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 <net/busy_poll.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/stat.h>
103 #include <net/dst_metadata.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/module.h>
110 #include <linux/netpoll.h>
111 #include <linux/rcupdate.h>
112 #include <linux/delay.h>
113 #include <net/iw_handler.h>
114 #include <asm/current.h>
115 #include <linux/audit.h>
116 #include <linux/dmaengine.h>
117 #include <linux/err.h>
118 #include <linux/ctype.h>
119 #include <linux/if_arp.h>
120 #include <linux/if_vlan.h>
121 #include <linux/ip.h>
123 #include <net/mpls.h>
124 #include <linux/ipv6.h>
125 #include <linux/in.h>
126 #include <linux/jhash.h>
127 #include <linux/random.h>
128 #include <trace/events/napi.h>
129 #include <trace/events/net.h>
130 #include <trace/events/skb.h>
131 #include <linux/pci.h>
132 #include <linux/inetdevice.h>
133 #include <linux/cpu_rmap.h>
134 #include <linux/static_key.h>
135 #include <linux/hashtable.h>
136 #include <linux/vmalloc.h>
137 #include <linux/if_macvlan.h>
138 #include <linux/errqueue.h>
139 #include <linux/hrtimer.h>
140 #include <linux/netfilter_ingress.h>
142 #include "net-sysfs.h"
144 /* Instead of increasing this, you should create a hash table. */
145 #define MAX_GRO_SKBS 8
147 /* This should be increased if a protocol with a bigger head is added. */
148 #define GRO_MAX_HEAD (MAX_HEADER + 128)
150 static DEFINE_SPINLOCK(ptype_lock);
151 static DEFINE_SPINLOCK(offload_lock);
152 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
153 struct list_head ptype_all __read_mostly; /* Taps */
154 static struct list_head offload_base __read_mostly;
156 static int netif_rx_internal(struct sk_buff *skb);
157 static int call_netdevice_notifiers_info(unsigned long val,
158 struct net_device *dev,
159 struct netdev_notifier_info *info);
162 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
165 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
167 * Writers must hold the rtnl semaphore while they loop through the
168 * dev_base_head list, and hold dev_base_lock for writing when they do the
169 * actual updates. This allows pure readers to access the list even
170 * while a writer is preparing to update it.
172 * To put it another way, dev_base_lock is held for writing only to
173 * protect against pure readers; the rtnl semaphore provides the
174 * protection against other writers.
176 * See, for example usages, register_netdevice() and
177 * unregister_netdevice(), which must be called with the rtnl
180 DEFINE_RWLOCK(dev_base_lock);
181 EXPORT_SYMBOL(dev_base_lock);
183 /* protects napi_hash addition/deletion and napi_gen_id */
184 static DEFINE_SPINLOCK(napi_hash_lock);
186 static unsigned int napi_gen_id = NR_CPUS;
187 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
189 static seqcount_t devnet_rename_seq;
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 spin_lock(&sd->input_pkt_queue.lock);
215 static inline void rps_unlock(struct softnet_data *sd)
218 spin_unlock(&sd->input_pkt_queue.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)) {
896 * dev_getbyhwaddr_rcu - find a device by its hardware address
897 * @net: the applicable net namespace
898 * @type: media type of device
899 * @ha: hardware address
901 * Search for an interface by MAC address. Returns NULL if the device
902 * is not found or a pointer to the device.
903 * The caller must hold RCU or RTNL.
904 * The returned device has not had its ref count increased
905 * and the caller must therefore be careful about locking
909 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
912 struct net_device *dev;
914 for_each_netdev_rcu(net, dev)
915 if (dev->type == type &&
916 !memcmp(dev->dev_addr, ha, dev->addr_len))
921 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
923 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
925 struct net_device *dev;
928 for_each_netdev(net, dev)
929 if (dev->type == type)
934 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
936 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
938 struct net_device *dev, *ret = NULL;
941 for_each_netdev_rcu(net, dev)
942 if (dev->type == type) {
950 EXPORT_SYMBOL(dev_getfirstbyhwtype);
953 * __dev_get_by_flags - find any device with given flags
954 * @net: the applicable net namespace
955 * @if_flags: IFF_* values
956 * @mask: bitmask of bits in if_flags to check
958 * Search for any interface with the given flags. Returns NULL if a device
959 * is not found or a pointer to the device. Must be called inside
960 * rtnl_lock(), and result refcount is unchanged.
963 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
966 struct net_device *dev, *ret;
971 for_each_netdev(net, dev) {
972 if (((dev->flags ^ if_flags) & mask) == 0) {
979 EXPORT_SYMBOL(__dev_get_by_flags);
982 * dev_valid_name - check if name is okay for network device
985 * Network device names need to be valid file names to
986 * to allow sysfs to work. We also disallow any kind of
989 bool dev_valid_name(const char *name)
993 if (strlen(name) >= IFNAMSIZ)
995 if (!strcmp(name, ".") || !strcmp(name, ".."))
999 if (*name == '/' || *name == ':' || isspace(*name))
1005 EXPORT_SYMBOL(dev_valid_name);
1008 * __dev_alloc_name - allocate a name for a device
1009 * @net: network namespace to allocate the device name in
1010 * @name: name format string
1011 * @buf: scratch buffer and result name string
1013 * Passed a format string - eg "lt%d" it will try and find a suitable
1014 * id. It scans list of devices to build up a free map, then chooses
1015 * the first empty slot. The caller must hold the dev_base or rtnl lock
1016 * while allocating the name and adding the device in order to avoid
1018 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1019 * Returns the number of the unit assigned or a negative errno code.
1022 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1026 const int max_netdevices = 8*PAGE_SIZE;
1027 unsigned long *inuse;
1028 struct net_device *d;
1030 p = strnchr(name, IFNAMSIZ-1, '%');
1033 * Verify the string as this thing may have come from
1034 * the user. There must be either one "%d" and no other "%"
1037 if (p[1] != 'd' || strchr(p + 2, '%'))
1040 /* Use one page as a bit array of possible slots */
1041 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1045 for_each_netdev(net, d) {
1046 if (!sscanf(d->name, name, &i))
1048 if (i < 0 || i >= max_netdevices)
1051 /* avoid cases where sscanf is not exact inverse of printf */
1052 snprintf(buf, IFNAMSIZ, name, i);
1053 if (!strncmp(buf, d->name, IFNAMSIZ))
1057 i = find_first_zero_bit(inuse, max_netdevices);
1058 free_page((unsigned long) inuse);
1062 snprintf(buf, IFNAMSIZ, name, i);
1063 if (!__dev_get_by_name(net, buf))
1066 /* It is possible to run out of possible slots
1067 * when the name is long and there isn't enough space left
1068 * for the digits, or if all bits are used.
1074 * dev_alloc_name - allocate a name for a device
1076 * @name: name format string
1078 * Passed a format string - eg "lt%d" it will try and find a suitable
1079 * id. It scans list of devices to build up a free map, then chooses
1080 * the first empty slot. The caller must hold the dev_base or rtnl lock
1081 * while allocating the name and adding the device in order to avoid
1083 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1084 * Returns the number of the unit assigned or a negative errno code.
1087 int dev_alloc_name(struct net_device *dev, const char *name)
1093 BUG_ON(!dev_net(dev));
1095 ret = __dev_alloc_name(net, name, buf);
1097 strlcpy(dev->name, buf, IFNAMSIZ);
1100 EXPORT_SYMBOL(dev_alloc_name);
1102 static int dev_alloc_name_ns(struct net *net,
1103 struct net_device *dev,
1109 ret = __dev_alloc_name(net, name, buf);
1111 strlcpy(dev->name, buf, IFNAMSIZ);
1115 static int dev_get_valid_name(struct net *net,
1116 struct net_device *dev,
1121 if (!dev_valid_name(name))
1124 if (strchr(name, '%'))
1125 return dev_alloc_name_ns(net, dev, name);
1126 else if (__dev_get_by_name(net, name))
1128 else if (dev->name != name)
1129 strlcpy(dev->name, name, IFNAMSIZ);
1135 * dev_change_name - change name of a device
1137 * @newname: name (or format string) must be at least IFNAMSIZ
1139 * Change name of a device, can pass format strings "eth%d".
1142 int dev_change_name(struct net_device *dev, const char *newname)
1144 unsigned char old_assign_type;
1145 char oldname[IFNAMSIZ];
1151 BUG_ON(!dev_net(dev));
1154 if (dev->flags & IFF_UP)
1157 write_seqcount_begin(&devnet_rename_seq);
1159 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1160 write_seqcount_end(&devnet_rename_seq);
1164 memcpy(oldname, dev->name, IFNAMSIZ);
1166 err = dev_get_valid_name(net, dev, newname);
1168 write_seqcount_end(&devnet_rename_seq);
1172 if (oldname[0] && !strchr(oldname, '%'))
1173 netdev_info(dev, "renamed from %s\n", oldname);
1175 old_assign_type = dev->name_assign_type;
1176 dev->name_assign_type = NET_NAME_RENAMED;
1179 ret = device_rename(&dev->dev, dev->name);
1181 memcpy(dev->name, oldname, IFNAMSIZ);
1182 dev->name_assign_type = old_assign_type;
1183 write_seqcount_end(&devnet_rename_seq);
1187 write_seqcount_end(&devnet_rename_seq);
1189 netdev_adjacent_rename_links(dev, oldname);
1191 write_lock_bh(&dev_base_lock);
1192 hlist_del_rcu(&dev->name_hlist);
1193 write_unlock_bh(&dev_base_lock);
1197 write_lock_bh(&dev_base_lock);
1198 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1199 write_unlock_bh(&dev_base_lock);
1201 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1202 ret = notifier_to_errno(ret);
1205 /* err >= 0 after dev_alloc_name() or stores the first errno */
1208 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",
1224 * dev_set_alias - change ifalias of a device
1226 * @alias: name up to IFALIASZ
1227 * @len: limit of bytes to copy from info
1229 * Set ifalias for a device,
1231 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1237 if (len >= IFALIASZ)
1241 kfree(dev->ifalias);
1242 dev->ifalias = NULL;
1246 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1249 dev->ifalias = new_ifalias;
1251 strlcpy(dev->ifalias, alias, len+1);
1257 * netdev_features_change - device changes features
1258 * @dev: device to cause notification
1260 * Called to indicate a device has changed features.
1262 void netdev_features_change(struct net_device *dev)
1264 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1266 EXPORT_SYMBOL(netdev_features_change);
1269 * netdev_state_change - device changes state
1270 * @dev: device to cause notification
1272 * Called to indicate a device has changed state. This function calls
1273 * the notifier chains for netdev_chain and sends a NEWLINK message
1274 * to the routing socket.
1276 void netdev_state_change(struct net_device *dev)
1278 if (dev->flags & IFF_UP) {
1279 struct netdev_notifier_change_info change_info;
1281 change_info.flags_changed = 0;
1282 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1284 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1287 EXPORT_SYMBOL(netdev_state_change);
1290 * netdev_notify_peers - notify network peers about existence of @dev
1291 * @dev: network device
1293 * Generate traffic such that interested network peers are aware of
1294 * @dev, such as by generating a gratuitous ARP. This may be used when
1295 * a device wants to inform the rest of the network about some sort of
1296 * reconfiguration such as a failover event or virtual machine
1299 void netdev_notify_peers(struct net_device *dev)
1302 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1305 EXPORT_SYMBOL(netdev_notify_peers);
1307 static int __dev_open(struct net_device *dev)
1309 const struct net_device_ops *ops = dev->netdev_ops;
1314 if (!netif_device_present(dev))
1317 /* Block netpoll from trying to do any rx path servicing.
1318 * If we don't do this there is a chance ndo_poll_controller
1319 * or ndo_poll may be running while we open the device
1321 netpoll_poll_disable(dev);
1323 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1324 ret = notifier_to_errno(ret);
1328 set_bit(__LINK_STATE_START, &dev->state);
1330 if (ops->ndo_validate_addr)
1331 ret = ops->ndo_validate_addr(dev);
1333 if (!ret && ops->ndo_open)
1334 ret = ops->ndo_open(dev);
1336 netpoll_poll_enable(dev);
1339 clear_bit(__LINK_STATE_START, &dev->state);
1341 dev->flags |= IFF_UP;
1342 dev_set_rx_mode(dev);
1344 add_device_randomness(dev->dev_addr, dev->addr_len);
1351 * dev_open - prepare an interface for use.
1352 * @dev: device to open
1354 * Takes a device from down to up state. The device's private open
1355 * function is invoked and then the multicast lists are loaded. Finally
1356 * the device is moved into the up state and a %NETDEV_UP message is
1357 * sent to the netdev notifier chain.
1359 * Calling this function on an active interface is a nop. On a failure
1360 * a negative errno code is returned.
1362 int dev_open(struct net_device *dev)
1366 if (dev->flags & IFF_UP)
1369 ret = __dev_open(dev);
1373 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1374 call_netdevice_notifiers(NETDEV_UP, dev);
1378 EXPORT_SYMBOL(dev_open);
1380 static int __dev_close_many(struct list_head *head)
1382 struct net_device *dev;
1387 list_for_each_entry(dev, head, close_list) {
1388 /* Temporarily disable netpoll until the interface is down */
1389 netpoll_poll_disable(dev);
1391 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1393 clear_bit(__LINK_STATE_START, &dev->state);
1395 /* Synchronize to scheduled poll. We cannot touch poll list, it
1396 * can be even on different cpu. So just clear netif_running().
1398 * dev->stop() will invoke napi_disable() on all of it's
1399 * napi_struct instances on this device.
1401 smp_mb__after_atomic(); /* Commit netif_running(). */
1404 dev_deactivate_many(head);
1406 list_for_each_entry(dev, head, close_list) {
1407 const struct net_device_ops *ops = dev->netdev_ops;
1410 * Call the device specific close. This cannot fail.
1411 * Only if device is UP
1413 * We allow it to be called even after a DETACH hot-plug
1419 dev->flags &= ~IFF_UP;
1420 netpoll_poll_enable(dev);
1426 static int __dev_close(struct net_device *dev)
1431 list_add(&dev->close_list, &single);
1432 retval = __dev_close_many(&single);
1438 int dev_close_many(struct list_head *head, bool unlink)
1440 struct net_device *dev, *tmp;
1442 /* Remove the devices that don't need to be closed */
1443 list_for_each_entry_safe(dev, tmp, head, close_list)
1444 if (!(dev->flags & IFF_UP))
1445 list_del_init(&dev->close_list);
1447 __dev_close_many(head);
1449 list_for_each_entry_safe(dev, tmp, head, close_list) {
1450 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1451 call_netdevice_notifiers(NETDEV_DOWN, dev);
1453 list_del_init(&dev->close_list);
1458 EXPORT_SYMBOL(dev_close_many);
1461 * dev_close - shutdown an interface.
1462 * @dev: device to shutdown
1464 * This function moves an active device into down state. A
1465 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1466 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1469 int dev_close(struct net_device *dev)
1471 if (dev->flags & IFF_UP) {
1474 list_add(&dev->close_list, &single);
1475 dev_close_many(&single, true);
1480 EXPORT_SYMBOL(dev_close);
1484 * dev_disable_lro - disable Large Receive Offload on a device
1487 * Disable Large Receive Offload (LRO) on a net device. Must be
1488 * called under RTNL. This is needed if received packets may be
1489 * forwarded to another interface.
1491 void dev_disable_lro(struct net_device *dev)
1493 struct net_device *lower_dev;
1494 struct list_head *iter;
1496 dev->wanted_features &= ~NETIF_F_LRO;
1497 netdev_update_features(dev);
1499 if (unlikely(dev->features & NETIF_F_LRO))
1500 netdev_WARN(dev, "failed to disable LRO!\n");
1502 netdev_for_each_lower_dev(dev, lower_dev, iter)
1503 dev_disable_lro(lower_dev);
1505 EXPORT_SYMBOL(dev_disable_lro);
1507 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1508 struct net_device *dev)
1510 struct netdev_notifier_info info;
1512 netdev_notifier_info_init(&info, dev);
1513 return nb->notifier_call(nb, val, &info);
1516 static int dev_boot_phase = 1;
1519 * register_netdevice_notifier - register a network notifier block
1522 * Register a notifier to be called when network device events occur.
1523 * The notifier passed is linked into the kernel structures and must
1524 * not be reused until it has been unregistered. A negative errno code
1525 * is returned on a failure.
1527 * When registered all registration and up events are replayed
1528 * to the new notifier to allow device to have a race free
1529 * view of the network device list.
1532 int register_netdevice_notifier(struct notifier_block *nb)
1534 struct net_device *dev;
1535 struct net_device *last;
1540 err = raw_notifier_chain_register(&netdev_chain, nb);
1546 for_each_netdev(net, dev) {
1547 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1548 err = notifier_to_errno(err);
1552 if (!(dev->flags & IFF_UP))
1555 call_netdevice_notifier(nb, NETDEV_UP, dev);
1566 for_each_netdev(net, dev) {
1570 if (dev->flags & IFF_UP) {
1571 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1573 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1575 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1580 raw_notifier_chain_unregister(&netdev_chain, nb);
1583 EXPORT_SYMBOL(register_netdevice_notifier);
1586 * unregister_netdevice_notifier - unregister a network notifier block
1589 * Unregister a notifier previously registered by
1590 * register_netdevice_notifier(). The notifier is unlinked into the
1591 * kernel structures and may then be reused. A negative errno code
1592 * is returned on a failure.
1594 * After unregistering unregister and down device events are synthesized
1595 * for all devices on the device list to the removed notifier to remove
1596 * the need for special case cleanup code.
1599 int unregister_netdevice_notifier(struct notifier_block *nb)
1601 struct net_device *dev;
1606 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1611 for_each_netdev(net, dev) {
1612 if (dev->flags & IFF_UP) {
1613 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1615 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1617 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1624 EXPORT_SYMBOL(unregister_netdevice_notifier);
1627 * call_netdevice_notifiers_info - call all network notifier blocks
1628 * @val: value passed unmodified to notifier function
1629 * @dev: net_device pointer passed unmodified to notifier function
1630 * @info: notifier information data
1632 * Call all network notifier blocks. Parameters and return value
1633 * are as for raw_notifier_call_chain().
1636 static int call_netdevice_notifiers_info(unsigned long val,
1637 struct net_device *dev,
1638 struct netdev_notifier_info *info)
1641 netdev_notifier_info_init(info, dev);
1642 return raw_notifier_call_chain(&netdev_chain, val, info);
1646 * call_netdevice_notifiers - call all network notifier blocks
1647 * @val: value passed unmodified to notifier function
1648 * @dev: net_device pointer passed unmodified to notifier function
1650 * Call all network notifier blocks. Parameters and return value
1651 * are as for raw_notifier_call_chain().
