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 <linux/ipv6.h>
122 #include <linux/in.h>
123 #include <linux/jhash.h>
124 #include <linux/random.h>
125 #include <trace/events/napi.h>
126 #include <trace/events/net.h>
127 #include <trace/events/skb.h>
128 #include <linux/pci.h>
129 #include <linux/inetdevice.h>
130 #include <linux/cpu_rmap.h>
131 #include <linux/static_key.h>
132 #include <linux/hashtable.h>
133 #include <linux/vmalloc.h>
134 #include <linux/if_macvlan.h>
136 #include "net-sysfs.h"
138 /* Instead of increasing this, you should create a hash table. */
139 #define MAX_GRO_SKBS 8
141 /* This should be increased if a protocol with a bigger head is added. */
142 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 static DEFINE_SPINLOCK(ptype_lock);
145 static DEFINE_SPINLOCK(offload_lock);
146 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
147 struct list_head ptype_all __read_mostly; /* Taps */
148 static struct list_head offload_base __read_mostly;
150 static int netif_rx_internal(struct sk_buff *skb);
151 static int call_netdevice_notifiers_info(unsigned long val,
152 struct net_device *dev,
153 struct netdev_notifier_info *info);
156 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
159 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
161 * Writers must hold the rtnl semaphore while they loop through the
162 * dev_base_head list, and hold dev_base_lock for writing when they do the
163 * actual updates. This allows pure readers to access the list even
164 * while a writer is preparing to update it.
166 * To put it another way, dev_base_lock is held for writing only to
167 * protect against pure readers; the rtnl semaphore provides the
168 * protection against other writers.
170 * See, for example usages, register_netdevice() and
171 * unregister_netdevice(), which must be called with the rtnl
174 DEFINE_RWLOCK(dev_base_lock);
175 EXPORT_SYMBOL(dev_base_lock);
177 /* protects napi_hash addition/deletion and napi_gen_id */
178 static DEFINE_SPINLOCK(napi_hash_lock);
180 static unsigned int napi_gen_id;
181 static DEFINE_HASHTABLE(napi_hash, 8);
183 static seqcount_t devnet_rename_seq;
185 static inline void dev_base_seq_inc(struct net *net)
187 while (++net->dev_base_seq == 0);
190 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
192 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
194 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
197 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
199 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
202 static inline void rps_lock(struct softnet_data *sd)
205 spin_lock(&sd->input_pkt_queue.lock);
209 static inline void rps_unlock(struct softnet_data *sd)
212 spin_unlock(&sd->input_pkt_queue.lock);
216 /* Device list insertion */
217 static void list_netdevice(struct net_device *dev)
219 struct net *net = dev_net(dev);
223 write_lock_bh(&dev_base_lock);
224 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
225 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
226 hlist_add_head_rcu(&dev->index_hlist,
227 dev_index_hash(net, dev->ifindex));
228 write_unlock_bh(&dev_base_lock);
230 dev_base_seq_inc(net);
233 /* Device list removal
234 * caller must respect a RCU grace period before freeing/reusing dev
236 static void unlist_netdevice(struct net_device *dev)
240 /* Unlink dev from the device chain */
241 write_lock_bh(&dev_base_lock);
242 list_del_rcu(&dev->dev_list);
243 hlist_del_rcu(&dev->name_hlist);
244 hlist_del_rcu(&dev->index_hlist);
245 write_unlock_bh(&dev_base_lock);
247 dev_base_seq_inc(dev_net(dev));
254 static RAW_NOTIFIER_HEAD(netdev_chain);
257 * Device drivers call our routines to queue packets here. We empty the
258 * queue in the local softnet handler.
261 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
262 EXPORT_PER_CPU_SYMBOL(softnet_data);
264 #ifdef CONFIG_LOCKDEP
266 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
267 * according to dev->type
269 static const unsigned short netdev_lock_type[] =
270 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
271 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
272 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
273 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
274 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
275 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
276 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
277 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
278 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
279 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
280 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
281 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
282 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
283 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
284 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
286 static const char *const netdev_lock_name[] =
287 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
288 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
289 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
290 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
291 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
292 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
293 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
294 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
295 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
296 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
297 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
298 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
299 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
300 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
301 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
303 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
304 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
306 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
310 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
311 if (netdev_lock_type[i] == dev_type)
313 /* the last key is used by default */
314 return ARRAY_SIZE(netdev_lock_type) - 1;
317 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
318 unsigned short dev_type)
322 i = netdev_lock_pos(dev_type);
323 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
324 netdev_lock_name[i]);
327 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
331 i = netdev_lock_pos(dev->type);
332 lockdep_set_class_and_name(&dev->addr_list_lock,
333 &netdev_addr_lock_key[i],
334 netdev_lock_name[i]);
337 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
338 unsigned short dev_type)
341 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
346 /*******************************************************************************
348 Protocol management and registration routines
350 *******************************************************************************/
353 * Add a protocol ID to the list. Now that the input handler is
354 * smarter we can dispense with all the messy stuff that used to be
357 * BEWARE!!! Protocol handlers, mangling input packets,
358 * MUST BE last in hash buckets and checking protocol handlers
359 * MUST start from promiscuous ptype_all chain in net_bh.
360 * It is true now, do not change it.
361 * Explanation follows: if protocol handler, mangling packet, will
362 * be the first on list, it is not able to sense, that packet
363 * is cloned and should be copied-on-write, so that it will
364 * change it and subsequent readers will get broken packet.
368 static inline struct list_head *ptype_head(const struct packet_type *pt)
370 if (pt->type == htons(ETH_P_ALL))
373 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
377 * dev_add_pack - add packet handler
378 * @pt: packet type declaration
380 * Add a protocol handler to the networking stack. The passed &packet_type
381 * is linked into kernel lists and may not be freed until it has been
382 * removed from the kernel lists.
384 * This call does not sleep therefore it can not
385 * guarantee all CPU's that are in middle of receiving packets
386 * will see the new packet type (until the next received packet).
389 void dev_add_pack(struct packet_type *pt)
391 struct list_head *head = ptype_head(pt);
393 spin_lock(&ptype_lock);
394 list_add_rcu(&pt->list, head);
395 spin_unlock(&ptype_lock);
397 EXPORT_SYMBOL(dev_add_pack);
400 * __dev_remove_pack - remove packet handler
401 * @pt: packet type declaration
403 * Remove a protocol handler that was previously added to the kernel
404 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
405 * from the kernel lists and can be freed or reused once this function
408 * The packet type might still be in use by receivers
409 * and must not be freed until after all the CPU's have gone
410 * through a quiescent state.
412 void __dev_remove_pack(struct packet_type *pt)
414 struct list_head *head = ptype_head(pt);
415 struct packet_type *pt1;
417 spin_lock(&ptype_lock);
419 list_for_each_entry(pt1, head, list) {
421 list_del_rcu(&pt->list);
426 pr_warn("dev_remove_pack: %p not found\n", pt);
428 spin_unlock(&ptype_lock);
430 EXPORT_SYMBOL(__dev_remove_pack);
433 * dev_remove_pack - remove packet handler
434 * @pt: packet type declaration
436 * Remove a protocol handler that was previously added to the kernel
437 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
438 * from the kernel lists and can be freed or reused once this function
441 * This call sleeps to guarantee that no CPU is looking at the packet
444 void dev_remove_pack(struct packet_type *pt)
446 __dev_remove_pack(pt);
450 EXPORT_SYMBOL(dev_remove_pack);
454 * dev_add_offload - register offload handlers
455 * @po: protocol offload declaration
457 * Add protocol offload handlers to the networking stack. The passed
458 * &proto_offload is linked into kernel lists and may not be freed until
459 * it has been removed from the kernel lists.
461 * This call does not sleep therefore it can not
462 * guarantee all CPU's that are in middle of receiving packets
463 * will see the new offload handlers (until the next received packet).
465 void dev_add_offload(struct packet_offload *po)
467 struct list_head *head = &offload_base;
469 spin_lock(&offload_lock);
470 list_add_rcu(&po->list, head);
471 spin_unlock(&offload_lock);
473 EXPORT_SYMBOL(dev_add_offload);
476 * __dev_remove_offload - remove offload handler
477 * @po: packet offload declaration
479 * Remove a protocol offload handler that was previously added to the
480 * kernel offload handlers by dev_add_offload(). The passed &offload_type
481 * is removed from the kernel lists and can be freed or reused once this
484 * The packet type might still be in use by receivers
485 * and must not be freed until after all the CPU's have gone
486 * through a quiescent state.
488 static void __dev_remove_offload(struct packet_offload *po)
490 struct list_head *head = &offload_base;
491 struct packet_offload *po1;
493 spin_lock(&offload_lock);
495 list_for_each_entry(po1, head, list) {
497 list_del_rcu(&po->list);
502 pr_warn("dev_remove_offload: %p not found\n", po);
504 spin_unlock(&offload_lock);
508 * dev_remove_offload - remove packet offload handler
509 * @po: packet offload declaration
511 * Remove a packet offload handler that was previously added to the kernel
512 * offload handlers by dev_add_offload(). The passed &offload_type is
513 * removed from the kernel lists and can be freed or reused once this
516 * This call sleeps to guarantee that no CPU is looking at the packet
519 void dev_remove_offload(struct packet_offload *po)
521 __dev_remove_offload(po);
525 EXPORT_SYMBOL(dev_remove_offload);
527 /******************************************************************************
529 Device Boot-time Settings Routines
531 *******************************************************************************/
533 /* Boot time configuration table */
534 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
537 * netdev_boot_setup_add - add new setup entry
538 * @name: name of the device
539 * @map: configured settings for the device
541 * Adds new setup entry to the dev_boot_setup list. The function
542 * returns 0 on error and 1 on success. This is a generic routine to
545 static int netdev_boot_setup_add(char *name, struct ifmap *map)
547 struct netdev_boot_setup *s;
551 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
552 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
553 memset(s[i].name, 0, sizeof(s[i].name));
554 strlcpy(s[i].name, name, IFNAMSIZ);
555 memcpy(&s[i].map, map, sizeof(s[i].map));
560 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
564 * netdev_boot_setup_check - check boot time settings
565 * @dev: the netdevice
567 * Check boot time settings for the device.
568 * The found settings are set for the device to be used
569 * later in the device probing.
570 * Returns 0 if no settings found, 1 if they are.
572 int netdev_boot_setup_check(struct net_device *dev)
574 struct netdev_boot_setup *s = dev_boot_setup;
577 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
578 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
579 !strcmp(dev->name, s[i].name)) {
580 dev->irq = s[i].map.irq;
581 dev->base_addr = s[i].map.base_addr;
582 dev->mem_start = s[i].map.mem_start;
583 dev->mem_end = s[i].map.mem_end;
589 EXPORT_SYMBOL(netdev_boot_setup_check);
593 * netdev_boot_base - get address from boot time settings
594 * @prefix: prefix for network device
595 * @unit: id for network device
597 * Check boot time settings for the base address of device.
598 * The found settings are set for the device to be used
599 * later in the device probing.
600 * Returns 0 if no settings found.
602 unsigned long netdev_boot_base(const char *prefix, int unit)
604 const struct netdev_boot_setup *s = dev_boot_setup;
608 sprintf(name, "%s%d", prefix, unit);
611 * If device already registered then return base of 1
612 * to indicate not to probe for this interface
614 if (__dev_get_by_name(&init_net, name))
617 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
618 if (!strcmp(name, s[i].name))
619 return s[i].map.base_addr;
624 * Saves at boot time configured settings for any netdevice.
626 int __init netdev_boot_setup(char *str)
631 str = get_options(str, ARRAY_SIZE(ints), ints);
636 memset(&map, 0, sizeof(map));
640 map.base_addr = ints[2];
642 map.mem_start = ints[3];
644 map.mem_end = ints[4];
646 /* Add new entry to the list */
647 return netdev_boot_setup_add(str, &map);
650 __setup("netdev=", netdev_boot_setup);
652 /*******************************************************************************
654 Device Interface Subroutines
656 *******************************************************************************/
659 * __dev_get_by_name - find a device by its name
660 * @net: the applicable net namespace
661 * @name: name to find
663 * Find an interface by name. Must be called under RTNL semaphore
664 * or @dev_base_lock. If the name is found a pointer to the device
665 * is returned. If the name is not found then %NULL is returned. The
666 * reference counters are not incremented so the caller must be
667 * careful with locks.
670 struct net_device *__dev_get_by_name(struct net *net, const char *name)
672 struct net_device *dev;
673 struct hlist_head *head = dev_name_hash(net, name);
675 hlist_for_each_entry(dev, head, name_hlist)
676 if (!strncmp(dev->name, name, IFNAMSIZ))
681 EXPORT_SYMBOL(__dev_get_by_name);
684 * dev_get_by_name_rcu - find a device by its name
685 * @net: the applicable net namespace
686 * @name: name to find
688 * Find an interface by name.
689 * If the name is found a pointer to the device is returned.
690 * If the name is not found then %NULL is returned.
691 * The reference counters are not incremented so the caller must be
692 * careful with locks. The caller must hold RCU lock.
695 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
697 struct net_device *dev;
698 struct hlist_head *head = dev_name_hash(net, name);
700 hlist_for_each_entry_rcu(dev, head, name_hlist)
701 if (!strncmp(dev->name, name, IFNAMSIZ))
706 EXPORT_SYMBOL(dev_get_by_name_rcu);
709 * dev_get_by_name - find a device by its name
710 * @net: the applicable net namespace
711 * @name: name to find
713 * Find an interface by name. This can be called from any
714 * context and does its own locking. The returned handle has
715 * the usage count incremented and the caller must use dev_put() to
716 * release it when it is no longer needed. %NULL is returned if no
717 * matching device is found.
720 struct net_device *dev_get_by_name(struct net *net, const char *name)
722 struct net_device *dev;
725 dev = dev_get_by_name_rcu(net, name);
731 EXPORT_SYMBOL(dev_get_by_name);
734 * __dev_get_by_index - find a device by its ifindex
735 * @net: the applicable net namespace
736 * @ifindex: index of device
738 * Search for an interface by index. Returns %NULL if the device
739 * is not found or a pointer to the device. The device has not
740 * had its reference counter increased so the caller must be careful
741 * about locking. The caller must hold either the RTNL semaphore
745 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
747 struct net_device *dev;
748 struct hlist_head *head = dev_index_hash(net, ifindex);
750 hlist_for_each_entry(dev, head, index_hlist)
751 if (dev->ifindex == ifindex)
756 EXPORT_SYMBOL(__dev_get_by_index);
759 * dev_get_by_index_rcu - find a device by its ifindex
760 * @net: the applicable net namespace
761 * @ifindex: index of device
763 * Search for an interface by index. Returns %NULL if the device
764 * is not found or a pointer to the device. The device has not
765 * had its reference counter increased so the caller must be careful
766 * about locking. The caller must hold RCU lock.
769 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
771 struct net_device *dev;
772 struct hlist_head *head = dev_index_hash(net, ifindex);
774 hlist_for_each_entry_rcu(dev, head, index_hlist)
775 if (dev->ifindex == ifindex)
780 EXPORT_SYMBOL(dev_get_by_index_rcu);
784 * dev_get_by_index - find a device by its ifindex
785 * @net: the applicable net namespace
786 * @ifindex: index of device
788 * Search for an interface by index. Returns NULL if the device
789 * is not found or a pointer to the device. The device returned has
790 * had a reference added and the pointer is safe until the user calls
791 * dev_put to indicate they have finished with it.
794 struct net_device *dev_get_by_index(struct net *net, int ifindex)
796 struct net_device *dev;
799 dev = dev_get_by_index_rcu(net, ifindex);
805 EXPORT_SYMBOL(dev_get_by_index);
808 * netdev_get_name - get a netdevice name, knowing its ifindex.
809 * @net: network namespace
810 * @name: a pointer to the buffer where the name will be stored.
811 * @ifindex: the ifindex of the interface to get the name from.
813 * The use of raw_seqcount_begin() and cond_resched() before
814 * retrying is required as we want to give the writers a chance
815 * to complete when CONFIG_PREEMPT is not set.
817 int netdev_get_name(struct net *net, char *name, int ifindex)
819 struct net_device *dev;
823 seq = raw_seqcount_begin(&devnet_rename_seq);
825 dev = dev_get_by_index_rcu(net, ifindex);
831 strcpy(name, dev->name);
833 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
842 * dev_getbyhwaddr_rcu - find a device by its hardware address
843 * @net: the applicable net namespace
844 * @type: media type of device
845 * @ha: hardware address
847 * Search for an interface by MAC address. Returns NULL if the device
848 * is not found or a pointer to the device.
849 * The caller must hold RCU or RTNL.
850 * The returned device has not had its ref count increased
851 * and the caller must therefore be careful about locking
855 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
858 struct net_device *dev;
860 for_each_netdev_rcu(net, dev)
861 if (dev->type == type &&
862 !memcmp(dev->dev_addr, ha, dev->addr_len))
867 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
869 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
871 struct net_device *dev;
874 for_each_netdev(net, dev)
875 if (dev->type == type)
880 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
882 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
884 struct net_device *dev, *ret = NULL;
887 for_each_netdev_rcu(net, dev)
888 if (dev->type == type) {
896 EXPORT_SYMBOL(dev_getfirstbyhwtype);
899 * dev_get_by_flags_rcu - find any device with given flags
900 * @net: the applicable net namespace
901 * @if_flags: IFF_* values
902 * @mask: bitmask of bits in if_flags to check
904 * Search for any interface with the given flags. Returns NULL if a device
905 * is not found or a pointer to the device. Must be called inside
906 * rcu_read_lock(), and result refcount is unchanged.
909 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
912 struct net_device *dev, *ret;
915 for_each_netdev_rcu(net, dev) {
916 if (((dev->flags ^ if_flags) & mask) == 0) {
923 EXPORT_SYMBOL(dev_get_by_flags_rcu);
926 * dev_valid_name - check if name is okay for network device
929 * Network device names need to be valid file names to
930 * to allow sysfs to work. We also disallow any kind of
933 bool dev_valid_name(const char *name)
937 if (strlen(name) >= IFNAMSIZ)
939 if (!strcmp(name, ".") || !strcmp(name, ".."))
943 if (*name == '/' || isspace(*name))
949 EXPORT_SYMBOL(dev_valid_name);
952 * __dev_alloc_name - allocate a name for a device
953 * @net: network namespace to allocate the device name in
954 * @name: name format string
955 * @buf: scratch buffer and result name string
957 * Passed a format string - eg "lt%d" it will try and find a suitable
958 * id. It scans list of devices to build up a free map, then chooses
959 * the first empty slot. The caller must hold the dev_base or rtnl lock
960 * while allocating the name and adding the device in order to avoid
962 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
963 * Returns the number of the unit assigned or a negative errno code.
