be linked, although other non-master devices
5579 * might be linked as well. The caller must hold the RTNL lock.
5580 * On a failure a negative errno code is returned. On success the reference
5581 * counts are adjusted and the function returns zero.
5583 int netdev_master_upper_dev_link(struct net_device *dev,
5584 struct net_device *upper_dev)
5586 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5588 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5590 int netdev_master_upper_dev_link_private(struct net_device *dev,
5591 struct net_device *upper_dev,
5594 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5596 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5599 * netdev_upper_dev_unlink - Removes a link to upper device
5601 * @upper_dev: new upper device
5603 * Removes a link to device which is upper to this one. The caller must hold
5606 void netdev_upper_dev_unlink(struct net_device *dev,
5607 struct net_device *upper_dev)
5609 struct netdev_notifier_changeupper_info changeupper_info;
5610 struct netdev_adjacent *i, *j;
5613 changeupper_info.upper_dev = upper_dev;
5614 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
5615 changeupper_info.linking = false;
5617 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5618 &changeupper_info.info);
5620 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5622 /* Here is the tricky part. We must remove all dev's lower
5623 * devices from all upper_dev's upper devices and vice
5624 * versa, to maintain the graph relationship.
5626 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5627 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5628 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5630 /* remove also the devices itself from lower/upper device
5633 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5634 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5636 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5637 __netdev_adjacent_dev_unlink(dev, i->dev);
5639 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5640 &changeupper_info.info);
5642 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5645 * netdev_bonding_info_change - Dispatch event about slave change
5647 * @bonding_info: info to dispatch
5649 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5650 * The caller must hold the RTNL lock.
5652 void netdev_bonding_info_change(struct net_device *dev,
5653 struct netdev_bonding_info *bonding_info)
5655 struct netdev_notifier_bonding_info info;
5657 memcpy(&info.bonding_info, bonding_info,
5658 sizeof(struct netdev_bonding_info));
5659 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5662 EXPORT_SYMBOL(netdev_bonding_info_change);
5664 static void netdev_adjacent_add_links(struct net_device *dev)
5666 struct netdev_adjacent *iter;
5668 struct net *net = dev_net(dev);
5670 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5671 if (!net_eq(net,dev_net(iter->dev)))
5673 netdev_adjacent_sysfs_add(iter->dev, dev,
5674 &iter->dev->adj_list.lower);
5675 netdev_adjacent_sysfs_add(dev, iter->dev,
5676 &dev->adj_list.upper);
5679 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5680 if (!net_eq(net,dev_net(iter->dev)))
5682 netdev_adjacent_sysfs_add(iter->dev, dev,
5683 &iter->dev->adj_list.upper);
5684 netdev_adjacent_sysfs_add(dev, iter->dev,
5685 &dev->adj_list.lower);
5689 static void netdev_adjacent_del_links(struct net_device *dev)
5691 struct netdev_adjacent *iter;
5693 struct net *net = dev_net(dev);
5695 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5696 if (!net_eq(net,dev_net(iter->dev)))
5698 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5699 &iter->dev->adj_list.lower);
5700 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5701 &dev->adj_list.upper);
5704 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5705 if (!net_eq(net,dev_net(iter->dev)))
5707 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5708 &iter->dev->adj_list.upper);
5709 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5710 &dev->adj_list.lower);
5714 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5716 struct netdev_adjacent *iter;
5718 struct net *net = dev_net(dev);
5720 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5721 if (!net_eq(net,dev_net(iter->dev)))
5723 netdev_adjacent_sysfs_del(iter->dev, oldname,
5724 &iter->dev->adj_list.lower);
5725 netdev_adjacent_sysfs_add(iter->dev, dev,
5726 &iter->dev->adj_list.lower);
5729 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5730 if (!net_eq(net,dev_net(iter->dev)))
5732 netdev_adjacent_sysfs_del(iter->dev, oldname,
5733 &iter->dev->adj_list.upper);
5734 netdev_adjacent_sysfs_add(iter->dev, dev,
5735 &iter->dev->adj_list.upper);
5739 void *netdev_lower_dev_get_private(struct net_device *dev,
5740 struct net_device *lower_dev)
5742 struct netdev_adjacent *lower;
5746 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
5750 return lower->private;
5752 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5755 int dev_get_nest_level(struct net_device *dev,
5756 bool (*type_check)(struct net_device *dev))
5758 struct net_device *lower = NULL;
5759 struct list_head *iter;
5765 netdev_for_each_lower_dev(dev, lower, iter) {
5766 nest = dev_get_nest_level(lower, type_check);
5767 if (max_nest < nest)
5771 if (type_check(dev))
5776 EXPORT_SYMBOL(dev_get_nest_level);
5778 static void dev_change_rx_flags(struct net_device *dev, int flags)
5780 const struct net_device_ops *ops = dev->netdev_ops;
5782 if (ops->ndo_change_rx_flags)
5783 ops->ndo_change_rx_flags(dev, flags);
5786 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5788 unsigned int old_flags = dev->flags;
5794 dev->flags |= IFF_PROMISC;
5795 dev->promiscuity += inc;
5796 if (dev->promiscuity == 0) {
5799 * If inc causes overflow, untouch promisc and return error.
5802 dev->flags &= ~IFF_PROMISC;
5804 dev->promiscuity -= inc;
5805 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5810 if (dev->flags != old_flags) {
5811 pr_info("device %s %s promiscuous mode\n",
5813 dev->flags & IFF_PROMISC ? "entered" : "left");
5814 if (audit_enabled) {
5815 current_uid_gid(&uid, &gid);
5816 audit_log(current->audit_context, GFP_ATOMIC,
5817 AUDIT_ANOM_PROMISCUOUS,
5818 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5819 dev->name, (dev->flags & IFF_PROMISC),
5820 (old_flags & IFF_PROMISC),
5821 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5822 from_kuid(&init_user_ns, uid),
5823 from_kgid(&init_user_ns, gid),
5824 audit_get_sessionid(current));
5827 dev_change_rx_flags(dev, IFF_PROMISC);
5830 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5835 * dev_set_promiscuity - update promiscuity count on a device
5839 * Add or remove promiscuity from a device. While the count in the device
5840 * remains above zero the interface remains promiscuous. Once it hits zero
5841 * the device reverts back to normal filtering operation. A negative inc
5842 * value is used to drop promiscuity on the device.
5843 * Return 0 if successful or a negative errno code on error.
5845 int dev_set_promiscuity(struct net_device *dev, int inc)
5847 unsigned int old_flags = dev->flags;
5850 err = __dev_set_promiscuity(dev, inc, true);
5853 if (dev->flags != old_flags)
5854 dev_set_rx_mode(dev);
5857 EXPORT_SYMBOL(dev_set_promiscuity);
5859 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5861 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5865 dev->flags |= IFF_ALLMULTI;
5866 dev->allmulti += inc;
5867 if (dev->allmulti == 0) {
5870 * If inc causes overflow, untouch allmulti and return error.
