2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
121 #include <net/mpls.h>
122 #include <linux/ipv6.h>
123 #include <linux/in.h>
124 #include <linux/jhash.h>
125 #include <linux/random.h>
126 #include <trace/events/napi.h>
127 #include <trace/events/net.h>
128 #include <trace/events/skb.h>
129 #include <linux/pci.h>
130 #include <linux/inetdevice.h>
131 #include <linux/cpu_rmap.h>
132 #include <linux/static_key.h>
133 #include <linux/hashtable.h>
134 #include <linux/vmalloc.h>
135 #include <linux/if_macvlan.h>
136 #include <linux/errqueue.h>
137 #include <linux/hrtimer.h>
138 #include <linux/netfilter_ingress.h>
140 #include "net-sysfs.h"
142 /* Instead of increasing this, you should create a hash table. */
143 #define MAX_GRO_SKBS 8
145 /* This should be increased if a protocol with a bigger head is added. */
146 #define GRO_MAX_HEAD (MAX_HEADER + 128)
148 static DEFINE_SPINLOCK(ptype_lock);
149 static DEFINE_SPINLOCK(offload_lock);
150 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
151 struct list_head ptype_all __read_mostly; /* Taps */
152 static struct list_head offload_base __read_mostly;
154 static int netif_rx_internal(struct sk_buff *skb);
155 static int call_netdevice_notifiers_info(unsigned long val,
156 struct net_device *dev,
157 struct netdev_notifier_info *info);
160 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
163 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
165 * Writers must hold the rtnl semaphore while they loop through the
166 * dev_base_head list, and hold dev_base_lock for writing when they do the
167 * actual updates. This allows pure readers to access the list even
168 * while a writer is preparing to update it.
170 * To put it another way, dev_base_lock is held for writing only to
171 * protect against pure readers; the rtnl semaphore provides the
172 * protection against other writers.
174 * See, for example usages, register_netdevice() and
175 * unregister_netdevice(), which must be called with the rtnl
178 DEFINE_RWLOCK(dev_base_lock);
179 EXPORT_SYMBOL(dev_base_lock);
181 /* protects napi_hash addition/deletion and napi_gen_id */
182 static DEFINE_SPINLOCK(napi_hash_lock);
184 static unsigned int napi_gen_id;
185 static DEFINE_HASHTABLE(napi_hash, 8);
187 static seqcount_t devnet_rename_seq;
189 static inline void dev_base_seq_inc(struct net *net)
191 while (++net->dev_base_seq == 0);
194 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
196 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
198 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
201 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
203 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
206 static inline void rps_lock(struct softnet_data *sd)
209 spin_lock(&sd->input_pkt_queue.lock);
213 static inline void rps_unlock(struct softnet_data *sd)
216 spin_unlock(&sd->input_pkt_queue.lock);
220 /* Device list insertion */
221 static void list_netdevice(struct net_device *dev)
223 struct net *net = dev_net(dev);
227 write_lock_bh(&dev_base_lock);
228 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
229 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
230 hlist_add_head_rcu(&dev->index_hlist,
231 dev_index_hash(net, dev->ifindex));
232 write_unlock_bh(&dev_base_lock);
234 dev_base_seq_inc(net);
237 /* Device list removal
238 * caller must respect a RCU grace period before freeing/reusing dev
240 static void unlist_netdevice(struct net_device *dev)
244 /* Unlink dev from the device chain */
245 write_lock_bh(&dev_base_lock);
246 list_del_rcu(&dev->dev_list);
247 hlist_del_rcu(&dev->name_hlist);
248 hlist_del_rcu(&dev->index_hlist);
249 write_unlock_bh(&dev_base_lock);
251 dev_base_seq_inc(dev_net(dev));
258 static RAW_NOTIFIER_HEAD(netdev_chain);
261 * Device drivers call our routines to queue packets here. We empty the
262 * queue in the local softnet handler.
265 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
266 EXPORT_PER_CPU_SYMBOL(softnet_data);
268 #ifdef CONFIG_LOCKDEP
270 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
271 * according to dev->type
273 static const unsigned short netdev_lock_type[] =
274 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
275 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
276 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
277 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
278 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
279 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
280 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
281 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
282 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
283 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
284 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
285 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
286 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
287 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
288 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
290 static const char *const netdev_lock_name[] =
291 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
292 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
293 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
294 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
295 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
296 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
297 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
298 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
299 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
300 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
301 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
302 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
303 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
304 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
305 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
307 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
308 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
310 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
314 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
315 if (netdev_lock_type[i] == dev_type)
317 /* the last key is used by default */
318 return ARRAY_SIZE(netdev_lock_type) - 1;
321 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
322 unsigned short dev_type)
326 i = netdev_lock_pos(dev_type);
327 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
328 netdev_lock_name[i]);
331 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
335 i = netdev_lock_pos(dev->type);
336 lockdep_set_class_and_name(&dev->addr_list_lock,
337 &netdev_addr_lock_key[i],
338 netdev_lock_name[i]);
341 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
342 unsigned short dev_type)
345 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
350 /*******************************************************************************
352 Protocol management and registration routines
354 *******************************************************************************/
357 * Add a protocol ID to the list. Now that the input handler is
358 * smarter we can dispense with all the messy stuff that used to be
361 * BEWARE!!! Protocol handlers, mangling input packets,
362 * MUST BE last in hash buckets and checking protocol handlers
363 * MUST start from promiscuous ptype_all chain in net_bh.
364 * It is true now, do not change it.
365 * Explanation follows: if protocol handler, mangling packet, will
366 * be the first on list, it is not able to sense, that packet
367 * is cloned and should be copied-on-write, so that it will
368 * change it and subsequent readers will get broken packet.
372 static inline struct list_head *ptype_head(const struct packet_type *pt)
374 if (pt->type == htons(ETH_P_ALL))
375 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
377 return pt->dev ? &pt->dev->ptype_specific :
378 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
382 * dev_add_pack - add packet handler
383 * @pt: packet type declaration
385 * Add a protocol handler to the networking stack. The passed &packet_type
386 * is linked into kernel lists and may not be freed until it has been
387 * removed from the kernel lists.
389 * This call does not sleep therefore it can not
390 * guarantee all CPU's that are in middle of receiving packets
391 * will see the new packet type (until the next received packet).
394 void dev_add_pack(struct packet_type *pt)
396 struct list_head *head = ptype_head(pt);
398 spin_lock(&ptype_lock);
399 list_add_rcu(&pt->list, head);
400 spin_unlock(&ptype_lock);
402 EXPORT_SYMBOL(dev_add_pack);
405 * __dev_remove_pack - remove packet handler
406 * @pt: packet type declaration
408 * Remove a protocol handler that was previously added to the kernel
409 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
410 * from the kernel lists and can be freed or reused once this function
413 * The packet type might still be in use by receivers
414 * and must not be freed until after all the CPU's have gone
415 * through a quiescent state.
417 void __dev_remove_pack(struct packet_type *pt)
419 struct list_head *head = ptype_head(pt);
420 struct packet_type *pt1;
422 spin_lock(&ptype_lock);
424 list_for_each_entry(pt1, head, list) {
426 list_del_rcu(&pt->list);
431 pr_warn("dev_remove_pack: %p not found\n", pt);
433 spin_unlock(&ptype_lock);
435 EXPORT_SYMBOL(__dev_remove_pack);
438 * dev_remove_pack - remove packet handler
439 * @pt: packet type declaration
441 * Remove a protocol handler that was previously added to the kernel
442 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
443 * from the kernel lists and can be freed or reused once this function
446 * This call sleeps to guarantee that no CPU is looking at the packet
449 void dev_remove_pack(struct packet_type *pt)
451 __dev_remove_pack(pt);
455 EXPORT_SYMBOL(dev_remove_pack);
459 * dev_add_offload - register offload handlers
460 * @po: protocol offload declaration
462 * Add protocol offload handlers to the networking stack. The passed
463 * &proto_offload is linked into kernel lists and may not be freed until
464 * it has been removed from the kernel lists.
466 * This call does not sleep therefore it can not
467 * guarantee all CPU's that are in middle of receiving packets
468 * will see the new offload handlers (until the next received packet).
470 void dev_add_offload(struct packet_offload *po)
472 struct packet_offload *elem;
474 spin_lock(&offload_lock);
475 list_for_each_entry(elem, &offload_base, list) {
476 if (po->priority < elem->priority)
479 list_add_rcu(&po->list, elem->list.prev);
480 spin_unlock(&offload_lock);
482 EXPORT_SYMBOL(dev_add_offload);
485 * __dev_remove_offload - remove offload handler
486 * @po: packet offload declaration
488 * Remove a protocol offload handler that was previously added to the
489 * kernel offload handlers by dev_add_offload(). The passed &offload_type
490 * is removed from the kernel lists and can be freed or reused once this
493 * The packet type might still be in use by receivers
494 * and must not be freed until after all the CPU's have gone
495 * through a quiescent state.
497 static void __dev_remove_offload(struct packet_offload *po)
499 struct list_head *head = &offload_base;
500 struct packet_offload *po1;
502 spin_lock(&offload_lock);
504 list_for_each_entry(po1, head, list) {
506 list_del_rcu(&po->list);
511 pr_warn("dev_remove_offload: %p not found\n", po);
513 spin_unlock(&offload_lock);
517 * dev_remove_offload - remove packet offload handler
518 * @po: packet offload declaration
520 * Remove a packet offload handler that was previously added to the kernel
521 * offload handlers by dev_add_offload(). The passed &offload_type is
522 * removed from the kernel lists and can be freed or reused once this
525 * This call sleeps to guarantee that no CPU is looking at the packet
528 void dev_remove_offload(struct packet_offload *po)
530 __dev_remove_offload(po);
534 EXPORT_SYMBOL(dev_remove_offload);
536 /******************************************************************************
538 Device Boot-time Settings Routines
540 *******************************************************************************/
542 /* Boot time configuration table */
543 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
546 * netdev_boot_setup_add - add new setup entry
547 * @name: name of the device
548 * @map: configured settings for the device
550 * Adds new setup entry to the dev_boot_setup list. The function
551 * returns 0 on error and 1 on success. This is a generic routine to
554 static int netdev_boot_setup_add(char *name, struct ifmap *map)
556 struct netdev_boot_setup *s;
560 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
561 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
562 memset(s[i].name, 0, sizeof(s[i].name));
563 strlcpy(s[i].name, name, IFNAMSIZ);
564 memcpy(&s[i].map, map, sizeof(s[i].map));
569 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
573 * netdev_boot_setup_check - check boot time settings
574 * @dev: the netdevice
576 * Check boot time settings for the device.
577 * The found settings are set for the device to be used
578 * later in the device probing.
579 * Returns 0 if no settings found, 1 if they are.
581 int netdev_boot_setup_check(struct net_device *dev)
583 struct netdev_boot_setup *s = dev_boot_setup;
586 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
587 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
588 !strcmp(dev->name, s[i].name)) {
589 dev->irq = s[i].map.irq;
590 dev->base_addr = s[i].map.base_addr;
591 dev->mem_start = s[i].map.mem_start;
592 dev->mem_end = s[i].map.mem_end;
598 EXPORT_SYMBOL(netdev_boot_setup_check);
602 * netdev_boot_base - get address from boot time settings
603 * @prefix: prefix for network device
604 * @unit: id for network device
606 * Check boot time settings for the base address of device.
607 * The found settings are set for the device to be used
608 * later in the device probing.
609 * Returns 0 if no settings found.
611 unsigned long netdev_boot_base(const char *prefix, int unit)
613 const struct netdev_boot_setup *s = dev_boot_setup;
617 sprintf(name, "%s%d", prefix, unit);
620 * If device already registered then return base of 1
621 * to indicate not to probe for this interface
623 if (__dev_get_by_name(&init_net, name))
626 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
627 if (!strcmp(name, s[i].name))
628 return s[i].map.base_addr;
633 * Saves at boot time configured settings for any netdevice.
635 int __init netdev_boot_setup(char *str)
640 str = get_options(str, ARRAY_SIZE(ints), ints);
645 memset(&map, 0, sizeof(map));
649 map.base_addr = ints[2];
651 map.mem_start = ints[3];
653 map.mem_end = ints[4];
655 /* Add new entry to the list */
656 return netdev_boot_setup_add(str, &map);
659 __setup("netdev=", netdev_boot_setup);
661 /*******************************************************************************
663 Device Interface Subroutines
665 *******************************************************************************/
668 * dev_get_iflink - get 'iflink' value of a interface
669 * @dev: targeted interface
671 * Indicates the ifindex the interface is linked to.
672 * Physical interfaces have the same 'ifindex' and 'iflink' values.
675 int dev_get_iflink(const struct net_device *dev)
677 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
678 return dev->netdev_ops->ndo_get_iflink(dev);
680 /* If dev->rtnl_link_ops is set, it's a virtual interface. */
681 if (dev->rtnl_link_ops)
686 EXPORT_SYMBOL(dev_get_iflink);
689 * __dev_get_by_name - find a device by its name
690 * @net: the applicable net namespace
691 * @name: name to find
693 * Find an interface by name. Must be called under RTNL semaphore
694 * or @dev_base_lock. If the name is found a pointer to the device
695 * is returned. If the name is not found then %NULL is returned. The
696 * reference counters are not incremented so the caller must be
697 * careful with locks.
700 struct net_device *__dev_get_by_name(struct net *net, const char *name)
702 struct net_device *dev;
703 struct hlist_head *head = dev_name_hash(net, name);
705 hlist_for_each_entry(dev, head, name_hlist)
706 if (!strncmp(dev->name, name, IFNAMSIZ))
711 EXPORT_SYMBOL(__dev_get_by_name);
714 * dev_get_by_name_rcu - find a device by its name
715 * @net: the applicable net namespace
716 * @name: name to find
718 * Find an interface by name.
719 * If the name is found a pointer to the device is returned.
720 * If the name is not found then %NULL is returned.
721 * The reference counters are not incremented so the caller must be
722 * careful with locks. The caller must hold RCU lock.
725 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
727 struct net_device *dev;
728 struct hlist_head *head = dev_name_hash(net, name);
730 hlist_for_each_entry_rcu(dev, head, name_hlist)
731 if (!strncmp(dev->name, name, IFNAMSIZ))
736 EXPORT_SYMBOL(dev_get_by_name_rcu);
739 * dev_get_by_name - find a device by its name
740 * @net: the applicable net namespace
741 * @name: name to find
743 * Find an interface by name. This can be called from any
744 * context and does its own locking. The returned handle has
745 * the usage count incremented and the caller must use dev_put() to
746 * release it when it is no longer needed. %NULL is returned if no
747 * matching device is found.
750 struct net_device *dev_get_by_name(struct net *net, const char *name)
752 struct net_device *dev;
755 dev = dev_get_by_name_rcu(net, name);
761 EXPORT_SYMBOL(dev_get_by_name);
764 * __dev_get_by_index - find a device by its ifindex
765 * @net: the applicable net namespace
766 * @ifindex: index of device
768 * Search for an interface by index. Returns %NULL if the device
769 * is not found or a pointer to the device. The device has not
770 * had its reference counter increased so the caller must be careful
771 * about locking. The caller must hold either the RTNL semaphore
775 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
777 struct net_device *dev;
778 struct hlist_head *head = dev_index_hash(net, ifindex);
780 hlist_for_each_entry(dev, head, index_hlist)
781 if (dev->ifindex == ifindex)
786 EXPORT_SYMBOL(__dev_get_by_index);
789 * dev_get_by_index_rcu - find a device by its ifindex
790 * @net: the applicable net namespace
791 * @ifindex: index of device
793 * Search for an interface by index. Returns %NULL if the device
794 * is not found or a pointer to the device. The device has not
795 * had its reference counter increased so the caller must be careful
796 * about locking. The caller must hold RCU lock.
799 struct net_device *dev_get_by_index_rcu(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_rcu(dev, head, index_hlist)
805 if (dev->ifindex == ifindex)
810 EXPORT_SYMBOL(dev_get_by_index_rcu);
814 * dev_get_by_index - find a device by its ifindex
815 * @net: the applicable net namespace
816 * @ifindex: index of device
818 * Search for an interface by index. Returns NULL if the device
819 * is not found or a pointer to the device. The device returned has
820 * had a reference added and the pointer is safe until the user calls
821 * dev_put to indicate they have finished with it.
824 struct net_device *dev_get_by_index(struct net *net, int ifindex)
826 struct net_device *dev;
829 dev = dev_get_by_index_rcu(net, ifindex);
835 EXPORT_SYMBOL(dev_get_by_index);
838 * netdev_get_name - get a netdevice name, knowing its ifindex.
839 * @net: network namespace
840 * @name: a pointer to the buffer where the name will be stored.
841 * @ifindex: the ifindex of the interface to get the name from.
843 * The use of raw_seqcount_begin() and cond_resched() before
844 * retrying is required as we want to give the writers a chance
845 * to complete when CONFIG_PREEMPT is not set.
847 int netdev_get_name(struct net *net, char *name, int ifindex)
849 struct net_device *dev;
853 seq = raw_seqcount_begin(&devnet_rename_seq);
855 dev = dev_get_by_index_rcu(net, ifindex);
861 strcpy(name, dev->name);
863 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
872 * dev_getbyhwaddr_rcu - find a device by its hardware address
873 * @net: the applicable net namespace
874 * @type: media type of device
875 * @ha: hardware address
877 * Search for an interface by MAC address. Returns NULL if the device
878 * is not found or a pointer to the device.
879 * The caller must hold RCU or RTNL.
880 * The returned device has not had its ref count increased
881 * and the caller must therefore be careful about locking
885 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
888 struct net_device *dev;
890 for_each_netdev_rcu(net, dev)
891 if (dev->type == type &&
892 !memcmp(dev->dev_addr, ha, dev->addr_len))
897 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
899 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
901 struct net_device *dev;
904 for_each_netdev(net, dev)
905 if (dev->type == type)
910 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
912 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
914 struct net_device *dev, *ret = NULL;
917 for_each_netdev_rcu(net, dev)
918 if (dev->type == type) {
926 EXPORT_SYMBOL(dev_getfirstbyhwtype);
929 * __dev_get_by_flags - find any device with given flags
930 * @net: the applicable net namespace
931 * @if_flags: IFF_* values
932 * @mask: bitmask of bits in if_flags to check
934 * Search for any interface with the given flags. Returns NULL if a device
935 * is not found or a pointer to the device. Must be called inside
936 * rtnl_lock(), and result refcount is unchanged.
939 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
942 struct net_device *dev, *ret;
947 for_each_netdev(net, dev) {
948 if (((dev->flags ^ if_flags) & mask) == 0) {
955 EXPORT_SYMBOL(__dev_get_by_flags);
958 * dev_valid_name - check if name is okay for network device
961 * Network device names need to be valid file names to
962 * to allow sysfs to work. We also disallow any kind of
965 bool dev_valid_name(const char *name)
969 if (strlen(name) >= IFNAMSIZ)
971 if (!strcmp(name, ".") || !strcmp(name, ".."))
975 if (*name == '/' || *name == ':' || isspace(*name))
981 EXPORT_SYMBOL(dev_valid_name);
984 * __dev_alloc_name - allocate a name for a device
985 * @net: network namespace to allocate the device name in
986 * @name: name format string
987 * @buf: scratch buffer and result name string
989 * Passed a format string - eg "lt%d" it will try and find a suitable
990 * id. It scans list of devices to build up a free map, then chooses
991 * the first empty slot. The caller must hold the dev_base or rtnl lock
992 * while allocating the name and adding the device in order to avoid
994 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
995 * Returns the number of the unit assigned or a negative errno code.