966 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
970 const int max_netdevices = 8*PAGE_SIZE;
971 unsigned long *inuse;
972 struct net_device *d;
974 p = strnchr(name, IFNAMSIZ-1, '%');
977 * Verify the string as this thing may have come from
978 * the user. There must be either one "%d" and no other "%"
981 if (p[1] != 'd' || strchr(p + 2, '%'))
984 /* Use one page as a bit array of possible slots */
985 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
989 for_each_netdev(net, d) {
990 if (!sscanf(d->name, name, &i))
992 if (i < 0 || i >= max_netdevices)
995 /* avoid cases where sscanf is not exact inverse of printf */
996 snprintf(buf, IFNAMSIZ, name, i);
997 if (!strncmp(buf, d->name, IFNAMSIZ))
1001 i = find_first_zero_bit(inuse, max_netdevices);
1002 free_page((unsigned long) inuse);
1006 snprintf(buf, IFNAMSIZ, name, i);
1007 if (!__dev_get_by_name(net, buf))
1010 /* It is possible to run out of possible slots
1011 * when the name is long and there isn't enough space left
1012 * for the digits, or if all bits are used.
1018 * dev_alloc_name - allocate a name for a device
1020 * @name: name format string
1022 * Passed a format string - eg "lt%d" it will try and find a suitable
1023 * id. It scans list of devices to build up a free map, then chooses
1024 * the first empty slot. The caller must hold the dev_base or rtnl lock
1025 * while allocating the name and adding the device in order to avoid
1027 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1028 * Returns the number of the unit assigned or a negative errno code.
1031 int dev_alloc_name(struct net_device *dev, const char *name)
1037 BUG_ON(!dev_net(dev));
1039 ret = __dev_alloc_name(net, name, buf);
1041 strlcpy(dev->name, buf, IFNAMSIZ);
1044 EXPORT_SYMBOL(dev_alloc_name);
1046 static int dev_alloc_name_ns(struct net *net,
1047 struct net_device *dev,
1053 ret = __dev_alloc_name(net, name, buf);
1055 strlcpy(dev->name, buf, IFNAMSIZ);
1059 static int dev_get_valid_name(struct net *net,
1060 struct net_device *dev,
1065 if (!dev_valid_name(name))
1068 if (strchr(name, '%'))
1069 return dev_alloc_name_ns(net, dev, name);
1070 else if (__dev_get_by_name(net, name))
1072 else if (dev->name != name)
1073 strlcpy(dev->name, name, IFNAMSIZ);
1079 * dev_change_name - change name of a device
1081 * @newname: name (or format string) must be at least IFNAMSIZ
1083 * Change name of a device, can pass format strings "eth%d".
1086 int dev_change_name(struct net_device *dev, const char *newname)
1088 unsigned char old_assign_type;
1089 char oldname[IFNAMSIZ];
1095 BUG_ON(!dev_net(dev));
1098 if (dev->flags & IFF_UP)
1101 write_seqcount_begin(&devnet_rename_seq);
1103 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1104 write_seqcount_end(&devnet_rename_seq);
1108 memcpy(oldname, dev->name, IFNAMSIZ);
1110 err = dev_get_valid_name(net, dev, newname);
1112 write_seqcount_end(&devnet_rename_seq);
1116 old_assign_type = dev->name_assign_type;
1117 dev->name_assign_type = NET_NAME_RENAMED;
1120 ret = device_rename(&dev->dev, dev->name);
1122 memcpy(dev->name, oldname, IFNAMSIZ);
1123 dev->name_assign_type = old_assign_type;
1124 write_seqcount_end(&devnet_rename_seq);
1128 write_seqcount_end(&devnet_rename_seq);
1130 netdev_adjacent_rename_links(dev, oldname);
1132 write_lock_bh(&dev_base_lock);
1133 hlist_del_rcu(&dev->name_hlist);
1134 write_unlock_bh(&dev_base_lock);
1138 write_lock_bh(&dev_base_lock);
1139 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1140 write_unlock_bh(&dev_base_lock);
1142 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1143 ret = notifier_to_errno(ret);
1146 /* err >= 0 after dev_alloc_name() or stores the first errno */
1149 write_seqcount_begin(&devnet_rename_seq);
1150 memcpy(dev->name, oldname, IFNAMSIZ);
1151 memcpy(oldname, newname, IFNAMSIZ);
1152 dev->name_assign_type = old_assign_type;
1153 old_assign_type = NET_NAME_RENAMED;
1156 pr_err("%s: name change rollback failed: %d\n",
1165 * dev_set_alias - change ifalias of a device
1167 * @alias: name up to IFALIASZ
1168 * @len: limit of bytes to copy from info
1170 * Set ifalias for a device,
1172 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1178 if (len >= IFALIASZ)
1182 kfree(dev->ifalias);
1183 dev->ifalias = NULL;
1187 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1190 dev->ifalias = new_ifalias;
1192 strlcpy(dev->ifalias, alias, len+1);
1198 * netdev_features_change - device changes features
1199 * @dev: device to cause notification
1201 * Called to indicate a device has changed features.
1203 void netdev_features_change(struct net_device *dev)
1205 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1207 EXPORT_SYMBOL(netdev_features_change);
1210 * netdev_state_change - device changes state
1211 * @dev: device to cause notification
1213 * Called to indicate a device has changed state. This function calls
1214 * the notifier chains for netdev_chain and sends a NEWLINK message
1215 * to the routing socket.
1217 void netdev_state_change(struct net_device *dev)
1219 if (dev->flags & IFF_UP) {
1220 struct netdev_notifier_change_info change_info;
1222 change_info.flags_changed = 0;
1223 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1225 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1228 EXPORT_SYMBOL(netdev_state_change);
1231 * netdev_notify_peers - notify network peers about existence of @dev
1232 * @dev: network device
1234 * Generate traffic such that interested network peers are aware of
1235 * @dev, such as by generating a gratuitous ARP. This may be used when
1236 * a device wants to inform the rest of the network about some sort of
1237 * reconfiguration such as a failover event or virtual machine
1240 void netdev_notify_peers(struct net_device *dev)
1243 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1246 EXPORT_SYMBOL(netdev_notify_peers);
1248 static int __dev_open(struct net_device *dev)
1250 const struct net_device_ops *ops = dev->netdev_ops;
1255 if (!netif_device_present(dev))
1258 /* Block netpoll from trying to do any rx path servicing.
1259 * If we don't do this there is a chance ndo_poll_controller
1260 * or ndo_poll may be running while we open the device
1262 netpoll_poll_disable(dev);
1264 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1265 ret = notifier_to_errno(ret);
1269 set_bit(__LINK_STATE_START, &dev->state);
1271 if (ops->ndo_validate_addr)
1272 ret = ops->ndo_validate_addr(dev);
1274 if (!ret && ops->ndo_open)
1275 ret = ops->ndo_open(dev);
1277 netpoll_poll_enable(dev);
1280 clear_bit(__LINK_STATE_START, &dev->state);
1282 dev->flags |= IFF_UP;
1283 net_dmaengine_get();
1284 dev_set_rx_mode(dev);
1286 add_device_randomness(dev->dev_addr, dev->addr_len);
1293 * dev_open - prepare an interface for use.
1294 * @dev: device to open
1296 * Takes a device from down to up state. The device's private open
1297 * function is invoked and then the multicast lists are loaded. Finally
1298 * the device is moved into the up state and a %NETDEV_UP message is
1299 * sent to the netdev notifier chain.
1301 * Calling this function on an active interface is a nop. On a failure
1302 * a negative errno code is returned.
1304 int dev_open(struct net_device *dev)
1308 if (dev->flags & IFF_UP)
1311 ret = __dev_open(dev);
1315 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1316 call_netdevice_notifiers(NETDEV_UP, dev);
1320 EXPORT_SYMBOL(dev_open);
1322 static int __dev_close_many(struct list_head *head)
1324 struct net_device *dev;
1329 list_for_each_entry(dev, head, close_list) {
1330 /* Temporarily disable netpoll until the interface is down */
1331 netpoll_poll_disable(dev);
1333 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1335 clear_bit(__LINK_STATE_START, &dev->state);
1337 /* Synchronize to scheduled poll. We cannot touch poll list, it
1338 * can be even on different cpu. So just clear netif_running().
1340 * dev->stop() will invoke napi_disable() on all of it's
1341 * napi_struct instances on this device.
1343 smp_mb__after_atomic(); /* Commit netif_running(). */
1346 dev_deactivate_many(head);
1348 list_for_each_entry(dev, head, close_list) {
1349 const struct net_device_ops *ops = dev->netdev_ops;
1352 * Call the device specific close. This cannot fail.
1353 * Only if device is UP
1355 * We allow it to be called even after a DETACH hot-plug
1361 dev->flags &= ~IFF_UP;
1362 net_dmaengine_put();
1363 netpoll_poll_enable(dev);
1369 static int __dev_close(struct net_device *dev)
1374 list_add(&dev->close_list, &single);
1375 retval = __dev_close_many(&single);
1381 static int dev_close_many(struct list_head *head)
1383 struct net_device *dev, *tmp;
1385 /* Remove the devices that don't need to be closed */
1386 list_for_each_entry_safe(dev, tmp, head, close_list)
1387 if (!(dev->flags & IFF_UP))
1388 list_del_init(&dev->close_list);
1390 __dev_close_many(head);
1392 list_for_each_entry_safe(dev, tmp, head, close_list) {
1393 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1394 call_netdevice_notifiers(NETDEV_DOWN, dev);
1395 list_del_init(&dev->close_list);
1402 * dev_close - shutdown an interface.
1403 * @dev: device to shutdown
1405 * This function moves an active device into down state. A
1406 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1407 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1410 int dev_close(struct net_device *dev)
1412 if (dev->flags & IFF_UP) {
1415 list_add(&dev->close_list, &single);
1416 dev_close_many(&single);
1421 EXPORT_SYMBOL(dev_close);
1425 * dev_disable_lro - disable Large Receive Offload on a device
1428 * Disable Large Receive Offload (LRO) on a net device. Must be
1429 * called under RTNL. This is needed if received packets may be
1430 * forwarded to another interface.
1432 void dev_disable_lro(struct net_device *dev)
1435 * If we're trying to disable lro on a vlan device
1436 * use the underlying physical device instead
1438 if (is_vlan_dev(dev))
1439 dev = vlan_dev_real_dev(dev);
1441 /* the same for macvlan devices */
1442 if (netif_is_macvlan(dev))
1443 dev = macvlan_dev_real_dev(dev);
1445 dev->wanted_features &= ~NETIF_F_LRO;
1446 netdev_update_features(dev);
1448 if (unlikely(dev->features & NETIF_F_LRO))
1449 netdev_WARN(dev, "failed to disable LRO!\n");
1451 EXPORT_SYMBOL(dev_disable_lro);
1453 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1454 struct net_device *dev)
1456 struct netdev_notifier_info info;
1458 netdev_notifier_info_init(&info, dev);
1459 return nb->notifier_call(nb, val, &info);
1462 static int dev_boot_phase = 1;
1465 * register_netdevice_notifier - register a network notifier block
1468 * Register a notifier to be called when network device events occur.
1469 * The notifier passed is linked into the kernel structures and must
1470 * not be reused until it has been unregistered. A negative errno code
1471 * is returned on a failure.
1473 * When registered all registration and up events are replayed
1474 * to the new notifier to allow device to have a race free
1475 * view of the network device list.
1478 int register_netdevice_notifier(struct notifier_block *nb)
1480 struct net_device *dev;
1481 struct net_device *last;
1486 err = raw_notifier_chain_register(&netdev_chain, nb);
1492 for_each_netdev(net, dev) {
1493 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1494 err = notifier_to_errno(err);
1498 if (!(dev->flags & IFF_UP))
1501 call_netdevice_notifier(nb, NETDEV_UP, dev);
1512 for_each_netdev(net, dev) {
1516 if (dev->flags & IFF_UP) {
1517 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1519 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1521 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1526 raw_notifier_chain_unregister(&netdev_chain, nb);
1529 EXPORT_SYMBOL(register_netdevice_notifier);
1532 * unregister_netdevice_notifier - unregister a network notifier block
1535 * Unregister a notifier previously registered by
1536 * register_netdevice_notifier(). The notifier is unlinked into the
1537 * kernel structures and may then be reused. A negative errno code
1538 * is returned on a failure.
1540 * After unregistering unregister and down device events are synthesized
1541 * for all devices on the device list to the removed notifier to remove
1542 * the need for special case cleanup code.
1545 int unregister_netdevice_notifier(struct notifier_block *nb)
1547 struct net_device *dev;
1552 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1557 for_each_netdev(net, dev) {
1558 if (dev->flags & IFF_UP) {
1559 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1561 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1563 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1570 EXPORT_SYMBOL(unregister_netdevice_notifier);
1573 * call_netdevice_notifiers_info - call all network notifier blocks
1574 * @val: value passed unmodified to notifier function
1575 * @dev: net_device pointer passed unmodified to notifier function
1576 * @info: notifier information data
1578 * Call all network notifier blocks. Parameters and return value
1579 * are as for raw_notifier_call_chain().
1582 static int call_netdevice_notifiers_info(unsigned long val,
1583 struct net_device *dev,
1584 struct netdev_notifier_info *info)
1587 netdev_notifier_info_init(info, dev);
1588 return raw_notifier_call_chain(&netdev_chain, val, info);
1592 * call_netdevice_notifiers - call all network notifier blocks
1593 * @val: value passed unmodified to notifier function
1594 * @dev: net_device pointer passed unmodified to notifier function
1596 * Call all network notifier blocks. Parameters and return value
1597 * are as for raw_notifier_call_chain().
1600 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1602 struct netdev_notifier_info info;
1604 return call_netdevice_notifiers_info(val, dev, &info);
1606 EXPORT_SYMBOL(call_netdevice_notifiers);
1608 static struct static_key netstamp_needed __read_mostly;
1609 #ifdef HAVE_JUMP_LABEL
1610 /* We are not allowed to call static_key_slow_dec() from irq context
1611 * If net_disable_timestamp() is called from irq context, defer the
1612 * static_key_slow_dec() calls.
1614 static atomic_t netstamp_needed_deferred;
1617 void net_enable_timestamp(void)
1619 #ifdef HAVE_JUMP_LABEL
1620 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1624 static_key_slow_dec(&netstamp_needed);
1628 static_key_slow_inc(&netstamp_needed);
1630 EXPORT_SYMBOL(net_enable_timestamp);
1632 void net_disable_timestamp(void)
1634 #ifdef HAVE_JUMP_LABEL
1635 if (in_interrupt()) {
1636 atomic_inc(&netstamp_needed_deferred);
1640 static_key_slow_dec(&netstamp_needed);
1642 EXPORT_SYMBOL(net_disable_timestamp);
1644 static inline void net_timestamp_set(struct sk_buff *skb)
1646 skb->tstamp.tv64 = 0;
1647 if (static_key_false(&netstamp_needed))
1648 __net_timestamp(skb);
1651 #define net_timestamp_check(COND, SKB) \
1652 if (static_key_false(&netstamp_needed)) { \
1653 if ((COND) && !(SKB)->tstamp.tv64) \
1654 __net_timestamp(SKB); \
1657 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1661 if (!(dev->flags & IFF_UP))
1664 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1665 if (skb->len <= len)
1668 /* if TSO is enabled, we don't care about the length as the packet
1669 * could be forwarded without being segmented before
1671 if (skb_is_gso(skb))
1676 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1678 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1680 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1681 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1682 atomic_long_inc(&dev->rx_dropped);
1688 if (unlikely(!is_skb_forwardable(dev, skb))) {
1689 atomic_long_inc(&dev->rx_dropped);
1694 skb_scrub_packet(skb, true);
1695 skb->protocol = eth_type_trans(skb, dev);
1699 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1702 * dev_forward_skb - loopback an skb to another netif
1704 * @dev: destination network device
1705 * @skb: buffer to forward
1708 * NET_RX_SUCCESS (no congestion)
1709 * NET_RX_DROP (packet was dropped, but freed)
1711 * dev_forward_skb can be used for injecting an skb from the
1712 * start_xmit function of one device into the receive queue
1713 * of another device.
1715 * The receiving device may be in another namespace, so
1716 * we have to clear all information in the skb that could
1717 * impact namespace isolation.
1719 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1721 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1723 EXPORT_SYMBOL_GPL(dev_forward_skb);
1725 static inline int deliver_skb(struct sk_buff *skb,
1726 struct packet_type *pt_prev,
1727 struct net_device *orig_dev)
1729 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1731 atomic_inc(&skb->users);
1732 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1735 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1737 if (!ptype->af_packet_priv || !skb->sk)
1740 if (ptype->id_match)
1741 return ptype->id_match(ptype, skb->sk);
1742 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1749 * Support routine. Sends outgoing frames to any network
1750 * taps currently in use.
1753 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1755 struct packet_type *ptype;
1756 struct sk_buff *skb2 = NULL;
1757 struct packet_type *pt_prev = NULL;
1760 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1761 /* Never send packets back to the socket
1762 * they originated from - MvS (miquels@drinkel.ow.org)
1764 if ((ptype->dev == dev || !ptype->dev) &&
1765 (!skb_loop_sk(ptype, skb))) {
1767 deliver_skb(skb2, pt_prev, skb->dev);
1772 skb2 = skb_clone(skb, GFP_ATOMIC);
1776 net_timestamp_set(skb2);
1778 /* skb->nh should be correctly
1779 set by sender, so that the second statement is
1780 just protection against buggy protocols.
1782 skb_reset_mac_header(skb2);
1784 if (skb_network_header(skb2) < skb2->data ||
1785 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1786 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1787 ntohs(skb2->protocol),
1789 skb_reset_network_header(skb2);
1792 skb2->transport_header = skb2->network_header;
1793 skb2->pkt_type = PACKET_OUTGOING;
1798 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1803 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1804 * @dev: Network device
1805 * @txq: number of queues available
1807 * If real_num_tx_queues is changed the tc mappings may no longer be
1808 * valid. To resolve this verify the tc mapping remains valid and if
1809 * not NULL the mapping. With no priorities mapping to this
1810 * offset/count pair it will no longer be used. In the worst case TC0
1811 * is invalid nothing can be done so disable priority mappings. If is
1812 * expected that drivers will fix this mapping if they can before
1813 * calling netif_set_real_num_tx_queues.