5873 dev->flags &= ~IFF_ALLMULTI;
5875 dev->allmulti -= inc;
5876 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5881 if (dev->flags ^ old_flags) {
5882 dev_change_rx_flags(dev, IFF_ALLMULTI);
5883 dev_set_rx_mode(dev);
5885 __dev_notify_flags(dev, old_flags,
5886 dev->gflags ^ old_gflags);
5892 * dev_set_allmulti - update allmulti count on a device
5896 * Add or remove reception of all multicast frames to a device. While the
5897 * count in the device remains above zero the interface remains listening
5898 * to all interfaces. Once it hits zero the device reverts back to normal
5899 * filtering operation. A negative @inc value is used to drop the counter
5900 * when releasing a resource needing all multicasts.
5901 * Return 0 if successful or a negative errno code on error.
5904 int dev_set_allmulti(struct net_device *dev, int inc)
5906 return __dev_set_allmulti(dev, inc, true);
5908 EXPORT_SYMBOL(dev_set_allmulti);
5911 * Upload unicast and multicast address lists to device and
5912 * configure RX filtering. When the device doesn't support unicast
5913 * filtering it is put in promiscuous mode while unicast addresses
5916 void __dev_set_rx_mode(struct net_device *dev)
5918 const struct net_device_ops *ops = dev->netdev_ops;
5920 /* dev_open will call this function so the list will stay sane. */
5921 if (!(dev->flags&IFF_UP))
5924 if (!netif_device_present(dev))
5927 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5928 /* Unicast addresses changes may only happen under the rtnl,
5929 * therefore calling __dev_set_promiscuity here is safe.
5931 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5932 __dev_set_promiscuity(dev, 1, false);
5933 dev->uc_promisc = true;
5934 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5935 __dev_set_promiscuity(dev, -1, false);
5936 dev->uc_promisc = false;
5940 if (ops->ndo_set_rx_mode)
5941 ops->ndo_set_rx_mode(dev);
5944 void dev_set_rx_mode(struct net_device *dev)
5946 netif_addr_lock_bh(dev);
5947 __dev_set_rx_mode(dev);
5948 netif_addr_unlock_bh(dev);
5952 * dev_get_flags - get flags reported to userspace
5955 * Get the combination of flag bits exported through APIs to userspace.
5957 unsigned int dev_get_flags(const struct net_device *dev)
5961 flags = (dev->flags & ~(IFF_PROMISC |
5966 (dev->gflags & (IFF_PROMISC |
5969 if (netif_running(dev)) {
5970 if (netif_oper_up(dev))
5971 flags |= IFF_RUNNING;
5972 if (netif_carrier_ok(dev))
5973 flags |= IFF_LOWER_UP;
5974 if (netif_dormant(dev))
5975 flags |= IFF_DORMANT;
5980 EXPORT_SYMBOL(dev_get_flags);
5982 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5984 unsigned int old_flags = dev->flags;
5990 * Set the flags on our device.
5993 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5994 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5996 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6000 * Load in the correct multicast list now the flags have changed.
6003 if ((old_flags ^ flags) & IFF_MULTICAST)
6004 dev_change_rx_flags(dev, IFF_MULTICAST);
6006 dev_set_rx_mode(dev);
6009 * Have we downed the interface. We handle IFF_UP ourselves
6010 * according to user attempts to set it, rather than blindly
6015 if ((old_flags ^ flags) & IFF_UP)
6016 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
6018 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6019 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6020 unsigned int old_flags = dev->flags;
6022 dev->gflags ^= IFF_PROMISC;
6024 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6025 if (dev->flags != old_flags)
6026 dev_set_rx_mode(dev);
6029 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6030 is important. Some (broken) drivers set IFF_PROMISC, when
6031 IFF_ALLMULTI is requested not asking us and not reporting.
6033 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6034 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6036 dev->gflags ^= IFF_ALLMULTI;
6037 __dev_set_allmulti(dev, inc, false);
6043 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6044 unsigned int gchanges)
6046 unsigned int changes = dev->flags ^ old_flags;
6049 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6051 if (changes & IFF_UP) {
6052 if (dev->flags & IFF_UP)
6053 call_netdevice_notifiers(NETDEV_UP, dev);
6055 call_netdevice_notifiers(NETDEV_DOWN, dev);
6058 if (dev->flags & IFF_UP &&
6059 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6060 struct netdev_notifier_change_info change_info;
6062 change_info.flags_changed = changes;
6063 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6069 * dev_change_flags - change device settings
6071 * @flags: device state flags
6073 * Change settings on device based state flags. The flags are
6074 * in the userspace exported format.
6076 int dev_change_flags(struct net_device *dev, unsigned int flags)
6079 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6081 ret = __dev_change_flags(dev, flags);
6085 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6086 __dev_notify_flags(dev, old_flags, changes);
6089 EXPORT_SYMBOL(dev_change_flags);
6091 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6093 const struct net_device_ops *ops = dev->netdev_ops;
6095 if (ops->ndo_change_mtu)
6096 return ops->ndo_change_mtu(dev, new_mtu);
6103 * dev_set_mtu - Change maximum transfer unit
6105 * @new_mtu: new transfer unit
6107 * Change the maximum transfer size of the network device.
6109 int dev_set_mtu(struct net_device *dev, int new_mtu)
6113 if (new_mtu == dev->mtu)
6116 /* MTU must be positive. */
6120 if (!netif_device_present(dev))
6123 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6124 err = notifier_to_errno(err);
6128 orig_mtu = dev->mtu;
6129 err = __dev_set_mtu(dev, new_mtu);
6132 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6133 err = notifier_to_errno(err);
6135 /* setting mtu back and notifying everyone again,
6136 * so that they have a chance to revert changes.