998 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1002 const int max_netdevices = 8*PAGE_SIZE;
1003 unsigned long *inuse;
1004 struct net_device *d;
1006 p = strnchr(name, IFNAMSIZ-1, '%');
1009 * Verify the string as this thing may have come from
1010 * the user. There must be either one "%d" and no other "%"
1013 if (p[1] != 'd' || strchr(p + 2, '%'))
1016 /* Use one page as a bit array of possible slots */
1017 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1021 for_each_netdev(net, d) {
1022 if (!sscanf(d->name, name, &i))
1024 if (i < 0 || i >= max_netdevices)
1027 /* avoid cases where sscanf is not exact inverse of printf */
1028 snprintf(buf, IFNAMSIZ, name, i);
1029 if (!strncmp(buf, d->name, IFNAMSIZ))
1033 i = find_first_zero_bit(inuse, max_netdevices);
1034 free_page((unsigned long) inuse);
1038 snprintf(buf, IFNAMSIZ, name, i);
1039 if (!__dev_get_by_name(net, buf))
1042 /* It is possible to run out of possible slots
1043 * when the name is long and there isn't enough space left
1044 * for the digits, or if all bits are used.
1050 * dev_alloc_name - allocate a name for a device
1052 * @name: name format string
1054 * Passed a format string - eg "lt%d" it will try and find a suitable
1055 * id. It scans list of devices to build up a free map, then chooses
1056 * the first empty slot. The caller must hold the dev_base or rtnl lock
1057 * while allocating the name and adding the device in order to avoid
1059 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1060 * Returns the number of the unit assigned or a negative errno code.
1063 int dev_alloc_name(struct net_device *dev, const char *name)
1069 BUG_ON(!dev_net(dev));
1071 ret = __dev_alloc_name(net, name, buf);
1073 strlcpy(dev->name, buf, IFNAMSIZ);
1076 EXPORT_SYMBOL(dev_alloc_name);
1078 static int dev_alloc_name_ns(struct net *net,
1079 struct net_device *dev,
1085 ret = __dev_alloc_name(net, name, buf);
1087 strlcpy(dev->name, buf, IFNAMSIZ);
1091 static int dev_get_valid_name(struct net *net,
1092 struct net_device *dev,
1097 if (!dev_valid_name(name))
1100 if (strchr(name, '%'))
1101 return dev_alloc_name_ns(net, dev, name);
1102 else if (__dev_get_by_name(net, name))
1104 else if (dev->name != name)
1105 strlcpy(dev->name, name, IFNAMSIZ);
1111 * dev_change_name - change name of a device
1113 * @newname: name (or format string) must be at least IFNAMSIZ
1115 * Change name of a device, can pass format strings "eth%d".
1118 int dev_change_name(struct net_device *dev, const char *newname)
1120 unsigned char old_assign_type;
1121 char oldname[IFNAMSIZ];
1127 BUG_ON(!dev_net(dev));
1130 if (dev->flags & IFF_UP)
1133 write_seqcount_begin(&devnet_rename_seq);
1135 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1136 write_seqcount_end(&devnet_rename_seq);
1140 memcpy(oldname, dev->name, IFNAMSIZ);
1142 err = dev_get_valid_name(net, dev, newname);
1144 write_seqcount_end(&devnet_rename_seq);
1148 if (oldname[0] && !strchr(oldname, '%'))
1149 netdev_info(dev, "renamed from %s\n", oldname);
1151 old_assign_type = dev->name_assign_type;
1152 dev->name_assign_type = NET_NAME_RENAMED;
1155 ret = device_rename(&dev->dev, dev->name);
1157 memcpy(dev->name, oldname, IFNAMSIZ);
1158 dev->name_assign_type = old_assign_type;
1159 write_seqcount_end(&devnet_rename_seq);
1163 write_seqcount_end(&devnet_rename_seq);
1165 netdev_adjacent_rename_links(dev, oldname);
1167 write_lock_bh(&dev_base_lock);
1168 hlist_del_rcu(&dev->name_hlist);
1169 write_unlock_bh(&dev_base_lock);
1173 write_lock_bh(&dev_base_lock);
1174 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1175 write_unlock_bh(&dev_base_lock);
1177 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1178 ret = notifier_to_errno(ret);
1181 /* err >= 0 after dev_alloc_name() or stores the first errno */
1184 write_seqcount_begin(&devnet_rename_seq);
1185 memcpy(dev->name, oldname, IFNAMSIZ);
1186 memcpy(oldname, newname, IFNAMSIZ);
1187 dev->name_assign_type = old_assign_type;
1188 old_assign_type = NET_NAME_RENAMED;
1191 pr_err("%s: name change rollback failed: %d\n",
1200 * dev_set_alias - change ifalias of a device
1202 * @alias: name up to IFALIASZ
1203 * @len: limit of bytes to copy from info
1205 * Set ifalias for a device,
1207 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1213 if (len >= IFALIASZ)
1217 kfree(dev->ifalias);
1218 dev->ifalias = NULL;
1222 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1225 dev->ifalias = new_ifalias;
1227 strlcpy(dev->ifalias, alias, len+1);
1233 * netdev_features_change - device changes features
1234 * @dev: device to cause notification
1236 * Called to indicate a device has changed features.
1238 void netdev_features_change(struct net_device *dev)
1240 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1242 EXPORT_SYMBOL(netdev_features_change);
1245 * netdev_state_change - device changes state
1246 * @dev: device to cause notification
1248 * Called to indicate a device has changed state. This function calls
1249 * the notifier chains for netdev_chain and sends a NEWLINK message
1250 * to the routing socket.
1252 void netdev_state_change(struct net_device *dev)
1254 if (dev->flags & IFF_UP) {
1255 struct netdev_notifier_change_info change_info;
1257 change_info.flags_changed = 0;
1258 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1260 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1263 EXPORT_SYMBOL(netdev_state_change);
1266 * netdev_notify_peers - notify network peers about existence of @dev
1267 * @dev: network device
1269 * Generate traffic such that interested network peers are aware of
1270 * @dev, such as by generating a gratuitous ARP. This may be used when
1271 * a device wants to inform the rest of the network about some sort of
1272 * reconfiguration such as a failover event or virtual machine
1275 void netdev_notify_peers(struct net_device *dev)
1278 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1281 EXPORT_SYMBOL(netdev_notify_peers);
1283 static int __dev_open(struct net_device *dev)
1285 const struct net_device_ops *ops = dev->netdev_ops;
1290 if (!netif_device_present(dev))
1293 /* Block netpoll from trying to do any rx path servicing.
1294 * If we don't do this there is a chance ndo_poll_controller
1295 * or ndo_poll may be running while we open the device
1297 netpoll_poll_disable(dev);
1299 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1300 ret = notifier_to_errno(ret);
1304 set_bit(__LINK_STATE_START, &dev->state);
1306 if (ops->ndo_validate_addr)
1307 ret = ops->ndo_validate_addr(dev);
1309 if (!ret && ops->ndo_open)
1310 ret = ops->ndo_open(dev);
1312 netpoll_poll_enable(dev);
1315 clear_bit(__LINK_STATE_START, &dev->state);
1317 dev->flags |= IFF_UP;
1318 dev_set_rx_mode(dev);
1320 add_device_randomness(dev->dev_addr, dev->addr_len);
1327 * dev_open - prepare an interface for use.
1328 * @dev: device to open
1330 * Takes a device from down to up state. The device's private open
1331 * function is invoked and then the multicast lists are loaded. Finally
1332 * the device is moved into the up state and a %NETDEV_UP message is
1333 * sent to the netdev notifier chain.
1335 * Calling this function on an active interface is a nop. On a failure
1336 * a negative errno code is returned.
1338 int dev_open(struct net_device *dev)
1342 if (dev->flags & IFF_UP)
1345 ret = __dev_open(dev);
1349 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1350 call_netdevice_notifiers(NETDEV_UP, dev);
1354 EXPORT_SYMBOL(dev_open);
1356 static int __dev_close_many(struct list_head *head)
1358 struct net_device *dev;
1363 list_for_each_entry(dev, head, close_list) {
1364 /* Temporarily disable netpoll until the interface is down */
1365 netpoll_poll_disable(dev);
1367 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1369 clear_bit(__LINK_STATE_START, &dev->state);
1371 /* Synchronize to scheduled poll. We cannot touch poll list, it
1372 * can be even on different cpu. So just clear netif_running().
1374 * dev->stop() will invoke napi_disable() on all of it's
1375 * napi_struct instances on this device.
1377 smp_mb__after_atomic(); /* Commit netif_running(). */
1380 dev_deactivate_many(head);
1382 list_for_each_entry(dev, head, close_list) {
1383 const struct net_device_ops *ops = dev->netdev_ops;
1386 * Call the device specific close. This cannot fail.
1387 * Only if device is UP
1389 * We allow it to be called even after a DETACH hot-plug
1395 dev->flags &= ~IFF_UP;
1396 netpoll_poll_enable(dev);
1402 static int __dev_close(struct net_device *dev)
1407 list_add(&dev->close_list, &single);
1408 retval = __dev_close_many(&single);
1414 int dev_close_many(struct list_head *head, bool unlink)
1416 struct net_device *dev, *tmp;
1418 /* Remove the devices that don't need to be closed */
1419 list_for_each_entry_safe(dev, tmp, head, close_list)
1420 if (!(dev->flags & IFF_UP))
1421 list_del_init(&dev->close_list);
1423 __dev_close_many(head);
1425 list_for_each_entry_safe(dev, tmp, head, close_list) {
1426 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1427 call_netdevice_notifiers(NETDEV_DOWN, dev);
1429 list_del_init(&dev->close_list);
1434 EXPORT_SYMBOL(dev_close_many);
1437 * dev_close - shutdown an interface.
1438 * @dev: device to shutdown
1440 * This function moves an active device into down state. A
1441 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1442 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1445 int dev_close(struct net_device *dev)
1447 if (dev->flags & IFF_UP) {
1450 list_add(&dev->close_list, &single);
1451 dev_close_many(&single, true);
1456 EXPORT_SYMBOL(dev_close);
1460 * dev_disable_lro - disable Large Receive Offload on a device
1463 * Disable Large Receive Offload (LRO) on a net device. Must be
1464 * called under RTNL. This is needed if received packets may be
1465 * forwarded to another interface.
1467 void dev_disable_lro(struct net_device *dev)
1469 struct net_device *lower_dev;
1470 struct list_head *iter;
1472 dev->wanted_features &= ~NETIF_F_LRO;
1473 netdev_update_features(dev);
1475 if (unlikely(dev->features & NETIF_F_LRO))
1476 netdev_WARN(dev, "failed to disable LRO!\n");
1478 netdev_for_each_lower_dev(dev, lower_dev, iter)
1479 dev_disable_lro(lower_dev);
1481 EXPORT_SYMBOL(dev_disable_lro);
1483 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1484 struct net_device *dev)
1486 struct netdev_notifier_info info;
1488 netdev_notifier_info_init(&info, dev);
1489 return nb->notifier_call(nb, val, &info);
1492 static int dev_boot_phase = 1;
1495 * register_netdevice_notifier - register a network notifier block
1498 * Register a notifier to be called when network device events occur.
1499 * The notifier passed is linked into the kernel structures and must
1500 * not be reused until it has been unregistered. A negative errno code
1501 * is returned on a failure.
1503 * When registered all registration and up events are replayed
1504 * to the new notifier to allow device to have a race free
1505 * view of the network device list.
1508 int register_netdevice_notifier(struct notifier_block *nb)
1510 struct net_device *dev;
1511 struct net_device *last;
1516 err = raw_notifier_chain_register(&netdev_chain, nb);
1522 for_each_netdev(net, dev) {
1523 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1524 err = notifier_to_errno(err);
1528 if (!(dev->flags & IFF_UP))
1531 call_netdevice_notifier(nb, NETDEV_UP, dev);
1542 for_each_netdev(net, dev) {
1546 if (dev->flags & IFF_UP) {
1547 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1549 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1551 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1556 raw_notifier_chain_unregister(&netdev_chain, nb);
1559 EXPORT_SYMBOL(register_netdevice_notifier);
1562 * unregister_netdevice_notifier - unregister a network notifier block
1565 * Unregister a notifier previously registered by
1566 * register_netdevice_notifier(). The notifier is unlinked into the
1567 * kernel structures and may then be reused. A negative errno code
1568 * is returned on a failure.
1570 * After unregistering unregister and down device events are synthesized
1571 * for all devices on the device list to the removed notifier to remove
1572 * the need for special case cleanup code.
1575 int unregister_netdevice_notifier(struct notifier_block *nb)
1577 struct net_device *dev;
1582 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1587 for_each_netdev(net, dev) {
1588 if (dev->flags & IFF_UP) {
1589 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1591 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1593 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1600 EXPORT_SYMBOL(unregister_netdevice_notifier);
1603 * call_netdevice_notifiers_info - call all network notifier blocks
1604 * @val: value passed unmodified to notifier function
1605 * @dev: net_device pointer passed unmodified to notifier function
1606 * @info: notifier information data
1608 * Call all network notifier blocks. Parameters and return value
1609 * are as for raw_notifier_call_chain().
1612 static int call_netdevice_notifiers_info(unsigned long val,
1613 struct net_device *dev,
1614 struct netdev_notifier_info *info)
1617 netdev_notifier_info_init(info, dev);
1618 return raw_notifier_call_chain(&netdev_chain, val, info);
1622 * call_netdevice_notifiers - call all network notifier blocks
1623 * @val: value passed unmodified to notifier function
1624 * @dev: net_device pointer passed unmodified to notifier function
1626 * Call all network notifier blocks. Parameters and return value
1627 * are as for raw_notifier_call_chain().
1630 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1632 struct netdev_notifier_info info;
1634 return call_netdevice_notifiers_info(val, dev, &info);
1636 EXPORT_SYMBOL(call_netdevice_notifiers);
1638 #ifdef CONFIG_NET_INGRESS
1639 static struct static_key ingress_needed __read_mostly;
1641 void net_inc_ingress_queue(void)
1643 static_key_slow_inc(&ingress_needed);
1645 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1647 void net_dec_ingress_queue(void)
1649 static_key_slow_dec(&ingress_needed);
1651 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1654 static struct static_key netstamp_needed __read_mostly;
1655 #ifdef HAVE_JUMP_LABEL
1656 /* We are not allowed to call static_key_slow_dec() from irq context
1657 * If net_disable_timestamp() is called from irq context, defer the
1658 * static_key_slow_dec() calls.
1660 static atomic_t netstamp_needed_deferred;
1663 void net_enable_timestamp(void)
1665 #ifdef HAVE_JUMP_LABEL
1666 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1670 static_key_slow_dec(&netstamp_needed);
1674 static_key_slow_inc(&netstamp_needed);
1676 EXPORT_SYMBOL(net_enable_timestamp);
1678 void net_disable_timestamp(void)
1680 #ifdef HAVE_JUMP_LABEL
1681 if (in_interrupt()) {
1682 atomic_inc(&netstamp_needed_deferred);
1686 static_key_slow_dec(&netstamp_needed);
1688 EXPORT_SYMBOL(net_disable_timestamp);
1690 static inline void net_timestamp_set(struct sk_buff *skb)
1692 skb->tstamp.tv64 = 0;
1693 if (static_key_false(&netstamp_needed))
1694 __net_timestamp(skb);
1697 #define net_timestamp_check(COND, SKB) \
1698 if (static_key_false(&netstamp_needed)) { \
1699 if ((COND) && !(SKB)->tstamp.tv64) \
1700 __net_timestamp(SKB); \
1703 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1707 if (!(dev->flags & IFF_UP))
1710 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1711 if (skb->len <= len)
1714 /* if TSO is enabled, we don't care about the length as the packet
1715 * could be forwarded without being segmented before
1717 if (skb_is_gso(skb))
1722 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1724 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1726 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1727 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1728 atomic_long_inc(&dev->rx_dropped);
1734 if (unlikely(!is_skb_forwardable(dev, skb))) {
1735 atomic_long_inc(&dev->rx_dropped);
1740 skb_scrub_packet(skb, true);
1742 skb->protocol = eth_type_trans(skb, dev);
1743 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1747 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1750 * dev_forward_skb - loopback an skb to another netif
1752 * @dev: destination network device
1753 * @skb: buffer to forward
1756 * NET_RX_SUCCESS (no congestion)
1757 * NET_RX_DROP (packet was dropped, but freed)
1759 * dev_forward_skb can be used for injecting an skb from the
1760 * start_xmit function of one device into the receive queue
1761 * of another device.
1763 * The receiving device may be in another namespace, so
1764 * we have to clear all information in the skb that could
1765 * impact namespace isolation.
1767 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1769 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1771 EXPORT_SYMBOL_GPL(dev_forward_skb);
1773 static inline int deliver_skb(struct sk_buff *skb,
1774 struct packet_type *pt_prev,
1775 struct net_device *orig_dev)
1777 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1779 atomic_inc(&skb->users);
1780 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1783 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1784 struct packet_type **pt,
1785 struct net_device *orig_dev,
1787 struct list_head *ptype_list)
1789 struct packet_type *ptype, *pt_prev = *pt;
1791 list_for_each_entry_rcu(ptype, ptype_list, list) {
1792 if (ptype->type != type)
1795 deliver_skb(skb, pt_prev, orig_dev);
1801 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1803 if (!ptype->af_packet_priv || !skb->sk)
1806 if (ptype->id_match)
1807 return ptype->id_match(ptype, skb->sk);
1808 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1815 * Support routine. Sends outgoing frames to any network
1816 * taps currently in use.
1819 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1821 struct packet_type *ptype;
1822 struct sk_buff *skb2 = NULL;
1823 struct packet_type *pt_prev = NULL;
1824 struct list_head *ptype_list = &ptype_all;
1828 list_for_each_entry_rcu(ptype, ptype_list, list) {
1829 /* Never send packets back to the socket
1830 * they originated from - MvS (miquels@drinkel.ow.org)
1832 if (skb_loop_sk(ptype, skb))
1836 deliver_skb(skb2, pt_prev, skb->dev);
1841 /* need to clone skb, done only once */
1842 skb2 = skb_clone(skb, GFP_ATOMIC);
1846 net_timestamp_set(skb2);
1848 /* skb->nh should be correctly
1849 * set by sender, so that the second statement is
1850 * just protection against buggy protocols.
1852 skb_reset_mac_header(skb2);
1854 if (skb_network_header(skb2) < skb2->data ||
1855 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1856 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1857 ntohs(skb2->protocol),
1859 skb_reset_network_header(skb2);
1862 skb2->transport_header = skb2->network_header;
1863 skb2->pkt_type = PACKET_OUTGOING;
1867 if (ptype_list == &ptype_all) {
1868 ptype_list = &dev->ptype_all;
1873 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1878 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1879 * @dev: Network device
1880 * @txq: number of queues available
1882 * If real_num_tx_queues is changed the tc mappings may no longer be
1883 * valid. To resolve this verify the tc mapping remains valid and if
1884 * not NULL the mapping. With no priorities mapping to this
1885 * offset/count pair it will no longer be used. In the worst case TC0
1886 * is invalid nothing can be done so disable priority mappings. If is
1887 * expected that drivers will fix this mapping if they can before
1888 * calling netif_set_real_num_tx_queues.