1815 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1818 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1820 /* If TC0 is invalidated disable TC mapping */
1821 if (tc->offset + tc->count > txq) {
1822 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1827 /* Invalidated prio to tc mappings set to TC0 */
1828 for (i = 1; i < TC_BITMASK + 1; i++) {
1829 int q = netdev_get_prio_tc_map(dev, i);
1831 tc = &dev->tc_to_txq[q];
1832 if (tc->offset + tc->count > txq) {
1833 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1835 netdev_set_prio_tc_map(dev, i, 0);
1841 static DEFINE_MUTEX(xps_map_mutex);
1842 #define xmap_dereference(P) \
1843 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1845 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1848 struct xps_map *map = NULL;
1852 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1854 for (pos = 0; map && pos < map->len; pos++) {
1855 if (map->queues[pos] == index) {
1857 map->queues[pos] = map->queues[--map->len];
1859 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1860 kfree_rcu(map, rcu);
1870 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1872 struct xps_dev_maps *dev_maps;
1874 bool active = false;
1876 mutex_lock(&xps_map_mutex);
1877 dev_maps = xmap_dereference(dev->xps_maps);
1882 for_each_possible_cpu(cpu) {
1883 for (i = index; i < dev->num_tx_queues; i++) {
1884 if (!remove_xps_queue(dev_maps, cpu, i))
1887 if (i == dev->num_tx_queues)
1892 RCU_INIT_POINTER(dev->xps_maps, NULL);
1893 kfree_rcu(dev_maps, rcu);
1896 for (i = index; i < dev->num_tx_queues; i++)
1897 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1901 mutex_unlock(&xps_map_mutex);
1904 static struct xps_map *expand_xps_map(struct xps_map *map,
1907 struct xps_map *new_map;
1908 int alloc_len = XPS_MIN_MAP_ALLOC;
1911 for (pos = 0; map && pos < map->len; pos++) {
1912 if (map->queues[pos] != index)
1917 /* Need to add queue to this CPU's existing map */
1919 if (pos < map->alloc_len)
1922 alloc_len = map->alloc_len * 2;
1925 /* Need to allocate new map to store queue on this CPU's map */
1926 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1931 for (i = 0; i < pos; i++)
1932 new_map->queues[i] = map->queues[i];
1933 new_map->alloc_len = alloc_len;
1939 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1942 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1943 struct xps_map *map, *new_map;
1944 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1945 int cpu, numa_node_id = -2;
1946 bool active = false;
1948 mutex_lock(&xps_map_mutex);
1950 dev_maps = xmap_dereference(dev->xps_maps);
1952 /* allocate memory for queue storage */
1953 for_each_online_cpu(cpu) {
1954 if (!cpumask_test_cpu(cpu, mask))
1958 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1959 if (!new_dev_maps) {
1960 mutex_unlock(&xps_map_mutex);
1964 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1967 map = expand_xps_map(map, cpu, index);
1971 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1975 goto out_no_new_maps;
1977 for_each_possible_cpu(cpu) {
1978 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1979 /* add queue to CPU maps */
1982 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1983 while ((pos < map->len) && (map->queues[pos] != index))
1986 if (pos == map->len)
1987 map->queues[map->len++] = index;
1989 if (numa_node_id == -2)
1990 numa_node_id = cpu_to_node(cpu);
1991 else if (numa_node_id != cpu_to_node(cpu))
1994 } else if (dev_maps) {
1995 /* fill in the new device map from the old device map */
1996 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1997 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2002 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2004 /* Cleanup old maps */
2006 for_each_possible_cpu(cpu) {
2007 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2008 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2009 if (map && map != new_map)
2010 kfree_rcu(map, rcu);
2013 kfree_rcu(dev_maps, rcu);
2016 dev_maps = new_dev_maps;
2020 /* update Tx queue numa node */
2021 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2022 (numa_node_id >= 0) ? numa_node_id :
2028 /* removes queue from unused CPUs */
2029 for_each_possible_cpu(cpu) {
2030 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2033 if (remove_xps_queue(dev_maps, cpu, index))
2037 /* free map if not active */
2039 RCU_INIT_POINTER(dev->xps_maps, NULL);
2040 kfree_rcu(dev_maps, rcu);
2044 mutex_unlock(&xps_map_mutex);
2048 /* remove any maps that we added */
2049 for_each_possible_cpu(cpu) {
2050 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2051 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2053 if (new_map && new_map != map)
2057 mutex_unlock(&xps_map_mutex);
2059 kfree(new_dev_maps);
2062 EXPORT_SYMBOL(netif_set_xps_queue);
2066 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2067 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2069 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2073 if (txq < 1 || txq > dev->num_tx_queues)
2076 if (dev->reg_state == NETREG_REGISTERED ||
2077 dev->reg_state == NETREG_UNREGISTERING) {
2080 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2086 netif_setup_tc(dev, txq);
2088 if (txq < dev->real_num_tx_queues) {
2089 qdisc_reset_all_tx_gt(dev, txq);
2091 netif_reset_xps_queues_gt(dev, txq);
2096 dev->real_num_tx_queues = txq;
2099 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2103 * netif_set_real_num_rx_queues - set actual number of RX queues used
2104 * @dev: Network device
2105 * @rxq: Actual number of RX queues
2107 * This must be called either with the rtnl_lock held or before
2108 * registration of the net device. Returns 0 on success, or a
2109 * negative error code. If called before registration, it always
2112 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2116 if (rxq < 1 || rxq > dev->num_rx_queues)
2119 if (dev->reg_state == NETREG_REGISTERED) {
2122 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2128 dev->real_num_rx_queues = rxq;
2131 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2135 * netif_get_num_default_rss_queues - default number of RSS queues
2137 * This routine should set an upper limit on the number of RSS queues
2138 * used by default by multiqueue devices.
2140 int netif_get_num_default_rss_queues(void)
2142 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2144 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2146 static inline void __netif_reschedule(struct Qdisc *q)
2148 struct softnet_data *sd;
2149 unsigned long flags;
2151 local_irq_save(flags);
2152 sd = &__get_cpu_var(softnet_data);
2153 q->next_sched = NULL;
2154 *sd->output_queue_tailp = q;
2155 sd->output_queue_tailp = &q->next_sched;
2156 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2157 local_irq_restore(flags);
2160 void __netif_schedule(struct Qdisc *q)
2162 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2163 __netif_reschedule(q);
2165 EXPORT_SYMBOL(__netif_schedule);
2167 struct dev_kfree_skb_cb {
2168 enum skb_free_reason reason;
2171 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2173 return (struct dev_kfree_skb_cb *)skb->cb;
2176 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2178 unsigned long flags;
2180 if (likely(atomic_read(&skb->users) == 1)) {
2182 atomic_set(&skb->users, 0);
2183 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2186 get_kfree_skb_cb(skb)->reason = reason;
2187 local_irq_save(flags);
2188 skb->next = __this_cpu_read(softnet_data.completion_queue);
2189 __this_cpu_write(softnet_data.completion_queue, skb);
2190 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2191 local_irq_restore(flags);
2193 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2195 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2197 if (in_irq() || irqs_disabled())
2198 __dev_kfree_skb_irq(skb, reason);
2202 EXPORT_SYMBOL(__dev_kfree_skb_any);
2206 * netif_device_detach - mark device as removed
2207 * @dev: network device
2209 * Mark device as removed from system and therefore no longer available.
2211 void netif_device_detach(struct net_device *dev)
2213 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2214 netif_running(dev)) {
2215 netif_tx_stop_all_queues(dev);
2218 EXPORT_SYMBOL(netif_device_detach);
2221 * netif_device_attach - mark device as attached
2222 * @dev: network device
2224 * Mark device as attached from system and restart if needed.
2226 void netif_device_attach(struct net_device *dev)
2228 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2229 netif_running(dev)) {
2230 netif_tx_wake_all_queues(dev);
2231 __netdev_watchdog_up(dev);
2234 EXPORT_SYMBOL(netif_device_attach);
2236 static void skb_warn_bad_offload(const struct sk_buff *skb)
2238 static const netdev_features_t null_features = 0;
2239 struct net_device *dev = skb->dev;
2240 const char *driver = "";
2242 if (!net_ratelimit())
2245 if (dev && dev->dev.parent)
2246 driver = dev_driver_string(dev->dev.parent);
2248 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2249 "gso_type=%d ip_summed=%d\n",
2250 driver, dev ? &dev->features : &null_features,
2251 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2252 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2253 skb_shinfo(skb)->gso_type, skb->ip_summed);
2257 * Invalidate hardware checksum when packet is to be mangled, and
2258 * complete checksum manually on outgoing path.
2260 int skb_checksum_help(struct sk_buff *skb)
2263 int ret = 0, offset;
2265 if (skb->ip_summed == CHECKSUM_COMPLETE)
2266 goto out_set_summed;
2268 if (unlikely(skb_shinfo(skb)->gso_size)) {
2269 skb_warn_bad_offload(skb);
2273 /* Before computing a checksum, we should make sure no frag could
2274 * be modified by an external entity : checksum could be wrong.
2276 if (skb_has_shared_frag(skb)) {
2277 ret = __skb_linearize(skb);
2282 offset = skb_checksum_start_offset(skb);
2283 BUG_ON(offset >= skb_headlen(skb));
2284 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2286 offset += skb->csum_offset;
2287 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2289 if (skb_cloned(skb) &&
2290 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2291 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2296 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2298 skb->ip_summed = CHECKSUM_NONE;
2302 EXPORT_SYMBOL(skb_checksum_help);
2304 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2306 unsigned int vlan_depth = skb->mac_len;
2307 __be16 type = skb->protocol;
2309 /* Tunnel gso handlers can set protocol to ethernet. */
2310 if (type == htons(ETH_P_TEB)) {
2313 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2316 eth = (struct ethhdr *)skb_mac_header(skb);
2317 type = eth->h_proto;
2320 /* if skb->protocol is 802.1Q/AD then the header should already be
2321 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at
2322 * ETH_HLEN otherwise
2324 if (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2326 if (unlikely(WARN_ON(vlan_depth < VLAN_HLEN)))
2328 vlan_depth -= VLAN_HLEN;
2330 vlan_depth = ETH_HLEN;
2333 struct vlan_hdr *vh;
2335 if (unlikely(!pskb_may_pull(skb,
2336 vlan_depth + VLAN_HLEN)))
2339 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2340 type = vh->h_vlan_encapsulated_proto;
2341 vlan_depth += VLAN_HLEN;
2342 } while (type == htons(ETH_P_8021Q) ||
2343 type == htons(ETH_P_8021AD));
2346 *depth = vlan_depth;
2352 * skb_mac_gso_segment - mac layer segmentation handler.
2353 * @skb: buffer to segment
2354 * @features: features for the output path (see dev->features)
2356 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2357 netdev_features_t features)
2359 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2360 struct packet_offload *ptype;
2361 int vlan_depth = skb->mac_len;
2362 __be16 type = skb_network_protocol(skb, &vlan_depth);
2364 if (unlikely(!type))
2365 return ERR_PTR(-EINVAL);
2367 __skb_pull(skb, vlan_depth);
2370 list_for_each_entry_rcu(ptype, &offload_base, list) {
2371 if (ptype->type == type && ptype->callbacks.gso_segment) {
2372 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2375 err = ptype->callbacks.gso_send_check(skb);
2376 segs = ERR_PTR(err);
2377 if (err || skb_gso_ok(skb, features))
2379 __skb_push(skb, (skb->data -
2380 skb_network_header(skb)));
2382 segs = ptype->callbacks.gso_segment(skb, features);
2388 __skb_push(skb, skb->data - skb_mac_header(skb));
2392 EXPORT_SYMBOL(skb_mac_gso_segment);
2395 /* openvswitch calls this on rx path, so we need a different check.
2397 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2400 return skb->ip_summed != CHECKSUM_PARTIAL;
2402 return skb->ip_summed == CHECKSUM_NONE;
2406 * __skb_gso_segment - Perform segmentation on skb.
2407 * @skb: buffer to segment
2408 * @features: features for the output path (see dev->features)
2409 * @tx_path: whether it is called in TX path
2411 * This function segments the given skb and returns a list of segments.
2413 * It may return NULL if the skb requires no segmentation. This is
2414 * only possible when GSO is used for verifying header integrity.
2416 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2417 netdev_features_t features, bool tx_path)
2419 if (unlikely(skb_needs_check(skb, tx_path))) {
2422 skb_warn_bad_offload(skb);
2424 err = skb_cow_head(skb, 0);
2426 return ERR_PTR(err);
2429 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2430 SKB_GSO_CB(skb)->encap_level = 0;
2432 skb_reset_mac_header(skb);
2433 skb_reset_mac_len(skb);
2435 return skb_mac_gso_segment(skb, features);
2437 EXPORT_SYMBOL(__skb_gso_segment);
2439 /* Take action when hardware reception checksum errors are detected. */
2441 void netdev_rx_csum_fault(struct net_device *dev)
2443 if (net_ratelimit()) {
2444 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2448 EXPORT_SYMBOL(netdev_rx_csum_fault);
2451 /* Actually, we should eliminate this check as soon as we know, that:
2452 * 1. IOMMU is present and allows to map all the memory.
2453 * 2. No high memory really exists on this machine.
2456 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2458 #ifdef CONFIG_HIGHMEM
2460 if (!(dev->features & NETIF_F_HIGHDMA)) {
2461 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2462 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2463 if (PageHighMem(skb_frag_page(frag)))
2468 if (PCI_DMA_BUS_IS_PHYS) {
2469 struct device *pdev = dev->dev.parent;
2473 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2474 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2475 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2476 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2485 void (*destructor)(struct sk_buff *skb);
2488 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2490 static void dev_gso_skb_destructor(struct sk_buff *skb)
2492 struct dev_gso_cb *cb;
2494 kfree_skb_list(skb->next);
2497 cb = DEV_GSO_CB(skb);
2499 cb->destructor(skb);
2503 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2504 * @skb: buffer to segment
2505 * @features: device features as applicable to this skb
2507 * This function segments the given skb and stores the list of segments
2510 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2512 struct sk_buff *segs;
2514 segs = skb_gso_segment(skb, features);
2516 /* Verifying header integrity only. */
2521 return PTR_ERR(segs);
2524 DEV_GSO_CB(skb)->destructor = skb->destructor;
2525 skb->destructor = dev_gso_skb_destructor;
2530 /* If MPLS offload request, verify we are testing hardware MPLS features
2531 * instead of standard features for the netdev.
2533 #ifdef CONFIG_NET_MPLS_GSO
2534 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2535 netdev_features_t features,
2538 if (type == htons(ETH_P_MPLS_UC) || type == htons(ETH_P_MPLS_MC))
2539 features &= skb->dev->mpls_features;
2544 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2545 netdev_features_t features,
2552 static netdev_features_t harmonize_features(struct sk_buff *skb,
2553 netdev_features_t features)
2558 type = skb_network_protocol(skb, &tmp);
2559 features = net_mpls_features(skb, features, type);
2561 if (skb->ip_summed != CHECKSUM_NONE &&
2562 !can_checksum_protocol(features, type)) {
2563 features &= ~NETIF_F_ALL_CSUM;
2564 } else if (illegal_highdma(skb->dev, skb)) {
2565 features &= ~NETIF_F_SG;
2571 netdev_features_t netif_skb_features(struct sk_buff *skb)
2573 __be16 protocol = skb->protocol;
2574 netdev_features_t features = skb->dev->features;
2576 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2577 features &= ~NETIF_F_GSO_MASK;
2579 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2580 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2581 protocol = veh->h_vlan_encapsulated_proto;
2582 } else if (!vlan_tx_tag_present(skb)) {
2583 return harmonize_features(skb, features);
2586 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2587 NETIF_F_HW_VLAN_STAG_TX);
2589 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2590 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2591 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2592 NETIF_F_HW_VLAN_STAG_TX;
2594 return harmonize_features(skb, features);
2596 EXPORT_SYMBOL(netif_skb_features);
2598 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2599 struct netdev_queue *txq)
2601 const struct net_device_ops *ops = dev->netdev_ops;
2602 int rc = NETDEV_TX_OK;
2603 unsigned int skb_len;
2605 if (likely(!skb->next)) {
2606 netdev_features_t features;
2609 * If device doesn't need skb->dst, release it right now while
2610 * its hot in this cpu cache
2612 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2615 features = netif_skb_features(skb);
2617 if (vlan_tx_tag_present(skb) &&
2618 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2619 skb = __vlan_put_tag(skb, skb->vlan_proto,
2620 vlan_tx_tag_get(skb));
2627 /* If encapsulation offload request, verify we are testing
2628 * hardware encapsulation features instead of standard
2629 * features for the netdev
2631 if (skb->encapsulation)
2632 features &= dev->hw_enc_features;
2634 if (netif_needs_gso(skb, features)) {
2635 if (unlikely(dev_gso_segment(skb, features)))
2640 if (skb_needs_linearize(skb, features) &&
2641 __skb_linearize(skb))
2644 /* If packet is not checksummed and device does not
2645 * support checksumming for this protocol, complete
2646 * checksumming here.
2648 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2649 if (skb->encapsulation)
2650 skb_set_inner_transport_header(skb,
2651 skb_checksum_start_offset(skb));
2653 skb_set_transport_header(skb,
2654 skb_checksum_start_offset(skb));
2655 if (!(features & NETIF_F_ALL_CSUM) &&
2656 skb_checksum_help(skb))
2661 if (!list_empty(&ptype_all))
2662 dev_queue_xmit_nit(skb, dev);
2665 trace_net_dev_start_xmit(skb, dev);
2666 rc = ops->ndo_start_xmit(skb, dev);
2667 trace_net_dev_xmit(skb, rc, dev, skb_len);
2668 if (rc == NETDEV_TX_OK)
2669 txq_trans_update(txq);
2675 struct sk_buff *nskb = skb->next;
2677 skb->next = nskb->next;
2680 if (!list_empty(&ptype_all))
2681 dev_queue_xmit_nit(nskb, dev);
2683 skb_len = nskb->len;
2684 trace_net_dev_start_xmit(nskb, dev);
2685 rc = ops->ndo_start_xmit(nskb, dev);
2686 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2687 if (unlikely(rc != NETDEV_TX_OK)) {
2688 if (rc & ~NETDEV_TX_MASK)
2689 goto out_kfree_gso_skb;
2690 nskb->next = skb->next;
2694 txq_trans_update(txq);
2695 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2696 return NETDEV_TX_BUSY;
2697 } while (skb->next);
2700 if (likely(skb->next == NULL)) {
2701 skb->destructor = DEV_GSO_CB(skb)->destructor;
2710 EXPORT_SYMBOL_GPL(dev_hard_start_xmit);
2712 static void qdisc_pkt_len_init(struct sk_buff *skb)
2714 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2716 qdisc_skb_cb(skb)->pkt_len = skb->len;
2718 /* To get more precise estimation of bytes sent on wire,
2719 * we add to pkt_len the headers size of all segments
2721 if (shinfo->gso_size) {
2722 unsigned int hdr_len;
2723 u16 gso_segs = shinfo->gso_segs;
2725 /* mac layer + network layer */
2726 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2728 /* + transport layer */
2729 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2730 hdr_len += tcp_hdrlen(skb);
2732 hdr_len += sizeof(struct udphdr);
2734 if (shinfo->gso_type & SKB_GSO_DODGY)
2735 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2738 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2742 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2743 struct net_device *dev,
2744 struct netdev_queue *txq)
2746 spinlock_t *root_lock = qdisc_lock(q);
2750 qdisc_pkt_len_init(skb);
2751 qdisc_calculate_pkt_len(skb, q);
2753 * Heuristic to force contended enqueues to serialize on a
2754 * separate lock before trying to get qdisc main lock.