1654 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1656 struct netdev_notifier_info info;
1658 return call_netdevice_notifiers_info(val, dev, &info);
1660 EXPORT_SYMBOL(call_netdevice_notifiers);
1662 #ifdef CONFIG_NET_INGRESS
1663 static struct static_key ingress_needed __read_mostly;
1665 void net_inc_ingress_queue(void)
1667 static_key_slow_inc(&ingress_needed);
1669 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1671 void net_dec_ingress_queue(void)
1673 static_key_slow_dec(&ingress_needed);
1675 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1678 static struct static_key netstamp_needed __read_mostly;
1679 #ifdef HAVE_JUMP_LABEL
1680 /* We are not allowed to call static_key_slow_dec() from irq context
1681 * If net_disable_timestamp() is called from irq context, defer the
1682 * static_key_slow_dec() calls.
1684 static atomic_t netstamp_needed_deferred;
1687 void net_enable_timestamp(void)
1689 #ifdef HAVE_JUMP_LABEL
1690 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1694 static_key_slow_dec(&netstamp_needed);
1698 static_key_slow_inc(&netstamp_needed);
1700 EXPORT_SYMBOL(net_enable_timestamp);
1702 void net_disable_timestamp(void)
1704 #ifdef HAVE_JUMP_LABEL
1705 if (in_interrupt()) {
1706 atomic_inc(&netstamp_needed_deferred);
1710 static_key_slow_dec(&netstamp_needed);
1712 EXPORT_SYMBOL(net_disable_timestamp);
1714 static inline void net_timestamp_set(struct sk_buff *skb)
1716 skb->tstamp.tv64 = 0;
1717 if (static_key_false(&netstamp_needed))
1718 __net_timestamp(skb);
1721 #define net_timestamp_check(COND, SKB) \
1722 if (static_key_false(&netstamp_needed)) { \
1723 if ((COND) && !(SKB)->tstamp.tv64) \
1724 __net_timestamp(SKB); \
1727 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1731 if (!(dev->flags & IFF_UP))
1734 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1735 if (skb->len <= len)
1738 /* if TSO is enabled, we don't care about the length as the packet
1739 * could be forwarded without being segmented before
1741 if (skb_is_gso(skb))
1746 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1748 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1750 if (skb_orphan_frags(skb, GFP_ATOMIC) ||
1751 unlikely(!is_skb_forwardable(dev, skb))) {
1752 atomic_long_inc(&dev->rx_dropped);
1757 skb_scrub_packet(skb, true);
1759 skb->protocol = eth_type_trans(skb, dev);
1760 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1764 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1767 * dev_forward_skb - loopback an skb to another netif
1769 * @dev: destination network device
1770 * @skb: buffer to forward
1773 * NET_RX_SUCCESS (no congestion)
1774 * NET_RX_DROP (packet was dropped, but freed)
1776 * dev_forward_skb can be used for injecting an skb from the
1777 * start_xmit function of one device into the receive queue
1778 * of another device.
1780 * The receiving device may be in another namespace, so
1781 * we have to clear all information in the skb that could
1782 * impact namespace isolation.
1784 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1786 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1788 EXPORT_SYMBOL_GPL(dev_forward_skb);
1790 static inline int deliver_skb(struct sk_buff *skb,
1791 struct packet_type *pt_prev,
1792 struct net_device *orig_dev)
1794 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1796 atomic_inc(&skb->users);
1797 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1800 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1801 struct packet_type **pt,
1802 struct net_device *orig_dev,
1804 struct list_head *ptype_list)
1806 struct packet_type *ptype, *pt_prev = *pt;
1808 list_for_each_entry_rcu(ptype, ptype_list, list) {
1809 if (ptype->type != type)
1812 deliver_skb(skb, pt_prev, orig_dev);
1818 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1820 if (!ptype->af_packet_priv || !skb->sk)
1823 if (ptype->id_match)
1824 return ptype->id_match(ptype, skb->sk);
1825 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1832 * Support routine. Sends outgoing frames to any network
1833 * taps currently in use.
1836 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1838 struct packet_type *ptype;
1839 struct sk_buff *skb2 = NULL;
1840 struct packet_type *pt_prev = NULL;
1841 struct list_head *ptype_list = &ptype_all;
1845 list_for_each_entry_rcu(ptype, ptype_list, list) {
1846 /* Never send packets back to the socket
1847 * they originated from - MvS (miquels@drinkel.ow.org)
1849 if (skb_loop_sk(ptype, skb))
1853 deliver_skb(skb2, pt_prev, skb->dev);
1858 /* need to clone skb, done only once */
1859 skb2 = skb_clone(skb, GFP_ATOMIC);
1863 net_timestamp_set(skb2);
1865 /* skb->nh should be correctly
1866 * set by sender, so that the second statement is
1867 * just protection against buggy protocols.
1869 skb_reset_mac_header(skb2);
1871 if (skb_network_header(skb2) < skb2->data ||
1872 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1873 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1874 ntohs(skb2->protocol),
1876 skb_reset_network_header(skb2);
1879 skb2->transport_header = skb2->network_header;
1880 skb2->pkt_type = PACKET_OUTGOING;
1884 if (ptype_list == &ptype_all) {
1885 ptype_list = &dev->ptype_all;
1890 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1895 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1896 * @dev: Network device
1897 * @txq: number of queues available
1899 * If real_num_tx_queues is changed the tc mappings may no longer be
1900 * valid. To resolve this verify the tc mapping remains valid and if
1901 * not NULL the mapping. With no priorities mapping to this
1902 * offset/count pair it will no longer be used. In the worst case TC0
1903 * is invalid nothing can be done so disable priority mappings. If is
1904 * expected that drivers will fix this mapping if they can before
1905 * calling netif_set_real_num_tx_queues.
1907 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1910 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1912 /* If TC0 is invalidated disable TC mapping */
1913 if (tc->offset + tc->count > txq) {
1914 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1919 /* Invalidated prio to tc mappings set to TC0 */
1920 for (i = 1; i < TC_BITMASK + 1; i++) {
1921 int q = netdev_get_prio_tc_map(dev, i);
1923 tc = &dev->tc_to_txq[q];
1924 if (tc->offset + tc->count > txq) {
1925 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1927 netdev_set_prio_tc_map(dev, i, 0);
1933 static DEFINE_MUTEX(xps_map_mutex);
1934 #define xmap_dereference(P) \
1935 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1937 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1940 struct xps_map *map = NULL;
1944 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1946 for (pos = 0; map && pos < map->len; pos++) {
1947 if (map->queues[pos] == index) {
1949 map->queues[pos] = map->queues[--map->len];
1951 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1952 kfree_rcu(map, rcu);
1962 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1964 struct xps_dev_maps *dev_maps;
1966 bool active = false;
1968 mutex_lock(&xps_map_mutex);
1969 dev_maps = xmap_dereference(dev->xps_maps);
1974 for_each_possible_cpu(cpu) {
1975 for (i = index; i < dev->num_tx_queues; i++) {
1976 if (!remove_xps_queue(dev_maps, cpu, i))
1979 if (i == dev->num_tx_queues)
1984 RCU_INIT_POINTER(dev->xps_maps, NULL);
1985 kfree_rcu(dev_maps, rcu);
1988 for (i = index; i < dev->num_tx_queues; i++)
1989 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1993 mutex_unlock(&xps_map_mutex);
1996 static struct xps_map *expand_xps_map(struct xps_map *map,
1999 struct xps_map *new_map;
2000 int alloc_len = XPS_MIN_MAP_ALLOC;
2003 for (pos = 0; map && pos < map->len; pos++) {
2004 if (map->queues[pos] != index)
2009 /* Need to add queue to this CPU's existing map */
2011 if (pos < map->alloc_len)
2014 alloc_len = map->alloc_len * 2;
2017 /* Need to allocate new map to store queue on this CPU's map */
2018 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2023 for (i = 0; i < pos; i++)
2024 new_map->queues[i] = map->queues[i];
2025 new_map->alloc_len = alloc_len;
2031 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2034 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2035 struct xps_map *map, *new_map;
2036 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2037 int cpu, numa_node_id = -2;
2038 bool active = false;
2040 mutex_lock(&xps_map_mutex);
2042 dev_maps = xmap_dereference(dev->xps_maps);
2044 /* allocate memory for queue storage */
2045 for_each_online_cpu(cpu) {
2046 if (!cpumask_test_cpu(cpu, mask))
2050 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2051 if (!new_dev_maps) {
2052 mutex_unlock(&xps_map_mutex);
2056 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2059 map = expand_xps_map(map, cpu, index);
2063 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2067 goto out_no_new_maps;
2069 for_each_possible_cpu(cpu) {
2070 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2071 /* add queue to CPU maps */
2074 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2075 while ((pos < map->len) && (map->queues[pos] != index))
2078 if (pos == map->len)
2079 map->queues[map->len++] = index;
2081 if (numa_node_id == -2)
2082 numa_node_id = cpu_to_node(cpu);
2083 else if (numa_node_id != cpu_to_node(cpu))
2086 } else if (dev_maps) {
2087 /* fill in the new device map from the old device map */
2088 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2089 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2094 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2096 /* Cleanup old maps */
2098 for_each_possible_cpu(cpu) {
2099 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2100 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2101 if (map && map != new_map)
2102 kfree_rcu(map, rcu);
2105 kfree_rcu(dev_maps, rcu);
2108 dev_maps = new_dev_maps;
2112 /* update Tx queue numa node */
2113 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2114 (numa_node_id >= 0) ? numa_node_id :
2120 /* removes queue from unused CPUs */
2121 for_each_possible_cpu(cpu) {
2122 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2125 if (remove_xps_queue(dev_maps, cpu, index))
2129 /* free map if not active */
2131 RCU_INIT_POINTER(dev->xps_maps, NULL);
2132 kfree_rcu(dev_maps, rcu);
2136 mutex_unlock(&xps_map_mutex);
2140 /* remove any maps that we added */
2141 for_each_possible_cpu(cpu) {
2142 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2143 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2145 if (new_map && new_map != map)
2149 mutex_unlock(&xps_map_mutex);
2151 kfree(new_dev_maps);
2154 EXPORT_SYMBOL(netif_set_xps_queue);
2158 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2159 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2161 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2165 if (txq < 1 || txq > dev->num_tx_queues)
2168 if (dev->reg_state == NETREG_REGISTERED ||
2169 dev->reg_state == NETREG_UNREGISTERING) {
2172 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2178 netif_setup_tc(dev, txq);
2180 if (txq < dev->real_num_tx_queues) {
2181 qdisc_reset_all_tx_gt(dev, txq);
2183 netif_reset_xps_queues_gt(dev, txq);
2188 dev->real_num_tx_queues = txq;
2191 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2195 * netif_set_real_num_rx_queues - set actual number of RX queues used
2196 * @dev: Network device
2197 * @rxq: Actual number of RX queues
2199 * This must be called either with the rtnl_lock held or before
2200 * registration of the net device. Returns 0 on success, or a
2201 * negative error code. If called before registration, it always
2204 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2208 if (rxq < 1 || rxq > dev->num_rx_queues)
2211 if (dev->reg_state == NETREG_REGISTERED) {
2214 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2220 dev->real_num_rx_queues = rxq;
2223 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2227 * netif_get_num_default_rss_queues - default number of RSS queues
2229 * This routine should set an upper limit on the number of RSS queues
2230 * used by default by multiqueue devices.
2232 int netif_get_num_default_rss_queues(void)
2234 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2236 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2238 static inline void __netif_reschedule(struct Qdisc *q)
2240 struct softnet_data *sd;
2241 unsigned long flags;
2243 local_irq_save(flags);
2244 sd = this_cpu_ptr(&softnet_data);
2245 q->next_sched = NULL;
2246 *sd->output_queue_tailp = q;
2247 sd->output_queue_tailp = &q->next_sched;
2248 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2249 local_irq_restore(flags);
2252 void __netif_schedule(struct Qdisc *q)
2254 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2255 __netif_reschedule(q);
2257 EXPORT_SYMBOL(__netif_schedule);
2259 struct dev_kfree_skb_cb {
2260 enum skb_free_reason reason;
2263 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2265 return (struct dev_kfree_skb_cb *)skb->cb;
2268 void netif_schedule_queue(struct netdev_queue *txq)
2271 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2272 struct Qdisc *q = rcu_dereference(txq->qdisc);
2274 __netif_schedule(q);
2278 EXPORT_SYMBOL(netif_schedule_queue);
2281 * netif_wake_subqueue - allow sending packets on subqueue
2282 * @dev: network device
2283 * @queue_index: sub queue index
2285 * Resume individual transmit queue of a device with multiple transmit queues.
2287 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2289 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2291 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2295 q = rcu_dereference(txq->qdisc);
2296 __netif_schedule(q);
2300 EXPORT_SYMBOL(netif_wake_subqueue);
2302 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2304 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2308 q = rcu_dereference(dev_queue->qdisc);
2309 __netif_schedule(q);
2313 EXPORT_SYMBOL(netif_tx_wake_queue);
2315 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2317 unsigned long flags;
2319 if (likely(atomic_read(&skb->users) == 1)) {
2321 atomic_set(&skb->users, 0);
2322 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2325 get_kfree_skb_cb(skb)->reason = reason;
2326 local_irq_save(flags);
2327 skb->next = __this_cpu_read(softnet_data.completion_queue);
2328 __this_cpu_write(softnet_data.completion_queue, skb);
2329 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2330 local_irq_restore(flags);
2332 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2334 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2336 if (in_irq() || irqs_disabled())
2337 __dev_kfree_skb_irq(skb, reason);
2341 EXPORT_SYMBOL(__dev_kfree_skb_any);
2345 * netif_device_detach - mark device as removed
2346 * @dev: network device
2348 * Mark device as removed from system and therefore no longer available.
2350 void netif_device_detach(struct net_device *dev)
2352 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2353 netif_running(dev)) {
2354 netif_tx_stop_all_queues(dev);
2357 EXPORT_SYMBOL(netif_device_detach);
2360 * netif_device_attach - mark device as attached
2361 * @dev: network device
2363 * Mark device as attached from system and restart if needed.
2365 void netif_device_attach(struct net_device *dev)
2367 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2368 netif_running(dev)) {
2369 netif_tx_wake_all_queues(dev);
2370 __netdev_watchdog_up(dev);
2373 EXPORT_SYMBOL(netif_device_attach);
2376 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2377 * to be used as a distribution range.
2379 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2380 unsigned int num_tx_queues)
2384 u16 qcount = num_tx_queues;
2386 if (skb_rx_queue_recorded(skb)) {
2387 hash = skb_get_rx_queue(skb);
2388 while (unlikely(hash >= num_tx_queues))
2389 hash -= num_tx_queues;
2394 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2395 qoffset = dev->tc_to_txq[tc].offset;
2396 qcount = dev->tc_to_txq[tc].count;
2399 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2401 EXPORT_SYMBOL(__skb_tx_hash);
2403 static void skb_warn_bad_offload(const struct sk_buff *skb)
2405 static const netdev_features_t null_features = 0;
2406 struct net_device *dev = skb->dev;
2407 const char *name = "";
2409 if (!net_ratelimit())
2413 if (dev->dev.parent)
2414 name = dev_driver_string(dev->dev.parent);
2416 name = netdev_name(dev);
2418 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2419 "gso_type=%d ip_summed=%d\n",
2420 name, dev ? &dev->features : &null_features,
2421 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2422 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2423 skb_shinfo(skb)->gso_type, skb->ip_summed);
2427 * Invalidate hardware checksum when packet is to be mangled, and
2428 * complete checksum manually on outgoing path.
2430 int skb_checksum_help(struct sk_buff *skb)
2433 int ret = 0, offset;
2435 if (skb->ip_summed == CHECKSUM_COMPLETE)
2436 goto out_set_summed;
2438 if (unlikely(skb_shinfo(skb)->gso_size)) {
2439 skb_warn_bad_offload(skb);
2443 /* Before computing a checksum, we should make sure no frag could
2444 * be modified by an external entity : checksum could be wrong.
2446 if (skb_has_shared_frag(skb)) {
2447 ret = __skb_linearize(skb);
2452 offset = skb_checksum_start_offset(skb);
2453 BUG_ON(offset >= skb_headlen(skb));
2454 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2456 offset += skb->csum_offset;
2457 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2459 if (skb_cloned(skb) &&
2460 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2461 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2466 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2468 skb->ip_summed = CHECKSUM_NONE;
2472 EXPORT_SYMBOL(skb_checksum_help);
2474 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2476 __be16 type = skb->protocol;
2478 /* Tunnel gso handlers can set protocol to ethernet. */
2479 if (type == htons(ETH_P_TEB)) {
2482 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2485 eth = (struct ethhdr *)skb_mac_header(skb);
2486 type = eth->h_proto;
2489 return __vlan_get_protocol(skb, type, depth);
2493 * skb_mac_gso_segment - mac layer segmentation handler.
2494 * @skb: buffer to segment
2495 * @features: features for the output path (see dev->features)
2497 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2498 netdev_features_t features)
2500 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2501 struct packet_offload *ptype;
2502 int vlan_depth = skb->mac_len;
2503 __be16 type = skb_network_protocol(skb, &vlan_depth);
2505 if (unlikely(!type))
2506 return ERR_PTR(-EINVAL);
2508 __skb_pull(skb, vlan_depth);
2511 list_for_each_entry_rcu(ptype, &offload_base, list) {
2512 if (ptype->type == type && ptype->callbacks.gso_segment) {
2513 segs = ptype->callbacks.gso_segment(skb, features);
2519 __skb_push(skb, skb->data - skb_mac_header(skb));
2523 EXPORT_SYMBOL(skb_mac_gso_segment);
2526 /* openvswitch calls this on rx path, so we need a different check.
2528 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2531 return skb->ip_summed != CHECKSUM_PARTIAL;
2533 return skb->ip_summed == CHECKSUM_NONE;
2537 * __skb_gso_segment - Perform segmentation on skb.
2538 * @skb: buffer to segment
2539 * @features: features for the output path (see dev->features)
2540 * @tx_path: whether it is called in TX path
2542 * This function segments the given skb and returns a list of segments.
2544 * It may return NULL if the skb requires no segmentation. This is
2545 * only possible when GSO is used for verifying header integrity.
2547 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2548 netdev_features_t features, bool tx_path)
2550 if (unlikely(skb_needs_check(skb, tx_path))) {
2553 skb_warn_bad_offload(skb);
2555 err = skb_cow_head(skb, 0);
2557 return ERR_PTR(err);
2560 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2561 SKB_GSO_CB(skb)->encap_level = 0;
2563 skb_reset_mac_header(skb);
2564 skb_reset_mac_len(skb);
2566 return skb_mac_gso_segment(skb, features);
2568 EXPORT_SYMBOL(__skb_gso_segment);
2570 /* Take action when hardware reception checksum errors are detected. */
2572 void netdev_rx_csum_fault(struct net_device *dev)
2574 if (net_ratelimit()) {
2575 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2579 EXPORT_SYMBOL(netdev_rx_csum_fault);
2582 /* Actually, we should eliminate this check as soon as we know, that:
2583 * 1. IOMMU is present and allows to map all the memory.
