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);
682 EXPORT_SYMBOL(dev_get_iflink);
685 * __dev_get_by_name - find a device by its name
686 * @net: the applicable net namespace
687 * @name: name to find
689 * Find an interface by name. Must be called under RTNL semaphore
690 * or @dev_base_lock. If the name is found a pointer to the device
691 * is returned. If the name is not found then %NULL is returned. The
692 * reference counters are not incremented so the caller must be
693 * careful with locks.
696 struct net_device *__dev_get_by_name(struct net *net, const char *name)
698 struct net_device *dev;
699 struct hlist_head *head = dev_name_hash(net, name);
701 hlist_for_each_entry(dev, head, name_hlist)
702 if (!strncmp(dev->name, name, IFNAMSIZ))
707 EXPORT_SYMBOL(__dev_get_by_name);
710 * dev_get_by_name_rcu - find a device by its name
711 * @net: the applicable net namespace
712 * @name: name to find
714 * Find an interface by name.
715 * If the name is found a pointer to the device is returned.
716 * If the name is not found then %NULL is returned.
717 * The reference counters are not incremented so the caller must be
718 * careful with locks. The caller must hold RCU lock.
721 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
723 struct net_device *dev;
724 struct hlist_head *head = dev_name_hash(net, name);
726 hlist_for_each_entry_rcu(dev, head, name_hlist)
727 if (!strncmp(dev->name, name, IFNAMSIZ))
732 EXPORT_SYMBOL(dev_get_by_name_rcu);
735 * dev_get_by_name - find a device by its name
736 * @net: the applicable net namespace
737 * @name: name to find
739 * Find an interface by name. This can be called from any
740 * context and does its own locking. The returned handle has
741 * the usage count incremented and the caller must use dev_put() to
742 * release it when it is no longer needed. %NULL is returned if no
743 * matching device is found.
746 struct net_device *dev_get_by_name(struct net *net, const char *name)
748 struct net_device *dev;
751 dev = dev_get_by_name_rcu(net, name);
757 EXPORT_SYMBOL(dev_get_by_name);
760 * __dev_get_by_index - find a device by its ifindex
761 * @net: the applicable net namespace
762 * @ifindex: index of device
764 * Search for an interface by index. Returns %NULL if the device
765 * is not found or a pointer to the device. The device has not
766 * had its reference counter increased so the caller must be careful
767 * about locking. The caller must hold either the RTNL semaphore
771 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
773 struct net_device *dev;
774 struct hlist_head *head = dev_index_hash(net, ifindex);
776 hlist_for_each_entry(dev, head, index_hlist)
777 if (dev->ifindex == ifindex)
782 EXPORT_SYMBOL(__dev_get_by_index);
785 * dev_get_by_index_rcu - find a device by its ifindex
786 * @net: the applicable net namespace
787 * @ifindex: index of device
789 * Search for an interface by index. Returns %NULL if the device
790 * is not found or a pointer to the device. The device has not
791 * had its reference counter increased so the caller must be careful
792 * about locking. The caller must hold RCU lock.
795 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
797 struct net_device *dev;
798 struct hlist_head *head = dev_index_hash(net, ifindex);
800 hlist_for_each_entry_rcu(dev, head, index_hlist)
801 if (dev->ifindex == ifindex)
806 EXPORT_SYMBOL(dev_get_by_index_rcu);
810 * dev_get_by_index - find a device by its ifindex
811 * @net: the applicable net namespace
812 * @ifindex: index of device
814 * Search for an interface by index. Returns NULL if the device
815 * is not found or a pointer to the device. The device returned has
816 * had a reference added and the pointer is safe until the user calls
817 * dev_put to indicate they have finished with it.
820 struct net_device *dev_get_by_index(struct net *net, int ifindex)
822 struct net_device *dev;
825 dev = dev_get_by_index_rcu(net, ifindex);
831 EXPORT_SYMBOL(dev_get_by_index);
834 * netdev_get_name - get a netdevice name, knowing its ifindex.
835 * @net: network namespace
836 * @name: a pointer to the buffer where the name will be stored.
837 * @ifindex: the ifindex of the interface to get the name from.
839 * The use of raw_seqcount_begin() and cond_resched() before
840 * retrying is required as we want to give the writers a chance
841 * to complete when CONFIG_PREEMPT is not set.
843 int netdev_get_name(struct net *net, char *name, int ifindex)
845 struct net_device *dev;
849 seq = raw_seqcount_begin(&devnet_rename_seq);
851 dev = dev_get_by_index_rcu(net, ifindex);
857 strcpy(name, dev->name);
859 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
868 * dev_getbyhwaddr_rcu - find a device by its hardware address
869 * @net: the applicable net namespace
870 * @type: media type of device
871 * @ha: hardware address
873 * Search for an interface by MAC address. Returns NULL if the device
874 * is not found or a pointer to the device.
875 * The caller must hold RCU or RTNL.
876 * The returned device has not had its ref count increased
877 * and the caller must therefore be careful about locking
881 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
884 struct net_device *dev;
886 for_each_netdev_rcu(net, dev)
887 if (dev->type == type &&
888 !memcmp(dev->dev_addr, ha, dev->addr_len))
893 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
895 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
897 struct net_device *dev;
900 for_each_netdev(net, dev)
901 if (dev->type == type)
906 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
908 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
910 struct net_device *dev, *ret = NULL;
913 for_each_netdev_rcu(net, dev)
914 if (dev->type == type) {
922 EXPORT_SYMBOL(dev_getfirstbyhwtype);
925 * __dev_get_by_flags - find any device with given flags
926 * @net: the applicable net namespace
927 * @if_flags: IFF_* values
928 * @mask: bitmask of bits in if_flags to check
930 * Search for any interface with the given flags. Returns NULL if a device
931 * is not found or a pointer to the device. Must be called inside
932 * rtnl_lock(), and result refcount is unchanged.
935 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
938 struct net_device *dev, *ret;
943 for_each_netdev(net, dev) {
944 if (((dev->flags ^ if_flags) & mask) == 0) {
951 EXPORT_SYMBOL(__dev_get_by_flags);
954 * dev_valid_name - check if name is okay for network device
957 * Network device names need to be valid file names to
958 * to allow sysfs to work. We also disallow any kind of
961 bool dev_valid_name(const char *name)
965 if (strlen(name) >= IFNAMSIZ)
967 if (!strcmp(name, ".") || !strcmp(name, ".."))
971 if (*name == '/' || *name == ':' || isspace(*name))
977 EXPORT_SYMBOL(dev_valid_name);
980 * __dev_alloc_name - allocate a name for a device
981 * @net: network namespace to allocate the device name in
982 * @name: name format string
983 * @buf: scratch buffer and result name string
985 * Passed a format string - eg "lt%d" it will try and find a suitable
986 * id. It scans list of devices to build up a free map, then chooses
987 * the first empty slot. The caller must hold the dev_base or rtnl lock
988 * while allocating the name and adding the device in order to avoid
990 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
991 * Returns the number of the unit assigned or a negative errno code.
994 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
998 const int max_netdevices = 8*PAGE_SIZE;
999 unsigned long *inuse;
1000 struct net_device *d;
1002 p = strnchr(name, IFNAMSIZ-1, '%');
1005 * Verify the string as this thing may have come from
1006 * the user. There must be either one "%d" and no other "%"
1009 if (p[1] != 'd' || strchr(p + 2, '%'))
1012 /* Use one page as a bit array of possible slots */
1013 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1017 for_each_netdev(net, d) {
1018 if (!sscanf(d->name, name, &i))
1020 if (i < 0 || i >= max_netdevices)
1023 /* avoid cases where sscanf is not exact inverse of printf */
1024 snprintf(buf, IFNAMSIZ, name, i);
1025 if (!strncmp(buf, d->name, IFNAMSIZ))
1029 i = find_first_zero_bit(inuse, max_netdevices);
1030 free_page((unsigned long) inuse);
1034 snprintf(buf, IFNAMSIZ, name, i);
1035 if (!__dev_get_by_name(net, buf))
1038 /* It is possible to run out of possible slots
1039 * when the name is long and there isn't enough space left
1040 * for the digits, or if all bits are used.
1046 * dev_alloc_name - allocate a name for a device
1048 * @name: name format string
1050 * Passed a format string - eg "lt%d" it will try and find a suitable
1051 * id. It scans list of devices to build up a free map, then chooses
1052 * the first empty slot. The caller must hold the dev_base or rtnl lock
1053 * while allocating the name and adding the device in order to avoid
1055 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1056 * Returns the number of the unit assigned or a negative errno code.
1059 int dev_alloc_name(struct net_device *dev, const char *name)
1065 BUG_ON(!dev_net(dev));
1067 ret = __dev_alloc_name(net, name, buf);
1069 strlcpy(dev->name, buf, IFNAMSIZ);
1072 EXPORT_SYMBOL(dev_alloc_name);
1074 static int dev_alloc_name_ns(struct net *net,
1075 struct net_device *dev,
1081 ret = __dev_alloc_name(net, name, buf);
1083 strlcpy(dev->name, buf, IFNAMSIZ);
1087 static int dev_get_valid_name(struct net *net,
1088 struct net_device *dev,
1093 if (!dev_valid_name(name))
1096 if (strchr(name, '%'))
1097 return dev_alloc_name_ns(net, dev, name);
1098 else if (__dev_get_by_name(net, name))
1100 else if (dev->name != name)
1101 strlcpy(dev->name, name, IFNAMSIZ);
1107 * dev_change_name - change name of a device
1109 * @newname: name (or format string) must be at least IFNAMSIZ
1111 * Change name of a device, can pass format strings "eth%d".
1114 int dev_change_name(struct net_device *dev, const char *newname)
1116 unsigned char old_assign_type;
1117 char oldname[IFNAMSIZ];
1123 BUG_ON(!dev_net(dev));
1126 if (dev->flags & IFF_UP)
1129 write_seqcount_begin(&devnet_rename_seq);
1131 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1132 write_seqcount_end(&devnet_rename_seq);
1136 memcpy(oldname, dev->name, IFNAMSIZ);
1138 err = dev_get_valid_name(net, dev, newname);
1140 write_seqcount_end(&devnet_rename_seq);
1144 if (oldname[0] && !strchr(oldname, '%'))
1145 netdev_info(dev, "renamed from %s\n", oldname);
1147 old_assign_type = dev->name_assign_type;
1148 dev->name_assign_type = NET_NAME_RENAMED;
1151 ret = device_rename(&dev->dev, dev->name);
1153 memcpy(dev->name, oldname, IFNAMSIZ);
1154 dev->name_assign_type = old_assign_type;
1155 write_seqcount_end(&devnet_rename_seq);
1159 write_seqcount_end(&devnet_rename_seq);
1161 netdev_adjacent_rename_links(dev, oldname);
1163 write_lock_bh(&dev_base_lock);
1164 hlist_del_rcu(&dev->name_hlist);
1165 write_unlock_bh(&dev_base_lock);
1169 write_lock_bh(&dev_base_lock);
1170 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1171 write_unlock_bh(&dev_base_lock);
1173 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1174 ret = notifier_to_errno(ret);
1177 /* err >= 0 after dev_alloc_name() or stores the first errno */
1180 write_seqcount_begin(&devnet_rename_seq);
1181 memcpy(dev->name, oldname, IFNAMSIZ);
1182 memcpy(oldname, newname, IFNAMSIZ);
1183 dev->name_assign_type = old_assign_type;
1184 old_assign_type = NET_NAME_RENAMED;
1187 pr_err("%s: name change rollback failed: %d\n",
1196 * dev_set_alias - change ifalias of a device
1198 * @alias: name up to IFALIASZ
1199 * @len: limit of bytes to copy from info
1201 * Set ifalias for a device,
1203 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1209 if (len >= IFALIASZ)
1213 kfree(dev->ifalias);
1214 dev->ifalias = NULL;
1218 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1221 dev->ifalias = new_ifalias;
1223 strlcpy(dev->ifalias, alias, len+1);
1229 * netdev_features_change - device changes features
1230 * @dev: device to cause notification
1232 * Called to indicate a device has changed features.
1234 void netdev_features_change(struct net_device *dev)
1236 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1238 EXPORT_SYMBOL(netdev_features_change);
1241 * netdev_state_change - device changes state
1242 * @dev: device to cause notification
1244 * Called to indicate a device has changed state. This function calls
1245 * the notifier chains for netdev_chain and sends a NEWLINK message
1246 * to the routing socket.
1248 void netdev_state_change(struct net_device *dev)
1250 if (dev->flags & IFF_UP) {
1251 struct netdev_notifier_change_info change_info;
1253 change_info.flags_changed = 0;
1254 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1256 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1259 EXPORT_SYMBOL(netdev_state_change);
1262 * netdev_notify_peers - notify network peers about existence of @dev
1263 * @dev: network device
1265 * Generate traffic such that interested network peers are aware of
1266 * @dev, such as by generating a gratuitous ARP. This may be used when
1267 * a device wants to inform the rest of the network about some sort of
1268 * reconfiguration such as a failover event or virtual machine
1271 void netdev_notify_peers(struct net_device *dev)
1274 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1277 EXPORT_SYMBOL(netdev_notify_peers);
1279 static int __dev_open(struct net_device *dev)
1281 const struct net_device_ops *ops = dev->netdev_ops;
1286 if (!netif_device_present(dev))
1289 /* Block netpoll from trying to do any rx path servicing.
1290 * If we don't do this there is a chance ndo_poll_controller
1291 * or ndo_poll may be running while we open the device
1293 netpoll_poll_disable(dev);
1295 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1296 ret = notifier_to_errno(ret);
1300 set_bit(__LINK_STATE_START, &dev->state);
1302 if (ops->ndo_validate_addr)
1303 ret = ops->ndo_validate_addr(dev);
1305 if (!ret && ops->ndo_open)
1306 ret = ops->ndo_open(dev);
1308 netpoll_poll_enable(dev);
1311 clear_bit(__LINK_STATE_START, &dev->state);
1313 dev->flags |= IFF_UP;
1314 dev_set_rx_mode(dev);
1316 add_device_randomness(dev->dev_addr, dev->addr_len);
1323 * dev_open - prepare an interface for use.
1324 * @dev: device to open
1326 * Takes a device from down to up state. The device's private open
1327 * function is invoked and then the multicast lists are loaded. Finally
1328 * the device is moved into the up state and a %NETDEV_UP message is
1329 * sent to the netdev notifier chain.
1331 * Calling this function on an active interface is a nop. On a failure
1332 * a negative errno code is returned.
1334 int dev_open(struct net_device *dev)
1338 if (dev->flags & IFF_UP)
1341 ret = __dev_open(dev);
1345 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1346 call_netdevice_notifiers(NETDEV_UP, dev);
1350 EXPORT_SYMBOL(dev_open);
1352 static int __dev_close_many(struct list_head *head)
1354 struct net_device *dev;
1359 list_for_each_entry(dev, head, close_list) {
1360 /* Temporarily disable netpoll until the interface is down */
1361 netpoll_poll_disable(dev);
1363 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1365 clear_bit(__LINK_STATE_START, &dev->state);
1367 /* Synchronize to scheduled poll. We cannot touch poll list, it
1368 * can be even on different cpu. So just clear netif_running().
1370 * dev->stop() will invoke napi_disable() on all of it's
1371 * napi_struct instances on this device.
1373 smp_mb__after_atomic(); /* Commit netif_running(). */
1376 dev_deactivate_many(head);
1378 list_for_each_entry(dev, head, close_list) {
1379 const struct net_device_ops *ops = dev->netdev_ops;
1382 * Call the device specific close. This cannot fail.
1383 * Only if device is UP
1385 * We allow it to be called even after a DETACH hot-plug
1391 dev->flags &= ~IFF_UP;
1392 netpoll_poll_enable(dev);
1398 static int __dev_close(struct net_device *dev)
1403 list_add(&dev->close_list, &single);
1404 retval = __dev_close_many(&single);
1410 int dev_close_many(struct list_head *head, bool unlink)
1412 struct net_device *dev, *tmp;
1414 /* Remove the devices that don't need to be closed */
1415 list_for_each_entry_safe(dev, tmp, head, close_list)
1416 if (!(dev->flags & IFF_UP))
1417 list_del_init(&dev->close_list);
1419 __dev_close_many(head);
1421 list_for_each_entry_safe(dev, tmp, head, close_list) {
1422 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1423 call_netdevice_notifiers(NETDEV_DOWN, dev);
1425 list_del_init(&dev->close_list);
1430 EXPORT_SYMBOL(dev_close_many);
1433 * dev_close - shutdown an interface.
1434 * @dev: device to shutdown
1436 * This function moves an active device into down state. A
1437 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1438 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1441 int dev_close(struct net_device *dev)
1443 if (dev->flags & IFF_UP) {
1446 list_add(&dev->close_list, &single);
1447 dev_close_many(&single, true);
1452 EXPORT_SYMBOL(dev_close);
1456 * dev_disable_lro - disable Large Receive Offload on a device
1459 * Disable Large Receive Offload (LRO) on a net device. Must be
1460 * called under RTNL. This is needed if received packets may be
1461 * forwarded to another interface.
1463 void dev_disable_lro(struct net_device *dev)
1465 struct net_device *lower_dev;
1466 struct list_head *iter;
1468 dev->wanted_features &= ~NETIF_F_LRO;
1469 netdev_update_features(dev);
1471 if (unlikely(dev->features & NETIF_F_LRO))
1472 netdev_WARN(dev, "failed to disable LRO!\n");
1474 netdev_for_each_lower_dev(dev, lower_dev, iter)
1475 dev_disable_lro(lower_dev);
1477 EXPORT_SYMBOL(dev_disable_lro);
1479 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1480 struct net_device *dev)
1482 struct netdev_notifier_info info;
1484 netdev_notifier_info_init(&info, dev);
1485 return nb->notifier_call(nb, val, &info);
1488 static int dev_boot_phase = 1;
1491 * register_netdevice_notifier - register a network notifier block
1494 * Register a notifier to be called when network device events occur.
1495 * The notifier passed is linked into the kernel structures and must
1496 * not be reused until it has been unregistered. A negative errno code
1497 * is returned on a failure.
1499 * When registered all registration and up events are replayed
1500 * to the new notifier to allow device to have a race free
1501 * view of the network device list.