2755 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2756 * often and dequeue packets faster.
2758 contended = qdisc_is_running(q);
2759 if (unlikely(contended))
2760 spin_lock(&q->busylock);
2762 spin_lock(root_lock);
2763 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2766 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2767 qdisc_run_begin(q)) {
2769 * This is a work-conserving queue; there are no old skbs
2770 * waiting to be sent out; and the qdisc is not running -
2771 * xmit the skb directly.
2773 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2776 qdisc_bstats_update(q, skb);
2778 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2779 if (unlikely(contended)) {
2780 spin_unlock(&q->busylock);
2787 rc = NET_XMIT_SUCCESS;
2790 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2791 if (qdisc_run_begin(q)) {
2792 if (unlikely(contended)) {
2793 spin_unlock(&q->busylock);
2799 spin_unlock(root_lock);
2800 if (unlikely(contended))
2801 spin_unlock(&q->busylock);
2805 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2806 static void skb_update_prio(struct sk_buff *skb)
2808 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2810 if (!skb->priority && skb->sk && map) {
2811 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2813 if (prioidx < map->priomap_len)
2814 skb->priority = map->priomap[prioidx];
2818 #define skb_update_prio(skb)
2821 static DEFINE_PER_CPU(int, xmit_recursion);
2822 #define RECURSION_LIMIT 10
2825 * dev_loopback_xmit - loop back @skb
2826 * @skb: buffer to transmit
2828 int dev_loopback_xmit(struct sk_buff *skb)
2830 skb_reset_mac_header(skb);
2831 __skb_pull(skb, skb_network_offset(skb));
2832 skb->pkt_type = PACKET_LOOPBACK;
2833 skb->ip_summed = CHECKSUM_UNNECESSARY;
2834 WARN_ON(!skb_dst(skb));
2839 EXPORT_SYMBOL(dev_loopback_xmit);
2842 * __dev_queue_xmit - transmit a buffer
2843 * @skb: buffer to transmit
2844 * @accel_priv: private data used for L2 forwarding offload
2846 * Queue a buffer for transmission to a network device. The caller must
2847 * have set the device and priority and built the buffer before calling
2848 * this function. The function can be called from an interrupt.
2850 * A negative errno code is returned on a failure. A success does not
2851 * guarantee the frame will be transmitted as it may be dropped due
2852 * to congestion or traffic shaping.
2854 * -----------------------------------------------------------------------------------
2855 * I notice this method can also return errors from the queue disciplines,
2856 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2859 * Regardless of the return value, the skb is consumed, so it is currently
2860 * difficult to retry a send to this method. (You can bump the ref count
2861 * before sending to hold a reference for retry if you are careful.)
2863 * When calling this method, interrupts MUST be enabled. This is because
2864 * the BH enable code must have IRQs enabled so that it will not deadlock.
2867 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2869 struct net_device *dev = skb->dev;
2870 struct netdev_queue *txq;
2874 skb_reset_mac_header(skb);
2876 /* Disable soft irqs for various locks below. Also
2877 * stops preemption for RCU.
2881 skb_update_prio(skb);
2883 txq = netdev_pick_tx(dev, skb, accel_priv);
2884 q = rcu_dereference_bh(txq->qdisc);
2886 #ifdef CONFIG_NET_CLS_ACT
2887 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2889 trace_net_dev_queue(skb);
2891 rc = __dev_xmit_skb(skb, q, dev, txq);
2895 /* The device has no queue. Common case for software devices:
2896 loopback, all the sorts of tunnels...
2898 Really, it is unlikely that netif_tx_lock protection is necessary
2899 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2901 However, it is possible, that they rely on protection
2904 Check this and shot the lock. It is not prone from deadlocks.
2905 Either shot noqueue qdisc, it is even simpler 8)
2907 if (dev->flags & IFF_UP) {
2908 int cpu = smp_processor_id(); /* ok because BHs are off */
2910 if (txq->xmit_lock_owner != cpu) {
2912 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2913 goto recursion_alert;
2915 HARD_TX_LOCK(dev, txq, cpu);
2917 if (!netif_xmit_stopped(txq)) {
2918 __this_cpu_inc(xmit_recursion);
2919 rc = dev_hard_start_xmit(skb, dev, txq);
2920 __this_cpu_dec(xmit_recursion);
2921 if (dev_xmit_complete(rc)) {
2922 HARD_TX_UNLOCK(dev, txq);
2926 HARD_TX_UNLOCK(dev, txq);
2927 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2930 /* Recursion is detected! It is possible,
2934 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2940 rcu_read_unlock_bh();
2942 atomic_long_inc(&dev->tx_dropped);
2946 rcu_read_unlock_bh();
2950 int dev_queue_xmit(struct sk_buff *skb)
2952 return __dev_queue_xmit(skb, NULL);
2954 EXPORT_SYMBOL(dev_queue_xmit);
2956 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
2958 return __dev_queue_xmit(skb, accel_priv);
2960 EXPORT_SYMBOL(dev_queue_xmit_accel);
2963 /*=======================================================================
2965 =======================================================================*/
2967 int netdev_max_backlog __read_mostly = 1000;
2968 EXPORT_SYMBOL(netdev_max_backlog);
2970 int netdev_tstamp_prequeue __read_mostly = 1;
2971 int netdev_budget __read_mostly = 300;
2972 int weight_p __read_mostly = 64; /* old backlog weight */
2974 /* Called with irq disabled */
2975 static inline void ____napi_schedule(struct softnet_data *sd,
2976 struct napi_struct *napi)
2978 list_add_tail(&napi->poll_list, &sd->poll_list);
2979 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2984 /* One global table that all flow-based protocols share. */
2985 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2986 EXPORT_SYMBOL(rps_sock_flow_table);
2988 struct static_key rps_needed __read_mostly;
2990 static struct rps_dev_flow *
2991 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2992 struct rps_dev_flow *rflow, u16 next_cpu)
2994 if (next_cpu != RPS_NO_CPU) {
2995 #ifdef CONFIG_RFS_ACCEL
2996 struct netdev_rx_queue *rxqueue;
2997 struct rps_dev_flow_table *flow_table;
2998 struct rps_dev_flow *old_rflow;
3003 /* Should we steer this flow to a different hardware queue? */
3004 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3005 !(dev->features & NETIF_F_NTUPLE))
3007 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3008 if (rxq_index == skb_get_rx_queue(skb))
3011 rxqueue = dev->_rx + rxq_index;
3012 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3015 flow_id = skb_get_hash(skb) & flow_table->mask;
3016 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3017 rxq_index, flow_id);
3021 rflow = &flow_table->flows[flow_id];
3023 if (old_rflow->filter == rflow->filter)
3024 old_rflow->filter = RPS_NO_FILTER;
3028 per_cpu(softnet_data, next_cpu).input_queue_head;
3031 rflow->cpu = next_cpu;
3036 * get_rps_cpu is called from netif_receive_skb and returns the target
3037 * CPU from the RPS map of the receiving queue for a given skb.
3038 * rcu_read_lock must be held on entry.
3040 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3041 struct rps_dev_flow **rflowp)
3043 struct netdev_rx_queue *rxqueue;
3044 struct rps_map *map;
3045 struct rps_dev_flow_table *flow_table;
3046 struct rps_sock_flow_table *sock_flow_table;
3051 if (skb_rx_queue_recorded(skb)) {
3052 u16 index = skb_get_rx_queue(skb);
3053 if (unlikely(index >= dev->real_num_rx_queues)) {
3054 WARN_ONCE(dev->real_num_rx_queues > 1,
3055 "%s received packet on queue %u, but number "
3056 "of RX queues is %u\n",
3057 dev->name, index, dev->real_num_rx_queues);
3060 rxqueue = dev->_rx + index;
3064 map = rcu_dereference(rxqueue->rps_map);
3066 if (map->len == 1 &&
3067 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3068 tcpu = map->cpus[0];
3069 if (cpu_online(tcpu))
3073 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3077 skb_reset_network_header(skb);
3078 hash = skb_get_hash(skb);
3082 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3083 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3084 if (flow_table && sock_flow_table) {
3086 struct rps_dev_flow *rflow;
3088 rflow = &flow_table->flows[hash & flow_table->mask];
3091 next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask];
3094 * If the desired CPU (where last recvmsg was done) is
3095 * different from current CPU (one in the rx-queue flow
3096 * table entry), switch if one of the following holds:
3097 * - Current CPU is unset (equal to RPS_NO_CPU).
3098 * - Current CPU is offline.
3099 * - The current CPU's queue tail has advanced beyond the
3100 * last packet that was enqueued using this table entry.
3101 * This guarantees that all previous packets for the flow
3102 * have been dequeued, thus preserving in order delivery.
3104 if (unlikely(tcpu != next_cpu) &&
3105 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3106 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3107 rflow->last_qtail)) >= 0)) {
3109 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3112 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3120 tcpu = map->cpus[((u64) hash * map->len) >> 32];
3122 if (cpu_online(tcpu)) {
3132 #ifdef CONFIG_RFS_ACCEL
3135 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3136 * @dev: Device on which the filter was set
3137 * @rxq_index: RX queue index
3138 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3139 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3141 * Drivers that implement ndo_rx_flow_steer() should periodically call
3142 * this function for each installed filter and remove the filters for
3143 * which it returns %true.
3145 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3146 u32 flow_id, u16 filter_id)
3148 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3149 struct rps_dev_flow_table *flow_table;
3150 struct rps_dev_flow *rflow;
3155 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3156 if (flow_table && flow_id <= flow_table->mask) {
3157 rflow = &flow_table->flows[flow_id];
3158 cpu = ACCESS_ONCE(rflow->cpu);
3159 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3160 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3161 rflow->last_qtail) <
3162 (int)(10 * flow_table->mask)))
3168 EXPORT_SYMBOL(rps_may_expire_flow);
3170 #endif /* CONFIG_RFS_ACCEL */
3172 /* Called from hardirq (IPI) context */
3173 static void rps_trigger_softirq(void *data)
3175 struct softnet_data *sd = data;
3177 ____napi_schedule(sd, &sd->backlog);
3181 #endif /* CONFIG_RPS */
3184 * Check if this softnet_data structure is another cpu one
3185 * If yes, queue it to our IPI list and return 1
3188 static int rps_ipi_queued(struct softnet_data *sd)
3191 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3194 sd->rps_ipi_next = mysd->rps_ipi_list;
3195 mysd->rps_ipi_list = sd;
3197 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3200 #endif /* CONFIG_RPS */
3204 #ifdef CONFIG_NET_FLOW_LIMIT
3205 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3208 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3210 #ifdef CONFIG_NET_FLOW_LIMIT
3211 struct sd_flow_limit *fl;
3212 struct softnet_data *sd;
3213 unsigned int old_flow, new_flow;
3215 if (qlen < (netdev_max_backlog >> 1))
3218 sd = &__get_cpu_var(softnet_data);
3221 fl = rcu_dereference(sd->flow_limit);
3223 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3224 old_flow = fl->history[fl->history_head];
3225 fl->history[fl->history_head] = new_flow;
3228 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3230 if (likely(fl->buckets[old_flow]))
3231 fl->buckets[old_flow]--;
3233 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3245 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3246 * queue (may be a remote CPU queue).
3248 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3249 unsigned int *qtail)
3251 struct softnet_data *sd;
3252 unsigned long flags;
3255 sd = &per_cpu(softnet_data, cpu);
3257 local_irq_save(flags);
3260 qlen = skb_queue_len(&sd->input_pkt_queue);
3261 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3262 if (skb_queue_len(&sd->input_pkt_queue)) {
3264 __skb_queue_tail(&sd->input_pkt_queue, skb);
3265 input_queue_tail_incr_save(sd, qtail);
3267 local_irq_restore(flags);
3268 return NET_RX_SUCCESS;
3271 /* Schedule NAPI for backlog device
3272 * We can use non atomic operation since we own the queue lock
3274 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3275 if (!rps_ipi_queued(sd))
3276 ____napi_schedule(sd, &sd->backlog);
3284 local_irq_restore(flags);
3286 atomic_long_inc(&skb->dev->rx_dropped);
3291 static int netif_rx_internal(struct sk_buff *skb)
3295 net_timestamp_check(netdev_tstamp_prequeue, skb);
3297 trace_netif_rx(skb);
3299 if (static_key_false(&rps_needed)) {
3300 struct rps_dev_flow voidflow, *rflow = &voidflow;
3306 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3308 cpu = smp_processor_id();
3310 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3318 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3325 * netif_rx - post buffer to the network code
3326 * @skb: buffer to post
3328 * This function receives a packet from a device driver and queues it for
3329 * the upper (protocol) levels to process. It always succeeds. The buffer
3330 * may be dropped during processing for congestion control or by the
3334 * NET_RX_SUCCESS (no congestion)
3335 * NET_RX_DROP (packet was dropped)
3339 int netif_rx(struct sk_buff *skb)
3341 trace_netif_rx_entry(skb);
3343 return netif_rx_internal(skb);
3345 EXPORT_SYMBOL(netif_rx);
3347 int netif_rx_ni(struct sk_buff *skb)
3351 trace_netif_rx_ni_entry(skb);
3354 err = netif_rx_internal(skb);
3355 if (local_softirq_pending())
3361 EXPORT_SYMBOL(netif_rx_ni);
3363 static void net_tx_action(struct softirq_action *h)
3365 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3367 if (sd->completion_queue) {
3368 struct sk_buff *clist;
3370 local_irq_disable();
3371 clist = sd->completion_queue;
3372 sd->completion_queue = NULL;
3376 struct sk_buff *skb = clist;
3377 clist = clist->next;
3379 WARN_ON(atomic_read(&skb->users));
3380 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3381 trace_consume_skb(skb);
3383 trace_kfree_skb(skb, net_tx_action);
3388 if (sd->output_queue) {
3391 local_irq_disable();
3392 head = sd->output_queue;
3393 sd->output_queue = NULL;
3394 sd->output_queue_tailp = &sd->output_queue;
3398 struct Qdisc *q = head;
3399 spinlock_t *root_lock;
3401 head = head->next_sched;
3403 root_lock = qdisc_lock(q);
3404 if (spin_trylock(root_lock)) {
3405 smp_mb__before_atomic();
3406 clear_bit(__QDISC_STATE_SCHED,
3409 spin_unlock(root_lock);
3411 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3413 __netif_reschedule(q);
3415 smp_mb__before_atomic();
3416 clear_bit(__QDISC_STATE_SCHED,
3424 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3425 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3426 /* This hook is defined here for ATM LANE */
3427 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3428 unsigned char *addr) __read_mostly;
3429 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3432 #ifdef CONFIG_NET_CLS_ACT
3433 /* TODO: Maybe we should just force sch_ingress to be compiled in
3434 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3435 * a compare and 2 stores extra right now if we dont have it on
3436 * but have CONFIG_NET_CLS_ACT
3437 * NOTE: This doesn't stop any functionality; if you dont have
3438 * the ingress scheduler, you just can't add policies on ingress.
3441 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3443 struct net_device *dev = skb->dev;
3444 u32 ttl = G_TC_RTTL(skb->tc_verd);
3445 int result = TC_ACT_OK;
3448 if (unlikely(MAX_RED_LOOP < ttl++)) {
3449 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3450 skb->skb_iif, dev->ifindex);
3454 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3455 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3458 if (q != &noop_qdisc) {
3459 spin_lock(qdisc_lock(q));
3460 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3461 result = qdisc_enqueue_root(skb, q);
3462 spin_unlock(qdisc_lock(q));
3468 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3469 struct packet_type **pt_prev,
3470 int *ret, struct net_device *orig_dev)
3472 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3474 if (!rxq || rxq->qdisc == &noop_qdisc)
3478 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3482 switch (ing_filter(skb, rxq)) {
3496 * netdev_rx_handler_register - register receive handler
3497 * @dev: device to register a handler for
3498 * @rx_handler: receive handler to register
3499 * @rx_handler_data: data pointer that is used by rx handler
3501 * Register a receive handler for a device. This handler will then be
3502 * called from __netif_receive_skb. A negative errno code is returned
3505 * The caller must hold the rtnl_mutex.
3507 * For a general description of rx_handler, see enum rx_handler_result.
3509 int netdev_rx_handler_register(struct net_device *dev,
3510 rx_handler_func_t *rx_handler,
3511 void *rx_handler_data)
3515 if (dev->rx_handler)
3518 /* Note: rx_handler_data must be set before rx_handler */
3519 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3520 rcu_assign_pointer(dev->rx_handler, rx_handler);
3524 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3527 * netdev_rx_handler_unregister - unregister receive handler
3528 * @dev: device to unregister a handler from
3530 * Unregister a receive handler from a device.
3532 * The caller must hold the rtnl_mutex.
3534 void netdev_rx_handler_unregister(struct net_device *dev)
3538 RCU_INIT_POINTER(dev->rx_handler, NULL);
3539 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3540 * section has a guarantee to see a non NULL rx_handler_data
3544 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3546 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3549 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3550 * the special handling of PFMEMALLOC skbs.