6138 __dev_set_mtu(dev, orig_mtu);
6139 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6144 EXPORT_SYMBOL(dev_set_mtu);
6147 * dev_set_group - Change group this device belongs to
6149 * @new_group: group this device should belong to
6151 void dev_set_group(struct net_device *dev, int new_group)
6153 dev->group = new_group;
6155 EXPORT_SYMBOL(dev_set_group);
6158 * dev_set_mac_address - Change Media Access Control Address
6162 * Change the hardware (MAC) address of the device
6164 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6166 const struct net_device_ops *ops = dev->netdev_ops;
6169 if (!ops->ndo_set_mac_address)
6171 if (sa->sa_family != dev->type)
6173 if (!netif_device_present(dev))
6175 err = ops->ndo_set_mac_address(dev, sa);
6178 dev->addr_assign_type = NET_ADDR_SET;
6179 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6180 add_device_randomness(dev->dev_addr, dev->addr_len);
6183 EXPORT_SYMBOL(dev_set_mac_address);
6186 * dev_change_carrier - Change device carrier
6188 * @new_carrier: new value
6190 * Change device carrier
6192 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6194 const struct net_device_ops *ops = dev->netdev_ops;
6196 if (!ops->ndo_change_carrier)
6198 if (!netif_device_present(dev))
6200 return ops->ndo_change_carrier(dev, new_carrier);
6202 EXPORT_SYMBOL(dev_change_carrier);
6205 * dev_get_phys_port_id - Get device physical port ID
6209 * Get device physical port ID
6211 int dev_get_phys_port_id(struct net_device *dev,
6212 struct netdev_phys_item_id *ppid)
6214 const struct net_device_ops *ops = dev->netdev_ops;
6216 if (!ops->ndo_get_phys_port_id)
6218 return ops->ndo_get_phys_port_id(dev, ppid);
6220 EXPORT_SYMBOL(dev_get_phys_port_id);
6223 * dev_get_phys_port_name - Get device physical port name
6227 * Get device physical port name
6229 int dev_get_phys_port_name(struct net_device *dev,
6230 char *name, size_t len)
6232 const struct net_device_ops *ops = dev->netdev_ops;
6234 if (!ops->ndo_get_phys_port_name)
6236 return ops->ndo_get_phys_port_name(dev, name, len);
6238 EXPORT_SYMBOL(dev_get_phys_port_name);
6241 * dev_change_proto_down - update protocol port state information
6243 * @proto_down: new value
6245 * This info can be used by switch drivers to set the phys state of the
6248 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6250 const struct net_device_ops *ops = dev->netdev_ops;
6252 if (!ops->ndo_change_proto_down)
6254 if (!netif_device_present(dev))
6256 return ops->ndo_change_proto_down(dev, proto_down);
6258 EXPORT_SYMBOL(dev_change_proto_down);
6261 * dev_new_index - allocate an ifindex
6262 * @net: the applicable net namespace
6264 * Returns a suitable unique value for a new device interface
6265 * number. The caller must hold the rtnl semaphore or the
6266 * dev_base_lock to be sure it remains unique.
6268 static int dev_new_index(struct net *net)
6270 int ifindex = net->ifindex;
6274 if (!__dev_get_by_index(net, ifindex))
6275 return net->ifindex = ifindex;
6279 /* Delayed registration/unregisteration */
6280 static LIST_HEAD(net_todo_list);
6281 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6283 static void net_set_todo(struct net_device *dev)
6285 list_add_tail(&dev->todo_list, &net_todo_list);
6286 dev_net(dev)->dev_unreg_count++;
6289 static void rollback_registered_many(struct list_head *head)
6291 struct net_device *dev, *tmp;
6292 LIST_HEAD(close_head);
6294 BUG_ON(dev_boot_phase);
6297 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6298 /* Some devices call without registering
6299 * for initialization unwind. Remove those
6300 * devices and proceed with the remaining.
6302 if (dev->reg_state == NETREG_UNINITIALIZED) {
6303 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6307 list_del(&dev->unreg_list);
6310 dev->dismantle = true;
6311 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6314 /* If device is running, close it first. */
6315 list_for_each_entry(dev, head, unreg_list)
6316 list_add_tail(&dev->close_list, &close_head);
6317 dev_close_many(&close_head, true);
6319 list_for_each_entry(dev, head, unreg_list) {
6320 /* And unlink it from device chain. */
6321 unlist_netdevice(dev);
6323 dev->reg_state = NETREG_UNREGISTERING;
6324 on_each_cpu(flush_backlog, dev, 1);
6329 list_for_each_entry(dev, head, unreg_list) {
6330 struct sk_buff *skb = NULL;
6332 /* Shutdown queueing discipline. */
6336 /* Notify protocols, that we are about to destroy
6337 this device. They should clean all the things.
6339 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6341 if (!dev->rtnl_link_ops ||
6342 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6343 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6347 * Flush the unicast and multicast chains
6352 if (dev->netdev_ops->ndo_uninit)
6353 dev->netdev_ops->ndo_uninit(dev);
6356 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6358 /* Notifier chain MUST detach us all upper devices. */
6359 WARN_ON(netdev_has_any_upper_dev(dev));
6361 /* Remove entries from kobject tree */
6362 netdev_unregister_kobject(dev);
6364 /* Remove XPS queueing entries */
6365 netif_reset_xps_queues_gt(dev, 0);
6371 list_for_each_entry(dev, head, unreg_list)
6375 static void rollback_registered(struct net_device *dev)
6379 list_add(&dev->unreg_list, &single);
6380 rollback_registered_many(&single);
6384 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6385 struct net_device *upper, netdev_features_t features)
6387 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6388 netdev_features_t feature;
6391 for_each_netdev_feature(&upper_disables, feature_bit) {
6392 feature = __NETIF_F_BIT(feature_bit);
6393 if (!(upper->wanted_features & feature)
6394 && (features & feature)) {
6395 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6396 &feature, upper->name);
6397 features &= ~feature;
6404 static void netdev_sync_lower_features(struct net_device *upper,
6405 struct net_device *lower, netdev_features_t features)
6407 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6408 netdev_features_t feature;
6411 for_each_netdev_feature(&upper_disables, feature_bit) {
6412 feature = __NETIF_F_BIT(feature_bit);
6413 if (!(features & feature) && (lower->features & feature)) {
6414 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6415 &feature, lower->name);
6416 lower->wanted_features &= ~feature;
6417 netdev_update_features(lower);
6419 if (unlikely(lower->features & feature))
6420 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6421 &feature, lower->name);
6426 static netdev_features_t netdev_fix_features(struct net_device *dev,
6427 netdev_features_t features)
6429 /* Fix illegal checksum combinations */
6430 if ((features & NETIF_F_HW_CSUM) &&
6431 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6432 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6433 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6436 /* TSO requires that SG is present as well. */
6437 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6438 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6439 features &= ~NETIF_F_ALL_TSO;
6442 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6443 !(features & NETIF_F_IP_CSUM)) {
6444 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6445 features &= ~NETIF_F_TSO;
6446 features &= ~NETIF_F_TSO_ECN;
6449 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6450 !(features & NETIF_F_IPV6_CSUM)) {
6451 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6452 features &= ~NETIF_F_TSO6;