1890 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1893 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1895 /* If TC0 is invalidated disable TC mapping */
1896 if (tc->offset + tc->count > txq) {
1897 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1902 /* Invalidated prio to tc mappings set to TC0 */
1903 for (i = 1; i < TC_BITMASK + 1; i++) {
1904 int q = netdev_get_prio_tc_map(dev, i);
1906 tc = &dev->tc_to_txq[q];
1907 if (tc->offset + tc->count > txq) {
1908 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1910 netdev_set_prio_tc_map(dev, i, 0);
1916 static DEFINE_MUTEX(xps_map_mutex);
1917 #define xmap_dereference(P) \
1918 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1920 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1923 struct xps_map *map = NULL;
1927 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1929 for (pos = 0; map && pos < map->len; pos++) {
1930 if (map->queues[pos] == index) {
1932 map->queues[pos] = map->queues[--map->len];
1934 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1935 kfree_rcu(map, rcu);
1945 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1947 struct xps_dev_maps *dev_maps;
1949 bool active = false;
1951 mutex_lock(&xps_map_mutex);
1952 dev_maps = xmap_dereference(dev->xps_maps);
1957 for_each_possible_cpu(cpu) {
1958 for (i = index; i < dev->num_tx_queues; i++) {
1959 if (!remove_xps_queue(dev_maps, cpu, i))
1962 if (i == dev->num_tx_queues)
1967 RCU_INIT_POINTER(dev->xps_maps, NULL);
1968 kfree_rcu(dev_maps, rcu);
1971 for (i = index; i < dev->num_tx_queues; i++)
1972 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1976 mutex_unlock(&xps_map_mutex);
1979 static struct xps_map *expand_xps_map(struct xps_map *map,
1982 struct xps_map *new_map;
1983 int alloc_len = XPS_MIN_MAP_ALLOC;
1986 for (pos = 0; map && pos < map->len; pos++) {
1987 if (map->queues[pos] != index)
1992 /* Need to add queue to this CPU's existing map */
1994 if (pos < map->alloc_len)
1997 alloc_len = map->alloc_len * 2;
2000 /* Need to allocate new map to store queue on this CPU's map */
2001 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2006 for (i = 0; i < pos; i++)
2007 new_map->queues[i] = map->queues[i];
2008 new_map->alloc_len = alloc_len;
2014 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2017 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2018 struct xps_map *map, *new_map;
2019 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2020 int cpu, numa_node_id = -2;
2021 bool active = false;
2023 mutex_lock(&xps_map_mutex);
2025 dev_maps = xmap_dereference(dev->xps_maps);
2027 /* allocate memory for queue storage */
2028 for_each_online_cpu(cpu) {
2029 if (!cpumask_test_cpu(cpu, mask))
2033 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2034 if (!new_dev_maps) {
2035 mutex_unlock(&xps_map_mutex);
2039 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2042 map = expand_xps_map(map, cpu, index);
2046 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2050 goto out_no_new_maps;
2052 for_each_possible_cpu(cpu) {
2053 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2054 /* add queue to CPU maps */
2057 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2058 while ((pos < map->len) && (map->queues[pos] != index))
2061 if (pos == map->len)
2062 map->queues[map->len++] = index;
2064 if (numa_node_id == -2)
2065 numa_node_id = cpu_to_node(cpu);
2066 else if (numa_node_id != cpu_to_node(cpu))
2069 } else if (dev_maps) {
2070 /* fill in the new device map from the old device map */
2071 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2072 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2077 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2079 /* Cleanup old maps */
2081 for_each_possible_cpu(cpu) {
2082 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2083 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2084 if (map && map != new_map)
2085 kfree_rcu(map, rcu);
2088 kfree_rcu(dev_maps, rcu);
2091 dev_maps = new_dev_maps;
2095 /* update Tx queue numa node */
2096 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2097 (numa_node_id >= 0) ? numa_node_id :
2103 /* removes queue from unused CPUs */
2104 for_each_possible_cpu(cpu) {
2105 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2108 if (remove_xps_queue(dev_maps, cpu, index))
2112 /* free map if not active */
2114 RCU_INIT_POINTER(dev->xps_maps, NULL);
2115 kfree_rcu(dev_maps, rcu);
2119 mutex_unlock(&xps_map_mutex);
2123 /* remove any maps that we added */
2124 for_each_possible_cpu(cpu) {
2125 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2126 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2128 if (new_map && new_map != map)
2132 mutex_unlock(&xps_map_mutex);
2134 kfree(new_dev_maps);
2137 EXPORT_SYMBOL(netif_set_xps_queue);
2141 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2142 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2144 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2148 if (txq < 1 || txq > dev->num_tx_queues)
2151 if (dev->reg_state == NETREG_REGISTERED ||
2152 dev->reg_state == NETREG_UNREGISTERING) {
2155 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2161 netif_setup_tc(dev, txq);
2163 if (txq < dev->real_num_tx_queues) {
2164 qdisc_reset_all_tx_gt(dev, txq);
2166 netif_reset_xps_queues_gt(dev, txq);
2171 dev->real_num_tx_queues = txq;
2174 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2178 * netif_set_real_num_rx_queues - set actual number of RX queues used
2179 * @dev: Network device
2180 * @rxq: Actual number of RX queues
2182 * This must be called either with the rtnl_lock held or before
2183 * registration of the net device. Returns 0 on success, or a
2184 * negative error code. If called before registration, it always
2187 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2191 if (rxq < 1 || rxq > dev->num_rx_queues)
2194 if (dev->reg_state == NETREG_REGISTERED) {
2197 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2203 dev->real_num_rx_queues = rxq;
2206 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2210 * netif_get_num_default_rss_queues - default number of RSS queues
2212 * This routine should set an upper limit on the number of RSS queues
2213 * used by default by multiqueue devices.
2215 int netif_get_num_default_rss_queues(void)
2217 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2219 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2221 static inline void __netif_reschedule(struct Qdisc *q)
2223 struct softnet_data *sd;
2224 unsigned long flags;
2226 local_irq_save(flags);
2227 sd = this_cpu_ptr(&softnet_data);
2228 q->next_sched = NULL;
2229 *sd->output_queue_tailp = q;
2230 sd->output_queue_tailp = &q->next_sched;
2231 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2232 local_irq_restore(flags);
2235 void __netif_schedule(struct Qdisc *q)
2237 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2238 __netif_reschedule(q);
2240 EXPORT_SYMBOL(__netif_schedule);
2242 struct dev_kfree_skb_cb {
2243 enum skb_free_reason reason;
2246 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2248 return (struct dev_kfree_skb_cb *)skb->cb;
2251 void netif_schedule_queue(struct netdev_queue *txq)
2254 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2255 struct Qdisc *q = rcu_dereference(txq->qdisc);
2257 __netif_schedule(q);
2261 EXPORT_SYMBOL(netif_schedule_queue);
2264 * netif_wake_subqueue - allow sending packets on subqueue
2265 * @dev: network device
2266 * @queue_index: sub queue index
2268 * Resume individual transmit queue of a device with multiple transmit queues.
2270 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2272 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2274 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2278 q = rcu_dereference(txq->qdisc);
2279 __netif_schedule(q);
2283 EXPORT_SYMBOL(netif_wake_subqueue);
2285 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2287 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2291 q = rcu_dereference(dev_queue->qdisc);
2292 __netif_schedule(q);
2296 EXPORT_SYMBOL(netif_tx_wake_queue);
2298 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2300 unsigned long flags;
2302 if (likely(atomic_read(&skb->users) == 1)) {
2304 atomic_set(&skb->users, 0);
2305 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2308 get_kfree_skb_cb(skb)->reason = reason;
2309 local_irq_save(flags);
2310 skb->next = __this_cpu_read(softnet_data.completion_queue);
2311 __this_cpu_write(softnet_data.completion_queue, skb);
2312 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2313 local_irq_restore(flags);
2315 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2317 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2319 if (in_irq() || irqs_disabled())
2320 __dev_kfree_skb_irq(skb, reason);
2324 EXPORT_SYMBOL(__dev_kfree_skb_any);
2328 * netif_device_detach - mark device as removed
2329 * @dev: network device
2331 * Mark device as removed from system and therefore no longer available.
2333 void netif_device_detach(struct net_device *dev)
2335 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2336 netif_running(dev)) {
2337 netif_tx_stop_all_queues(dev);
2340 EXPORT_SYMBOL(netif_device_detach);
2343 * netif_device_attach - mark device as attached
2344 * @dev: network device
2346 * Mark device as attached from system and restart if needed.
2348 void netif_device_attach(struct net_device *dev)
2350 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2351 netif_running(dev)) {
2352 netif_tx_wake_all_queues(dev);
2353 __netdev_watchdog_up(dev);
2356 EXPORT_SYMBOL(netif_device_attach);
2359 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2360 * to be used as a distribution range.
2362 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2363 unsigned int num_tx_queues)
2367 u16 qcount = num_tx_queues;
2369 if (skb_rx_queue_recorded(skb)) {
2370 hash = skb_get_rx_queue(skb);
2371 while (unlikely(hash >= num_tx_queues))
2372 hash -= num_tx_queues;
2377 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2378 qoffset = dev->tc_to_txq[tc].offset;
2379 qcount = dev->tc_to_txq[tc].count;
2382 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2384 EXPORT_SYMBOL(__skb_tx_hash);
2386 static void skb_warn_bad_offload(const struct sk_buff *skb)
2388 static const netdev_features_t null_features = 0;
2389 struct net_device *dev = skb->dev;
2390 const char *driver = "";
2392 if (!net_ratelimit())
2395 if (dev && dev->dev.parent)
2396 driver = dev_driver_string(dev->dev.parent);
2398 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2399 "gso_type=%d ip_summed=%d\n",
2400 driver, dev ? &dev->features : &null_features,
2401 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2402 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2403 skb_shinfo(skb)->gso_type, skb->ip_summed);
2407 * Invalidate hardware checksum when packet is to be mangled, and
2408 * complete checksum manually on outgoing path.
2410 int skb_checksum_help(struct sk_buff *skb)
2413 int ret = 0, offset;
2415 if (skb->ip_summed == CHECKSUM_COMPLETE)
2416 goto out_set_summed;
2418 if (unlikely(skb_shinfo(skb)->gso_size)) {
2419 skb_warn_bad_offload(skb);
2423 /* Before computing a checksum, we should make sure no frag could
2424 * be modified by an external entity : checksum could be wrong.
2426 if (skb_has_shared_frag(skb)) {
2427 ret = __skb_linearize(skb);
2432 offset = skb_checksum_start_offset(skb);
2433 BUG_ON(offset >= skb_headlen(skb));
2434 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2436 offset += skb->csum_offset;
2437 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2439 if (skb_cloned(skb) &&
2440 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2441 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2446 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2448 skb->ip_summed = CHECKSUM_NONE;
2452 EXPORT_SYMBOL(skb_checksum_help);
2454 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2456 __be16 type = skb->protocol;
2458 /* Tunnel gso handlers can set protocol to ethernet. */
2459 if (type == htons(ETH_P_TEB)) {
2462 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2465 eth = (struct ethhdr *)skb_mac_header(skb);
2466 type = eth->h_proto;
2469 return __vlan_get_protocol(skb, type, depth);
2473 * skb_mac_gso_segment - mac layer segmentation handler.
2474 * @skb: buffer to segment
2475 * @features: features for the output path (see dev->features)
2477 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2478 netdev_features_t features)
2480 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2481 struct packet_offload *ptype;
2482 int vlan_depth = skb->mac_len;
2483 __be16 type = skb_network_protocol(skb, &vlan_depth);
2485 if (unlikely(!type))
2486 return ERR_PTR(-EINVAL);
2488 __skb_pull(skb, vlan_depth);
2491 list_for_each_entry_rcu(ptype, &offload_base, list) {
2492 if (ptype->type == type && ptype->callbacks.gso_segment) {
2493 segs = ptype->callbacks.gso_segment(skb, features);
2499 __skb_push(skb, skb->data - skb_mac_header(skb));
2503 EXPORT_SYMBOL(skb_mac_gso_segment);
2506 /* openvswitch calls this on rx path, so we need a different check.
2508 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2511 return skb->ip_summed != CHECKSUM_PARTIAL;
2513 return skb->ip_summed == CHECKSUM_NONE;
2517 * __skb_gso_segment - Perform segmentation on skb.
2518 * @skb: buffer to segment
2519 * @features: features for the output path (see dev->features)
2520 * @tx_path: whether it is called in TX path
2522 * This function segments the given skb and returns a list of segments.
2524 * It may return NULL if the skb requires no segmentation. This is
2525 * only possible when GSO is used for verifying header integrity.
2527 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2528 netdev_features_t features, bool tx_path)
2530 if (unlikely(skb_needs_check(skb, tx_path))) {
2533 skb_warn_bad_offload(skb);
2535 err = skb_cow_head(skb, 0);
2537 return ERR_PTR(err);
2540 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2541 SKB_GSO_CB(skb)->encap_level = 0;
2543 skb_reset_mac_header(skb);
2544 skb_reset_mac_len(skb);
2546 return skb_mac_gso_segment(skb, features);
2548 EXPORT_SYMBOL(__skb_gso_segment);
2550 /* Take action when hardware reception checksum errors are detected. */
2552 void netdev_rx_csum_fault(struct net_device *dev)
2554 if (net_ratelimit()) {
2555 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2559 EXPORT_SYMBOL(netdev_rx_csum_fault);
2562 /* Actually, we should eliminate this check as soon as we know, that:
2563 * 1. IOMMU is present and allows to map all the memory.
2564 * 2. No high memory really exists on this machine.
2567 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2569 #ifdef CONFIG_HIGHMEM
2571 if (!(dev->features & NETIF_F_HIGHDMA)) {
2572 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2573 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2574 if (PageHighMem(skb_frag_page(frag)))
2579 if (PCI_DMA_BUS_IS_PHYS) {
2580 struct device *pdev = dev->dev.parent;
2584 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2585 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2586 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2587 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2595 /* If MPLS offload request, verify we are testing hardware MPLS features
2596 * instead of standard features for the netdev.
2598 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2599 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2600 netdev_features_t features,
2603 if (eth_p_mpls(type))
2604 features &= skb->dev->mpls_features;
2609 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2610 netdev_features_t features,
2617 static netdev_features_t harmonize_features(struct sk_buff *skb,
2618 netdev_features_t features)
2623 type = skb_network_protocol(skb, &tmp);
2624 features = net_mpls_features(skb, features, type);
2626 if (skb->ip_summed != CHECKSUM_NONE &&
2627 !can_checksum_protocol(features, type)) {
2628 features &= ~NETIF_F_ALL_CSUM;
2629 } else if (illegal_highdma(skb->dev, skb)) {
2630 features &= ~NETIF_F_SG;
2636 netdev_features_t passthru_features_check(struct sk_buff *skb,
2637 struct net_device *dev,
2638 netdev_features_t features)
2642 EXPORT_SYMBOL(passthru_features_check);
2644 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2645 struct net_device *dev,
2646 netdev_features_t features)
2648 return vlan_features_check(skb, features);
2651 netdev_features_t netif_skb_features(struct sk_buff *skb)
2653 struct net_device *dev = skb->dev;
2654 netdev_features_t features = dev->features;
2655 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2657 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2658 features &= ~NETIF_F_GSO_MASK;
2660 /* If encapsulation offload request, verify we are testing
2661 * hardware encapsulation features instead of standard
2662 * features for the netdev
2664 if (skb->encapsulation)
2665 features &= dev->hw_enc_features;
2667 if (skb_vlan_tagged(skb))
2668 features = netdev_intersect_features(features,
2669 dev->vlan_features |
2670 NETIF_F_HW_VLAN_CTAG_TX |
2671 NETIF_F_HW_VLAN_STAG_TX);
2673 if (dev->netdev_ops->ndo_features_check)
2674 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2677 features &= dflt_features_check(skb, dev, features);
2679 return harmonize_features(skb, features);
2681 EXPORT_SYMBOL(netif_skb_features);
2683 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2684 struct netdev_queue *txq, bool more)
2689 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2690 dev_queue_xmit_nit(skb, dev);
2693 trace_net_dev_start_xmit(skb, dev);
2694 rc = netdev_start_xmit(skb, dev, txq, more);
2695 trace_net_dev_xmit(skb, rc, dev, len);
2700 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2701 struct netdev_queue *txq, int *ret)
2703 struct sk_buff *skb = first;
2704 int rc = NETDEV_TX_OK;
2707 struct sk_buff *next = skb->next;
2710 rc = xmit_one(skb, dev, txq, next != NULL);
2711 if (unlikely(!dev_xmit_complete(rc))) {
2717 if (netif_xmit_stopped(txq) && skb) {
2718 rc = NETDEV_TX_BUSY;
2728 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2729 netdev_features_t features)
2731 if (skb_vlan_tag_present(skb) &&
2732 !vlan_hw_offload_capable(features, skb->vlan_proto))
2733 skb = __vlan_hwaccel_push_inside(skb);
2737 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2739 netdev_features_t features;
2744 features = netif_skb_features(skb);
2745 skb = validate_xmit_vlan(skb, features);
2749 if (netif_needs_gso(skb, features)) {
2750 struct sk_buff *segs;
2752 segs = skb_gso_segment(skb, features);
2760 if (skb_needs_linearize(skb, features) &&
2761 __skb_linearize(skb))
2764 /* If packet is not checksummed and device does not
2765 * support checksumming for this protocol, complete
2766 * checksumming here.
2768 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2769 if (skb->encapsulation)
2770 skb_set_inner_transport_header(skb,
2771 skb_checksum_start_offset(skb));
2773 skb_set_transport_header(skb,
2774 skb_checksum_start_offset(skb));
2775 if (!(features & NETIF_F_ALL_CSUM) &&
2776 skb_checksum_help(skb))
2789 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2791 struct sk_buff *next, *head = NULL, *tail;
2793 for (; skb != NULL; skb = next) {
2797 /* in case skb wont be segmented, point to itself */
2800 skb = validate_xmit_skb(skb, dev);
2808 /* If skb was segmented, skb->prev points to
2809 * the last segment. If not, it still contains skb.
2816 static void qdisc_pkt_len_init(struct sk_buff *skb)
2818 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2820 qdisc_skb_cb(skb)->pkt_len = skb->len;
2822 /* To get more precise estimation of bytes sent on wire,
2823 * we add to pkt_len the headers size of all segments
2825 if (shinfo->gso_size) {
2826 unsigned int hdr_len;
2827 u16 gso_segs = shinfo->gso_segs;
2829 /* mac layer + network layer */
2830 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2832 /* + transport layer */
2833 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2834 hdr_len += tcp_hdrlen(skb);
2836 hdr_len += sizeof(struct udphdr);
2838 if (shinfo->gso_type & SKB_GSO_DODGY)
2839 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2842 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2846 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2847 struct net_device *dev,
2848 struct netdev_queue *txq)
2850 spinlock_t *root_lock = qdisc_lock(q);
2854 qdisc_pkt_len_init(skb);
2855 qdisc_calculate_pkt_len(skb, q);
2857 * Heuristic to force contended enqueues to serialize on a
2858 * separate lock before trying to get qdisc main lock.
2859 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2860 * often and dequeue packets faster.
2862 contended = qdisc_is_running(q);
2863 if (unlikely(contended))
2864 spin_lock(&q->busylock);
2866 spin_lock(root_lock);
2867 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2870 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2871 qdisc_run_begin(q)) {
2873 * This is a work-conserving queue; there are no old skbs
2874 * waiting to be sent out; and the qdisc is not running -
2875 * xmit the skb directly.