3552 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3554 switch (skb->protocol) {
3555 case htons(ETH_P_ARP):
3556 case htons(ETH_P_IP):
3557 case htons(ETH_P_IPV6):
3558 case htons(ETH_P_8021Q):
3559 case htons(ETH_P_8021AD):
3566 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3568 struct packet_type *ptype, *pt_prev;
3569 rx_handler_func_t *rx_handler;
3570 struct net_device *orig_dev;
3571 struct net_device *null_or_dev;
3572 bool deliver_exact = false;
3573 int ret = NET_RX_DROP;
3576 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3578 trace_netif_receive_skb(skb);
3580 orig_dev = skb->dev;
3582 skb_reset_network_header(skb);
3583 if (!skb_transport_header_was_set(skb))
3584 skb_reset_transport_header(skb);
3585 skb_reset_mac_len(skb);
3592 skb->skb_iif = skb->dev->ifindex;
3594 __this_cpu_inc(softnet_data.processed);
3596 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3597 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3598 skb = vlan_untag(skb);
3603 #ifdef CONFIG_NET_CLS_ACT
3604 if (skb->tc_verd & TC_NCLS) {
3605 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3613 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3614 if (!ptype->dev || ptype->dev == skb->dev) {
3616 ret = deliver_skb(skb, pt_prev, orig_dev);
3622 #ifdef CONFIG_NET_CLS_ACT
3623 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3629 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3632 if (vlan_tx_tag_present(skb)) {
3634 ret = deliver_skb(skb, pt_prev, orig_dev);
3637 if (vlan_do_receive(&skb))
3639 else if (unlikely(!skb))
3643 rx_handler = rcu_dereference(skb->dev->rx_handler);
3646 ret = deliver_skb(skb, pt_prev, orig_dev);
3649 switch (rx_handler(&skb)) {
3650 case RX_HANDLER_CONSUMED:
3651 ret = NET_RX_SUCCESS;
3653 case RX_HANDLER_ANOTHER:
3655 case RX_HANDLER_EXACT:
3656 deliver_exact = true;
3657 case RX_HANDLER_PASS:
3664 if (unlikely(vlan_tx_tag_present(skb))) {
3665 if (vlan_tx_tag_get_id(skb))
3666 skb->pkt_type = PACKET_OTHERHOST;
3667 /* Note: we might in the future use prio bits
3668 * and set skb->priority like in vlan_do_receive()
3669 * For the time being, just ignore Priority Code Point
3674 /* deliver only exact match when indicated */
3675 null_or_dev = deliver_exact ? skb->dev : NULL;
3677 type = skb->protocol;
3678 list_for_each_entry_rcu(ptype,
3679 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3680 if (ptype->type == type &&
3681 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3682 ptype->dev == orig_dev)) {
3684 ret = deliver_skb(skb, pt_prev, orig_dev);
3690 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3693 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3696 atomic_long_inc(&skb->dev->rx_dropped);
3698 /* Jamal, now you will not able to escape explaining
3699 * me how you were going to use this. :-)
3709 static int __netif_receive_skb(struct sk_buff *skb)
3713 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3714 unsigned long pflags = current->flags;
3717 * PFMEMALLOC skbs are special, they should
3718 * - be delivered to SOCK_MEMALLOC sockets only
3719 * - stay away from userspace
3720 * - have bounded memory usage
3722 * Use PF_MEMALLOC as this saves us from propagating the allocation
3723 * context down to all allocation sites.
3725 current->flags |= PF_MEMALLOC;
3726 ret = __netif_receive_skb_core(skb, true);
3727 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3729 ret = __netif_receive_skb_core(skb, false);
3734 static int netif_receive_skb_internal(struct sk_buff *skb)
3736 net_timestamp_check(netdev_tstamp_prequeue, skb);
3738 if (skb_defer_rx_timestamp(skb))
3739 return NET_RX_SUCCESS;
3742 if (static_key_false(&rps_needed)) {
3743 struct rps_dev_flow voidflow, *rflow = &voidflow;
3748 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3751 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3758 return __netif_receive_skb(skb);
3762 * netif_receive_skb - process receive buffer from network
3763 * @skb: buffer to process
3765 * netif_receive_skb() is the main receive data processing function.
3766 * It always succeeds. The buffer may be dropped during processing
3767 * for congestion control or by the protocol layers.
3769 * This function may only be called from softirq context and interrupts
3770 * should be enabled.
3772 * Return values (usually ignored):
3773 * NET_RX_SUCCESS: no congestion
3774 * NET_RX_DROP: packet was dropped
3776 int netif_receive_skb(struct sk_buff *skb)
3778 trace_netif_receive_skb_entry(skb);
3780 return netif_receive_skb_internal(skb);
3782 EXPORT_SYMBOL(netif_receive_skb);
3784 /* Network device is going away, flush any packets still pending
3785 * Called with irqs disabled.
3787 static void flush_backlog(void *arg)
3789 struct net_device *dev = arg;
3790 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3791 struct sk_buff *skb, *tmp;
3794 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3795 if (skb->dev == dev) {
3796 __skb_unlink(skb, &sd->input_pkt_queue);
3798 input_queue_head_incr(sd);
3803 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3804 if (skb->dev == dev) {
3805 __skb_unlink(skb, &sd->process_queue);
3807 input_queue_head_incr(sd);
3812 static int napi_gro_complete(struct sk_buff *skb)
3814 struct packet_offload *ptype;
3815 __be16 type = skb->protocol;
3816 struct list_head *head = &offload_base;
3819 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3821 if (NAPI_GRO_CB(skb)->count == 1) {
3822 skb_shinfo(skb)->gso_size = 0;
3827 list_for_each_entry_rcu(ptype, head, list) {
3828 if (ptype->type != type || !ptype->callbacks.gro_complete)
3831 err = ptype->callbacks.gro_complete(skb, 0);
3837 WARN_ON(&ptype->list == head);
3839 return NET_RX_SUCCESS;
3843 return netif_receive_skb_internal(skb);
3846 /* napi->gro_list contains packets ordered by age.
3847 * youngest packets at the head of it.
3848 * Complete skbs in reverse order to reduce latencies.
3850 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3852 struct sk_buff *skb, *prev = NULL;
3854 /* scan list and build reverse chain */
3855 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3860 for (skb = prev; skb; skb = prev) {
3863 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3867 napi_gro_complete(skb);
3871 napi->gro_list = NULL;
3873 EXPORT_SYMBOL(napi_gro_flush);
3875 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3878 unsigned int maclen = skb->dev->hard_header_len;
3879 u32 hash = skb_get_hash_raw(skb);
3881 for (p = napi->gro_list; p; p = p->next) {
3882 unsigned long diffs;
3884 NAPI_GRO_CB(p)->flush = 0;
3886 if (hash != skb_get_hash_raw(p)) {
3887 NAPI_GRO_CB(p)->same_flow = 0;
3891 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3892 diffs |= p->vlan_tci ^ skb->vlan_tci;
3893 if (maclen == ETH_HLEN)
3894 diffs |= compare_ether_header(skb_mac_header(p),
3895 skb_mac_header(skb));
3897 diffs = memcmp(skb_mac_header(p),
3898 skb_mac_header(skb),
3900 NAPI_GRO_CB(p)->same_flow = !diffs;
3904 static void skb_gro_reset_offset(struct sk_buff *skb)
3906 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3907 const skb_frag_t *frag0 = &pinfo->frags[0];
3909 NAPI_GRO_CB(skb)->data_offset = 0;
3910 NAPI_GRO_CB(skb)->frag0 = NULL;
3911 NAPI_GRO_CB(skb)->frag0_len = 0;
3913 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3915 !PageHighMem(skb_frag_page(frag0))) {
3916 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3917 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3921 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
3923 struct skb_shared_info *pinfo = skb_shinfo(skb);
3925 BUG_ON(skb->end - skb->tail < grow);
3927 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3929 skb->data_len -= grow;
3932 pinfo->frags[0].page_offset += grow;
3933 skb_frag_size_sub(&pinfo->frags[0], grow);
3935 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
3936 skb_frag_unref(skb, 0);
3937 memmove(pinfo->frags, pinfo->frags + 1,
3938 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
3942 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3944 struct sk_buff **pp = NULL;
3945 struct packet_offload *ptype;
3946 __be16 type = skb->protocol;
3947 struct list_head *head = &offload_base;
3949 enum gro_result ret;
3952 if (!(skb->dev->features & NETIF_F_GRO))
3955 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3958 gro_list_prepare(napi, skb);
3959 NAPI_GRO_CB(skb)->csum = skb->csum; /* Needed for CHECKSUM_COMPLETE */
3962 list_for_each_entry_rcu(ptype, head, list) {
3963 if (ptype->type != type || !ptype->callbacks.gro_receive)
3966 skb_set_network_header(skb, skb_gro_offset(skb));
3967 skb_reset_mac_len(skb);
3968 NAPI_GRO_CB(skb)->same_flow = 0;
3969 NAPI_GRO_CB(skb)->flush = 0;
3970 NAPI_GRO_CB(skb)->free = 0;
3971 NAPI_GRO_CB(skb)->udp_mark = 0;
3973 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3978 if (&ptype->list == head)
3981 same_flow = NAPI_GRO_CB(skb)->same_flow;
3982 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3985 struct sk_buff *nskb = *pp;
3989 napi_gro_complete(nskb);
3996 if (NAPI_GRO_CB(skb)->flush)
3999 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4000 struct sk_buff *nskb = napi->gro_list;
4002 /* locate the end of the list to select the 'oldest' flow */
4003 while (nskb->next) {
4009 napi_gro_complete(nskb);
4013 NAPI_GRO_CB(skb)->count = 1;
4014 NAPI_GRO_CB(skb)->age = jiffies;
4015 NAPI_GRO_CB(skb)->last = skb;
4016 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4017 skb->next = napi->gro_list;
4018 napi->gro_list = skb;
4022 grow = skb_gro_offset(skb) - skb_headlen(skb);
4024 gro_pull_from_frag0(skb, grow);
4033 struct packet_offload *gro_find_receive_by_type(__be16 type)
4035 struct list_head *offload_head = &offload_base;
4036 struct packet_offload *ptype;
4038 list_for_each_entry_rcu(ptype, offload_head, list) {
4039 if (ptype->type != type || !ptype->callbacks.gro_receive)
4045 EXPORT_SYMBOL(gro_find_receive_by_type);
4047 struct packet_offload *gro_find_complete_by_type(__be16 type)
4049 struct list_head *offload_head = &offload_base;
4050 struct packet_offload *ptype;
4052 list_for_each_entry_rcu(ptype, offload_head, list) {
4053 if (ptype->type != type || !ptype->callbacks.gro_complete)
4059 EXPORT_SYMBOL(gro_find_complete_by_type);
4061 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4065 if (netif_receive_skb_internal(skb))
4073 case GRO_MERGED_FREE:
4074 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4075 kmem_cache_free(skbuff_head_cache, skb);
4088 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4090 trace_napi_gro_receive_entry(skb);
4092 skb_gro_reset_offset(skb);
4094 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4096 EXPORT_SYMBOL(napi_gro_receive);
4098 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4100 __skb_pull(skb, skb_headlen(skb));
4101 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4102 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4104 skb->dev = napi->dev;
4106 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4111 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4113 struct sk_buff *skb = napi->skb;
4116 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4121 EXPORT_SYMBOL(napi_get_frags);
4123 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4124 struct sk_buff *skb,
4130 __skb_push(skb, ETH_HLEN);
4131 skb->protocol = eth_type_trans(skb, skb->dev);
4132 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4137 case GRO_MERGED_FREE:
4138 napi_reuse_skb(napi, skb);
4148 /* Upper GRO stack assumes network header starts at gro_offset=0
4149 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4150 * We copy ethernet header into skb->data to have a common layout.
4152 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4154 struct sk_buff *skb = napi->skb;
4155 const struct ethhdr *eth;
4156 unsigned int hlen = sizeof(*eth);
4160 skb_reset_mac_header(skb);
4161 skb_gro_reset_offset(skb);
4163 eth = skb_gro_header_fast(skb, 0);
4164 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4165 eth = skb_gro_header_slow(skb, hlen, 0);
4166 if (unlikely(!eth)) {
4167 napi_reuse_skb(napi, skb);
4171 gro_pull_from_frag0(skb, hlen);
4172 NAPI_GRO_CB(skb)->frag0 += hlen;
4173 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4175 __skb_pull(skb, hlen);
4178 * This works because the only protocols we care about don't require
4180 * We'll fix it up properly in napi_frags_finish()
4182 skb->protocol = eth->h_proto;
4187 gro_result_t napi_gro_frags(struct napi_struct *napi)
4189 struct sk_buff *skb = napi_frags_skb(napi);
4194 trace_napi_gro_frags_entry(skb);
4196 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4198 EXPORT_SYMBOL(napi_gro_frags);
4201 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4202 * Note: called with local irq disabled, but exits with local irq enabled.
4204 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4207 struct softnet_data *remsd = sd->rps_ipi_list;
4210 sd->rps_ipi_list = NULL;
4214 /* Send pending IPI's to kick RPS processing on remote cpus. */
4216 struct softnet_data *next = remsd->rps_ipi_next;
4218 if (cpu_online(remsd->cpu))
4219 smp_call_function_single_async(remsd->cpu,
4228 static int process_backlog(struct napi_struct *napi, int quota)
4231 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4234 /* Check if we have pending ipi, its better to send them now,
4235 * not waiting net_rx_action() end.
4237 if (sd->rps_ipi_list) {
4238 local_irq_disable();
4239 net_rps_action_and_irq_enable(sd);
4242 napi->weight = weight_p;
4243 local_irq_disable();
4245 struct sk_buff *skb;
4247 while ((skb = __skb_dequeue(&sd->process_queue))) {
4249 __netif_receive_skb(skb);
4250 local_irq_disable();
4251 input_queue_head_incr(sd);
4252 if (++work >= quota) {
4259 if (skb_queue_empty(&sd->input_pkt_queue)) {
4261 * Inline a custom version of __napi_complete().
4262 * only current cpu owns and manipulates this napi,
4263 * and NAPI_STATE_SCHED is the only possible flag set
4265 * We can use a plain write instead of clear_bit(),
4266 * and we dont need an smp_mb() memory barrier.
4268 list_del(&napi->poll_list);
4275 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4276 &sd->process_queue);
4285 * __napi_schedule - schedule for receive
4286 * @n: entry to schedule
4288 * The entry's receive function will be scheduled to run
4290 void __napi_schedule(struct napi_struct *n)
4292 unsigned long flags;
4294 local_irq_save(flags);
4295 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4296 local_irq_restore(flags);
4298 EXPORT_SYMBOL(__napi_schedule);
4300 void __napi_complete(struct napi_struct *n)
4302 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4303 BUG_ON(n->gro_list);
4305 list_del(&n->poll_list);
4306 smp_mb__before_atomic();
4307 clear_bit(NAPI_STATE_SCHED, &n->state);
4309 EXPORT_SYMBOL(__napi_complete);
4311 void napi_complete(struct napi_struct *n)
4313 unsigned long flags;
4316 * don't let napi dequeue from the cpu poll list
4317 * just in case its running on a different cpu
4319 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4322 napi_gro_flush(n, false);
4323 local_irq_save(flags);
4325 local_irq_restore(flags);
4327 EXPORT_SYMBOL(napi_complete);
4329 /* must be called under rcu_read_lock(), as we dont take a reference */
4330 struct napi_struct *napi_by_id(unsigned int napi_id)
4332 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4333 struct napi_struct *napi;
4335 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4336 if (napi->napi_id == napi_id)
4341 EXPORT_SYMBOL_GPL(napi_by_id);
4343 void napi_hash_add(struct napi_struct *napi)
4345 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4347 spin_lock(&napi_hash_lock);
4349 /* 0 is not a valid id, we also skip an id that is taken
4350 * we expect both events to be extremely rare
4353 while (!napi->napi_id) {
4354 napi->napi_id = ++napi_gen_id;
4355 if (napi_by_id(napi->napi_id))
4359 hlist_add_head_rcu(&napi->napi_hash_node,
4360 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4362 spin_unlock(&napi_hash_lock);
4365 EXPORT_SYMBOL_GPL(napi_hash_add);
4367 /* Warning : caller is responsible to make sure rcu grace period
4368 * is respected before freeing memory containing @napi
4370 void napi_hash_del(struct napi_struct *napi)
4372 spin_lock(&napi_hash_lock);
4374 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4375 hlist_del_rcu(&napi->napi_hash_node);
4377 spin_unlock(&napi_hash_lock);
4379 EXPORT_SYMBOL_GPL(napi_hash_del);
4381 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4382 int (*poll)(struct napi_struct *, int), int weight)
4384 INIT_LIST_HEAD(&napi->poll_list);
4385 napi->gro_count = 0;
4386 napi->gro_list = NULL;
4389 if (weight > NAPI_POLL_WEIGHT)
4390 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4392 napi->weight = weight;
4393 list_add(&napi->dev_list, &dev->napi_list);
4395 #ifdef CONFIG_NETPOLL
4396 spin_lock_init(&napi->poll_lock);
4397 napi->poll_owner = -1;
4399 set_bit(NAPI_STATE_SCHED, &napi->state);
4401 EXPORT_SYMBOL(netif_napi_add);
4403 void netif_napi_del(struct napi_struct *napi)
4405 list_del_init(&napi->dev_list);
4406 napi_free_frags(napi);
4408 kfree_skb_list(napi->gro_list);
4409 napi->gro_list = NULL;
4410 napi->gro_count = 0;
4412 EXPORT_SYMBOL(netif_napi_del);
4414 static void net_rx_action(struct softirq_action *h)
4416 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4417 unsigned long time_limit = jiffies + 2;
4418 int budget = netdev_budget;
4421 local_irq_disable();
4423 while (!list_empty(&sd->poll_list)) {
4424 struct napi_struct *n;
4427 /* If softirq window is exhuasted then punt.
4428 * Allow this to run for 2 jiffies since which will allow
4429 * an average latency of 1.5/HZ.
4431 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4436 /* Even though interrupts have been re-enabled, this
4437 * access is safe because interrupts can only add new
4438 * entries to the tail of this list, and only ->poll()
4439 * calls can remove this head entry from the list.
4441 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4443 have = netpoll_poll_lock(n);
4447 /* This NAPI_STATE_SCHED test is for avoiding a race
4448 * with netpoll's poll_napi(). Only the entity which
4449 * obtains the lock and sees NAPI_STATE_SCHED set will
4450 * actually make the ->poll() call. Therefore we avoid
4451 * accidentally calling ->poll() when NAPI is not scheduled.
4454 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4455 work = n->poll(n, weight);
4459 WARN_ON_ONCE(work > weight);
4463 local_irq_disable();
4465 /* Drivers must not modify the NAPI state if they
4466 * consume the entire weight. In such cases this code
4467 * still "owns" the NAPI instance and therefore can
4468 * move the instance around on the list at-will.
4470 if (unlikely(work == weight)) {
4471 if (unlikely(napi_disable_pending(n))) {
4474 local_irq_disable();
4477 /* flush too old packets
4478 * If HZ < 1000, flush all packets.
4481 napi_gro_flush(n, HZ >= 1000);
4482 local_irq_disable();
4484 list_move_tail(&n->poll_list, &sd->poll_list);
4488 netpoll_poll_unlock(have);
4491 net_rps_action_and_irq_enable(sd);
4493 #ifdef CONFIG_NET_DMA
4495 * There may not be any more sk_buffs coming right now, so push
4496 * any pending DMA copies to hardware
4498 dma_issue_pending_all();
4505 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4509 struct netdev_adjacent {
4510 struct net_device *dev;
4512 /* upper master flag, there can only be one master device per list */
4515 /* counter for the number of times this device was added to us */
4518 /* private field for the users */
4521 struct list_head list;
4522 struct rcu_head rcu;
4525 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4526 struct net_device *adj_dev,
4527 struct list_head *adj_list)
4529 struct netdev_adjacent *adj;
4531 list_for_each_entry(adj, adj_list, list) {
4532 if (adj->dev == adj_dev)
4539 * netdev_has_upper_dev - Check if device is linked to an upper device
4541 * @upper_dev: upper device to check
4543 * Find out if a device is linked to specified upper device and return true
4544 * in case it is. Note that this checks only immediate upper device,
4545 * not through a complete stack of devices. The caller must hold the RTNL lock.