1504 int register_netdevice_notifier(struct notifier_block *nb)
1506 struct net_device *dev;
1507 struct net_device *last;
1512 err = raw_notifier_chain_register(&netdev_chain, nb);
1518 for_each_netdev(net, dev) {
1519 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1520 err = notifier_to_errno(err);
1524 if (!(dev->flags & IFF_UP))
1527 call_netdevice_notifier(nb, NETDEV_UP, dev);
1538 for_each_netdev(net, dev) {
1542 if (dev->flags & IFF_UP) {
1543 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1545 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1547 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1552 raw_notifier_chain_unregister(&netdev_chain, nb);
1555 EXPORT_SYMBOL(register_netdevice_notifier);
1558 * unregister_netdevice_notifier - unregister a network notifier block
1561 * Unregister a notifier previously registered by
1562 * register_netdevice_notifier(). The notifier is unlinked into the
1563 * kernel structures and may then be reused. A negative errno code
1564 * is returned on a failure.
1566 * After unregistering unregister and down device events are synthesized
1567 * for all devices on the device list to the removed notifier to remove
1568 * the need for special case cleanup code.
1571 int unregister_netdevice_notifier(struct notifier_block *nb)
1573 struct net_device *dev;
1578 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1583 for_each_netdev(net, dev) {
1584 if (dev->flags & IFF_UP) {
1585 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1587 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1589 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1596 EXPORT_SYMBOL(unregister_netdevice_notifier);
1599 * call_netdevice_notifiers_info - call all network notifier blocks
1600 * @val: value passed unmodified to notifier function
1601 * @dev: net_device pointer passed unmodified to notifier function
1602 * @info: notifier information data
1604 * Call all network notifier blocks. Parameters and return value
1605 * are as for raw_notifier_call_chain().
1608 static int call_netdevice_notifiers_info(unsigned long val,
1609 struct net_device *dev,
1610 struct netdev_notifier_info *info)
1613 netdev_notifier_info_init(info, dev);
1614 return raw_notifier_call_chain(&netdev_chain, val, info);
1618 * call_netdevice_notifiers - call all network notifier blocks
1619 * @val: value passed unmodified to notifier function
1620 * @dev: net_device pointer passed unmodified to notifier function
1622 * Call all network notifier blocks. Parameters and return value
1623 * are as for raw_notifier_call_chain().
1626 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1628 struct netdev_notifier_info info;
1630 return call_netdevice_notifiers_info(val, dev, &info);
1632 EXPORT_SYMBOL(call_netdevice_notifiers);
1634 #ifdef CONFIG_NET_INGRESS
1635 static struct static_key ingress_needed __read_mostly;
1637 void net_inc_ingress_queue(void)
1639 static_key_slow_inc(&ingress_needed);
1641 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1643 void net_dec_ingress_queue(void)
1645 static_key_slow_dec(&ingress_needed);
1647 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1650 static struct static_key netstamp_needed __read_mostly;
1651 #ifdef HAVE_JUMP_LABEL
1652 /* We are not allowed to call static_key_slow_dec() from irq context
1653 * If net_disable_timestamp() is called from irq context, defer the
1654 * static_key_slow_dec() calls.
1656 static atomic_t netstamp_needed_deferred;
1659 void net_enable_timestamp(void)
1661 #ifdef HAVE_JUMP_LABEL
1662 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1666 static_key_slow_dec(&netstamp_needed);
1670 static_key_slow_inc(&netstamp_needed);
1672 EXPORT_SYMBOL(net_enable_timestamp);
1674 void net_disable_timestamp(void)
1676 #ifdef HAVE_JUMP_LABEL
1677 if (in_interrupt()) {
1678 atomic_inc(&netstamp_needed_deferred);
1682 static_key_slow_dec(&netstamp_needed);
1684 EXPORT_SYMBOL(net_disable_timestamp);
1686 static inline void net_timestamp_set(struct sk_buff *skb)
1688 skb->tstamp.tv64 = 0;
1689 if (static_key_false(&netstamp_needed))
1690 __net_timestamp(skb);
1693 #define net_timestamp_check(COND, SKB) \
1694 if (static_key_false(&netstamp_needed)) { \
1695 if ((COND) && !(SKB)->tstamp.tv64) \
1696 __net_timestamp(SKB); \
1699 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1703 if (!(dev->flags & IFF_UP))
1706 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1707 if (skb->len <= len)
1710 /* if TSO is enabled, we don't care about the length as the packet
1711 * could be forwarded without being segmented before
1713 if (skb_is_gso(skb))
1718 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1720 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1722 if (skb_orphan_frags(skb, GFP_ATOMIC) ||
1723 unlikely(!is_skb_forwardable(dev, skb))) {
1724 atomic_long_inc(&dev->rx_dropped);
1729 skb_scrub_packet(skb, true);
1731 skb->protocol = eth_type_trans(skb, dev);
1732 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1736 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1739 * dev_forward_skb - loopback an skb to another netif
1741 * @dev: destination network device
1742 * @skb: buffer to forward
1745 * NET_RX_SUCCESS (no congestion)
1746 * NET_RX_DROP (packet was dropped, but freed)
1748 * dev_forward_skb can be used for injecting an skb from the
1749 * start_xmit function of one device into the receive queue
1750 * of another device.
1752 * The receiving device may be in another namespace, so
1753 * we have to clear all information in the skb that could
1754 * impact namespace isolation.
1756 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1758 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1760 EXPORT_SYMBOL_GPL(dev_forward_skb);
1762 static inline int deliver_skb(struct sk_buff *skb,
1763 struct packet_type *pt_prev,
1764 struct net_device *orig_dev)
1766 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1768 atomic_inc(&skb->users);
1769 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1772 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1773 struct packet_type **pt,
1774 struct net_device *orig_dev,
1776 struct list_head *ptype_list)
1778 struct packet_type *ptype, *pt_prev = *pt;
1780 list_for_each_entry_rcu(ptype, ptype_list, list) {
1781 if (ptype->type != type)
1784 deliver_skb(skb, pt_prev, orig_dev);
1790 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1792 if (!ptype->af_packet_priv || !skb->sk)
1795 if (ptype->id_match)
1796 return ptype->id_match(ptype, skb->sk);
1797 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1804 * Support routine. Sends outgoing frames to any network
1805 * taps currently in use.
1808 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1810 struct packet_type *ptype;
1811 struct sk_buff *skb2 = NULL;
1812 struct packet_type *pt_prev = NULL;
1813 struct list_head *ptype_list = &ptype_all;
1817 list_for_each_entry_rcu(ptype, ptype_list, list) {
1818 /* Never send packets back to the socket
1819 * they originated from - MvS (miquels@drinkel.ow.org)
1821 if (skb_loop_sk(ptype, skb))
1825 deliver_skb(skb2, pt_prev, skb->dev);
1830 /* need to clone skb, done only once */
1831 skb2 = skb_clone(skb, GFP_ATOMIC);
1835 net_timestamp_set(skb2);
1837 /* skb->nh should be correctly
1838 * set by sender, so that the second statement is
1839 * just protection against buggy protocols.
1841 skb_reset_mac_header(skb2);
1843 if (skb_network_header(skb2) < skb2->data ||
1844 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1845 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1846 ntohs(skb2->protocol),
1848 skb_reset_network_header(skb2);
1851 skb2->transport_header = skb2->network_header;
1852 skb2->pkt_type = PACKET_OUTGOING;
1856 if (ptype_list == &ptype_all) {
1857 ptype_list = &dev->ptype_all;
1862 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1867 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1868 * @dev: Network device
1869 * @txq: number of queues available
1871 * If real_num_tx_queues is changed the tc mappings may no longer be
1872 * valid. To resolve this verify the tc mapping remains valid and if
1873 * not NULL the mapping. With no priorities mapping to this
1874 * offset/count pair it will no longer be used. In the worst case TC0
1875 * is invalid nothing can be done so disable priority mappings. If is
1876 * expected that drivers will fix this mapping if they can before
1877 * calling netif_set_real_num_tx_queues.
1879 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1882 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1884 /* If TC0 is invalidated disable TC mapping */
1885 if (tc->offset + tc->count > txq) {
1886 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1891 /* Invalidated prio to tc mappings set to TC0 */
1892 for (i = 1; i < TC_BITMASK + 1; i++) {
1893 int q = netdev_get_prio_tc_map(dev, i);
1895 tc = &dev->tc_to_txq[q];
1896 if (tc->offset + tc->count > txq) {
1897 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1899 netdev_set_prio_tc_map(dev, i, 0);
1905 static DEFINE_MUTEX(xps_map_mutex);
1906 #define xmap_dereference(P) \
1907 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1909 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1912 struct xps_map *map = NULL;
1916 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1918 for (pos = 0; map && pos < map->len; pos++) {
1919 if (map->queues[pos] == index) {
1921 map->queues[pos] = map->queues[--map->len];
1923 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1924 kfree_rcu(map, rcu);
1934 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1936 struct xps_dev_maps *dev_maps;
1938 bool active = false;
1940 mutex_lock(&xps_map_mutex);
1941 dev_maps = xmap_dereference(dev->xps_maps);
1946 for_each_possible_cpu(cpu) {
1947 for (i = index; i < dev->num_tx_queues; i++) {
1948 if (!remove_xps_queue(dev_maps, cpu, i))
1951 if (i == dev->num_tx_queues)
1956 RCU_INIT_POINTER(dev->xps_maps, NULL);
1957 kfree_rcu(dev_maps, rcu);
1960 for (i = index; i < dev->num_tx_queues; i++)
1961 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1965 mutex_unlock(&xps_map_mutex);
1968 static struct xps_map *expand_xps_map(struct xps_map *map,
1971 struct xps_map *new_map;
1972 int alloc_len = XPS_MIN_MAP_ALLOC;
1975 for (pos = 0; map && pos < map->len; pos++) {
1976 if (map->queues[pos] != index)
1981 /* Need to add queue to this CPU's existing map */
1983 if (pos < map->alloc_len)
1986 alloc_len = map->alloc_len * 2;
1989 /* Need to allocate new map to store queue on this CPU's map */
1990 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1995 for (i = 0; i < pos; i++)
1996 new_map->queues[i] = map->queues[i];
1997 new_map->alloc_len = alloc_len;
2003 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2006 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2007 struct xps_map *map, *new_map;
2008 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2009 int cpu, numa_node_id = -2;
2010 bool active = false;
2012 mutex_lock(&xps_map_mutex);
2014 dev_maps = xmap_dereference(dev->xps_maps);
2016 /* allocate memory for queue storage */
2017 for_each_online_cpu(cpu) {
2018 if (!cpumask_test_cpu(cpu, mask))
2022 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2023 if (!new_dev_maps) {
2024 mutex_unlock(&xps_map_mutex);
2028 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2031 map = expand_xps_map(map, cpu, index);
2035 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2039 goto out_no_new_maps;
2041 for_each_possible_cpu(cpu) {
2042 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2043 /* add queue to CPU maps */
2046 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2047 while ((pos < map->len) && (map->queues[pos] != index))
2050 if (pos == map->len)
2051 map->queues[map->len++] = index;
2053 if (numa_node_id == -2)
2054 numa_node_id = cpu_to_node(cpu);
2055 else if (numa_node_id != cpu_to_node(cpu))
2058 } else if (dev_maps) {
2059 /* fill in the new device map from the old device map */
2060 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2061 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2066 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2068 /* Cleanup old maps */
2070 for_each_possible_cpu(cpu) {
2071 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2072 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2073 if (map && map != new_map)
2074 kfree_rcu(map, rcu);
2077 kfree_rcu(dev_maps, rcu);
2080 dev_maps = new_dev_maps;
2084 /* update Tx queue numa node */
2085 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2086 (numa_node_id >= 0) ? numa_node_id :
2092 /* removes queue from unused CPUs */
2093 for_each_possible_cpu(cpu) {
2094 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2097 if (remove_xps_queue(dev_maps, cpu, index))
2101 /* free map if not active */
2103 RCU_INIT_POINTER(dev->xps_maps, NULL);
2104 kfree_rcu(dev_maps, rcu);
2108 mutex_unlock(&xps_map_mutex);
2112 /* remove any maps that we added */
2113 for_each_possible_cpu(cpu) {
2114 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2115 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2117 if (new_map && new_map != map)
2121 mutex_unlock(&xps_map_mutex);
2123 kfree(new_dev_maps);
2126 EXPORT_SYMBOL(netif_set_xps_queue);
2130 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2131 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2133 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2137 if (txq < 1 || txq > dev->num_tx_queues)
2140 if (dev->reg_state == NETREG_REGISTERED ||
2141 dev->reg_state == NETREG_UNREGISTERING) {
2144 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2150 netif_setup_tc(dev, txq);
2152 if (txq < dev->real_num_tx_queues) {
2153 qdisc_reset_all_tx_gt(dev, txq);
2155 netif_reset_xps_queues_gt(dev, txq);
2160 dev->real_num_tx_queues = txq;
2163 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2167 * netif_set_real_num_rx_queues - set actual number of RX queues used
2168 * @dev: Network device
2169 * @rxq: Actual number of RX queues
2171 * This must be called either with the rtnl_lock held or before
2172 * registration of the net device. Returns 0 on success, or a
2173 * negative error code. If called before registration, it always
2176 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2180 if (rxq < 1 || rxq > dev->num_rx_queues)
2183 if (dev->reg_state == NETREG_REGISTERED) {
2186 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2192 dev->real_num_rx_queues = rxq;
2195 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2199 * netif_get_num_default_rss_queues - default number of RSS queues
2201 * This routine should set an upper limit on the number of RSS queues
2202 * used by default by multiqueue devices.
2204 int netif_get_num_default_rss_queues(void)
2206 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2208 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2210 static inline void __netif_reschedule(struct Qdisc *q)
2212 struct softnet_data *sd;
2213 unsigned long flags;
2215 local_irq_save(flags);
2216 sd = this_cpu_ptr(&softnet_data);
2217 q->next_sched = NULL;
2218 *sd->output_queue_tailp = q;
2219 sd->output_queue_tailp = &q->next_sched;
2220 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2221 local_irq_restore(flags);
2224 void __netif_schedule(struct Qdisc *q)
2226 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2227 __netif_reschedule(q);
2229 EXPORT_SYMBOL(__netif_schedule);
2231 struct dev_kfree_skb_cb {
2232 enum skb_free_reason reason;
2235 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2237 return (struct dev_kfree_skb_cb *)skb->cb;
2240 void netif_schedule_queue(struct netdev_queue *txq)
2243 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2244 struct Qdisc *q = rcu_dereference(txq->qdisc);
2246 __netif_schedule(q);
2250 EXPORT_SYMBOL(netif_schedule_queue);
2253 * netif_wake_subqueue - allow sending packets on subqueue
2254 * @dev: network device
2255 * @queue_index: sub queue index
2257 * Resume individual transmit queue of a device with multiple transmit queues.
2259 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2261 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2263 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2267 q = rcu_dereference(txq->qdisc);
2268 __netif_schedule(q);
2272 EXPORT_SYMBOL(netif_wake_subqueue);
2274 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2276 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2280 q = rcu_dereference(dev_queue->qdisc);
2281 __netif_schedule(q);
2285 EXPORT_SYMBOL(netif_tx_wake_queue);
2287 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2289 unsigned long flags;
2291 if (likely(atomic_read(&skb->users) == 1)) {
2293 atomic_set(&skb->users, 0);
2294 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2297 get_kfree_skb_cb(skb)->reason = reason;
2298 local_irq_save(flags);
2299 skb->next = __this_cpu_read(softnet_data.completion_queue);
2300 __this_cpu_write(softnet_data.completion_queue, skb);
2301 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2302 local_irq_restore(flags);
2304 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2306 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2308 if (in_irq() || irqs_disabled())
2309 __dev_kfree_skb_irq(skb, reason);
2313 EXPORT_SYMBOL(__dev_kfree_skb_any);
2317 * netif_device_detach - mark device as removed
2318 * @dev: network device
2320 * Mark device as removed from system and therefore no longer available.
2322 void netif_device_detach(struct net_device *dev)
2324 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2325 netif_running(dev)) {
2326 netif_tx_stop_all_queues(dev);
2329 EXPORT_SYMBOL(netif_device_detach);
2332 * netif_device_attach - mark device as attached
2333 * @dev: network device
2335 * Mark device as attached from system and restart if needed.
2337 void netif_device_attach(struct net_device *dev)
2339 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2340 netif_running(dev)) {
2341 netif_tx_wake_all_queues(dev);
2342 __netdev_watchdog_up(dev);
2345 EXPORT_SYMBOL(netif_device_attach);
2348 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2349 * to be used as a distribution range.
2351 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2352 unsigned int num_tx_queues)
2356 u16 qcount = num_tx_queues;
2358 if (skb_rx_queue_recorded(skb)) {
2359 hash = skb_get_rx_queue(skb);
2360 while (unlikely(hash >= num_tx_queues))
2361 hash -= num_tx_queues;
2366 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2367 qoffset = dev->tc_to_txq[tc].offset;
2368 qcount = dev->tc_to_txq[tc].count;
2371 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2373 EXPORT_SYMBOL(__skb_tx_hash);
2375 static void skb_warn_bad_offload(const struct sk_buff *skb)
2377 static const netdev_features_t null_features = 0;
2378 struct net_device *dev = skb->dev;
2379 const char *driver = "";
2381 if (!net_ratelimit())
2384 if (dev && dev->dev.parent)
2385 driver = dev_driver_string(dev->dev.parent);
2387 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2388 "gso_type=%d ip_summed=%d\n",
2389 driver, dev ? &dev->features : &null_features,
2390 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2391 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2392 skb_shinfo(skb)->gso_type, skb->ip_summed);
2396 * Invalidate hardware checksum when packet is to be mangled, and
2397 * complete checksum manually on outgoing path.
2399 int skb_checksum_help(struct sk_buff *skb)
2402 int ret = 0, offset;
2404 if (skb->ip_summed == CHECKSUM_COMPLETE)
2405 goto out_set_summed;
2407 if (unlikely(skb_shinfo(skb)->gso_size)) {
2408 skb_warn_bad_offload(skb);
2412 /* Before computing a checksum, we should make sure no frag could
2413 * be modified by an external entity : checksum could be wrong.