2878 qdisc_bstats_update(q, skb);
2880 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2881 if (unlikely(contended)) {
2882 spin_unlock(&q->busylock);
2889 rc = NET_XMIT_SUCCESS;
2891 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2892 if (qdisc_run_begin(q)) {
2893 if (unlikely(contended)) {
2894 spin_unlock(&q->busylock);
2900 spin_unlock(root_lock);
2901 if (unlikely(contended))
2902 spin_unlock(&q->busylock);
2906 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2907 static void skb_update_prio(struct sk_buff *skb)
2909 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2911 if (!skb->priority && skb->sk && map) {
2912 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2914 if (prioidx < map->priomap_len)
2915 skb->priority = map->priomap[prioidx];
2919 #define skb_update_prio(skb)
2922 DEFINE_PER_CPU(int, xmit_recursion);
2923 EXPORT_SYMBOL(xmit_recursion);
2925 #define RECURSION_LIMIT 10
2928 * dev_loopback_xmit - loop back @skb
2929 * @skb: buffer to transmit
2931 int dev_loopback_xmit(struct sock *sk, struct sk_buff *skb)
2933 skb_reset_mac_header(skb);
2934 __skb_pull(skb, skb_network_offset(skb));
2935 skb->pkt_type = PACKET_LOOPBACK;
2936 skb->ip_summed = CHECKSUM_UNNECESSARY;
2937 WARN_ON(!skb_dst(skb));
2942 EXPORT_SYMBOL(dev_loopback_xmit);
2944 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2947 struct xps_dev_maps *dev_maps;
2948 struct xps_map *map;
2949 int queue_index = -1;
2952 dev_maps = rcu_dereference(dev->xps_maps);
2954 map = rcu_dereference(
2955 dev_maps->cpu_map[skb->sender_cpu - 1]);
2958 queue_index = map->queues[0];
2960 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
2962 if (unlikely(queue_index >= dev->real_num_tx_queues))
2974 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
2976 struct sock *sk = skb->sk;
2977 int queue_index = sk_tx_queue_get(sk);
2979 if (queue_index < 0 || skb->ooo_okay ||
2980 queue_index >= dev->real_num_tx_queues) {
2981 int new_index = get_xps_queue(dev, skb);
2983 new_index = skb_tx_hash(dev, skb);
2985 if (queue_index != new_index && sk &&
2986 rcu_access_pointer(sk->sk_dst_cache))
2987 sk_tx_queue_set(sk, new_index);
2989 queue_index = new_index;
2995 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
2996 struct sk_buff *skb,
2999 int queue_index = 0;
3002 if (skb->sender_cpu == 0)
3003 skb->sender_cpu = raw_smp_processor_id() + 1;
3006 if (dev->real_num_tx_queues != 1) {
3007 const struct net_device_ops *ops = dev->netdev_ops;
3008 if (ops->ndo_select_queue)
3009 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3012 queue_index = __netdev_pick_tx(dev, skb);
3015 queue_index = netdev_cap_txqueue(dev, queue_index);
3018 skb_set_queue_mapping(skb, queue_index);
3019 return netdev_get_tx_queue(dev, queue_index);
3023 * __dev_queue_xmit - transmit a buffer
3024 * @skb: buffer to transmit
3025 * @accel_priv: private data used for L2 forwarding offload
3027 * Queue a buffer for transmission to a network device. The caller must
3028 * have set the device and priority and built the buffer before calling
3029 * this function. The function can be called from an interrupt.
3031 * A negative errno code is returned on a failure. A success does not
3032 * guarantee the frame will be transmitted as it may be dropped due
3033 * to congestion or traffic shaping.
3035 * -----------------------------------------------------------------------------------
3036 * I notice this method can also return errors from the queue disciplines,
3037 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3040 * Regardless of the return value, the skb is consumed, so it is currently
3041 * difficult to retry a send to this method. (You can bump the ref count
3042 * before sending to hold a reference for retry if you are careful.)
3044 * When calling this method, interrupts MUST be enabled. This is because
3045 * the BH enable code must have IRQs enabled so that it will not deadlock.
3048 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3050 struct net_device *dev = skb->dev;
3051 struct netdev_queue *txq;
3055 skb_reset_mac_header(skb);
3057 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3058 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3060 /* Disable soft irqs for various locks below. Also
3061 * stops preemption for RCU.
3065 skb_update_prio(skb);
3067 /* If device/qdisc don't need skb->dst, release it right now while
3068 * its hot in this cpu cache.
3070 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3075 txq = netdev_pick_tx(dev, skb, accel_priv);
3076 q = rcu_dereference_bh(txq->qdisc);
3078 #ifdef CONFIG_NET_CLS_ACT
3079 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3081 trace_net_dev_queue(skb);
3083 rc = __dev_xmit_skb(skb, q, dev, txq);
3087 /* The device has no queue. Common case for software devices:
3088 loopback, all the sorts of tunnels...
3090 Really, it is unlikely that netif_tx_lock protection is necessary
3091 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3093 However, it is possible, that they rely on protection
3096 Check this and shot the lock. It is not prone from deadlocks.
3097 Either shot noqueue qdisc, it is even simpler 8)
3099 if (dev->flags & IFF_UP) {
3100 int cpu = smp_processor_id(); /* ok because BHs are off */
3102 if (txq->xmit_lock_owner != cpu) {
3104 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
3105 goto recursion_alert;
3107 skb = validate_xmit_skb(skb, dev);
3111 HARD_TX_LOCK(dev, txq, cpu);
3113 if (!netif_xmit_stopped(txq)) {
3114 __this_cpu_inc(xmit_recursion);
3115 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3116 __this_cpu_dec(xmit_recursion);
3117 if (dev_xmit_complete(rc)) {
3118 HARD_TX_UNLOCK(dev, txq);
3122 HARD_TX_UNLOCK(dev, txq);
3123 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3126 /* Recursion is detected! It is possible,
3130 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3137 rcu_read_unlock_bh();
3139 atomic_long_inc(&dev->tx_dropped);
3140 kfree_skb_list(skb);
3143 rcu_read_unlock_bh();
3147 int dev_queue_xmit_sk(struct sock *sk, struct sk_buff *skb)
3149 return __dev_queue_xmit(skb, NULL);
3151 EXPORT_SYMBOL(dev_queue_xmit_sk);
3153 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3155 return __dev_queue_xmit(skb, accel_priv);
3157 EXPORT_SYMBOL(dev_queue_xmit_accel);
3160 /*=======================================================================
3162 =======================================================================*/
3164 int netdev_max_backlog __read_mostly = 1000;
3165 EXPORT_SYMBOL(netdev_max_backlog);
3167 int netdev_tstamp_prequeue __read_mostly = 1;
3168 int netdev_budget __read_mostly = 300;
3169 int weight_p __read_mostly = 64; /* old backlog weight */
3171 /* Called with irq disabled */
3172 static inline void ____napi_schedule(struct softnet_data *sd,
3173 struct napi_struct *napi)
3175 list_add_tail(&napi->poll_list, &sd->poll_list);
3176 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3181 /* One global table that all flow-based protocols share. */
3182 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3183 EXPORT_SYMBOL(rps_sock_flow_table);
3184 u32 rps_cpu_mask __read_mostly;
3185 EXPORT_SYMBOL(rps_cpu_mask);
3187 struct static_key rps_needed __read_mostly;
3189 static struct rps_dev_flow *
3190 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3191 struct rps_dev_flow *rflow, u16 next_cpu)
3193 if (next_cpu < nr_cpu_ids) {
3194 #ifdef CONFIG_RFS_ACCEL
3195 struct netdev_rx_queue *rxqueue;
3196 struct rps_dev_flow_table *flow_table;
3197 struct rps_dev_flow *old_rflow;
3202 /* Should we steer this flow to a different hardware queue? */
3203 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3204 !(dev->features & NETIF_F_NTUPLE))
3206 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3207 if (rxq_index == skb_get_rx_queue(skb))
3210 rxqueue = dev->_rx + rxq_index;
3211 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3214 flow_id = skb_get_hash(skb) & flow_table->mask;
3215 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3216 rxq_index, flow_id);
3220 rflow = &flow_table->flows[flow_id];
3222 if (old_rflow->filter == rflow->filter)
3223 old_rflow->filter = RPS_NO_FILTER;
3227 per_cpu(softnet_data, next_cpu).input_queue_head;
3230 rflow->cpu = next_cpu;
3235 * get_rps_cpu is called from netif_receive_skb and returns the target
3236 * CPU from the RPS map of the receiving queue for a given skb.
3237 * rcu_read_lock must be held on entry.
3239 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3240 struct rps_dev_flow **rflowp)
3242 const struct rps_sock_flow_table *sock_flow_table;
3243 struct netdev_rx_queue *rxqueue = dev->_rx;
3244 struct rps_dev_flow_table *flow_table;
3245 struct rps_map *map;
3250 if (skb_rx_queue_recorded(skb)) {
3251 u16 index = skb_get_rx_queue(skb);
3253 if (unlikely(index >= dev->real_num_rx_queues)) {
3254 WARN_ONCE(dev->real_num_rx_queues > 1,
3255 "%s received packet on queue %u, but number "
3256 "of RX queues is %u\n",
3257 dev->name, index, dev->real_num_rx_queues);
3263 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3265 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3266 map = rcu_dereference(rxqueue->rps_map);
3267 if (!flow_table && !map)
3270 skb_reset_network_header(skb);
3271 hash = skb_get_hash(skb);
3275 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3276 if (flow_table && sock_flow_table) {
3277 struct rps_dev_flow *rflow;
3281 /* First check into global flow table if there is a match */
3282 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3283 if ((ident ^ hash) & ~rps_cpu_mask)
3286 next_cpu = ident & rps_cpu_mask;
3288 /* OK, now we know there is a match,
3289 * we can look at the local (per receive queue) flow table
3291 rflow = &flow_table->flows[hash & flow_table->mask];
3295 * If the desired CPU (where last recvmsg was done) is
3296 * different from current CPU (one in the rx-queue flow
3297 * table entry), switch if one of the following holds:
3298 * - Current CPU is unset (>= nr_cpu_ids).
3299 * - Current CPU is offline.
3300 * - The current CPU's queue tail has advanced beyond the
3301 * last packet that was enqueued using this table entry.
3302 * This guarantees that all previous packets for the flow
3303 * have been dequeued, thus preserving in order delivery.
3305 if (unlikely(tcpu != next_cpu) &&
3306 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3307 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3308 rflow->last_qtail)) >= 0)) {
3310 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3313 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3323 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3324 if (cpu_online(tcpu)) {
3334 #ifdef CONFIG_RFS_ACCEL
3337 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3338 * @dev: Device on which the filter was set
3339 * @rxq_index: RX queue index
3340 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3341 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3343 * Drivers that implement ndo_rx_flow_steer() should periodically call
3344 * this function for each installed filter and remove the filters for
3345 * which it returns %true.
3347 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3348 u32 flow_id, u16 filter_id)
3350 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3351 struct rps_dev_flow_table *flow_table;
3352 struct rps_dev_flow *rflow;
3357 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3358 if (flow_table && flow_id <= flow_table->mask) {
3359 rflow = &flow_table->flows[flow_id];
3360 cpu = ACCESS_ONCE(rflow->cpu);
3361 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3362 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3363 rflow->last_qtail) <
3364 (int)(10 * flow_table->mask)))
3370 EXPORT_SYMBOL(rps_may_expire_flow);
3372 #endif /* CONFIG_RFS_ACCEL */
3374 /* Called from hardirq (IPI) context */
3375 static void rps_trigger_softirq(void *data)
3377 struct softnet_data *sd = data;
3379 ____napi_schedule(sd, &sd->backlog);
3383 #endif /* CONFIG_RPS */
3386 * Check if this softnet_data structure is another cpu one
3387 * If yes, queue it to our IPI list and return 1
3390 static int rps_ipi_queued(struct softnet_data *sd)
3393 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3396 sd->rps_ipi_next = mysd->rps_ipi_list;
3397 mysd->rps_ipi_list = sd;
3399 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3402 #endif /* CONFIG_RPS */
3406 #ifdef CONFIG_NET_FLOW_LIMIT
3407 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3410 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3412 #ifdef CONFIG_NET_FLOW_LIMIT
3413 struct sd_flow_limit *fl;
3414 struct softnet_data *sd;
3415 unsigned int old_flow, new_flow;
3417 if (qlen < (netdev_max_backlog >> 1))
3420 sd = this_cpu_ptr(&softnet_data);
3423 fl = rcu_dereference(sd->flow_limit);
3425 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3426 old_flow = fl->history[fl->history_head];
3427 fl->history[fl->history_head] = new_flow;
3430 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3432 if (likely(fl->buckets[old_flow]))
3433 fl->buckets[old_flow]--;
3435 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3447 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3448 * queue (may be a remote CPU queue).
3450 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3451 unsigned int *qtail)
3453 struct softnet_data *sd;
3454 unsigned long flags;
3457 sd = &per_cpu(softnet_data, cpu);
3459 local_irq_save(flags);
3462 qlen = skb_queue_len(&sd->input_pkt_queue);
3463 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3466 __skb_queue_tail(&sd->input_pkt_queue, skb);
3467 input_queue_tail_incr_save(sd, qtail);
3469 local_irq_restore(flags);
3470 return NET_RX_SUCCESS;
3473 /* Schedule NAPI for backlog device
3474 * We can use non atomic operation since we own the queue lock
3476 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3477 if (!rps_ipi_queued(sd))
3478 ____napi_schedule(sd, &sd->backlog);
3486 local_irq_restore(flags);
3488 atomic_long_inc(&skb->dev->rx_dropped);
3493 static int netif_rx_internal(struct sk_buff *skb)
3497 net_timestamp_check(netdev_tstamp_prequeue, skb);
3499 trace_netif_rx(skb);
3501 if (static_key_false(&rps_needed)) {
3502 struct rps_dev_flow voidflow, *rflow = &voidflow;
3508 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3510 cpu = smp_processor_id();
3512 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3520 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3527 * netif_rx - post buffer to the network code
3528 * @skb: buffer to post
3530 * This function receives a packet from a device driver and queues it for
3531 * the upper (protocol) levels to process. It always succeeds. The buffer
3532 * may be dropped during processing for congestion control or by the
3536 * NET_RX_SUCCESS (no congestion)
3537 * NET_RX_DROP (packet was dropped)
3541 int netif_rx(struct sk_buff *skb)
3543 trace_netif_rx_entry(skb);
3545 return netif_rx_internal(skb);
3547 EXPORT_SYMBOL(netif_rx);
3549 int netif_rx_ni(struct sk_buff *skb)
3553 trace_netif_rx_ni_entry(skb);
3556 err = netif_rx_internal(skb);
3557 if (local_softirq_pending())
3563 EXPORT_SYMBOL(netif_rx_ni);
3565 static void net_tx_action(struct softirq_action *h)
3567 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3569 if (sd->completion_queue) {
3570 struct sk_buff *clist;
3572 local_irq_disable();
3573 clist = sd->completion_queue;
3574 sd->completion_queue = NULL;
3578 struct sk_buff *skb = clist;
3579 clist = clist->next;
3581 WARN_ON(atomic_read(&skb->users));
3582 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3583 trace_consume_skb(skb);
3585 trace_kfree_skb(skb, net_tx_action);
3590 if (sd->output_queue) {
3593 local_irq_disable();
3594 head = sd->output_queue;
3595 sd->output_queue = NULL;
3596 sd->output_queue_tailp = &sd->output_queue;
3600 struct Qdisc *q = head;
3601 spinlock_t *root_lock;
3603 head = head->next_sched;
3605 root_lock = qdisc_lock(q);
3606 if (spin_trylock(root_lock)) {
3607 smp_mb__before_atomic();
3608 clear_bit(__QDISC_STATE_SCHED,
3611 spin_unlock(root_lock);
3613 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3615 __netif_reschedule(q);
3617 smp_mb__before_atomic();
3618 clear_bit(__QDISC_STATE_SCHED,
3626 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3627 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3628 /* This hook is defined here for ATM LANE */
3629 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3630 unsigned char *addr) __read_mostly;
3631 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3634 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3635 struct packet_type **pt_prev,
3636 int *ret, struct net_device *orig_dev)
3638 #ifdef CONFIG_NET_CLS_ACT
3639 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3640 struct tcf_result cl_res;
3642 /* If there's at least one ingress present somewhere (so
3643 * we get here via enabled static key), remaining devices
3644 * that are not configured with an ingress qdisc will bail
3650 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3654 qdisc_skb_cb(skb)->pkt_len = skb->len;
3655 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3656 qdisc_bstats_update_cpu(cl->q, skb);
3658 switch (tc_classify(skb, cl, &cl_res)) {
3660 case TC_ACT_RECLASSIFY:
3661 skb->tc_index = TC_H_MIN(cl_res.classid);
3664 qdisc_qstats_drop_cpu(cl->q);
3672 #endif /* CONFIG_NET_CLS_ACT */
3677 * netdev_rx_handler_register - register receive handler
3678 * @dev: device to register a handler for
3679 * @rx_handler: receive handler to register
3680 * @rx_handler_data: data pointer that is used by rx handler
3682 * Register a receive handler for a device. This handler will then be
3683 * called from __netif_receive_skb. A negative errno code is returned
3686 * The caller must hold the rtnl_mutex.
3688 * For a general description of rx_handler, see enum rx_handler_result.
3690 int netdev_rx_handler_register(struct net_device *dev,
3691 rx_handler_func_t *rx_handler,
3692 void *rx_handler_data)
3696 if (dev->rx_handler)
3699 /* Note: rx_handler_data must be set before rx_handler */
3700 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3701 rcu_assign_pointer(dev->rx_handler, rx_handler);
3705 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3708 * netdev_rx_handler_unregister - unregister receive handler
3709 * @dev: device to unregister a handler from
3711 * Unregister a receive handler from a device.
3713 * The caller must hold the rtnl_mutex.
3715 void netdev_rx_handler_unregister(struct net_device *dev)
3719 RCU_INIT_POINTER(dev->rx_handler, NULL);
3720 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3721 * section has a guarantee to see a non NULL rx_handler_data
3725 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3727 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3730 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3731 * the special handling of PFMEMALLOC skbs.
3733 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3735 switch (skb->protocol) {
3736 case htons(ETH_P_ARP):
3737 case htons(ETH_P_IP):
3738 case htons(ETH_P_IPV6):
3739 case htons(ETH_P_8021Q):
3740 case htons(ETH_P_8021AD):
3747 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
3748 int *ret, struct net_device *orig_dev)
3750 #ifdef CONFIG_NETFILTER_INGRESS
3751 if (nf_hook_ingress_active(skb)) {
3753 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3757 return nf_hook_ingress(skb);
3759 #endif /* CONFIG_NETFILTER_INGRESS */
3763 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3765 struct packet_type *ptype, *pt_prev;
3766 rx_handler_func_t *rx_handler;
3767 struct net_device *orig_dev;
3768 bool deliver_exact = false;
3769 int ret = NET_RX_DROP;
3772 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3774 trace_netif_receive_skb(skb);
3776 orig_dev = skb->dev;
3778 skb_reset_network_header(skb);
3779 if (!skb_transport_header_was_set(skb))
3780 skb_reset_transport_header(skb);
3781 skb_reset_mac_len(skb);
3788 skb->skb_iif = skb->dev->ifindex;
3790 __this_cpu_inc(softnet_data.processed);
3792 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3793 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3794 skb = skb_vlan_untag(skb);
3799 #ifdef CONFIG_NET_CLS_ACT
3800 if (skb->tc_verd & TC_NCLS) {
3801 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3809 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3811 ret = deliver_skb(skb, pt_prev, orig_dev);
3815 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3817 ret = deliver_skb(skb, pt_prev, orig_dev);
3822 #ifdef CONFIG_NET_INGRESS
3823 if (static_key_false(&ingress_needed)) {
3824 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3828 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
3832 #ifdef CONFIG_NET_CLS_ACT
3836 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3839 if (skb_vlan_tag_present(skb)) {
3841 ret = deliver_skb(skb, pt_prev, orig_dev);
3844 if (vlan_do_receive(&skb))
3846 else if (unlikely(!skb))
3850 rx_handler = rcu_dereference(skb->dev->rx_handler);
3853 ret = deliver_skb(skb, pt_prev, orig_dev);
3856 switch (rx_handler(&skb)) {
3857 case RX_HANDLER_CONSUMED:
3858 ret = NET_RX_SUCCESS;
3860 case RX_HANDLER_ANOTHER:
3862 case RX_HANDLER_EXACT:
3863 deliver_exact = true;
3864 case RX_HANDLER_PASS:
3871 if (unlikely(skb_vlan_tag_present(skb))) {
3872 if (skb_vlan_tag_get_id(skb))
3873 skb->pkt_type = PACKET_OTHERHOST;
3874 /* Note: we might in the future use prio bits
3875 * and set skb->priority like in vlan_do_receive()
3876 * For the time being, just ignore Priority Code Point
3881 type = skb->protocol;
3883 /* deliver only exact match when indicated */
3884 if (likely(!deliver_exact)) {
3885 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3886 &ptype_base[ntohs(type) &
3890 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3891 &orig_dev->ptype_specific);
3893 if (unlikely(skb->dev != orig_dev)) {
3894 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3895 &skb->dev->ptype_specific);
3899 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3902 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3905 atomic_long_inc(&skb->dev->rx_dropped);
3907 /* Jamal, now you will not able to escape explaining
3908 * me how you were going to use this. :-)
3918 static int __netif_receive_skb(struct sk_buff *skb)
3922 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3923 unsigned long pflags = current->flags;
3926 * PFMEMALLOC skbs are special, they should
3927 * - be delivered to SOCK_MEMALLOC sockets only
3928 * - stay away from userspace
3929 * - have bounded memory usage
3931 * Use PF_MEMALLOC as this saves us from propagating the allocation
3932 * context down to all allocation sites.
