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);
153 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
156 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
158 * Writers must hold the rtnl semaphore while they loop through the
159 * dev_base_head list, and hold dev_base_lock for writing when they do the
160 * actual updates. This allows pure readers to access the list even
161 * while a writer is preparing to update it.
163 * To put it another way, dev_base_lock is held for writing only to
164 * protect against pure readers; the rtnl semaphore provides the
165 * protection against other writers.
167 * See, for example usages, register_netdevice() and
168 * unregister_netdevice(), which must be called with the rtnl
171 DEFINE_RWLOCK(dev_base_lock);
172 EXPORT_SYMBOL(dev_base_lock);
174 /* protects napi_hash addition/deletion and napi_gen_id */
175 static DEFINE_SPINLOCK(napi_hash_lock);
177 static unsigned int napi_gen_id;
178 static DEFINE_HASHTABLE(napi_hash, 8);
180 static seqcount_t devnet_rename_seq;
182 static inline void dev_base_seq_inc(struct net *net)
184 while (++net->dev_base_seq == 0);
187 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
189 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
191 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
194 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
196 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
199 static inline void rps_lock(struct softnet_data *sd)
202 spin_lock(&sd->input_pkt_queue.lock);
206 static inline void rps_unlock(struct softnet_data *sd)
209 spin_unlock(&sd->input_pkt_queue.lock);
213 /* Device list insertion */
214 static void list_netdevice(struct net_device *dev)
216 struct net *net = dev_net(dev);
220 write_lock_bh(&dev_base_lock);
221 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
222 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
223 hlist_add_head_rcu(&dev->index_hlist,
224 dev_index_hash(net, dev->ifindex));
225 write_unlock_bh(&dev_base_lock);
227 dev_base_seq_inc(net);
230 /* Device list removal
231 * caller must respect a RCU grace period before freeing/reusing dev
233 static void unlist_netdevice(struct net_device *dev)
237 /* Unlink dev from the device chain */
238 write_lock_bh(&dev_base_lock);
239 list_del_rcu(&dev->dev_list);
240 hlist_del_rcu(&dev->name_hlist);
241 hlist_del_rcu(&dev->index_hlist);
242 write_unlock_bh(&dev_base_lock);
244 dev_base_seq_inc(dev_net(dev));
251 static RAW_NOTIFIER_HEAD(netdev_chain);
254 * Device drivers call our routines to queue packets here. We empty the
255 * queue in the local softnet handler.
258 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
259 EXPORT_PER_CPU_SYMBOL(softnet_data);
261 #ifdef CONFIG_LOCKDEP
263 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
264 * according to dev->type
266 static const unsigned short netdev_lock_type[] =
267 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
268 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
269 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
270 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
271 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
272 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
273 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
274 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
275 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
276 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
277 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
278 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
279 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
280 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
281 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
283 static const char *const netdev_lock_name[] =
284 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
285 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
286 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
287 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
288 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
289 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
290 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
291 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
292 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
293 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
294 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
295 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
296 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
297 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
298 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
300 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
301 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
303 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
307 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
308 if (netdev_lock_type[i] == dev_type)
310 /* the last key is used by default */
311 return ARRAY_SIZE(netdev_lock_type) - 1;
314 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
315 unsigned short dev_type)
319 i = netdev_lock_pos(dev_type);
320 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
321 netdev_lock_name[i]);
324 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
328 i = netdev_lock_pos(dev->type);
329 lockdep_set_class_and_name(&dev->addr_list_lock,
330 &netdev_addr_lock_key[i],
331 netdev_lock_name[i]);
334 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
335 unsigned short dev_type)
338 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
343 /*******************************************************************************
345 Protocol management and registration routines
347 *******************************************************************************/
350 * Add a protocol ID to the list. Now that the input handler is
351 * smarter we can dispense with all the messy stuff that used to be
354 * BEWARE!!! Protocol handlers, mangling input packets,
355 * MUST BE last in hash buckets and checking protocol handlers
356 * MUST start from promiscuous ptype_all chain in net_bh.
357 * It is true now, do not change it.
358 * Explanation follows: if protocol handler, mangling packet, will
359 * be the first on list, it is not able to sense, that packet
360 * is cloned and should be copied-on-write, so that it will
361 * change it and subsequent readers will get broken packet.
365 static inline struct list_head *ptype_head(const struct packet_type *pt)
367 if (pt->type == htons(ETH_P_ALL))
370 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
374 * dev_add_pack - add packet handler
375 * @pt: packet type declaration
377 * Add a protocol handler to the networking stack. The passed &packet_type
378 * is linked into kernel lists and may not be freed until it has been
379 * removed from the kernel lists.
381 * This call does not sleep therefore it can not
382 * guarantee all CPU's that are in middle of receiving packets
383 * will see the new packet type (until the next received packet).
386 void dev_add_pack(struct packet_type *pt)
388 struct list_head *head = ptype_head(pt);
390 spin_lock(&ptype_lock);
391 list_add_rcu(&pt->list, head);
392 spin_unlock(&ptype_lock);
394 EXPORT_SYMBOL(dev_add_pack);
397 * __dev_remove_pack - remove packet handler
398 * @pt: packet type declaration
400 * Remove a protocol handler that was previously added to the kernel
401 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
402 * from the kernel lists and can be freed or reused once this function
405 * The packet type might still be in use by receivers
406 * and must not be freed until after all the CPU's have gone
407 * through a quiescent state.
409 void __dev_remove_pack(struct packet_type *pt)
411 struct list_head *head = ptype_head(pt);
412 struct packet_type *pt1;
414 spin_lock(&ptype_lock);
416 list_for_each_entry(pt1, head, list) {
418 list_del_rcu(&pt->list);
423 pr_warn("dev_remove_pack: %p not found\n", pt);
425 spin_unlock(&ptype_lock);
427 EXPORT_SYMBOL(__dev_remove_pack);
430 * dev_remove_pack - remove packet handler
431 * @pt: packet type declaration
433 * Remove a protocol handler that was previously added to the kernel
434 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
435 * from the kernel lists and can be freed or reused once this function
438 * This call sleeps to guarantee that no CPU is looking at the packet
441 void dev_remove_pack(struct packet_type *pt)
443 __dev_remove_pack(pt);
447 EXPORT_SYMBOL(dev_remove_pack);
451 * dev_add_offload - register offload handlers
452 * @po: protocol offload declaration
454 * Add protocol offload handlers to the networking stack. The passed
455 * &proto_offload is linked into kernel lists and may not be freed until
456 * it has been removed from the kernel lists.
458 * This call does not sleep therefore it can not
459 * guarantee all CPU's that are in middle of receiving packets
460 * will see the new offload handlers (until the next received packet).
462 void dev_add_offload(struct packet_offload *po)
464 struct list_head *head = &offload_base;
466 spin_lock(&offload_lock);
467 list_add_rcu(&po->list, head);
468 spin_unlock(&offload_lock);
470 EXPORT_SYMBOL(dev_add_offload);
473 * __dev_remove_offload - remove offload handler
474 * @po: packet offload declaration
476 * Remove a protocol offload handler that was previously added to the
477 * kernel offload handlers by dev_add_offload(). The passed &offload_type
478 * is removed from the kernel lists and can be freed or reused once this
481 * The packet type might still be in use by receivers
482 * and must not be freed until after all the CPU's have gone
483 * through a quiescent state.
485 static void __dev_remove_offload(struct packet_offload *po)
487 struct list_head *head = &offload_base;
488 struct packet_offload *po1;
490 spin_lock(&offload_lock);
492 list_for_each_entry(po1, head, list) {
494 list_del_rcu(&po->list);
499 pr_warn("dev_remove_offload: %p not found\n", po);
501 spin_unlock(&offload_lock);
505 * dev_remove_offload - remove packet offload handler
506 * @po: packet offload declaration
508 * Remove a packet offload handler that was previously added to the kernel
509 * offload handlers by dev_add_offload(). The passed &offload_type is
510 * removed from the kernel lists and can be freed or reused once this
513 * This call sleeps to guarantee that no CPU is looking at the packet
516 void dev_remove_offload(struct packet_offload *po)
518 __dev_remove_offload(po);
522 EXPORT_SYMBOL(dev_remove_offload);
524 /******************************************************************************
526 Device Boot-time Settings Routines
528 *******************************************************************************/
530 /* Boot time configuration table */
531 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
534 * netdev_boot_setup_add - add new setup entry
535 * @name: name of the device
536 * @map: configured settings for the device
538 * Adds new setup entry to the dev_boot_setup list. The function
539 * returns 0 on error and 1 on success. This is a generic routine to
542 static int netdev_boot_setup_add(char *name, struct ifmap *map)
544 struct netdev_boot_setup *s;
548 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
549 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
550 memset(s[i].name, 0, sizeof(s[i].name));
551 strlcpy(s[i].name, name, IFNAMSIZ);
552 memcpy(&s[i].map, map, sizeof(s[i].map));
557 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
561 * netdev_boot_setup_check - check boot time settings
562 * @dev: the netdevice
564 * Check boot time settings for the device.
565 * The found settings are set for the device to be used
566 * later in the device probing.
567 * Returns 0 if no settings found, 1 if they are.
569 int netdev_boot_setup_check(struct net_device *dev)
571 struct netdev_boot_setup *s = dev_boot_setup;
574 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
575 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
576 !strcmp(dev->name, s[i].name)) {
577 dev->irq = s[i].map.irq;
578 dev->base_addr = s[i].map.base_addr;
579 dev->mem_start = s[i].map.mem_start;
580 dev->mem_end = s[i].map.mem_end;
586 EXPORT_SYMBOL(netdev_boot_setup_check);
590 * netdev_boot_base - get address from boot time settings
591 * @prefix: prefix for network device
592 * @unit: id for network device
594 * Check boot time settings for the base address of device.
595 * The found settings are set for the device to be used
596 * later in the device probing.
597 * Returns 0 if no settings found.