2415 if (skb_has_shared_frag(skb)) {
2416 ret = __skb_linearize(skb);
2421 offset = skb_checksum_start_offset(skb);
2422 BUG_ON(offset >= skb_headlen(skb));
2423 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2425 offset += skb->csum_offset;
2426 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2428 if (skb_cloned(skb) &&
2429 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2430 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2435 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2437 skb->ip_summed = CHECKSUM_NONE;
2441 EXPORT_SYMBOL(skb_checksum_help);
2443 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2445 __be16 type = skb->protocol;
2447 /* Tunnel gso handlers can set protocol to ethernet. */
2448 if (type == htons(ETH_P_TEB)) {
2451 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2454 eth = (struct ethhdr *)skb_mac_header(skb);
2455 type = eth->h_proto;
2458 return __vlan_get_protocol(skb, type, depth);
2462 * skb_mac_gso_segment - mac layer segmentation handler.
2463 * @skb: buffer to segment
2464 * @features: features for the output path (see dev->features)
2466 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2467 netdev_features_t features)
2469 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2470 struct packet_offload *ptype;
2471 int vlan_depth = skb->mac_len;
2472 __be16 type = skb_network_protocol(skb, &vlan_depth);
2474 if (unlikely(!type))
2475 return ERR_PTR(-EINVAL);
2477 __skb_pull(skb, vlan_depth);
2480 list_for_each_entry_rcu(ptype, &offload_base, list) {
2481 if (ptype->type == type && ptype->callbacks.gso_segment) {
2482 segs = ptype->callbacks.gso_segment(skb, features);
2488 __skb_push(skb, skb->data - skb_mac_header(skb));
2492 EXPORT_SYMBOL(skb_mac_gso_segment);
2495 /* openvswitch calls this on rx path, so we need a different check.
2497 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2500 return skb->ip_summed != CHECKSUM_PARTIAL;
2502 return skb->ip_summed == CHECKSUM_NONE;
2506 * __skb_gso_segment - Perform segmentation on skb.
2507 * @skb: buffer to segment
2508 * @features: features for the output path (see dev->features)
2509 * @tx_path: whether it is called in TX path
2511 * This function segments the given skb and returns a list of segments.
2513 * It may return NULL if the skb requires no segmentation. This is
2514 * only possible when GSO is used for verifying header integrity.
2516 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2517 netdev_features_t features, bool tx_path)
2519 if (unlikely(skb_needs_check(skb, tx_path))) {
2522 skb_warn_bad_offload(skb);
2524 err = skb_cow_head(skb, 0);
2526 return ERR_PTR(err);
2529 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2530 SKB_GSO_CB(skb)->encap_level = 0;
2532 skb_reset_mac_header(skb);
2533 skb_reset_mac_len(skb);
2535 return skb_mac_gso_segment(skb, features);
2537 EXPORT_SYMBOL(__skb_gso_segment);
2539 /* Take action when hardware reception checksum errors are detected. */
2541 void netdev_rx_csum_fault(struct net_device *dev)
2543 if (net_ratelimit()) {
2544 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2548 EXPORT_SYMBOL(netdev_rx_csum_fault);
2551 /* Actually, we should eliminate this check as soon as we know, that:
2552 * 1. IOMMU is present and allows to map all the memory.
2553 * 2. No high memory really exists on this machine.
2556 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2558 #ifdef CONFIG_HIGHMEM
2560 if (!(dev->features & NETIF_F_HIGHDMA)) {
2561 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2562 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2563 if (PageHighMem(skb_frag_page(frag)))
2568 if (PCI_DMA_BUS_IS_PHYS) {
2569 struct device *pdev = dev->dev.parent;
2573 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2574 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2575 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2576 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2584 /* If MPLS offload request, verify we are testing hardware MPLS features
2585 * instead of standard features for the netdev.
2587 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2588 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2589 netdev_features_t features,
2592 if (eth_p_mpls(type))
2593 features &= skb->dev->mpls_features;
2598 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2599 netdev_features_t features,
2606 static netdev_features_t harmonize_features(struct sk_buff *skb,
2607 netdev_features_t features)
2612 type = skb_network_protocol(skb, &tmp);
2613 features = net_mpls_features(skb, features, type);
2615 if (skb->ip_summed != CHECKSUM_NONE &&
2616 !can_checksum_protocol(features, type)) {
2617 features &= ~NETIF_F_ALL_CSUM;
2618 } else if (illegal_highdma(skb->dev, skb)) {
2619 features &= ~NETIF_F_SG;
2625 netdev_features_t passthru_features_check(struct sk_buff *skb,
2626 struct net_device *dev,
2627 netdev_features_t features)
2631 EXPORT_SYMBOL(passthru_features_check);
2633 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2634 struct net_device *dev,
2635 netdev_features_t features)
2637 return vlan_features_check(skb, features);
2640 netdev_features_t netif_skb_features(struct sk_buff *skb)
2642 struct net_device *dev = skb->dev;
2643 netdev_features_t features = dev->features;
2644 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2646 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2647 features &= ~NETIF_F_GSO_MASK;
2649 /* If encapsulation offload request, verify we are testing
2650 * hardware encapsulation features instead of standard
2651 * features for the netdev
2653 if (skb->encapsulation)
2654 features &= dev->hw_enc_features;
2656 if (skb_vlan_tagged(skb))
2657 features = netdev_intersect_features(features,
2658 dev->vlan_features |
2659 NETIF_F_HW_VLAN_CTAG_TX |
2660 NETIF_F_HW_VLAN_STAG_TX);
2662 if (dev->netdev_ops->ndo_features_check)
2663 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2666 features &= dflt_features_check(skb, dev, features);
2668 return harmonize_features(skb, features);
2670 EXPORT_SYMBOL(netif_skb_features);
2672 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2673 struct netdev_queue *txq, bool more)
2678 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2679 dev_queue_xmit_nit(skb, dev);
2682 trace_net_dev_start_xmit(skb, dev);
2683 rc = netdev_start_xmit(skb, dev, txq, more);
2684 trace_net_dev_xmit(skb, rc, dev, len);
2689 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2690 struct netdev_queue *txq, int *ret)
2692 struct sk_buff *skb = first;
2693 int rc = NETDEV_TX_OK;
2696 struct sk_buff *next = skb->next;
2699 rc = xmit_one(skb, dev, txq, next != NULL);
2700 if (unlikely(!dev_xmit_complete(rc))) {
2706 if (netif_xmit_stopped(txq) && skb) {
2707 rc = NETDEV_TX_BUSY;
2717 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2718 netdev_features_t features)
2720 if (skb_vlan_tag_present(skb) &&
2721 !vlan_hw_offload_capable(features, skb->vlan_proto))
2722 skb = __vlan_hwaccel_push_inside(skb);
2726 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2728 netdev_features_t features;
2733 features = netif_skb_features(skb);
2734 skb = validate_xmit_vlan(skb, features);
2738 if (netif_needs_gso(skb, features)) {
2739 struct sk_buff *segs;
2741 segs = skb_gso_segment(skb, features);
2749 if (skb_needs_linearize(skb, features) &&
2750 __skb_linearize(skb))
2753 /* If packet is not checksummed and device does not
2754 * support checksumming for this protocol, complete
2755 * checksumming here.
2757 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2758 if (skb->encapsulation)
2759 skb_set_inner_transport_header(skb,
2760 skb_checksum_start_offset(skb));
2762 skb_set_transport_header(skb,
2763 skb_checksum_start_offset(skb));
2764 if (!(features & NETIF_F_ALL_CSUM) &&
2765 skb_checksum_help(skb))
2778 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2780 struct sk_buff *next, *head = NULL, *tail;
2782 for (; skb != NULL; skb = next) {
2786 /* in case skb wont be segmented, point to itself */
2789 skb = validate_xmit_skb(skb, dev);
2797 /* If skb was segmented, skb->prev points to
2798 * the last segment. If not, it still contains skb.
2805 static void qdisc_pkt_len_init(struct sk_buff *skb)
2807 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2809 qdisc_skb_cb(skb)->pkt_len = skb->len;
2811 /* To get more precise estimation of bytes sent on wire,
2812 * we add to pkt_len the headers size of all segments
2814 if (shinfo->gso_size) {
2815 unsigned int hdr_len;
2816 u16 gso_segs = shinfo->gso_segs;
2818 /* mac layer + network layer */
2819 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2821 /* + transport layer */
2822 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2823 hdr_len += tcp_hdrlen(skb);
2825 hdr_len += sizeof(struct udphdr);
2827 if (shinfo->gso_type & SKB_GSO_DODGY)
2828 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2831 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2835 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2836 struct net_device *dev,
2837 struct netdev_queue *txq)
2839 spinlock_t *root_lock = qdisc_lock(q);
2843 qdisc_pkt_len_init(skb);
2844 qdisc_calculate_pkt_len(skb, q);
2846 * Heuristic to force contended enqueues to serialize on a
2847 * separate lock before trying to get qdisc main lock.
2848 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2849 * often and dequeue packets faster.
2851 contended = qdisc_is_running(q);
2852 if (unlikely(contended))
2853 spin_lock(&q->busylock);
2855 spin_lock(root_lock);
2856 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2859 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2860 qdisc_run_begin(q)) {
2862 * This is a work-conserving queue; there are no old skbs
2863 * waiting to be sent out; and the qdisc is not running -
2864 * xmit the skb directly.
2867 qdisc_bstats_update(q, skb);
2869 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2870 if (unlikely(contended)) {
2871 spin_unlock(&q->busylock);
2878 rc = NET_XMIT_SUCCESS;
2880 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2881 if (qdisc_run_begin(q)) {
2882 if (unlikely(contended)) {
2883 spin_unlock(&q->busylock);
2889 spin_unlock(root_lock);
2890 if (unlikely(contended))
2891 spin_unlock(&q->busylock);
2895 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2896 static void skb_update_prio(struct sk_buff *skb)
2898 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2900 if (!skb->priority && skb->sk && map) {
2901 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2903 if (prioidx < map->priomap_len)
2904 skb->priority = map->priomap[prioidx];
2908 #define skb_update_prio(skb)
2911 DEFINE_PER_CPU(int, xmit_recursion);
2912 EXPORT_SYMBOL(xmit_recursion);
2914 #define RECURSION_LIMIT 10
2917 * dev_loopback_xmit - loop back @skb
2918 * @skb: buffer to transmit
2920 int dev_loopback_xmit(struct sock *sk, struct sk_buff *skb)
2922 skb_reset_mac_header(skb);
2923 __skb_pull(skb, skb_network_offset(skb));
2924 skb->pkt_type = PACKET_LOOPBACK;
2925 skb->ip_summed = CHECKSUM_UNNECESSARY;
2926 WARN_ON(!skb_dst(skb));
2931 EXPORT_SYMBOL(dev_loopback_xmit);
2933 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2936 struct xps_dev_maps *dev_maps;
2937 struct xps_map *map;
2938 int queue_index = -1;
2941 dev_maps = rcu_dereference(dev->xps_maps);
2943 map = rcu_dereference(
2944 dev_maps->cpu_map[skb->sender_cpu - 1]);
2947 queue_index = map->queues[0];
2949 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
2951 if (unlikely(queue_index >= dev->real_num_tx_queues))
2963 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
2965 struct sock *sk = skb->sk;
2966 int queue_index = sk_tx_queue_get(sk);
2968 if (queue_index < 0 || skb->ooo_okay ||
2969 queue_index >= dev->real_num_tx_queues) {
2970 int new_index = get_xps_queue(dev, skb);
2972 new_index = skb_tx_hash(dev, skb);
2974 if (queue_index != new_index && sk &&
2975 rcu_access_pointer(sk->sk_dst_cache))
2976 sk_tx_queue_set(sk, new_index);
2978 queue_index = new_index;
2984 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
2985 struct sk_buff *skb,
2988 int queue_index = 0;
2991 if (skb->sender_cpu == 0)
2992 skb->sender_cpu = raw_smp_processor_id() + 1;
2995 if (dev->real_num_tx_queues != 1) {
2996 const struct net_device_ops *ops = dev->netdev_ops;
2997 if (ops->ndo_select_queue)
2998 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3001 queue_index = __netdev_pick_tx(dev, skb);
3004 queue_index = netdev_cap_txqueue(dev, queue_index);
3007 skb_set_queue_mapping(skb, queue_index);
3008 return netdev_get_tx_queue(dev, queue_index);
3012 * __dev_queue_xmit - transmit a buffer
3013 * @skb: buffer to transmit
3014 * @accel_priv: private data used for L2 forwarding offload
3016 * Queue a buffer for transmission to a network device. The caller must
3017 * have set the device and priority and built the buffer before calling
3018 * this function. The function can be called from an interrupt.
3020 * A negative errno code is returned on a failure. A success does not
3021 * guarantee the frame will be transmitted as it may be dropped due
3022 * to congestion or traffic shaping.
3024 * -----------------------------------------------------------------------------------
3025 * I notice this method can also return errors from the queue disciplines,
3026 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3029 * Regardless of the return value, the skb is consumed, so it is currently
3030 * difficult to retry a send to this method. (You can bump the ref count
3031 * before sending to hold a reference for retry if you are careful.)
3033 * When calling this method, interrupts MUST be enabled. This is because
3034 * the BH enable code must have IRQs enabled so that it will not deadlock.
3037 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3039 struct net_device *dev = skb->dev;
3040 struct netdev_queue *txq;
3044 skb_reset_mac_header(skb);
3046 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3047 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3049 /* Disable soft irqs for various locks below. Also
3050 * stops preemption for RCU.
3054 skb_update_prio(skb);
3056 /* If device/qdisc don't need skb->dst, release it right now while
3057 * its hot in this cpu cache.
3059 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3064 txq = netdev_pick_tx(dev, skb, accel_priv);
3065 q = rcu_dereference_bh(txq->qdisc);
3067 #ifdef CONFIG_NET_CLS_ACT
3068 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3070 trace_net_dev_queue(skb);
3072 rc = __dev_xmit_skb(skb, q, dev, txq);
3076 /* The device has no queue. Common case for software devices:
3077 loopback, all the sorts of tunnels...
3079 Really, it is unlikely that netif_tx_lock protection is necessary
3080 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3082 However, it is possible, that they rely on protection
3085 Check this and shot the lock. It is not prone from deadlocks.
3086 Either shot noqueue qdisc, it is even simpler 8)
3088 if (dev->flags & IFF_UP) {
3089 int cpu = smp_processor_id(); /* ok because BHs are off */
3091 if (txq->xmit_lock_owner != cpu) {
3093 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
3094 goto recursion_alert;
3096 skb = validate_xmit_skb(skb, dev);
3100 HARD_TX_LOCK(dev, txq, cpu);
3102 if (!netif_xmit_stopped(txq)) {
3103 __this_cpu_inc(xmit_recursion);
3104 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3105 __this_cpu_dec(xmit_recursion);
3106 if (dev_xmit_complete(rc)) {
3107 HARD_TX_UNLOCK(dev, txq);
3111 HARD_TX_UNLOCK(dev, txq);
3112 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3115 /* Recursion is detected! It is possible,
3119 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3126 rcu_read_unlock_bh();
3128 atomic_long_inc(&dev->tx_dropped);
3129 kfree_skb_list(skb);
3132 rcu_read_unlock_bh();
3136 int dev_queue_xmit_sk(struct sock *sk, struct sk_buff *skb)
3138 return __dev_queue_xmit(skb, NULL);
3140 EXPORT_SYMBOL(dev_queue_xmit_sk);
3142 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3144 return __dev_queue_xmit(skb, accel_priv);
3146 EXPORT_SYMBOL(dev_queue_xmit_accel);
3149 /*=======================================================================
3151 =======================================================================*/
3153 int netdev_max_backlog __read_mostly = 1000;
3154 EXPORT_SYMBOL(netdev_max_backlog);
3156 int netdev_tstamp_prequeue __read_mostly = 1;
3157 int netdev_budget __read_mostly = 300;
3158 int weight_p __read_mostly = 64; /* old backlog weight */
3160 /* Called with irq disabled */
3161 static inline void ____napi_schedule(struct softnet_data *sd,
3162 struct napi_struct *napi)
3164 list_add_tail(&napi->poll_list, &sd->poll_list);
3165 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3170 /* One global table that all flow-based protocols share. */
3171 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3172 EXPORT_SYMBOL(rps_sock_flow_table);
3173 u32 rps_cpu_mask __read_mostly;
3174 EXPORT_SYMBOL(rps_cpu_mask);
3176 struct static_key rps_needed __read_mostly;
3178 static struct rps_dev_flow *
3179 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3180 struct rps_dev_flow *rflow, u16 next_cpu)
3182 if (next_cpu < nr_cpu_ids) {
3183 #ifdef CONFIG_RFS_ACCEL
3184 struct netdev_rx_queue *rxqueue;
3185 struct rps_dev_flow_table *flow_table;
3186 struct rps_dev_flow *old_rflow;
3191 /* Should we steer this flow to a different hardware queue? */
3192 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3193 !(dev->features & NETIF_F_NTUPLE))
3195 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3196 if (rxq_index == skb_get_rx_queue(skb))
3199 rxqueue = dev->_rx + rxq_index;
3200 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3203 flow_id = skb_get_hash(skb) & flow_table->mask;
3204 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3205 rxq_index, flow_id);
3209 rflow = &flow_table->flows[flow_id];
3211 if (old_rflow->filter == rflow->filter)
3212 old_rflow->filter = RPS_NO_FILTER;
3216 per_cpu(softnet_data, next_cpu).input_queue_head;
3219 rflow->cpu = next_cpu;
3224 * get_rps_cpu is called from netif_receive_skb and returns the target
3225 * CPU from the RPS map of the receiving queue for a given skb.
3226 * rcu_read_lock must be held on entry.
3228 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3229 struct rps_dev_flow **rflowp)
3231 const struct rps_sock_flow_table *sock_flow_table;
3232 struct netdev_rx_queue *rxqueue = dev->_rx;
3233 struct rps_dev_flow_table *flow_table;
3234 struct rps_map *map;
3239 if (skb_rx_queue_recorded(skb)) {
3240 u16 index = skb_get_rx_queue(skb);
3242 if (unlikely(index >= dev->real_num_rx_queues)) {
3243 WARN_ONCE(dev->real_num_rx_queues > 1,
3244 "%s received packet on queue %u, but number "
3245 "of RX queues is %u\n",
3246 dev->name, index, dev->real_num_rx_queues);
3252 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3254 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3255 map = rcu_dereference(rxqueue->rps_map);
3256 if (!flow_table && !map)
3259 skb_reset_network_header(skb);
3260 hash = skb_get_hash(skb);
3264 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3265 if (flow_table && sock_flow_table) {
3266 struct rps_dev_flow *rflow;
3270 /* First check into global flow table if there is a match */
3271 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3272 if ((ident ^ hash) & ~rps_cpu_mask)
3275 next_cpu = ident & rps_cpu_mask;
3277 /* OK, now we know there is a match,
3278 * we can look at the local (per receive queue) flow table
3280 rflow = &flow_table->flows[hash & flow_table->mask];
3284 * If the desired CPU (where last recvmsg was done) is
3285 * different from current CPU (one in the rx-queue flow
3286 * table entry), switch if one of the following holds:
3287 * - Current CPU is unset (>= nr_cpu_ids).