3934 current->flags |= PF_MEMALLOC;
3935 ret = __netif_receive_skb_core(skb, true);
3936 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3938 ret = __netif_receive_skb_core(skb, false);
3943 static int netif_receive_skb_internal(struct sk_buff *skb)
3945 net_timestamp_check(netdev_tstamp_prequeue, skb);
3947 if (skb_defer_rx_timestamp(skb))
3948 return NET_RX_SUCCESS;
3951 if (static_key_false(&rps_needed)) {
3952 struct rps_dev_flow voidflow, *rflow = &voidflow;
3957 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3960 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3967 return __netif_receive_skb(skb);
3971 * netif_receive_skb - process receive buffer from network
3972 * @skb: buffer to process
3974 * netif_receive_skb() is the main receive data processing function.
3975 * It always succeeds. The buffer may be dropped during processing
3976 * for congestion control or by the protocol layers.
3978 * This function may only be called from softirq context and interrupts
3979 * should be enabled.
3981 * Return values (usually ignored):
3982 * NET_RX_SUCCESS: no congestion
3983 * NET_RX_DROP: packet was dropped
3985 int netif_receive_skb_sk(struct sock *sk, struct sk_buff *skb)
3987 trace_netif_receive_skb_entry(skb);
3989 return netif_receive_skb_internal(skb);
3991 EXPORT_SYMBOL(netif_receive_skb_sk);
3993 /* Network device is going away, flush any packets still pending
3994 * Called with irqs disabled.
3996 static void flush_backlog(void *arg)
3998 struct net_device *dev = arg;
3999 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4000 struct sk_buff *skb, *tmp;
4003 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4004 if (skb->dev == dev) {
4005 __skb_unlink(skb, &sd->input_pkt_queue);
4007 input_queue_head_incr(sd);
4012 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4013 if (skb->dev == dev) {
4014 __skb_unlink(skb, &sd->process_queue);
4016 input_queue_head_incr(sd);
4021 static int napi_gro_complete(struct sk_buff *skb)
4023 struct packet_offload *ptype;
4024 __be16 type = skb->protocol;
4025 struct list_head *head = &offload_base;
4028 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4030 if (NAPI_GRO_CB(skb)->count == 1) {
4031 skb_shinfo(skb)->gso_size = 0;
4036 list_for_each_entry_rcu(ptype, head, list) {
4037 if (ptype->type != type || !ptype->callbacks.gro_complete)
4040 err = ptype->callbacks.gro_complete(skb, 0);
4046 WARN_ON(&ptype->list == head);
4048 return NET_RX_SUCCESS;
4052 return netif_receive_skb_internal(skb);
4055 /* napi->gro_list contains packets ordered by age.
4056 * youngest packets at the head of it.
4057 * Complete skbs in reverse order to reduce latencies.
4059 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4061 struct sk_buff *skb, *prev = NULL;
4063 /* scan list and build reverse chain */
4064 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4069 for (skb = prev; skb; skb = prev) {
4072 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4076 napi_gro_complete(skb);
4080 napi->gro_list = NULL;
4082 EXPORT_SYMBOL(napi_gro_flush);
4084 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4087 unsigned int maclen = skb->dev->hard_header_len;
4088 u32 hash = skb_get_hash_raw(skb);
4090 for (p = napi->gro_list; p; p = p->next) {
4091 unsigned long diffs;
4093 NAPI_GRO_CB(p)->flush = 0;
4095 if (hash != skb_get_hash_raw(p)) {
4096 NAPI_GRO_CB(p)->same_flow = 0;
4100 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4101 diffs |= p->vlan_tci ^ skb->vlan_tci;
4102 if (maclen == ETH_HLEN)
4103 diffs |= compare_ether_header(skb_mac_header(p),
4104 skb_mac_header(skb));
4106 diffs = memcmp(skb_mac_header(p),
4107 skb_mac_header(skb),
4109 NAPI_GRO_CB(p)->same_flow = !diffs;
4113 static void skb_gro_reset_offset(struct sk_buff *skb)
4115 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4116 const skb_frag_t *frag0 = &pinfo->frags[0];
4118 NAPI_GRO_CB(skb)->data_offset = 0;
4119 NAPI_GRO_CB(skb)->frag0 = NULL;
4120 NAPI_GRO_CB(skb)->frag0_len = 0;
4122 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4124 !PageHighMem(skb_frag_page(frag0))) {
4125 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4126 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4130 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4132 struct skb_shared_info *pinfo = skb_shinfo(skb);
4134 BUG_ON(skb->end - skb->tail < grow);
4136 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4138 skb->data_len -= grow;
4141 pinfo->frags[0].page_offset += grow;
4142 skb_frag_size_sub(&pinfo->frags[0], grow);
4144 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4145 skb_frag_unref(skb, 0);
4146 memmove(pinfo->frags, pinfo->frags + 1,
4147 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4151 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4153 struct sk_buff **pp = NULL;
4154 struct packet_offload *ptype;
4155 __be16 type = skb->protocol;
4156 struct list_head *head = &offload_base;
4158 enum gro_result ret;
4161 if (!(skb->dev->features & NETIF_F_GRO))
4164 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4167 gro_list_prepare(napi, skb);
4170 list_for_each_entry_rcu(ptype, head, list) {
4171 if (ptype->type != type || !ptype->callbacks.gro_receive)
4174 skb_set_network_header(skb, skb_gro_offset(skb));
4175 skb_reset_mac_len(skb);
4176 NAPI_GRO_CB(skb)->same_flow = 0;
4177 NAPI_GRO_CB(skb)->flush = 0;
4178 NAPI_GRO_CB(skb)->free = 0;
4179 NAPI_GRO_CB(skb)->udp_mark = 0;
4180 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4182 /* Setup for GRO checksum validation */
4183 switch (skb->ip_summed) {
4184 case CHECKSUM_COMPLETE:
4185 NAPI_GRO_CB(skb)->csum = skb->csum;
4186 NAPI_GRO_CB(skb)->csum_valid = 1;
4187 NAPI_GRO_CB(skb)->csum_cnt = 0;
4189 case CHECKSUM_UNNECESSARY:
4190 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4191 NAPI_GRO_CB(skb)->csum_valid = 0;
4194 NAPI_GRO_CB(skb)->csum_cnt = 0;
4195 NAPI_GRO_CB(skb)->csum_valid = 0;
4198 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4203 if (&ptype->list == head)
4206 same_flow = NAPI_GRO_CB(skb)->same_flow;
4207 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4210 struct sk_buff *nskb = *pp;
4214 napi_gro_complete(nskb);
4221 if (NAPI_GRO_CB(skb)->flush)
4224 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4225 struct sk_buff *nskb = napi->gro_list;
4227 /* locate the end of the list to select the 'oldest' flow */
4228 while (nskb->next) {
4234 napi_gro_complete(nskb);
4238 NAPI_GRO_CB(skb)->count = 1;
4239 NAPI_GRO_CB(skb)->age = jiffies;
4240 NAPI_GRO_CB(skb)->last = skb;
4241 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4242 skb->next = napi->gro_list;
4243 napi->gro_list = skb;
4247 grow = skb_gro_offset(skb) - skb_headlen(skb);
4249 gro_pull_from_frag0(skb, grow);
4258 struct packet_offload *gro_find_receive_by_type(__be16 type)
4260 struct list_head *offload_head = &offload_base;
4261 struct packet_offload *ptype;
4263 list_for_each_entry_rcu(ptype, offload_head, list) {
4264 if (ptype->type != type || !ptype->callbacks.gro_receive)
4270 EXPORT_SYMBOL(gro_find_receive_by_type);
4272 struct packet_offload *gro_find_complete_by_type(__be16 type)
4274 struct list_head *offload_head = &offload_base;
4275 struct packet_offload *ptype;
4277 list_for_each_entry_rcu(ptype, offload_head, list) {
4278 if (ptype->type != type || !ptype->callbacks.gro_complete)
4284 EXPORT_SYMBOL(gro_find_complete_by_type);
4286 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4290 if (netif_receive_skb_internal(skb))
4298 case GRO_MERGED_FREE:
4299 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4300 kmem_cache_free(skbuff_head_cache, skb);
4313 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4315 trace_napi_gro_receive_entry(skb);
4317 skb_gro_reset_offset(skb);
4319 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4321 EXPORT_SYMBOL(napi_gro_receive);
4323 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4325 if (unlikely(skb->pfmemalloc)) {
4329 __skb_pull(skb, skb_headlen(skb));
4330 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4331 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4333 skb->dev = napi->dev;
4335 skb->encapsulation = 0;
4336 skb_shinfo(skb)->gso_type = 0;
4337 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4342 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4344 struct sk_buff *skb = napi->skb;
4347 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4352 EXPORT_SYMBOL(napi_get_frags);
4354 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4355 struct sk_buff *skb,
4361 __skb_push(skb, ETH_HLEN);
4362 skb->protocol = eth_type_trans(skb, skb->dev);
4363 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4368 case GRO_MERGED_FREE:
4369 napi_reuse_skb(napi, skb);
4379 /* Upper GRO stack assumes network header starts at gro_offset=0
4380 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4381 * We copy ethernet header into skb->data to have a common layout.
4383 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4385 struct sk_buff *skb = napi->skb;
4386 const struct ethhdr *eth;
4387 unsigned int hlen = sizeof(*eth);
4391 skb_reset_mac_header(skb);
4392 skb_gro_reset_offset(skb);
4394 eth = skb_gro_header_fast(skb, 0);
4395 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4396 eth = skb_gro_header_slow(skb, hlen, 0);
4397 if (unlikely(!eth)) {
4398 napi_reuse_skb(napi, skb);
4402 gro_pull_from_frag0(skb, hlen);
4403 NAPI_GRO_CB(skb)->frag0 += hlen;
4404 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4406 __skb_pull(skb, hlen);
4409 * This works because the only protocols we care about don't require
4411 * We'll fix it up properly in napi_frags_finish()
4413 skb->protocol = eth->h_proto;
4418 gro_result_t napi_gro_frags(struct napi_struct *napi)
4420 struct sk_buff *skb = napi_frags_skb(napi);
4425 trace_napi_gro_frags_entry(skb);
4427 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4429 EXPORT_SYMBOL(napi_gro_frags);
4431 /* Compute the checksum from gro_offset and return the folded value
4432 * after adding in any pseudo checksum.
4434 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4439 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4441 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4442 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4444 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4445 !skb->csum_complete_sw)
4446 netdev_rx_csum_fault(skb->dev);
4449 NAPI_GRO_CB(skb)->csum = wsum;
4450 NAPI_GRO_CB(skb)->csum_valid = 1;
4454 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4457 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4458 * Note: called with local irq disabled, but exits with local irq enabled.
4460 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4463 struct softnet_data *remsd = sd->rps_ipi_list;
4466 sd->rps_ipi_list = NULL;
4470 /* Send pending IPI's to kick RPS processing on remote cpus. */
4472 struct softnet_data *next = remsd->rps_ipi_next;
4474 if (cpu_online(remsd->cpu))
4475 smp_call_function_single_async(remsd->cpu,
4484 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4487 return sd->rps_ipi_list != NULL;
4493 static int process_backlog(struct napi_struct *napi, int quota)
4496 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4498 /* Check if we have pending ipi, its better to send them now,
4499 * not waiting net_rx_action() end.
4501 if (sd_has_rps_ipi_waiting(sd)) {
4502 local_irq_disable();
4503 net_rps_action_and_irq_enable(sd);
4506 napi->weight = weight_p;
4507 local_irq_disable();
4509 struct sk_buff *skb;
4511 while ((skb = __skb_dequeue(&sd->process_queue))) {
4513 __netif_receive_skb(skb);
4514 local_irq_disable();
4515 input_queue_head_incr(sd);
4516 if (++work >= quota) {
4523 if (skb_queue_empty(&sd->input_pkt_queue)) {
4525 * Inline a custom version of __napi_complete().
4526 * only current cpu owns and manipulates this napi,
4527 * and NAPI_STATE_SCHED is the only possible flag set
4529 * We can use a plain write instead of clear_bit(),
4530 * and we dont need an smp_mb() memory barrier.
4538 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4539 &sd->process_queue);
4548 * __napi_schedule - schedule for receive
4549 * @n: entry to schedule
4551 * The entry's receive function will be scheduled to run.
4552 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4554 void __napi_schedule(struct napi_struct *n)
4556 unsigned long flags;
4558 local_irq_save(flags);
4559 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4560 local_irq_restore(flags);
4562 EXPORT_SYMBOL(__napi_schedule);
4565 * __napi_schedule_irqoff - schedule for receive
4566 * @n: entry to schedule
4568 * Variant of __napi_schedule() assuming hard irqs are masked
4570 void __napi_schedule_irqoff(struct napi_struct *n)
4572 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4574 EXPORT_SYMBOL(__napi_schedule_irqoff);
4576 void __napi_complete(struct napi_struct *n)
4578 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4580 list_del_init(&n->poll_list);
4581 smp_mb__before_atomic();
4582 clear_bit(NAPI_STATE_SCHED, &n->state);
4584 EXPORT_SYMBOL(__napi_complete);
4586 void napi_complete_done(struct napi_struct *n, int work_done)
4588 unsigned long flags;
4591 * don't let napi dequeue from the cpu poll list
4592 * just in case its running on a different cpu
4594 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4598 unsigned long timeout = 0;
4601 timeout = n->dev->gro_flush_timeout;
4604 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4605 HRTIMER_MODE_REL_PINNED);
4607 napi_gro_flush(n, false);
4609 if (likely(list_empty(&n->poll_list))) {
4610 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4612 /* If n->poll_list is not empty, we need to mask irqs */
4613 local_irq_save(flags);
4615 local_irq_restore(flags);
4618 EXPORT_SYMBOL(napi_complete_done);
4620 /* must be called under rcu_read_lock(), as we dont take a reference */
4621 struct napi_struct *napi_by_id(unsigned int napi_id)
4623 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4624 struct napi_struct *napi;
4626 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4627 if (napi->napi_id == napi_id)
4632 EXPORT_SYMBOL_GPL(napi_by_id);
4634 void napi_hash_add(struct napi_struct *napi)
4636 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4638 spin_lock(&napi_hash_lock);
4640 /* 0 is not a valid id, we also skip an id that is taken
4641 * we expect both events to be extremely rare
4644 while (!napi->napi_id) {
4645 napi->napi_id = ++napi_gen_id;
4646 if (napi_by_id(napi->napi_id))
4650 hlist_add_head_rcu(&napi->napi_hash_node,
4651 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4653 spin_unlock(&napi_hash_lock);
4656 EXPORT_SYMBOL_GPL(napi_hash_add);
4658 /* Warning : caller is responsible to make sure rcu grace period
4659 * is respected before freeing memory containing @napi
4661 void napi_hash_del(struct napi_struct *napi)
4663 spin_lock(&napi_hash_lock);
4665 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4666 hlist_del_rcu(&napi->napi_hash_node);
4668 spin_unlock(&napi_hash_lock);
4670 EXPORT_SYMBOL_GPL(napi_hash_del);
4672 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4674 struct napi_struct *napi;
4676 napi = container_of(timer, struct napi_struct, timer);
4678 napi_schedule(napi);
4680 return HRTIMER_NORESTART;
4683 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4684 int (*poll)(struct napi_struct *, int), int weight)
4686 INIT_LIST_HEAD(&napi->poll_list);
4687 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4688 napi->timer.function = napi_watchdog;
4689 napi->gro_count = 0;
4690 napi->gro_list = NULL;
4693 if (weight > NAPI_POLL_WEIGHT)
4694 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4696 napi->weight = weight;
4697 list_add(&napi->dev_list, &dev->napi_list);
4699 #ifdef CONFIG_NETPOLL
4700 spin_lock_init(&napi->poll_lock);
4701 napi->poll_owner = -1;
4703 set_bit(NAPI_STATE_SCHED, &napi->state);
4705 EXPORT_SYMBOL(netif_napi_add);
4707 void napi_disable(struct napi_struct *n)
4710 set_bit(NAPI_STATE_DISABLE, &n->state);
4712 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4715 hrtimer_cancel(&n->timer);
4717 clear_bit(NAPI_STATE_DISABLE, &n->state);
4719 EXPORT_SYMBOL(napi_disable);
4721 void netif_napi_del(struct napi_struct *napi)
4723 list_del_init(&napi->dev_list);
4724 napi_free_frags(napi);
4726 kfree_skb_list(napi->gro_list);
4727 napi->gro_list = NULL;
4728 napi->gro_count = 0;
4730 EXPORT_SYMBOL(netif_napi_del);
4732 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4737 list_del_init(&n->poll_list);
4739 have = netpoll_poll_lock(n);
4743 /* This NAPI_STATE_SCHED test is for avoiding a race
4744 * with netpoll's poll_napi(). Only the entity which
4745 * obtains the lock and sees NAPI_STATE_SCHED set will
4746 * actually make the ->poll() call. Therefore we avoid
4747 * accidentally calling ->poll() when NAPI is not scheduled.
4750 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4751 work = n->poll(n, weight);
4755 WARN_ON_ONCE(work > weight);
4757 if (likely(work < weight))
4760 /* Drivers must not modify the NAPI state if they
4761 * consume the entire weight. In such cases this code
4762 * still "owns" the NAPI instance and therefore can
4763 * move the instance around on the list at-will.
4765 if (unlikely(napi_disable_pending(n))) {
4771 /* flush too old packets
4772 * If HZ < 1000, flush all packets.
4774 napi_gro_flush(n, HZ >= 1000);
4777 /* Some drivers may have called napi_schedule
4778 * prior to exhausting their budget.
4780 if (unlikely(!list_empty(&n->poll_list))) {
4781 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4782 n->dev ? n->dev->name : "backlog");
4786 list_add_tail(&n->poll_list, repoll);
4789 netpoll_poll_unlock(have);
4794 static void net_rx_action(struct softirq_action *h)
4796 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4797 unsigned long time_limit = jiffies + 2;
4798 int budget = netdev_budget;
4802 local_irq_disable();
4803 list_splice_init(&sd->poll_list, &list);
4807 struct napi_struct *n;
4809 if (list_empty(&list)) {
4810 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4815 n = list_first_entry(&list, struct napi_struct, poll_list);
4816 budget -= napi_poll(n, &repoll);
4818 /* If softirq window is exhausted then punt.
4819 * Allow this to run for 2 jiffies since which will allow
4820 * an average latency of 1.5/HZ.
4822 if (unlikely(budget <= 0 ||
4823 time_after_eq(jiffies, time_limit))) {
4829 local_irq_disable();
4831 list_splice_tail_init(&sd->poll_list, &list);
4832 list_splice_tail(&repoll, &list);
4833 list_splice(&list, &sd->poll_list);
4834 if (!list_empty(&sd->poll_list))
4835 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4837 net_rps_action_and_irq_enable(sd);
4840 struct netdev_adjacent {
4841 struct net_device *dev;
4843 /* upper master flag, there can only be one master device per list */
4846 /* counter for the number of times this device was added to us */
4849 /* private field for the users */
4852 struct list_head list;
4853 struct rcu_head rcu;
4856 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4857 struct net_device *adj_dev,
4858 struct list_head *adj_list)
4860 struct netdev_adjacent *adj;
4862 list_for_each_entry(adj, adj_list, list) {
4863 if (adj->dev == adj_dev)
4870 * netdev_has_upper_dev - Check if device is linked to an upper device
4872 * @upper_dev: upper device to check
4874 * Find out if a device is linked to specified upper device and return true
4875 * in case it is. Note that this checks only immediate upper device,
4876 * not through a complete stack of devices. The caller must hold the RTNL lock.