4547 bool netdev_has_upper_dev(struct net_device *dev,
4548 struct net_device *upper_dev)
4552 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4554 EXPORT_SYMBOL(netdev_has_upper_dev);
4557 * netdev_has_any_upper_dev - Check if device is linked to some device
4560 * Find out if a device is linked to an upper device and return true in case
4561 * it is. The caller must hold the RTNL lock.
4563 static bool netdev_has_any_upper_dev(struct net_device *dev)
4567 return !list_empty(&dev->all_adj_list.upper);
4571 * netdev_master_upper_dev_get - Get master upper device
4574 * Find a master upper device and return pointer to it or NULL in case
4575 * it's not there. The caller must hold the RTNL lock.
4577 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4579 struct netdev_adjacent *upper;
4583 if (list_empty(&dev->adj_list.upper))
4586 upper = list_first_entry(&dev->adj_list.upper,
4587 struct netdev_adjacent, list);
4588 if (likely(upper->master))
4592 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4594 void *netdev_adjacent_get_private(struct list_head *adj_list)
4596 struct netdev_adjacent *adj;
4598 adj = list_entry(adj_list, struct netdev_adjacent, list);
4600 return adj->private;
4602 EXPORT_SYMBOL(netdev_adjacent_get_private);
4605 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4607 * @iter: list_head ** of the current position
4609 * Gets the next device from the dev's upper list, starting from iter
4610 * position. The caller must hold RCU read lock.
4612 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4613 struct list_head **iter)
4615 struct netdev_adjacent *upper;
4617 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4619 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4621 if (&upper->list == &dev->adj_list.upper)
4624 *iter = &upper->list;
4628 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4631 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4633 * @iter: list_head ** of the current position
4635 * Gets the next device from the dev's upper list, starting from iter
4636 * position. The caller must hold RCU read lock.
4638 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4639 struct list_head **iter)
4641 struct netdev_adjacent *upper;
4643 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4645 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4647 if (&upper->list == &dev->all_adj_list.upper)
4650 *iter = &upper->list;
4654 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4657 * netdev_lower_get_next_private - Get the next ->private from the
4658 * lower neighbour list
4660 * @iter: list_head ** of the current position
4662 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4663 * list, starting from iter position. The caller must hold either hold the
4664 * RTNL lock or its own locking that guarantees that the neighbour lower
4665 * list will remain unchainged.
4667 void *netdev_lower_get_next_private(struct net_device *dev,
4668 struct list_head **iter)
4670 struct netdev_adjacent *lower;
4672 lower = list_entry(*iter, struct netdev_adjacent, list);
4674 if (&lower->list == &dev->adj_list.lower)
4677 *iter = lower->list.next;
4679 return lower->private;
4681 EXPORT_SYMBOL(netdev_lower_get_next_private);
4684 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4685 * lower neighbour list, RCU
4688 * @iter: list_head ** of the current position
4690 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4691 * list, starting from iter position. The caller must hold RCU read lock.
4693 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4694 struct list_head **iter)
4696 struct netdev_adjacent *lower;
4698 WARN_ON_ONCE(!rcu_read_lock_held());
4700 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4702 if (&lower->list == &dev->adj_list.lower)
4705 *iter = &lower->list;
4707 return lower->private;
4709 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4712 * netdev_lower_get_next - Get the next device from the lower neighbour
4715 * @iter: list_head ** of the current position
4717 * Gets the next netdev_adjacent from the dev's lower neighbour
4718 * list, starting from iter position. The caller must hold RTNL lock or
4719 * its own locking that guarantees that the neighbour lower
4720 * list will remain unchainged.
4722 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4724 struct netdev_adjacent *lower;
4726 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4728 if (&lower->list == &dev->adj_list.lower)
4731 *iter = &lower->list;
4735 EXPORT_SYMBOL(netdev_lower_get_next);
4738 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4739 * lower neighbour list, RCU
4743 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4744 * list. The caller must hold RCU read lock.
4746 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4748 struct netdev_adjacent *lower;
4750 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4751 struct netdev_adjacent, list);
4753 return lower->private;
4756 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4759 * netdev_master_upper_dev_get_rcu - Get master upper device
4762 * Find a master upper device and return pointer to it or NULL in case
4763 * it's not there. The caller must hold the RCU read lock.
4765 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4767 struct netdev_adjacent *upper;
4769 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4770 struct netdev_adjacent, list);
4771 if (upper && likely(upper->master))
4775 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4777 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4778 struct net_device *adj_dev,
4779 struct list_head *dev_list)
4781 char linkname[IFNAMSIZ+7];
4782 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4783 "upper_%s" : "lower_%s", adj_dev->name);
4784 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4787 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4789 struct list_head *dev_list)
4791 char linkname[IFNAMSIZ+7];
4792 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4793 "upper_%s" : "lower_%s", name);
4794 sysfs_remove_link(&(dev->dev.kobj), linkname);
4797 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4798 (dev_list == &dev->adj_list.upper || \
4799 dev_list == &dev->adj_list.lower)
4801 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4802 struct net_device *adj_dev,
4803 struct list_head *dev_list,
4804 void *private, bool master)
4806 struct netdev_adjacent *adj;
4809 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4816 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4821 adj->master = master;
4823 adj->private = private;
4826 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4827 adj_dev->name, dev->name, adj_dev->name);
4829 if (netdev_adjacent_is_neigh_list(dev, dev_list)) {
4830 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
4835 /* Ensure that master link is always the first item in list. */
4837 ret = sysfs_create_link(&(dev->dev.kobj),
4838 &(adj_dev->dev.kobj), "master");
4840 goto remove_symlinks;
4842 list_add_rcu(&adj->list, dev_list);
4844 list_add_tail_rcu(&adj->list, dev_list);
4850 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4851 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4859 static void __netdev_adjacent_dev_remove(struct net_device *dev,
4860 struct net_device *adj_dev,
4861 struct list_head *dev_list)
4863 struct netdev_adjacent *adj;
4865 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4868 pr_err("tried to remove device %s from %s\n",
4869 dev->name, adj_dev->name);
4873 if (adj->ref_nr > 1) {
4874 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4881 sysfs_remove_link(&(dev->dev.kobj), "master");
4883 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4884 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4886 list_del_rcu(&adj->list);
4887 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4888 adj_dev->name, dev->name, adj_dev->name);
4890 kfree_rcu(adj, rcu);
4893 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4894 struct net_device *upper_dev,
4895 struct list_head *up_list,
4896 struct list_head *down_list,
4897 void *private, bool master)
4901 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4906 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4909 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4916 static int __netdev_adjacent_dev_link(struct net_device *dev,
4917 struct net_device *upper_dev)
4919 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4920 &dev->all_adj_list.upper,
4921 &upper_dev->all_adj_list.lower,
4925 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4926 struct net_device *upper_dev,
4927 struct list_head *up_list,
4928 struct list_head *down_list)
4930 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4931 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4934 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
4935 struct net_device *upper_dev)
4937 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4938 &dev->all_adj_list.upper,
4939 &upper_dev->all_adj_list.lower);
4942 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4943 struct net_device *upper_dev,
4944 void *private, bool master)
4946 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4951 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4952 &dev->adj_list.upper,
4953 &upper_dev->adj_list.lower,
4956 __netdev_adjacent_dev_unlink(dev, upper_dev);
4963 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4964 struct net_device *upper_dev)
4966 __netdev_adjacent_dev_unlink(dev, upper_dev);
4967 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4968 &dev->adj_list.upper,
4969 &upper_dev->adj_list.lower);
4972 static int __netdev_upper_dev_link(struct net_device *dev,
4973 struct net_device *upper_dev, bool master,
4976 struct netdev_adjacent *i, *j, *to_i, *to_j;
4981 if (dev == upper_dev)
4984 /* To prevent loops, check if dev is not upper device to upper_dev. */
4985 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
4988 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
4991 if (master && netdev_master_upper_dev_get(dev))
4994 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
4999 /* Now that we linked these devs, make all the upper_dev's
5000 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5001 * versa, and don't forget the devices itself. All of these
5002 * links are non-neighbours.
5004 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5005 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5006 pr_debug("Interlinking %s with %s, non-neighbour\n",
5007 i->dev->name, j->dev->name);
5008 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5014 /* add dev to every upper_dev's upper device */
5015 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5016 pr_debug("linking %s's upper device %s with %s\n",
5017 upper_dev->name, i->dev->name, dev->name);
5018 ret = __netdev_adjacent_dev_link(dev, i->dev);
5020 goto rollback_upper_mesh;
5023 /* add upper_dev to every dev's lower device */
5024 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5025 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5026 i->dev->name, upper_dev->name);
5027 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5029 goto rollback_lower_mesh;
5032 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5035 rollback_lower_mesh:
5037 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5040 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5045 rollback_upper_mesh:
5047 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5050 __netdev_adjacent_dev_unlink(dev, i->dev);
5058 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5059 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5060 if (i == to_i && j == to_j)
5062 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5068 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5074 * netdev_upper_dev_link - Add a link to the upper device
5076 * @upper_dev: new upper device
5078 * Adds a link to device which is upper to this one. The caller must hold
5079 * the RTNL lock. On a failure a negative errno code is returned.
5080 * On success the reference counts are adjusted and the function
5083 int netdev_upper_dev_link(struct net_device *dev,
5084 struct net_device *upper_dev)
5086 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5088 EXPORT_SYMBOL(netdev_upper_dev_link);
5091 * netdev_master_upper_dev_link - Add a master link to the upper device
5093 * @upper_dev: new upper device
5095 * Adds a link to device which is upper to this one. In this case, only
5096 * one master upper device can be linked, although other non-master devices
5097 * might be linked as well. The caller must hold the RTNL lock.
5098 * On a failure a negative errno code is returned. On success the reference
5099 * counts are adjusted and the function returns zero.
5101 int netdev_master_upper_dev_link(struct net_device *dev,
5102 struct net_device *upper_dev)
5104 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5106 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5108 int netdev_master_upper_dev_link_private(struct net_device *dev,
5109 struct net_device *upper_dev,
5112 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5114 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5117 * netdev_upper_dev_unlink - Removes a link to upper device
5119 * @upper_dev: new upper device
5121 * Removes a link to device which is upper to this one. The caller must hold
5124 void netdev_upper_dev_unlink(struct net_device *dev,
5125 struct net_device *upper_dev)
5127 struct netdev_adjacent *i, *j;
5130 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5132 /* Here is the tricky part. We must remove all dev's lower
5133 * devices from all upper_dev's upper devices and vice
5134 * versa, to maintain the graph relationship.
5136 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5137 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5138 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5140 /* remove also the devices itself from lower/upper device
5143 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5144 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5146 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5147 __netdev_adjacent_dev_unlink(dev, i->dev);
5149 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5151 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5153 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5155 struct netdev_adjacent *iter;
5157 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5158 netdev_adjacent_sysfs_del(iter->dev, oldname,
5159 &iter->dev->adj_list.lower);
5160 netdev_adjacent_sysfs_add(iter->dev, dev,
5161 &iter->dev->adj_list.lower);
5164 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5165 netdev_adjacent_sysfs_del(iter->dev, oldname,
5166 &iter->dev->adj_list.upper);
5167 netdev_adjacent_sysfs_add(iter->dev, dev,
5168 &iter->dev->adj_list.upper);
5172 void *netdev_lower_dev_get_private(struct net_device *dev,
5173 struct net_device *lower_dev)
5175 struct netdev_adjacent *lower;
5179 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5183 return lower->private;
5185 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5188 int dev_get_nest_level(struct net_device *dev,
5189 bool (*type_check)(struct net_device *dev))
5191 struct net_device *lower = NULL;
5192 struct list_head *iter;
5198 netdev_for_each_lower_dev(dev, lower, iter) {
5199 nest = dev_get_nest_level(lower, type_check);
5200 if (max_nest < nest)
5204 if (type_check(dev))
5209 EXPORT_SYMBOL(dev_get_nest_level);
5211 static void dev_change_rx_flags(struct net_device *dev, int flags)
5213 const struct net_device_ops *ops = dev->netdev_ops;
5215 if (ops->ndo_change_rx_flags)
5216 ops->ndo_change_rx_flags(dev, flags);
5219 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5221 unsigned int old_flags = dev->flags;
5227 dev->flags |= IFF_PROMISC;
5228 dev->promiscuity += inc;
5229 if (dev->promiscuity == 0) {
5232 * If inc causes overflow, untouch promisc and return error.
5235 dev->flags &= ~IFF_PROMISC;
5237 dev->promiscuity -= inc;
5238 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5243 if (dev->flags != old_flags) {
5244 pr_info("device %s %s promiscuous mode\n",
5246 dev->flags & IFF_PROMISC ? "entered" : "left");
5247 if (audit_enabled) {
5248 current_uid_gid(&uid, &gid);
5249 audit_log(current->audit_context, GFP_ATOMIC,
5250 AUDIT_ANOM_PROMISCUOUS,
5251 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5252 dev->name, (dev->flags & IFF_PROMISC),
5253 (old_flags & IFF_PROMISC),
5254 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5255 from_kuid(&init_user_ns, uid),
5256 from_kgid(&init_user_ns, gid),
5257 audit_get_sessionid(current));
5260 dev_change_rx_flags(dev, IFF_PROMISC);
5263 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5268 * dev_set_promiscuity - update promiscuity count on a device
5272 * Add or remove promiscuity from a device. While the count in the device
5273 * remains above zero the interface remains promiscuous. Once it hits zero
5274 * the device reverts back to normal filtering operation. A negative inc
5275 * value is used to drop promiscuity on the device.
5276 * Return 0 if successful or a negative errno code on error.
5278 int dev_set_promiscuity(struct net_device *dev, int inc)
5280 unsigned int old_flags = dev->flags;
5283 err = __dev_set_promiscuity(dev, inc, true);
5286 if (dev->flags != old_flags)
5287 dev_set_rx_mode(dev);
5290 EXPORT_SYMBOL(dev_set_promiscuity);
5292 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5294 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5298 dev->flags |= IFF_ALLMULTI;
5299 dev->allmulti += inc;
5300 if (dev->allmulti == 0) {
5303 * If inc causes overflow, untouch allmulti and return error.
5306 dev->flags &= ~IFF_ALLMULTI;
5308 dev->allmulti -= inc;
5309 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5314 if (dev->flags ^ old_flags) {
5315 dev_change_rx_flags(dev, IFF_ALLMULTI);
5316 dev_set_rx_mode(dev);
5318 __dev_notify_flags(dev, old_flags,
5319 dev->gflags ^ old_gflags);
5325 * dev_set_allmulti - update allmulti count on a device
5329 * Add or remove reception of all multicast frames to a device. While the
5330 * count in the device remains above zero the interface remains listening
5331 * to all interfaces. Once it hits zero the device reverts back to normal
5332 * filtering operation. A negative @inc value is used to drop the counter
5333 * when releasing a resource needing all multicasts.
5334 * Return 0 if successful or a negative errno code on error.
5337 int dev_set_allmulti(struct net_device *dev, int inc)
5339 return __dev_set_allmulti(dev, inc, true);
5341 EXPORT_SYMBOL(dev_set_allmulti);
5344 * Upload unicast and multicast address lists to device and
5345 * configure RX filtering. When the device doesn't support unicast
5346 * filtering it is put in promiscuous mode while unicast addresses
5349 void __dev_set_rx_mode(struct net_device *dev)
5351 const struct net_device_ops *ops = dev->netdev_ops;
5353 /* dev_open will call this function so the list will stay sane. */
5354 if (!(dev->flags&IFF_UP))
5357 if (!netif_device_present(dev))
5360 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5361 /* Unicast addresses changes may only happen under the rtnl,
5362 * therefore calling __dev_set_promiscuity here is safe.
5364 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5365 __dev_set_promiscuity(dev, 1, false);
5366 dev->uc_promisc = true;
5367 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5368 __dev_set_promiscuity(dev, -1, false);
5369 dev->uc_promisc = false;
5373 if (ops->ndo_set_rx_mode)
5374 ops->ndo_set_rx_mode(dev);
5377 void dev_set_rx_mode(struct net_device *dev)
5379 netif_addr_lock_bh(dev);
5380 __dev_set_rx_mode(dev);
5381 netif_addr_unlock_bh(dev);
5385 * dev_get_flags - get flags reported to userspace
5388 * Get the combination of flag bits exported through APIs to userspace.
5390 unsigned int dev_get_flags(const struct net_device *dev)
5394 flags = (dev->flags & ~(IFF_PROMISC |
5399 (dev->gflags & (IFF_PROMISC |
5402 if (netif_running(dev)) {
5403 if (netif_oper_up(dev))
5404 flags |= IFF_RUNNING;
5405 if (netif_carrier_ok(dev))
5406 flags |= IFF_LOWER_UP;
5407 if (netif_dormant(dev))
5408 flags |= IFF_DORMANT;
5413 EXPORT_SYMBOL(dev_get_flags);
5415 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5417 unsigned int old_flags = dev->flags;
5423 * Set the flags on our device.
5426 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5427 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5429 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5433 * Load in the correct multicast list now the flags have changed.
5436 if ((old_flags ^ flags) & IFF_MULTICAST)
5437 dev_change_rx_flags(dev, IFF_MULTICAST);
5439 dev_set_rx_mode(dev);
5442 * Have we downed the interface. We handle IFF_UP ourselves
5443 * according to user attempts to set it, rather than blindly
5448 if ((old_flags ^ flags) & IFF_UP)
5449 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5451 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5452 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5453 unsigned int old_flags = dev->flags;
5455 dev->gflags ^= IFF_PROMISC;
5457 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5458 if (dev->flags != old_flags)
5459 dev_set_rx_mode(dev);
5462 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5463 is important. Some (broken) drivers set IFF_PROMISC, when
5464 IFF_ALLMULTI is requested not asking us and not reporting.
5466 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5467 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5469 dev->gflags ^= IFF_ALLMULTI;
5470 __dev_set_allmulti(dev, inc, false);
5476 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5477 unsigned int gchanges)
5479 unsigned int changes = dev->flags ^ old_flags;
5482 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5484 if (changes & IFF_UP) {
5485 if (dev->flags & IFF_UP)
5486 call_netdevice_notifiers(NETDEV_UP, dev);
5488 call_netdevice_notifiers(NETDEV_DOWN, dev);
5491 if (dev->flags & IFF_UP &&
5492 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5493 struct netdev_notifier_change_info change_info;
5495 change_info.flags_changed = changes;
5496 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5502 * dev_change_flags - change device settings
5504 * @flags: device state flags
5506 * Change settings on device based state flags. The flags are
5507 * in the userspace exported format.