3288 * - Current CPU is offline.
3289 * - The current CPU's queue tail has advanced beyond the
3290 * last packet that was enqueued using this table entry.
3291 * This guarantees that all previous packets for the flow
3292 * have been dequeued, thus preserving in order delivery.
3294 if (unlikely(tcpu != next_cpu) &&
3295 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3296 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3297 rflow->last_qtail)) >= 0)) {
3299 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3302 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3312 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3313 if (cpu_online(tcpu)) {
3323 #ifdef CONFIG_RFS_ACCEL
3326 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3327 * @dev: Device on which the filter was set
3328 * @rxq_index: RX queue index
3329 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3330 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3332 * Drivers that implement ndo_rx_flow_steer() should periodically call
3333 * this function for each installed filter and remove the filters for
3334 * which it returns %true.
3336 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3337 u32 flow_id, u16 filter_id)
3339 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3340 struct rps_dev_flow_table *flow_table;
3341 struct rps_dev_flow *rflow;
3346 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3347 if (flow_table && flow_id <= flow_table->mask) {
3348 rflow = &flow_table->flows[flow_id];
3349 cpu = ACCESS_ONCE(rflow->cpu);
3350 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3351 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3352 rflow->last_qtail) <
3353 (int)(10 * flow_table->mask)))
3359 EXPORT_SYMBOL(rps_may_expire_flow);
3361 #endif /* CONFIG_RFS_ACCEL */
3363 /* Called from hardirq (IPI) context */
3364 static void rps_trigger_softirq(void *data)
3366 struct softnet_data *sd = data;
3368 ____napi_schedule(sd, &sd->backlog);
3372 #endif /* CONFIG_RPS */
3375 * Check if this softnet_data structure is another cpu one
3376 * If yes, queue it to our IPI list and return 1
3379 static int rps_ipi_queued(struct softnet_data *sd)
3382 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3385 sd->rps_ipi_next = mysd->rps_ipi_list;
3386 mysd->rps_ipi_list = sd;
3388 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3391 #endif /* CONFIG_RPS */
3395 #ifdef CONFIG_NET_FLOW_LIMIT
3396 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3399 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3401 #ifdef CONFIG_NET_FLOW_LIMIT
3402 struct sd_flow_limit *fl;
3403 struct softnet_data *sd;
3404 unsigned int old_flow, new_flow;
3406 if (qlen < (netdev_max_backlog >> 1))
3409 sd = this_cpu_ptr(&softnet_data);
3412 fl = rcu_dereference(sd->flow_limit);
3414 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3415 old_flow = fl->history[fl->history_head];
3416 fl->history[fl->history_head] = new_flow;
3419 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3421 if (likely(fl->buckets[old_flow]))
3422 fl->buckets[old_flow]--;
3424 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3436 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3437 * queue (may be a remote CPU queue).
3439 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3440 unsigned int *qtail)
3442 struct softnet_data *sd;
3443 unsigned long flags;
3446 sd = &per_cpu(softnet_data, cpu);
3448 local_irq_save(flags);
3451 qlen = skb_queue_len(&sd->input_pkt_queue);
3452 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3455 __skb_queue_tail(&sd->input_pkt_queue, skb);
3456 input_queue_tail_incr_save(sd, qtail);
3458 local_irq_restore(flags);
3459 return NET_RX_SUCCESS;
3462 /* Schedule NAPI for backlog device
3463 * We can use non atomic operation since we own the queue lock
3465 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3466 if (!rps_ipi_queued(sd))
3467 ____napi_schedule(sd, &sd->backlog);
3475 local_irq_restore(flags);
3477 atomic_long_inc(&skb->dev->rx_dropped);
3482 static int netif_rx_internal(struct sk_buff *skb)
3486 net_timestamp_check(netdev_tstamp_prequeue, skb);
3488 trace_netif_rx(skb);
3490 if (static_key_false(&rps_needed)) {
3491 struct rps_dev_flow voidflow, *rflow = &voidflow;
3497 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3499 cpu = smp_processor_id();
3501 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3509 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3516 * netif_rx - post buffer to the network code
3517 * @skb: buffer to post
3519 * This function receives a packet from a device driver and queues it for
3520 * the upper (protocol) levels to process. It always succeeds. The buffer
3521 * may be dropped during processing for congestion control or by the
3525 * NET_RX_SUCCESS (no congestion)
3526 * NET_RX_DROP (packet was dropped)
3530 int netif_rx(struct sk_buff *skb)
3532 trace_netif_rx_entry(skb);
3534 return netif_rx_internal(skb);
3536 EXPORT_SYMBOL(netif_rx);
3538 int netif_rx_ni(struct sk_buff *skb)
3542 trace_netif_rx_ni_entry(skb);
3545 err = netif_rx_internal(skb);
3546 if (local_softirq_pending())
3552 EXPORT_SYMBOL(netif_rx_ni);
3554 static void net_tx_action(struct softirq_action *h)
3556 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3558 if (sd->completion_queue) {
3559 struct sk_buff *clist;
3561 local_irq_disable();
3562 clist = sd->completion_queue;
3563 sd->completion_queue = NULL;
3567 struct sk_buff *skb = clist;
3568 clist = clist->next;
3570 WARN_ON(atomic_read(&skb->users));
3571 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3572 trace_consume_skb(skb);
3574 trace_kfree_skb(skb, net_tx_action);
3579 if (sd->output_queue) {
3582 local_irq_disable();
3583 head = sd->output_queue;
3584 sd->output_queue = NULL;
3585 sd->output_queue_tailp = &sd->output_queue;
3589 struct Qdisc *q = head;
3590 spinlock_t *root_lock;
3592 head = head->next_sched;
3594 root_lock = qdisc_lock(q);
3595 if (spin_trylock(root_lock)) {
3596 smp_mb__before_atomic();
3597 clear_bit(__QDISC_STATE_SCHED,
3600 spin_unlock(root_lock);
3602 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3604 __netif_reschedule(q);
3606 smp_mb__before_atomic();
3607 clear_bit(__QDISC_STATE_SCHED,
3615 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3616 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3617 /* This hook is defined here for ATM LANE */
3618 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3619 unsigned char *addr) __read_mostly;
3620 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3623 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3624 struct packet_type **pt_prev,
3625 int *ret, struct net_device *orig_dev)
3627 #ifdef CONFIG_NET_CLS_ACT
3628 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3629 struct tcf_result cl_res;
3631 /* If there's at least one ingress present somewhere (so
3632 * we get here via enabled static key), remaining devices
3633 * that are not configured with an ingress qdisc will bail
3639 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3643 qdisc_skb_cb(skb)->pkt_len = skb->len;
3644 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3645 qdisc_bstats_update_cpu(cl->q, skb);
3647 switch (tc_classify(skb, cl, &cl_res)) {
3649 case TC_ACT_RECLASSIFY:
3650 skb->tc_index = TC_H_MIN(cl_res.classid);
3653 qdisc_qstats_drop_cpu(cl->q);
3661 #endif /* CONFIG_NET_CLS_ACT */
3666 * netdev_rx_handler_register - register receive handler
3667 * @dev: device to register a handler for
3668 * @rx_handler: receive handler to register
3669 * @rx_handler_data: data pointer that is used by rx handler
3671 * Register a receive handler for a device. This handler will then be
3672 * called from __netif_receive_skb. A negative errno code is returned
3675 * The caller must hold the rtnl_mutex.
3677 * For a general description of rx_handler, see enum rx_handler_result.
3679 int netdev_rx_handler_register(struct net_device *dev,
3680 rx_handler_func_t *rx_handler,
3681 void *rx_handler_data)
3685 if (dev->rx_handler)
3688 /* Note: rx_handler_data must be set before rx_handler */
3689 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3690 rcu_assign_pointer(dev->rx_handler, rx_handler);
3694 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3697 * netdev_rx_handler_unregister - unregister receive handler
3698 * @dev: device to unregister a handler from
3700 * Unregister a receive handler from a device.
3702 * The caller must hold the rtnl_mutex.
3704 void netdev_rx_handler_unregister(struct net_device *dev)
3708 RCU_INIT_POINTER(dev->rx_handler, NULL);
3709 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3710 * section has a guarantee to see a non NULL rx_handler_data
3714 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3716 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3719 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3720 * the special handling of PFMEMALLOC skbs.
3722 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3724 switch (skb->protocol) {
3725 case htons(ETH_P_ARP):
3726 case htons(ETH_P_IP):
3727 case htons(ETH_P_IPV6):
3728 case htons(ETH_P_8021Q):
3729 case htons(ETH_P_8021AD):
3736 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
3737 int *ret, struct net_device *orig_dev)
3739 #ifdef CONFIG_NETFILTER_INGRESS
3740 if (nf_hook_ingress_active(skb)) {
3742 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3746 return nf_hook_ingress(skb);
3748 #endif /* CONFIG_NETFILTER_INGRESS */
3752 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3754 struct packet_type *ptype, *pt_prev;
3755 rx_handler_func_t *rx_handler;
3756 struct net_device *orig_dev;
3757 bool deliver_exact = false;
3758 int ret = NET_RX_DROP;
3761 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3763 trace_netif_receive_skb(skb);
3765 orig_dev = skb->dev;
3767 skb_reset_network_header(skb);
3768 if (!skb_transport_header_was_set(skb))
3769 skb_reset_transport_header(skb);
3770 skb_reset_mac_len(skb);
3777 skb->skb_iif = skb->dev->ifindex;
3779 __this_cpu_inc(softnet_data.processed);
3781 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3782 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3783 skb = skb_vlan_untag(skb);
3788 #ifdef CONFIG_NET_CLS_ACT
3789 if (skb->tc_verd & TC_NCLS) {
3790 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3798 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3800 ret = deliver_skb(skb, pt_prev, orig_dev);
3804 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3806 ret = deliver_skb(skb, pt_prev, orig_dev);
3811 #ifdef CONFIG_NET_INGRESS
3812 if (static_key_false(&ingress_needed)) {
3813 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3817 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
3821 #ifdef CONFIG_NET_CLS_ACT
3825 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3828 if (skb_vlan_tag_present(skb)) {
3830 ret = deliver_skb(skb, pt_prev, orig_dev);
3833 if (vlan_do_receive(&skb))
3835 else if (unlikely(!skb))
3839 rx_handler = rcu_dereference(skb->dev->rx_handler);
3842 ret = deliver_skb(skb, pt_prev, orig_dev);
3845 switch (rx_handler(&skb)) {
3846 case RX_HANDLER_CONSUMED:
3847 ret = NET_RX_SUCCESS;
3849 case RX_HANDLER_ANOTHER:
3851 case RX_HANDLER_EXACT:
3852 deliver_exact = true;
3853 case RX_HANDLER_PASS:
3860 if (unlikely(skb_vlan_tag_present(skb))) {
3861 if (skb_vlan_tag_get_id(skb))
3862 skb->pkt_type = PACKET_OTHERHOST;
3863 /* Note: we might in the future use prio bits
3864 * and set skb->priority like in vlan_do_receive()
3865 * For the time being, just ignore Priority Code Point
3870 type = skb->protocol;
3872 /* deliver only exact match when indicated */
3873 if (likely(!deliver_exact)) {
3874 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3875 &ptype_base[ntohs(type) &
3879 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3880 &orig_dev->ptype_specific);
3882 if (unlikely(skb->dev != orig_dev)) {
3883 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3884 &skb->dev->ptype_specific);
3888 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3891 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3894 atomic_long_inc(&skb->dev->rx_dropped);
3896 /* Jamal, now you will not able to escape explaining
3897 * me how you were going to use this. :-)
3907 static int __netif_receive_skb(struct sk_buff *skb)
3911 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3912 unsigned long pflags = current->flags;
3915 * PFMEMALLOC skbs are special, they should
3916 * - be delivered to SOCK_MEMALLOC sockets only
3917 * - stay away from userspace
3918 * - have bounded memory usage
3920 * Use PF_MEMALLOC as this saves us from propagating the allocation
3921 * context down to all allocation sites.
3923 current->flags |= PF_MEMALLOC;
3924 ret = __netif_receive_skb_core(skb, true);
3925 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3927 ret = __netif_receive_skb_core(skb, false);
3932 static int netif_receive_skb_internal(struct sk_buff *skb)
3934 net_timestamp_check(netdev_tstamp_prequeue, skb);
3936 if (skb_defer_rx_timestamp(skb))
3937 return NET_RX_SUCCESS;
3940 if (static_key_false(&rps_needed)) {
3941 struct rps_dev_flow voidflow, *rflow = &voidflow;
3946 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3949 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3956 return __netif_receive_skb(skb);
3960 * netif_receive_skb - process receive buffer from network
3961 * @skb: buffer to process
3963 * netif_receive_skb() is the main receive data processing function.
3964 * It always succeeds. The buffer may be dropped during processing
3965 * for congestion control or by the protocol layers.
3967 * This function may only be called from softirq context and interrupts
3968 * should be enabled.
3970 * Return values (usually ignored):
3971 * NET_RX_SUCCESS: no congestion
3972 * NET_RX_DROP: packet was dropped
3974 int netif_receive_skb_sk(struct sock *sk, struct sk_buff *skb)
3976 trace_netif_receive_skb_entry(skb);
3978 return netif_receive_skb_internal(skb);
3980 EXPORT_SYMBOL(netif_receive_skb_sk);
3982 /* Network device is going away, flush any packets still pending
3983 * Called with irqs disabled.
3985 static void flush_backlog(void *arg)
3987 struct net_device *dev = arg;
3988 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3989 struct sk_buff *skb, *tmp;
3992 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3993 if (skb->dev == dev) {
3994 __skb_unlink(skb, &sd->input_pkt_queue);
3996 input_queue_head_incr(sd);
4001 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4002 if (skb->dev == dev) {
4003 __skb_unlink(skb, &sd->process_queue);
4005 input_queue_head_incr(sd);
4010 static int napi_gro_complete(struct sk_buff *skb)
4012 struct packet_offload *ptype;
4013 __be16 type = skb->protocol;
4014 struct list_head *head = &offload_base;
4017 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4019 if (NAPI_GRO_CB(skb)->count == 1) {
4020 skb_shinfo(skb)->gso_size = 0;
4025 list_for_each_entry_rcu(ptype, head, list) {
4026 if (ptype->type != type || !ptype->callbacks.gro_complete)
4029 err = ptype->callbacks.gro_complete(skb, 0);
4035 WARN_ON(&ptype->list == head);
4037 return NET_RX_SUCCESS;
4041 return netif_receive_skb_internal(skb);
4044 /* napi->gro_list contains packets ordered by age.
4045 * youngest packets at the head of it.
4046 * Complete skbs in reverse order to reduce latencies.