2584 * 2. No high memory really exists on this machine.
2587 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2589 #ifdef CONFIG_HIGHMEM
2591 if (!(dev->features & NETIF_F_HIGHDMA)) {
2592 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2593 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2594 if (PageHighMem(skb_frag_page(frag)))
2599 if (PCI_DMA_BUS_IS_PHYS) {
2600 struct device *pdev = dev->dev.parent;
2604 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2605 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2606 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2607 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2615 /* If MPLS offload request, verify we are testing hardware MPLS features
2616 * instead of standard features for the netdev.
2618 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2619 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2620 netdev_features_t features,
2623 if (eth_p_mpls(type))
2624 features &= skb->dev->mpls_features;
2629 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2630 netdev_features_t features,
2637 static netdev_features_t harmonize_features(struct sk_buff *skb,
2638 netdev_features_t features)
2643 type = skb_network_protocol(skb, &tmp);
2644 features = net_mpls_features(skb, features, type);
2646 if (skb->ip_summed != CHECKSUM_NONE &&
2647 !can_checksum_protocol(features, type)) {
2648 features &= ~NETIF_F_ALL_CSUM;
2649 } else if (illegal_highdma(skb->dev, skb)) {
2650 features &= ~NETIF_F_SG;
2656 netdev_features_t passthru_features_check(struct sk_buff *skb,
2657 struct net_device *dev,
2658 netdev_features_t features)
2662 EXPORT_SYMBOL(passthru_features_check);
2664 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2665 struct net_device *dev,
2666 netdev_features_t features)
2668 return vlan_features_check(skb, features);
2671 netdev_features_t netif_skb_features(struct sk_buff *skb)
2673 struct net_device *dev = skb->dev;
2674 netdev_features_t features = dev->features;
2675 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2677 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2678 features &= ~NETIF_F_GSO_MASK;
2680 /* If encapsulation offload request, verify we are testing
2681 * hardware encapsulation features instead of standard
2682 * features for the netdev
2684 if (skb->encapsulation)
2685 features &= dev->hw_enc_features;
2687 if (skb_vlan_tagged(skb))
2688 features = netdev_intersect_features(features,
2689 dev->vlan_features |
2690 NETIF_F_HW_VLAN_CTAG_TX |
2691 NETIF_F_HW_VLAN_STAG_TX);
2693 if (dev->netdev_ops->ndo_features_check)
2694 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2697 features &= dflt_features_check(skb, dev, features);
2699 return harmonize_features(skb, features);
2701 EXPORT_SYMBOL(netif_skb_features);
2703 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2704 struct netdev_queue *txq, bool more)
2709 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2710 dev_queue_xmit_nit(skb, dev);
2713 trace_net_dev_start_xmit(skb, dev);
2714 rc = netdev_start_xmit(skb, dev, txq, more);
2715 trace_net_dev_xmit(skb, rc, dev, len);
2720 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2721 struct netdev_queue *txq, int *ret)
2723 struct sk_buff *skb = first;
2724 int rc = NETDEV_TX_OK;
2727 struct sk_buff *next = skb->next;
2730 rc = xmit_one(skb, dev, txq, next != NULL);
2731 if (unlikely(!dev_xmit_complete(rc))) {
2737 if (netif_xmit_stopped(txq) && skb) {
2738 rc = NETDEV_TX_BUSY;
2748 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2749 netdev_features_t features)
2751 if (skb_vlan_tag_present(skb) &&
2752 !vlan_hw_offload_capable(features, skb->vlan_proto))
2753 skb = __vlan_hwaccel_push_inside(skb);
2757 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2759 netdev_features_t features;
2764 features = netif_skb_features(skb);
2765 skb = validate_xmit_vlan(skb, features);
2769 if (netif_needs_gso(skb, features)) {
2770 struct sk_buff *segs;
2772 segs = skb_gso_segment(skb, features);
2780 if (skb_needs_linearize(skb, features) &&
2781 __skb_linearize(skb))
2784 /* If packet is not checksummed and device does not
2785 * support checksumming for this protocol, complete
2786 * checksumming here.
2788 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2789 if (skb->encapsulation)
2790 skb_set_inner_transport_header(skb,
2791 skb_checksum_start_offset(skb));
2793 skb_set_transport_header(skb,
2794 skb_checksum_start_offset(skb));
2795 if (!(features & NETIF_F_ALL_CSUM) &&
2796 skb_checksum_help(skb))
2809 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2811 struct sk_buff *next, *head = NULL, *tail;
2813 for (; skb != NULL; skb = next) {
2817 /* in case skb wont be segmented, point to itself */
2820 skb = validate_xmit_skb(skb, dev);
2828 /* If skb was segmented, skb->prev points to
2829 * the last segment. If not, it still contains skb.
2836 static void qdisc_pkt_len_init(struct sk_buff *skb)
2838 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2840 qdisc_skb_cb(skb)->pkt_len = skb->len;
2842 /* To get more precise estimation of bytes sent on wire,
2843 * we add to pkt_len the headers size of all segments
2845 if (shinfo->gso_size) {
2846 unsigned int hdr_len;
2847 u16 gso_segs = shinfo->gso_segs;
2849 /* mac layer + network layer */
2850 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2852 /* + transport layer */
2853 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2854 hdr_len += tcp_hdrlen(skb);
2856 hdr_len += sizeof(struct udphdr);
2858 if (shinfo->gso_type & SKB_GSO_DODGY)
2859 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2862 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2866 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2867 struct net_device *dev,
2868 struct netdev_queue *txq)
2870 spinlock_t *root_lock = qdisc_lock(q);
2874 qdisc_pkt_len_init(skb);
2875 qdisc_calculate_pkt_len(skb, q);
2877 * Heuristic to force contended enqueues to serialize on a
2878 * separate lock before trying to get qdisc main lock.
2879 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2880 * often and dequeue packets faster.
2882 contended = qdisc_is_running(q);
2883 if (unlikely(contended))
2884 spin_lock(&q->busylock);
2886 spin_lock(root_lock);
2887 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2890 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2891 qdisc_run_begin(q)) {
2893 * This is a work-conserving queue; there are no old skbs
2894 * waiting to be sent out; and the qdisc is not running -
2895 * xmit the skb directly.
2898 qdisc_bstats_update(q, skb);
2900 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2901 if (unlikely(contended)) {
2902 spin_unlock(&q->busylock);
2909 rc = NET_XMIT_SUCCESS;
2911 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2912 if (qdisc_run_begin(q)) {
2913 if (unlikely(contended)) {
2914 spin_unlock(&q->busylock);
2920 spin_unlock(root_lock);
2921 if (unlikely(contended))
2922 spin_unlock(&q->busylock);
2926 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2927 static void skb_update_prio(struct sk_buff *skb)
2929 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2931 if (!skb->priority && skb->sk && map) {
2932 unsigned int prioidx =
2933 sock_cgroup_prioidx(&skb->sk->sk_cgrp_data);
2935 if (prioidx < map->priomap_len)
2936 skb->priority = map->priomap[prioidx];
2940 #define skb_update_prio(skb)
2943 DEFINE_PER_CPU(int, xmit_recursion);
2944 EXPORT_SYMBOL(xmit_recursion);
2946 #define RECURSION_LIMIT 10
2949 * dev_loopback_xmit - loop back @skb
2950 * @net: network namespace this loopback is happening in
2951 * @sk: sk needed to be a netfilter okfn
2952 * @skb: buffer to transmit
2954 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
2956 skb_reset_mac_header(skb);
2957 __skb_pull(skb, skb_network_offset(skb));
2958 skb->pkt_type = PACKET_LOOPBACK;
2959 skb->ip_summed = CHECKSUM_UNNECESSARY;
2960 WARN_ON(!skb_dst(skb));
2965 EXPORT_SYMBOL(dev_loopback_xmit);
2967 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2970 struct xps_dev_maps *dev_maps;
2971 struct xps_map *map;
2972 int queue_index = -1;
2975 dev_maps = rcu_dereference(dev->xps_maps);
2977 map = rcu_dereference(
2978 dev_maps->cpu_map[skb->sender_cpu - 1]);
2981 queue_index = map->queues[0];
2983 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
2985 if (unlikely(queue_index >= dev->real_num_tx_queues))
2997 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
2999 struct sock *sk = skb->sk;
3000 int queue_index = sk_tx_queue_get(sk);
3002 if (queue_index < 0 || skb->ooo_okay ||
3003 queue_index >= dev->real_num_tx_queues) {
3004 int new_index = get_xps_queue(dev, skb);
3006 new_index = skb_tx_hash(dev, skb);
3008 if (queue_index != new_index && sk &&
3010 rcu_access_pointer(sk->sk_dst_cache))
3011 sk_tx_queue_set(sk, new_index);
3013 queue_index = new_index;
3019 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3020 struct sk_buff *skb,
3023 int queue_index = 0;
3026 u32 sender_cpu = skb->sender_cpu - 1;
3028 if (sender_cpu >= (u32)NR_CPUS)
3029 skb->sender_cpu = raw_smp_processor_id() + 1;
3032 if (dev->real_num_tx_queues != 1) {
3033 const struct net_device_ops *ops = dev->netdev_ops;
3034 if (ops->ndo_select_queue)
3035 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3038 queue_index = __netdev_pick_tx(dev, skb);
3041 queue_index = netdev_cap_txqueue(dev, queue_index);
3044 skb_set_queue_mapping(skb, queue_index);
3045 return netdev_get_tx_queue(dev, queue_index);
3049 * __dev_queue_xmit - transmit a buffer
3050 * @skb: buffer to transmit
3051 * @accel_priv: private data used for L2 forwarding offload
3053 * Queue a buffer for transmission to a network device. The caller must
3054 * have set the device and priority and built the buffer before calling
3055 * this function. The function can be called from an interrupt.
3057 * A negative errno code is returned on a failure. A success does not
3058 * guarantee the frame will be transmitted as it may be dropped due
3059 * to congestion or traffic shaping.
3061 * -----------------------------------------------------------------------------------
3062 * I notice this method can also return errors from the queue disciplines,
3063 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3066 * Regardless of the return value, the skb is consumed, so it is currently
3067 * difficult to retry a send to this method. (You can bump the ref count
3068 * before sending to hold a reference for retry if you are careful.)
3070 * When calling this method, interrupts MUST be enabled. This is because
3071 * the BH enable code must have IRQs enabled so that it will not deadlock.
3074 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3076 struct net_device *dev = skb->dev;
3077 struct netdev_queue *txq;
3081 skb_reset_mac_header(skb);
3083 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3084 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3086 /* Disable soft irqs for various locks below. Also
3087 * stops preemption for RCU.
3091 skb_update_prio(skb);
3093 /* If device/qdisc don't need skb->dst, release it right now while
3094 * its hot in this cpu cache.
3096 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3101 #ifdef CONFIG_NET_SWITCHDEV
3102 /* Don't forward if offload device already forwarded */
3103 if (skb->offload_fwd_mark &&
3104 skb->offload_fwd_mark == dev->offload_fwd_mark) {
3106 rc = NET_XMIT_SUCCESS;
3111 txq = netdev_pick_tx(dev, skb, accel_priv);
3112 q = rcu_dereference_bh(txq->qdisc);
3114 #ifdef CONFIG_NET_CLS_ACT
3115 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3117 trace_net_dev_queue(skb);
3119 rc = __dev_xmit_skb(skb, q, dev, txq);
3123 /* The device has no queue. Common case for software devices:
3124 loopback, all the sorts of tunnels...
3126 Really, it is unlikely that netif_tx_lock protection is necessary
3127 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3129 However, it is possible, that they rely on protection
3132 Check this and shot the lock. It is not prone from deadlocks.
3133 Either shot noqueue qdisc, it is even simpler 8)
3135 if (dev->flags & IFF_UP) {
3136 int cpu = smp_processor_id(); /* ok because BHs are off */
3138 if (txq->xmit_lock_owner != cpu) {
3140 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
3141 goto recursion_alert;
3143 skb = validate_xmit_skb(skb, dev);
3147 HARD_TX_LOCK(dev, txq, cpu);
3149 if (!netif_xmit_stopped(txq)) {
3150 __this_cpu_inc(xmit_recursion);
3151 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3152 __this_cpu_dec(xmit_recursion);
3153 if (dev_xmit_complete(rc)) {
3154 HARD_TX_UNLOCK(dev, txq);
3158 HARD_TX_UNLOCK(dev, txq);
3159 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3162 /* Recursion is detected! It is possible,
3166 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3173 rcu_read_unlock_bh();
3175 atomic_long_inc(&dev->tx_dropped);
3176 kfree_skb_list(skb);
3179 rcu_read_unlock_bh();
3183 int dev_queue_xmit(struct sk_buff *skb)
3185 return __dev_queue_xmit(skb, NULL);
3187 EXPORT_SYMBOL(dev_queue_xmit);
3189 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3191 return __dev_queue_xmit(skb, accel_priv);
3193 EXPORT_SYMBOL(dev_queue_xmit_accel);
3196 /*=======================================================================
3198 =======================================================================*/
3200 int netdev_max_backlog __read_mostly = 1000;
3201 EXPORT_SYMBOL(netdev_max_backlog);
3203 int netdev_tstamp_prequeue __read_mostly = 1;
3204 int netdev_budget __read_mostly = 300;
3205 int weight_p __read_mostly = 64; /* old backlog weight */
3207 /* Called with irq disabled */
3208 static inline void ____napi_schedule(struct softnet_data *sd,
3209 struct napi_struct *napi)
3211 list_add_tail(&napi->poll_list, &sd->poll_list);
3212 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3217 /* One global table that all flow-based protocols share. */
3218 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3219 EXPORT_SYMBOL(rps_sock_flow_table);
3220 u32 rps_cpu_mask __read_mostly;
3221 EXPORT_SYMBOL(rps_cpu_mask);
3223 struct static_key rps_needed __read_mostly;
3225 static struct rps_dev_flow *
3226 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3227 struct rps_dev_flow *rflow, u16 next_cpu)
3229 if (next_cpu < nr_cpu_ids) {
3230 #ifdef CONFIG_RFS_ACCEL
3231 struct netdev_rx_queue *rxqueue;
3232 struct rps_dev_flow_table *flow_table;
3233 struct rps_dev_flow *old_rflow;
3238 /* Should we steer this flow to a different hardware queue? */
3239 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3240 !(dev->features & NETIF_F_NTUPLE))
3242 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3243 if (rxq_index == skb_get_rx_queue(skb))
3246 rxqueue = dev->_rx + rxq_index;
3247 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3250 flow_id = skb_get_hash(skb) & flow_table->mask;
3251 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3252 rxq_index, flow_id);
3256 rflow = &flow_table->flows[flow_id];
3258 if (old_rflow->filter == rflow->filter)
3259 old_rflow->filter = RPS_NO_FILTER;
3263 per_cpu(softnet_data, next_cpu).input_queue_head;
3266 rflow->cpu = next_cpu;
3271 * get_rps_cpu is called from netif_receive_skb and returns the target
3272 * CPU from the RPS map of the receiving queue for a given skb.
3273 * rcu_read_lock must be held on entry.
3275 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3276 struct rps_dev_flow **rflowp)
3278 const struct rps_sock_flow_table *sock_flow_table;
3279 struct netdev_rx_queue *rxqueue = dev->_rx;
3280 struct rps_dev_flow_table *flow_table;
3281 struct rps_map *map;
3286 if (skb_rx_queue_recorded(skb)) {
3287 u16 index = skb_get_rx_queue(skb);
3289 if (unlikely(index >= dev->real_num_rx_queues)) {
3290 WARN_ONCE(dev->real_num_rx_queues > 1,
3291 "%s received packet on queue %u, but number "
3292 "of RX queues is %u\n",
3293 dev->name, index, dev->real_num_rx_queues);
3299 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3301 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3302 map = rcu_dereference(rxqueue->rps_map);
3303 if (!flow_table && !map)
3306 skb_reset_network_header(skb);
3307 hash = skb_get_hash(skb);
3311 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3312 if (flow_table && sock_flow_table) {
3313 struct rps_dev_flow *rflow;
3317 /* First check into global flow table if there is a match */
3318 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3319 if ((ident ^ hash) & ~rps_cpu_mask)
3322 next_cpu = ident & rps_cpu_mask;
3324 /* OK, now we know there is a match,
3325 * we can look at the local (per receive queue) flow table
3327 rflow = &flow_table->flows[hash & flow_table->mask];
3331 * If the desired CPU (where last recvmsg was done) is
3332 * different from current CPU (one in the rx-queue flow
3333 * table entry), switch if one of the following holds:
3334 * - Current CPU is unset (>= nr_cpu_ids).
3335 * - Current CPU is offline.
3336 * - The current CPU's queue tail has advanced beyond the
3337 * last packet that was enqueued using this table entry.
3338 * This guarantees that all previous packets for the flow
3339 * have been dequeued, thus preserving in order delivery.
3341 if (unlikely(tcpu != next_cpu) &&
3342 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3343 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3344 rflow->last_qtail)) >= 0)) {
3346 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3349 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3359 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3360 if (cpu_online(tcpu)) {
3370 #ifdef CONFIG_RFS_ACCEL
3373 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3374 * @dev: Device on which the filter was set
3375 * @rxq_index: RX queue index
3376 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3377 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3379 * Drivers that implement ndo_rx_flow_steer() should periodically call
3380 * this function for each installed filter and remove the filters for
3381 * which it returns %true.
3383 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3384 u32 flow_id, u16 filter_id)
3386 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3387 struct rps_dev_flow_table *flow_table;
3388 struct rps_dev_flow *rflow;
3393 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3394 if (flow_table && flow_id <= flow_table->mask) {
3395 rflow = &flow_table->flows[flow_id];
3396 cpu = ACCESS_ONCE(rflow->cpu);
3397 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3398 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3399 rflow->last_qtail) <
3400 (int)(10 * flow_table->mask)))
3406 EXPORT_SYMBOL(rps_may_expire_flow);
3408 #endif /* CONFIG_RFS_ACCEL */
3410 /* Called from hardirq (IPI) context */
3411 static void rps_trigger_softirq(void *data)
3413 struct softnet_data *sd = data;
3415 ____napi_schedule(sd, &sd->backlog);
3419 #endif /* CONFIG_RPS */
3422 * Check if this softnet_data structure is another cpu one
3423 * If yes, queue it to our IPI list and return 1
3426 static int rps_ipi_queued(struct softnet_data *sd)
3429 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3432 sd->rps_ipi_next = mysd->rps_ipi_list;
3433 mysd->rps_ipi_list = sd;
3435 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3438 #endif /* CONFIG_RPS */
3442 #ifdef CONFIG_NET_FLOW_LIMIT
3443 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3446 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3448 #ifdef CONFIG_NET_FLOW_LIMIT
3449 struct sd_flow_limit *fl;
3450 struct softnet_data *sd;
3451 unsigned int old_flow, new_flow;
3453 if (qlen < (netdev_max_backlog >> 1))
3456 sd = this_cpu_ptr(&softnet_data);
3459 fl = rcu_dereference(sd->flow_limit);
3461 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3462 old_flow = fl->history[fl->history_head];
3463 fl->history[fl->history_head] = new_flow;
3466 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3468 if (likely(fl->buckets[old_flow]))
3469 fl->buckets[old_flow]--;
3471 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3483 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3484 * queue (may be a remote CPU queue).