599 unsigned long netdev_boot_base(const char *prefix, int unit)
601 const struct netdev_boot_setup *s = dev_boot_setup;
605 sprintf(name, "%s%d", prefix, unit);
608 * If device already registered then return base of 1
609 * to indicate not to probe for this interface
611 if (__dev_get_by_name(&init_net, name))
614 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
615 if (!strcmp(name, s[i].name))
616 return s[i].map.base_addr;
621 * Saves at boot time configured settings for any netdevice.
623 int __init netdev_boot_setup(char *str)
628 str = get_options(str, ARRAY_SIZE(ints), ints);
633 memset(&map, 0, sizeof(map));
637 map.base_addr = ints[2];
639 map.mem_start = ints[3];
641 map.mem_end = ints[4];
643 /* Add new entry to the list */
644 return netdev_boot_setup_add(str, &map);
647 __setup("netdev=", netdev_boot_setup);
649 /*******************************************************************************
651 Device Interface Subroutines
653 *******************************************************************************/
656 * __dev_get_by_name - find a device by its name
657 * @net: the applicable net namespace
658 * @name: name to find
660 * Find an interface by name. Must be called under RTNL semaphore
661 * or @dev_base_lock. If the name is found a pointer to the device
662 * is returned. If the name is not found then %NULL is returned. The
663 * reference counters are not incremented so the caller must be
664 * careful with locks.
667 struct net_device *__dev_get_by_name(struct net *net, const char *name)
669 struct net_device *dev;
670 struct hlist_head *head = dev_name_hash(net, name);
672 hlist_for_each_entry(dev, head, name_hlist)
673 if (!strncmp(dev->name, name, IFNAMSIZ))
678 EXPORT_SYMBOL(__dev_get_by_name);
681 * dev_get_by_name_rcu - find a device by its name
682 * @net: the applicable net namespace
683 * @name: name to find
685 * Find an interface by name.
686 * If the name is found a pointer to the device is returned.
687 * If the name is not found then %NULL is returned.
688 * The reference counters are not incremented so the caller must be
689 * careful with locks. The caller must hold RCU lock.
692 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
694 struct net_device *dev;
695 struct hlist_head *head = dev_name_hash(net, name);
697 hlist_for_each_entry_rcu(dev, head, name_hlist)
698 if (!strncmp(dev->name, name, IFNAMSIZ))
703 EXPORT_SYMBOL(dev_get_by_name_rcu);
706 * dev_get_by_name - find a device by its name
707 * @net: the applicable net namespace
708 * @name: name to find
710 * Find an interface by name. This can be called from any
711 * context and does its own locking. The returned handle has
712 * the usage count incremented and the caller must use dev_put() to
713 * release it when it is no longer needed. %NULL is returned if no
714 * matching device is found.
717 struct net_device *dev_get_by_name(struct net *net, const char *name)
719 struct net_device *dev;
722 dev = dev_get_by_name_rcu(net, name);
728 EXPORT_SYMBOL(dev_get_by_name);
731 * __dev_get_by_index - find a device by its ifindex
732 * @net: the applicable net namespace
733 * @ifindex: index of device
735 * Search for an interface by index. Returns %NULL if the device
736 * is not found or a pointer to the device. The device has not
737 * had its reference counter increased so the caller must be careful
738 * about locking. The caller must hold either the RTNL semaphore
742 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
744 struct net_device *dev;
745 struct hlist_head *head = dev_index_hash(net, ifindex);
747 hlist_for_each_entry(dev, head, index_hlist)
748 if (dev->ifindex == ifindex)
753 EXPORT_SYMBOL(__dev_get_by_index);
756 * dev_get_by_index_rcu - find a device by its ifindex
757 * @net: the applicable net namespace
758 * @ifindex: index of device
760 * Search for an interface by index. Returns %NULL if the device
761 * is not found or a pointer to the device. The device has not
762 * had its reference counter increased so the caller must be careful
763 * about locking. The caller must hold RCU lock.
766 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
768 struct net_device *dev;
769 struct hlist_head *head = dev_index_hash(net, ifindex);
771 hlist_for_each_entry_rcu(dev, head, index_hlist)
772 if (dev->ifindex == ifindex)
777 EXPORT_SYMBOL(dev_get_by_index_rcu);
781 * dev_get_by_index - find a device by its ifindex
782 * @net: the applicable net namespace
783 * @ifindex: index of device
785 * Search for an interface by index. Returns NULL if the device
786 * is not found or a pointer to the device. The device returned has
787 * had a reference added and the pointer is safe until the user calls
788 * dev_put to indicate they have finished with it.
791 struct net_device *dev_get_by_index(struct net *net, int ifindex)
793 struct net_device *dev;
796 dev = dev_get_by_index_rcu(net, ifindex);
802 EXPORT_SYMBOL(dev_get_by_index);
805 * netdev_get_name - get a netdevice name, knowing its ifindex.
806 * @net: network namespace
807 * @name: a pointer to the buffer where the name will be stored.
808 * @ifindex: the ifindex of the interface to get the name from.
810 * The use of raw_seqcount_begin() and cond_resched() before
811 * retrying is required as we want to give the writers a chance
812 * to complete when CONFIG_PREEMPT is not set.
814 int netdev_get_name(struct net *net, char *name, int ifindex)
816 struct net_device *dev;
820 seq = raw_seqcount_begin(&devnet_rename_seq);
822 dev = dev_get_by_index_rcu(net, ifindex);
828 strcpy(name, dev->name);
830 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
839 * dev_getbyhwaddr_rcu - find a device by its hardware address
840 * @net: the applicable net namespace
841 * @type: media type of device
842 * @ha: hardware address
844 * Search for an interface by MAC address. Returns NULL if the device
845 * is not found or a pointer to the device.
846 * The caller must hold RCU or RTNL.
847 * The returned device has not had its ref count increased
848 * and the caller must therefore be careful about locking
852 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
855 struct net_device *dev;
857 for_each_netdev_rcu(net, dev)
858 if (dev->type == type &&
859 !memcmp(dev->dev_addr, ha, dev->addr_len))
864 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
866 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
868 struct net_device *dev;
871 for_each_netdev(net, dev)
872 if (dev->type == type)
877 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
879 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
881 struct net_device *dev, *ret = NULL;
884 for_each_netdev_rcu(net, dev)
885 if (dev->type == type) {
893 EXPORT_SYMBOL(dev_getfirstbyhwtype);
896 * dev_get_by_flags_rcu - find any device with given flags
897 * @net: the applicable net namespace
898 * @if_flags: IFF_* values
899 * @mask: bitmask of bits in if_flags to check
901 * Search for any interface with the given flags. Returns NULL if a device
902 * is not found or a pointer to the device. Must be called inside
903 * rcu_read_lock(), and result refcount is unchanged.
906 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
909 struct net_device *dev, *ret;
912 for_each_netdev_rcu(net, dev) {
913 if (((dev->flags ^ if_flags) & mask) == 0) {
920 EXPORT_SYMBOL(dev_get_by_flags_rcu);
923 * dev_valid_name - check if name is okay for network device
926 * Network device names need to be valid file names to
927 * to allow sysfs to work. We also disallow any kind of
930 bool dev_valid_name(const char *name)
934 if (strlen(name) >= IFNAMSIZ)
936 if (!strcmp(name, ".") || !strcmp(name, ".."))
940 if (*name == '/' || isspace(*name))
946 EXPORT_SYMBOL(dev_valid_name);
949 * __dev_alloc_name - allocate a name for a device
950 * @net: network namespace to allocate the device name in
951 * @name: name format string
952 * @buf: scratch buffer and result name string
954 * Passed a format string - eg "lt%d" it will try and find a suitable
955 * id. It scans list of devices to build up a free map, then chooses
956 * the first empty slot. The caller must hold the dev_base or rtnl lock
957 * while allocating the name and adding the device in order to avoid
959 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
960 * Returns the number of the unit assigned or a negative errno code.
963 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
967 const int max_netdevices = 8*PAGE_SIZE;
968 unsigned long *inuse;
969 struct net_device *d;
971 p = strnchr(name, IFNAMSIZ-1, '%');
974 * Verify the string as this thing may have come from
975 * the user. There must be either one "%d" and no other "%"
978 if (p[1] != 'd' || strchr(p + 2, '%'))
981 /* Use one page as a bit array of possible slots */
982 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
986 for_each_netdev(net, d) {
987 if (!sscanf(d->name, name, &i))
989 if (i < 0 || i >= max_netdevices)
992 /* avoid cases where sscanf is not exact inverse of printf */
993 snprintf(buf, IFNAMSIZ, name, i);
994 if (!strncmp(buf, d->name, IFNAMSIZ))
998 i = find_first_zero_bit(inuse, max_netdevices);
999 free_page((unsigned long) inuse);
1003 snprintf(buf, IFNAMSIZ, name, i);
1004 if (!__dev_get_by_name(net, buf))
1007 /* It is possible to run out of possible slots
1008 * when the name is long and there isn't enough space left
1009 * for the digits, or if all bits are used.
1015 * dev_alloc_name - allocate a name for a device
1017 * @name: name format string
1019 * Passed a format string - eg "lt%d" it will try and find a suitable
1020 * id. It scans list of devices to build up a free map, then chooses
1021 * the first empty slot. The caller must hold the dev_base or rtnl lock
1022 * while allocating the name and adding the device in order to avoid
1024 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1025 * Returns the number of the unit assigned or a negative errno code.
1028 int dev_alloc_name(struct net_device *dev, const char *name)
1034 BUG_ON(!dev_net(dev));
1036 ret = __dev_alloc_name(net, name, buf);
1038 strlcpy(dev->name, buf, IFNAMSIZ);
1041 EXPORT_SYMBOL(dev_alloc_name);
1043 static int dev_alloc_name_ns(struct net *net,
1044 struct net_device *dev,
1050 ret = __dev_alloc_name(net, name, buf);
1052 strlcpy(dev->name, buf, IFNAMSIZ);
1056 static int dev_get_valid_name(struct net *net,
1057 struct net_device *dev,
1062 if (!dev_valid_name(name))
1065 if (strchr(name, '%'))
1066 return dev_alloc_name_ns(net, dev, name);
1067 else if (__dev_get_by_name(net, name))
1069 else if (dev->name != name)
1070 strlcpy(dev->name, name, IFNAMSIZ);
1076 * dev_change_name - change name of a device
1078 * @newname: name (or format string) must be at least IFNAMSIZ
1080 * Change name of a device, can pass format strings "eth%d".