4878 bool netdev_has_upper_dev(struct net_device *dev,
4879 struct net_device *upper_dev)
4883 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4885 EXPORT_SYMBOL(netdev_has_upper_dev);
4888 * netdev_has_any_upper_dev - Check if device is linked to some device
4891 * Find out if a device is linked to an upper device and return true in case
4892 * it is. The caller must hold the RTNL lock.
4894 static bool netdev_has_any_upper_dev(struct net_device *dev)
4898 return !list_empty(&dev->all_adj_list.upper);
4902 * netdev_master_upper_dev_get - Get master upper device
4905 * Find a master upper device and return pointer to it or NULL in case
4906 * it's not there. The caller must hold the RTNL lock.
4908 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4910 struct netdev_adjacent *upper;
4914 if (list_empty(&dev->adj_list.upper))
4917 upper = list_first_entry(&dev->adj_list.upper,
4918 struct netdev_adjacent, list);
4919 if (likely(upper->master))
4923 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4925 void *netdev_adjacent_get_private(struct list_head *adj_list)
4927 struct netdev_adjacent *adj;
4929 adj = list_entry(adj_list, struct netdev_adjacent, list);
4931 return adj->private;
4933 EXPORT_SYMBOL(netdev_adjacent_get_private);
4936 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4938 * @iter: list_head ** of the current position
4940 * Gets the next device from the dev's upper list, starting from iter
4941 * position. The caller must hold RCU read lock.
4943 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4944 struct list_head **iter)
4946 struct netdev_adjacent *upper;
4948 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4950 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4952 if (&upper->list == &dev->adj_list.upper)
4955 *iter = &upper->list;
4959 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4962 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4964 * @iter: list_head ** of the current position
4966 * Gets the next device from the dev's upper list, starting from iter
4967 * position. The caller must hold RCU read lock.
4969 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4970 struct list_head **iter)
4972 struct netdev_adjacent *upper;
4974 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4976 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4978 if (&upper->list == &dev->all_adj_list.upper)
4981 *iter = &upper->list;
4985 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4988 * netdev_lower_get_next_private - Get the next ->private from the
4989 * lower neighbour list
4991 * @iter: list_head ** of the current position
4993 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4994 * list, starting from iter position. The caller must hold either hold the
4995 * RTNL lock or its own locking that guarantees that the neighbour lower
4996 * list will remain unchainged.
4998 void *netdev_lower_get_next_private(struct net_device *dev,
4999 struct list_head **iter)
5001 struct netdev_adjacent *lower;
5003 lower = list_entry(*iter, struct netdev_adjacent, list);
5005 if (&lower->list == &dev->adj_list.lower)
5008 *iter = lower->list.next;
5010 return lower->private;
5012 EXPORT_SYMBOL(netdev_lower_get_next_private);
5015 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5016 * lower neighbour list, RCU
5019 * @iter: list_head ** of the current position
5021 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5022 * list, starting from iter position. The caller must hold RCU read lock.
5024 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5025 struct list_head **iter)
5027 struct netdev_adjacent *lower;
5029 WARN_ON_ONCE(!rcu_read_lock_held());
5031 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5033 if (&lower->list == &dev->adj_list.lower)
5036 *iter = &lower->list;
5038 return lower->private;
5040 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5043 * netdev_lower_get_next - Get the next device from the lower neighbour
5046 * @iter: list_head ** of the current position
5048 * Gets the next netdev_adjacent from the dev's lower neighbour
5049 * list, starting from iter position. The caller must hold RTNL lock or
5050 * its own locking that guarantees that the neighbour lower
5051 * list will remain unchainged.
5053 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5055 struct netdev_adjacent *lower;
5057 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5059 if (&lower->list == &dev->adj_list.lower)
5062 *iter = &lower->list;
5066 EXPORT_SYMBOL(netdev_lower_get_next);
5069 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5070 * lower neighbour list, RCU
5074 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5075 * list. The caller must hold RCU read lock.
5077 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5079 struct netdev_adjacent *lower;
5081 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5082 struct netdev_adjacent, list);
5084 return lower->private;
5087 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5090 * netdev_master_upper_dev_get_rcu - Get master upper device
5093 * Find a master upper device and return pointer to it or NULL in case
5094 * it's not there. The caller must hold the RCU read lock.
5096 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5098 struct netdev_adjacent *upper;
5100 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5101 struct netdev_adjacent, list);
5102 if (upper && likely(upper->master))
5106 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5108 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5109 struct net_device *adj_dev,
5110 struct list_head *dev_list)
5112 char linkname[IFNAMSIZ+7];
5113 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5114 "upper_%s" : "lower_%s", adj_dev->name);
5115 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5118 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5120 struct list_head *dev_list)
5122 char linkname[IFNAMSIZ+7];
5123 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5124 "upper_%s" : "lower_%s", name);
5125 sysfs_remove_link(&(dev->dev.kobj), linkname);
5128 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5129 struct net_device *adj_dev,
5130 struct list_head *dev_list)
5132 return (dev_list == &dev->adj_list.upper ||
5133 dev_list == &dev->adj_list.lower) &&
5134 net_eq(dev_net(dev), dev_net(adj_dev));
5137 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5138 struct net_device *adj_dev,
5139 struct list_head *dev_list,
5140 void *private, bool master)
5142 struct netdev_adjacent *adj;
5145 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5152 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5157 adj->master = master;
5159 adj->private = private;
5162 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5163 adj_dev->name, dev->name, adj_dev->name);
5165 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5166 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5171 /* Ensure that master link is always the first item in list. */
5173 ret = sysfs_create_link(&(dev->dev.kobj),
5174 &(adj_dev->dev.kobj), "master");
5176 goto remove_symlinks;
5178 list_add_rcu(&adj->list, dev_list);
5180 list_add_tail_rcu(&adj->list, dev_list);
5186 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5187 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5195 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5196 struct net_device *adj_dev,
5197 struct list_head *dev_list)
5199 struct netdev_adjacent *adj;
5201 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5204 pr_err("tried to remove device %s from %s\n",
5205 dev->name, adj_dev->name);
5209 if (adj->ref_nr > 1) {
5210 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5217 sysfs_remove_link(&(dev->dev.kobj), "master");
5219 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5220 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5222 list_del_rcu(&adj->list);
5223 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5224 adj_dev->name, dev->name, adj_dev->name);
5226 kfree_rcu(adj, rcu);
5229 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5230 struct net_device *upper_dev,
5231 struct list_head *up_list,
5232 struct list_head *down_list,
5233 void *private, bool master)
5237 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5242 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5245 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5252 static int __netdev_adjacent_dev_link(struct net_device *dev,
5253 struct net_device *upper_dev)
5255 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5256 &dev->all_adj_list.upper,
5257 &upper_dev->all_adj_list.lower,
5261 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5262 struct net_device *upper_dev,
5263 struct list_head *up_list,
5264 struct list_head *down_list)
5266 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5267 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5270 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5271 struct net_device *upper_dev)
5273 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5274 &dev->all_adj_list.upper,
5275 &upper_dev->all_adj_list.lower);
5278 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5279 struct net_device *upper_dev,
5280 void *private, bool master)
5282 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5287 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5288 &dev->adj_list.upper,
5289 &upper_dev->adj_list.lower,
5292 __netdev_adjacent_dev_unlink(dev, upper_dev);
5299 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5300 struct net_device *upper_dev)
5302 __netdev_adjacent_dev_unlink(dev, upper_dev);
5303 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5304 &dev->adj_list.upper,
5305 &upper_dev->adj_list.lower);
5308 static int __netdev_upper_dev_link(struct net_device *dev,
5309 struct net_device *upper_dev, bool master,
5312 struct netdev_adjacent *i, *j, *to_i, *to_j;
5317 if (dev == upper_dev)
5320 /* To prevent loops, check if dev is not upper device to upper_dev. */
5321 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5324 if (__netdev_find_adj(dev, upper_dev, &dev->adj_list.upper))
5327 if (master && netdev_master_upper_dev_get(dev))
5330 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5335 /* Now that we linked these devs, make all the upper_dev's
5336 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5337 * versa, and don't forget the devices itself. All of these
5338 * links are non-neighbours.
5340 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5341 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5342 pr_debug("Interlinking %s with %s, non-neighbour\n",
5343 i->dev->name, j->dev->name);
5344 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5350 /* add dev to every upper_dev's upper device */
5351 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5352 pr_debug("linking %s's upper device %s with %s\n",
5353 upper_dev->name, i->dev->name, dev->name);
5354 ret = __netdev_adjacent_dev_link(dev, i->dev);
5356 goto rollback_upper_mesh;
5359 /* add upper_dev to every dev's lower device */
5360 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5361 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5362 i->dev->name, upper_dev->name);
5363 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5365 goto rollback_lower_mesh;
5368 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5371 rollback_lower_mesh:
5373 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5376 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5381 rollback_upper_mesh:
5383 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5386 __netdev_adjacent_dev_unlink(dev, i->dev);
5394 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5395 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5396 if (i == to_i && j == to_j)
5398 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5404 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5410 * netdev_upper_dev_link - Add a link to the upper device
5412 * @upper_dev: new upper device
5414 * Adds a link to device which is upper to this one. The caller must hold
5415 * the RTNL lock. On a failure a negative errno code is returned.
5416 * On success the reference counts are adjusted and the function
5419 int netdev_upper_dev_link(struct net_device *dev,
5420 struct net_device *upper_dev)
5422 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5424 EXPORT_SYMBOL(netdev_upper_dev_link);
5427 * netdev_master_upper_dev_link - Add a master link to the upper device
5429 * @upper_dev: new upper device
5431 * Adds a link to device which is upper to this one. In this case, only
5432 * one master upper device can be linked, although other non-master devices
5433 * might be linked as well. The caller must hold the RTNL lock.
5434 * On a failure a negative errno code is returned. On success the reference
5435 * counts are adjusted and the function returns zero.
5437 int netdev_master_upper_dev_link(struct net_device *dev,
5438 struct net_device *upper_dev)
5440 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5442 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5444 int netdev_master_upper_dev_link_private(struct net_device *dev,
5445 struct net_device *upper_dev,
5448 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5450 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5453 * netdev_upper_dev_unlink - Removes a link to upper device
5455 * @upper_dev: new upper device
5457 * Removes a link to device which is upper to this one. The caller must hold
5460 void netdev_upper_dev_unlink(struct net_device *dev,
5461 struct net_device *upper_dev)
5463 struct netdev_adjacent *i, *j;
5466 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5468 /* Here is the tricky part. We must remove all dev's lower
5469 * devices from all upper_dev's upper devices and vice
5470 * versa, to maintain the graph relationship.
5472 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5473 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5474 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5476 /* remove also the devices itself from lower/upper device
5479 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5480 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5482 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5483 __netdev_adjacent_dev_unlink(dev, i->dev);
5485 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5487 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5490 * netdev_bonding_info_change - Dispatch event about slave change
5492 * @bonding_info: info to dispatch
5494 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5495 * The caller must hold the RTNL lock.
5497 void netdev_bonding_info_change(struct net_device *dev,
5498 struct netdev_bonding_info *bonding_info)
5500 struct netdev_notifier_bonding_info info;
5502 memcpy(&info.bonding_info, bonding_info,
5503 sizeof(struct netdev_bonding_info));
5504 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5507 EXPORT_SYMBOL(netdev_bonding_info_change);
5509 static void netdev_adjacent_add_links(struct net_device *dev)
5511 struct netdev_adjacent *iter;
5513 struct net *net = dev_net(dev);
5515 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5516 if (!net_eq(net,dev_net(iter->dev)))
5518 netdev_adjacent_sysfs_add(iter->dev, dev,
5519 &iter->dev->adj_list.lower);
5520 netdev_adjacent_sysfs_add(dev, iter->dev,
5521 &dev->adj_list.upper);
5524 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5525 if (!net_eq(net,dev_net(iter->dev)))
5527 netdev_adjacent_sysfs_add(iter->dev, dev,
5528 &iter->dev->adj_list.upper);
5529 netdev_adjacent_sysfs_add(dev, iter->dev,
5530 &dev->adj_list.lower);
5534 static void netdev_adjacent_del_links(struct net_device *dev)
5536 struct netdev_adjacent *iter;
5538 struct net *net = dev_net(dev);
5540 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5541 if (!net_eq(net,dev_net(iter->dev)))
5543 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5544 &iter->dev->adj_list.lower);
5545 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5546 &dev->adj_list.upper);
5549 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5550 if (!net_eq(net,dev_net(iter->dev)))
5552 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5553 &iter->dev->adj_list.upper);
5554 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5555 &dev->adj_list.lower);
5559 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5561 struct netdev_adjacent *iter;
5563 struct net *net = dev_net(dev);
5565 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5566 if (!net_eq(net,dev_net(iter->dev)))
5568 netdev_adjacent_sysfs_del(iter->dev, oldname,
5569 &iter->dev->adj_list.lower);
5570 netdev_adjacent_sysfs_add(iter->dev, dev,
5571 &iter->dev->adj_list.lower);
5574 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5575 if (!net_eq(net,dev_net(iter->dev)))
5577 netdev_adjacent_sysfs_del(iter->dev, oldname,
5578 &iter->dev->adj_list.upper);
5579 netdev_adjacent_sysfs_add(iter->dev, dev,
5580 &iter->dev->adj_list.upper);
5584 void *netdev_lower_dev_get_private(struct net_device *dev,
5585 struct net_device *lower_dev)
5587 struct netdev_adjacent *lower;
5591 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5595 return lower->private;
5597 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5600 int dev_get_nest_level(struct net_device *dev,
5601 bool (*type_check)(struct net_device *dev))
5603 struct net_device *lower = NULL;
5604 struct list_head *iter;
5610 netdev_for_each_lower_dev(dev, lower, iter) {
5611 nest = dev_get_nest_level(lower, type_check);
5612 if (max_nest < nest)
5616 if (type_check(dev))
5621 EXPORT_SYMBOL(dev_get_nest_level);
5623 static void dev_change_rx_flags(struct net_device *dev, int flags)
5625 const struct net_device_ops *ops = dev->netdev_ops;
5627 if (ops->ndo_change_rx_flags)
5628 ops->ndo_change_rx_flags(dev, flags);
5631 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5633 unsigned int old_flags = dev->flags;
5639 dev->flags |= IFF_PROMISC;
5640 dev->promiscuity += inc;
5641 if (dev->promiscuity == 0) {
5644 * If inc causes overflow, untouch promisc and return error.
5647 dev->flags &= ~IFF_PROMISC;
5649 dev->promiscuity -= inc;
5650 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5655 if (dev->flags != old_flags) {
5656 pr_info("device %s %s promiscuous mode\n",
5658 dev->flags & IFF_PROMISC ? "entered" : "left");
5659 if (audit_enabled) {
5660 current_uid_gid(&uid, &gid);
5661 audit_log(current->audit_context, GFP_ATOMIC,
5662 AUDIT_ANOM_PROMISCUOUS,
5663 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5664 dev->name, (dev->flags & IFF_PROMISC),
5665 (old_flags & IFF_PROMISC),
5666 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5667 from_kuid(&init_user_ns, uid),
5668 from_kgid(&init_user_ns, gid),
5669 audit_get_sessionid(current));
5672 dev_change_rx_flags(dev, IFF_PROMISC);
5675 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5680 * dev_set_promiscuity - update promiscuity count on a device
5684 * Add or remove promiscuity from a device. While the count in the device
5685 * remains above zero the interface remains promiscuous. Once it hits zero
5686 * the device reverts back to normal filtering operation. A negative inc
5687 * value is used to drop promiscuity on the device.
5688 * Return 0 if successful or a negative errno code on error.
5690 int dev_set_promiscuity(struct net_device *dev, int inc)
5692 unsigned int old_flags = dev->flags;
5695 err = __dev_set_promiscuity(dev, inc, true);
5698 if (dev->flags != old_flags)
5699 dev_set_rx_mode(dev);
5702 EXPORT_SYMBOL(dev_set_promiscuity);
5704 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5706 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5710 dev->flags |= IFF_ALLMULTI;
5711 dev->allmulti += inc;
5712 if (dev->allmulti == 0) {
5715 * If inc causes overflow, untouch allmulti and return error.
5718 dev->flags &= ~IFF_ALLMULTI;
5720 dev->allmulti -= inc;
5721 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5726 if (dev->flags ^ old_flags) {
5727 dev_change_rx_flags(dev, IFF_ALLMULTI);
5728 dev_set_rx_mode(dev);
5730 __dev_notify_flags(dev, old_flags,
5731 dev->gflags ^ old_gflags);
5737 * dev_set_allmulti - update allmulti count on a device
5741 * Add or remove reception of all multicast frames to a device. While the
5742 * count in the device remains above zero the interface remains listening
5743 * to all interfaces. Once it hits zero the device reverts back to normal
5744 * filtering operation. A negative @inc value is used to drop the counter
5745 * when releasing a resource needing all multicasts.
5746 * Return 0 if successful or a negative errno code on error.
5749 int dev_set_allmulti(struct net_device *dev, int inc)
5751 return __dev_set_allmulti(dev, inc, true);
5753 EXPORT_SYMBOL(dev_set_allmulti);
5756 * Upload unicast and multicast address lists to device and
5757 * configure RX filtering. When the device doesn't support unicast
5758 * filtering it is put in promiscuous mode while unicast addresses
5761 void __dev_set_rx_mode(struct net_device *dev)
5763 const struct net_device_ops *ops = dev->netdev_ops;
5765 /* dev_open will call this function so the list will stay sane. */
5766 if (!(dev->flags&IFF_UP))
5769 if (!netif_device_present(dev))
5772 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5773 /* Unicast addresses changes may only happen under the rtnl,
5774 * therefore calling __dev_set_promiscuity here is safe.
5776 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5777 __dev_set_promiscuity(dev, 1, false);
5778 dev->uc_promisc = true;
5779 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5780 __dev_set_promiscuity(dev, -1, false);
5781 dev->uc_promisc = false;
5785 if (ops->ndo_set_rx_mode)
5786 ops->ndo_set_rx_mode(dev);
5789 void dev_set_rx_mode(struct net_device *dev)
5791 netif_addr_lock_bh(dev);
5792 __dev_set_rx_mode(dev);
5793 netif_addr_unlock_bh(dev);
5797 * dev_get_flags - get flags reported to userspace
5800 * Get the combination of flag bits exported through APIs to userspace.
5802 unsigned int dev_get_flags(const struct net_device *dev)
5806 flags = (dev->flags & ~(IFF_PROMISC |
5811 (dev->gflags & (IFF_PROMISC |
5814 if (netif_running(dev)) {
5815 if (netif_oper_up(dev))
5816 flags |= IFF_RUNNING;
5817 if (netif_carrier_ok(dev))
5818 flags |= IFF_LOWER_UP;
5819 if (netif_dormant(dev))
5820 flags |= IFF_DORMANT;
5825 EXPORT_SYMBOL(dev_get_flags);
5827 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5829 unsigned int old_flags = dev->flags;
5835 * Set the flags on our device.
5838 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5839 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5841 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5845 * Load in the correct multicast list now the flags have changed.