5509 int dev_change_flags(struct net_device *dev, unsigned int flags)
5512 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5514 ret = __dev_change_flags(dev, flags);
5518 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5519 __dev_notify_flags(dev, old_flags, changes);
5522 EXPORT_SYMBOL(dev_change_flags);
5524 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5526 const struct net_device_ops *ops = dev->netdev_ops;
5528 if (ops->ndo_change_mtu)
5529 return ops->ndo_change_mtu(dev, new_mtu);
5536 * dev_set_mtu - Change maximum transfer unit
5538 * @new_mtu: new transfer unit
5540 * Change the maximum transfer size of the network device.
5542 int dev_set_mtu(struct net_device *dev, int new_mtu)
5546 if (new_mtu == dev->mtu)
5549 /* MTU must be positive. */
5553 if (!netif_device_present(dev))
5556 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5557 err = notifier_to_errno(err);
5561 orig_mtu = dev->mtu;
5562 err = __dev_set_mtu(dev, new_mtu);
5565 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5566 err = notifier_to_errno(err);
5568 /* setting mtu back and notifying everyone again,
5569 * so that they have a chance to revert changes.
5571 __dev_set_mtu(dev, orig_mtu);
5572 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5577 EXPORT_SYMBOL(dev_set_mtu);
5580 * dev_set_group - Change group this device belongs to
5582 * @new_group: group this device should belong to
5584 void dev_set_group(struct net_device *dev, int new_group)
5586 dev->group = new_group;
5588 EXPORT_SYMBOL(dev_set_group);
5591 * dev_set_mac_address - Change Media Access Control Address
5595 * Change the hardware (MAC) address of the device
5597 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5599 const struct net_device_ops *ops = dev->netdev_ops;
5602 if (!ops->ndo_set_mac_address)
5604 if (sa->sa_family != dev->type)
5606 if (!netif_device_present(dev))
5608 err = ops->ndo_set_mac_address(dev, sa);
5611 dev->addr_assign_type = NET_ADDR_SET;
5612 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5613 add_device_randomness(dev->dev_addr, dev->addr_len);
5616 EXPORT_SYMBOL(dev_set_mac_address);
5619 * dev_change_carrier - Change device carrier
5621 * @new_carrier: new value
5623 * Change device carrier
5625 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5627 const struct net_device_ops *ops = dev->netdev_ops;
5629 if (!ops->ndo_change_carrier)
5631 if (!netif_device_present(dev))
5633 return ops->ndo_change_carrier(dev, new_carrier);
5635 EXPORT_SYMBOL(dev_change_carrier);
5638 * dev_get_phys_port_id - Get device physical port ID
5642 * Get device physical port ID
5644 int dev_get_phys_port_id(struct net_device *dev,
5645 struct netdev_phys_port_id *ppid)
5647 const struct net_device_ops *ops = dev->netdev_ops;
5649 if (!ops->ndo_get_phys_port_id)
5651 return ops->ndo_get_phys_port_id(dev, ppid);
5653 EXPORT_SYMBOL(dev_get_phys_port_id);
5656 * dev_new_index - allocate an ifindex
5657 * @net: the applicable net namespace
5659 * Returns a suitable unique value for a new device interface
5660 * number. The caller must hold the rtnl semaphore or the
5661 * dev_base_lock to be sure it remains unique.
5663 static int dev_new_index(struct net *net)
5665 int ifindex = net->ifindex;
5669 if (!__dev_get_by_index(net, ifindex))
5670 return net->ifindex = ifindex;
5674 /* Delayed registration/unregisteration */
5675 static LIST_HEAD(net_todo_list);
5676 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5678 static void net_set_todo(struct net_device *dev)
5680 list_add_tail(&dev->todo_list, &net_todo_list);
5681 dev_net(dev)->dev_unreg_count++;
5684 static void rollback_registered_many(struct list_head *head)
5686 struct net_device *dev, *tmp;
5687 LIST_HEAD(close_head);
5689 BUG_ON(dev_boot_phase);
5692 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5693 /* Some devices call without registering
5694 * for initialization unwind. Remove those
5695 * devices and proceed with the remaining.
5697 if (dev->reg_state == NETREG_UNINITIALIZED) {
5698 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5702 list_del(&dev->unreg_list);
5705 dev->dismantle = true;
5706 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5709 /* If device is running, close it first. */
5710 list_for_each_entry(dev, head, unreg_list)
5711 list_add_tail(&dev->close_list, &close_head);
5712 dev_close_many(&close_head);
5714 list_for_each_entry(dev, head, unreg_list) {
5715 /* And unlink it from device chain. */
5716 unlist_netdevice(dev);
5718 dev->reg_state = NETREG_UNREGISTERING;
5723 list_for_each_entry(dev, head, unreg_list) {
5724 /* Shutdown queueing discipline. */
5728 /* Notify protocols, that we are about to destroy
5729 this device. They should clean all the things.
5731 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5734 * Flush the unicast and multicast chains
5739 if (dev->netdev_ops->ndo_uninit)
5740 dev->netdev_ops->ndo_uninit(dev);
5742 if (!dev->rtnl_link_ops ||
5743 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5744 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5746 /* Notifier chain MUST detach us all upper devices. */
5747 WARN_ON(netdev_has_any_upper_dev(dev));
5749 /* Remove entries from kobject tree */
5750 netdev_unregister_kobject(dev);
5752 /* Remove XPS queueing entries */
5753 netif_reset_xps_queues_gt(dev, 0);
5759 list_for_each_entry(dev, head, unreg_list)
5763 static void rollback_registered(struct net_device *dev)
5767 list_add(&dev->unreg_list, &single);
5768 rollback_registered_many(&single);
5772 static netdev_features_t netdev_fix_features(struct net_device *dev,
5773 netdev_features_t features)
5775 /* Fix illegal checksum combinations */
5776 if ((features & NETIF_F_HW_CSUM) &&
5777 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5778 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5779 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5782 /* TSO requires that SG is present as well. */
5783 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5784 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5785 features &= ~NETIF_F_ALL_TSO;
5788 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5789 !(features & NETIF_F_IP_CSUM)) {
5790 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5791 features &= ~NETIF_F_TSO;
5792 features &= ~NETIF_F_TSO_ECN;
5795 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5796 !(features & NETIF_F_IPV6_CSUM)) {
5797 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5798 features &= ~NETIF_F_TSO6;
5801 /* TSO ECN requires that TSO is present as well. */
5802 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5803 features &= ~NETIF_F_TSO_ECN;
5805 /* Software GSO depends on SG. */
5806 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5807 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5808 features &= ~NETIF_F_GSO;
5811 /* UFO needs SG and checksumming */
5812 if (features & NETIF_F_UFO) {
5813 /* maybe split UFO into V4 and V6? */
5814 if (!((features & NETIF_F_GEN_CSUM) ||
5815 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5816 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5818 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5819 features &= ~NETIF_F_UFO;
5822 if (!(features & NETIF_F_SG)) {
5824 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5825 features &= ~NETIF_F_UFO;
5829 #ifdef CONFIG_NET_RX_BUSY_POLL
5830 if (dev->netdev_ops->ndo_busy_poll)
5831 features |= NETIF_F_BUSY_POLL;
5834 features &= ~NETIF_F_BUSY_POLL;
5839 int __netdev_update_features(struct net_device *dev)
5841 netdev_features_t features;
5846 features = netdev_get_wanted_features(dev);
5848 if (dev->netdev_ops->ndo_fix_features)
5849 features = dev->netdev_ops->ndo_fix_features(dev, features);
5851 /* driver might be less strict about feature dependencies */
5852 features = netdev_fix_features(dev, features);
5854 if (dev->features == features)
5857 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5858 &dev->features, &features);
5860 if (dev->netdev_ops->ndo_set_features)
5861 err = dev->netdev_ops->ndo_set_features(dev, features);
5863 if (unlikely(err < 0)) {
5865 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5866 err, &features, &dev->features);
5871 dev->features = features;
5877 * netdev_update_features - recalculate device features
5878 * @dev: the device to check
5880 * Recalculate dev->features set and send notifications if it
5881 * has changed. Should be called after driver or hardware dependent
5882 * conditions might have changed that influence the features.
5884 void netdev_update_features(struct net_device *dev)
5886 if (__netdev_update_features(dev))
5887 netdev_features_change(dev);
5889 EXPORT_SYMBOL(netdev_update_features);
5892 * netdev_change_features - recalculate device features
5893 * @dev: the device to check
5895 * Recalculate dev->features set and send notifications even
5896 * if they have not changed. Should be called instead of
5897 * netdev_update_features() if also dev->vlan_features might
5898 * have changed to allow the changes to be propagated to stacked
5901 void netdev_change_features(struct net_device *dev)
5903 __netdev_update_features(dev);
5904 netdev_features_change(dev);
5906 EXPORT_SYMBOL(netdev_change_features);
5909 * netif_stacked_transfer_operstate - transfer operstate
5910 * @rootdev: the root or lower level device to transfer state from
5911 * @dev: the device to transfer operstate to
5913 * Transfer operational state from root to device. This is normally
5914 * called when a stacking relationship exists between the root
5915 * device and the device(a leaf device).
5917 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5918 struct net_device *dev)
5920 if (rootdev->operstate == IF_OPER_DORMANT)
5921 netif_dormant_on(dev);
5923 netif_dormant_off(dev);
5925 if (netif_carrier_ok(rootdev)) {
5926 if (!netif_carrier_ok(dev))
5927 netif_carrier_on(dev);
5929 if (netif_carrier_ok(dev))
5930 netif_carrier_off(dev);
5933 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5936 static int netif_alloc_rx_queues(struct net_device *dev)
5938 unsigned int i, count = dev->num_rx_queues;
5939 struct netdev_rx_queue *rx;
5943 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5949 for (i = 0; i < count; i++)
5955 static void netdev_init_one_queue(struct net_device *dev,
5956 struct netdev_queue *queue, void *_unused)
5958 /* Initialize queue lock */
5959 spin_lock_init(&queue->_xmit_lock);
5960 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5961 queue->xmit_lock_owner = -1;
5962 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5965 dql_init(&queue->dql, HZ);
5969 static void netif_free_tx_queues(struct net_device *dev)
5974 static int netif_alloc_netdev_queues(struct net_device *dev)
5976 unsigned int count = dev->num_tx_queues;
5977 struct netdev_queue *tx;
5978 size_t sz = count * sizeof(*tx);
5980 BUG_ON(count < 1 || count > 0xffff);
5982 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
5990 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5991 spin_lock_init(&dev->tx_global_lock);
5997 * register_netdevice - register a network device
5998 * @dev: device to register
6000 * Take a completed network device structure and add it to the kernel
6001 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6002 * chain. 0 is returned on success. A negative errno code is returned
6003 * on a failure to set up the device, or if the name is a duplicate.
6005 * Callers must hold the rtnl semaphore. You may want
6006 * register_netdev() instead of this.
6009 * The locking appears insufficient to guarantee two parallel registers
6010 * will not get the same name.
6013 int register_netdevice(struct net_device *dev)
6016 struct net *net = dev_net(dev);
6018 BUG_ON(dev_boot_phase);
6023 /* When net_device's are persistent, this will be fatal. */
6024 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6027 spin_lock_init(&dev->addr_list_lock);
6028 netdev_set_addr_lockdep_class(dev);
6032 ret = dev_get_valid_name(net, dev, dev->name);
6036 /* Init, if this function is available */
6037 if (dev->netdev_ops->ndo_init) {
6038 ret = dev->netdev_ops->ndo_init(dev);
6046 if (((dev->hw_features | dev->features) &
6047 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6048 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6049 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6050 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6057 dev->ifindex = dev_new_index(net);
6058 else if (__dev_get_by_index(net, dev->ifindex))
6061 if (dev->iflink == -1)
6062 dev->iflink = dev->ifindex;
6064 /* Transfer changeable features to wanted_features and enable
6065 * software offloads (GSO and GRO).
6067 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6068 dev->features |= NETIF_F_SOFT_FEATURES;
6069 dev->wanted_features = dev->features & dev->hw_features;
6071 if (!(dev->flags & IFF_LOOPBACK)) {
6072 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6075 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6077 dev->vlan_features |= NETIF_F_HIGHDMA;
6079 /* Make NETIF_F_SG inheritable to tunnel devices.
6081 dev->hw_enc_features |= NETIF_F_SG;
6083 /* Make NETIF_F_SG inheritable to MPLS.
6085 dev->mpls_features |= NETIF_F_SG;
6087 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6088 ret = notifier_to_errno(ret);
6092 ret = netdev_register_kobject(dev);
6095 dev->reg_state = NETREG_REGISTERED;
6097 __netdev_update_features(dev);
6100 * Default initial state at registry is that the
6101 * device is present.
6104 set_bit(__LINK_STATE_PRESENT, &dev->state);
6106 linkwatch_init_dev(dev);
6108 dev_init_scheduler(dev);
6110 list_netdevice(dev);
6111 add_device_randomness(dev->dev_addr, dev->addr_len);
6113 /* If the device has permanent device address, driver should
6114 * set dev_addr and also addr_assign_type should be set to
6115 * NET_ADDR_PERM (default value).
6117 if (dev->addr_assign_type == NET_ADDR_PERM)
6118 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6120 /* Notify protocols, that a new device appeared. */
6121 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6122 ret = notifier_to_errno(ret);
6124 rollback_registered(dev);
6125 dev->reg_state = NETREG_UNREGISTERED;
6128 * Prevent userspace races by waiting until the network
6129 * device is fully setup before sending notifications.
6131 if (!dev->rtnl_link_ops ||
6132 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6133 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6139 if (dev->netdev_ops->ndo_uninit)
6140 dev->netdev_ops->ndo_uninit(dev);
6143 EXPORT_SYMBOL(register_netdevice);
6146 * init_dummy_netdev - init a dummy network device for NAPI
6147 * @dev: device to init
6149 * This takes a network device structure and initialize the minimum
6150 * amount of fields so it can be used to schedule NAPI polls without
6151 * registering a full blown interface. This is to be used by drivers
6152 * that need to tie several hardware interfaces to a single NAPI
6153 * poll scheduler due to HW limitations.
6155 int init_dummy_netdev(struct net_device *dev)
6157 /* Clear everything. Note we don't initialize spinlocks
6158 * are they aren't supposed to be taken by any of the
6159 * NAPI code and this dummy netdev is supposed to be
6160 * only ever used for NAPI polls
6162 memset(dev, 0, sizeof(struct net_device));
6164 /* make sure we BUG if trying to hit standard
6165 * register/unregister code path
6167 dev->reg_state = NETREG_DUMMY;
6169 /* NAPI wants this */
6170 INIT_LIST_HEAD(&dev->napi_list);
6172 /* a dummy interface is started by default */
6173 set_bit(__LINK_STATE_PRESENT, &dev->state);
6174 set_bit(__LINK_STATE_START, &dev->state);
6176 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6177 * because users of this 'device' dont need to change
6183 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6187 * register_netdev - register a network device
6188 * @dev: device to register
6190 * Take a completed network device structure and add it to the kernel
6191 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6192 * chain. 0 is returned on success. A negative errno code is returned
6193 * on a failure to set up the device, or if the name is a duplicate.
6195 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6196 * and expands the device name if you passed a format string to
6199 int register_netdev(struct net_device *dev)
6204 err = register_netdevice(dev);
6208 EXPORT_SYMBOL(register_netdev);
6210 int netdev_refcnt_read(const struct net_device *dev)
6214 for_each_possible_cpu(i)
6215 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6218 EXPORT_SYMBOL(netdev_refcnt_read);
6221 * netdev_wait_allrefs - wait until all references are gone.
6222 * @dev: target net_device
6224 * This is called when unregistering network devices.
6226 * Any protocol or device that holds a reference should register
6227 * for netdevice notification, and cleanup and put back the
6228 * reference if they receive an UNREGISTER event.
6229 * We can get stuck here if buggy protocols don't correctly
6232 static void netdev_wait_allrefs(struct net_device *dev)
6234 unsigned long rebroadcast_time, warning_time;
6237 linkwatch_forget_dev(dev);
6239 rebroadcast_time = warning_time = jiffies;
6240 refcnt = netdev_refcnt_read(dev);
6242 while (refcnt != 0) {
6243 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6246 /* Rebroadcast unregister notification */
6247 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6253 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6254 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6256 /* We must not have linkwatch events
6257 * pending on unregister. If this
6258 * happens, we simply run the queue
6259 * unscheduled, resulting in a noop
6262 linkwatch_run_queue();
6267 rebroadcast_time = jiffies;
6272 refcnt = netdev_refcnt_read(dev);
6274 if (time_after(jiffies, warning_time + 10 * HZ)) {
6275 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6277 warning_time = jiffies;
6286 * register_netdevice(x1);
6287 * register_netdevice(x2);
6289 * unregister_netdevice(y1);
6290 * unregister_netdevice(y2);
6296 * We are invoked by rtnl_unlock().
6297 * This allows us to deal with problems:
6298 * 1) We can delete sysfs objects which invoke hotplug
6299 * without deadlocking with linkwatch via keventd.
6300 * 2) Since we run with the RTNL semaphore not held, we can sleep
6301 * safely in order to wait for the netdev refcnt to drop to zero.
6303 * We must not return until all unregister events added during
6304 * the interval the lock was held have been completed.
6306 void netdev_run_todo(void)
6308 struct list_head list;
6310 /* Snapshot list, allow later requests */
6311 list_replace_init(&net_todo_list, &list);
6316 /* Wait for rcu callbacks to finish before next phase */
6317 if (!list_empty(&list))
6320 while (!list_empty(&list)) {
6321 struct net_device *dev
6322 = list_first_entry(&list, struct net_device, todo_list);
6323 list_del(&dev->todo_list);
6326 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6329 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6330 pr_err("network todo '%s' but state %d\n",
6331 dev->name, dev->reg_state);
6336 dev->reg_state = NETREG_UNREGISTERED;
6338 on_each_cpu(flush_backlog, dev, 1);
6340 netdev_wait_allrefs(dev);
6343 BUG_ON(netdev_refcnt_read(dev));
6344 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6345 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6346 WARN_ON(dev->dn_ptr);
6348 if (dev->destructor)
6349 dev->destructor(dev);
6351 /* Report a network device has been unregistered */
6353 dev_net(dev)->dev_unreg_count--;
6355 wake_up(&netdev_unregistering_wq);
6357 /* Free network device */
6358 kobject_put(&dev->dev.kobj);
6362 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6363 * fields in the same order, with only the type differing.