1083 int dev_change_name(struct net_device *dev, const char *newname)
1085 char oldname[IFNAMSIZ];
1091 BUG_ON(!dev_net(dev));
1094 if (dev->flags & IFF_UP)
1097 write_seqcount_begin(&devnet_rename_seq);
1099 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1100 write_seqcount_end(&devnet_rename_seq);
1104 memcpy(oldname, dev->name, IFNAMSIZ);
1106 err = dev_get_valid_name(net, dev, newname);
1108 write_seqcount_end(&devnet_rename_seq);
1113 ret = device_rename(&dev->dev, dev->name);
1115 memcpy(dev->name, oldname, IFNAMSIZ);
1116 write_seqcount_end(&devnet_rename_seq);
1120 write_seqcount_end(&devnet_rename_seq);
1122 netdev_adjacent_rename_links(dev, oldname);
1124 write_lock_bh(&dev_base_lock);
1125 hlist_del_rcu(&dev->name_hlist);
1126 write_unlock_bh(&dev_base_lock);
1130 write_lock_bh(&dev_base_lock);
1131 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1132 write_unlock_bh(&dev_base_lock);
1134 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1135 ret = notifier_to_errno(ret);
1138 /* err >= 0 after dev_alloc_name() or stores the first errno */
1141 write_seqcount_begin(&devnet_rename_seq);
1142 memcpy(dev->name, oldname, IFNAMSIZ);
1143 memcpy(oldname, newname, IFNAMSIZ);
1146 pr_err("%s: name change rollback failed: %d\n",
1155 * dev_set_alias - change ifalias of a device
1157 * @alias: name up to IFALIASZ
1158 * @len: limit of bytes to copy from info
1160 * Set ifalias for a device,
1162 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1168 if (len >= IFALIASZ)
1172 kfree(dev->ifalias);
1173 dev->ifalias = NULL;
1177 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1180 dev->ifalias = new_ifalias;
1182 strlcpy(dev->ifalias, alias, len+1);
1188 * netdev_features_change - device changes features
1189 * @dev: device to cause notification
1191 * Called to indicate a device has changed features.
1193 void netdev_features_change(struct net_device *dev)
1195 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1197 EXPORT_SYMBOL(netdev_features_change);
1200 * netdev_state_change - device changes state
1201 * @dev: device to cause notification
1203 * Called to indicate a device has changed state. This function calls
1204 * the notifier chains for netdev_chain and sends a NEWLINK message
1205 * to the routing socket.
1207 void netdev_state_change(struct net_device *dev)
1209 if (dev->flags & IFF_UP) {
1210 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1211 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1214 EXPORT_SYMBOL(netdev_state_change);
1217 * netdev_notify_peers - notify network peers about existence of @dev
1218 * @dev: network device
1220 * Generate traffic such that interested network peers are aware of
1221 * @dev, such as by generating a gratuitous ARP. This may be used when
1222 * a device wants to inform the rest of the network about some sort of
1223 * reconfiguration such as a failover event or virtual machine
1226 void netdev_notify_peers(struct net_device *dev)
1229 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1232 EXPORT_SYMBOL(netdev_notify_peers);
1234 static int __dev_open(struct net_device *dev)
1236 const struct net_device_ops *ops = dev->netdev_ops;
1241 if (!netif_device_present(dev))
1244 /* Block netpoll from trying to do any rx path servicing.
1245 * If we don't do this there is a chance ndo_poll_controller
1246 * or ndo_poll may be running while we open the device
1248 netpoll_poll_disable(dev);
1250 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1251 ret = notifier_to_errno(ret);
1255 set_bit(__LINK_STATE_START, &dev->state);
1257 if (ops->ndo_validate_addr)
1258 ret = ops->ndo_validate_addr(dev);
1260 if (!ret && ops->ndo_open)
1261 ret = ops->ndo_open(dev);
1263 netpoll_poll_enable(dev);
1266 clear_bit(__LINK_STATE_START, &dev->state);
1268 dev->flags |= IFF_UP;
1269 net_dmaengine_get();
1270 dev_set_rx_mode(dev);
1272 add_device_randomness(dev->dev_addr, dev->addr_len);
1279 * dev_open - prepare an interface for use.
1280 * @dev: device to open
1282 * Takes a device from down to up state. The device's private open
1283 * function is invoked and then the multicast lists are loaded. Finally
1284 * the device is moved into the up state and a %NETDEV_UP message is
1285 * sent to the netdev notifier chain.
1287 * Calling this function on an active interface is a nop. On a failure
1288 * a negative errno code is returned.
1290 int dev_open(struct net_device *dev)
1294 if (dev->flags & IFF_UP)
1297 ret = __dev_open(dev);
1301 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1302 call_netdevice_notifiers(NETDEV_UP, dev);
1306 EXPORT_SYMBOL(dev_open);
1308 static int __dev_close_many(struct list_head *head)
1310 struct net_device *dev;
1315 list_for_each_entry(dev, head, close_list) {
1316 /* Temporarily disable netpoll until the interface is down */
1317 netpoll_poll_disable(dev);
1319 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1321 clear_bit(__LINK_STATE_START, &dev->state);
1323 /* Synchronize to scheduled poll. We cannot touch poll list, it
1324 * can be even on different cpu. So just clear netif_running().
1326 * dev->stop() will invoke napi_disable() on all of it's
1327 * napi_struct instances on this device.
1329 smp_mb__after_atomic(); /* Commit netif_running(). */
1332 dev_deactivate_many(head);
1334 list_for_each_entry(dev, head, close_list) {
1335 const struct net_device_ops *ops = dev->netdev_ops;
1338 * Call the device specific close. This cannot fail.
1339 * Only if device is UP
1341 * We allow it to be called even after a DETACH hot-plug
1347 dev->flags &= ~IFF_UP;
1348 net_dmaengine_put();
1349 netpoll_poll_enable(dev);
1355 static int __dev_close(struct net_device *dev)
1360 list_add(&dev->close_list, &single);
1361 retval = __dev_close_many(&single);
1367 static int dev_close_many(struct list_head *head)
1369 struct net_device *dev, *tmp;
1371 /* Remove the devices that don't need to be closed */
1372 list_for_each_entry_safe(dev, tmp, head, close_list)
1373 if (!(dev->flags & IFF_UP))
1374 list_del_init(&dev->close_list);
1376 __dev_close_many(head);
1378 list_for_each_entry_safe(dev, tmp, head, close_list) {
1379 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1380 call_netdevice_notifiers(NETDEV_DOWN, dev);
1381 list_del_init(&dev->close_list);
1388 * dev_close - shutdown an interface.
1389 * @dev: device to shutdown
1391 * This function moves an active device into down state. A
1392 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1393 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1396 int dev_close(struct net_device *dev)
1398 if (dev->flags & IFF_UP) {
1401 list_add(&dev->close_list, &single);
1402 dev_close_many(&single);
1407 EXPORT_SYMBOL(dev_close);
1411 * dev_disable_lro - disable Large Receive Offload on a device
1414 * Disable Large Receive Offload (LRO) on a net device. Must be
1415 * called under RTNL. This is needed if received packets may be
1416 * forwarded to another interface.
1418 void dev_disable_lro(struct net_device *dev)
1421 * If we're trying to disable lro on a vlan device
1422 * use the underlying physical device instead
1424 if (is_vlan_dev(dev))
1425 dev = vlan_dev_real_dev(dev);
1427 /* the same for macvlan devices */
1428 if (netif_is_macvlan(dev))
1429 dev = macvlan_dev_real_dev(dev);
1431 dev->wanted_features &= ~NETIF_F_LRO;
1432 netdev_update_features(dev);
1434 if (unlikely(dev->features & NETIF_F_LRO))
1435 netdev_WARN(dev, "failed to disable LRO!\n");
1437 EXPORT_SYMBOL(dev_disable_lro);
1439 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1440 struct net_device *dev)
1442 struct netdev_notifier_info info;
1444 netdev_notifier_info_init(&info, dev);
1445 return nb->notifier_call(nb, val, &info);
1448 static int dev_boot_phase = 1;
1451 * register_netdevice_notifier - register a network notifier block
1454 * Register a notifier to be called when network device events occur.
1455 * The notifier passed is linked into the kernel structures and must
1456 * not be reused until it has been unregistered. A negative errno code
1457 * is returned on a failure.
1459 * When registered all registration and up events are replayed
1460 * to the new notifier to allow device to have a race free
1461 * view of the network device list.
1464 int register_netdevice_notifier(struct notifier_block *nb)
1466 struct net_device *dev;
1467 struct net_device *last;
1472 err = raw_notifier_chain_register(&netdev_chain, nb);
1478 for_each_netdev(net, dev) {
1479 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1480 err = notifier_to_errno(err);
1484 if (!(dev->flags & IFF_UP))
1487 call_netdevice_notifier(nb, NETDEV_UP, dev);
1498 for_each_netdev(net, dev) {
1502 if (dev->flags & IFF_UP) {
1503 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1505 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1507 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1512 raw_notifier_chain_unregister(&netdev_chain, nb);
1515 EXPORT_SYMBOL(register_netdevice_notifier);
1518 * unregister_netdevice_notifier - unregister a network notifier block
1521 * Unregister a notifier previously registered by
1522 * register_netdevice_notifier(). The notifier is unlinked into the
1523 * kernel structures and may then be reused. A negative errno code
1524 * is returned on a failure.