5848 if ((old_flags ^ flags) & IFF_MULTICAST)
5849 dev_change_rx_flags(dev, IFF_MULTICAST);
5851 dev_set_rx_mode(dev);
5854 * Have we downed the interface. We handle IFF_UP ourselves
5855 * according to user attempts to set it, rather than blindly
5860 if ((old_flags ^ flags) & IFF_UP)
5861 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5863 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5864 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5865 unsigned int old_flags = dev->flags;
5867 dev->gflags ^= IFF_PROMISC;
5869 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5870 if (dev->flags != old_flags)
5871 dev_set_rx_mode(dev);
5874 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5875 is important. Some (broken) drivers set IFF_PROMISC, when
5876 IFF_ALLMULTI is requested not asking us and not reporting.
5878 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5879 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5881 dev->gflags ^= IFF_ALLMULTI;
5882 __dev_set_allmulti(dev, inc, false);
5888 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5889 unsigned int gchanges)
5891 unsigned int changes = dev->flags ^ old_flags;
5894 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5896 if (changes & IFF_UP) {
5897 if (dev->flags & IFF_UP)
5898 call_netdevice_notifiers(NETDEV_UP, dev);
5900 call_netdevice_notifiers(NETDEV_DOWN, dev);
5903 if (dev->flags & IFF_UP &&
5904 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5905 struct netdev_notifier_change_info change_info;
5907 change_info.flags_changed = changes;
5908 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5914 * dev_change_flags - change device settings
5916 * @flags: device state flags
5918 * Change settings on device based state flags. The flags are
5919 * in the userspace exported format.
5921 int dev_change_flags(struct net_device *dev, unsigned int flags)
5924 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5926 ret = __dev_change_flags(dev, flags);
5930 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5931 __dev_notify_flags(dev, old_flags, changes);
5934 EXPORT_SYMBOL(dev_change_flags);
5936 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5938 const struct net_device_ops *ops = dev->netdev_ops;
5940 if (ops->ndo_change_mtu)
5941 return ops->ndo_change_mtu(dev, new_mtu);
5948 * dev_set_mtu - Change maximum transfer unit
5950 * @new_mtu: new transfer unit
5952 * Change the maximum transfer size of the network device.
5954 int dev_set_mtu(struct net_device *dev, int new_mtu)
5958 if (new_mtu == dev->mtu)
5961 /* MTU must be positive. */
5965 if (!netif_device_present(dev))
5968 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5969 err = notifier_to_errno(err);
5973 orig_mtu = dev->mtu;
5974 err = __dev_set_mtu(dev, new_mtu);
5977 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5978 err = notifier_to_errno(err);
5980 /* setting mtu back and notifying everyone again,
5981 * so that they have a chance to revert changes.
5983 __dev_set_mtu(dev, orig_mtu);
5984 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5989 EXPORT_SYMBOL(dev_set_mtu);
5992 * dev_set_group - Change group this device belongs to
5994 * @new_group: group this device should belong to
5996 void dev_set_group(struct net_device *dev, int new_group)
5998 dev->group = new_group;
6000 EXPORT_SYMBOL(dev_set_group);
6003 * dev_set_mac_address - Change Media Access Control Address
6007 * Change the hardware (MAC) address of the device
6009 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6011 const struct net_device_ops *ops = dev->netdev_ops;
6014 if (!ops->ndo_set_mac_address)
6016 if (sa->sa_family != dev->type)
6018 if (!netif_device_present(dev))
6020 err = ops->ndo_set_mac_address(dev, sa);
6023 dev->addr_assign_type = NET_ADDR_SET;
6024 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6025 add_device_randomness(dev->dev_addr, dev->addr_len);
6028 EXPORT_SYMBOL(dev_set_mac_address);
6031 * dev_change_carrier - Change device carrier
6033 * @new_carrier: new value
6035 * Change device carrier
6037 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6039 const struct net_device_ops *ops = dev->netdev_ops;
6041 if (!ops->ndo_change_carrier)
6043 if (!netif_device_present(dev))
6045 return ops->ndo_change_carrier(dev, new_carrier);
6047 EXPORT_SYMBOL(dev_change_carrier);
6050 * dev_get_phys_port_id - Get device physical port ID
6054 * Get device physical port ID
6056 int dev_get_phys_port_id(struct net_device *dev,
6057 struct netdev_phys_item_id *ppid)
6059 const struct net_device_ops *ops = dev->netdev_ops;
6061 if (!ops->ndo_get_phys_port_id)
6063 return ops->ndo_get_phys_port_id(dev, ppid);
6065 EXPORT_SYMBOL(dev_get_phys_port_id);
6068 * dev_get_phys_port_name - Get device physical port name
6072 * Get device physical port name
6074 int dev_get_phys_port_name(struct net_device *dev,
6075 char *name, size_t len)
6077 const struct net_device_ops *ops = dev->netdev_ops;
6079 if (!ops->ndo_get_phys_port_name)
6081 return ops->ndo_get_phys_port_name(dev, name, len);
6083 EXPORT_SYMBOL(dev_get_phys_port_name);
6086 * dev_new_index - allocate an ifindex
6087 * @net: the applicable net namespace
6089 * Returns a suitable unique value for a new device interface
6090 * number. The caller must hold the rtnl semaphore or the
6091 * dev_base_lock to be sure it remains unique.
6093 static int dev_new_index(struct net *net)
6095 int ifindex = net->ifindex;
6099 if (!__dev_get_by_index(net, ifindex))
6100 return net->ifindex = ifindex;
6104 /* Delayed registration/unregisteration */
6105 static LIST_HEAD(net_todo_list);
6106 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6108 static void net_set_todo(struct net_device *dev)
6110 list_add_tail(&dev->todo_list, &net_todo_list);
6111 dev_net(dev)->dev_unreg_count++;
6114 static void rollback_registered_many(struct list_head *head)
6116 struct net_device *dev, *tmp;
6117 LIST_HEAD(close_head);
6119 BUG_ON(dev_boot_phase);
6122 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6123 /* Some devices call without registering
6124 * for initialization unwind. Remove those
6125 * devices and proceed with the remaining.
6127 if (dev->reg_state == NETREG_UNINITIALIZED) {
6128 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6132 list_del(&dev->unreg_list);
6135 dev->dismantle = true;
6136 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6139 /* If device is running, close it first. */
6140 list_for_each_entry(dev, head, unreg_list)
6141 list_add_tail(&dev->close_list, &close_head);
6142 dev_close_many(&close_head, true);
6144 list_for_each_entry(dev, head, unreg_list) {
6145 /* And unlink it from device chain. */
6146 unlist_netdevice(dev);
6148 dev->reg_state = NETREG_UNREGISTERING;
6153 list_for_each_entry(dev, head, unreg_list) {
6154 struct sk_buff *skb = NULL;
6156 /* Shutdown queueing discipline. */
6160 /* Notify protocols, that we are about to destroy
6161 this device. They should clean all the things.
6163 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6165 if (!dev->rtnl_link_ops ||
6166 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6167 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6171 * Flush the unicast and multicast chains
6176 if (dev->netdev_ops->ndo_uninit)
6177 dev->netdev_ops->ndo_uninit(dev);
6180 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6182 /* Notifier chain MUST detach us all upper devices. */
6183 WARN_ON(netdev_has_any_upper_dev(dev));
6185 /* Remove entries from kobject tree */
6186 netdev_unregister_kobject(dev);
6188 /* Remove XPS queueing entries */
6189 netif_reset_xps_queues_gt(dev, 0);
6195 list_for_each_entry(dev, head, unreg_list)
6199 static void rollback_registered(struct net_device *dev)
6203 list_add(&dev->unreg_list, &single);
6204 rollback_registered_many(&single);
6208 static netdev_features_t netdev_fix_features(struct net_device *dev,
6209 netdev_features_t features)
6211 /* Fix illegal checksum combinations */
6212 if ((features & NETIF_F_HW_CSUM) &&
6213 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6214 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6215 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6218 /* TSO requires that SG is present as well. */
6219 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6220 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6221 features &= ~NETIF_F_ALL_TSO;
6224 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6225 !(features & NETIF_F_IP_CSUM)) {
6226 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6227 features &= ~NETIF_F_TSO;
6228 features &= ~NETIF_F_TSO_ECN;
6231 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6232 !(features & NETIF_F_IPV6_CSUM)) {
6233 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6234 features &= ~NETIF_F_TSO6;
6237 /* TSO ECN requires that TSO is present as well. */
6238 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6239 features &= ~NETIF_F_TSO_ECN;
6241 /* Software GSO depends on SG. */
6242 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6243 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6244 features &= ~NETIF_F_GSO;
6247 /* UFO needs SG and checksumming */
6248 if (features & NETIF_F_UFO) {
6249 /* maybe split UFO into V4 and V6? */
6250 if (!((features & NETIF_F_GEN_CSUM) ||
6251 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6252 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6254 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6255 features &= ~NETIF_F_UFO;
6258 if (!(features & NETIF_F_SG)) {
6260 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6261 features &= ~NETIF_F_UFO;
6265 #ifdef CONFIG_NET_RX_BUSY_POLL
6266 if (dev->netdev_ops->ndo_busy_poll)
6267 features |= NETIF_F_BUSY_POLL;
6270 features &= ~NETIF_F_BUSY_POLL;
6275 int __netdev_update_features(struct net_device *dev)
6277 netdev_features_t features;
6282 features = netdev_get_wanted_features(dev);
6284 if (dev->netdev_ops->ndo_fix_features)
6285 features = dev->netdev_ops->ndo_fix_features(dev, features);
6287 /* driver might be less strict about feature dependencies */
6288 features = netdev_fix_features(dev, features);
6290 if (dev->features == features)
6293 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6294 &dev->features, &features);
6296 if (dev->netdev_ops->ndo_set_features)
6297 err = dev->netdev_ops->ndo_set_features(dev, features);
6299 if (unlikely(err < 0)) {
6301 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6302 err, &features, &dev->features);
6307 dev->features = features;
6313 * netdev_update_features - recalculate device features
6314 * @dev: the device to check
6316 * Recalculate dev->features set and send notifications if it
6317 * has changed. Should be called after driver or hardware dependent
6318 * conditions might have changed that influence the features.
6320 void netdev_update_features(struct net_device *dev)
6322 if (__netdev_update_features(dev))
6323 netdev_features_change(dev);
6325 EXPORT_SYMBOL(netdev_update_features);
6328 * netdev_change_features - recalculate device features
6329 * @dev: the device to check
6331 * Recalculate dev->features set and send notifications even
6332 * if they have not changed. Should be called instead of
6333 * netdev_update_features() if also dev->vlan_features might
6334 * have changed to allow the changes to be propagated to stacked
6337 void netdev_change_features(struct net_device *dev)
6339 __netdev_update_features(dev);
6340 netdev_features_change(dev);
6342 EXPORT_SYMBOL(netdev_change_features);
6345 * netif_stacked_transfer_operstate - transfer operstate
6346 * @rootdev: the root or lower level device to transfer state from
6347 * @dev: the device to transfer operstate to
6349 * Transfer operational state from root to device. This is normally
6350 * called when a stacking relationship exists between the root
6351 * device and the device(a leaf device).
6353 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6354 struct net_device *dev)
6356 if (rootdev->operstate == IF_OPER_DORMANT)
6357 netif_dormant_on(dev);
6359 netif_dormant_off(dev);
6361 if (netif_carrier_ok(rootdev)) {
6362 if (!netif_carrier_ok(dev))
6363 netif_carrier_on(dev);
6365 if (netif_carrier_ok(dev))
6366 netif_carrier_off(dev);
6369 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6372 static int netif_alloc_rx_queues(struct net_device *dev)
6374 unsigned int i, count = dev->num_rx_queues;
6375 struct netdev_rx_queue *rx;
6376 size_t sz = count * sizeof(*rx);
6380 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6388 for (i = 0; i < count; i++)
6394 static void netdev_init_one_queue(struct net_device *dev,
6395 struct netdev_queue *queue, void *_unused)
6397 /* Initialize queue lock */
6398 spin_lock_init(&queue->_xmit_lock);
6399 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6400 queue->xmit_lock_owner = -1;
6401 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6404 dql_init(&queue->dql, HZ);
6408 static void netif_free_tx_queues(struct net_device *dev)
6413 static int netif_alloc_netdev_queues(struct net_device *dev)
6415 unsigned int count = dev->num_tx_queues;
6416 struct netdev_queue *tx;
6417 size_t sz = count * sizeof(*tx);
6419 BUG_ON(count < 1 || count > 0xffff);
6421 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6429 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6430 spin_lock_init(&dev->tx_global_lock);
6435 void netif_tx_stop_all_queues(struct net_device *dev)
6439 for (i = 0; i < dev->num_tx_queues; i++) {
6440 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
6441 netif_tx_stop_queue(txq);
6444 EXPORT_SYMBOL(netif_tx_stop_all_queues);
6447 * register_netdevice - register a network device
6448 * @dev: device to register
6450 * Take a completed network device structure and add it to the kernel
6451 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6452 * chain. 0 is returned on success. A negative errno code is returned
6453 * on a failure to set up the device, or if the name is a duplicate.
6455 * Callers must hold the rtnl semaphore. You may want
6456 * register_netdev() instead of this.
6459 * The locking appears insufficient to guarantee two parallel registers
6460 * will not get the same name.
6463 int register_netdevice(struct net_device *dev)
6466 struct net *net = dev_net(dev);
6468 BUG_ON(dev_boot_phase);
6473 /* When net_device's are persistent, this will be fatal. */
6474 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6477 spin_lock_init(&dev->addr_list_lock);
6478 netdev_set_addr_lockdep_class(dev);
6480 ret = dev_get_valid_name(net, dev, dev->name);
6484 /* Init, if this function is available */
6485 if (dev->netdev_ops->ndo_init) {
6486 ret = dev->netdev_ops->ndo_init(dev);
6494 if (((dev->hw_features | dev->features) &
6495 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6496 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6497 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6498 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6505 dev->ifindex = dev_new_index(net);
6506 else if (__dev_get_by_index(net, dev->ifindex))
6509 /* Transfer changeable features to wanted_features and enable
6510 * software offloads (GSO and GRO).
6512 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6513 dev->features |= NETIF_F_SOFT_FEATURES;
6514 dev->wanted_features = dev->features & dev->hw_features;
6516 if (!(dev->flags & IFF_LOOPBACK)) {
6517 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6520 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6522 dev->vlan_features |= NETIF_F_HIGHDMA;
6524 /* Make NETIF_F_SG inheritable to tunnel devices.
6526 dev->hw_enc_features |= NETIF_F_SG;
6528 /* Make NETIF_F_SG inheritable to MPLS.
6530 dev->mpls_features |= NETIF_F_SG;
6532 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6533 ret = notifier_to_errno(ret);
6537 ret = netdev_register_kobject(dev);
6540 dev->reg_state = NETREG_REGISTERED;
6542 __netdev_update_features(dev);
6545 * Default initial state at registry is that the
6546 * device is present.
6549 set_bit(__LINK_STATE_PRESENT, &dev->state);
6551 linkwatch_init_dev(dev);
6553 dev_init_scheduler(dev);
6555 list_netdevice(dev);
6556 add_device_randomness(dev->dev_addr, dev->addr_len);
6558 /* If the device has permanent device address, driver should
6559 * set dev_addr and also addr_assign_type should be set to
6560 * NET_ADDR_PERM (default value).
6562 if (dev->addr_assign_type == NET_ADDR_PERM)
6563 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6565 /* Notify protocols, that a new device appeared. */
6566 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6567 ret = notifier_to_errno(ret);
6569 rollback_registered(dev);
6570 dev->reg_state = NETREG_UNREGISTERED;
6573 * Prevent userspace races by waiting until the network
6574 * device is fully setup before sending notifications.
6576 if (!dev->rtnl_link_ops ||
6577 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6578 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6584 if (dev->netdev_ops->ndo_uninit)
6585 dev->netdev_ops->ndo_uninit(dev);
6588 EXPORT_SYMBOL(register_netdevice);
6591 * init_dummy_netdev - init a dummy network device for NAPI
6592 * @dev: device to init
6594 * This takes a network device structure and initialize the minimum
6595 * amount of fields so it can be used to schedule NAPI polls without
6596 * registering a full blown interface. This is to be used by drivers
6597 * that need to tie several hardware interfaces to a single NAPI
6598 * poll scheduler due to HW limitations.
6600 int init_dummy_netdev(struct net_device *dev)
6602 /* Clear everything. Note we don't initialize spinlocks
6603 * are they aren't supposed to be taken by any of the
6604 * NAPI code and this dummy netdev is supposed to be
6605 * only ever used for NAPI polls
6607 memset(dev, 0, sizeof(struct net_device));
6609 /* make sure we BUG if trying to hit standard
6610 * register/unregister code path
6612 dev->reg_state = NETREG_DUMMY;
6614 /* NAPI wants this */
6615 INIT_LIST_HEAD(&dev->napi_list);
6617 /* a dummy interface is started by default */
6618 set_bit(__LINK_STATE_PRESENT, &dev->state);
6619 set_bit(__LINK_STATE_START, &dev->state);
6621 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6622 * because users of this 'device' dont need to change
6628 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6632 * register_netdev - register a network device
6633 * @dev: device to register
6635 * Take a completed network device structure and add it to the kernel
6636 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6637 * chain. 0 is returned on success. A negative errno code is returned
6638 * on a failure to set up the device, or if the name is a duplicate.
6640 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6641 * and expands the device name if you passed a format string to
6644 int register_netdev(struct net_device *dev)
6649 err = register_netdevice(dev);
6653 EXPORT_SYMBOL(register_netdev);
6655 int netdev_refcnt_read(const struct net_device *dev)
6659 for_each_possible_cpu(i)
6660 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6663 EXPORT_SYMBOL(netdev_refcnt_read);
6666 * netdev_wait_allrefs - wait until all references are gone.
6667 * @dev: target net_device
6669 * This is called when unregistering network devices.
6671 * Any protocol or device that holds a reference should register
6672 * for netdevice notification, and cleanup and put back the
6673 * reference if they receive an UNREGISTER event.
6674 * We can get stuck here if buggy protocols don't correctly
6677 static void netdev_wait_allrefs(struct net_device *dev)
6679 unsigned long rebroadcast_time, warning_time;
6682 linkwatch_forget_dev(dev);
6684 rebroadcast_time = warning_time = jiffies;
6685 refcnt = netdev_refcnt_read(dev);
6687 while (refcnt != 0) {
6688 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6691 /* Rebroadcast unregister notification */
6692 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6698 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6699 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6701 /* We must not have linkwatch events
6702 * pending on unregister. If this
6703 * happens, we simply run the queue
6704 * unscheduled, resulting in a noop
6707 linkwatch_run_queue();
6712 rebroadcast_time = jiffies;
6717 refcnt = netdev_refcnt_read(dev);
6719 if (time_after(jiffies, warning_time + 10 * HZ)) {
6720 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6722 warning_time = jiffies;
6731 * register_netdevice(x1);
6732 * register_netdevice(x2);
6734 * unregister_netdevice(y1);
6735 * unregister_netdevice(y2);
6741 * We are invoked by rtnl_unlock().
6742 * This allows us to deal with problems:
6743 * 1) We can delete sysfs objects which invoke hotplug
6744 * without deadlocking with linkwatch via keventd.
6745 * 2) Since we run with the RTNL semaphore not held, we can sleep
6746 * safely in order to wait for the netdev refcnt to drop to zero.
6748 * We must not return until all unregister events added during
6749 * the interval the lock was held have been completed.
6751 void netdev_run_todo(void)
6753 struct list_head list;
6755 /* Snapshot list, allow later requests */
6756 list_replace_init(&net_todo_list, &list);
6761 /* Wait for rcu callbacks to finish before next phase */
6762 if (!list_empty(&list))
6765 while (!list_empty(&list)) {
6766 struct net_device *dev
6767 = list_first_entry(&list, struct net_device, todo_list);
6768 list_del(&dev->todo_list);
6771 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6774 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6775 pr_err("network todo '%s' but state %d\n",
6776 dev->name, dev->reg_state);
6781 dev->reg_state = NETREG_UNREGISTERED;
6783 on_each_cpu(flush_backlog, dev, 1);
6785 netdev_wait_allrefs(dev);
6788 BUG_ON(netdev_refcnt_read(dev));
6789 BUG_ON(!list_empty(&dev->ptype_all));
6790 BUG_ON(!list_empty(&dev->ptype_specific));
6791 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6792 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6793 WARN_ON(dev->dn_ptr);
6795 if (dev->destructor)
6796 dev->destructor(dev);
6798 /* Report a network device has been unregistered */
6800 dev_net(dev)->dev_unreg_count--;
6802 wake_up(&netdev_unregistering_wq);
6804 /* Free network device */
6805 kobject_put(&dev->dev.kobj);
6809 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6810 * fields in the same order, with only the type differing.