6455 /* TSO ECN requires that TSO is present as well. */
6456 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6457 features &= ~NETIF_F_TSO_ECN;
6459 /* Software GSO depends on SG. */
6460 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6461 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6462 features &= ~NETIF_F_GSO;
6465 /* UFO needs SG and checksumming */
6466 if (features & NETIF_F_UFO) {
6467 /* maybe split UFO into V4 and V6? */
6468 if (!((features & NETIF_F_GEN_CSUM) ||
6469 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6470 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6472 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6473 features &= ~NETIF_F_UFO;
6476 if (!(features & NETIF_F_SG)) {
6478 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6479 features &= ~NETIF_F_UFO;
6483 #ifdef CONFIG_NET_RX_BUSY_POLL
6484 if (dev->netdev_ops->ndo_busy_poll)
6485 features |= NETIF_F_BUSY_POLL;
6488 features &= ~NETIF_F_BUSY_POLL;
6493 int __netdev_update_features(struct net_device *dev)
6495 struct net_device *upper, *lower;
6496 netdev_features_t features;
6497 struct list_head *iter;
6502 features = netdev_get_wanted_features(dev);
6504 if (dev->netdev_ops->ndo_fix_features)
6505 features = dev->netdev_ops->ndo_fix_features(dev, features);
6507 /* driver might be less strict about feature dependencies */
6508 features = netdev_fix_features(dev, features);
6510 /* some features can't be enabled if they're off an an upper device */
6511 netdev_for_each_upper_dev_rcu(dev, upper, iter)
6512 features = netdev_sync_upper_features(dev, upper, features);
6514 if (dev->features == features)
6517 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6518 &dev->features, &features);
6520 if (dev->netdev_ops->ndo_set_features)
6521 err = dev->netdev_ops->ndo_set_features(dev, features);
6525 if (unlikely(err < 0)) {
6527 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6528 err, &features, &dev->features);
6529 /* return non-0 since some features might have changed and
6530 * it's better to fire a spurious notification than miss it
6536 /* some features must be disabled on lower devices when disabled
6537 * on an upper device (think: bonding master or bridge)
6539 netdev_for_each_lower_dev(dev, lower, iter)
6540 netdev_sync_lower_features(dev, lower, features);
6543 dev->features = features;
6545 return err < 0 ? 0 : 1;
6549 * netdev_update_features - recalculate device features
6550 * @dev: the device to check
6552 * Recalculate dev->features set and send notifications if it
6553 * has changed. Should be called after driver or hardware dependent
6554 * conditions might have changed that influence the features.
6556 void netdev_update_features(struct net_device *dev)
6558 if (__netdev_update_features(dev))
6559 netdev_features_change(dev);
6561 EXPORT_SYMBOL(netdev_update_features);
6564 * netdev_change_features - recalculate device features
6565 * @dev: the device to check
6567 * Recalculate dev->features set and send notifications even
6568 * if they have not changed. Should be called instead of
6569 * netdev_update_features() if also dev->vlan_features might
6570 * have changed to allow the changes to be propagated to stacked
6573 void netdev_change_features(struct net_device *dev)
6575 __netdev_update_features(dev);
6576 netdev_features_change(dev);
6578 EXPORT_SYMBOL(netdev_change_features);
6581 * netif_stacked_transfer_operstate - transfer operstate
6582 * @rootdev: the root or lower level device to transfer state from
6583 * @dev: the device to transfer operstate to
6585 * Transfer operational state from root to device. This is normally
6586 * called when a stacking relationship exists between the root
6587 * device and the device(a leaf device).
6589 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6590 struct net_device *dev)
6592 if (rootdev->operstate == IF_OPER_DORMANT)
6593 netif_dormant_on(dev);
6595 netif_dormant_off(dev);
6597 if (netif_carrier_ok(rootdev)) {
6598 if (!netif_carrier_ok(dev))
6599 netif_carrier_on(dev);
6601 if (netif_carrier_ok(dev))
6602 netif_carrier_off(dev);
6605 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6608 static int netif_alloc_rx_queues(struct net_device *dev)
6610 unsigned int i, count = dev->num_rx_queues;
6611 struct netdev_rx_queue *rx;
6612 size_t sz = count * sizeof(*rx);
6616 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6624 for (i = 0; i < count; i++)
6630 static void netdev_init_one_queue(struct net_device *dev,
6631 struct netdev_queue *queue, void *_unused)
6633 /* Initialize queue lock */
6634 spin_lock_init(&queue->_xmit_lock);
6635 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6636 queue->xmit_lock_owner = -1;
6637 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6640 dql_init(&queue->dql, HZ);
6644 static void netif_free_tx_queues(struct net_device *dev)
6649 static int netif_alloc_netdev_queues(struct net_device *dev)
6651 unsigned int count = dev->num_tx_queues;
6652 struct netdev_queue *tx;
6653 size_t sz = count * sizeof(*tx);
6655 if (count < 1 || count > 0xffff)
6658 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6666 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6667 spin_lock_init(&dev->tx_global_lock);
6672 void netif_tx_stop_all_queues(struct net_device *dev)
6676 for (i = 0; i < dev->num_tx_queues; i++) {
6677 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
6678 netif_tx_stop_queue(txq);
6681 EXPORT_SYMBOL(netif_tx_stop_all_queues);
6684 * register_netdevice - register a network device
6685 * @dev: device to register
6687 * Take a completed network device structure and add it to the kernel
6688 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6689 * chain. 0 is returned on success. A negative errno code is returned
6690 * on a failure to set up the device, or if the name is a duplicate.
6692 * Callers must hold the rtnl semaphore. You may want
6693 * register_netdev() instead of this.
6696 * The locking appears insufficient to guarantee two parallel registers
6697 * will not get the same name.
6700 int register_netdevice(struct net_device *dev)
6703 struct net *net = dev_net(dev);
6705 BUG_ON(dev_boot_phase);
6710 /* When net_device's are persistent, this will be fatal. */
6711 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6714 spin_lock_init(&dev->addr_list_lock);
6715 netdev_set_addr_lockdep_class(dev);
6717 ret = dev_get_valid_name(net, dev, dev->name);
6721 /* Init, if this function is available */
6722 if (dev->netdev_ops->ndo_init) {
6723 ret = dev->netdev_ops->ndo_init(dev);
6731 if (((dev->hw_features | dev->features) &
6732 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6733 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6734 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6735 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6742 dev->ifindex = dev_new_index(net);
6743 else if (__dev_get_by_index(net, dev->ifindex))
6746 /* Transfer changeable features to wanted_features and enable
6747 * software offloads (GSO and GRO).
6749 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6750 dev->features |= NETIF_F_SOFT_FEATURES;
6751 dev->wanted_features = dev->features & dev->hw_features;
6753 if (!(dev->flags & IFF_LOOPBACK)) {
6754 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6757 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6759 dev->vlan_features |= NETIF_F_HIGHDMA;
6761 /* Make NETIF_F_SG inheritable to tunnel devices.