1526 * After unregistering unregister and down device events are synthesized
1527 * for all devices on the device list to the removed notifier to remove
1528 * the need for special case cleanup code.
1531 int unregister_netdevice_notifier(struct notifier_block *nb)
1533 struct net_device *dev;
1538 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1543 for_each_netdev(net, dev) {
1544 if (dev->flags & IFF_UP) {
1545 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1547 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1549 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1556 EXPORT_SYMBOL(unregister_netdevice_notifier);
1559 * call_netdevice_notifiers_info - call all network notifier blocks
1560 * @val: value passed unmodified to notifier function
1561 * @dev: net_device pointer passed unmodified to notifier function
1562 * @info: notifier information data
1564 * Call all network notifier blocks. Parameters and return value
1565 * are as for raw_notifier_call_chain().
1568 static int call_netdevice_notifiers_info(unsigned long val,
1569 struct net_device *dev,
1570 struct netdev_notifier_info *info)
1573 netdev_notifier_info_init(info, dev);
1574 return raw_notifier_call_chain(&netdev_chain, val, info);
1578 * call_netdevice_notifiers - call all network notifier blocks
1579 * @val: value passed unmodified to notifier function
1580 * @dev: net_device pointer passed unmodified to notifier function
1582 * Call all network notifier blocks. Parameters and return value
1583 * are as for raw_notifier_call_chain().
1586 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1588 struct netdev_notifier_info info;
1590 return call_netdevice_notifiers_info(val, dev, &info);
1592 EXPORT_SYMBOL(call_netdevice_notifiers);
1594 static struct static_key netstamp_needed __read_mostly;
1595 #ifdef HAVE_JUMP_LABEL
1596 /* We are not allowed to call static_key_slow_dec() from irq context
1597 * If net_disable_timestamp() is called from irq context, defer the
1598 * static_key_slow_dec() calls.
1600 static atomic_t netstamp_needed_deferred;
1603 void net_enable_timestamp(void)
1605 #ifdef HAVE_JUMP_LABEL
1606 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1610 static_key_slow_dec(&netstamp_needed);
1614 static_key_slow_inc(&netstamp_needed);
1616 EXPORT_SYMBOL(net_enable_timestamp);
1618 void net_disable_timestamp(void)
1620 #ifdef HAVE_JUMP_LABEL
1621 if (in_interrupt()) {
1622 atomic_inc(&netstamp_needed_deferred);
1626 static_key_slow_dec(&netstamp_needed);
1628 EXPORT_SYMBOL(net_disable_timestamp);
1630 static inline void net_timestamp_set(struct sk_buff *skb)
1632 skb->tstamp.tv64 = 0;
1633 if (static_key_false(&netstamp_needed))
1634 __net_timestamp(skb);
1637 #define net_timestamp_check(COND, SKB) \
1638 if (static_key_false(&netstamp_needed)) { \
1639 if ((COND) && !(SKB)->tstamp.tv64) \
1640 __net_timestamp(SKB); \
1643 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1647 if (!(dev->flags & IFF_UP))
1650 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1651 if (skb->len <= len)
1654 /* if TSO is enabled, we don't care about the length as the packet
1655 * could be forwarded without being segmented before
1657 if (skb_is_gso(skb))
1662 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1665 * dev_forward_skb - loopback an skb to another netif
1667 * @dev: destination network device
1668 * @skb: buffer to forward
1671 * NET_RX_SUCCESS (no congestion)
1672 * NET_RX_DROP (packet was dropped, but freed)
1674 * dev_forward_skb can be used for injecting an skb from the
1675 * start_xmit function of one device into the receive queue
1676 * of another device.
1678 * The receiving device may be in another namespace, so
1679 * we have to clear all information in the skb that could
1680 * impact namespace isolation.
1682 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1684 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1685 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1686 atomic_long_inc(&dev->rx_dropped);
1692 if (unlikely(!is_skb_forwardable(dev, skb))) {
1693 atomic_long_inc(&dev->rx_dropped);
1698 skb_scrub_packet(skb, true);
1699 skb->protocol = eth_type_trans(skb, dev);
1701 return netif_rx_internal(skb);
1703 EXPORT_SYMBOL_GPL(dev_forward_skb);
1705 static inline int deliver_skb(struct sk_buff *skb,
1706 struct packet_type *pt_prev,
1707 struct net_device *orig_dev)
1709 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1711 atomic_inc(&skb->users);
1712 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1715 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1717 if (!ptype->af_packet_priv || !skb->sk)
1720 if (ptype->id_match)
1721 return ptype->id_match(ptype, skb->sk);
1722 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1729 * Support routine. Sends outgoing frames to any network
1730 * taps currently in use.
1733 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1735 struct packet_type *ptype;
1736 struct sk_buff *skb2 = NULL;
1737 struct packet_type *pt_prev = NULL;
1740 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1741 /* Never send packets back to the socket
1742 * they originated from - MvS (miquels@drinkel.ow.org)
1744 if ((ptype->dev == dev || !ptype->dev) &&
1745 (!skb_loop_sk(ptype, skb))) {
1747 deliver_skb(skb2, pt_prev, skb->dev);
1752 skb2 = skb_clone(skb, GFP_ATOMIC);
1756 net_timestamp_set(skb2);
1758 /* skb->nh should be correctly
1759 set by sender, so that the second statement is
1760 just protection against buggy protocols.
1762 skb_reset_mac_header(skb2);
1764 if (skb_network_header(skb2) < skb2->data ||
1765 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1766 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1767 ntohs(skb2->protocol),
1769 skb_reset_network_header(skb2);
1772 skb2->transport_header = skb2->network_header;
1773 skb2->pkt_type = PACKET_OUTGOING;
1778 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1783 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1784 * @dev: Network device
1785 * @txq: number of queues available
1787 * If real_num_tx_queues is changed the tc mappings may no longer be
1788 * valid. To resolve this verify the tc mapping remains valid and if
1789 * not NULL the mapping. With no priorities mapping to this
1790 * offset/count pair it will no longer be used. In the worst case TC0
1791 * is invalid nothing can be done so disable priority mappings. If is
1792 * expected that drivers will fix this mapping if they can before
1793 * calling netif_set_real_num_tx_queues.
1795 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1798 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1800 /* If TC0 is invalidated disable TC mapping */
1801 if (tc->offset + tc->count > txq) {
1802 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1807 /* Invalidated prio to tc mappings set to TC0 */
1808 for (i = 1; i < TC_BITMASK + 1; i++) {
1809 int q = netdev_get_prio_tc_map(dev, i);
1811 tc = &dev->tc_to_txq[q];
1812 if (tc->offset + tc->count > txq) {
1813 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1815 netdev_set_prio_tc_map(dev, i, 0);
1821 static DEFINE_MUTEX(xps_map_mutex);
1822 #define xmap_dereference(P) \
1823 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1825 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1828 struct xps_map *map = NULL;
1832 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1834 for (pos = 0; map && pos < map->len; pos++) {
1835 if (map->queues[pos] == index) {
1837 map->queues[pos] = map->queues[--map->len];
1839 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1840 kfree_rcu(map, rcu);
1850 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1852 struct xps_dev_maps *dev_maps;
1854 bool active = false;
1856 mutex_lock(&xps_map_mutex);
1857 dev_maps = xmap_dereference(dev->xps_maps);
1862 for_each_possible_cpu(cpu) {
1863 for (i = index; i < dev->num_tx_queues; i++) {
1864 if (!remove_xps_queue(dev_maps, cpu, i))
1867 if (i == dev->num_tx_queues)
1872 RCU_INIT_POINTER(dev->xps_maps, NULL);
1873 kfree_rcu(dev_maps, rcu);
1876 for (i = index; i < dev->num_tx_queues; i++)
1877 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1881 mutex_unlock(&xps_map_mutex);
1884 static struct xps_map *expand_xps_map(struct xps_map *map,
1887 struct xps_map *new_map;
1888 int alloc_len = XPS_MIN_MAP_ALLOC;
1891 for (pos = 0; map && pos < map->len; pos++) {
1892 if (map->queues[pos] != index)
1897 /* Need to add queue to this CPU's existing map */
1899 if (pos < map->alloc_len)
1902 alloc_len = map->alloc_len * 2;
1905 /* Need to allocate new map to store queue on this CPU's map */
1906 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1911 for (i = 0; i < pos; i++)
1912 new_map->queues[i] = map->queues[i];
1913 new_map->alloc_len = alloc_len;
1919 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1922 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1923 struct xps_map *map, *new_map;
1924 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1925 int cpu, numa_node_id = -2;
1926 bool active = false;
1928 mutex_lock(&xps_map_mutex);
1930 dev_maps = xmap_dereference(dev->xps_maps);
1932 /* allocate memory for queue storage */
1933 for_each_online_cpu(cpu) {
1934 if (!cpumask_test_cpu(cpu, mask))
1938 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1939 if (!new_dev_maps) {
1940 mutex_unlock(&xps_map_mutex);
1944 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1947 map = expand_xps_map(map, cpu, index);
1951 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1955 goto out_no_new_maps;
1957 for_each_possible_cpu(cpu) {
1958 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1959 /* add queue to CPU maps */
1962 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1963 while ((pos < map->len) && (map->queues[pos] != index))
1966 if (pos == map->len)
1967 map->queues[map->len++] = index;
1969 if (numa_node_id == -2)
1970 numa_node_id = cpu_to_node(cpu);
1971 else if (numa_node_id != cpu_to_node(cpu))
1974 } else if (dev_maps) {
1975 /* fill in the new device map from the old device map */
1976 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1977 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1982 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1984 /* Cleanup old maps */
1986 for_each_possible_cpu(cpu) {
1987 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1988 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1989 if (map && map != new_map)
1990 kfree_rcu(map, rcu);
1993 kfree_rcu(dev_maps, rcu);
1996 dev_maps = new_dev_maps;
2000 /* update Tx queue numa node */
2001 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2002 (numa_node_id >= 0) ? numa_node_id :
2008 /* removes queue from unused CPUs */
2009 for_each_possible_cpu(cpu) {
2010 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2013 if (remove_xps_queue(dev_maps, cpu, index))
2017 /* free map if not active */
2019 RCU_INIT_POINTER(dev->xps_maps, NULL);
2020 kfree_rcu(dev_maps, rcu);
2024 mutex_unlock(&xps_map_mutex);
2028 /* remove any maps that we added */
2029 for_each_possible_cpu(cpu) {
2030 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2031 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2033 if (new_map && new_map != map)
2037 mutex_unlock(&xps_map_mutex);
2039 kfree(new_dev_maps);
2042 EXPORT_SYMBOL(netif_set_xps_queue);
2046 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2047 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2049 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2053 if (txq < 1 || txq > dev->num_tx_queues)
2056 if (dev->reg_state == NETREG_REGISTERED ||
2057 dev->reg_state == NETREG_UNREGISTERING) {
2060 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2066 netif_setup_tc(dev, txq);
2068 if (txq < dev->real_num_tx_queues) {
2069 qdisc_reset_all_tx_gt(dev, txq);
2071 netif_reset_xps_queues_gt(dev, txq);
2076 dev->real_num_tx_queues = txq;
2079 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2083 * netif_set_real_num_rx_queues - set actual number of RX queues used
2084 * @dev: Network device
2085 * @rxq: Actual number of RX queues
2087 * This must be called either with the rtnl_lock held or before
2088 * registration of the net device. Returns 0 on success, or a
2089 * negative error code. If called before registration, it always
2092 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2096 if (rxq < 1 || rxq > dev->num_rx_queues)
2099 if (dev->reg_state == NETREG_REGISTERED) {
2102 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2108 dev->real_num_rx_queues = rxq;
2111 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2115 * netif_get_num_default_rss_queues - default number of RSS queues
2117 * This routine should set an upper limit on the number of RSS queues
2118 * used by default by multiqueue devices.