3486 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3487 unsigned int *qtail)
3489 struct softnet_data *sd;
3490 unsigned long flags;
3493 sd = &per_cpu(softnet_data, cpu);
3495 local_irq_save(flags);
3498 if (!netif_running(skb->dev))
3500 qlen = skb_queue_len(&sd->input_pkt_queue);
3501 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3504 __skb_queue_tail(&sd->input_pkt_queue, skb);
3505 input_queue_tail_incr_save(sd, qtail);
3507 local_irq_restore(flags);
3508 return NET_RX_SUCCESS;
3511 /* Schedule NAPI for backlog device
3512 * We can use non atomic operation since we own the queue lock
3514 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3515 if (!rps_ipi_queued(sd))
3516 ____napi_schedule(sd, &sd->backlog);
3525 local_irq_restore(flags);
3527 atomic_long_inc(&skb->dev->rx_dropped);
3532 static int netif_rx_internal(struct sk_buff *skb)
3536 net_timestamp_check(netdev_tstamp_prequeue, skb);
3538 trace_netif_rx(skb);
3540 if (static_key_false(&rps_needed)) {
3541 struct rps_dev_flow voidflow, *rflow = &voidflow;
3547 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3549 cpu = smp_processor_id();
3551 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3559 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3566 * netif_rx - post buffer to the network code
3567 * @skb: buffer to post
3569 * This function receives a packet from a device driver and queues it for
3570 * the upper (protocol) levels to process. It always succeeds. The buffer
3571 * may be dropped during processing for congestion control or by the
3575 * NET_RX_SUCCESS (no congestion)
3576 * NET_RX_DROP (packet was dropped)
3580 int netif_rx(struct sk_buff *skb)
3582 trace_netif_rx_entry(skb);
3584 return netif_rx_internal(skb);
3586 EXPORT_SYMBOL(netif_rx);
3588 int netif_rx_ni(struct sk_buff *skb)
3592 trace_netif_rx_ni_entry(skb);
3595 err = netif_rx_internal(skb);
3596 if (local_softirq_pending())
3602 EXPORT_SYMBOL(netif_rx_ni);
3604 static void net_tx_action(struct softirq_action *h)
3606 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3608 if (sd->completion_queue) {
3609 struct sk_buff *clist;
3611 local_irq_disable();
3612 clist = sd->completion_queue;
3613 sd->completion_queue = NULL;
3617 struct sk_buff *skb = clist;
3618 clist = clist->next;
3620 WARN_ON(atomic_read(&skb->users));
3621 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3622 trace_consume_skb(skb);
3624 trace_kfree_skb(skb, net_tx_action);
3629 if (sd->output_queue) {
3632 local_irq_disable();
3633 head = sd->output_queue;
3634 sd->output_queue = NULL;
3635 sd->output_queue_tailp = &sd->output_queue;
3639 struct Qdisc *q = head;
3640 spinlock_t *root_lock;
3642 head = head->next_sched;
3644 root_lock = qdisc_lock(q);
3645 if (spin_trylock(root_lock)) {
3646 smp_mb__before_atomic();
3647 clear_bit(__QDISC_STATE_SCHED,
3650 spin_unlock(root_lock);
3652 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3654 __netif_reschedule(q);
3656 smp_mb__before_atomic();
3657 clear_bit(__QDISC_STATE_SCHED,
3665 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3666 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3667 /* This hook is defined here for ATM LANE */
3668 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3669 unsigned char *addr) __read_mostly;
3670 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3673 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3674 struct packet_type **pt_prev,
3675 int *ret, struct net_device *orig_dev)
3677 #ifdef CONFIG_NET_CLS_ACT
3678 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3679 struct tcf_result cl_res;
3681 /* If there's at least one ingress present somewhere (so
3682 * we get here via enabled static key), remaining devices
3683 * that are not configured with an ingress qdisc will bail
3689 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3693 qdisc_skb_cb(skb)->pkt_len = skb->len;
3694 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3695 qdisc_bstats_cpu_update(cl->q, skb);
3697 switch (tc_classify(skb, cl, &cl_res, false)) {
3699 case TC_ACT_RECLASSIFY:
3700 skb->tc_index = TC_H_MIN(cl_res.classid);
3703 qdisc_qstats_cpu_drop(cl->q);
3708 case TC_ACT_REDIRECT:
3709 /* skb_mac_header check was done by cls/act_bpf, so
3710 * we can safely push the L2 header back before
3711 * redirecting to another netdev
3713 __skb_push(skb, skb->mac_len);
3714 skb_do_redirect(skb);
3719 #endif /* CONFIG_NET_CLS_ACT */
3724 * netdev_rx_handler_register - register receive handler
3725 * @dev: device to register a handler for
3726 * @rx_handler: receive handler to register
3727 * @rx_handler_data: data pointer that is used by rx handler
3729 * Register a receive handler for a device. This handler will then be
3730 * called from __netif_receive_skb. A negative errno code is returned
3733 * The caller must hold the rtnl_mutex.
3735 * For a general description of rx_handler, see enum rx_handler_result.
3737 int netdev_rx_handler_register(struct net_device *dev,
3738 rx_handler_func_t *rx_handler,
3739 void *rx_handler_data)
3743 if (dev->rx_handler)
3746 /* Note: rx_handler_data must be set before rx_handler */
3747 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3748 rcu_assign_pointer(dev->rx_handler, rx_handler);
3752 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3755 * netdev_rx_handler_unregister - unregister receive handler
3756 * @dev: device to unregister a handler from
3758 * Unregister a receive handler from a device.
3760 * The caller must hold the rtnl_mutex.
3762 void netdev_rx_handler_unregister(struct net_device *dev)
3766 RCU_INIT_POINTER(dev->rx_handler, NULL);
3767 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3768 * section has a guarantee to see a non NULL rx_handler_data
3772 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3774 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3777 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3778 * the special handling of PFMEMALLOC skbs.
3780 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3782 switch (skb->protocol) {
3783 case htons(ETH_P_ARP):
3784 case htons(ETH_P_IP):
3785 case htons(ETH_P_IPV6):
3786 case htons(ETH_P_8021Q):
3787 case htons(ETH_P_8021AD):
3794 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
3795 int *ret, struct net_device *orig_dev)
3797 #ifdef CONFIG_NETFILTER_INGRESS
3798 if (nf_hook_ingress_active(skb)) {
3800 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3804 return nf_hook_ingress(skb);
3806 #endif /* CONFIG_NETFILTER_INGRESS */
3810 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3812 struct packet_type *ptype, *pt_prev;
3813 rx_handler_func_t *rx_handler;
3814 struct net_device *orig_dev;
3815 bool deliver_exact = false;
3816 int ret = NET_RX_DROP;
3819 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3821 trace_netif_receive_skb(skb);
3823 orig_dev = skb->dev;
3825 skb_reset_network_header(skb);
3826 if (!skb_transport_header_was_set(skb))
3827 skb_reset_transport_header(skb);
3828 skb_reset_mac_len(skb);
3833 skb->skb_iif = skb->dev->ifindex;
3835 __this_cpu_inc(softnet_data.processed);
3837 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3838 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3839 skb = skb_vlan_untag(skb);
3844 #ifdef CONFIG_NET_CLS_ACT
3845 if (skb->tc_verd & TC_NCLS) {
3846 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3854 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3856 ret = deliver_skb(skb, pt_prev, orig_dev);
3860 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3862 ret = deliver_skb(skb, pt_prev, orig_dev);
3867 #ifdef CONFIG_NET_INGRESS
3868 if (static_key_false(&ingress_needed)) {
3869 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3873 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
3877 #ifdef CONFIG_NET_CLS_ACT
3881 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3884 if (skb_vlan_tag_present(skb)) {
3886 ret = deliver_skb(skb, pt_prev, orig_dev);
3889 if (vlan_do_receive(&skb))
3891 else if (unlikely(!skb))
3895 rx_handler = rcu_dereference(skb->dev->rx_handler);
3898 ret = deliver_skb(skb, pt_prev, orig_dev);
3901 switch (rx_handler(&skb)) {
3902 case RX_HANDLER_CONSUMED:
3903 ret = NET_RX_SUCCESS;
3905 case RX_HANDLER_ANOTHER:
3907 case RX_HANDLER_EXACT:
3908 deliver_exact = true;
3909 case RX_HANDLER_PASS:
3916 if (unlikely(skb_vlan_tag_present(skb))) {
3917 if (skb_vlan_tag_get_id(skb))
3918 skb->pkt_type = PACKET_OTHERHOST;
3919 /* Note: we might in the future use prio bits
3920 * and set skb->priority like in vlan_do_receive()
3921 * For the time being, just ignore Priority Code Point
3926 type = skb->protocol;
3928 /* deliver only exact match when indicated */
3929 if (likely(!deliver_exact)) {
3930 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3931 &ptype_base[ntohs(type) &
3935 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3936 &orig_dev->ptype_specific);
3938 if (unlikely(skb->dev != orig_dev)) {
3939 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3940 &skb->dev->ptype_specific);
3944 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3947 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3950 atomic_long_inc(&skb->dev->rx_dropped);
3952 /* Jamal, now you will not able to escape explaining
3953 * me how you were going to use this. :-)
3962 static int __netif_receive_skb(struct sk_buff *skb)
3966 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3967 unsigned long pflags = current->flags;
3970 * PFMEMALLOC skbs are special, they should
3971 * - be delivered to SOCK_MEMALLOC sockets only
3972 * - stay away from userspace
3973 * - have bounded memory usage
3975 * Use PF_MEMALLOC as this saves us from propagating the allocation
3976 * context down to all allocation sites.
3978 current->flags |= PF_MEMALLOC;
3979 ret = __netif_receive_skb_core(skb, true);
3980 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3982 ret = __netif_receive_skb_core(skb, false);
3987 static int netif_receive_skb_internal(struct sk_buff *skb)
3991 net_timestamp_check(netdev_tstamp_prequeue, skb);
3993 if (skb_defer_rx_timestamp(skb))
3994 return NET_RX_SUCCESS;
3999 if (static_key_false(&rps_needed)) {
4000 struct rps_dev_flow voidflow, *rflow = &voidflow;
4001 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4004 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4010 ret = __netif_receive_skb(skb);
4016 * netif_receive_skb - process receive buffer from network
4017 * @skb: buffer to process
4019 * netif_receive_skb() is the main receive data processing function.
4020 * It always succeeds. The buffer may be dropped during processing
4021 * for congestion control or by the protocol layers.
4023 * This function may only be called from softirq context and interrupts
4024 * should be enabled.
4026 * Return values (usually ignored):
4027 * NET_RX_SUCCESS: no congestion
4028 * NET_RX_DROP: packet was dropped
4030 int netif_receive_skb(struct sk_buff *skb)
4032 trace_netif_receive_skb_entry(skb);
4034 return netif_receive_skb_internal(skb);
4036 EXPORT_SYMBOL(netif_receive_skb);
4038 /* Network device is going away, flush any packets still pending
4039 * Called with irqs disabled.
4041 static void flush_backlog(void *arg)
4043 struct net_device *dev = arg;
4044 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4045 struct sk_buff *skb, *tmp;
4048 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4049 if (skb->dev == dev) {
4050 __skb_unlink(skb, &sd->input_pkt_queue);
4052 input_queue_head_incr(sd);
4057 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4058 if (skb->dev == dev) {
4059 __skb_unlink(skb, &sd->process_queue);
4061 input_queue_head_incr(sd);
4066 static int napi_gro_complete(struct sk_buff *skb)
4068 struct packet_offload *ptype;
4069 __be16 type = skb->protocol;
4070 struct list_head *head = &offload_base;
4073 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4075 if (NAPI_GRO_CB(skb)->count == 1) {
4076 skb_shinfo(skb)->gso_size = 0;
4081 list_for_each_entry_rcu(ptype, head, list) {
4082 if (ptype->type != type || !ptype->callbacks.gro_complete)
4085 err = ptype->callbacks.gro_complete(skb, 0);
4091 WARN_ON(&ptype->list == head);
4093 return NET_RX_SUCCESS;
4097 return netif_receive_skb_internal(skb);
4100 /* napi->gro_list contains packets ordered by age.
4101 * youngest packets at the head of it.
4102 * Complete skbs in reverse order to reduce latencies.
4104 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4106 struct sk_buff *skb, *prev = NULL;
4108 /* scan list and build reverse chain */
4109 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4114 for (skb = prev; skb; skb = prev) {
4117 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4121 napi_gro_complete(skb);
4125 napi->gro_list = NULL;
4127 EXPORT_SYMBOL(napi_gro_flush);
4129 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4132 unsigned int maclen = skb->dev->hard_header_len;
4133 u32 hash = skb_get_hash_raw(skb);
4135 for (p = napi->gro_list; p; p = p->next) {
4136 unsigned long diffs;
4138 NAPI_GRO_CB(p)->flush = 0;
4140 if (hash != skb_get_hash_raw(p)) {
4141 NAPI_GRO_CB(p)->same_flow = 0;
4145 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4146 diffs |= p->vlan_tci ^ skb->vlan_tci;
4147 if (maclen == ETH_HLEN)
4148 diffs |= compare_ether_header(skb_mac_header(p),
4149 skb_mac_header(skb));
4151 diffs = memcmp(skb_mac_header(p),
4152 skb_mac_header(skb),
4154 NAPI_GRO_CB(p)->same_flow = !diffs;
4158 static void skb_gro_reset_offset(struct sk_buff *skb)
4160 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4161 const skb_frag_t *frag0 = &pinfo->frags[0];
4163 NAPI_GRO_CB(skb)->data_offset = 0;
4164 NAPI_GRO_CB(skb)->frag0 = NULL;
4165 NAPI_GRO_CB(skb)->frag0_len = 0;
4167 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4169 !PageHighMem(skb_frag_page(frag0))) {
4170 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4171 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4175 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4177 struct skb_shared_info *pinfo = skb_shinfo(skb);
4179 BUG_ON(skb->end - skb->tail < grow);
4181 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4183 skb->data_len -= grow;
4186 pinfo->frags[0].page_offset += grow;
4187 skb_frag_size_sub(&pinfo->frags[0], grow);
4189 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4190 skb_frag_unref(skb, 0);
4191 memmove(pinfo->frags, pinfo->frags + 1,
4192 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4196 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4198 struct sk_buff **pp = NULL;
4199 struct packet_offload *ptype;
4200 __be16 type = skb->protocol;
4201 struct list_head *head = &offload_base;
4203 enum gro_result ret;
4206 if (!(skb->dev->features & NETIF_F_GRO))
4209 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4212 gro_list_prepare(napi, skb);
4215 list_for_each_entry_rcu(ptype, head, list) {
4216 if (ptype->type != type || !ptype->callbacks.gro_receive)
4219 skb_set_network_header(skb, skb_gro_offset(skb));
4220 skb_reset_mac_len(skb);
4221 NAPI_GRO_CB(skb)->same_flow = 0;
4222 NAPI_GRO_CB(skb)->flush = 0;
4223 NAPI_GRO_CB(skb)->free = 0;
4224 NAPI_GRO_CB(skb)->udp_mark = 0;
4225 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4227 /* Setup for GRO checksum validation */
4228 switch (skb->ip_summed) {
4229 case CHECKSUM_COMPLETE:
4230 NAPI_GRO_CB(skb)->csum = skb->csum;
4231 NAPI_GRO_CB(skb)->csum_valid = 1;
4232 NAPI_GRO_CB(skb)->csum_cnt = 0;
4234 case CHECKSUM_UNNECESSARY:
4235 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4236 NAPI_GRO_CB(skb)->csum_valid = 0;
4239 NAPI_GRO_CB(skb)->csum_cnt = 0;
4240 NAPI_GRO_CB(skb)->csum_valid = 0;
4243 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4248 if (&ptype->list == head)
4251 same_flow = NAPI_GRO_CB(skb)->same_flow;
4252 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4255 struct sk_buff *nskb = *pp;
4259 napi_gro_complete(nskb);
4266 if (NAPI_GRO_CB(skb)->flush)
4269 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4270 struct sk_buff *nskb = napi->gro_list;
4272 /* locate the end of the list to select the 'oldest' flow */
4273 while (nskb->next) {
4279 napi_gro_complete(nskb);
4283 NAPI_GRO_CB(skb)->count = 1;
4284 NAPI_GRO_CB(skb)->age = jiffies;
4285 NAPI_GRO_CB(skb)->last = skb;
4286 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4287 skb->next = napi->gro_list;
4288 napi->gro_list = skb;
4292 grow = skb_gro_offset(skb) - skb_headlen(skb);
4294 gro_pull_from_frag0(skb, grow);
4303 struct packet_offload *gro_find_receive_by_type(__be16 type)
4305 struct list_head *offload_head = &offload_base;
4306 struct packet_offload *ptype;
4308 list_for_each_entry_rcu(ptype, offload_head, list) {
4309 if (ptype->type != type || !ptype->callbacks.gro_receive)
4315 EXPORT_SYMBOL(gro_find_receive_by_type);
4317 struct packet_offload *gro_find_complete_by_type(__be16 type)
4319 struct list_head *offload_head = &offload_base;
4320 struct packet_offload *ptype;
4322 list_for_each_entry_rcu(ptype, offload_head, list) {
4323 if (ptype->type != type || !ptype->callbacks.gro_complete)
4329 EXPORT_SYMBOL(gro_find_complete_by_type);
4331 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4335 if (netif_receive_skb_internal(skb))
4343 case GRO_MERGED_FREE:
4344 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4345 kmem_cache_free(skbuff_head_cache, skb);
4358 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4360 skb_mark_napi_id(skb, napi);
4361 trace_napi_gro_receive_entry(skb);
4363 skb_gro_reset_offset(skb);
4365 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4367 EXPORT_SYMBOL(napi_gro_receive);
4369 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4371 if (unlikely(skb->pfmemalloc)) {
4375 __skb_pull(skb, skb_headlen(skb));
4376 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4377 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4379 skb->dev = napi->dev;
4381 skb->encapsulation = 0;
4382 skb_shinfo(skb)->gso_type = 0;
4383 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4388 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4390 struct sk_buff *skb = napi->skb;
4393 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4396 skb_mark_napi_id(skb, napi);
4401 EXPORT_SYMBOL(napi_get_frags);
4403 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4404 struct sk_buff *skb,
4410 __skb_push(skb, ETH_HLEN);
4411 skb->protocol = eth_type_trans(skb, skb->dev);
4412 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4417 case GRO_MERGED_FREE:
4418 napi_reuse_skb(napi, skb);
4428 /* Upper GRO stack assumes network header starts at gro_offset=0
4429 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4430 * We copy ethernet header into skb->data to have a common layout.