6763 dev->hw_enc_features |= NETIF_F_SG;
6765 /* Make NETIF_F_SG inheritable to MPLS.
6767 dev->mpls_features |= NETIF_F_SG;
6769 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6770 ret = notifier_to_errno(ret);
6774 ret = netdev_register_kobject(dev);
6777 dev->reg_state = NETREG_REGISTERED;
6779 __netdev_update_features(dev);
6782 * Default initial state at registry is that the
6783 * device is present.
6786 set_bit(__LINK_STATE_PRESENT, &dev->state);
6788 linkwatch_init_dev(dev);
6790 dev_init_scheduler(dev);
6792 list_netdevice(dev);
6793 add_device_randomness(dev->dev_addr, dev->addr_len);
6795 /* If the device has permanent device address, driver should
6796 * set dev_addr and also addr_assign_type should be set to
6797 * NET_ADDR_PERM (default value).
6799 if (dev->addr_assign_type == NET_ADDR_PERM)
6800 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6802 /* Notify protocols, that a new device appeared. */
6803 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6804 ret = notifier_to_errno(ret);
6806 rollback_registered(dev);
6807 dev->reg_state = NETREG_UNREGISTERED;
6810 * Prevent userspace races by waiting until the network
6811 * device is fully setup before sending notifications.
6813 if (!dev->rtnl_link_ops ||
6814 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6815 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6821 if (dev->netdev_ops->ndo_uninit)
6822 dev->netdev_ops->ndo_uninit(dev);
6825 EXPORT_SYMBOL(register_netdevice);
6828 * init_dummy_netdev - init a dummy network device for NAPI
6829 * @dev: device to init
6831 * This takes a network device structure and initialize the minimum
6832 * amount of fields so it can be used to schedule NAPI polls without
6833 * registering a full blown interface. This is to be used by drivers
6834 * that need to tie several hardware interfaces to a single NAPI
6835 * poll scheduler due to HW limitations.
6837 int init_dummy_netdev(struct net_device *dev)
6839 /* Clear everything. Note we don't initialize spinlocks
6840 * are they aren't supposed to be taken by any of the
6841 * NAPI code and this dummy netdev is supposed to be
6842 * only ever used for NAPI polls
6844 memset(dev, 0, sizeof(struct net_device));
6846 /* make sure we BUG if trying to hit standard
6847 * register/unregister code path
6849 dev->reg_state = NETREG_DUMMY;
6851 /* NAPI wants this */
6852 INIT_LIST_HEAD(&dev->napi_list);
6854 /* a dummy interface is started by default */
6855 set_bit(__LINK_STATE_PRESENT, &dev->state);
6856 set_bit(__LINK_STATE_START, &dev->state);
6858 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6859 * because users of this 'device' dont need to change
6865 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6869 * register_netdev - register a network device
6870 * @dev: device to register
6872 * Take a completed network device structure and add it to the kernel
6873 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6874 * chain. 0 is returned on success. A negative errno code is returned
6875 * on a failure to set up the device, or if the name is a duplicate.
6877 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6878 * and expands the device name if you passed a format string to
6881 int register_netdev(struct net_device *dev)
6886 err = register_netdevice(dev);
6890 EXPORT_SYMBOL(register_netdev);
6892 int netdev_refcnt_read(const struct net_device *dev)
6896 for_each_possible_cpu(i)
6897 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6900 EXPORT_SYMBOL(netdev_refcnt_read);
6903 * netdev_wait_allrefs - wait until all references are gone.
6904 * @dev: target net_device
6906 * This is called when unregistering network devices.
6908 * Any protocol or device that holds a reference should register
6909 * for netdevice notification, and cleanup and put back the
6910 * reference if they receive an UNREGISTER event.
6911 * We can get stuck here if buggy protocols don't correctly
6914 static void netdev_wait_allrefs(struct net_device *dev)
6916 unsigned long rebroadcast_time, warning_time;
6919 linkwatch_forget_dev(dev);
6921 rebroadcast_time = warning_time = jiffies;
6922 refcnt = netdev_refcnt_read(dev);
6924 while (refcnt != 0) {
6925 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6928 /* Rebroadcast unregister notification */
6929 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6935 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6936 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6938 /* We must not have linkwatch events
6939 * pending on unregister. If this
6940 * happens, we simply run the queue
6941 * unscheduled, resulting in a noop
6944 linkwatch_run_queue();
6949 rebroadcast_time = jiffies;
6954 refcnt = netdev_refcnt_read(dev);
6956 if (time_after(jiffies, warning_time + 10 * HZ)) {
6957 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6959 warning_time = jiffies;
6968 * register_netdevice(x1);
6969 * register_netdevice(x2);
6971 * unregister_netdevice(y1);
6972 * unregister_netdevice(y2);
6978 * We are invoked by rtnl_unlock().
6979 * This allows us to deal with problems:
6980 * 1) We can delete sysfs objects which invoke hotplug
6981 * without deadlocking with linkwatch via keventd.
6982 * 2) Since we run with the RTNL semaphore not held, we can sleep
6983 * safely in order to wait for the netdev refcnt to drop to zero.
6985 * We must not return until all unregister events added during
6986 * the interval the lock was held have been completed.
6988 void netdev_run_todo(void)
6990 struct list_head list;
6992 /* Snapshot list, allow later requests */
6993 list_replace_init(&net_todo_list, &list);
6998 /* Wait for rcu callbacks to finish before next phase */
6999 if (!list_empty(&list))
7002 while (!list_empty(&list)) {
7003 struct net_device *dev
7004 = list_first_entry(&list, struct net_device, todo_list);
7005 list_del(&dev->todo_list);
7008 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7011 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
7012 pr_err("network todo '%s' but state %d\n",
7013 dev->name, dev->reg_state);
7018 dev->reg_state = NETREG_UNREGISTERED;
7020 netdev_wait_allrefs(dev);
7023 BUG_ON(netdev_refcnt_read(dev));
7024 BUG_ON(!list_empty(&dev->ptype_all));
7025 BUG_ON(!list_empty(&dev->ptype_specific));
7026 WARN_ON(rcu_access_pointer(dev->ip_ptr));
7027 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
7028 WARN_ON(dev->dn_ptr);
7030 if (dev->destructor)
7031 dev->destructor(dev);
7033 /* Report a network device has been unregistered */
7035 dev_net(dev)->dev_unreg_count--;
7037 wake_up(&netdev_unregistering_wq);
7039 /* Free network device */
7040 kobject_put(&dev->dev.kobj);
7044 /* Convert net_device_stats to rtnl_link_stats64. They have the same
7045 * fields in the same order, with only the type differing.