4048 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4050 struct sk_buff *skb, *prev = NULL;
4052 /* scan list and build reverse chain */
4053 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4058 for (skb = prev; skb; skb = prev) {
4061 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4065 napi_gro_complete(skb);
4069 napi->gro_list = NULL;
4071 EXPORT_SYMBOL(napi_gro_flush);
4073 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4076 unsigned int maclen = skb->dev->hard_header_len;
4077 u32 hash = skb_get_hash_raw(skb);
4079 for (p = napi->gro_list; p; p = p->next) {
4080 unsigned long diffs;
4082 NAPI_GRO_CB(p)->flush = 0;
4084 if (hash != skb_get_hash_raw(p)) {
4085 NAPI_GRO_CB(p)->same_flow = 0;
4089 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4090 diffs |= p->vlan_tci ^ skb->vlan_tci;
4091 if (maclen == ETH_HLEN)
4092 diffs |= compare_ether_header(skb_mac_header(p),
4093 skb_mac_header(skb));
4095 diffs = memcmp(skb_mac_header(p),
4096 skb_mac_header(skb),
4098 NAPI_GRO_CB(p)->same_flow = !diffs;
4102 static void skb_gro_reset_offset(struct sk_buff *skb)
4104 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4105 const skb_frag_t *frag0 = &pinfo->frags[0];
4107 NAPI_GRO_CB(skb)->data_offset = 0;
4108 NAPI_GRO_CB(skb)->frag0 = NULL;
4109 NAPI_GRO_CB(skb)->frag0_len = 0;
4111 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4113 !PageHighMem(skb_frag_page(frag0))) {
4114 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4115 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4119 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4121 struct skb_shared_info *pinfo = skb_shinfo(skb);
4123 BUG_ON(skb->end - skb->tail < grow);
4125 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4127 skb->data_len -= grow;
4130 pinfo->frags[0].page_offset += grow;
4131 skb_frag_size_sub(&pinfo->frags[0], grow);
4133 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4134 skb_frag_unref(skb, 0);
4135 memmove(pinfo->frags, pinfo->frags + 1,
4136 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4140 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4142 struct sk_buff **pp = NULL;
4143 struct packet_offload *ptype;
4144 __be16 type = skb->protocol;
4145 struct list_head *head = &offload_base;
4147 enum gro_result ret;
4150 if (!(skb->dev->features & NETIF_F_GRO))
4153 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4156 gro_list_prepare(napi, skb);
4159 list_for_each_entry_rcu(ptype, head, list) {
4160 if (ptype->type != type || !ptype->callbacks.gro_receive)
4163 skb_set_network_header(skb, skb_gro_offset(skb));
4164 skb_reset_mac_len(skb);
4165 NAPI_GRO_CB(skb)->same_flow = 0;
4166 NAPI_GRO_CB(skb)->flush = 0;
4167 NAPI_GRO_CB(skb)->free = 0;
4168 NAPI_GRO_CB(skb)->udp_mark = 0;
4169 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4171 /* Setup for GRO checksum validation */
4172 switch (skb->ip_summed) {
4173 case CHECKSUM_COMPLETE:
4174 NAPI_GRO_CB(skb)->csum = skb->csum;
4175 NAPI_GRO_CB(skb)->csum_valid = 1;
4176 NAPI_GRO_CB(skb)->csum_cnt = 0;
4178 case CHECKSUM_UNNECESSARY:
4179 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4180 NAPI_GRO_CB(skb)->csum_valid = 0;
4183 NAPI_GRO_CB(skb)->csum_cnt = 0;
4184 NAPI_GRO_CB(skb)->csum_valid = 0;
4187 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4192 if (&ptype->list == head)
4195 same_flow = NAPI_GRO_CB(skb)->same_flow;
4196 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4199 struct sk_buff *nskb = *pp;
4203 napi_gro_complete(nskb);
4210 if (NAPI_GRO_CB(skb)->flush)
4213 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4214 struct sk_buff *nskb = napi->gro_list;
4216 /* locate the end of the list to select the 'oldest' flow */
4217 while (nskb->next) {
4223 napi_gro_complete(nskb);
4227 NAPI_GRO_CB(skb)->count = 1;
4228 NAPI_GRO_CB(skb)->age = jiffies;
4229 NAPI_GRO_CB(skb)->last = skb;
4230 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4231 skb->next = napi->gro_list;
4232 napi->gro_list = skb;
4236 grow = skb_gro_offset(skb) - skb_headlen(skb);
4238 gro_pull_from_frag0(skb, grow);
4247 struct packet_offload *gro_find_receive_by_type(__be16 type)
4249 struct list_head *offload_head = &offload_base;
4250 struct packet_offload *ptype;
4252 list_for_each_entry_rcu(ptype, offload_head, list) {
4253 if (ptype->type != type || !ptype->callbacks.gro_receive)
4259 EXPORT_SYMBOL(gro_find_receive_by_type);
4261 struct packet_offload *gro_find_complete_by_type(__be16 type)
4263 struct list_head *offload_head = &offload_base;
4264 struct packet_offload *ptype;
4266 list_for_each_entry_rcu(ptype, offload_head, list) {
4267 if (ptype->type != type || !ptype->callbacks.gro_complete)
4273 EXPORT_SYMBOL(gro_find_complete_by_type);
4275 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4279 if (netif_receive_skb_internal(skb))
4287 case GRO_MERGED_FREE:
4288 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4289 kmem_cache_free(skbuff_head_cache, skb);
4302 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4304 trace_napi_gro_receive_entry(skb);
4306 skb_gro_reset_offset(skb);
4308 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4310 EXPORT_SYMBOL(napi_gro_receive);
4312 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4314 if (unlikely(skb->pfmemalloc)) {
4318 __skb_pull(skb, skb_headlen(skb));
4319 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4320 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4322 skb->dev = napi->dev;
4324 skb->encapsulation = 0;
4325 skb_shinfo(skb)->gso_type = 0;
4326 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4331 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4333 struct sk_buff *skb = napi->skb;
4336 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4341 EXPORT_SYMBOL(napi_get_frags);
4343 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4344 struct sk_buff *skb,
4350 __skb_push(skb, ETH_HLEN);
4351 skb->protocol = eth_type_trans(skb, skb->dev);
4352 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4357 case GRO_MERGED_FREE:
4358 napi_reuse_skb(napi, skb);
4368 /* Upper GRO stack assumes network header starts at gro_offset=0
4369 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4370 * We copy ethernet header into skb->data to have a common layout.
4372 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4374 struct sk_buff *skb = napi->skb;
4375 const struct ethhdr *eth;
4376 unsigned int hlen = sizeof(*eth);
4380 skb_reset_mac_header(skb);
4381 skb_gro_reset_offset(skb);
4383 eth = skb_gro_header_fast(skb, 0);
4384 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4385 eth = skb_gro_header_slow(skb, hlen, 0);
4386 if (unlikely(!eth)) {
4387 napi_reuse_skb(napi, skb);
4391 gro_pull_from_frag0(skb, hlen);
4392 NAPI_GRO_CB(skb)->frag0 += hlen;
4393 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4395 __skb_pull(skb, hlen);
4398 * This works because the only protocols we care about don't require
4400 * We'll fix it up properly in napi_frags_finish()
4402 skb->protocol = eth->h_proto;
4407 gro_result_t napi_gro_frags(struct napi_struct *napi)
4409 struct sk_buff *skb = napi_frags_skb(napi);
4414 trace_napi_gro_frags_entry(skb);
4416 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4418 EXPORT_SYMBOL(napi_gro_frags);
4420 /* Compute the checksum from gro_offset and return the folded value
4421 * after adding in any pseudo checksum.
4423 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4428 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4430 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4431 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4433 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4434 !skb->csum_complete_sw)
4435 netdev_rx_csum_fault(skb->dev);
4438 NAPI_GRO_CB(skb)->csum = wsum;
4439 NAPI_GRO_CB(skb)->csum_valid = 1;
4443 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4446 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4447 * Note: called with local irq disabled, but exits with local irq enabled.
4449 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4452 struct softnet_data *remsd = sd->rps_ipi_list;
4455 sd->rps_ipi_list = NULL;
4459 /* Send pending IPI's to kick RPS processing on remote cpus. */
4461 struct softnet_data *next = remsd->rps_ipi_next;
4463 if (cpu_online(remsd->cpu))
4464 smp_call_function_single_async(remsd->cpu,
4473 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4476 return sd->rps_ipi_list != NULL;
4482 static int process_backlog(struct napi_struct *napi, int quota)
4485 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4487 /* Check if we have pending ipi, its better to send them now,
4488 * not waiting net_rx_action() end.
4490 if (sd_has_rps_ipi_waiting(sd)) {
4491 local_irq_disable();
4492 net_rps_action_and_irq_enable(sd);
4495 napi->weight = weight_p;
4496 local_irq_disable();
4498 struct sk_buff *skb;
4500 while ((skb = __skb_dequeue(&sd->process_queue))) {
4502 __netif_receive_skb(skb);
4503 local_irq_disable();
4504 input_queue_head_incr(sd);
4505 if (++work >= quota) {
4512 if (skb_queue_empty(&sd->input_pkt_queue)) {
4514 * Inline a custom version of __napi_complete().
4515 * only current cpu owns and manipulates this napi,
4516 * and NAPI_STATE_SCHED is the only possible flag set
4518 * We can use a plain write instead of clear_bit(),
4519 * and we dont need an smp_mb() memory barrier.
4527 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4528 &sd->process_queue);
4537 * __napi_schedule - schedule for receive
4538 * @n: entry to schedule
4540 * The entry's receive function will be scheduled to run.
4541 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4543 void __napi_schedule(struct napi_struct *n)
4545 unsigned long flags;
4547 local_irq_save(flags);
4548 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4549 local_irq_restore(flags);
4551 EXPORT_SYMBOL(__napi_schedule);
4554 * __napi_schedule_irqoff - schedule for receive
4555 * @n: entry to schedule
4557 * Variant of __napi_schedule() assuming hard irqs are masked
4559 void __napi_schedule_irqoff(struct napi_struct *n)
4561 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4563 EXPORT_SYMBOL(__napi_schedule_irqoff);
4565 void __napi_complete(struct napi_struct *n)
4567 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4569 list_del_init(&n->poll_list);
4570 smp_mb__before_atomic();
4571 clear_bit(NAPI_STATE_SCHED, &n->state);
4573 EXPORT_SYMBOL(__napi_complete);
4575 void napi_complete_done(struct napi_struct *n, int work_done)
4577 unsigned long flags;
4580 * don't let napi dequeue from the cpu poll list
4581 * just in case its running on a different cpu
4583 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4587 unsigned long timeout = 0;
4590 timeout = n->dev->gro_flush_timeout;
4593 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4594 HRTIMER_MODE_REL_PINNED);
4596 napi_gro_flush(n, false);
4598 if (likely(list_empty(&n->poll_list))) {
4599 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4601 /* If n->poll_list is not empty, we need to mask irqs */
4602 local_irq_save(flags);
4604 local_irq_restore(flags);
4607 EXPORT_SYMBOL(napi_complete_done);
4609 /* must be called under rcu_read_lock(), as we dont take a reference */
4610 struct napi_struct *napi_by_id(unsigned int napi_id)
4612 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4613 struct napi_struct *napi;
4615 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4616 if (napi->napi_id == napi_id)
4621 EXPORT_SYMBOL_GPL(napi_by_id);
4623 void napi_hash_add(struct napi_struct *napi)
4625 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4627 spin_lock(&napi_hash_lock);
4629 /* 0 is not a valid id, we also skip an id that is taken
4630 * we expect both events to be extremely rare
4633 while (!napi->napi_id) {
4634 napi->napi_id = ++napi_gen_id;
4635 if (napi_by_id(napi->napi_id))
4639 hlist_add_head_rcu(&napi->napi_hash_node,
4640 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4642 spin_unlock(&napi_hash_lock);
4645 EXPORT_SYMBOL_GPL(napi_hash_add);
4647 /* Warning : caller is responsible to make sure rcu grace period
4648 * is respected before freeing memory containing @napi
4650 void napi_hash_del(struct napi_struct *napi)
4652 spin_lock(&napi_hash_lock);
4654 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4655 hlist_del_rcu(&napi->napi_hash_node);
4657 spin_unlock(&napi_hash_lock);
4659 EXPORT_SYMBOL_GPL(napi_hash_del);
4661 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4663 struct napi_struct *napi;
4665 napi = container_of(timer, struct napi_struct, timer);
4667 napi_schedule(napi);
4669 return HRTIMER_NORESTART;
4672 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4673 int (*poll)(struct napi_struct *, int), int weight)
4675 INIT_LIST_HEAD(&napi->poll_list);
4676 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4677 napi->timer.function = napi_watchdog;
4678 napi->gro_count = 0;
4679 napi->gro_list = NULL;
4682 if (weight > NAPI_POLL_WEIGHT)
4683 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4685 napi->weight = weight;
4686 list_add(&napi->dev_list, &dev->napi_list);
4688 #ifdef CONFIG_NETPOLL
4689 spin_lock_init(&napi->poll_lock);
4690 napi->poll_owner = -1;
4692 set_bit(NAPI_STATE_SCHED, &napi->state);
4694 EXPORT_SYMBOL(netif_napi_add);
4696 void napi_disable(struct napi_struct *n)
4699 set_bit(NAPI_STATE_DISABLE, &n->state);
4701 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4704 hrtimer_cancel(&n->timer);
4706 clear_bit(NAPI_STATE_DISABLE, &n->state);
4708 EXPORT_SYMBOL(napi_disable);
4710 void netif_napi_del(struct napi_struct *napi)
4712 list_del_init(&napi->dev_list);
4713 napi_free_frags(napi);
4715 kfree_skb_list(napi->gro_list);
4716 napi->gro_list = NULL;
4717 napi->gro_count = 0;
4719 EXPORT_SYMBOL(netif_napi_del);
4721 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4726 list_del_init(&n->poll_list);
4728 have = netpoll_poll_lock(n);
4732 /* This NAPI_STATE_SCHED test is for avoiding a race
4733 * with netpoll's poll_napi(). Only the entity which
4734 * obtains the lock and sees NAPI_STATE_SCHED set will
4735 * actually make the ->poll() call. Therefore we avoid
4736 * accidentally calling ->poll() when NAPI is not scheduled.
4739 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4740 work = n->poll(n, weight);
4744 WARN_ON_ONCE(work > weight);
4746 if (likely(work < weight))
4749 /* Drivers must not modify the NAPI state if they
4750 * consume the entire weight. In such cases this code
4751 * still "owns" the NAPI instance and therefore can
4752 * move the instance around on the list at-will.
4754 if (unlikely(napi_disable_pending(n))) {
4760 /* flush too old packets
4761 * If HZ < 1000, flush all packets.
4763 napi_gro_flush(n, HZ >= 1000);
4766 /* Some drivers may have called napi_schedule
4767 * prior to exhausting their budget.
4769 if (unlikely(!list_empty(&n->poll_list))) {
4770 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4771 n->dev ? n->dev->name : "backlog");
4775 list_add_tail(&n->poll_list, repoll);
4778 netpoll_poll_unlock(have);
4783 static void net_rx_action(struct softirq_action *h)
4785 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4786 unsigned long time_limit = jiffies + 2;
4787 int budget = netdev_budget;
4791 local_irq_disable();
4792 list_splice_init(&sd->poll_list, &list);
4796 struct napi_struct *n;
4798 if (list_empty(&list)) {
4799 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4804 n = list_first_entry(&list, struct napi_struct, poll_list);
4805 budget -= napi_poll(n, &repoll);
4807 /* If softirq window is exhausted then punt.
4808 * Allow this to run for 2 jiffies since which will allow
4809 * an average latency of 1.5/HZ.
4811 if (unlikely(budget <= 0 ||
4812 time_after_eq(jiffies, time_limit))) {
4818 local_irq_disable();
4820 list_splice_tail_init(&sd->poll_list, &list);
4821 list_splice_tail(&repoll, &list);
4822 list_splice(&list, &sd->poll_list);
4823 if (!list_empty(&sd->poll_list))
4824 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4826 net_rps_action_and_irq_enable(sd);
4829 struct netdev_adjacent {
4830 struct net_device *dev;
4832 /* upper master flag, there can only be one master device per list */
4835 /* counter for the number of times this device was added to us */
4838 /* private field for the users */
4841 struct list_head list;
4842 struct rcu_head rcu;
4845 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4846 struct net_device *adj_dev,
4847 struct list_head *adj_list)
4849 struct netdev_adjacent *adj;
4851 list_for_each_entry(adj, adj_list, list) {
4852 if (adj->dev == adj_dev)
4859 * netdev_has_upper_dev - Check if device is linked to an upper device
4861 * @upper_dev: upper device to check
4863 * Find out if a device is linked to specified upper device and return true
4864 * in case it is. Note that this checks only immediate upper device,
4865 * not through a complete stack of devices. The caller must hold the RTNL lock.
4867 bool netdev_has_upper_dev(struct net_device *dev,
4868 struct net_device *upper_dev)
4872 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4874 EXPORT_SYMBOL(netdev_has_upper_dev);
4877 * netdev_has_any_upper_dev - Check if device is linked to some device
4880 * Find out if a device is linked to an upper device and return true in case
4881 * it is. The caller must hold the RTNL lock.
4883 static bool netdev_has_any_upper_dev(struct net_device *dev)
4887 return !list_empty(&dev->all_adj_list.upper);
4891 * netdev_master_upper_dev_get - Get master upper device
4894 * Find a master upper device and return pointer to it or NULL in case
4895 * it's not there. The caller must hold the RTNL lock.
4897 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4899 struct netdev_adjacent *upper;
4903 if (list_empty(&dev->adj_list.upper))
4906 upper = list_first_entry(&dev->adj_list.upper,
4907 struct netdev_adjacent, list);
4908 if (likely(upper->master))
4912 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4914 void *netdev_adjacent_get_private(struct list_head *adj_list)
4916 struct netdev_adjacent *adj;
4918 adj = list_entry(adj_list, struct netdev_adjacent, list);
4920 return adj->private;
4922 EXPORT_SYMBOL(netdev_adjacent_get_private);
4925 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4927 * @iter: list_head ** of the current position
4929 * Gets the next device from the dev's upper list, starting from iter
4930 * position. The caller must hold RCU read lock.
4932 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4933 struct list_head **iter)
4935 struct netdev_adjacent *upper;
4937 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4939 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4941 if (&upper->list == &dev->adj_list.upper)
4944 *iter = &upper->list;
4948 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4951 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4953 * @iter: list_head ** of the current position
4955 * Gets the next device from the dev's upper list, starting from iter
4956 * position. The caller must hold RCU read lock.
4958 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4959 struct list_head **iter)
4961 struct netdev_adjacent *upper;
4963 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4965 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4967 if (&upper->list == &dev->all_adj_list.upper)
4970 *iter = &upper->list;
4974 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4977 * netdev_lower_get_next_private - Get the next ->private from the
4978 * lower neighbour list
4980 * @iter: list_head ** of the current position
4982 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4983 * list, starting from iter position. The caller must hold either hold the
4984 * RTNL lock or its own locking that guarantees that the neighbour lower
4985 * list will remain unchainged.
4987 void *netdev_lower_get_next_private(struct net_device *dev,
4988 struct list_head **iter)
4990 struct netdev_adjacent *lower;
4992 lower = list_entry(*iter, struct netdev_adjacent, list);
4994 if (&lower->list == &dev->adj_list.lower)
4997 *iter = lower->list.next;
4999 return lower->private;
5001 EXPORT_SYMBOL(netdev_lower_get_next_private);
5004 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5005 * lower neighbour list, RCU
5008 * @iter: list_head ** of the current position
5010 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5011 * list, starting from iter position. The caller must hold RCU read lock.
5013 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5014 struct list_head **iter)
5016 struct netdev_adjacent *lower;
5018 WARN_ON_ONCE(!rcu_read_lock_held());
5020 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5022 if (&lower->list == &dev->adj_list.lower)
5025 *iter = &lower->list;
5027 return lower->private;
5029 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5032 * netdev_lower_get_next - Get the next device from the lower neighbour
5035 * @iter: list_head ** of the current position
5037 * Gets the next netdev_adjacent from the dev's lower neighbour
5038 * list, starting from iter position. The caller must hold RTNL lock or
5039 * its own locking that guarantees that the neighbour lower
5040 * list will remain unchainged.
5042 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5044 struct netdev_adjacent *lower;
5046 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5048 if (&lower->list == &dev->adj_list.lower)
5051 *iter = &lower->list;
5055 EXPORT_SYMBOL(netdev_lower_get_next);
5058 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5059 * lower neighbour list, RCU
5063 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5064 * list. The caller must hold RCU read lock.
5066 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5068 struct netdev_adjacent *lower;
5070 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5071 struct netdev_adjacent, list);
5073 return lower->private;
5076 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5079 * netdev_master_upper_dev_get_rcu - Get master upper device
5082 * Find a master upper device and return pointer to it or NULL in case
5083 * it's not there. The caller must hold the RCU read lock.