4432 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4434 struct sk_buff *skb = napi->skb;
4435 const struct ethhdr *eth;
4436 unsigned int hlen = sizeof(*eth);
4440 skb_reset_mac_header(skb);
4441 skb_gro_reset_offset(skb);
4443 eth = skb_gro_header_fast(skb, 0);
4444 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4445 eth = skb_gro_header_slow(skb, hlen, 0);
4446 if (unlikely(!eth)) {
4447 napi_reuse_skb(napi, skb);
4451 gro_pull_from_frag0(skb, hlen);
4452 NAPI_GRO_CB(skb)->frag0 += hlen;
4453 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4455 __skb_pull(skb, hlen);
4458 * This works because the only protocols we care about don't require
4460 * We'll fix it up properly in napi_frags_finish()
4462 skb->protocol = eth->h_proto;
4467 gro_result_t napi_gro_frags(struct napi_struct *napi)
4469 struct sk_buff *skb = napi_frags_skb(napi);
4474 trace_napi_gro_frags_entry(skb);
4476 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4478 EXPORT_SYMBOL(napi_gro_frags);
4480 /* Compute the checksum from gro_offset and return the folded value
4481 * after adding in any pseudo checksum.
4483 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4488 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4490 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4491 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4493 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4494 !skb->csum_complete_sw)
4495 netdev_rx_csum_fault(skb->dev);
4498 NAPI_GRO_CB(skb)->csum = wsum;
4499 NAPI_GRO_CB(skb)->csum_valid = 1;
4503 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4506 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4507 * Note: called with local irq disabled, but exits with local irq enabled.
4509 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4512 struct softnet_data *remsd = sd->rps_ipi_list;
4515 sd->rps_ipi_list = NULL;
4519 /* Send pending IPI's to kick RPS processing on remote cpus. */
4521 struct softnet_data *next = remsd->rps_ipi_next;
4523 if (cpu_online(remsd->cpu))
4524 smp_call_function_single_async(remsd->cpu,
4533 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4536 return sd->rps_ipi_list != NULL;
4542 static int process_backlog(struct napi_struct *napi, int quota)
4545 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4547 /* Check if we have pending ipi, its better to send them now,
4548 * not waiting net_rx_action() end.
4550 if (sd_has_rps_ipi_waiting(sd)) {
4551 local_irq_disable();
4552 net_rps_action_and_irq_enable(sd);
4555 napi->weight = weight_p;
4556 local_irq_disable();
4558 struct sk_buff *skb;
4560 while ((skb = __skb_dequeue(&sd->process_queue))) {
4563 __netif_receive_skb(skb);
4565 local_irq_disable();
4566 input_queue_head_incr(sd);
4567 if (++work >= quota) {
4574 if (skb_queue_empty(&sd->input_pkt_queue)) {
4576 * Inline a custom version of __napi_complete().
4577 * only current cpu owns and manipulates this napi,
4578 * and NAPI_STATE_SCHED is the only possible flag set
4580 * We can use a plain write instead of clear_bit(),
4581 * and we dont need an smp_mb() memory barrier.
4589 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4590 &sd->process_queue);
4599 * __napi_schedule - schedule for receive
4600 * @n: entry to schedule
4602 * The entry's receive function will be scheduled to run.
4603 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4605 void __napi_schedule(struct napi_struct *n)
4607 unsigned long flags;
4609 local_irq_save(flags);
4610 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4611 local_irq_restore(flags);
4613 EXPORT_SYMBOL(__napi_schedule);
4616 * __napi_schedule_irqoff - schedule for receive
4617 * @n: entry to schedule
4619 * Variant of __napi_schedule() assuming hard irqs are masked
4621 void __napi_schedule_irqoff(struct napi_struct *n)
4623 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4625 EXPORT_SYMBOL(__napi_schedule_irqoff);
4627 void __napi_complete(struct napi_struct *n)
4629 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4631 list_del_init(&n->poll_list);
4632 smp_mb__before_atomic();
4633 clear_bit(NAPI_STATE_SCHED, &n->state);
4635 EXPORT_SYMBOL(__napi_complete);
4637 void napi_complete_done(struct napi_struct *n, int work_done)
4639 unsigned long flags;
4642 * don't let napi dequeue from the cpu poll list
4643 * just in case its running on a different cpu
4645 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4649 unsigned long timeout = 0;
4652 timeout = n->dev->gro_flush_timeout;
4655 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4656 HRTIMER_MODE_REL_PINNED);
4658 napi_gro_flush(n, false);
4660 if (likely(list_empty(&n->poll_list))) {
4661 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4663 /* If n->poll_list is not empty, we need to mask irqs */
4664 local_irq_save(flags);
4666 local_irq_restore(flags);
4669 EXPORT_SYMBOL(napi_complete_done);
4671 /* must be called under rcu_read_lock(), as we dont take a reference */
4672 static struct napi_struct *napi_by_id(unsigned int napi_id)
4674 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4675 struct napi_struct *napi;
4677 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4678 if (napi->napi_id == napi_id)
4684 #if defined(CONFIG_NET_RX_BUSY_POLL)
4685 #define BUSY_POLL_BUDGET 8
4686 bool sk_busy_loop(struct sock *sk, int nonblock)
4688 unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0;
4689 int (*busy_poll)(struct napi_struct *dev);
4690 struct napi_struct *napi;
4695 napi = napi_by_id(sk->sk_napi_id);
4699 /* Note: ndo_busy_poll method is optional in linux-4.5 */
4700 busy_poll = napi->dev->netdev_ops->ndo_busy_poll;
4706 rc = busy_poll(napi);
4707 } else if (napi_schedule_prep(napi)) {
4708 void *have = netpoll_poll_lock(napi);
4710 if (test_bit(NAPI_STATE_SCHED, &napi->state)) {
4711 rc = napi->poll(napi, BUSY_POLL_BUDGET);
4712 trace_napi_poll(napi);
4713 if (rc == BUSY_POLL_BUDGET) {
4714 napi_complete_done(napi, rc);
4715 napi_schedule(napi);
4718 netpoll_poll_unlock(have);
4721 NET_ADD_STATS_BH(sock_net(sk),
4722 LINUX_MIB_BUSYPOLLRXPACKETS, rc);
4725 if (rc == LL_FLUSH_FAILED)
4726 break; /* permanent failure */
4729 } while (!nonblock && skb_queue_empty(&sk->sk_receive_queue) &&
4730 !need_resched() && !busy_loop_timeout(end_time));
4732 rc = !skb_queue_empty(&sk->sk_receive_queue);
4737 EXPORT_SYMBOL(sk_busy_loop);
4739 #endif /* CONFIG_NET_RX_BUSY_POLL */
4741 void napi_hash_add(struct napi_struct *napi)
4743 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
4744 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
4747 spin_lock(&napi_hash_lock);
4749 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */
4751 if (unlikely(++napi_gen_id < NR_CPUS + 1))
4752 napi_gen_id = NR_CPUS + 1;
4753 } while (napi_by_id(napi_gen_id));
4754 napi->napi_id = napi_gen_id;
4756 hlist_add_head_rcu(&napi->napi_hash_node,
4757 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4759 spin_unlock(&napi_hash_lock);
4761 EXPORT_SYMBOL_GPL(napi_hash_add);
4763 /* Warning : caller is responsible to make sure rcu grace period
4764 * is respected before freeing memory containing @napi
4766 bool napi_hash_del(struct napi_struct *napi)
4768 bool rcu_sync_needed = false;
4770 spin_lock(&napi_hash_lock);
4772 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
4773 rcu_sync_needed = true;
4774 hlist_del_rcu(&napi->napi_hash_node);
4776 spin_unlock(&napi_hash_lock);
4777 return rcu_sync_needed;
4779 EXPORT_SYMBOL_GPL(napi_hash_del);
4781 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4783 struct napi_struct *napi;
4785 napi = container_of(timer, struct napi_struct, timer);
4787 napi_schedule(napi);
4789 return HRTIMER_NORESTART;
4792 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4793 int (*poll)(struct napi_struct *, int), int weight)
4795 INIT_LIST_HEAD(&napi->poll_list);
4796 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4797 napi->timer.function = napi_watchdog;
4798 napi->gro_count = 0;
4799 napi->gro_list = NULL;
4802 if (weight > NAPI_POLL_WEIGHT)
4803 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4805 napi->weight = weight;
4806 list_add(&napi->dev_list, &dev->napi_list);
4808 #ifdef CONFIG_NETPOLL
4809 spin_lock_init(&napi->poll_lock);
4810 napi->poll_owner = -1;
4812 set_bit(NAPI_STATE_SCHED, &napi->state);
4813 napi_hash_add(napi);
4815 EXPORT_SYMBOL(netif_napi_add);
4817 void napi_disable(struct napi_struct *n)
4820 set_bit(NAPI_STATE_DISABLE, &n->state);
4822 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4824 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
4827 hrtimer_cancel(&n->timer);
4829 clear_bit(NAPI_STATE_DISABLE, &n->state);
4831 EXPORT_SYMBOL(napi_disable);
4833 /* Must be called in process context */
4834 void netif_napi_del(struct napi_struct *napi)
4837 if (napi_hash_del(napi))
4839 list_del_init(&napi->dev_list);
4840 napi_free_frags(napi);
4842 kfree_skb_list(napi->gro_list);
4843 napi->gro_list = NULL;
4844 napi->gro_count = 0;
4846 EXPORT_SYMBOL(netif_napi_del);
4848 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4853 list_del_init(&n->poll_list);
4855 have = netpoll_poll_lock(n);
4859 /* This NAPI_STATE_SCHED test is for avoiding a race
4860 * with netpoll's poll_napi(). Only the entity which
4861 * obtains the lock and sees NAPI_STATE_SCHED set will
4862 * actually make the ->poll() call. Therefore we avoid
4863 * accidentally calling ->poll() when NAPI is not scheduled.
4866 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4867 work = n->poll(n, weight);
4871 WARN_ON_ONCE(work > weight);
4873 if (likely(work < weight))
4876 /* Drivers must not modify the NAPI state if they
4877 * consume the entire weight. In such cases this code
4878 * still "owns" the NAPI instance and therefore can
4879 * move the instance around on the list at-will.
4881 if (unlikely(napi_disable_pending(n))) {
4887 /* flush too old packets
4888 * If HZ < 1000, flush all packets.
4890 napi_gro_flush(n, HZ >= 1000);
4893 /* Some drivers may have called napi_schedule
4894 * prior to exhausting their budget.
4896 if (unlikely(!list_empty(&n->poll_list))) {
4897 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4898 n->dev ? n->dev->name : "backlog");
4902 list_add_tail(&n->poll_list, repoll);
4905 netpoll_poll_unlock(have);
4910 static void net_rx_action(struct softirq_action *h)
4912 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4913 unsigned long time_limit = jiffies + 2;
4914 int budget = netdev_budget;
4918 local_irq_disable();
4919 list_splice_init(&sd->poll_list, &list);
4923 struct napi_struct *n;
4925 if (list_empty(&list)) {
4926 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4931 n = list_first_entry(&list, struct napi_struct, poll_list);
4932 budget -= napi_poll(n, &repoll);
4934 /* If softirq window is exhausted then punt.
4935 * Allow this to run for 2 jiffies since which will allow
4936 * an average latency of 1.5/HZ.
4938 if (unlikely(budget <= 0 ||
4939 time_after_eq(jiffies, time_limit))) {
4945 local_irq_disable();
4947 list_splice_tail_init(&sd->poll_list, &list);
4948 list_splice_tail(&repoll, &list);
4949 list_splice(&list, &sd->poll_list);
4950 if (!list_empty(&sd->poll_list))
4951 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4953 net_rps_action_and_irq_enable(sd);
4956 struct netdev_adjacent {
4957 struct net_device *dev;
4959 /* upper master flag, there can only be one master device per list */
4962 /* counter for the number of times this device was added to us */
4965 /* private field for the users */
4968 struct list_head list;
4969 struct rcu_head rcu;
4972 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
4973 struct list_head *adj_list)
4975 struct netdev_adjacent *adj;
4977 list_for_each_entry(adj, adj_list, list) {
4978 if (adj->dev == adj_dev)
4985 * netdev_has_upper_dev - Check if device is linked to an upper device
4987 * @upper_dev: upper device to check
4989 * Find out if a device is linked to specified upper device and return true
4990 * in case it is. Note that this checks only immediate upper device,
4991 * not through a complete stack of devices. The caller must hold the RTNL lock.
4993 bool netdev_has_upper_dev(struct net_device *dev,
4994 struct net_device *upper_dev)
4998 return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper);
5000 EXPORT_SYMBOL(netdev_has_upper_dev);
5003 * netdev_has_any_upper_dev - Check if device is linked to some device
5006 * Find out if a device is linked to an upper device and return true in case
5007 * it is. The caller must hold the RTNL lock.
5009 static bool netdev_has_any_upper_dev(struct net_device *dev)
5013 return !list_empty(&dev->all_adj_list.upper);
5017 * netdev_master_upper_dev_get - Get master upper device
5020 * Find a master upper device and return pointer to it or NULL in case
5021 * it's not there. The caller must hold the RTNL lock.
5023 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5025 struct netdev_adjacent *upper;
5029 if (list_empty(&dev->adj_list.upper))
5032 upper = list_first_entry(&dev->adj_list.upper,
5033 struct netdev_adjacent, list);
5034 if (likely(upper->master))
5038 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5040 void *netdev_adjacent_get_private(struct list_head *adj_list)
5042 struct netdev_adjacent *adj;
5044 adj = list_entry(adj_list, struct netdev_adjacent, list);
5046 return adj->private;
5048 EXPORT_SYMBOL(netdev_adjacent_get_private);
5051 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5053 * @iter: list_head ** of the current position
5055 * Gets the next device from the dev's upper list, starting from iter
5056 * position. The caller must hold RCU read lock.
5058 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5059 struct list_head **iter)
5061 struct netdev_adjacent *upper;
5063 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5065 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5067 if (&upper->list == &dev->adj_list.upper)
5070 *iter = &upper->list;
5074 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5077 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
5079 * @iter: list_head ** of the current position
5081 * Gets the next device from the dev's upper list, starting from iter
5082 * position. The caller must hold RCU read lock.
5084 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
5085 struct list_head **iter)
5087 struct netdev_adjacent *upper;
5089 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5091 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5093 if (&upper->list == &dev->all_adj_list.upper)
5096 *iter = &upper->list;
5100 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
5103 * netdev_lower_get_next_private - Get the next ->private from the
5104 * lower neighbour list
5106 * @iter: list_head ** of the current position
5108 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5109 * list, starting from iter position. The caller must hold either hold the
5110 * RTNL lock or its own locking that guarantees that the neighbour lower
5111 * list will remain unchanged.
5113 void *netdev_lower_get_next_private(struct net_device *dev,
5114 struct list_head **iter)
5116 struct netdev_adjacent *lower;
5118 lower = list_entry(*iter, struct netdev_adjacent, list);
5120 if (&lower->list == &dev->adj_list.lower)
5123 *iter = lower->list.next;
5125 return lower->private;
5127 EXPORT_SYMBOL(netdev_lower_get_next_private);
5130 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5131 * lower neighbour list, RCU
5134 * @iter: list_head ** of the current position
5136 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5137 * list, starting from iter position. The caller must hold RCU read lock.
5139 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5140 struct list_head **iter)
5142 struct netdev_adjacent *lower;
5144 WARN_ON_ONCE(!rcu_read_lock_held());
5146 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5148 if (&lower->list == &dev->adj_list.lower)
5151 *iter = &lower->list;
5153 return lower->private;
5155 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5158 * netdev_lower_get_next - Get the next device from the lower neighbour
5161 * @iter: list_head ** of the current position
5163 * Gets the next netdev_adjacent from the dev's lower neighbour
5164 * list, starting from iter position. The caller must hold RTNL lock or
5165 * its own locking that guarantees that the neighbour lower
5166 * list will remain unchanged.
5168 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5170 struct netdev_adjacent *lower;
5172 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5174 if (&lower->list == &dev->adj_list.lower)
5177 *iter = &lower->list;
5181 EXPORT_SYMBOL(netdev_lower_get_next);
5184 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5185 * lower neighbour list, RCU
5189 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5190 * list. The caller must hold RCU read lock.
5192 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5194 struct netdev_adjacent *lower;
5196 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5197 struct netdev_adjacent, list);
5199 return lower->private;
5202 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5205 * netdev_master_upper_dev_get_rcu - Get master upper device
5208 * Find a master upper device and return pointer to it or NULL in case
5209 * it's not there. The caller must hold the RCU read lock.