2120 int netif_get_num_default_rss_queues(void)
2122 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2124 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2126 static inline void __netif_reschedule(struct Qdisc *q)
2128 struct softnet_data *sd;
2129 unsigned long flags;
2131 local_irq_save(flags);
2132 sd = &__get_cpu_var(softnet_data);
2133 q->next_sched = NULL;
2134 *sd->output_queue_tailp = q;
2135 sd->output_queue_tailp = &q->next_sched;
2136 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2137 local_irq_restore(flags);
2140 void __netif_schedule(struct Qdisc *q)
2142 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2143 __netif_reschedule(q);
2145 EXPORT_SYMBOL(__netif_schedule);
2147 struct dev_kfree_skb_cb {
2148 enum skb_free_reason reason;
2151 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2153 return (struct dev_kfree_skb_cb *)skb->cb;
2156 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2158 unsigned long flags;
2160 if (likely(atomic_read(&skb->users) == 1)) {
2162 atomic_set(&skb->users, 0);
2163 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2166 get_kfree_skb_cb(skb)->reason = reason;
2167 local_irq_save(flags);
2168 skb->next = __this_cpu_read(softnet_data.completion_queue);
2169 __this_cpu_write(softnet_data.completion_queue, skb);
2170 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2171 local_irq_restore(flags);
2173 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2175 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2177 if (in_irq() || irqs_disabled())
2178 __dev_kfree_skb_irq(skb, reason);
2182 EXPORT_SYMBOL(__dev_kfree_skb_any);
2186 * netif_device_detach - mark device as removed
2187 * @dev: network device
2189 * Mark device as removed from system and therefore no longer available.
2191 void netif_device_detach(struct net_device *dev)
2193 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2194 netif_running(dev)) {
2195 netif_tx_stop_all_queues(dev);
2198 EXPORT_SYMBOL(netif_device_detach);
2201 * netif_device_attach - mark device as attached
2202 * @dev: network device
2204 * Mark device as attached from system and restart if needed.
2206 void netif_device_attach(struct net_device *dev)
2208 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2209 netif_running(dev)) {
2210 netif_tx_wake_all_queues(dev);
2211 __netdev_watchdog_up(dev);
2214 EXPORT_SYMBOL(netif_device_attach);
2216 static void skb_warn_bad_offload(const struct sk_buff *skb)
2218 static const netdev_features_t null_features = 0;
2219 struct net_device *dev = skb->dev;
2220 const char *driver = "";
2222 if (!net_ratelimit())
2225 if (dev && dev->dev.parent)
2226 driver = dev_driver_string(dev->dev.parent);
2228 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2229 "gso_type=%d ip_summed=%d\n",
2230 driver, dev ? &dev->features : &null_features,
2231 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2232 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2233 skb_shinfo(skb)->gso_type, skb->ip_summed);
2237 * Invalidate hardware checksum when packet is to be mangled, and
2238 * complete checksum manually on outgoing path.
2240 int skb_checksum_help(struct sk_buff *skb)
2243 int ret = 0, offset;
2245 if (skb->ip_summed == CHECKSUM_COMPLETE)
2246 goto out_set_summed;
2248 if (unlikely(skb_shinfo(skb)->gso_size)) {
2249 skb_warn_bad_offload(skb);
2253 /* Before computing a checksum, we should make sure no frag could
2254 * be modified by an external entity : checksum could be wrong.
2256 if (skb_has_shared_frag(skb)) {
2257 ret = __skb_linearize(skb);
2262 offset = skb_checksum_start_offset(skb);
2263 BUG_ON(offset >= skb_headlen(skb));
2264 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2266 offset += skb->csum_offset;
2267 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2269 if (skb_cloned(skb) &&
2270 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2271 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2276 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2278 skb->ip_summed = CHECKSUM_NONE;
2282 EXPORT_SYMBOL(skb_checksum_help);
2284 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2286 __be16 type = skb->protocol;
2287 int vlan_depth = skb->mac_len;
2289 /* Tunnel gso handlers can set protocol to ethernet. */
2290 if (type == htons(ETH_P_TEB)) {
2293 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2296 eth = (struct ethhdr *)skb_mac_header(skb);
2297 type = eth->h_proto;
2300 while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2301 struct vlan_hdr *vh;
2303 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2306 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2307 type = vh->h_vlan_encapsulated_proto;
2308 vlan_depth += VLAN_HLEN;
2311 *depth = vlan_depth;
2317 * skb_mac_gso_segment - mac layer segmentation handler.
2318 * @skb: buffer to segment
2319 * @features: features for the output path (see dev->features)
2321 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2322 netdev_features_t features)
2324 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2325 struct packet_offload *ptype;
2326 int vlan_depth = skb->mac_len;
2327 __be16 type = skb_network_protocol(skb, &vlan_depth);
2329 if (unlikely(!type))
2330 return ERR_PTR(-EINVAL);
2332 __skb_pull(skb, vlan_depth);
2335 list_for_each_entry_rcu(ptype, &offload_base, list) {
2336 if (ptype->type == type && ptype->callbacks.gso_segment) {
2337 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2340 err = ptype->callbacks.gso_send_check(skb);
2341 segs = ERR_PTR(err);
2342 if (err || skb_gso_ok(skb, features))
2344 __skb_push(skb, (skb->data -
2345 skb_network_header(skb)));
2347 segs = ptype->callbacks.gso_segment(skb, features);
2353 __skb_push(skb, skb->data - skb_mac_header(skb));
2357 EXPORT_SYMBOL(skb_mac_gso_segment);
2360 /* openvswitch calls this on rx path, so we need a different check.
2362 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2365 return skb->ip_summed != CHECKSUM_PARTIAL;
2367 return skb->ip_summed == CHECKSUM_NONE;
2371 * __skb_gso_segment - Perform segmentation on skb.
2372 * @skb: buffer to segment
2373 * @features: features for the output path (see dev->features)
2374 * @tx_path: whether it is called in TX path
2376 * This function segments the given skb and returns a list of segments.
2378 * It may return NULL if the skb requires no segmentation. This is
2379 * only possible when GSO is used for verifying header integrity.
2381 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2382 netdev_features_t features, bool tx_path)
2384 if (unlikely(skb_needs_check(skb, tx_path))) {
2387 skb_warn_bad_offload(skb);
2389 if (skb_header_cloned(skb) &&
2390 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2391 return ERR_PTR(err);
2394 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2395 SKB_GSO_CB(skb)->encap_level = 0;
2397 skb_reset_mac_header(skb);
2398 skb_reset_mac_len(skb);
2400 return skb_mac_gso_segment(skb, features);
2402 EXPORT_SYMBOL(__skb_gso_segment);
2404 /* Take action when hardware reception checksum errors are detected. */
2406 void netdev_rx_csum_fault(struct net_device *dev)
2408 if (net_ratelimit()) {
2409 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2413 EXPORT_SYMBOL(netdev_rx_csum_fault);
2416 /* Actually, we should eliminate this check as soon as we know, that:
2417 * 1. IOMMU is present and allows to map all the memory.
2418 * 2. No high memory really exists on this machine.