5085 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5087 struct netdev_adjacent *upper;
5089 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5090 struct netdev_adjacent, list);
5091 if (upper && likely(upper->master))
5095 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5097 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5098 struct net_device *adj_dev,
5099 struct list_head *dev_list)
5101 char linkname[IFNAMSIZ+7];
5102 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5103 "upper_%s" : "lower_%s", adj_dev->name);
5104 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5107 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5109 struct list_head *dev_list)
5111 char linkname[IFNAMSIZ+7];
5112 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5113 "upper_%s" : "lower_%s", name);
5114 sysfs_remove_link(&(dev->dev.kobj), linkname);
5117 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5118 struct net_device *adj_dev,
5119 struct list_head *dev_list)
5121 return (dev_list == &dev->adj_list.upper ||
5122 dev_list == &dev->adj_list.lower) &&
5123 net_eq(dev_net(dev), dev_net(adj_dev));
5126 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5127 struct net_device *adj_dev,
5128 struct list_head *dev_list,
5129 void *private, bool master)
5131 struct netdev_adjacent *adj;
5134 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5141 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5146 adj->master = master;
5148 adj->private = private;
5151 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5152 adj_dev->name, dev->name, adj_dev->name);
5154 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5155 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5160 /* Ensure that master link is always the first item in list. */
5162 ret = sysfs_create_link(&(dev->dev.kobj),
5163 &(adj_dev->dev.kobj), "master");
5165 goto remove_symlinks;
5167 list_add_rcu(&adj->list, dev_list);
5169 list_add_tail_rcu(&adj->list, dev_list);
5175 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5176 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5184 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5185 struct net_device *adj_dev,
5186 struct list_head *dev_list)
5188 struct netdev_adjacent *adj;
5190 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5193 pr_err("tried to remove device %s from %s\n",
5194 dev->name, adj_dev->name);
5198 if (adj->ref_nr > 1) {
5199 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5206 sysfs_remove_link(&(dev->dev.kobj), "master");
5208 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5209 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5211 list_del_rcu(&adj->list);
5212 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5213 adj_dev->name, dev->name, adj_dev->name);
5215 kfree_rcu(adj, rcu);
5218 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5219 struct net_device *upper_dev,
5220 struct list_head *up_list,
5221 struct list_head *down_list,
5222 void *private, bool master)
5226 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5231 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5234 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5241 static int __netdev_adjacent_dev_link(struct net_device *dev,
5242 struct net_device *upper_dev)
5244 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5245 &dev->all_adj_list.upper,
5246 &upper_dev->all_adj_list.lower,
5250 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5251 struct net_device *upper_dev,
5252 struct list_head *up_list,
5253 struct list_head *down_list)
5255 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5256 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5259 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5260 struct net_device *upper_dev)
5262 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5263 &dev->all_adj_list.upper,
5264 &upper_dev->all_adj_list.lower);
5267 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5268 struct net_device *upper_dev,
5269 void *private, bool master)
5271 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5276 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5277 &dev->adj_list.upper,
5278 &upper_dev->adj_list.lower,
5281 __netdev_adjacent_dev_unlink(dev, upper_dev);
5288 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5289 struct net_device *upper_dev)
5291 __netdev_adjacent_dev_unlink(dev, upper_dev);
5292 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5293 &dev->adj_list.upper,
5294 &upper_dev->adj_list.lower);
5297 static int __netdev_upper_dev_link(struct net_device *dev,
5298 struct net_device *upper_dev, bool master,
5301 struct netdev_adjacent *i, *j, *to_i, *to_j;
5306 if (dev == upper_dev)
5309 /* To prevent loops, check if dev is not upper device to upper_dev. */
5310 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5313 if (__netdev_find_adj(dev, upper_dev, &dev->adj_list.upper))
5316 if (master && netdev_master_upper_dev_get(dev))
5319 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5324 /* Now that we linked these devs, make all the upper_dev's
5325 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5326 * versa, and don't forget the devices itself. All of these
5327 * links are non-neighbours.
5329 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5330 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5331 pr_debug("Interlinking %s with %s, non-neighbour\n",
5332 i->dev->name, j->dev->name);
5333 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5339 /* add dev to every upper_dev's upper device */
5340 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5341 pr_debug("linking %s's upper device %s with %s\n",
5342 upper_dev->name, i->dev->name, dev->name);
5343 ret = __netdev_adjacent_dev_link(dev, i->dev);
5345 goto rollback_upper_mesh;
5348 /* add upper_dev to every dev's lower device */
5349 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5350 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5351 i->dev->name, upper_dev->name);
5352 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5354 goto rollback_lower_mesh;
5357 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5360 rollback_lower_mesh:
5362 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5365 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5370 rollback_upper_mesh:
5372 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5375 __netdev_adjacent_dev_unlink(dev, i->dev);
5383 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5384 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5385 if (i == to_i && j == to_j)
5387 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5393 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5399 * netdev_upper_dev_link - Add a link to the upper device
5401 * @upper_dev: new upper device
5403 * Adds a link to device which is upper to this one. The caller must hold
5404 * the RTNL lock. On a failure a negative errno code is returned.
5405 * On success the reference counts are adjusted and the function
5408 int netdev_upper_dev_link(struct net_device *dev,
5409 struct net_device *upper_dev)
5411 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5413 EXPORT_SYMBOL(netdev_upper_dev_link);
5416 * netdev_master_upper_dev_link - Add a master link to the upper device
5418 * @upper_dev: new upper device
5420 * Adds a link to device which is upper to this one. In this case, only
5421 * one master upper device can be linked, although other non-master devices
5422 * might be linked as well. The caller must hold the RTNL lock.
5423 * On a failure a negative errno code is returned. On success the reference
5424 * counts are adjusted and the function returns zero.
5426 int netdev_master_upper_dev_link(struct net_device *dev,
5427 struct net_device *upper_dev)
5429 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5431 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5433 int netdev_master_upper_dev_link_private(struct net_device *dev,
5434 struct net_device *upper_dev,
5437 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5439 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5442 * netdev_upper_dev_unlink - Removes a link to upper device
5444 * @upper_dev: new upper device
5446 * Removes a link to device which is upper to this one. The caller must hold
5449 void netdev_upper_dev_unlink(struct net_device *dev,
5450 struct net_device *upper_dev)
5452 struct netdev_adjacent *i, *j;
5455 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5457 /* Here is the tricky part. We must remove all dev's lower
5458 * devices from all upper_dev's upper devices and vice
5459 * versa, to maintain the graph relationship.
5461 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5462 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5463 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5465 /* remove also the devices itself from lower/upper device
5468 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5469 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5471 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5472 __netdev_adjacent_dev_unlink(dev, i->dev);
5474 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5476 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5479 * netdev_bonding_info_change - Dispatch event about slave change
5481 * @bonding_info: info to dispatch
5483 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5484 * The caller must hold the RTNL lock.
5486 void netdev_bonding_info_change(struct net_device *dev,
5487 struct netdev_bonding_info *bonding_info)
5489 struct netdev_notifier_bonding_info info;
5491 memcpy(&info.bonding_info, bonding_info,
5492 sizeof(struct netdev_bonding_info));
5493 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5496 EXPORT_SYMBOL(netdev_bonding_info_change);
5498 static void netdev_adjacent_add_links(struct net_device *dev)
5500 struct netdev_adjacent *iter;
5502 struct net *net = dev_net(dev);
5504 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5505 if (!net_eq(net,dev_net(iter->dev)))
5507 netdev_adjacent_sysfs_add(iter->dev, dev,
5508 &iter->dev->adj_list.lower);
5509 netdev_adjacent_sysfs_add(dev, iter->dev,
5510 &dev->adj_list.upper);
5513 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5514 if (!net_eq(net,dev_net(iter->dev)))
5516 netdev_adjacent_sysfs_add(iter->dev, dev,
5517 &iter->dev->adj_list.upper);
5518 netdev_adjacent_sysfs_add(dev, iter->dev,
5519 &dev->adj_list.lower);
5523 static void netdev_adjacent_del_links(struct net_device *dev)
5525 struct netdev_adjacent *iter;
5527 struct net *net = dev_net(dev);
5529 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5530 if (!net_eq(net,dev_net(iter->dev)))
5532 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5533 &iter->dev->adj_list.lower);
5534 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5535 &dev->adj_list.upper);
5538 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5539 if (!net_eq(net,dev_net(iter->dev)))
5541 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5542 &iter->dev->adj_list.upper);
5543 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5544 &dev->adj_list.lower);
5548 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5550 struct netdev_adjacent *iter;
5552 struct net *net = dev_net(dev);
5554 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5555 if (!net_eq(net,dev_net(iter->dev)))
5557 netdev_adjacent_sysfs_del(iter->dev, oldname,
5558 &iter->dev->adj_list.lower);
5559 netdev_adjacent_sysfs_add(iter->dev, dev,
5560 &iter->dev->adj_list.lower);
5563 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5564 if (!net_eq(net,dev_net(iter->dev)))
5566 netdev_adjacent_sysfs_del(iter->dev, oldname,
5567 &iter->dev->adj_list.upper);
5568 netdev_adjacent_sysfs_add(iter->dev, dev,
5569 &iter->dev->adj_list.upper);
5573 void *netdev_lower_dev_get_private(struct net_device *dev,
5574 struct net_device *lower_dev)
5576 struct netdev_adjacent *lower;
5580 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5584 return lower->private;
5586 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5589 int dev_get_nest_level(struct net_device *dev,
5590 bool (*type_check)(struct net_device *dev))
5592 struct net_device *lower = NULL;
5593 struct list_head *iter;
5599 netdev_for_each_lower_dev(dev, lower, iter) {
5600 nest = dev_get_nest_level(lower, type_check);
5601 if (max_nest < nest)
5605 if (type_check(dev))
5610 EXPORT_SYMBOL(dev_get_nest_level);
5612 static void dev_change_rx_flags(struct net_device *dev, int flags)
5614 const struct net_device_ops *ops = dev->netdev_ops;
5616 if (ops->ndo_change_rx_flags)
5617 ops->ndo_change_rx_flags(dev, flags);
5620 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5622 unsigned int old_flags = dev->flags;
5628 dev->flags |= IFF_PROMISC;
5629 dev->promiscuity += inc;
5630 if (dev->promiscuity == 0) {
5633 * If inc causes overflow, untouch promisc and return error.
5636 dev->flags &= ~IFF_PROMISC;
5638 dev->promiscuity -= inc;
5639 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5644 if (dev->flags != old_flags) {
5645 pr_info("device %s %s promiscuous mode\n",
5647 dev->flags & IFF_PROMISC ? "entered" : "left");
5648 if (audit_enabled) {
5649 current_uid_gid(&uid, &gid);
5650 audit_log(current->audit_context, GFP_ATOMIC,
5651 AUDIT_ANOM_PROMISCUOUS,
5652 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5653 dev->name, (dev->flags & IFF_PROMISC),
5654 (old_flags & IFF_PROMISC),
5655 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5656 from_kuid(&init_user_ns, uid),
5657 from_kgid(&init_user_ns, gid),
5658 audit_get_sessionid(current));
5661 dev_change_rx_flags(dev, IFF_PROMISC);
5664 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5669 * dev_set_promiscuity - update promiscuity count on a device
5673 * Add or remove promiscuity from a device. While the count in the device
5674 * remains above zero the interface remains promiscuous. Once it hits zero
5675 * the device reverts back to normal filtering operation. A negative inc
5676 * value is used to drop promiscuity on the device.
5677 * Return 0 if successful or a negative errno code on error.
5679 int dev_set_promiscuity(struct net_device *dev, int inc)
5681 unsigned int old_flags = dev->flags;
5684 err = __dev_set_promiscuity(dev, inc, true);
5687 if (dev->flags != old_flags)
5688 dev_set_rx_mode(dev);
5691 EXPORT_SYMBOL(dev_set_promiscuity);
5693 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5695 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5699 dev->flags |= IFF_ALLMULTI;
5700 dev->allmulti += inc;
5701 if (dev->allmulti == 0) {
5704 * If inc causes overflow, untouch allmulti and return error.
5707 dev->flags &= ~IFF_ALLMULTI;
5709 dev->allmulti -= inc;
5710 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5715 if (dev->flags ^ old_flags) {
5716 dev_change_rx_flags(dev, IFF_ALLMULTI);
5717 dev_set_rx_mode(dev);
5719 __dev_notify_flags(dev, old_flags,
5720 dev->gflags ^ old_gflags);
5726 * dev_set_allmulti - update allmulti count on a device
5730 * Add or remove reception of all multicast frames to a device. While the
5731 * count in the device remains above zero the interface remains listening
5732 * to all interfaces. Once it hits zero the device reverts back to normal
5733 * filtering operation. A negative @inc value is used to drop the counter
5734 * when releasing a resource needing all multicasts.
5735 * Return 0 if successful or a negative errno code on error.
5738 int dev_set_allmulti(struct net_device *dev, int inc)
5740 return __dev_set_allmulti(dev, inc, true);
5742 EXPORT_SYMBOL(dev_set_allmulti);
5745 * Upload unicast and multicast address lists to device and
5746 * configure RX filtering. When the device doesn't support unicast
5747 * filtering it is put in promiscuous mode while unicast addresses
5750 void __dev_set_rx_mode(struct net_device *dev)
5752 const struct net_device_ops *ops = dev->netdev_ops;
5754 /* dev_open will call this function so the list will stay sane. */
5755 if (!(dev->flags&IFF_UP))
5758 if (!netif_device_present(dev))
5761 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5762 /* Unicast addresses changes may only happen under the rtnl,
5763 * therefore calling __dev_set_promiscuity here is safe.
5765 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5766 __dev_set_promiscuity(dev, 1, false);
5767 dev->uc_promisc = true;
5768 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5769 __dev_set_promiscuity(dev, -1, false);
5770 dev->uc_promisc = false;
5774 if (ops->ndo_set_rx_mode)
5775 ops->ndo_set_rx_mode(dev);
5778 void dev_set_rx_mode(struct net_device *dev)
5780 netif_addr_lock_bh(dev);
5781 __dev_set_rx_mode(dev);
5782 netif_addr_unlock_bh(dev);
5786 * dev_get_flags - get flags reported to userspace
5789 * Get the combination of flag bits exported through APIs to userspace.
5791 unsigned int dev_get_flags(const struct net_device *dev)
5795 flags = (dev->flags & ~(IFF_PROMISC |
5800 (dev->gflags & (IFF_PROMISC |
5803 if (netif_running(dev)) {
5804 if (netif_oper_up(dev))
5805 flags |= IFF_RUNNING;
5806 if (netif_carrier_ok(dev))
5807 flags |= IFF_LOWER_UP;
5808 if (netif_dormant(dev))
5809 flags |= IFF_DORMANT;
5814 EXPORT_SYMBOL(dev_get_flags);
5816 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5818 unsigned int old_flags = dev->flags;
5824 * Set the flags on our device.
5827 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5828 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5830 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5834 * Load in the correct multicast list now the flags have changed.
5837 if ((old_flags ^ flags) & IFF_MULTICAST)
5838 dev_change_rx_flags(dev, IFF_MULTICAST);
5840 dev_set_rx_mode(dev);
5843 * Have we downed the interface. We handle IFF_UP ourselves
5844 * according to user attempts to set it, rather than blindly
5849 if ((old_flags ^ flags) & IFF_UP)
5850 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5852 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5853 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5854 unsigned int old_flags = dev->flags;
5856 dev->gflags ^= IFF_PROMISC;
5858 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5859 if (dev->flags != old_flags)
5860 dev_set_rx_mode(dev);
5863 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5864 is important. Some (broken) drivers set IFF_PROMISC, when
5865 IFF_ALLMULTI is requested not asking us and not reporting.
5867 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5868 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5870 dev->gflags ^= IFF_ALLMULTI;
5871 __dev_set_allmulti(dev, inc, false);
5877 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5878 unsigned int gchanges)
5880 unsigned int changes = dev->flags ^ old_flags;
5883 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5885 if (changes & IFF_UP) {
5886 if (dev->flags & IFF_UP)
5887 call_netdevice_notifiers(NETDEV_UP, dev);
5889 call_netdevice_notifiers(NETDEV_DOWN, dev);
5892 if (dev->flags & IFF_UP &&
5893 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5894 struct netdev_notifier_change_info change_info;
5896 change_info.flags_changed = changes;
5897 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5903 * dev_change_flags - change device settings
5905 * @flags: device state flags
5907 * Change settings on device based state flags. The flags are
5908 * in the userspace exported format.
5910 int dev_change_flags(struct net_device *dev, unsigned int flags)
5913 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5915 ret = __dev_change_flags(dev, flags);
5919 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5920 __dev_notify_flags(dev, old_flags, changes);
5923 EXPORT_SYMBOL(dev_change_flags);
5925 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5927 const struct net_device_ops *ops = dev->netdev_ops;
5929 if (ops->ndo_change_mtu)
5930 return ops->ndo_change_mtu(dev, new_mtu);
5937 * dev_set_mtu - Change maximum transfer unit
5939 * @new_mtu: new transfer unit
5941 * Change the maximum transfer size of the network device.
5943 int dev_set_mtu(struct net_device *dev, int new_mtu)
5947 if (new_mtu == dev->mtu)
5950 /* MTU must be positive. */
5954 if (!netif_device_present(dev))
5957 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5958 err = notifier_to_errno(err);
5962 orig_mtu = dev->mtu;
5963 err = __dev_set_mtu(dev, new_mtu);
5966 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5967 err = notifier_to_errno(err);
5969 /* setting mtu back and notifying everyone again,
5970 * so that they have a chance to revert changes.