5211 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5213 struct netdev_adjacent *upper;
5215 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5216 struct netdev_adjacent, list);
5217 if (upper && likely(upper->master))
5221 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5223 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5224 struct net_device *adj_dev,
5225 struct list_head *dev_list)
5227 char linkname[IFNAMSIZ+7];
5228 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5229 "upper_%s" : "lower_%s", adj_dev->name);
5230 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5233 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5235 struct list_head *dev_list)
5237 char linkname[IFNAMSIZ+7];
5238 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5239 "upper_%s" : "lower_%s", name);
5240 sysfs_remove_link(&(dev->dev.kobj), linkname);
5243 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5244 struct net_device *adj_dev,
5245 struct list_head *dev_list)
5247 return (dev_list == &dev->adj_list.upper ||
5248 dev_list == &dev->adj_list.lower) &&
5249 net_eq(dev_net(dev), dev_net(adj_dev));
5252 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5253 struct net_device *adj_dev,
5254 struct list_head *dev_list,
5255 void *private, bool master)
5257 struct netdev_adjacent *adj;
5260 adj = __netdev_find_adj(adj_dev, dev_list);
5267 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5272 adj->master = master;
5274 adj->private = private;
5277 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5278 adj_dev->name, dev->name, adj_dev->name);
5280 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5281 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5286 /* Ensure that master link is always the first item in list. */
5288 ret = sysfs_create_link(&(dev->dev.kobj),
5289 &(adj_dev->dev.kobj), "master");
5291 goto remove_symlinks;
5293 list_add_rcu(&adj->list, dev_list);
5295 list_add_tail_rcu(&adj->list, dev_list);
5301 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5302 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5310 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5311 struct net_device *adj_dev,
5312 struct list_head *dev_list)
5314 struct netdev_adjacent *adj;
5316 adj = __netdev_find_adj(adj_dev, dev_list);
5319 pr_err("tried to remove device %s from %s\n",
5320 dev->name, adj_dev->name);
5324 if (adj->ref_nr > 1) {
5325 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5332 sysfs_remove_link(&(dev->dev.kobj), "master");
5334 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5335 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5337 list_del_rcu(&adj->list);
5338 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5339 adj_dev->name, dev->name, adj_dev->name);
5341 kfree_rcu(adj, rcu);
5344 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5345 struct net_device *upper_dev,
5346 struct list_head *up_list,
5347 struct list_head *down_list,
5348 void *private, bool master)
5352 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5357 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5360 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5367 static int __netdev_adjacent_dev_link(struct net_device *dev,
5368 struct net_device *upper_dev)
5370 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5371 &dev->all_adj_list.upper,
5372 &upper_dev->all_adj_list.lower,
5376 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5377 struct net_device *upper_dev,
5378 struct list_head *up_list,
5379 struct list_head *down_list)
5381 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5382 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5385 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5386 struct net_device *upper_dev)
5388 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5389 &dev->all_adj_list.upper,
5390 &upper_dev->all_adj_list.lower);
5393 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5394 struct net_device *upper_dev,
5395 void *private, bool master)
5397 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5402 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5403 &dev->adj_list.upper,
5404 &upper_dev->adj_list.lower,
5407 __netdev_adjacent_dev_unlink(dev, upper_dev);
5414 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5415 struct net_device *upper_dev)
5417 __netdev_adjacent_dev_unlink(dev, upper_dev);
5418 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5419 &dev->adj_list.upper,
5420 &upper_dev->adj_list.lower);
5423 static int __netdev_upper_dev_link(struct net_device *dev,
5424 struct net_device *upper_dev, bool master,
5425 void *upper_priv, void *upper_info)
5427 struct netdev_notifier_changeupper_info changeupper_info;
5428 struct netdev_adjacent *i, *j, *to_i, *to_j;
5433 if (dev == upper_dev)
5436 /* To prevent loops, check if dev is not upper device to upper_dev. */
5437 if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper))
5440 if (__netdev_find_adj(upper_dev, &dev->adj_list.upper))
5443 if (master && netdev_master_upper_dev_get(dev))
5446 changeupper_info.upper_dev = upper_dev;
5447 changeupper_info.master = master;
5448 changeupper_info.linking = true;
5449 changeupper_info.upper_info = upper_info;
5451 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5452 &changeupper_info.info);
5453 ret = notifier_to_errno(ret);
5457 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
5462 /* Now that we linked these devs, make all the upper_dev's
5463 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5464 * versa, and don't forget the devices itself. All of these
5465 * links are non-neighbours.
5467 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5468 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5469 pr_debug("Interlinking %s with %s, non-neighbour\n",
5470 i->dev->name, j->dev->name);
5471 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5477 /* add dev to every upper_dev's upper device */
5478 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5479 pr_debug("linking %s's upper device %s with %s\n",
5480 upper_dev->name, i->dev->name, dev->name);
5481 ret = __netdev_adjacent_dev_link(dev, i->dev);
5483 goto rollback_upper_mesh;
5486 /* add upper_dev to every dev's lower device */
5487 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5488 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5489 i->dev->name, upper_dev->name);
5490 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5492 goto rollback_lower_mesh;
5495 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5496 &changeupper_info.info);
5497 ret = notifier_to_errno(ret);
5499 goto rollback_lower_mesh;
5503 rollback_lower_mesh:
5505 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5508 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5513 rollback_upper_mesh:
5515 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5518 __netdev_adjacent_dev_unlink(dev, i->dev);
5526 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5527 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5528 if (i == to_i && j == to_j)
5530 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5536 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5542 * netdev_upper_dev_link - Add a link to the upper device
5544 * @upper_dev: new upper device
5546 * Adds a link to device which is upper to this one. The caller must hold
5547 * the RTNL lock. On a failure a negative errno code is returned.
5548 * On success the reference counts are adjusted and the function
5551 int netdev_upper_dev_link(struct net_device *dev,
5552 struct net_device *upper_dev)
5554 return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL);
5556 EXPORT_SYMBOL(netdev_upper_dev_link);
5559 * netdev_master_upper_dev_link - Add a master link to the upper device
5561 * @upper_dev: new upper device
5562 * @upper_priv: upper device private
5563 * @upper_info: upper info to be passed down via notifier
5565 * Adds a link to device which is upper to this one. In this case, only
5566 * one master upper device can be linked, although other non-master devices
5567 * might be linked as well. The caller must hold the RTNL lock.
5568 * On a failure a negative errno code is returned. On success the reference
5569 * counts are adjusted and the function returns zero.
5571 int netdev_master_upper_dev_link(struct net_device *dev,
5572 struct net_device *upper_dev,
5573 void *upper_priv, void *upper_info)
5575 return __netdev_upper_dev_link(dev, upper_dev, true,
5576 upper_priv, upper_info);
5578 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5581 * netdev_upper_dev_unlink - Removes a link to upper device
5583 * @upper_dev: new upper device
5585 * Removes a link to device which is upper to this one. The caller must hold
5588 void netdev_upper_dev_unlink(struct net_device *dev,
5589 struct net_device *upper_dev)
5591 struct netdev_notifier_changeupper_info changeupper_info;
5592 struct netdev_adjacent *i, *j;
5595 changeupper_info.upper_dev = upper_dev;
5596 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
5597 changeupper_info.linking = false;
5599 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5600 &changeupper_info.info);
5602 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5604 /* Here is the tricky part. We must remove all dev's lower
5605 * devices from all upper_dev's upper devices and vice
5606 * versa, to maintain the graph relationship.
5608 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5609 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5610 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5612 /* remove also the devices itself from lower/upper device
5615 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5616 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5618 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5619 __netdev_adjacent_dev_unlink(dev, i->dev);
5621 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5622 &changeupper_info.info);
5624 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5627 * netdev_bonding_info_change - Dispatch event about slave change
5629 * @bonding_info: info to dispatch
5631 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5632 * The caller must hold the RTNL lock.
5634 void netdev_bonding_info_change(struct net_device *dev,
5635 struct netdev_bonding_info *bonding_info)
5637 struct netdev_notifier_bonding_info info;
5639 memcpy(&info.bonding_info, bonding_info,
5640 sizeof(struct netdev_bonding_info));
5641 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5644 EXPORT_SYMBOL(netdev_bonding_info_change);
5646 static void netdev_adjacent_add_links(struct net_device *dev)
5648 struct netdev_adjacent *iter;
5650 struct net *net = dev_net(dev);
5652 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5653 if (!net_eq(net,dev_net(iter->dev)))
5655 netdev_adjacent_sysfs_add(iter->dev, dev,
5656 &iter->dev->adj_list.lower);
5657 netdev_adjacent_sysfs_add(dev, iter->dev,
5658 &dev->adj_list.upper);
5661 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5662 if (!net_eq(net,dev_net(iter->dev)))
5664 netdev_adjacent_sysfs_add(iter->dev, dev,
5665 &iter->dev->adj_list.upper);
5666 netdev_adjacent_sysfs_add(dev, iter->dev,
5667 &dev->adj_list.lower);
5671 static void netdev_adjacent_del_links(struct net_device *dev)
5673 struct netdev_adjacent *iter;
5675 struct net *net = dev_net(dev);
5677 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5678 if (!net_eq(net,dev_net(iter->dev)))
5680 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5681 &iter->dev->adj_list.lower);
5682 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5683 &dev->adj_list.upper);
5686 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5687 if (!net_eq(net,dev_net(iter->dev)))
5689 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5690 &iter->dev->adj_list.upper);
5691 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5692 &dev->adj_list.lower);
5696 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5698 struct netdev_adjacent *iter;
5700 struct net *net = dev_net(dev);
5702 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5703 if (!net_eq(net,dev_net(iter->dev)))
5705 netdev_adjacent_sysfs_del(iter->dev, oldname,
5706 &iter->dev->adj_list.lower);
5707 netdev_adjacent_sysfs_add(iter->dev, dev,
5708 &iter->dev->adj_list.lower);
5711 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5712 if (!net_eq(net,dev_net(iter->dev)))
5714 netdev_adjacent_sysfs_del(iter->dev, oldname,
5715 &iter->dev->adj_list.upper);
5716 netdev_adjacent_sysfs_add(iter->dev, dev,
5717 &iter->dev->adj_list.upper);
5721 void *netdev_lower_dev_get_private(struct net_device *dev,
5722 struct net_device *lower_dev)
5724 struct netdev_adjacent *lower;
5728 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
5732 return lower->private;
5734 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5737 int dev_get_nest_level(struct net_device *dev,
5738 bool (*type_check)(const struct net_device *dev))
5740 struct net_device *lower = NULL;
5741 struct list_head *iter;
5747 netdev_for_each_lower_dev(dev, lower, iter) {
5748 nest = dev_get_nest_level(lower, type_check);
5749 if (max_nest < nest)
5753 if (type_check(dev))
5758 EXPORT_SYMBOL(dev_get_nest_level);
5761 * netdev_lower_change - Dispatch event about lower device state change
5762 * @lower_dev: device
5763 * @lower_state_info: state to dispatch
5765 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
5766 * The caller must hold the RTNL lock.
5768 void netdev_lower_state_changed(struct net_device *lower_dev,
5769 void *lower_state_info)
5771 struct netdev_notifier_changelowerstate_info changelowerstate_info;
5774 changelowerstate_info.lower_state_info = lower_state_info;
5775 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev,
5776 &changelowerstate_info.info);
5778 EXPORT_SYMBOL(netdev_lower_state_changed);
5780 static void dev_change_rx_flags(struct net_device *dev, int flags)
5782 const struct net_device_ops *ops = dev->netdev_ops;
5784 if (ops->ndo_change_rx_flags)
5785 ops->ndo_change_rx_flags(dev, flags);
5788 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5790 unsigned int old_flags = dev->flags;
5796 dev->flags |= IFF_PROMISC;
5797 dev->promiscuity += inc;
5798 if (dev->promiscuity == 0) {
5801 * If inc causes overflow, untouch promisc and return error.
5804 dev->flags &= ~IFF_PROMISC;
5806 dev->promiscuity -= inc;
5807 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5812 if (dev->flags != old_flags) {
5813 pr_info("device %s %s promiscuous mode\n",
5815 dev->flags & IFF_PROMISC ? "entered" : "left");
5816 if (audit_enabled) {
5817 current_uid_gid(&uid, &gid);
5818 audit_log(current->audit_context, GFP_ATOMIC,
5819 AUDIT_ANOM_PROMISCUOUS,
5820 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5821 dev->name, (dev->flags & IFF_PROMISC),
5822 (old_flags & IFF_PROMISC),
5823 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5824 from_kuid(&init_user_ns, uid),
5825 from_kgid(&init_user_ns, gid),
5826 audit_get_sessionid(current));
5829 dev_change_rx_flags(dev, IFF_PROMISC);
5832 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5837 * dev_set_promiscuity - update promiscuity count on a device
5841 * Add or remove promiscuity from a device. While the count in the device
5842 * remains above zero the interface remains promiscuous. Once it hits zero
5843 * the device reverts back to normal filtering operation. A negative inc
5844 * value is used to drop promiscuity on the device.
5845 * Return 0 if successful or a negative errno code on error.
5847 int dev_set_promiscuity(struct net_device *dev, int inc)
5849 unsigned int old_flags = dev->flags;
5852 err = __dev_set_promiscuity(dev, inc, true);
5855 if (dev->flags != old_flags)
5856 dev_set_rx_mode(dev);
5859 EXPORT_SYMBOL(dev_set_promiscuity);
5861 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5863 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5867 dev->flags |= IFF_ALLMULTI;
5868 dev->allmulti += inc;
5869 if (dev->allmulti == 0) {
5872 * If inc causes overflow, untouch allmulti and return error.
5875 dev->flags &= ~IFF_ALLMULTI;
5877 dev->allmulti -= inc;
5878 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5883 if (dev->flags ^ old_flags) {
5884 dev_change_rx_flags(dev, IFF_ALLMULTI);
5885 dev_set_rx_mode(dev);
5887 __dev_notify_flags(dev, old_flags,
5888 dev->gflags ^ old_gflags);
5894 * dev_set_allmulti - update allmulti count on a device
5898 * Add or remove reception of all multicast frames to a device. While the
5899 * count in the device remains above zero the interface remains listening
5900 * to all interfaces. Once it hits zero the device reverts back to normal
5901 * filtering operation. A negative @inc value is used to drop the counter
5902 * when releasing a resource needing all multicasts.
5903 * Return 0 if successful or a negative errno code on error.
5906 int dev_set_allmulti(struct net_device *dev, int inc)
5908 return __dev_set_allmulti(dev, inc, true);
5910 EXPORT_SYMBOL(dev_set_allmulti);
5913 * Upload unicast and multicast address lists to device and
5914 * configure RX filtering. When the device doesn't support unicast
5915 * filtering it is put in promiscuous mode while unicast addresses
5918 void __dev_set_rx_mode(struct net_device *dev)
5920 const struct net_device_ops *ops = dev->netdev_ops;
5922 /* dev_open will call this function so the list will stay sane. */
5923 if (!(dev->flags&IFF_UP))
5926 if (!netif_device_present(dev))
5929 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5930 /* Unicast addresses changes may only happen under the rtnl,
5931 * therefore calling __dev_set_promiscuity here is safe.
5933 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5934 __dev_set_promiscuity(dev, 1, false);
5935 dev->uc_promisc = true;
5936 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5937 __dev_set_promiscuity(dev, -1, false);
5938 dev->uc_promisc = false;
5942 if (ops->ndo_set_rx_mode)
5943 ops->ndo_set_rx_mode(dev);
5946 void dev_set_rx_mode(struct net_device *dev)
5948 netif_addr_lock_bh(dev);
5949 __dev_set_rx_mode(dev);
5950 netif_addr_unlock_bh(dev);
5954 * dev_get_flags - get flags reported to userspace
5957 * Get the combination of flag bits exported through APIs to userspace.
5959 unsigned int dev_get_flags(const struct net_device *dev)
5963 flags = (dev->flags & ~(IFF_PROMISC |
5968 (dev->gflags & (IFF_PROMISC |
5971 if (netif_running(dev)) {
5972 if (netif_oper_up(dev))
5973 flags |= IFF_RUNNING;
5974 if (netif_carrier_ok(dev))
5975 flags |= IFF_LOWER_UP;
5976 if (netif_dormant(dev))
5977 flags |= IFF_DORMANT;
5982 EXPORT_SYMBOL(dev_get_flags);
5984 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5986 unsigned int old_flags = dev->flags;
5992 * Set the flags on our device.
5995 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5996 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5998 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6002 * Load in the correct multicast list now the flags have changed.
6005 if ((old_flags ^ flags) & IFF_MULTICAST)
6006 dev_change_rx_flags(dev, IFF_MULTICAST);
6008 dev_set_rx_mode(dev);
6011 * Have we downed the interface. We handle IFF_UP ourselves
6012 * according to user attempts to set it, rather than blindly
6017 if ((old_flags ^ flags) & IFF_UP)
6018 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
6020 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6021 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6022 unsigned int old_flags = dev->flags;
6024 dev->gflags ^= IFF_PROMISC;
6026 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6027 if (dev->flags != old_flags)
6028 dev_set_rx_mode(dev);
6031 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6032 is important. Some (broken) drivers set IFF_PROMISC, when
6033 IFF_ALLMULTI is requested not asking us and not reporting.
6035 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6036 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6038 dev->gflags ^= IFF_ALLMULTI;
6039 __dev_set_allmulti(dev, inc, false);
6045 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6046 unsigned int gchanges)
6048 unsigned int changes = dev->flags ^ old_flags;
6051 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6053 if (changes & IFF_UP) {
6054 if (dev->flags & IFF_UP)
6055 call_netdevice_notifiers(NETDEV_UP, dev);
6057 call_netdevice_notifiers(NETDEV_DOWN, dev);
6060 if (dev->flags & IFF_UP &&
6061 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6062 struct netdev_notifier_change_info change_info;
6064 change_info.flags_changed = changes;
6065 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6071 * dev_change_flags - change device settings
6073 * @flags: device state flags
6075 * Change settings on device based state flags. The flags are
6076 * in the userspace exported format.
6078 int dev_change_flags(struct net_device *dev, unsigned int flags)
6081 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6083 ret = __dev_change_flags(dev, flags);
6087 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6088 __dev_notify_flags(dev, old_flags, changes);
6091 EXPORT_SYMBOL(dev_change_flags);
6093 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6095 const struct net_device_ops *ops = dev->netdev_ops;
6097 if (ops->ndo_change_mtu)
6098 return ops->ndo_change_mtu(dev, new_mtu);
6105 * dev_set_mtu - Change maximum transfer unit
6107 * @new_mtu: new transfer unit
6109 * Change the maximum transfer size of the network device.
6111 int dev_set_mtu(struct net_device *dev, int new_mtu)
6115 if (new_mtu == dev->mtu)
6118 /* MTU must be positive. */
6122 if (!netif_device_present(dev))
6125 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6126 err = notifier_to_errno(err);
6130 orig_mtu = dev->mtu;
6131 err = __dev_set_mtu(dev, new_mtu);
6134 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6135 err = notifier_to_errno(err);
6137 /* setting mtu back and notifying everyone again,
6138 * so that they have a chance to revert changes.