2421 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2423 #ifdef CONFIG_HIGHMEM
2425 if (!(dev->features & NETIF_F_HIGHDMA)) {
2426 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2427 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2428 if (PageHighMem(skb_frag_page(frag)))
2433 if (PCI_DMA_BUS_IS_PHYS) {
2434 struct device *pdev = dev->dev.parent;
2438 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2439 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2440 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2441 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2450 void (*destructor)(struct sk_buff *skb);
2453 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2455 static void dev_gso_skb_destructor(struct sk_buff *skb)
2457 struct dev_gso_cb *cb;
2459 kfree_skb_list(skb->next);
2462 cb = DEV_GSO_CB(skb);
2464 cb->destructor(skb);
2468 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2469 * @skb: buffer to segment
2470 * @features: device features as applicable to this skb
2472 * This function segments the given skb and stores the list of segments
2475 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2477 struct sk_buff *segs;
2479 segs = skb_gso_segment(skb, features);
2481 /* Verifying header integrity only. */
2486 return PTR_ERR(segs);
2489 DEV_GSO_CB(skb)->destructor = skb->destructor;
2490 skb->destructor = dev_gso_skb_destructor;
2495 static netdev_features_t harmonize_features(struct sk_buff *skb,
2496 netdev_features_t features)
2500 if (skb->ip_summed != CHECKSUM_NONE &&
2501 !can_checksum_protocol(features, skb_network_protocol(skb, &tmp))) {
2502 features &= ~NETIF_F_ALL_CSUM;
2503 } else if (illegal_highdma(skb->dev, skb)) {
2504 features &= ~NETIF_F_SG;
2510 netdev_features_t netif_skb_features(struct sk_buff *skb)
2512 __be16 protocol = skb->protocol;
2513 netdev_features_t features = skb->dev->features;
2515 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2516 features &= ~NETIF_F_GSO_MASK;
2518 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2519 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2520 protocol = veh->h_vlan_encapsulated_proto;
2521 } else if (!vlan_tx_tag_present(skb)) {
2522 return harmonize_features(skb, features);
2525 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2526 NETIF_F_HW_VLAN_STAG_TX);
2528 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2529 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2530 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2531 NETIF_F_HW_VLAN_STAG_TX;
2533 return harmonize_features(skb, features);
2535 EXPORT_SYMBOL(netif_skb_features);
2537 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2538 struct netdev_queue *txq)
2540 const struct net_device_ops *ops = dev->netdev_ops;
2541 int rc = NETDEV_TX_OK;
2542 unsigned int skb_len;
2544 if (likely(!skb->next)) {
2545 netdev_features_t features;
2548 * If device doesn't need skb->dst, release it right now while
2549 * its hot in this cpu cache
2551 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2554 features = netif_skb_features(skb);
2556 if (vlan_tx_tag_present(skb) &&
2557 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2558 skb = __vlan_put_tag(skb, skb->vlan_proto,
2559 vlan_tx_tag_get(skb));
2566 /* If encapsulation offload request, verify we are testing
2567 * hardware encapsulation features instead of standard
2568 * features for the netdev
2570 if (skb->encapsulation)
2571 features &= dev->hw_enc_features;
2573 if (netif_needs_gso(skb, features)) {
2574 if (unlikely(dev_gso_segment(skb, features)))
2579 if (skb_needs_linearize(skb, features) &&
2580 __skb_linearize(skb))
2583 /* If packet is not checksummed and device does not
2584 * support checksumming for this protocol, complete
2585 * checksumming here.
2587 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2588 if (skb->encapsulation)
2589 skb_set_inner_transport_header(skb,
2590 skb_checksum_start_offset(skb));
2592 skb_set_transport_header(skb,
2593 skb_checksum_start_offset(skb));
2594 if (!(features & NETIF_F_ALL_CSUM) &&
2595 skb_checksum_help(skb))
2600 if (!list_empty(&ptype_all))
2601 dev_queue_xmit_nit(skb, dev);
2604 trace_net_dev_start_xmit(skb, dev);
2605 rc = ops->ndo_start_xmit(skb, dev);
2606 trace_net_dev_xmit(skb, rc, dev, skb_len);
2607 if (rc == NETDEV_TX_OK)
2608 txq_trans_update(txq);
2614 struct sk_buff *nskb = skb->next;
2616 skb->next = nskb->next;
2619 if (!list_empty(&ptype_all))
2620 dev_queue_xmit_nit(nskb, dev);
2622 skb_len = nskb->len;
2623 trace_net_dev_start_xmit(nskb, dev);
2624 rc = ops->ndo_start_xmit(nskb, dev);
2625 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2626 if (unlikely(rc != NETDEV_TX_OK)) {
2627 if (rc & ~NETDEV_TX_MASK)
2628 goto out_kfree_gso_skb;
2629 nskb->next = skb->next;
2633 txq_trans_update(txq);
2634 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2635 return NETDEV_TX_BUSY;
2636 } while (skb->next);
2639 if (likely(skb->next == NULL)) {
2640 skb->destructor = DEV_GSO_CB(skb)->destructor;
2649 EXPORT_SYMBOL_GPL(dev_hard_start_xmit);
2651 static void qdisc_pkt_len_init(struct sk_buff *skb)
2653 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2655 qdisc_skb_cb(skb)->pkt_len = skb->len;
2657 /* To get more precise estimation of bytes sent on wire,
2658 * we add to pkt_len the headers size of all segments
2660 if (shinfo->gso_size) {
2661 unsigned int hdr_len;
2662 u16 gso_segs = shinfo->gso_segs;
2664 /* mac layer + network layer */
2665 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2667 /* + transport layer */
2668 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2669 hdr_len += tcp_hdrlen(skb);
2671 hdr_len += sizeof(struct udphdr);
2673 if (shinfo->gso_type & SKB_GSO_DODGY)
2674 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2677 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2681 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2682 struct net_device *dev,
2683 struct netdev_queue *txq)
2685 spinlock_t *root_lock = qdisc_lock(q);
2689 qdisc_pkt_len_init(skb);
2690 qdisc_calculate_pkt_len(skb, q);
2692 * Heuristic to force contended enqueues to serialize on a
2693 * separate lock before trying to get qdisc main lock.
2694 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2695 * and dequeue packets faster.
2697 contended = qdisc_is_running(q);
2698 if (unlikely(contended))
2699 spin_lock(&q->busylock);
2701 spin_lock(root_lock);
2702 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2705 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2706 qdisc_run_begin(q)) {
2708 * This is a work-conserving queue; there are no old skbs
2709 * waiting to be sent out; and the qdisc is not running -
2710 * xmit the skb directly.
2712 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2715 qdisc_bstats_update(q, skb);
2717 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2718 if (unlikely(contended)) {
2719 spin_unlock(&q->busylock);
2726 rc = NET_XMIT_SUCCESS;
2729 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2730 if (qdisc_run_begin(q)) {
2731 if (unlikely(contended)) {
2732 spin_unlock(&q->busylock);
2738 spin_unlock(root_lock);
2739 if (unlikely(contended))
2740 spin_unlock(&q->busylock);
2744 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2745 static void skb_update_prio(struct sk_buff *skb)
2747 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2749 if (!skb->priority && skb->sk && map) {
2750 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2752 if (prioidx < map->priomap_len)
2753 skb->priority = map->priomap[prioidx];
2757 #define skb_update_prio(skb)
2760 static DEFINE_PER_CPU(int, xmit_recursion);
2761 #define RECURSION_LIMIT 10
2764 * dev_loopback_xmit - loop back @skb
2765 * @skb: buffer to transmit
2767 int dev_loopback_xmit(struct sk_buff *skb)
2769 skb_reset_mac_header(skb);
2770 __skb_pull(skb, skb_network_offset(skb));
2771 skb->pkt_type = PACKET_LOOPBACK;
2772 skb->ip_summed = CHECKSUM_UNNECESSARY;
2773 WARN_ON(!skb_dst(skb));
2778 EXPORT_SYMBOL(dev_loopback_xmit);
2781 * __dev_queue_xmit - transmit a buffer
2782 * @skb: buffer to transmit
2783 * @accel_priv: private data used for L2 forwarding offload
2785 * Queue a buffer for transmission to a network device. The caller must
2786 * have set the device and priority and built the buffer before calling
2787 * this function. The function can be called from an interrupt.
2789 * A negative errno code is returned on a failure. A success does not
2790 * guarantee the frame will be transmitted as it may be dropped due
2791 * to congestion or traffic shaping.
2793 * -----------------------------------------------------------------------------------
2794 * I notice this method can also return errors from the queue disciplines,
2795 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2798 * Regardless of the return value, the skb is consumed, so it is currently
2799 * difficult to retry a send to this method. (You can bump the ref count
2800 * before sending to hold a reference for retry if you are careful.)
2802 * When calling this method, interrupts MUST be enabled. This is because
2803 * the BH enable code must have IRQs enabled so that it will not deadlock.
2806 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2808 struct net_device *dev = skb->dev;
2809 struct netdev_queue *txq;
2813 skb_reset_mac_header(skb);
2815 /* Disable soft irqs for various locks below. Also
2816 * stops preemption for RCU.
2820 skb_update_prio(skb);
2822 txq = netdev_pick_tx(dev, skb, accel_priv);
2823 q = rcu_dereference_bh(txq->qdisc);
2825 #ifdef CONFIG_NET_CLS_ACT
2826 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2828 trace_net_dev_queue(skb);
2830 rc = __dev_xmit_skb(skb, q, dev, txq);
2834 /* The device has no queue. Common case for software devices:
2835 loopback, all the sorts of tunnels...