6365 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6366 const struct net_device_stats *netdev_stats)
6368 #if BITS_PER_LONG == 64
6369 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6370 memcpy(stats64, netdev_stats, sizeof(*stats64));
6372 size_t i, n = sizeof(*stats64) / sizeof(u64);
6373 const unsigned long *src = (const unsigned long *)netdev_stats;
6374 u64 *dst = (u64 *)stats64;
6376 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6377 sizeof(*stats64) / sizeof(u64));
6378 for (i = 0; i < n; i++)
6382 EXPORT_SYMBOL(netdev_stats_to_stats64);
6385 * dev_get_stats - get network device statistics
6386 * @dev: device to get statistics from
6387 * @storage: place to store stats
6389 * Get network statistics from device. Return @storage.
6390 * The device driver may provide its own method by setting
6391 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6392 * otherwise the internal statistics structure is used.
6394 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6395 struct rtnl_link_stats64 *storage)
6397 const struct net_device_ops *ops = dev->netdev_ops;
6399 if (ops->ndo_get_stats64) {
6400 memset(storage, 0, sizeof(*storage));
6401 ops->ndo_get_stats64(dev, storage);
6402 } else if (ops->ndo_get_stats) {
6403 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6405 netdev_stats_to_stats64(storage, &dev->stats);
6407 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6408 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6411 EXPORT_SYMBOL(dev_get_stats);
6413 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6415 struct netdev_queue *queue = dev_ingress_queue(dev);
6417 #ifdef CONFIG_NET_CLS_ACT
6420 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6423 netdev_init_one_queue(dev, queue, NULL);
6424 queue->qdisc = &noop_qdisc;
6425 queue->qdisc_sleeping = &noop_qdisc;
6426 rcu_assign_pointer(dev->ingress_queue, queue);
6431 static const struct ethtool_ops default_ethtool_ops;
6433 void netdev_set_default_ethtool_ops(struct net_device *dev,
6434 const struct ethtool_ops *ops)
6436 if (dev->ethtool_ops == &default_ethtool_ops)
6437 dev->ethtool_ops = ops;
6439 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6441 void netdev_freemem(struct net_device *dev)
6443 char *addr = (char *)dev - dev->padded;
6449 * alloc_netdev_mqs - allocate network device
6450 * @sizeof_priv: size of private data to allocate space for
6451 * @name: device name format string
6452 * @name_assign_type: origin of device name
6453 * @setup: callback to initialize device
6454 * @txqs: the number of TX subqueues to allocate
6455 * @rxqs: the number of RX subqueues to allocate
6457 * Allocates a struct net_device with private data area for driver use
6458 * and performs basic initialization. Also allocates subqueue structs
6459 * for each queue on the device.
6461 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6462 unsigned char name_assign_type,
6463 void (*setup)(struct net_device *),
6464 unsigned int txqs, unsigned int rxqs)
6466 struct net_device *dev;
6468 struct net_device *p;
6470 BUG_ON(strlen(name) >= sizeof(dev->name));
6473 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6479 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6484 alloc_size = sizeof(struct net_device);
6486 /* ensure 32-byte alignment of private area */
6487 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6488 alloc_size += sizeof_priv;
6490 /* ensure 32-byte alignment of whole construct */
6491 alloc_size += NETDEV_ALIGN - 1;
6493 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6495 p = vzalloc(alloc_size);
6499 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6500 dev->padded = (char *)dev - (char *)p;
6502 dev->pcpu_refcnt = alloc_percpu(int);
6503 if (!dev->pcpu_refcnt)
6506 if (dev_addr_init(dev))
6512 dev_net_set(dev, &init_net);
6514 dev->gso_max_size = GSO_MAX_SIZE;
6515 dev->gso_max_segs = GSO_MAX_SEGS;
6517 INIT_LIST_HEAD(&dev->napi_list);
6518 INIT_LIST_HEAD(&dev->unreg_list);
6519 INIT_LIST_HEAD(&dev->close_list);
6520 INIT_LIST_HEAD(&dev->link_watch_list);
6521 INIT_LIST_HEAD(&dev->adj_list.upper);
6522 INIT_LIST_HEAD(&dev->adj_list.lower);
6523 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6524 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6525 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6528 dev->num_tx_queues = txqs;
6529 dev->real_num_tx_queues = txqs;
6530 if (netif_alloc_netdev_queues(dev))
6534 dev->num_rx_queues = rxqs;
6535 dev->real_num_rx_queues = rxqs;
6536 if (netif_alloc_rx_queues(dev))
6540 strcpy(dev->name, name);
6541 dev->name_assign_type = name_assign_type;
6542 dev->group = INIT_NETDEV_GROUP;
6543 if (!dev->ethtool_ops)
6544 dev->ethtool_ops = &default_ethtool_ops;
6552 free_percpu(dev->pcpu_refcnt);
6554 netdev_freemem(dev);
6557 EXPORT_SYMBOL(alloc_netdev_mqs);
6560 * free_netdev - free network device
6563 * This function does the last stage of destroying an allocated device
6564 * interface. The reference to the device object is released.
6565 * If this is the last reference then it will be freed.
6567 void free_netdev(struct net_device *dev)
6569 struct napi_struct *p, *n;
6571 release_net(dev_net(dev));
6573 netif_free_tx_queues(dev);
6578 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6580 /* Flush device addresses */
6581 dev_addr_flush(dev);
6583 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6586 free_percpu(dev->pcpu_refcnt);
6587 dev->pcpu_refcnt = NULL;
6589 /* Compatibility with error handling in drivers */
6590 if (dev->reg_state == NETREG_UNINITIALIZED) {
6591 netdev_freemem(dev);
6595 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6596 dev->reg_state = NETREG_RELEASED;
6598 /* will free via device release */
6599 put_device(&dev->dev);
6601 EXPORT_SYMBOL(free_netdev);
6604 * synchronize_net - Synchronize with packet receive processing
6606 * Wait for packets currently being received to be done.
6607 * Does not block later packets from starting.
6609 void synchronize_net(void)
6612 if (rtnl_is_locked())
6613 synchronize_rcu_expedited();
6617 EXPORT_SYMBOL(synchronize_net);
6620 * unregister_netdevice_queue - remove device from the kernel
6624 * This function shuts down a device interface and removes it
6625 * from the kernel tables.
6626 * If head not NULL, device is queued to be unregistered later.
6628 * Callers must hold the rtnl semaphore. You may want
6629 * unregister_netdev() instead of this.
6632 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6637 list_move_tail(&dev->unreg_list, head);
6639 rollback_registered(dev);
6640 /* Finish processing unregister after unlock */
6644 EXPORT_SYMBOL(unregister_netdevice_queue);
6647 * unregister_netdevice_many - unregister many devices
6648 * @head: list of devices
6650 * Note: As most callers use a stack allocated list_head,
6651 * we force a list_del() to make sure stack wont be corrupted later.
6653 void unregister_netdevice_many(struct list_head *head)
6655 struct net_device *dev;
6657 if (!list_empty(head)) {
6658 rollback_registered_many(head);
6659 list_for_each_entry(dev, head, unreg_list)
6664 EXPORT_SYMBOL(unregister_netdevice_many);
6667 * unregister_netdev - remove device from the kernel
6670 * This function shuts down a device interface and removes it
6671 * from the kernel tables.
6673 * This is just a wrapper for unregister_netdevice that takes
6674 * the rtnl semaphore. In general you want to use this and not
6675 * unregister_netdevice.
6677 void unregister_netdev(struct net_device *dev)
6680 unregister_netdevice(dev);
6683 EXPORT_SYMBOL(unregister_netdev);
6686 * dev_change_net_namespace - move device to different nethost namespace
6688 * @net: network namespace
6689 * @pat: If not NULL name pattern to try if the current device name
6690 * is already taken in the destination network namespace.
6692 * This function shuts down a device interface and moves it
6693 * to a new network namespace. On success 0 is returned, on
6694 * a failure a netagive errno code is returned.
6696 * Callers must hold the rtnl semaphore.
6699 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6705 /* Don't allow namespace local devices to be moved. */
6707 if (dev->features & NETIF_F_NETNS_LOCAL)
6710 /* Ensure the device has been registrered */
6711 if (dev->reg_state != NETREG_REGISTERED)
6714 /* Get out if there is nothing todo */
6716 if (net_eq(dev_net(dev), net))
6719 /* Pick the destination device name, and ensure
6720 * we can use it in the destination network namespace.
6723 if (__dev_get_by_name(net, dev->name)) {
6724 /* We get here if we can't use the current device name */
6727 if (dev_get_valid_name(net, dev, pat) < 0)
6732 * And now a mini version of register_netdevice unregister_netdevice.
6735 /* If device is running close it first. */
6738 /* And unlink it from device chain */
6740 unlist_netdevice(dev);
6744 /* Shutdown queueing discipline. */
6747 /* Notify protocols, that we are about to destroy
6748 this device. They should clean all the things.
6750 Note that dev->reg_state stays at NETREG_REGISTERED.
6751 This is wanted because this way 8021q and macvlan know
6752 the device is just moving and can keep their slaves up.
6754 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6756 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6757 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6760 * Flush the unicast and multicast chains
6765 /* Send a netdev-removed uevent to the old namespace */
6766 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6768 /* Actually switch the network namespace */
6769 dev_net_set(dev, net);
6771 /* If there is an ifindex conflict assign a new one */
6772 if (__dev_get_by_index(net, dev->ifindex)) {
6773 int iflink = (dev->iflink == dev->ifindex);
6774 dev->ifindex = dev_new_index(net);
6776 dev->iflink = dev->ifindex;
6779 /* Send a netdev-add uevent to the new namespace */
6780 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6782 /* Fixup kobjects */
6783 err = device_rename(&dev->dev, dev->name);
6786 /* Add the device back in the hashes */
6787 list_netdevice(dev);
6789 /* Notify protocols, that a new device appeared. */
6790 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6793 * Prevent userspace races by waiting until the network
6794 * device is fully setup before sending notifications.
6796 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6803 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6805 static int dev_cpu_callback(struct notifier_block *nfb,
6806 unsigned long action,
6809 struct sk_buff **list_skb;
6810 struct sk_buff *skb;
6811 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6812 struct softnet_data *sd, *oldsd;
6814 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6817 local_irq_disable();
6818 cpu = smp_processor_id();
6819 sd = &per_cpu(softnet_data, cpu);
6820 oldsd = &per_cpu(softnet_data, oldcpu);
6822 /* Find end of our completion_queue. */
6823 list_skb = &sd->completion_queue;
6825 list_skb = &(*list_skb)->next;
6826 /* Append completion queue from offline CPU. */
6827 *list_skb = oldsd->completion_queue;
6828 oldsd->completion_queue = NULL;
6830 /* Append output queue from offline CPU. */
6831 if (oldsd->output_queue) {
6832 *sd->output_queue_tailp = oldsd->output_queue;
6833 sd->output_queue_tailp = oldsd->output_queue_tailp;
6834 oldsd->output_queue = NULL;
6835 oldsd->output_queue_tailp = &oldsd->output_queue;
6837 /* Append NAPI poll list from offline CPU. */
6838 if (!list_empty(&oldsd->poll_list)) {
6839 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6840 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6843 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6846 /* Process offline CPU's input_pkt_queue */
6847 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6848 netif_rx_internal(skb);
6849 input_queue_head_incr(oldsd);
6851 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6852 netif_rx_internal(skb);
6853 input_queue_head_incr(oldsd);
6861 * netdev_increment_features - increment feature set by one
6862 * @all: current feature set
6863 * @one: new feature set
6864 * @mask: mask feature set
6866 * Computes a new feature set after adding a device with feature set
6867 * @one to the master device with current feature set @all. Will not
6868 * enable anything that is off in @mask. Returns the new feature set.
6870 netdev_features_t netdev_increment_features(netdev_features_t all,
6871 netdev_features_t one, netdev_features_t mask)
6873 if (mask & NETIF_F_GEN_CSUM)
6874 mask |= NETIF_F_ALL_CSUM;
6875 mask |= NETIF_F_VLAN_CHALLENGED;
6877 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6878 all &= one | ~NETIF_F_ALL_FOR_ALL;
6880 /* If one device supports hw checksumming, set for all. */
6881 if (all & NETIF_F_GEN_CSUM)
6882 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6886 EXPORT_SYMBOL(netdev_increment_features);
6888 static struct hlist_head * __net_init netdev_create_hash(void)
6891 struct hlist_head *hash;
6893 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6895 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6896 INIT_HLIST_HEAD(&hash[i]);
6901 /* Initialize per network namespace state */
6902 static int __net_init netdev_init(struct net *net)
6904 if (net != &init_net)
6905 INIT_LIST_HEAD(&net->dev_base_head);
6907 net->dev_name_head = netdev_create_hash();
6908 if (net->dev_name_head == NULL)
6911 net->dev_index_head = netdev_create_hash();
6912 if (net->dev_index_head == NULL)
6918 kfree(net->dev_name_head);
6924 * netdev_drivername - network driver for the device
6925 * @dev: network device
6927 * Determine network driver for device.
6929 const char *netdev_drivername(const struct net_device *dev)
6931 const struct device_driver *driver;
6932 const struct device *parent;
6933 const char *empty = "";
6935 parent = dev->dev.parent;
6939 driver = parent->driver;
6940 if (driver && driver->name)
6941 return driver->name;
6945 static int __netdev_printk(const char *level, const struct net_device *dev,
6946 struct va_format *vaf)
6950 if (dev && dev->dev.parent) {
6951 r = dev_printk_emit(level[1] - '0',
6954 dev_driver_string(dev->dev.parent),
6955 dev_name(dev->dev.parent),
6956 netdev_name(dev), netdev_reg_state(dev),
6959 r = printk("%s%s%s: %pV", level, netdev_name(dev),
6960 netdev_reg_state(dev), vaf);
6962 r = printk("%s(NULL net_device): %pV", level, vaf);
6968 int netdev_printk(const char *level, const struct net_device *dev,
6969 const char *format, ...)
6971 struct va_format vaf;
6975 va_start(args, format);
6980 r = __netdev_printk(level, dev, &vaf);
6986 EXPORT_SYMBOL(netdev_printk);
6988 #define define_netdev_printk_level(func, level) \
6989 int func(const struct net_device *dev, const char *fmt, ...) \
6992 struct va_format vaf; \
6995 va_start(args, fmt); \
7000 r = __netdev_printk(level, dev, &vaf); \
7006 EXPORT_SYMBOL(func);
7008 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7009 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7010 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7011 define_netdev_printk_level(netdev_err, KERN_ERR);
7012 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7013 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7014 define_netdev_printk_level(netdev_info, KERN_INFO);
7016 static void __net_exit netdev_exit(struct net *net)
7018 kfree(net->dev_name_head);
7019 kfree(net->dev_index_head);
7022 static struct pernet_operations __net_initdata netdev_net_ops = {
7023 .init = netdev_init,
7024 .exit = netdev_exit,
7027 static void __net_exit default_device_exit(struct net *net)
7029 struct net_device *dev, *aux;
7031 * Push all migratable network devices back to the
7032 * initial network namespace
7035 for_each_netdev_safe(net, dev, aux) {
7037 char fb_name[IFNAMSIZ];
7039 /* Ignore unmoveable devices (i.e. loopback) */
7040 if (dev->features & NETIF_F_NETNS_LOCAL)
7043 /* Leave virtual devices for the generic cleanup */
7044 if (dev->rtnl_link_ops)
7047 /* Push remaining network devices to init_net */
7048 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7049 err = dev_change_net_namespace(dev, &init_net, fb_name);
7051 pr_emerg("%s: failed to move %s to init_net: %d\n",
7052 __func__, dev->name, err);
7059 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7061 /* Return with the rtnl_lock held when there are no network
7062 * devices unregistering in any network namespace in net_list.
7069 prepare_to_wait(&netdev_unregistering_wq, &wait,
7070 TASK_UNINTERRUPTIBLE);
7071 unregistering = false;
7073 list_for_each_entry(net, net_list, exit_list) {
7074 if (net->dev_unreg_count > 0) {
7075 unregistering = true;
7084 finish_wait(&netdev_unregistering_wq, &wait);
7087 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7089 /* At exit all network devices most be removed from a network
7090 * namespace. Do this in the reverse order of registration.
7091 * Do this across as many network namespaces as possible to
7092 * improve batching efficiency.
7094 struct net_device *dev;
7096 LIST_HEAD(dev_kill_list);
7098 /* To prevent network device cleanup code from dereferencing
7099 * loopback devices or network devices that have been freed
7100 * wait here for all pending unregistrations to complete,
7101 * before unregistring the loopback device and allowing the
7102 * network namespace be freed.
7104 * The netdev todo list containing all network devices
7105 * unregistrations that happen in default_device_exit_batch
7106 * will run in the rtnl_unlock() at the end of
7107 * default_device_exit_batch.
7109 rtnl_lock_unregistering(net_list);
7110 list_for_each_entry(net, net_list, exit_list) {
7111 for_each_netdev_reverse(net, dev) {
7112 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7113 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7115 unregister_netdevice_queue(dev, &dev_kill_list);
7118 unregister_netdevice_many(&dev_kill_list);
7122 static struct pernet_operations __net_initdata default_device_ops = {
7123 .exit = default_device_exit,
7124 .exit_batch = default_device_exit_batch,
7128 * Initialize the DEV module. At boot time this walks the device list and
7129 * unhooks any devices that fail to initialise (normally hardware not
7130 * present) and leaves us with a valid list of present and active devices.
7135 * This is called single threaded during boot, so no need
7136 * to take the rtnl semaphore.
7138 static int __init net_dev_init(void)
7140 int i, rc = -ENOMEM;
7142 BUG_ON(!dev_boot_phase);
7144 if (dev_proc_init())
7147 if (netdev_kobject_init())
7150 INIT_LIST_HEAD(&ptype_all);
7151 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7152 INIT_LIST_HEAD(&ptype_base[i]);
7154 INIT_LIST_HEAD(&offload_base);
7156 if (register_pernet_subsys(&netdev_net_ops))
7160 * Initialise the packet receive queues.
7163 for_each_possible_cpu(i) {
7164 struct softnet_data *sd = &per_cpu(softnet_data, i);
7166 skb_queue_head_init(&sd->input_pkt_queue);
7167 skb_queue_head_init(&sd->process_queue);
7168 INIT_LIST_HEAD(&sd->poll_list);
7169 sd->output_queue_tailp = &sd->output_queue;
7171 sd->csd.func = rps_trigger_softirq;
7176 sd->backlog.poll = process_backlog;
7177 sd->backlog.weight = weight_p;
7182 /* The loopback device is special if any other network devices
7183 * is present in a network namespace the loopback device must
7184 * be present. Since we now dynamically allocate and free the
7185 * loopback device ensure this invariant is maintained by
7186 * keeping the loopback device as the first device on the
7187 * list of network devices. Ensuring the loopback devices
7188 * is the first device that appears and the last network device
7191 if (register_pernet_device(&loopback_net_ops))
7194 if (register_pernet_device(&default_device_ops))
7197 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7198 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7200 hotcpu_notifier(dev_cpu_callback, 0);
7207 subsys_initcall(net_dev_init);