6812 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6813 const struct net_device_stats *netdev_stats)
6815 #if BITS_PER_LONG == 64
6816 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6817 memcpy(stats64, netdev_stats, sizeof(*stats64));
6819 size_t i, n = sizeof(*stats64) / sizeof(u64);
6820 const unsigned long *src = (const unsigned long *)netdev_stats;
6821 u64 *dst = (u64 *)stats64;
6823 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6824 sizeof(*stats64) / sizeof(u64));
6825 for (i = 0; i < n; i++)
6829 EXPORT_SYMBOL(netdev_stats_to_stats64);
6832 * dev_get_stats - get network device statistics
6833 * @dev: device to get statistics from
6834 * @storage: place to store stats
6836 * Get network statistics from device. Return @storage.
6837 * The device driver may provide its own method by setting
6838 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6839 * otherwise the internal statistics structure is used.
6841 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6842 struct rtnl_link_stats64 *storage)
6844 const struct net_device_ops *ops = dev->netdev_ops;
6846 if (ops->ndo_get_stats64) {
6847 memset(storage, 0, sizeof(*storage));
6848 ops->ndo_get_stats64(dev, storage);
6849 } else if (ops->ndo_get_stats) {
6850 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6852 netdev_stats_to_stats64(storage, &dev->stats);
6854 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6855 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6858 EXPORT_SYMBOL(dev_get_stats);
6860 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6862 struct netdev_queue *queue = dev_ingress_queue(dev);
6864 #ifdef CONFIG_NET_CLS_ACT
6867 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6870 netdev_init_one_queue(dev, queue, NULL);
6871 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
6872 queue->qdisc_sleeping = &noop_qdisc;
6873 rcu_assign_pointer(dev->ingress_queue, queue);
6878 static const struct ethtool_ops default_ethtool_ops;
6880 void netdev_set_default_ethtool_ops(struct net_device *dev,
6881 const struct ethtool_ops *ops)
6883 if (dev->ethtool_ops == &default_ethtool_ops)
6884 dev->ethtool_ops = ops;
6886 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6888 void netdev_freemem(struct net_device *dev)
6890 char *addr = (char *)dev - dev->padded;
6896 * alloc_netdev_mqs - allocate network device
6897 * @sizeof_priv: size of private data to allocate space for
6898 * @name: device name format string
6899 * @name_assign_type: origin of device name
6900 * @setup: callback to initialize device
6901 * @txqs: the number of TX subqueues to allocate
6902 * @rxqs: the number of RX subqueues to allocate
6904 * Allocates a struct net_device with private data area for driver use
6905 * and performs basic initialization. Also allocates subqueue structs
6906 * for each queue on the device.
6908 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6909 unsigned char name_assign_type,
6910 void (*setup)(struct net_device *),
6911 unsigned int txqs, unsigned int rxqs)
6913 struct net_device *dev;
6915 struct net_device *p;
6917 BUG_ON(strlen(name) >= sizeof(dev->name));
6920 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6926 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6931 alloc_size = sizeof(struct net_device);
6933 /* ensure 32-byte alignment of private area */
6934 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6935 alloc_size += sizeof_priv;
6937 /* ensure 32-byte alignment of whole construct */
6938 alloc_size += NETDEV_ALIGN - 1;
6940 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6942 p = vzalloc(alloc_size);
6946 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6947 dev->padded = (char *)dev - (char *)p;
6949 dev->pcpu_refcnt = alloc_percpu(int);
6950 if (!dev->pcpu_refcnt)
6953 if (dev_addr_init(dev))
6959 dev_net_set(dev, &init_net);
6961 dev->gso_max_size = GSO_MAX_SIZE;
6962 dev->gso_max_segs = GSO_MAX_SEGS;
6963 dev->gso_min_segs = 0;
6965 INIT_LIST_HEAD(&dev->napi_list);
6966 INIT_LIST_HEAD(&dev->unreg_list);
6967 INIT_LIST_HEAD(&dev->close_list);
6968 INIT_LIST_HEAD(&dev->link_watch_list);
6969 INIT_LIST_HEAD(&dev->adj_list.upper);
6970 INIT_LIST_HEAD(&dev->adj_list.lower);
6971 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6972 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6973 INIT_LIST_HEAD(&dev->ptype_all);
6974 INIT_LIST_HEAD(&dev->ptype_specific);
6975 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
6978 dev->num_tx_queues = txqs;
6979 dev->real_num_tx_queues = txqs;
6980 if (netif_alloc_netdev_queues(dev))
6984 dev->num_rx_queues = rxqs;
6985 dev->real_num_rx_queues = rxqs;
6986 if (netif_alloc_rx_queues(dev))
6990 strcpy(dev->name, name);
6991 dev->name_assign_type = name_assign_type;
6992 dev->group = INIT_NETDEV_GROUP;
6993 if (!dev->ethtool_ops)
6994 dev->ethtool_ops = &default_ethtool_ops;
6996 nf_hook_ingress_init(dev);
7005 free_percpu(dev->pcpu_refcnt);
7007 netdev_freemem(dev);
7010 EXPORT_SYMBOL(alloc_netdev_mqs);
7013 * free_netdev - free network device
7016 * This function does the last stage of destroying an allocated device
7017 * interface. The reference to the device object is released.
7018 * If this is the last reference then it will be freed.
7020 void free_netdev(struct net_device *dev)
7022 struct napi_struct *p, *n;
7024 netif_free_tx_queues(dev);
7029 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7031 /* Flush device addresses */
7032 dev_addr_flush(dev);
7034 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7037 free_percpu(dev->pcpu_refcnt);
7038 dev->pcpu_refcnt = NULL;
7040 /* Compatibility with error handling in drivers */
7041 if (dev->reg_state == NETREG_UNINITIALIZED) {
7042 netdev_freemem(dev);
7046 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7047 dev->reg_state = NETREG_RELEASED;
7049 /* will free via device release */
7050 put_device(&dev->dev);
7052 EXPORT_SYMBOL(free_netdev);
7055 * synchronize_net - Synchronize with packet receive processing
7057 * Wait for packets currently being received to be done.
7058 * Does not block later packets from starting.
7060 void synchronize_net(void)
7063 if (rtnl_is_locked())
7064 synchronize_rcu_expedited();
7068 EXPORT_SYMBOL(synchronize_net);
7071 * unregister_netdevice_queue - remove device from the kernel
7075 * This function shuts down a device interface and removes it
7076 * from the kernel tables.
7077 * If head not NULL, device is queued to be unregistered later.
7079 * Callers must hold the rtnl semaphore. You may want
7080 * unregister_netdev() instead of this.
7083 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7088 list_move_tail(&dev->unreg_list, head);
7090 rollback_registered(dev);
7091 /* Finish processing unregister after unlock */
7095 EXPORT_SYMBOL(unregister_netdevice_queue);
7098 * unregister_netdevice_many - unregister many devices
7099 * @head: list of devices
7101 * Note: As most callers use a stack allocated list_head,
7102 * we force a list_del() to make sure stack wont be corrupted later.
7104 void unregister_netdevice_many(struct list_head *head)
7106 struct net_device *dev;
7108 if (!list_empty(head)) {
7109 rollback_registered_many(head);
7110 list_for_each_entry(dev, head, unreg_list)
7115 EXPORT_SYMBOL(unregister_netdevice_many);
7118 * unregister_netdev - remove device from the kernel
7121 * This function shuts down a device interface and removes it
7122 * from the kernel tables.
7124 * This is just a wrapper for unregister_netdevice that takes
7125 * the rtnl semaphore. In general you want to use this and not
7126 * unregister_netdevice.
7128 void unregister_netdev(struct net_device *dev)
7131 unregister_netdevice(dev);
7134 EXPORT_SYMBOL(unregister_netdev);
7137 * dev_change_net_namespace - move device to different nethost namespace
7139 * @net: network namespace
7140 * @pat: If not NULL name pattern to try if the current device name
7141 * is already taken in the destination network namespace.
7143 * This function shuts down a device interface and moves it
7144 * to a new network namespace. On success 0 is returned, on
7145 * a failure a netagive errno code is returned.
7147 * Callers must hold the rtnl semaphore.
7150 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7156 /* Don't allow namespace local devices to be moved. */
7158 if (dev->features & NETIF_F_NETNS_LOCAL)
7161 /* Ensure the device has been registrered */
7162 if (dev->reg_state != NETREG_REGISTERED)
7165 /* Get out if there is nothing todo */
7167 if (net_eq(dev_net(dev), net))
7170 /* Pick the destination device name, and ensure
7171 * we can use it in the destination network namespace.
7174 if (__dev_get_by_name(net, dev->name)) {
7175 /* We get here if we can't use the current device name */
7178 if (dev_get_valid_name(net, dev, pat) < 0)
7183 * And now a mini version of register_netdevice unregister_netdevice.
7186 /* If device is running close it first. */
7189 /* And unlink it from device chain */
7191 unlist_netdevice(dev);
7195 /* Shutdown queueing discipline. */
7198 /* Notify protocols, that we are about to destroy
7199 this device. They should clean all the things.
7201 Note that dev->reg_state stays at NETREG_REGISTERED.
7202 This is wanted because this way 8021q and macvlan know
7203 the device is just moving and can keep their slaves up.
7205 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7207 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7208 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7211 * Flush the unicast and multicast chains
7216 /* Send a netdev-removed uevent to the old namespace */
7217 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7218 netdev_adjacent_del_links(dev);
7220 /* Actually switch the network namespace */
7221 dev_net_set(dev, net);
7223 /* If there is an ifindex conflict assign a new one */
7224 if (__dev_get_by_index(net, dev->ifindex))
7225 dev->ifindex = dev_new_index(net);
7227 /* Send a netdev-add uevent to the new namespace */
7228 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7229 netdev_adjacent_add_links(dev);
7231 /* Fixup kobjects */
7232 err = device_rename(&dev->dev, dev->name);
7235 /* Add the device back in the hashes */
7236 list_netdevice(dev);
7238 /* Notify protocols, that a new device appeared. */
7239 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7242 * Prevent userspace races by waiting until the network
7243 * device is fully setup before sending notifications.
7245 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7252 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7254 static int dev_cpu_callback(struct notifier_block *nfb,
7255 unsigned long action,
7258 struct sk_buff **list_skb;
7259 struct sk_buff *skb;
7260 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7261 struct softnet_data *sd, *oldsd;
7263 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7266 local_irq_disable();
7267 cpu = smp_processor_id();
7268 sd = &per_cpu(softnet_data, cpu);
7269 oldsd = &per_cpu(softnet_data, oldcpu);
7271 /* Find end of our completion_queue. */
7272 list_skb = &sd->completion_queue;
7274 list_skb = &(*list_skb)->next;
7275 /* Append completion queue from offline CPU. */
7276 *list_skb = oldsd->completion_queue;
7277 oldsd->completion_queue = NULL;
7279 /* Append output queue from offline CPU. */
7280 if (oldsd->output_queue) {
7281 *sd->output_queue_tailp = oldsd->output_queue;
7282 sd->output_queue_tailp = oldsd->output_queue_tailp;
7283 oldsd->output_queue = NULL;
7284 oldsd->output_queue_tailp = &oldsd->output_queue;
7286 /* Append NAPI poll list from offline CPU, with one exception :
7287 * process_backlog() must be called by cpu owning percpu backlog.
7288 * We properly handle process_queue & input_pkt_queue later.
7290 while (!list_empty(&oldsd->poll_list)) {
7291 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7295 list_del_init(&napi->poll_list);
7296 if (napi->poll == process_backlog)
7299 ____napi_schedule(sd, napi);
7302 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7305 /* Process offline CPU's input_pkt_queue */
7306 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7308 input_queue_head_incr(oldsd);
7310 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7312 input_queue_head_incr(oldsd);
7320 * netdev_increment_features - increment feature set by one
7321 * @all: current feature set
7322 * @one: new feature set
7323 * @mask: mask feature set
7325 * Computes a new feature set after adding a device with feature set
7326 * @one to the master device with current feature set @all. Will not
7327 * enable anything that is off in @mask. Returns the new feature set.
7329 netdev_features_t netdev_increment_features(netdev_features_t all,
7330 netdev_features_t one, netdev_features_t mask)
7332 if (mask & NETIF_F_GEN_CSUM)
7333 mask |= NETIF_F_ALL_CSUM;
7334 mask |= NETIF_F_VLAN_CHALLENGED;
7336 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7337 all &= one | ~NETIF_F_ALL_FOR_ALL;
7339 /* If one device supports hw checksumming, set for all. */
7340 if (all & NETIF_F_GEN_CSUM)
7341 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7345 EXPORT_SYMBOL(netdev_increment_features);
7347 static struct hlist_head * __net_init netdev_create_hash(void)
7350 struct hlist_head *hash;
7352 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7354 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7355 INIT_HLIST_HEAD(&hash[i]);
7360 /* Initialize per network namespace state */
7361 static int __net_init netdev_init(struct net *net)
7363 if (net != &init_net)
7364 INIT_LIST_HEAD(&net->dev_base_head);
7366 net->dev_name_head = netdev_create_hash();
7367 if (net->dev_name_head == NULL)
7370 net->dev_index_head = netdev_create_hash();
7371 if (net->dev_index_head == NULL)
7377 kfree(net->dev_name_head);
7383 * netdev_drivername - network driver for the device
7384 * @dev: network device
7386 * Determine network driver for device.
7388 const char *netdev_drivername(const struct net_device *dev)
7390 const struct device_driver *driver;
7391 const struct device *parent;
7392 const char *empty = "";
7394 parent = dev->dev.parent;
7398 driver = parent->driver;
7399 if (driver && driver->name)
7400 return driver->name;
7404 static void __netdev_printk(const char *level, const struct net_device *dev,
7405 struct va_format *vaf)
7407 if (dev && dev->dev.parent) {
7408 dev_printk_emit(level[1] - '0',
7411 dev_driver_string(dev->dev.parent),
7412 dev_name(dev->dev.parent),
7413 netdev_name(dev), netdev_reg_state(dev),
7416 printk("%s%s%s: %pV",
7417 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7419 printk("%s(NULL net_device): %pV", level, vaf);
7423 void netdev_printk(const char *level, const struct net_device *dev,
7424 const char *format, ...)
7426 struct va_format vaf;
7429 va_start(args, format);
7434 __netdev_printk(level, dev, &vaf);
7438 EXPORT_SYMBOL(netdev_printk);
7440 #define define_netdev_printk_level(func, level) \
7441 void func(const struct net_device *dev, const char *fmt, ...) \
7443 struct va_format vaf; \
7446 va_start(args, fmt); \
7451 __netdev_printk(level, dev, &vaf); \
7455 EXPORT_SYMBOL(func);
7457 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7458 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7459 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7460 define_netdev_printk_level(netdev_err, KERN_ERR);
7461 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7462 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7463 define_netdev_printk_level(netdev_info, KERN_INFO);
7465 static void __net_exit netdev_exit(struct net *net)
7467 kfree(net->dev_name_head);
7468 kfree(net->dev_index_head);
7471 static struct pernet_operations __net_initdata netdev_net_ops = {
7472 .init = netdev_init,
7473 .exit = netdev_exit,
7476 static void __net_exit default_device_exit(struct net *net)
7478 struct net_device *dev, *aux;
7480 * Push all migratable network devices back to the
7481 * initial network namespace
7484 for_each_netdev_safe(net, dev, aux) {
7486 char fb_name[IFNAMSIZ];
7488 /* Ignore unmoveable devices (i.e. loopback) */
7489 if (dev->features & NETIF_F_NETNS_LOCAL)
7492 /* Leave virtual devices for the generic cleanup */
7493 if (dev->rtnl_link_ops)
7496 /* Push remaining network devices to init_net */
7497 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7498 err = dev_change_net_namespace(dev, &init_net, fb_name);
7500 pr_emerg("%s: failed to move %s to init_net: %d\n",
7501 __func__, dev->name, err);
7508 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7510 /* Return with the rtnl_lock held when there are no network
7511 * devices unregistering in any network namespace in net_list.
7515 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7517 add_wait_queue(&netdev_unregistering_wq, &wait);
7519 unregistering = false;
7521 list_for_each_entry(net, net_list, exit_list) {
7522 if (net->dev_unreg_count > 0) {
7523 unregistering = true;
7531 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7533 remove_wait_queue(&netdev_unregistering_wq, &wait);
7536 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7538 /* At exit all network devices most be removed from a network
7539 * namespace. Do this in the reverse order of registration.
7540 * Do this across as many network namespaces as possible to
7541 * improve batching efficiency.
7543 struct net_device *dev;
7545 LIST_HEAD(dev_kill_list);
7547 /* To prevent network device cleanup code from dereferencing
7548 * loopback devices or network devices that have been freed
7549 * wait here for all pending unregistrations to complete,
7550 * before unregistring the loopback device and allowing the
7551 * network namespace be freed.
7553 * The netdev todo list containing all network devices
7554 * unregistrations that happen in default_device_exit_batch
7555 * will run in the rtnl_unlock() at the end of
7556 * default_device_exit_batch.
7558 rtnl_lock_unregistering(net_list);
7559 list_for_each_entry(net, net_list, exit_list) {
7560 for_each_netdev_reverse(net, dev) {
7561 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7562 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7564 unregister_netdevice_queue(dev, &dev_kill_list);
7567 unregister_netdevice_many(&dev_kill_list);
7571 static struct pernet_operations __net_initdata default_device_ops = {
7572 .exit = default_device_exit,
7573 .exit_batch = default_device_exit_batch,
7577 * Initialize the DEV module. At boot time this walks the device list and
7578 * unhooks any devices that fail to initialise (normally hardware not
7579 * present) and leaves us with a valid list of present and active devices.
7584 * This is called single threaded during boot, so no need
7585 * to take the rtnl semaphore.
7587 static int __init net_dev_init(void)
7589 int i, rc = -ENOMEM;
7591 BUG_ON(!dev_boot_phase);
7593 if (dev_proc_init())
7596 if (netdev_kobject_init())
7599 INIT_LIST_HEAD(&ptype_all);
7600 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7601 INIT_LIST_HEAD(&ptype_base[i]);
7603 INIT_LIST_HEAD(&offload_base);
7605 if (register_pernet_subsys(&netdev_net_ops))
7609 * Initialise the packet receive queues.
7612 for_each_possible_cpu(i) {
7613 struct softnet_data *sd = &per_cpu(softnet_data, i);
7615 skb_queue_head_init(&sd->input_pkt_queue);
7616 skb_queue_head_init(&sd->process_queue);
7617 INIT_LIST_HEAD(&sd->poll_list);
7618 sd->output_queue_tailp = &sd->output_queue;
7620 sd->csd.func = rps_trigger_softirq;
7625 sd->backlog.poll = process_backlog;
7626 sd->backlog.weight = weight_p;
7631 /* The loopback device is special if any other network devices
7632 * is present in a network namespace the loopback device must
7633 * be present. Since we now dynamically allocate and free the
7634 * loopback device ensure this invariant is maintained by
7635 * keeping the loopback device as the first device on the
7636 * list of network devices. Ensuring the loopback devices
7637 * is the first device that appears and the last network device
7640 if (register_pernet_device(&loopback_net_ops))
7643 if (register_pernet_device(&default_device_ops))
7646 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7647 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7649 hotcpu_notifier(dev_cpu_callback, 0);
7656 subsys_initcall(net_dev_init);