2837 Really, it is unlikely that netif_tx_lock protection is necessary
2838 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2840 However, it is possible, that they rely on protection
2843 Check this and shot the lock. It is not prone from deadlocks.
2844 Either shot noqueue qdisc, it is even simpler 8)
2846 if (dev->flags & IFF_UP) {
2847 int cpu = smp_processor_id(); /* ok because BHs are off */
2849 if (txq->xmit_lock_owner != cpu) {
2851 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2852 goto recursion_alert;
2854 HARD_TX_LOCK(dev, txq, cpu);
2856 if (!netif_xmit_stopped(txq)) {
2857 __this_cpu_inc(xmit_recursion);
2858 rc = dev_hard_start_xmit(skb, dev, txq);
2859 __this_cpu_dec(xmit_recursion);
2860 if (dev_xmit_complete(rc)) {
2861 HARD_TX_UNLOCK(dev, txq);
2865 HARD_TX_UNLOCK(dev, txq);
2866 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2869 /* Recursion is detected! It is possible,
2873 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2879 rcu_read_unlock_bh();
2881 atomic_long_inc(&dev->tx_dropped);
2885 rcu_read_unlock_bh();
2889 int dev_queue_xmit(struct sk_buff *skb)
2891 return __dev_queue_xmit(skb, NULL);
2893 EXPORT_SYMBOL(dev_queue_xmit);
2895 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
2897 return __dev_queue_xmit(skb, accel_priv);
2899 EXPORT_SYMBOL(dev_queue_xmit_accel);
2902 /*=======================================================================
2904 =======================================================================*/
2906 int netdev_max_backlog __read_mostly = 1000;
2907 EXPORT_SYMBOL(netdev_max_backlog);
2909 int netdev_tstamp_prequeue __read_mostly = 1;
2910 int netdev_budget __read_mostly = 300;
2911 int weight_p __read_mostly = 64; /* old backlog weight */
2913 /* Called with irq disabled */
2914 static inline void ____napi_schedule(struct softnet_data *sd,
2915 struct napi_struct *napi)
2917 list_add_tail(&napi->poll_list, &sd->poll_list);
2918 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2923 /* One global table that all flow-based protocols share. */
2924 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2925 EXPORT_SYMBOL(rps_sock_flow_table);
2927 struct static_key rps_needed __read_mostly;
2929 static struct rps_dev_flow *
2930 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2931 struct rps_dev_flow *rflow, u16 next_cpu)
2933 if (next_cpu != RPS_NO_CPU) {
2934 #ifdef CONFIG_RFS_ACCEL
2935 struct netdev_rx_queue *rxqueue;
2936 struct rps_dev_flow_table *flow_table;
2937 struct rps_dev_flow *old_rflow;
2942 /* Should we steer this flow to a different hardware queue? */
2943 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2944 !(dev->features & NETIF_F_NTUPLE))
2946 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2947 if (rxq_index == skb_get_rx_queue(skb))
2950 rxqueue = dev->_rx + rxq_index;
2951 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2954 flow_id = skb_get_hash(skb) & flow_table->mask;
2955 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2956 rxq_index, flow_id);
2960 rflow = &flow_table->flows[flow_id];
2962 if (old_rflow->filter == rflow->filter)
2963 old_rflow->filter = RPS_NO_FILTER;
2967 per_cpu(softnet_data, next_cpu).input_queue_head;
2970 rflow->cpu = next_cpu;
2975 * get_rps_cpu is called from netif_receive_skb and returns the target
2976 * CPU from the RPS map of the receiving queue for a given skb.
2977 * rcu_read_lock must be held on entry.
2979 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2980 struct rps_dev_flow **rflowp)
2982 struct netdev_rx_queue *rxqueue;
2983 struct rps_map *map;
2984 struct rps_dev_flow_table *flow_table;
2985 struct rps_sock_flow_table *sock_flow_table;
2990 if (skb_rx_queue_recorded(skb)) {
2991 u16 index = skb_get_rx_queue(skb);
2992 if (unlikely(index >= dev->real_num_rx_queues)) {
2993 WARN_ONCE(dev->real_num_rx_queues > 1,
2994 "%s received packet on queue %u, but number "
2995 "of RX queues is %u\n",
2996 dev->name, index, dev->real_num_rx_queues);
2999 rxqueue = dev->_rx + index;
3003 map = rcu_dereference(rxqueue->rps_map);
3005 if (map->len == 1 &&
3006 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3007 tcpu = map->cpus[0];
3008 if (cpu_online(tcpu))
3012 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3016 skb_reset_network_header(skb);
3017 hash = skb_get_hash(skb);
3021 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3022 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3023 if (flow_table && sock_flow_table) {
3025 struct rps_dev_flow *rflow;
3027 rflow = &flow_table->flows[hash & flow_table->mask];
3030 next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask];
3033 * If the desired CPU (where last recvmsg was done) is
3034 * different from current CPU (one in the rx-queue flow
3035 * table entry), switch if one of the following holds:
3036 * - Current CPU is unset (equal to RPS_NO_CPU).
3037 * - Current CPU is offline.
3038 * - The current CPU's queue tail has advanced beyond the
3039 * last packet that was enqueued using this table entry.
3040 * This guarantees that all previous packets for the flow
3041 * have been dequeued, thus preserving in order delivery.
3043 if (unlikely(tcpu != next_cpu) &&
3044 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3045 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3046 rflow->last_qtail)) >= 0)) {
3048 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3051 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3059 tcpu = map->cpus[((u64) hash * map->len) >> 32];
3061 if (cpu_online(tcpu)) {
3071 #ifdef CONFIG_RFS_ACCEL
3074 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3075 * @dev: Device on which the filter was set
3076 * @rxq_index: RX queue index
3077 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3078 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3080 * Drivers that implement ndo_rx_flow_steer() should periodically call
3081 * this function for each installed filter and remove the filters for
3082 * which it returns %true.
3084 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3085 u32 flow_id, u16 filter_id)
3087 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3088 struct rps_dev_flow_table *flow_table;
3089 struct rps_dev_flow *rflow;
3094 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3095 if (flow_table && flow_id <= flow_table->mask) {
3096 rflow = &flow_table->flows[flow_id];
3097 cpu = ACCESS_ONCE(rflow->cpu);
3098 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3099 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3100 rflow->last_qtail) <
3101 (int)(10 * flow_table->mask)))
3107 EXPORT_SYMBOL(rps_may_expire_flow);
3109 #endif /* CONFIG_RFS_ACCEL */
3111 /* Called from hardirq (IPI) context */
3112 static void rps_trigger_softirq(void *data)
3114 struct softnet_data *sd = data;
3116 ____napi_schedule(sd, &sd->backlog);
3120 #endif /* CONFIG_RPS */
3123 * Check if this softnet_data structure is another cpu one
3124 * If yes, queue it to our IPI list and return 1
3127 static int rps_ipi_queued(struct softnet_data *sd)
3130 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3133 sd->rps_ipi_next = mysd->rps_ipi_list;
3134 mysd->rps_ipi_list = sd;
3136 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3139 #endif /* CONFIG_RPS */
3143 #ifdef CONFIG_NET_FLOW_LIMIT
3144 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3147 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3149 #ifdef CONFIG_NET_FLOW_LIMIT
3150 struct sd_flow_limit *fl;
3151 struct softnet_data *sd;
3152 unsigned int old_flow, new_flow;
3154 if (qlen < (netdev_max_backlog >> 1))
3157 sd = &__get_cpu_var(softnet_data);
3160 fl = rcu_dereference(sd->flow_limit);
3162 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3163 old_flow = fl->history[fl->history_head];
3164 fl->history[fl->history_head] = new_flow;
3167 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3169 if (likely(fl->buckets[old_flow]))
3170 fl->buckets[old_flow]--;
3172 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3184 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3185 * queue (may be a remote CPU queue).
3187 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3188 unsigned int *qtail)
3190 struct softnet_data *sd;
3191 unsigned long flags;
3194 sd = &per_cpu(softnet_data, cpu);
3196 local_irq_save(flags);
3199 qlen = skb_queue_len(&sd->input_pkt_queue);
3200 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3201 if (skb_queue_len(&sd->input_pkt_queue)) {
3203 __skb_queue_tail(&sd->input_pkt_queue, skb);
3204 input_queue_tail_incr_save(sd, qtail);
3206 local_irq_restore(flags);
3207 return NET_RX_SUCCESS;
3210 /* Schedule NAPI for backlog device
3211 * We can use non atomic operation since we own the queue lock
3213 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3214 if (!rps_ipi_queued(sd))
3215 ____napi_schedule(sd, &sd->backlog);
3223 local_irq_restore(flags);
3225 atomic_long_inc(&skb->dev->rx_dropped);
3230 static int netif_rx_internal(struct sk_buff *skb)
3234 net_timestamp_check(netdev_tstamp_prequeue, skb);
3236 trace_netif_rx(skb);
3238 if (static_key_false(&rps_needed)) {
3239 struct rps_dev_flow voidflow, *rflow = &voidflow;
3245 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3247 cpu = smp_processor_id();
3249 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3257 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3264 * netif_rx - post buffer to the network code
3265 * @skb: buffer to post
3267 * This function receives a packet from a device driver and queues it for
3268 * the upper (protocol) levels to process. It always succeeds. The buffer
3269 * may be dropped during processing for congestion control or by the
3273 * NET_RX_SUCCESS (no congestion)
3274 * NET_RX_DROP (packet was dropped)
3278 int netif_rx(struct sk_buff *skb)
3280 trace_netif_rx_entry(skb);
3282 return netif_rx_internal(skb);
3284 EXPORT_SYMBOL(netif_rx);
3286 int netif_rx_ni(struct sk_buff *skb)
3290 trace_netif_rx_ni_entry(skb);
3293 err = netif_rx_internal(skb);
3294 if (local_softirq_pending())
3300 EXPORT_SYMBOL(netif_rx_ni);
3302 static void net_tx_action(struct softirq_action *h)
3304 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3306 if (sd->completion_queue) {
3307 struct sk_buff *clist;
3309 local_irq_disable();
3310 clist = sd->completion_queue;
3311 sd->completion_queue = NULL;
3315 struct sk_buff *skb = clist;
3316 clist = clist->next;
3318 WARN_ON(atomic_read(&skb->users));
3319 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3320 trace_consume_skb(skb);
3322 trace_kfree_skb(skb, net_tx_action);
3327 if (sd->output_queue) {
3330 local_irq_disable();
3331 head = sd->output_queue;
3332 sd->output_queue = NULL;
3333 sd->output_queue_tailp = &sd->output_queue;
3337 struct Qdisc *q = head;
3338 spinlock_t *root_lock;
3340 head = head->next_sched;
3342 root_lock = qdisc_lock(q);
3343 if (spin_trylock(root_lock)) {
3344 smp_mb__before_atomic();
3345 clear_bit(__QDISC_STATE_SCHED,
3348 spin_unlock(root_lock);
3350 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3352 __netif_reschedule(q);
3354 smp_mb__before_atomic();
3355 clear_bit(__QDISC_STATE_SCHED,
3363 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3364 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3365 /* This hook is defined here for ATM LANE */
3366 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3367 unsigned char *addr) __read_mostly;
3368 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3371 #ifdef CONFIG_NET_CLS_ACT
3372 /* TODO: Maybe we should just force sch_ingress to be compiled in
3373 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3374 * a compare and 2 stores extra right now if we dont have it on
3375 * but have CONFIG_NET_CLS_ACT
3376 * NOTE: This doesn't stop any functionality; if you dont have
3377 * the ingress scheduler, you just can't add policies on ingress.