6140 __dev_set_mtu(dev, orig_mtu);
6141 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6146 EXPORT_SYMBOL(dev_set_mtu);
6149 * dev_set_group - Change group this device belongs to
6151 * @new_group: group this device should belong to
6153 void dev_set_group(struct net_device *dev, int new_group)
6155 dev->group = new_group;
6157 EXPORT_SYMBOL(dev_set_group);
6160 * dev_set_mac_address - Change Media Access Control Address
6164 * Change the hardware (MAC) address of the device
6166 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6168 const struct net_device_ops *ops = dev->netdev_ops;
6171 if (!ops->ndo_set_mac_address)
6173 if (sa->sa_family != dev->type)
6175 if (!netif_device_present(dev))
6177 err = ops->ndo_set_mac_address(dev, sa);
6180 dev->addr_assign_type = NET_ADDR_SET;
6181 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6182 add_device_randomness(dev->dev_addr, dev->addr_len);
6185 EXPORT_SYMBOL(dev_set_mac_address);
6188 * dev_change_carrier - Change device carrier
6190 * @new_carrier: new value
6192 * Change device carrier
6194 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6196 const struct net_device_ops *ops = dev->netdev_ops;
6198 if (!ops->ndo_change_carrier)
6200 if (!netif_device_present(dev))
6202 return ops->ndo_change_carrier(dev, new_carrier);
6204 EXPORT_SYMBOL(dev_change_carrier);
6207 * dev_get_phys_port_id - Get device physical port ID
6211 * Get device physical port ID
6213 int dev_get_phys_port_id(struct net_device *dev,
6214 struct netdev_phys_item_id *ppid)
6216 const struct net_device_ops *ops = dev->netdev_ops;
6218 if (!ops->ndo_get_phys_port_id)
6220 return ops->ndo_get_phys_port_id(dev, ppid);
6222 EXPORT_SYMBOL(dev_get_phys_port_id);
6225 * dev_get_phys_port_name - Get device physical port name
6229 * Get device physical port name
6231 int dev_get_phys_port_name(struct net_device *dev,
6232 char *name, size_t len)
6234 const struct net_device_ops *ops = dev->netdev_ops;
6236 if (!ops->ndo_get_phys_port_name)
6238 return ops->ndo_get_phys_port_name(dev, name, len);
6240 EXPORT_SYMBOL(dev_get_phys_port_name);
6243 * dev_change_proto_down - update protocol port state information
6245 * @proto_down: new value
6247 * This info can be used by switch drivers to set the phys state of the
6250 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6252 const struct net_device_ops *ops = dev->netdev_ops;
6254 if (!ops->ndo_change_proto_down)
6256 if (!netif_device_present(dev))
6258 return ops->ndo_change_proto_down(dev, proto_down);
6260 EXPORT_SYMBOL(dev_change_proto_down);
6263 * dev_new_index - allocate an ifindex
6264 * @net: the applicable net namespace
6266 * Returns a suitable unique value for a new device interface
6267 * number. The caller must hold the rtnl semaphore or the
6268 * dev_base_lock to be sure it remains unique.
6270 static int dev_new_index(struct net *net)
6272 int ifindex = net->ifindex;
6276 if (!__dev_get_by_index(net, ifindex))
6277 return net->ifindex = ifindex;
6281 /* Delayed registration/unregisteration */
6282 static LIST_HEAD(net_todo_list);
6283 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6285 static void net_set_todo(struct net_device *dev)
6287 list_add_tail(&dev->todo_list, &net_todo_list);
6288 dev_net(dev)->dev_unreg_count++;
6291 static void rollback_registered_many(struct list_head *head)
6293 struct net_device *dev, *tmp;
6294 LIST_HEAD(close_head);
6296 BUG_ON(dev_boot_phase);
6299 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6300 /* Some devices call without registering
6301 * for initialization unwind. Remove those
6302 * devices and proceed with the remaining.
6304 if (dev->reg_state == NETREG_UNINITIALIZED) {
6305 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6309 list_del(&dev->unreg_list);
6312 dev->dismantle = true;
6313 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6316 /* If device is running, close it first. */
6317 list_for_each_entry(dev, head, unreg_list)
6318 list_add_tail(&dev->close_list, &close_head);
6319 dev_close_many(&close_head, true);
6321 list_for_each_entry(dev, head, unreg_list) {
6322 /* And unlink it from device chain. */
6323 unlist_netdevice(dev);
6325 dev->reg_state = NETREG_UNREGISTERING;
6326 on_each_cpu(flush_backlog, dev, 1);
6331 list_for_each_entry(dev, head, unreg_list) {
6332 struct sk_buff *skb = NULL;
6334 /* Shutdown queueing discipline. */
6338 /* Notify protocols, that we are about to destroy
6339 this device. They should clean all the things.
6341 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6343 if (!dev->rtnl_link_ops ||
6344 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6345 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6349 * Flush the unicast and multicast chains
6354 if (dev->netdev_ops->ndo_uninit)
6355 dev->netdev_ops->ndo_uninit(dev);
6358 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6360 /* Notifier chain MUST detach us all upper devices. */
6361 WARN_ON(netdev_has_any_upper_dev(dev));
6363 /* Remove entries from kobject tree */
6364 netdev_unregister_kobject(dev);
6366 /* Remove XPS queueing entries */
6367 netif_reset_xps_queues_gt(dev, 0);
6373 list_for_each_entry(dev, head, unreg_list)
6377 static void rollback_registered(struct net_device *dev)
6381 list_add(&dev->unreg_list, &single);
6382 rollback_registered_many(&single);
6386 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6387 struct net_device *upper, netdev_features_t features)
6389 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6390 netdev_features_t feature;
6393 for_each_netdev_feature(&upper_disables, feature_bit) {
6394 feature = __NETIF_F_BIT(feature_bit);
6395 if (!(upper->wanted_features & feature)
6396 && (features & feature)) {
6397 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6398 &feature, upper->name);
6399 features &= ~feature;
6406 static void netdev_sync_lower_features(struct net_device *upper,
6407 struct net_device *lower, netdev_features_t features)
6409 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6410 netdev_features_t feature;
6413 for_each_netdev_feature(&upper_disables, feature_bit) {
6414 feature = __NETIF_F_BIT(feature_bit);
6415 if (!(features & feature) && (lower->features & feature)) {
6416 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6417 &feature, lower->name);
6418 lower->wanted_features &= ~feature;
6419 netdev_update_features(lower);
6421 if (unlikely(lower->features & feature))
6422 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6423 &feature, lower->name);
6428 static netdev_features_t netdev_fix_features(struct net_device *dev,
6429 netdev_features_t features)
6431 /* Fix illegal checksum combinations */
6432 if ((features & NETIF_F_HW_CSUM) &&
6433 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6434 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6435 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6438 /* TSO requires that SG is present as well. */
6439 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6440 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6441 features &= ~NETIF_F_ALL_TSO;
6444 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6445 !(features & NETIF_F_IP_CSUM)) {
6446 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6447 features &= ~NETIF_F_TSO;
6448 features &= ~NETIF_F_TSO_ECN;
6451 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6452 !(features & NETIF_F_IPV6_CSUM)) {
6453 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6454 features &= ~NETIF_F_TSO6;
6457 /* TSO ECN requires that TSO is present as well. */
6458 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6459 features &= ~NETIF_F_TSO_ECN;
6461 /* Software GSO depends on SG. */
6462 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6463 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6464 features &= ~NETIF_F_GSO;
6467 /* UFO needs SG and checksumming */
6468 if (features & NETIF_F_UFO) {
6469 /* maybe split UFO into V4 and V6? */
6470 if (!((features & NETIF_F_GEN_CSUM) ||
6471 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6472 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6474 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6475 features &= ~NETIF_F_UFO;
6478 if (!(features & NETIF_F_SG)) {
6480 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6481 features &= ~NETIF_F_UFO;
6485 #ifdef CONFIG_NET_RX_BUSY_POLL
6486 if (dev->netdev_ops->ndo_busy_poll)
6487 features |= NETIF_F_BUSY_POLL;
6490 features &= ~NETIF_F_BUSY_POLL;
6495 int __netdev_update_features(struct net_device *dev)
6497 struct net_device *upper, *lower;
6498 netdev_features_t features;
6499 struct list_head *iter;
6504 features = netdev_get_wanted_features(dev);
6506 if (dev->netdev_ops->ndo_fix_features)
6507 features = dev->netdev_ops->ndo_fix_features(dev, features);
6509 /* driver might be less strict about feature dependencies */
6510 features = netdev_fix_features(dev, features);
6512 /* some features can't be enabled if they're off an an upper device */
6513 netdev_for_each_upper_dev_rcu(dev, upper, iter)
6514 features = netdev_sync_upper_features(dev, upper, features);
6516 if (dev->features == features)
6519 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6520 &dev->features, &features);
6522 if (dev->netdev_ops->ndo_set_features)
6523 err = dev->netdev_ops->ndo_set_features(dev, features);
6527 if (unlikely(err < 0)) {
6529 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6530 err, &features, &dev->features);
6531 /* return non-0 since some features might have changed and
6532 * it's better to fire a spurious notification than miss it
6538 /* some features must be disabled on lower devices when disabled
6539 * on an upper device (think: bonding master or bridge)
6541 netdev_for_each_lower_dev(dev, lower, iter)
6542 netdev_sync_lower_features(dev, lower, features);
6545 dev->features = features;
6547 return err < 0 ? 0 : 1;
6551 * netdev_update_features - recalculate device features
6552 * @dev: the device to check
6554 * Recalculate dev->features set and send notifications if it
6555 * has changed. Should be called after driver or hardware dependent
6556 * conditions might have changed that influence the features.
6558 void netdev_update_features(struct net_device *dev)
6560 if (__netdev_update_features(dev))
6561 netdev_features_change(dev);
6563 EXPORT_SYMBOL(netdev_update_features);
6566 * netdev_change_features - recalculate device features
6567 * @dev: the device to check
6569 * Recalculate dev->features set and send notifications even
6570 * if they have not changed. Should be called instead of
6571 * netdev_update_features() if also dev->vlan_features might
6572 * have changed to allow the changes to be propagated to stacked
6575 void netdev_change_features(struct net_device *dev)
6577 __netdev_update_features(dev);
6578 netdev_features_change(dev);
6580 EXPORT_SYMBOL(netdev_change_features);
6583 * netif_stacked_transfer_operstate - transfer operstate
6584 * @rootdev: the root or lower level device to transfer state from
6585 * @dev: the device to transfer operstate to
6587 * Transfer operational state from root to device. This is normally
6588 * called when a stacking relationship exists between the root
6589 * device and the device(a leaf device).
6591 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6592 struct net_device *dev)
6594 if (rootdev->operstate == IF_OPER_DORMANT)
6595 netif_dormant_on(dev);
6597 netif_dormant_off(dev);
6599 if (netif_carrier_ok(rootdev)) {
6600 if (!netif_carrier_ok(dev))
6601 netif_carrier_on(dev);
6603 if (netif_carrier_ok(dev))
6604 netif_carrier_off(dev);
6607 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6610 static int netif_alloc_rx_queues(struct net_device *dev)
6612 unsigned int i, count = dev->num_rx_queues;
6613 struct netdev_rx_queue *rx;
6614 size_t sz = count * sizeof(*rx);
6618 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6626 for (i = 0; i < count; i++)
6632 static void netdev_init_one_queue(struct net_device *dev,
6633 struct netdev_queue *queue, void *_unused)
6635 /* Initialize queue lock */
6636 spin_lock_init(&queue->_xmit_lock);
6637 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6638 queue->xmit_lock_owner = -1;
6639 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6642 dql_init(&queue->dql, HZ);
6646 static void netif_free_tx_queues(struct net_device *dev)
6651 static int netif_alloc_netdev_queues(struct net_device *dev)
6653 unsigned int count = dev->num_tx_queues;
6654 struct netdev_queue *tx;
6655 size_t sz = count * sizeof(*tx);
6657 if (count < 1 || count > 0xffff)
6660 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6668 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6669 spin_lock_init(&dev->tx_global_lock);
6674 void netif_tx_stop_all_queues(struct net_device *dev)
6678 for (i = 0; i < dev->num_tx_queues; i++) {
6679 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
6680 netif_tx_stop_queue(txq);
6683 EXPORT_SYMBOL(netif_tx_stop_all_queues);
6686 * register_netdevice - register a network device
6687 * @dev: device to register
6689 * Take a completed network device structure and add it to the kernel
6690 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6691 * chain. 0 is returned on success. A negative errno code is returned
6692 * on a failure to set up the device, or if the name is a duplicate.
6694 * Callers must hold the rtnl semaphore. You may want
6695 * register_netdev() instead of this.
6698 * The locking appears insufficient to guarantee two parallel registers
6699 * will not get the same name.
6702 int register_netdevice(struct net_device *dev)
6705 struct net *net = dev_net(dev);
6707 BUG_ON(dev_boot_phase);
6712 /* When net_device's are persistent, this will be fatal. */
6713 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6716 spin_lock_init(&dev->addr_list_lock);
6717 netdev_set_addr_lockdep_class(dev);
6719 ret = dev_get_valid_name(net, dev, dev->name);
6723 /* Init, if this function is available */
6724 if (dev->netdev_ops->ndo_init) {
6725 ret = dev->netdev_ops->ndo_init(dev);
6733 if (((dev->hw_features | dev->features) &
6734 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6735 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6736 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6737 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6744 dev->ifindex = dev_new_index(net);
6745 else if (__dev_get_by_index(net, dev->ifindex))
6748 /* Transfer changeable features to wanted_features and enable
6749 * software offloads (GSO and GRO).
6751 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6752 dev->features |= NETIF_F_SOFT_FEATURES;
6753 dev->wanted_features = dev->features & dev->hw_features;
6755 if (!(dev->flags & IFF_LOOPBACK)) {
6756 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6759 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6761 dev->vlan_features |= NETIF_F_HIGHDMA;
6763 /* Make NETIF_F_SG inheritable to tunnel devices.
6765 dev->hw_enc_features |= NETIF_F_SG;
6767 /* Make NETIF_F_SG inheritable to MPLS.
6769 dev->mpls_features |= NETIF_F_SG;
6771 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6772 ret = notifier_to_errno(ret);
6776 ret = netdev_register_kobject(dev);
6779 dev->reg_state = NETREG_REGISTERED;
6781 __netdev_update_features(dev);
6784 * Default initial state at registry is that the
6785 * device is present.
6788 set_bit(__LINK_STATE_PRESENT, &dev->state);
6790 linkwatch_init_dev(dev);
6792 dev_init_scheduler(dev);
6794 list_netdevice(dev);
6795 add_device_randomness(dev->dev_addr, dev->addr_len);
6797 /* If the device has permanent device address, driver should
6798 * set dev_addr and also addr_assign_type should be set to
6799 * NET_ADDR_PERM (default value).
6801 if (dev->addr_assign_type == NET_ADDR_PERM)
6802 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6804 /* Notify protocols, that a new device appeared. */
6805 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6806 ret = notifier_to_errno(ret);
6808 rollback_registered(dev);
6809 dev->reg_state = NETREG_UNREGISTERED;
6812 * Prevent userspace races by waiting until the network
6813 * device is fully setup before sending notifications.
6815 if (!dev->rtnl_link_ops ||
6816 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6817 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6823 if (dev->netdev_ops->ndo_uninit)
6824 dev->netdev_ops->ndo_uninit(dev);
6827 EXPORT_SYMBOL(register_netdevice);
6830 * init_dummy_netdev - init a dummy network device for NAPI
6831 * @dev: device to init
6833 * This takes a network device structure and initialize the minimum
6834 * amount of fields so it can be used to schedule NAPI polls without
6835 * registering a full blown interface. This is to be used by drivers
6836 * that need to tie several hardware interfaces to a single NAPI
6837 * poll scheduler due to HW limitations.
6839 int init_dummy_netdev(struct net_device *dev)
6841 /* Clear everything. Note we don't initialize spinlocks
6842 * are they aren't supposed to be taken by any of the
6843 * NAPI code and this dummy netdev is supposed to be
6844 * only ever used for NAPI polls
6846 memset(dev, 0, sizeof(struct net_device));
6848 /* make sure we BUG if trying to hit standard
6849 * register/unregister code path
6851 dev->reg_state = NETREG_DUMMY;
6853 /* NAPI wants this */
6854 INIT_LIST_HEAD(&dev->napi_list);
6856 /* a dummy interface is started by default */
6857 set_bit(__LINK_STATE_PRESENT, &dev->state);
6858 set_bit(__LINK_STATE_START, &dev->state);
6860 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6861 * because users of this 'device' dont need to change
6867 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6871 * register_netdev - register a network device
6872 * @dev: device to register
6874 * Take a completed network device structure and add it to the kernel
6875 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6876 * chain. 0 is returned on success. A negative errno code is returned
6877 * on a failure to set up the device, or if the name is a duplicate.
6879 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6880 * and expands the device name if you passed a format string to
6883 int register_netdev(struct net_device *dev)
6888 err = register_netdevice(dev);
6892 EXPORT_SYMBOL(register_netdev);
6894 int netdev_refcnt_read(const struct net_device *dev)
6898 for_each_possible_cpu(i)
6899 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6902 EXPORT_SYMBOL(netdev_refcnt_read);
6905 * netdev_wait_allrefs - wait until all references are gone.
6906 * @dev: target net_device
6908 * This is called when unregistering network devices.
6910 * Any protocol or device that holds a reference should register
6911 * for netdevice notification, and cleanup and put back the
6912 * reference if they receive an UNREGISTER event.
6913 * We can get stuck here if buggy protocols don't correctly
6916 static void netdev_wait_allrefs(struct net_device *dev)
6918 unsigned long rebroadcast_time, warning_time;
6921 linkwatch_forget_dev(dev);
6923 rebroadcast_time = warning_time = jiffies;
6924 refcnt = netdev_refcnt_read(dev);
6926 while (refcnt != 0) {
6927 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6930 /* Rebroadcast unregister notification */
6931 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6937 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6938 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6940 /* We must not have linkwatch events
6941 * pending on unregister. If this
6942 * happens, we simply run the queue
6943 * unscheduled, resulting in a noop
6946 linkwatch_run_queue();
6951 rebroadcast_time = jiffies;
6956 refcnt = netdev_refcnt_read(dev);
6958 if (time_after(jiffies, warning_time + 10 * HZ)) {
6959 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6961 warning_time = jiffies;
6970 * register_netdevice(x1);
6971 * register_netdevice(x2);
6973 * unregister_netdevice(y1);
6974 * unregister_netdevice(y2);
6980 * We are invoked by rtnl_unlock().
6981 * This allows us to deal with problems:
6982 * 1) We can delete sysfs objects which invoke hotplug
6983 * without deadlocking with linkwatch via keventd.
6984 * 2) Since we run with the RTNL semaphore not held, we can sleep
6985 * safely in order to wait for the netdev refcnt to drop to zero.
6987 * We must not return until all unregister events added during
6988 * the interval the lock was held have been completed.
6990 void netdev_run_todo(void)
6992 struct list_head list;
6994 /* Snapshot list, allow later requests */
6995 list_replace_init(&net_todo_list, &list);
7000 /* Wait for rcu callbacks to finish before next phase */
7001 if (!list_empty(&list))
7004 while (!list_empty(&list)) {
7005 struct net_device *dev
7006 = list_first_entry(&list, struct net_device, todo_list);
7007 list_del(&dev->todo_list);
7010 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7013 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
7014 pr_err("network todo '%s' but state %d\n",
7015 dev->name, dev->reg_state);
7020 dev->reg_state = NETREG_UNREGISTERED;
7022 netdev_wait_allrefs(dev);
7025 BUG_ON(netdev_refcnt_read(dev));
7026 BUG_ON(!list_empty(&dev->ptype_all));
7027 BUG_ON(!list_empty(&dev->ptype_specific));
7028 WARN_ON(rcu_access_pointer(dev->ip_ptr));
7029 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
7030 WARN_ON(dev->dn_ptr);
7032 if (dev->destructor)
7033 dev->destructor(dev);
7035 /* Report a network device has been unregistered */
7037 dev_net(dev)->dev_unreg_count--;
7039 wake_up(&netdev_unregistering_wq);
7041 /* Free network device */
7042 kobject_put(&dev->dev.kobj);
7046 /* Convert net_device_stats to rtnl_link_stats64. They have the same
7047 * fields in the same order, with only the type differing.