7047 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7048 const struct net_device_stats *netdev_stats)
7050 #if BITS_PER_LONG == 64
7051 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
7052 memcpy(stats64, netdev_stats, sizeof(*stats64));
7054 size_t i, n = sizeof(*stats64) / sizeof(u64);
7055 const unsigned long *src = (const unsigned long *)netdev_stats;
7056 u64 *dst = (u64 *)stats64;
7058 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
7059 sizeof(*stats64) / sizeof(u64));
7060 for (i = 0; i < n; i++)
7064 EXPORT_SYMBOL(netdev_stats_to_stats64);
7067 * dev_get_stats - get network device statistics
7068 * @dev: device to get statistics from
7069 * @storage: place to store stats
7071 * Get network statistics from device. Return @storage.
7072 * The device driver may provide its own method by setting
7073 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7074 * otherwise the internal statistics structure is used.
7076 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
7077 struct rtnl_link_stats64 *storage)
7079 const struct net_device_ops *ops = dev->netdev_ops;
7081 if (ops->ndo_get_stats64) {
7082 memset(storage, 0, sizeof(*storage));
7083 ops->ndo_get_stats64(dev, storage);
7084 } else if (ops->ndo_get_stats) {
7085 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7087 netdev_stats_to_stats64(storage, &dev->stats);
7089 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
7090 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
7093 EXPORT_SYMBOL(dev_get_stats);
7095 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7097 struct netdev_queue *queue = dev_ingress_queue(dev);
7099 #ifdef CONFIG_NET_CLS_ACT
7102 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7105 netdev_init_one_queue(dev, queue, NULL);
7106 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7107 queue->qdisc_sleeping = &noop_qdisc;
7108 rcu_assign_pointer(dev->ingress_queue, queue);
7113 static const struct ethtool_ops default_ethtool_ops;
7115 void netdev_set_default_ethtool_ops(struct net_device *dev,
7116 const struct ethtool_ops *ops)
7118 if (dev->ethtool_ops == &default_ethtool_ops)
7119 dev->ethtool_ops = ops;
7121 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7123 void netdev_freemem(struct net_device *dev)
7125 char *addr = (char *)dev - dev->padded;
7131 * alloc_netdev_mqs - allocate network device
7132 * @sizeof_priv: size of private data to allocate space for
7133 * @name: device name format string
7134 * @name_assign_type: origin of device name
7135 * @setup: callback to initialize device
7136 * @txqs: the number of TX subqueues to allocate
7137 * @rxqs: the number of RX subqueues to allocate
7139 * Allocates a struct net_device with private data area for driver use
7140 * and performs basic initialization. Also allocates subqueue structs
7141 * for each queue on the device.
7143 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7144 unsigned char name_assign_type,
7145 void (*setup)(struct net_device *),
7146 unsigned int txqs, unsigned int rxqs)
7148 struct net_device *dev;
7150 struct net_device *p;
7152 BUG_ON(strlen(name) >= sizeof(dev->name));
7155 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7161 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7166 alloc_size = sizeof(struct net_device);
7168 /* ensure 32-byte alignment of private area */
7169 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7170 alloc_size += sizeof_priv;
7172 /* ensure 32-byte alignment of whole construct */
7173 alloc_size += NETDEV_ALIGN - 1;
7175 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7177 p = vzalloc(alloc_size);
7181 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7182 dev->padded = (char *)dev - (char *)p;
7184 dev->pcpu_refcnt = alloc_percpu(int);
7185 if (!dev->pcpu_refcnt)
7188 if (dev_addr_init(dev))
7194 dev_net_set(dev, &init_net);
7196 dev->gso_max_size = GSO_MAX_SIZE;
7197 dev->gso_max_segs = GSO_MAX_SEGS;
7198 dev->gso_min_segs = 0;
7200 INIT_LIST_HEAD(&dev->napi_list);
7201 INIT_LIST_HEAD(&dev->unreg_list);
7202 INIT_LIST_HEAD(&dev->close_list);
7203 INIT_LIST_HEAD(&dev->link_watch_list);
7204 INIT_LIST_HEAD(&dev->adj_list.upper);
7205 INIT_LIST_HEAD(&dev->adj_list.lower);
7206 INIT_LIST_HEAD(&dev->all_adj_list.upper);
7207 INIT_LIST_HEAD(&dev->all_adj_list.lower);
7208 INIT_LIST_HEAD(&dev->ptype_all);
7209 INIT_LIST_HEAD(&dev->ptype_specific);
7210 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7213 if (!dev->tx_queue_len) {
7214 dev->priv_flags |= IFF_NO_QUEUE;
7215 dev->tx_queue_len = 1;
7218 dev->num_tx_queues = txqs;
7219 dev->real_num_tx_queues = txqs;
7220 if (netif_alloc_netdev_queues(dev))
7224 dev->num_rx_queues = rxqs;
7225 dev->real_num_rx_queues = rxqs;
7226 if (netif_alloc_rx_queues(dev))
7230 strcpy(dev->name, name);
7231 dev->name_assign_type = name_assign_type;
7232 dev->group = INIT_NETDEV_GROUP;
7233 if (!dev->ethtool_ops)
7234 dev->ethtool_ops = &default_ethtool_ops;
7236 nf_hook_ingress_init(dev);
7245 free_percpu(dev->pcpu_refcnt);
7247 netdev_freemem(dev);
7250 EXPORT_SYMBOL(alloc_netdev_mqs);
7253 * free_netdev - free network device
7256 * This function does the last stage of destroying an allocated device
7257 * interface. The reference to the device object is released.
7258 * If this is the last reference then it will be freed.
7260 void free_netdev(struct net_device *dev)
7262 struct napi_struct *p, *n;
7264 netif_free_tx_queues(dev);
7269 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7271 /* Flush device addresses */
7272 dev_addr_flush(dev);
7274 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7277 free_percpu(dev->pcpu_refcnt);
7278 dev->pcpu_refcnt = NULL;
7280 /* Compatibility with error handling in drivers */
7281 if (dev->reg_state == NETREG_UNINITIALIZED) {
7282 netdev_freemem(dev);
7286 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7287 dev->reg_state = NETREG_RELEASED;
7289 /* will free via device release */
7290 put_device(&dev->dev);
7292 EXPORT_SYMBOL(free_netdev);
7295 * synchronize_net - Synchronize with packet receive processing
7297 * Wait for packets currently being received to be done.
7298 * Does not block later packets from starting.
7300 void synchronize_net(void)
7303 if (rtnl_is_locked() && !IS_ENABLED(CONFIG_PREEMPT_RT_FULL))
7304 synchronize_rcu_expedited();
7308 EXPORT_SYMBOL(synchronize_net);
7311 * unregister_netdevice_queue - remove device from the kernel
7315 * This function shuts down a device interface and removes it
7316 * from the kernel tables.
7317 * If head not NULL, device is queued to be unregistered later.