5972 __dev_set_mtu(dev, orig_mtu);
5973 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5978 EXPORT_SYMBOL(dev_set_mtu);
5981 * dev_set_group - Change group this device belongs to
5983 * @new_group: group this device should belong to
5985 void dev_set_group(struct net_device *dev, int new_group)
5987 dev->group = new_group;
5989 EXPORT_SYMBOL(dev_set_group);
5992 * dev_set_mac_address - Change Media Access Control Address
5996 * Change the hardware (MAC) address of the device
5998 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6000 const struct net_device_ops *ops = dev->netdev_ops;
6003 if (!ops->ndo_set_mac_address)
6005 if (sa->sa_family != dev->type)
6007 if (!netif_device_present(dev))
6009 err = ops->ndo_set_mac_address(dev, sa);
6012 dev->addr_assign_type = NET_ADDR_SET;
6013 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6014 add_device_randomness(dev->dev_addr, dev->addr_len);
6017 EXPORT_SYMBOL(dev_set_mac_address);
6020 * dev_change_carrier - Change device carrier
6022 * @new_carrier: new value
6024 * Change device carrier
6026 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6028 const struct net_device_ops *ops = dev->netdev_ops;
6030 if (!ops->ndo_change_carrier)
6032 if (!netif_device_present(dev))
6034 return ops->ndo_change_carrier(dev, new_carrier);
6036 EXPORT_SYMBOL(dev_change_carrier);
6039 * dev_get_phys_port_id - Get device physical port ID
6043 * Get device physical port ID
6045 int dev_get_phys_port_id(struct net_device *dev,
6046 struct netdev_phys_item_id *ppid)
6048 const struct net_device_ops *ops = dev->netdev_ops;
6050 if (!ops->ndo_get_phys_port_id)
6052 return ops->ndo_get_phys_port_id(dev, ppid);
6054 EXPORT_SYMBOL(dev_get_phys_port_id);
6057 * dev_get_phys_port_name - Get device physical port name
6061 * Get device physical port name
6063 int dev_get_phys_port_name(struct net_device *dev,
6064 char *name, size_t len)
6066 const struct net_device_ops *ops = dev->netdev_ops;
6068 if (!ops->ndo_get_phys_port_name)
6070 return ops->ndo_get_phys_port_name(dev, name, len);
6072 EXPORT_SYMBOL(dev_get_phys_port_name);
6075 * dev_new_index - allocate an ifindex
6076 * @net: the applicable net namespace
6078 * Returns a suitable unique value for a new device interface
6079 * number. The caller must hold the rtnl semaphore or the
6080 * dev_base_lock to be sure it remains unique.
6082 static int dev_new_index(struct net *net)
6084 int ifindex = net->ifindex;
6088 if (!__dev_get_by_index(net, ifindex))
6089 return net->ifindex = ifindex;
6093 /* Delayed registration/unregisteration */
6094 static LIST_HEAD(net_todo_list);
6095 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6097 static void net_set_todo(struct net_device *dev)
6099 list_add_tail(&dev->todo_list, &net_todo_list);
6100 dev_net(dev)->dev_unreg_count++;
6103 static void rollback_registered_many(struct list_head *head)
6105 struct net_device *dev, *tmp;
6106 LIST_HEAD(close_head);
6108 BUG_ON(dev_boot_phase);
6111 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6112 /* Some devices call without registering
6113 * for initialization unwind. Remove those
6114 * devices and proceed with the remaining.
6116 if (dev->reg_state == NETREG_UNINITIALIZED) {
6117 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6121 list_del(&dev->unreg_list);
6124 dev->dismantle = true;
6125 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6128 /* If device is running, close it first. */
6129 list_for_each_entry(dev, head, unreg_list)
6130 list_add_tail(&dev->close_list, &close_head);
6131 dev_close_many(&close_head, true);
6133 list_for_each_entry(dev, head, unreg_list) {
6134 /* And unlink it from device chain. */
6135 unlist_netdevice(dev);
6137 dev->reg_state = NETREG_UNREGISTERING;
6142 list_for_each_entry(dev, head, unreg_list) {
6143 struct sk_buff *skb = NULL;
6145 /* Shutdown queueing discipline. */
6149 /* Notify protocols, that we are about to destroy
6150 this device. They should clean all the things.
6152 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6154 if (!dev->rtnl_link_ops ||
6155 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6156 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6160 * Flush the unicast and multicast chains
6165 if (dev->netdev_ops->ndo_uninit)
6166 dev->netdev_ops->ndo_uninit(dev);
6169 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6171 /* Notifier chain MUST detach us all upper devices. */
6172 WARN_ON(netdev_has_any_upper_dev(dev));
6174 /* Remove entries from kobject tree */
6175 netdev_unregister_kobject(dev);
6177 /* Remove XPS queueing entries */
6178 netif_reset_xps_queues_gt(dev, 0);
6184 list_for_each_entry(dev, head, unreg_list)
6188 static void rollback_registered(struct net_device *dev)
6192 list_add(&dev->unreg_list, &single);
6193 rollback_registered_many(&single);
6197 static netdev_features_t netdev_fix_features(struct net_device *dev,
6198 netdev_features_t features)
6200 /* Fix illegal checksum combinations */
6201 if ((features & NETIF_F_HW_CSUM) &&
6202 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6203 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6204 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6207 /* TSO requires that SG is present as well. */
6208 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6209 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6210 features &= ~NETIF_F_ALL_TSO;
6213 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6214 !(features & NETIF_F_IP_CSUM)) {
6215 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6216 features &= ~NETIF_F_TSO;
6217 features &= ~NETIF_F_TSO_ECN;
6220 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6221 !(features & NETIF_F_IPV6_CSUM)) {
6222 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6223 features &= ~NETIF_F_TSO6;
6226 /* TSO ECN requires that TSO is present as well. */
6227 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6228 features &= ~NETIF_F_TSO_ECN;
6230 /* Software GSO depends on SG. */
6231 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6232 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6233 features &= ~NETIF_F_GSO;
6236 /* UFO needs SG and checksumming */
6237 if (features & NETIF_F_UFO) {
6238 /* maybe split UFO into V4 and V6? */
6239 if (!((features & NETIF_F_GEN_CSUM) ||
6240 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6241 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6243 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6244 features &= ~NETIF_F_UFO;
6247 if (!(features & NETIF_F_SG)) {
6249 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6250 features &= ~NETIF_F_UFO;
6254 #ifdef CONFIG_NET_RX_BUSY_POLL
6255 if (dev->netdev_ops->ndo_busy_poll)
6256 features |= NETIF_F_BUSY_POLL;
6259 features &= ~NETIF_F_BUSY_POLL;
6264 int __netdev_update_features(struct net_device *dev)
6266 netdev_features_t features;
6271 features = netdev_get_wanted_features(dev);
6273 if (dev->netdev_ops->ndo_fix_features)
6274 features = dev->netdev_ops->ndo_fix_features(dev, features);
6276 /* driver might be less strict about feature dependencies */
6277 features = netdev_fix_features(dev, features);
6279 if (dev->features == features)
6282 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6283 &dev->features, &features);
6285 if (dev->netdev_ops->ndo_set_features)
6286 err = dev->netdev_ops->ndo_set_features(dev, features);
6288 if (unlikely(err < 0)) {
6290 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6291 err, &features, &dev->features);
6296 dev->features = features;
6302 * netdev_update_features - recalculate device features
6303 * @dev: the device to check
6305 * Recalculate dev->features set and send notifications if it
6306 * has changed. Should be called after driver or hardware dependent
6307 * conditions might have changed that influence the features.
6309 void netdev_update_features(struct net_device *dev)
6311 if (__netdev_update_features(dev))
6312 netdev_features_change(dev);
6314 EXPORT_SYMBOL(netdev_update_features);
6317 * netdev_change_features - recalculate device features
6318 * @dev: the device to check
6320 * Recalculate dev->features set and send notifications even
6321 * if they have not changed. Should be called instead of
6322 * netdev_update_features() if also dev->vlan_features might
6323 * have changed to allow the changes to be propagated to stacked
6326 void netdev_change_features(struct net_device *dev)
6328 __netdev_update_features(dev);
6329 netdev_features_change(dev);
6331 EXPORT_SYMBOL(netdev_change_features);
6334 * netif_stacked_transfer_operstate - transfer operstate
6335 * @rootdev: the root or lower level device to transfer state from
6336 * @dev: the device to transfer operstate to
6338 * Transfer operational state from root to device. This is normally
6339 * called when a stacking relationship exists between the root
6340 * device and the device(a leaf device).
6342 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6343 struct net_device *dev)
6345 if (rootdev->operstate == IF_OPER_DORMANT)
6346 netif_dormant_on(dev);
6348 netif_dormant_off(dev);
6350 if (netif_carrier_ok(rootdev)) {
6351 if (!netif_carrier_ok(dev))
6352 netif_carrier_on(dev);
6354 if (netif_carrier_ok(dev))
6355 netif_carrier_off(dev);
6358 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6361 static int netif_alloc_rx_queues(struct net_device *dev)
6363 unsigned int i, count = dev->num_rx_queues;
6364 struct netdev_rx_queue *rx;
6365 size_t sz = count * sizeof(*rx);
6369 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6377 for (i = 0; i < count; i++)
6383 static void netdev_init_one_queue(struct net_device *dev,
6384 struct netdev_queue *queue, void *_unused)
6386 /* Initialize queue lock */
6387 spin_lock_init(&queue->_xmit_lock);
6388 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6389 queue->xmit_lock_owner = -1;
6390 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6393 dql_init(&queue->dql, HZ);
6397 static void netif_free_tx_queues(struct net_device *dev)
6402 static int netif_alloc_netdev_queues(struct net_device *dev)
6404 unsigned int count = dev->num_tx_queues;
6405 struct netdev_queue *tx;
6406 size_t sz = count * sizeof(*tx);
6408 if (count < 1 || count > 0xffff)
6411 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6419 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6420 spin_lock_init(&dev->tx_global_lock);
6425 void netif_tx_stop_all_queues(struct net_device *dev)
6429 for (i = 0; i < dev->num_tx_queues; i++) {
6430 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
6431 netif_tx_stop_queue(txq);
6434 EXPORT_SYMBOL(netif_tx_stop_all_queues);
6437 * register_netdevice - register a network device
6438 * @dev: device to register
6440 * Take a completed network device structure and add it to the kernel
6441 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6442 * chain. 0 is returned on success. A negative errno code is returned
6443 * on a failure to set up the device, or if the name is a duplicate.
6445 * Callers must hold the rtnl semaphore. You may want
6446 * register_netdev() instead of this.
6449 * The locking appears insufficient to guarantee two parallel registers
6450 * will not get the same name.
6453 int register_netdevice(struct net_device *dev)
6456 struct net *net = dev_net(dev);
6458 BUG_ON(dev_boot_phase);
6463 /* When net_device's are persistent, this will be fatal. */
6464 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6467 spin_lock_init(&dev->addr_list_lock);
6468 netdev_set_addr_lockdep_class(dev);
6470 ret = dev_get_valid_name(net, dev, dev->name);
6474 /* Init, if this function is available */
6475 if (dev->netdev_ops->ndo_init) {
6476 ret = dev->netdev_ops->ndo_init(dev);
6484 if (((dev->hw_features | dev->features) &
6485 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6486 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6487 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6488 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6495 dev->ifindex = dev_new_index(net);
6496 else if (__dev_get_by_index(net, dev->ifindex))
6499 /* Transfer changeable features to wanted_features and enable
6500 * software offloads (GSO and GRO).
6502 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6503 dev->features |= NETIF_F_SOFT_FEATURES;
6504 dev->wanted_features = dev->features & dev->hw_features;
6506 if (!(dev->flags & IFF_LOOPBACK)) {
6507 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6510 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6512 dev->vlan_features |= NETIF_F_HIGHDMA;
6514 /* Make NETIF_F_SG inheritable to tunnel devices.
6516 dev->hw_enc_features |= NETIF_F_SG;
6518 /* Make NETIF_F_SG inheritable to MPLS.
6520 dev->mpls_features |= NETIF_F_SG;
6522 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6523 ret = notifier_to_errno(ret);
6527 ret = netdev_register_kobject(dev);
6530 dev->reg_state = NETREG_REGISTERED;
6532 __netdev_update_features(dev);
6535 * Default initial state at registry is that the
6536 * device is present.
6539 set_bit(__LINK_STATE_PRESENT, &dev->state);
6541 linkwatch_init_dev(dev);
6543 dev_init_scheduler(dev);
6545 list_netdevice(dev);
6546 add_device_randomness(dev->dev_addr, dev->addr_len);
6548 /* If the device has permanent device address, driver should
6549 * set dev_addr and also addr_assign_type should be set to
6550 * NET_ADDR_PERM (default value).
6552 if (dev->addr_assign_type == NET_ADDR_PERM)
6553 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6555 /* Notify protocols, that a new device appeared. */
6556 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6557 ret = notifier_to_errno(ret);
6559 rollback_registered(dev);
6560 dev->reg_state = NETREG_UNREGISTERED;
6563 * Prevent userspace races by waiting until the network
6564 * device is fully setup before sending notifications.
6566 if (!dev->rtnl_link_ops ||
6567 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6568 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6574 if (dev->netdev_ops->ndo_uninit)
6575 dev->netdev_ops->ndo_uninit(dev);
6578 EXPORT_SYMBOL(register_netdevice);
6581 * init_dummy_netdev - init a dummy network device for NAPI
6582 * @dev: device to init
6584 * This takes a network device structure and initialize the minimum
6585 * amount of fields so it can be used to schedule NAPI polls without
6586 * registering a full blown interface. This is to be used by drivers
6587 * that need to tie several hardware interfaces to a single NAPI
6588 * poll scheduler due to HW limitations.
6590 int init_dummy_netdev(struct net_device *dev)
6592 /* Clear everything. Note we don't initialize spinlocks
6593 * are they aren't supposed to be taken by any of the
6594 * NAPI code and this dummy netdev is supposed to be
6595 * only ever used for NAPI polls
6597 memset(dev, 0, sizeof(struct net_device));
6599 /* make sure we BUG if trying to hit standard
6600 * register/unregister code path
6602 dev->reg_state = NETREG_DUMMY;
6604 /* NAPI wants this */
6605 INIT_LIST_HEAD(&dev->napi_list);
6607 /* a dummy interface is started by default */
6608 set_bit(__LINK_STATE_PRESENT, &dev->state);
6609 set_bit(__LINK_STATE_START, &dev->state);
6611 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6612 * because users of this 'device' dont need to change
6618 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6622 * register_netdev - register a network device
6623 * @dev: device to register
6625 * Take a completed network device structure and add it to the kernel
6626 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6627 * chain. 0 is returned on success. A negative errno code is returned
6628 * on a failure to set up the device, or if the name is a duplicate.
6630 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6631 * and expands the device name if you passed a format string to
6634 int register_netdev(struct net_device *dev)
6639 err = register_netdevice(dev);
6643 EXPORT_SYMBOL(register_netdev);
6645 int netdev_refcnt_read(const struct net_device *dev)
6649 for_each_possible_cpu(i)
6650 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6653 EXPORT_SYMBOL(netdev_refcnt_read);
6656 * netdev_wait_allrefs - wait until all references are gone.
6657 * @dev: target net_device
6659 * This is called when unregistering network devices.
6661 * Any protocol or device that holds a reference should register
6662 * for netdevice notification, and cleanup and put back the
6663 * reference if they receive an UNREGISTER event.
6664 * We can get stuck here if buggy protocols don't correctly
6667 static void netdev_wait_allrefs(struct net_device *dev)
6669 unsigned long rebroadcast_time, warning_time;
6672 linkwatch_forget_dev(dev);
6674 rebroadcast_time = warning_time = jiffies;
6675 refcnt = netdev_refcnt_read(dev);
6677 while (refcnt != 0) {
6678 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6681 /* Rebroadcast unregister notification */
6682 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6688 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6689 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6691 /* We must not have linkwatch events
6692 * pending on unregister. If this
6693 * happens, we simply run the queue
6694 * unscheduled, resulting in a noop
6697 linkwatch_run_queue();
6702 rebroadcast_time = jiffies;
6707 refcnt = netdev_refcnt_read(dev);
6709 if (time_after(jiffies, warning_time + 10 * HZ)) {
6710 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6712 warning_time = jiffies;
6721 * register_netdevice(x1);
6722 * register_netdevice(x2);
6724 * unregister_netdevice(y1);
6725 * unregister_netdevice(y2);
6731 * We are invoked by rtnl_unlock().
6732 * This allows us to deal with problems:
6733 * 1) We can delete sysfs objects which invoke hotplug
6734 * without deadlocking with linkwatch via keventd.
6735 * 2) Since we run with the RTNL semaphore not held, we can sleep
6736 * safely in order to wait for the netdev refcnt to drop to zero.
6738 * We must not return until all unregister events added during
6739 * the interval the lock was held have been completed.
6741 void netdev_run_todo(void)
6743 struct list_head list;
6745 /* Snapshot list, allow later requests */
6746 list_replace_init(&net_todo_list, &list);
6751 /* Wait for rcu callbacks to finish before next phase */
6752 if (!list_empty(&list))
6755 while (!list_empty(&list)) {
6756 struct net_device *dev
6757 = list_first_entry(&list, struct net_device, todo_list);
6758 list_del(&dev->todo_list);
6761 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6764 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6765 pr_err("network todo '%s' but state %d\n",
6766 dev->name, dev->reg_state);
6771 dev->reg_state = NETREG_UNREGISTERED;
6773 on_each_cpu(flush_backlog, dev, 1);
6775 netdev_wait_allrefs(dev);
6778 BUG_ON(netdev_refcnt_read(dev));
6779 BUG_ON(!list_empty(&dev->ptype_all));
6780 BUG_ON(!list_empty(&dev->ptype_specific));
6781 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6782 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6783 WARN_ON(dev->dn_ptr);
6785 if (dev->destructor)
6786 dev->destructor(dev);
6788 /* Report a network device has been unregistered */
6790 dev_net(dev)->dev_unreg_count--;
6792 wake_up(&netdev_unregistering_wq);
6794 /* Free network device */
6795 kobject_put(&dev->dev.kobj);
6799 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6800 * fields in the same order, with only the type differing.
6802 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6803 const struct net_device_stats *netdev_stats)
6805 #if BITS_PER_LONG == 64
6806 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6807 memcpy(stats64, netdev_stats, sizeof(*stats64));
6809 size_t i, n = sizeof(*stats64) / sizeof(u64);
6810 const unsigned long *src = (const unsigned long *)netdev_stats;
6811 u64 *dst = (u64 *)stats64;
6813 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6814 sizeof(*stats64) / sizeof(u64));
6815 for (i = 0; i < n; i++)
6819 EXPORT_SYMBOL(netdev_stats_to_stats64);
6822 * dev_get_stats - get network device statistics
6823 * @dev: device to get statistics from
6824 * @storage: place to store stats
6826 * Get network statistics from device. Return @storage.
6827 * The device driver may provide its own method by setting
6828 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6829 * otherwise the internal statistics structure is used.