7049 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7050 const struct net_device_stats *netdev_stats)
7052 #if BITS_PER_LONG == 64
7053 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
7054 memcpy(stats64, netdev_stats, sizeof(*stats64));
7056 size_t i, n = sizeof(*stats64) / sizeof(u64);
7057 const unsigned long *src = (const unsigned long *)netdev_stats;
7058 u64 *dst = (u64 *)stats64;
7060 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
7061 sizeof(*stats64) / sizeof(u64));
7062 for (i = 0; i < n; i++)
7066 EXPORT_SYMBOL(netdev_stats_to_stats64);
7069 * dev_get_stats - get network device statistics
7070 * @dev: device to get statistics from
7071 * @storage: place to store stats
7073 * Get network statistics from device. Return @storage.
7074 * The device driver may provide its own method by setting
7075 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7076 * otherwise the internal statistics structure is used.
7078 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
7079 struct rtnl_link_stats64 *storage)
7081 const struct net_device_ops *ops = dev->netdev_ops;
7083 if (ops->ndo_get_stats64) {
7084 memset(storage, 0, sizeof(*storage));
7085 ops->ndo_get_stats64(dev, storage);
7086 } else if (ops->ndo_get_stats) {
7087 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7089 netdev_stats_to_stats64(storage, &dev->stats);
7091 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
7092 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
7095 EXPORT_SYMBOL(dev_get_stats);
7097 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7099 struct netdev_queue *queue = dev_ingress_queue(dev);
7101 #ifdef CONFIG_NET_CLS_ACT
7104 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7107 netdev_init_one_queue(dev, queue, NULL);
7108 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7109 queue->qdisc_sleeping = &noop_qdisc;
7110 rcu_assign_pointer(dev->ingress_queue, queue);
7115 static const struct ethtool_ops default_ethtool_ops;
7117 void netdev_set_default_ethtool_ops(struct net_device *dev,
7118 const struct ethtool_ops *ops)
7120 if (dev->ethtool_ops == &default_ethtool_ops)
7121 dev->ethtool_ops = ops;
7123 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7125 void netdev_freemem(struct net_device *dev)
7127 char *addr = (char *)dev - dev->padded;
7133 * alloc_netdev_mqs - allocate network device
7134 * @sizeof_priv: size of private data to allocate space for
7135 * @name: device name format string
7136 * @name_assign_type: origin of device name
7137 * @setup: callback to initialize device
7138 * @txqs: the number of TX subqueues to allocate
7139 * @rxqs: the number of RX subqueues to allocate
7141 * Allocates a struct net_device with private data area for driver use
7142 * and performs basic initialization. Also allocates subqueue structs
7143 * for each queue on the device.
7145 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7146 unsigned char name_assign_type,
7147 void (*setup)(struct net_device *),
7148 unsigned int txqs, unsigned int rxqs)
7150 struct net_device *dev;
7152 struct net_device *p;
7154 BUG_ON(strlen(name) >= sizeof(dev->name));
7157 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7163 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7168 alloc_size = sizeof(struct net_device);
7170 /* ensure 32-byte alignment of private area */
7171 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7172 alloc_size += sizeof_priv;
7174 /* ensure 32-byte alignment of whole construct */
7175 alloc_size += NETDEV_ALIGN - 1;
7177 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7179 p = vzalloc(alloc_size);
7183 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7184 dev->padded = (char *)dev - (char *)p;
7186 dev->pcpu_refcnt = alloc_percpu(int);
7187 if (!dev->pcpu_refcnt)
7190 if (dev_addr_init(dev))
7196 dev_net_set(dev, &init_net);
7198 dev->gso_max_size = GSO_MAX_SIZE;
7199 dev->gso_max_segs = GSO_MAX_SEGS;
7200 dev->gso_min_segs = 0;
7202 INIT_LIST_HEAD(&dev->napi_list);
7203 INIT_LIST_HEAD(&dev->unreg_list);
7204 INIT_LIST_HEAD(&dev->close_list);
7205 INIT_LIST_HEAD(&dev->link_watch_list);
7206 INIT_LIST_HEAD(&dev->adj_list.upper);
7207 INIT_LIST_HEAD(&dev->adj_list.lower);
7208 INIT_LIST_HEAD(&dev->all_adj_list.upper);
7209 INIT_LIST_HEAD(&dev->all_adj_list.lower);
7210 INIT_LIST_HEAD(&dev->ptype_all);
7211 INIT_LIST_HEAD(&dev->ptype_specific);
7212 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7215 if (!dev->tx_queue_len)
7216 dev->priv_flags |= IFF_NO_QUEUE;
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.
7259 * Must be called in process context.
7261 void free_netdev(struct net_device *dev)
7263 struct napi_struct *p, *n;
7266 netif_free_tx_queues(dev);
7271 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7273 /* Flush device addresses */
7274 dev_addr_flush(dev);
7276 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7279 free_percpu(dev->pcpu_refcnt);
7280 dev->pcpu_refcnt = NULL;
7282 /* Compatibility with error handling in drivers */
7283 if (dev->reg_state == NETREG_UNINITIALIZED) {
7284 netdev_freemem(dev);
7288 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7289 dev->reg_state = NETREG_RELEASED;
7291 /* will free via device release */
7292 put_device(&dev->dev);
7294 EXPORT_SYMBOL(free_netdev);
7297 * synchronize_net - Synchronize with packet receive processing
7299 * Wait for packets currently being received to be done.
7300 * Does not block later packets from starting.
7302 void synchronize_net(void)
7305 if (rtnl_is_locked())
7306 synchronize_rcu_expedited();
7310 EXPORT_SYMBOL(synchronize_net);
7313 * unregister_netdevice_queue - remove device from the kernel
7317 * This function shuts down a device interface and removes it
7318 * from the kernel tables.
7319 * If head not NULL, device is queued to be unregistered later.
7321 * Callers must hold the rtnl semaphore. You may want
7322 * unregister_netdev() instead of this.
7325 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7330 list_move_tail(&dev->unreg_list, head);
7332 rollback_registered(dev);
7333 /* Finish processing unregister after unlock */
7337 EXPORT_SYMBOL(unregister_netdevice_queue);
7340 * unregister_netdevice_many - unregister many devices
7341 * @head: list of devices
7343 * Note: As most callers use a stack allocated list_head,
7344 * we force a list_del() to make sure stack wont be corrupted later.
7346 void unregister_netdevice_many(struct list_head *head)
7348 struct net_device *dev;
7350 if (!list_empty(head)) {
7351 rollback_registered_many(head);
7352 list_for_each_entry(dev, head, unreg_list)
7357 EXPORT_SYMBOL(unregister_netdevice_many);
7360 * unregister_netdev - remove device from the kernel
7363 * This function shuts down a device interface and removes it
7364 * from the kernel tables.
7366 * This is just a wrapper for unregister_netdevice that takes
7367 * the rtnl semaphore. In general you want to use this and not
7368 * unregister_netdevice.
7370 void unregister_netdev(struct net_device *dev)
7373 unregister_netdevice(dev);
7376 EXPORT_SYMBOL(unregister_netdev);
7379 * dev_change_net_namespace - move device to different nethost namespace
7381 * @net: network namespace
7382 * @pat: If not NULL name pattern to try if the current device name
7383 * is already taken in the destination network namespace.
7385 * This function shuts down a device interface and moves it
7386 * to a new network namespace. On success 0 is returned, on
7387 * a failure a netagive errno code is returned.
7389 * Callers must hold the rtnl semaphore.
7392 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7398 /* Don't allow namespace local devices to be moved. */
7400 if (dev->features & NETIF_F_NETNS_LOCAL)
7403 /* Ensure the device has been registrered */
7404 if (dev->reg_state != NETREG_REGISTERED)
7407 /* Get out if there is nothing todo */
7409 if (net_eq(dev_net(dev), net))
7412 /* Pick the destination device name, and ensure
7413 * we can use it in the destination network namespace.
7416 if (__dev_get_by_name(net, dev->name)) {
7417 /* We get here if we can't use the current device name */
7420 if (dev_get_valid_name(net, dev, pat) < 0)
7425 * And now a mini version of register_netdevice unregister_netdevice.
7428 /* If device is running close it first. */
7431 /* And unlink it from device chain */
7433 unlist_netdevice(dev);
7437 /* Shutdown queueing discipline. */
7440 /* Notify protocols, that we are about to destroy
7441 this device. They should clean all the things.
7443 Note that dev->reg_state stays at NETREG_REGISTERED.
7444 This is wanted because this way 8021q and macvlan know
7445 the device is just moving and can keep their slaves up.
7447 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7449 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7450 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7453 * Flush the unicast and multicast chains
7458 /* Send a netdev-removed uevent to the old namespace */
7459 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7460 netdev_adjacent_del_links(dev);
7462 /* Actually switch the network namespace */
7463 dev_net_set(dev, net);
7465 /* If there is an ifindex conflict assign a new one */
7466 if (__dev_get_by_index(net, dev->ifindex))
7467 dev->ifindex = dev_new_index(net);
7469 /* Send a netdev-add uevent to the new namespace */
7470 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7471 netdev_adjacent_add_links(dev);
7473 /* Fixup kobjects */
7474 err = device_rename(&dev->dev, dev->name);
7477 /* Add the device back in the hashes */
7478 list_netdevice(dev);
7480 /* Notify protocols, that a new device appeared. */
7481 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7484 * Prevent userspace races by waiting until the network
7485 * device is fully setup before sending notifications.
7487 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7494 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7496 static int dev_cpu_callback(struct notifier_block *nfb,
7497 unsigned long action,
7500 struct sk_buff **list_skb;
7501 struct sk_buff *skb;
7502 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7503 struct softnet_data *sd, *oldsd;
7505 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7508 local_irq_disable();
7509 cpu = smp_processor_id();
7510 sd = &per_cpu(softnet_data, cpu);
7511 oldsd = &per_cpu(softnet_data, oldcpu);
7513 /* Find end of our completion_queue. */
7514 list_skb = &sd->completion_queue;
7516 list_skb = &(*list_skb)->next;
7517 /* Append completion queue from offline CPU. */
7518 *list_skb = oldsd->completion_queue;
7519 oldsd->completion_queue = NULL;
7521 /* Append output queue from offline CPU. */
7522 if (oldsd->output_queue) {
7523 *sd->output_queue_tailp = oldsd->output_queue;
7524 sd->output_queue_tailp = oldsd->output_queue_tailp;
7525 oldsd->output_queue = NULL;
7526 oldsd->output_queue_tailp = &oldsd->output_queue;
7528 /* Append NAPI poll list from offline CPU, with one exception :
7529 * process_backlog() must be called by cpu owning percpu backlog.
7530 * We properly handle process_queue & input_pkt_queue later.
7532 while (!list_empty(&oldsd->poll_list)) {
7533 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7537 list_del_init(&napi->poll_list);
7538 if (napi->poll == process_backlog)
7541 ____napi_schedule(sd, napi);
7544 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7547 /* Process offline CPU's input_pkt_queue */
7548 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7550 input_queue_head_incr(oldsd);
7552 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7554 input_queue_head_incr(oldsd);
7562 * netdev_increment_features - increment feature set by one
7563 * @all: current feature set
7564 * @one: new feature set
7565 * @mask: mask feature set
7567 * Computes a new feature set after adding a device with feature set
7568 * @one to the master device with current feature set @all. Will not
7569 * enable anything that is off in @mask. Returns the new feature set.
7571 netdev_features_t netdev_increment_features(netdev_features_t all,
7572 netdev_features_t one, netdev_features_t mask)
7574 if (mask & NETIF_F_GEN_CSUM)
7575 mask |= NETIF_F_ALL_CSUM;
7576 mask |= NETIF_F_VLAN_CHALLENGED;
7578 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7579 all &= one | ~NETIF_F_ALL_FOR_ALL;
7581 /* If one device supports hw checksumming, set for all. */
7582 if (all & NETIF_F_GEN_CSUM)
7583 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7587 EXPORT_SYMBOL(netdev_increment_features);
7589 static struct hlist_head * __net_init netdev_create_hash(void)
7592 struct hlist_head *hash;
7594 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7596 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7597 INIT_HLIST_HEAD(&hash[i]);
7602 /* Initialize per network namespace state */
7603 static int __net_init netdev_init(struct net *net)
7605 if (net != &init_net)
7606 INIT_LIST_HEAD(&net->dev_base_head);
7608 net->dev_name_head = netdev_create_hash();
7609 if (net->dev_name_head == NULL)
7612 net->dev_index_head = netdev_create_hash();
7613 if (net->dev_index_head == NULL)
7619 kfree(net->dev_name_head);
7625 * netdev_drivername - network driver for the device
7626 * @dev: network device
7628 * Determine network driver for device.
7630 const char *netdev_drivername(const struct net_device *dev)
7632 const struct device_driver *driver;
7633 const struct device *parent;
7634 const char *empty = "";
7636 parent = dev->dev.parent;
7640 driver = parent->driver;
7641 if (driver && driver->name)
7642 return driver->name;
7646 static void __netdev_printk(const char *level, const struct net_device *dev,
7647 struct va_format *vaf)
7649 if (dev && dev->dev.parent) {
7650 dev_printk_emit(level[1] - '0',
7653 dev_driver_string(dev->dev.parent),
7654 dev_name(dev->dev.parent),
7655 netdev_name(dev), netdev_reg_state(dev),
7658 printk("%s%s%s: %pV",
7659 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7661 printk("%s(NULL net_device): %pV", level, vaf);
7665 void netdev_printk(const char *level, const struct net_device *dev,
7666 const char *format, ...)
7668 struct va_format vaf;
7671 va_start(args, format);
7676 __netdev_printk(level, dev, &vaf);
7680 EXPORT_SYMBOL(netdev_printk);
7682 #define define_netdev_printk_level(func, level) \
7683 void func(const struct net_device *dev, const char *fmt, ...) \
7685 struct va_format vaf; \
7688 va_start(args, fmt); \
7693 __netdev_printk(level, dev, &vaf); \
7697 EXPORT_SYMBOL(func);
7699 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7700 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7701 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7702 define_netdev_printk_level(netdev_err, KERN_ERR);
7703 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7704 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7705 define_netdev_printk_level(netdev_info, KERN_INFO);
7707 static void __net_exit netdev_exit(struct net *net)
7709 kfree(net->dev_name_head);
7710 kfree(net->dev_index_head);
7713 static struct pernet_operations __net_initdata netdev_net_ops = {
7714 .init = netdev_init,
7715 .exit = netdev_exit,
7718 static void __net_exit default_device_exit(struct net *net)
7720 struct net_device *dev, *aux;
7722 * Push all migratable network devices back to the
7723 * initial network namespace
7726 for_each_netdev_safe(net, dev, aux) {
7728 char fb_name[IFNAMSIZ];
7730 /* Ignore unmoveable devices (i.e. loopback) */
7731 if (dev->features & NETIF_F_NETNS_LOCAL)
7734 /* Leave virtual devices for the generic cleanup */
7735 if (dev->rtnl_link_ops)
7738 /* Push remaining network devices to init_net */
7739 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7740 err = dev_change_net_namespace(dev, &init_net, fb_name);
7742 pr_emerg("%s: failed to move %s to init_net: %d\n",
7743 __func__, dev->name, err);
7750 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7752 /* Return with the rtnl_lock held when there are no network
7753 * devices unregistering in any network namespace in net_list.
7757 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7759 add_wait_queue(&netdev_unregistering_wq, &wait);
7761 unregistering = false;
7763 list_for_each_entry(net, net_list, exit_list) {
7764 if (net->dev_unreg_count > 0) {
7765 unregistering = true;
7773 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7775 remove_wait_queue(&netdev_unregistering_wq, &wait);
7778 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7780 /* At exit all network devices most be removed from a network
7781 * namespace. Do this in the reverse order of registration.
7782 * Do this across as many network namespaces as possible to
7783 * improve batching efficiency.
7785 struct net_device *dev;
7787 LIST_HEAD(dev_kill_list);
7789 /* To prevent network device cleanup code from dereferencing
7790 * loopback devices or network devices that have been freed
7791 * wait here for all pending unregistrations to complete,
7792 * before unregistring the loopback device and allowing the
7793 * network namespace be freed.
7795 * The netdev todo list containing all network devices
7796 * unregistrations that happen in default_device_exit_batch
7797 * will run in the rtnl_unlock() at the end of
7798 * default_device_exit_batch.
7800 rtnl_lock_unregistering(net_list);
7801 list_for_each_entry(net, net_list, exit_list) {
7802 for_each_netdev_reverse(net, dev) {
7803 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7804 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7806 unregister_netdevice_queue(dev, &dev_kill_list);
7809 unregister_netdevice_many(&dev_kill_list);
7813 static struct pernet_operations __net_initdata default_device_ops = {
7814 .exit = default_device_exit,
7815 .exit_batch = default_device_exit_batch,
7819 * Initialize the DEV module. At boot time this walks the device list and
7820 * unhooks any devices that fail to initialise (normally hardware not
7821 * present) and leaves us with a valid list of present and active devices.
7826 * This is called single threaded during boot, so no need
7827 * to take the rtnl semaphore.
7829 static int __init net_dev_init(void)
7831 int i, rc = -ENOMEM;
7833 BUG_ON(!dev_boot_phase);
7835 if (dev_proc_init())
7838 if (netdev_kobject_init())
7841 INIT_LIST_HEAD(&ptype_all);
7842 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7843 INIT_LIST_HEAD(&ptype_base[i]);
7845 INIT_LIST_HEAD(&offload_base);
7847 if (register_pernet_subsys(&netdev_net_ops))
7851 * Initialise the packet receive queues.
7854 for_each_possible_cpu(i) {
7855 struct softnet_data *sd = &per_cpu(softnet_data, i);
7857 skb_queue_head_init(&sd->input_pkt_queue);
7858 skb_queue_head_init(&sd->process_queue);
7859 INIT_LIST_HEAD(&sd->poll_list);
7860 sd->output_queue_tailp = &sd->output_queue;
7862 sd->csd.func = rps_trigger_softirq;
7867 sd->backlog.poll = process_backlog;
7868 sd->backlog.weight = weight_p;
7873 /* The loopback device is special if any other network devices
7874 * is present in a network namespace the loopback device must
7875 * be present. Since we now dynamically allocate and free the
7876 * loopback device ensure this invariant is maintained by
7877 * keeping the loopback device as the first device on the
7878 * list of network devices. Ensuring the loopback devices
7879 * is the first device that appears and the last network device
7882 if (register_pernet_device(&loopback_net_ops))
7885 if (register_pernet_device(&default_device_ops))
7888 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7889 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7891 hotcpu_notifier(dev_cpu_callback, 0);
7898 subsys_initcall(net_dev_init);