7319 * Callers must hold the rtnl semaphore. You may want
7320 * unregister_netdev() instead of this.
7323 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7328 list_move_tail(&dev->unreg_list, head);
7330 rollback_registered(dev);
7331 /* Finish processing unregister after unlock */
7335 EXPORT_SYMBOL(unregister_netdevice_queue);
7338 * unregister_netdevice_many - unregister many devices
7339 * @head: list of devices
7341 * Note: As most callers use a stack allocated list_head,
7342 * we force a list_del() to make sure stack wont be corrupted later.
7344 void unregister_netdevice_many(struct list_head *head)
7346 struct net_device *dev;
7348 if (!list_empty(head)) {
7349 rollback_registered_many(head);
7350 list_for_each_entry(dev, head, unreg_list)
7355 EXPORT_SYMBOL(unregister_netdevice_many);
7358 * unregister_netdev - remove device from the kernel
7361 * This function shuts down a device interface and removes it
7362 * from the kernel tables.
7364 * This is just a wrapper for unregister_netdevice that takes
7365 * the rtnl semaphore. In general you want to use this and not
7366 * unregister_netdevice.
7368 void unregister_netdev(struct net_device *dev)
7371 unregister_netdevice(dev);
7374 EXPORT_SYMBOL(unregister_netdev);
7377 * dev_change_net_namespace - move device to different nethost namespace
7379 * @net: network namespace
7380 * @pat: If not NULL name pattern to try if the current device name
7381 * is already taken in the destination network namespace.
7383 * This function shuts down a device interface and moves it
7384 * to a new network namespace. On success 0 is returned, on
7385 * a failure a netagive errno code is returned.
7387 * Callers must hold the rtnl semaphore.
7390 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7396 /* Don't allow namespace local devices to be moved. */
7398 if (dev->features & NETIF_F_NETNS_LOCAL)
7401 /* Ensure the device has been registrered */
7402 if (dev->reg_state != NETREG_REGISTERED)
7405 /* Get out if there is nothing todo */
7407 if (net_eq(dev_net(dev), net))
7410 /* Pick the destination device name, and ensure
7411 * we can use it in the destination network namespace.
7414 if (__dev_get_by_name(net, dev->name)) {
7415 /* We get here if we can't use the current device name */
7418 if (dev_get_valid_name(net, dev, pat) < 0)
7423 * And now a mini version of register_netdevice unregister_netdevice.
7426 /* If device is running close it first. */
7429 /* And unlink it from device chain */
7431 unlist_netdevice(dev);
7435 /* Shutdown queueing discipline. */
7438 /* Notify protocols, that we are about to destroy
7439 this device. They should clean all the things.
7441 Note that dev->reg_state stays at NETREG_REGISTERED.
7442 This is wanted because this way 8021q and macvlan know
7443 the device is just moving and can keep their slaves up.
7445 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7447 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7448 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7451 * Flush the unicast and multicast chains
7456 /* Send a netdev-removed uevent to the old namespace */
7457 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7458 netdev_adjacent_del_links(dev);
7460 /* Actually switch the network namespace */
7461 dev_net_set(dev, net);
7463 /* If there is an ifindex conflict assign a new one */
7464 if (__dev_get_by_index(net, dev->ifindex))
7465 dev->ifindex = dev_new_index(net);
7467 /* Send a netdev-add uevent to the new namespace */
7468 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7469 netdev_adjacent_add_links(dev);
7471 /* Fixup kobjects */
7472 err = device_rename(&dev->dev, dev->name);
7475 /* Add the device back in the hashes */
7476 list_netdevice(dev);
7478 /* Notify protocols, that a new device appeared. */
7479 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7482 * Prevent userspace races by waiting until the network
7483 * device is fully setup before sending notifications.
7485 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7492 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7494 static int dev_cpu_callback(struct notifier_block *nfb,
7495 unsigned long action,
7498 struct sk_buff **list_skb;
7499 struct sk_buff *skb;
7500 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7501 struct softnet_data *sd, *oldsd;
7503 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7506 local_irq_disable();
7507 cpu = smp_processor_id();
7508 sd = &per_cpu(softnet_data, cpu);
7509 oldsd = &per_cpu(softnet_data, oldcpu);
7511 /* Find end of our completion_queue. */
7512 list_skb = &sd->completion_queue;
7514 list_skb = &(*list_skb)->next;
7515 /* Append completion queue from offline CPU. */
7516 *list_skb = oldsd->completion_queue;
7517 oldsd->completion_queue = NULL;
7519 /* Append output queue from offline CPU. */
7520 if (oldsd->output_queue) {
7521 *sd->output_queue_tailp = oldsd->output_queue;
7522 sd->output_queue_tailp = oldsd->output_queue_tailp;
7523 oldsd->output_queue = NULL;
7524 oldsd->output_queue_tailp = &oldsd->output_queue;
7526 /* Append NAPI poll list from offline CPU, with one exception :
7527 * process_backlog() must be called by cpu owning percpu backlog.
7528 * We properly handle process_queue & input_pkt_queue later.
7530 while (!list_empty(&oldsd->poll_list)) {
7531 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7535 list_del_init(&napi->poll_list);
7536 if (napi->poll == process_backlog)
7539 ____napi_schedule(sd, napi);
7542 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7544 preempt_check_resched_rt();
7546 /* Process offline CPU's input_pkt_queue */
7547 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7549 input_queue_head_incr(oldsd);
7551 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
7553 input_queue_head_incr(oldsd);
7555 while ((skb = __skb_dequeue(&oldsd->tofree_queue))) {
7564 * netdev_increment_features - increment feature set by one
7565 * @all: current feature set
7566 * @one: new feature set
7567 * @mask: mask feature set
7569 * Computes a new feature set after adding a device with feature set
7570 * @one to the master device with current feature set @all. Will not
7571 * enable anything that is off in @mask. Returns the new feature set.
7573 netdev_features_t netdev_increment_features(netdev_features_t all,
7574 netdev_features_t one, netdev_features_t mask)
7576 if (mask & NETIF_F_GEN_CSUM)
7577 mask |= NETIF_F_ALL_CSUM;
7578 mask |= NETIF_F_VLAN_CHALLENGED;
7580 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7581 all &= one | ~NETIF_F_ALL_FOR_ALL;
7583 /* If one device supports hw checksumming, set for all. */
7584 if (all & NETIF_F_GEN_CSUM)
7585 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7589 EXPORT_SYMBOL(netdev_increment_features);
7591 static struct hlist_head * __net_init netdev_create_hash(void)
7594 struct hlist_head *hash;
7596 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7598 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7599 INIT_HLIST_HEAD(&hash[i]);
7604 /* Initialize per network namespace state */
7605 static int __net_init netdev_init(struct net *net)
7607 if (net != &init_net)
7608 INIT_LIST_HEAD(&net->dev_base_head);
7610 net->dev_name_head = netdev_create_hash();
7611 if (net->dev_name_head == NULL)
7614 net->dev_index_head = netdev_create_hash();
7615 if (net->dev_index_head == NULL)
7621 kfree(net->dev_name_head);
7627 * netdev_drivername - network driver for the device
7628 * @dev: network device
7630 * Determine network driver for device.