6831 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6832 struct rtnl_link_stats64 *storage)
6834 const struct net_device_ops *ops = dev->netdev_ops;
6836 if (ops->ndo_get_stats64) {
6837 memset(storage, 0, sizeof(*storage));
6838 ops->ndo_get_stats64(dev, storage);
6839 } else if (ops->ndo_get_stats) {
6840 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6842 netdev_stats_to_stats64(storage, &dev->stats);
6844 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6845 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6848 EXPORT_SYMBOL(dev_get_stats);
6850 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6852 struct netdev_queue *queue = dev_ingress_queue(dev);
6854 #ifdef CONFIG_NET_CLS_ACT
6857 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6860 netdev_init_one_queue(dev, queue, NULL);
6861 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
6862 queue->qdisc_sleeping = &noop_qdisc;
6863 rcu_assign_pointer(dev->ingress_queue, queue);
6868 static const struct ethtool_ops default_ethtool_ops;
6870 void netdev_set_default_ethtool_ops(struct net_device *dev,
6871 const struct ethtool_ops *ops)
6873 if (dev->ethtool_ops == &default_ethtool_ops)
6874 dev->ethtool_ops = ops;
6876 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6878 void netdev_freemem(struct net_device *dev)
6880 char *addr = (char *)dev - dev->padded;
6886 * alloc_netdev_mqs - allocate network device
6887 * @sizeof_priv: size of private data to allocate space for
6888 * @name: device name format string
6889 * @name_assign_type: origin of device name
6890 * @setup: callback to initialize device
6891 * @txqs: the number of TX subqueues to allocate
6892 * @rxqs: the number of RX subqueues to allocate
6894 * Allocates a struct net_device with private data area for driver use
6895 * and performs basic initialization. Also allocates subqueue structs
6896 * for each queue on the device.
6898 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6899 unsigned char name_assign_type,
6900 void (*setup)(struct net_device *),
6901 unsigned int txqs, unsigned int rxqs)
6903 struct net_device *dev;
6905 struct net_device *p;
6907 BUG_ON(strlen(name) >= sizeof(dev->name));
6910 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6916 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6921 alloc_size = sizeof(struct net_device);
6923 /* ensure 32-byte alignment of private area */
6924 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6925 alloc_size += sizeof_priv;
6927 /* ensure 32-byte alignment of whole construct */
6928 alloc_size += NETDEV_ALIGN - 1;
6930 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6932 p = vzalloc(alloc_size);
6936 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6937 dev->padded = (char *)dev - (char *)p;
6939 dev->pcpu_refcnt = alloc_percpu(int);
6940 if (!dev->pcpu_refcnt)
6943 if (dev_addr_init(dev))
6949 dev_net_set(dev, &init_net);
6951 dev->gso_max_size = GSO_MAX_SIZE;
6952 dev->gso_max_segs = GSO_MAX_SEGS;
6953 dev->gso_min_segs = 0;
6955 INIT_LIST_HEAD(&dev->napi_list);
6956 INIT_LIST_HEAD(&dev->unreg_list);
6957 INIT_LIST_HEAD(&dev->close_list);
6958 INIT_LIST_HEAD(&dev->link_watch_list);
6959 INIT_LIST_HEAD(&dev->adj_list.upper);
6960 INIT_LIST_HEAD(&dev->adj_list.lower);
6961 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6962 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6963 INIT_LIST_HEAD(&dev->ptype_all);
6964 INIT_LIST_HEAD(&dev->ptype_specific);
6965 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
6968 dev->num_tx_queues = txqs;
6969 dev->real_num_tx_queues = txqs;
6970 if (netif_alloc_netdev_queues(dev))
6974 dev->num_rx_queues = rxqs;
6975 dev->real_num_rx_queues = rxqs;
6976 if (netif_alloc_rx_queues(dev))
6980 strcpy(dev->name, name);
6981 dev->name_assign_type = name_assign_type;
6982 dev->group = INIT_NETDEV_GROUP;
6983 if (!dev->ethtool_ops)
6984 dev->ethtool_ops = &default_ethtool_ops;
6986 nf_hook_ingress_init(dev);
6995 free_percpu(dev->pcpu_refcnt);
6997 netdev_freemem(dev);
7000 EXPORT_SYMBOL(alloc_netdev_mqs);
7003 * free_netdev - free network device
7006 * This function does the last stage of destroying an allocated device
7007 * interface. The reference to the device object is released.
7008 * If this is the last reference then it will be freed.
7010 void free_netdev(struct net_device *dev)
7012 struct napi_struct *p, *n;
7014 netif_free_tx_queues(dev);
7019 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7021 /* Flush device addresses */
7022 dev_addr_flush(dev);
7024 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7027 free_percpu(dev->pcpu_refcnt);
7028 dev->pcpu_refcnt = NULL;
7030 /* Compatibility with error handling in drivers */
7031 if (dev->reg_state == NETREG_UNINITIALIZED) {
7032 netdev_freemem(dev);
7036 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7037 dev->reg_state = NETREG_RELEASED;
7039 /* will free via device release */
7040 put_device(&dev->dev);
7042 EXPORT_SYMBOL(free_netdev);
7045 * synchronize_net - Synchronize with packet receive processing
7047 * Wait for packets currently being received to be done.
7048 * Does not block later packets from starting.
7050 void synchronize_net(void)
7053 if (rtnl_is_locked())
7054 synchronize_rcu_expedited();
7058 EXPORT_SYMBOL(synchronize_net);
7061 * unregister_netdevice_queue - remove device from the kernel
7065 * This function shuts down a device interface and removes it
7066 * from the kernel tables.
7067 * If head not NULL, device is queued to be unregistered later.
7069 * Callers must hold the rtnl semaphore. You may want
7070 * unregister_netdev() instead of this.
7073 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7078 list_move_tail(&dev->unreg_list, head);
7080 rollback_registered(dev);
7081 /* Finish processing unregister after unlock */
7085 EXPORT_SYMBOL(unregister_netdevice_queue);
7088 * unregister_netdevice_many - unregister many devices
7089 * @head: list of devices
7091 * Note: As most callers use a stack allocated list_head,
7092 * we force a list_del() to make sure stack wont be corrupted later.
7094 void unregister_netdevice_many(struct list_head *head)
7096 struct net_device *dev;
7098 if (!list_empty(head)) {
7099 rollback_registered_many(head);
7100 list_for_each_entry(dev, head, unreg_list)
7105 EXPORT_SYMBOL(unregister_netdevice_many);
7108 * unregister_netdev - remove device from the kernel
7111 * This function shuts down a device interface and removes it
7112 * from the kernel tables.
7114 * This is just a wrapper for unregister_netdevice that takes
7115 * the rtnl semaphore. In general you want to use this and not
7116 * unregister_netdevice.
7118 void unregister_netdev(struct net_device *dev)
7121 unregister_netdevice(dev);
7124 EXPORT_SYMBOL(unregister_netdev);
7127 * dev_change_net_namespace - move device to different nethost namespace
7129 * @net: network namespace
7130 * @pat: If not NULL name pattern to try if the current device name
7131 * is already taken in the destination network namespace.
7133 * This function shuts down a device interface and moves it
7134 * to a new network namespace. On success 0 is returned, on
7135 * a failure a netagive errno code is returned.
7137 * Callers must hold the rtnl semaphore.
7140 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7146 /* Don't allow namespace local devices to be moved. */
7148 if (dev->features & NETIF_F_NETNS_LOCAL)
7151 /* Ensure the device has been registrered */
7152 if (dev->reg_state != NETREG_REGISTERED)
7155 /* Get out if there is nothing todo */
7157 if (net_eq(dev_net(dev), net))
7160 /* Pick the destination device name, and ensure
7161 * we can use it in the destination network namespace.
7164 if (__dev_get_by_name(net, dev->name)) {
7165 /* We get here if we can't use the current device name */
7168 if (dev_get_valid_name(net, dev, pat) < 0)
7173 * And now a mini version of register_netdevice unregister_netdevice.
7176 /* If device is running close it first. */
7179 /* And unlink it from device chain */
7181 unlist_netdevice(dev);
7185 /* Shutdown queueing discipline. */
7188 /* Notify protocols, that we are about to destroy
7189 this device. They should clean all the things.
7191 Note that dev->reg_state stays at NETREG_REGISTERED.
7192 This is wanted because this way 8021q and macvlan know
7193 the device is just moving and can keep their slaves up.
7195 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7197 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7198 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7201 * Flush the unicast and multicast chains
7206 /* Send a netdev-removed uevent to the old namespace */
7207 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7208 netdev_adjacent_del_links(dev);
7210 /* Actually switch the network namespace */
7211 dev_net_set(dev, net);
7213 /* If there is an ifindex conflict assign a new one */
7214 if (__dev_get_by_index(net, dev->ifindex))
7215 dev->ifindex = dev_new_index(net);
7217 /* Send a netdev-add uevent to the new namespace */
7218 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7219 netdev_adjacent_add_links(dev);
7221 /* Fixup kobjects */
7222 err = device_rename(&dev->dev, dev->name);
7225 /* Add the device back in the hashes */
7226 list_netdevice(dev);
7228 /* Notify protocols, that a new device appeared. */
7229 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7232 * Prevent userspace races by waiting until the network
7233 * device is fully setup before sending notifications.
7235 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7242 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7244 static int dev_cpu_callback(struct notifier_block *nfb,
7245 unsigned long action,
7248 struct sk_buff **list_skb;
7249 struct sk_buff *skb;
7250 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7251 struct softnet_data *sd, *oldsd;
7253 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7256 local_irq_disable();
7257 cpu = smp_processor_id();
7258 sd = &per_cpu(softnet_data, cpu);
7259 oldsd = &per_cpu(softnet_data, oldcpu);
7261 /* Find end of our completion_queue. */
7262 list_skb = &sd->completion_queue;
7264 list_skb = &(*list_skb)->next;
7265 /* Append completion queue from offline CPU. */
7266 *list_skb = oldsd->completion_queue;
7267 oldsd->completion_queue = NULL;
7269 /* Append output queue from offline CPU. */
7270 if (oldsd->output_queue) {
7271 *sd->output_queue_tailp = oldsd->output_queue;
7272 sd->output_queue_tailp = oldsd->output_queue_tailp;
7273 oldsd->output_queue = NULL;
7274 oldsd->output_queue_tailp = &oldsd->output_queue;
7276 /* Append NAPI poll list from offline CPU, with one exception :
7277 * process_backlog() must be called by cpu owning percpu backlog.
7278 * We properly handle process_queue & input_pkt_queue later.
7280 while (!list_empty(&oldsd->poll_list)) {
7281 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7285 list_del_init(&napi->poll_list);
7286 if (napi->poll == process_backlog)
7289 ____napi_schedule(sd, napi);
7292 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7295 /* Process offline CPU's input_pkt_queue */
7296 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7298 input_queue_head_incr(oldsd);
7300 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7302 input_queue_head_incr(oldsd);
7310 * netdev_increment_features - increment feature set by one
7311 * @all: current feature set
7312 * @one: new feature set
7313 * @mask: mask feature set
7315 * Computes a new feature set after adding a device with feature set
7316 * @one to the master device with current feature set @all. Will not
7317 * enable anything that is off in @mask. Returns the new feature set.
7319 netdev_features_t netdev_increment_features(netdev_features_t all,
7320 netdev_features_t one, netdev_features_t mask)
7322 if (mask & NETIF_F_GEN_CSUM)
7323 mask |= NETIF_F_ALL_CSUM;
7324 mask |= NETIF_F_VLAN_CHALLENGED;
7326 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7327 all &= one | ~NETIF_F_ALL_FOR_ALL;
7329 /* If one device supports hw checksumming, set for all. */
7330 if (all & NETIF_F_GEN_CSUM)
7331 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7335 EXPORT_SYMBOL(netdev_increment_features);
7337 static struct hlist_head * __net_init netdev_create_hash(void)
7340 struct hlist_head *hash;
7342 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7344 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7345 INIT_HLIST_HEAD(&hash[i]);
7350 /* Initialize per network namespace state */
7351 static int __net_init netdev_init(struct net *net)
7353 if (net != &init_net)
7354 INIT_LIST_HEAD(&net->dev_base_head);
7356 net->dev_name_head = netdev_create_hash();
7357 if (net->dev_name_head == NULL)
7360 net->dev_index_head = netdev_create_hash();
7361 if (net->dev_index_head == NULL)
7367 kfree(net->dev_name_head);
7373 * netdev_drivername - network driver for the device
7374 * @dev: network device
7376 * Determine network driver for device.
7378 const char *netdev_drivername(const struct net_device *dev)
7380 const struct device_driver *driver;
7381 const struct device *parent;
7382 const char *empty = "";
7384 parent = dev->dev.parent;
7388 driver = parent->driver;
7389 if (driver && driver->name)
7390 return driver->name;
7394 static void __netdev_printk(const char *level, const struct net_device *dev,
7395 struct va_format *vaf)
7397 if (dev && dev->dev.parent) {
7398 dev_printk_emit(level[1] - '0',
7401 dev_driver_string(dev->dev.parent),
7402 dev_name(dev->dev.parent),
7403 netdev_name(dev), netdev_reg_state(dev),
7406 printk("%s%s%s: %pV",
7407 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7409 printk("%s(NULL net_device): %pV", level, vaf);
7413 void netdev_printk(const char *level, const struct net_device *dev,
7414 const char *format, ...)
7416 struct va_format vaf;
7419 va_start(args, format);
7424 __netdev_printk(level, dev, &vaf);
7428 EXPORT_SYMBOL(netdev_printk);
7430 #define define_netdev_printk_level(func, level) \
7431 void func(const struct net_device *dev, const char *fmt, ...) \
7433 struct va_format vaf; \
7436 va_start(args, fmt); \
7441 __netdev_printk(level, dev, &vaf); \
7445 EXPORT_SYMBOL(func);
7447 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7448 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7449 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7450 define_netdev_printk_level(netdev_err, KERN_ERR);
7451 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7452 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7453 define_netdev_printk_level(netdev_info, KERN_INFO);
7455 static void __net_exit netdev_exit(struct net *net)
7457 kfree(net->dev_name_head);
7458 kfree(net->dev_index_head);
7461 static struct pernet_operations __net_initdata netdev_net_ops = {
7462 .init = netdev_init,
7463 .exit = netdev_exit,
7466 static void __net_exit default_device_exit(struct net *net)
7468 struct net_device *dev, *aux;
7470 * Push all migratable network devices back to the
7471 * initial network namespace
7474 for_each_netdev_safe(net, dev, aux) {
7476 char fb_name[IFNAMSIZ];
7478 /* Ignore unmoveable devices (i.e. loopback) */
7479 if (dev->features & NETIF_F_NETNS_LOCAL)
7482 /* Leave virtual devices for the generic cleanup */
7483 if (dev->rtnl_link_ops)
7486 /* Push remaining network devices to init_net */
7487 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7488 err = dev_change_net_namespace(dev, &init_net, fb_name);
7490 pr_emerg("%s: failed to move %s to init_net: %d\n",
7491 __func__, dev->name, err);
7498 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7500 /* Return with the rtnl_lock held when there are no network
7501 * devices unregistering in any network namespace in net_list.
7505 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7507 add_wait_queue(&netdev_unregistering_wq, &wait);
7509 unregistering = false;
7511 list_for_each_entry(net, net_list, exit_list) {
7512 if (net->dev_unreg_count > 0) {
7513 unregistering = true;
7521 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7523 remove_wait_queue(&netdev_unregistering_wq, &wait);
7526 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7528 /* At exit all network devices most be removed from a network
7529 * namespace. Do this in the reverse order of registration.
7530 * Do this across as many network namespaces as possible to
7531 * improve batching efficiency.
7533 struct net_device *dev;
7535 LIST_HEAD(dev_kill_list);
7537 /* To prevent network device cleanup code from dereferencing
7538 * loopback devices or network devices that have been freed
7539 * wait here for all pending unregistrations to complete,
7540 * before unregistring the loopback device and allowing the
7541 * network namespace be freed.
7543 * The netdev todo list containing all network devices
7544 * unregistrations that happen in default_device_exit_batch
7545 * will run in the rtnl_unlock() at the end of
7546 * default_device_exit_batch.
7548 rtnl_lock_unregistering(net_list);
7549 list_for_each_entry(net, net_list, exit_list) {
7550 for_each_netdev_reverse(net, dev) {
7551 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7552 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7554 unregister_netdevice_queue(dev, &dev_kill_list);
7557 unregister_netdevice_many(&dev_kill_list);
7561 static struct pernet_operations __net_initdata default_device_ops = {
7562 .exit = default_device_exit,
7563 .exit_batch = default_device_exit_batch,
7567 * Initialize the DEV module. At boot time this walks the device list and
7568 * unhooks any devices that fail to initialise (normally hardware not
7569 * present) and leaves us with a valid list of present and active devices.
7574 * This is called single threaded during boot, so no need
7575 * to take the rtnl semaphore.
7577 static int __init net_dev_init(void)
7579 int i, rc = -ENOMEM;
7581 BUG_ON(!dev_boot_phase);
7583 if (dev_proc_init())
7586 if (netdev_kobject_init())
7589 INIT_LIST_HEAD(&ptype_all);
7590 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7591 INIT_LIST_HEAD(&ptype_base[i]);
7593 INIT_LIST_HEAD(&offload_base);
7595 if (register_pernet_subsys(&netdev_net_ops))
7599 * Initialise the packet receive queues.
7602 for_each_possible_cpu(i) {
7603 struct softnet_data *sd = &per_cpu(softnet_data, i);
7605 skb_queue_head_init(&sd->input_pkt_queue);
7606 skb_queue_head_init(&sd->process_queue);
7607 INIT_LIST_HEAD(&sd->poll_list);
7608 sd->output_queue_tailp = &sd->output_queue;
7610 sd->csd.func = rps_trigger_softirq;
7615 sd->backlog.poll = process_backlog;
7616 sd->backlog.weight = weight_p;
7621 /* The loopback device is special if any other network devices
7622 * is present in a network namespace the loopback device must
7623 * be present. Since we now dynamically allocate and free the
7624 * loopback device ensure this invariant is maintained by
7625 * keeping the loopback device as the first device on the
7626 * list of network devices. Ensuring the loopback devices
7627 * is the first device that appears and the last network device
7630 if (register_pernet_device(&loopback_net_ops))
7633 if (register_pernet_device(&default_device_ops))
7636 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7637 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7639 hotcpu_notifier(dev_cpu_callback, 0);
7646 subsys_initcall(net_dev_init);