7632 const char *netdev_drivername(const struct net_device *dev)
7634 const struct device_driver *driver;
7635 const struct device *parent;
7636 const char *empty = "";
7638 parent = dev->dev.parent;
7642 driver = parent->driver;
7643 if (driver && driver->name)
7644 return driver->name;
7648 static void __netdev_printk(const char *level, const struct net_device *dev,
7649 struct va_format *vaf)
7651 if (dev && dev->dev.parent) {
7652 dev_printk_emit(level[1] - '0',
7655 dev_driver_string(dev->dev.parent),
7656 dev_name(dev->dev.parent),
7657 netdev_name(dev), netdev_reg_state(dev),
7660 printk("%s%s%s: %pV",
7661 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7663 printk("%s(NULL net_device): %pV", level, vaf);
7667 void netdev_printk(const char *level, const struct net_device *dev,
7668 const char *format, ...)
7670 struct va_format vaf;
7673 va_start(args, format);
7678 __netdev_printk(level, dev, &vaf);
7682 EXPORT_SYMBOL(netdev_printk);
7684 #define define_netdev_printk_level(func, level) \
7685 void func(const struct net_device *dev, const char *fmt, ...) \
7687 struct va_format vaf; \
7690 va_start(args, fmt); \
7695 __netdev_printk(level, dev, &vaf); \
7699 EXPORT_SYMBOL(func);
7701 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7702 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7703 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7704 define_netdev_printk_level(netdev_err, KERN_ERR);
7705 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7706 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7707 define_netdev_printk_level(netdev_info, KERN_INFO);
7709 static void __net_exit netdev_exit(struct net *net)
7711 kfree(net->dev_name_head);
7712 kfree(net->dev_index_head);
7715 static struct pernet_operations __net_initdata netdev_net_ops = {
7716 .init = netdev_init,
7717 .exit = netdev_exit,
7720 static void __net_exit default_device_exit(struct net *net)
7722 struct net_device *dev, *aux;
7724 * Push all migratable network devices back to the
7725 * initial network namespace
7728 for_each_netdev_safe(net, dev, aux) {
7730 char fb_name[IFNAMSIZ];
7732 /* Ignore unmoveable devices (i.e. loopback) */
7733 if (dev->features & NETIF_F_NETNS_LOCAL)
7736 /* Leave virtual devices for the generic cleanup */
7737 if (dev->rtnl_link_ops)
7740 /* Push remaining network devices to init_net */
7741 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7742 err = dev_change_net_namespace(dev, &init_net, fb_name);
7744 pr_emerg("%s: failed to move %s to init_net: %d\n",
7745 __func__, dev->name, err);
7752 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7754 /* Return with the rtnl_lock held when there are no network
7755 * devices unregistering in any network namespace in net_list.
7759 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7761 add_wait_queue(&netdev_unregistering_wq, &wait);
7763 unregistering = false;
7765 list_for_each_entry(net, net_list, exit_list) {
7766 if (net->dev_unreg_count > 0) {
7767 unregistering = true;
7775 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7777 remove_wait_queue(&netdev_unregistering_wq, &wait);
7780 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7782 /* At exit all network devices most be removed from a network
7783 * namespace. Do this in the reverse order of registration.
7784 * Do this across as many network namespaces as possible to
7785 * improve batching efficiency.
7787 struct net_device *dev;
7789 LIST_HEAD(dev_kill_list);
7791 /* To prevent network device cleanup code from dereferencing
7792 * loopback devices or network devices that have been freed
7793 * wait here for all pending unregistrations to complete,
7794 * before unregistring the loopback device and allowing the
7795 * network namespace be freed.
7797 * The netdev todo list containing all network devices
7798 * unregistrations that happen in default_device_exit_batch
7799 * will run in the rtnl_unlock() at the end of
7800 * default_device_exit_batch.
7802 rtnl_lock_unregistering(net_list);
7803 list_for_each_entry(net, net_list, exit_list) {
7804 for_each_netdev_reverse(net, dev) {
7805 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7806 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7808 unregister_netdevice_queue(dev, &dev_kill_list);
7811 unregister_netdevice_many(&dev_kill_list);
7815 static struct pernet_operations __net_initdata default_device_ops = {
7816 .exit = default_device_exit,
7817 .exit_batch = default_device_exit_batch,
7821 * Initialize the DEV module. At boot time this walks the device list and
7822 * unhooks any devices that fail to initialise (normally hardware not
7823 * present) and leaves us with a valid list of present and active devices.
7828 * This is called single threaded during boot, so no need
7829 * to take the rtnl semaphore.
7831 static int __init net_dev_init(void)
7833 int i, rc = -ENOMEM;
7835 BUG_ON(!dev_boot_phase);
7837 if (dev_proc_init())
7840 if (netdev_kobject_init())
7843 INIT_LIST_HEAD(&ptype_all);
7844 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7845 INIT_LIST_HEAD(&ptype_base[i]);
7847 INIT_LIST_HEAD(&offload_base);
7849 if (register_pernet_subsys(&netdev_net_ops))
7853 * Initialise the packet receive queues.
7856 for_each_possible_cpu(i) {
7857 struct softnet_data *sd = &per_cpu(softnet_data, i);
7859 skb_queue_head_init_raw(&sd->input_pkt_queue);
7860 skb_queue_head_init_raw(&sd->process_queue);
7861 skb_queue_head_init_raw(&sd->tofree_queue);
7862 INIT_LIST_HEAD(&sd->poll_list);
7863 sd->output_queue_tailp = &sd->output_queue;
7865 sd->csd.func = rps_trigger_softirq;
7870 sd->backlog.poll = process_backlog;
7871 sd->backlog.weight = weight_p;
7876 /* The loopback device is special if any other network devices
7877 * is present in a network namespace the loopback device must
7878 * be present. Since we now dynamically allocate and free the
7879 * loopback device ensure this invariant is maintained by
7880 * keeping the loopback device as the first device on the
7881 * list of network devices. Ensuring the loopback devices
7882 * is the first device that appears and the last network device
7885 if (register_pernet_device(&loopback_net_ops))
7888 if (register_pernet_device(&default_device_ops))
7891 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7892 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7894 hotcpu_notifier(dev_cpu_callback, 0);
7901 subsys_initcall(net_dev_init);