3380 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3382 struct net_device *dev = skb->dev;
3383 u32 ttl = G_TC_RTTL(skb->tc_verd);
3384 int result = TC_ACT_OK;
3387 if (unlikely(MAX_RED_LOOP < ttl++)) {
3388 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3389 skb->skb_iif, dev->ifindex);
3393 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3394 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3397 if (q != &noop_qdisc) {
3398 spin_lock(qdisc_lock(q));
3399 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3400 result = qdisc_enqueue_root(skb, q);
3401 spin_unlock(qdisc_lock(q));
3407 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3408 struct packet_type **pt_prev,
3409 int *ret, struct net_device *orig_dev)
3411 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3413 if (!rxq || rxq->qdisc == &noop_qdisc)
3417 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3421 switch (ing_filter(skb, rxq)) {
3435 * netdev_rx_handler_register - register receive handler
3436 * @dev: device to register a handler for
3437 * @rx_handler: receive handler to register
3438 * @rx_handler_data: data pointer that is used by rx handler
3440 * Register a receive handler for a device. This handler will then be
3441 * called from __netif_receive_skb. A negative errno code is returned
3444 * The caller must hold the rtnl_mutex.
3446 * For a general description of rx_handler, see enum rx_handler_result.
3448 int netdev_rx_handler_register(struct net_device *dev,
3449 rx_handler_func_t *rx_handler,
3450 void *rx_handler_data)
3454 if (dev->rx_handler)
3457 /* Note: rx_handler_data must be set before rx_handler */
3458 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3459 rcu_assign_pointer(dev->rx_handler, rx_handler);
3463 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3466 * netdev_rx_handler_unregister - unregister receive handler
3467 * @dev: device to unregister a handler from
3469 * Unregister a receive handler from a device.
3471 * The caller must hold the rtnl_mutex.
3473 void netdev_rx_handler_unregister(struct net_device *dev)
3477 RCU_INIT_POINTER(dev->rx_handler, NULL);
3478 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3479 * section has a guarantee to see a non NULL rx_handler_data
3483 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3485 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3488 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3489 * the special handling of PFMEMALLOC skbs.
3491 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3493 switch (skb->protocol) {
3494 case htons(ETH_P_ARP):
3495 case htons(ETH_P_IP):
3496 case htons(ETH_P_IPV6):
3497 case htons(ETH_P_8021Q):
3498 case htons(ETH_P_8021AD):
3505 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3507 struct packet_type *ptype, *pt_prev;
3508 rx_handler_func_t *rx_handler;
3509 struct net_device *orig_dev;
3510 struct net_device *null_or_dev;
3511 bool deliver_exact = false;
3512 int ret = NET_RX_DROP;
3515 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3517 trace_netif_receive_skb(skb);
3519 orig_dev = skb->dev;
3521 skb_reset_network_header(skb);
3522 if (!skb_transport_header_was_set(skb))
3523 skb_reset_transport_header(skb);
3524 skb_reset_mac_len(skb);
3531 skb->skb_iif = skb->dev->ifindex;
3533 __this_cpu_inc(softnet_data.processed);
3535 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3536 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3537 skb = vlan_untag(skb);
3542 #ifdef CONFIG_NET_CLS_ACT
3543 if (skb->tc_verd & TC_NCLS) {
3544 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3552 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3553 if (!ptype->dev || ptype->dev == skb->dev) {
3555 ret = deliver_skb(skb, pt_prev, orig_dev);
3561 #ifdef CONFIG_NET_CLS_ACT
3562 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3568 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3571 if (vlan_tx_tag_present(skb)) {
3573 ret = deliver_skb(skb, pt_prev, orig_dev);
3576 if (vlan_do_receive(&skb))
3578 else if (unlikely(!skb))
3582 rx_handler = rcu_dereference(skb->dev->rx_handler);
3585 ret = deliver_skb(skb, pt_prev, orig_dev);
3588 switch (rx_handler(&skb)) {
3589 case RX_HANDLER_CONSUMED:
3590 ret = NET_RX_SUCCESS;
3592 case RX_HANDLER_ANOTHER:
3594 case RX_HANDLER_EXACT:
3595 deliver_exact = true;
3596 case RX_HANDLER_PASS:
3603 if (unlikely(vlan_tx_tag_present(skb))) {
3604 if (vlan_tx_tag_get_id(skb))
3605 skb->pkt_type = PACKET_OTHERHOST;
3606 /* Note: we might in the future use prio bits
3607 * and set skb->priority like in vlan_do_receive()
3608 * For the time being, just ignore Priority Code Point
3613 /* deliver only exact match when indicated */
3614 null_or_dev = deliver_exact ? skb->dev : NULL;
3616 type = skb->protocol;
3617 list_for_each_entry_rcu(ptype,
3618 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3619 if (ptype->type == type &&
3620 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3621 ptype->dev == orig_dev)) {
3623 ret = deliver_skb(skb, pt_prev, orig_dev);
3629 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3632 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3635 atomic_long_inc(&skb->dev->rx_dropped);
3637 /* Jamal, now you will not able to escape explaining
3638 * me how you were going to use this. :-)
3648 static int __netif_receive_skb(struct sk_buff *skb)
3652 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3653 unsigned long pflags = current->flags;
3656 * PFMEMALLOC skbs are special, they should
3657 * - be delivered to SOCK_MEMALLOC sockets only
3658 * - stay away from userspace
3659 * - have bounded memory usage
3661 * Use PF_MEMALLOC as this saves us from propagating the allocation
3662 * context down to all allocation sites.
3664 current->flags |= PF_MEMALLOC;
3665 ret = __netif_receive_skb_core(skb, true);
3666 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3668 ret = __netif_receive_skb_core(skb, false);
3673 static int netif_receive_skb_internal(struct sk_buff *skb)
3675 net_timestamp_check(netdev_tstamp_prequeue, skb);
3677 if (skb_defer_rx_timestamp(skb))
3678 return NET_RX_SUCCESS;
3681 if (static_key_false(&rps_needed)) {
3682 struct rps_dev_flow voidflow, *rflow = &voidflow;
3687 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3690 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3697 return __netif_receive_skb(skb);
3701 * netif_receive_skb - process receive buffer from network
3702 * @skb: buffer to process
3704 * netif_receive_skb() is the main receive data processing function.
3705 * It always succeeds. The buffer may be dropped during processing
3706 * for congestion control or by the protocol layers.
3708 * This function may only be called from softirq context and interrupts
3709 * should be enabled.
3711 * Return values (usually ignored):
3712 * NET_RX_SUCCESS: no congestion
3713 * NET_RX_DROP: packet was dropped
3715 int netif_receive_skb(struct sk_buff *skb)
3717 trace_netif_receive_skb_entry(skb);
3719 return netif_receive_skb_internal(skb);
3721 EXPORT_SYMBOL(netif_receive_skb);
3723 /* Network device is going away, flush any packets still pending
3724 * Called with irqs disabled.
3726 static void flush_backlog(void *arg)
3728 struct net_device *dev = arg;
3729 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3730 struct sk_buff *skb, *tmp;
3733 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3734 if (skb->dev == dev) {
3735 __skb_unlink(skb, &sd->input_pkt_queue);
3737 input_queue_head_incr(sd);
3742 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3743 if (skb->dev == dev) {
3744 __skb_unlink(skb, &sd->process_queue);
3746 input_queue_head_incr(sd);
3751 static int napi_gro_complete(struct sk_buff *skb)
3753 struct packet_offload *ptype;
3754 __be16 type = skb->protocol;
3755 struct list_head *head = &offload_base;
3758 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3760 if (NAPI_GRO_CB(skb)->count == 1) {
3761 skb_shinfo(skb)->gso_size = 0;
3766 list_for_each_entry_rcu(ptype, head, list) {
3767 if (ptype->type != type || !ptype->callbacks.gro_complete)
3770 err = ptype->callbacks.gro_complete(skb, 0);
3776 WARN_ON(&ptype->list == head);
3778 return NET_RX_SUCCESS;
3782 return netif_receive_skb_internal(skb);
3785 /* napi->gro_list contains packets ordered by age.
3786 * youngest packets at the head of it.
3787 * Complete skbs in reverse order to reduce latencies.
3789 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3791 struct sk_buff *skb, *prev = NULL;
3793 /* scan list and build reverse chain */
3794 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3799 for (skb = prev; skb; skb = prev) {
3802 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3806 napi_gro_complete(skb);
3810 napi->gro_list = NULL;
3812 EXPORT_SYMBOL(napi_gro_flush);
3814 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3817 unsigned int maclen = skb->dev->hard_header_len;
3818 u32 hash = skb_get_hash_raw(skb);
3820 for (p = napi->gro_list; p; p = p->next) {
3821 unsigned long diffs;
3823 NAPI_GRO_CB(p)->flush = 0;
3825 if (hash != skb_get_hash_raw(p)) {
3826 NAPI_GRO_CB(p)->same_flow = 0;
3830 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3831 diffs |= p->vlan_tci ^ skb->vlan_tci;
3832 if (maclen == ETH_HLEN)
3833 diffs |= compare_ether_header(skb_mac_header(p),
3834 skb_mac_header(skb));
3836 diffs = memcmp(skb_mac_header(p),
3837 skb_mac_header(skb),
3839 NAPI_GRO_CB(p)->same_flow = !diffs;
3843 static void skb_gro_reset_offset(struct sk_buff *skb)
3845 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3846 const skb_frag_t *frag0 = &pinfo->frags[0];
3848 NAPI_GRO_CB(skb)->data_offset = 0;
3849 NAPI_GRO_CB(skb)->frag0 = NULL;
3850 NAPI_GRO_CB(skb)->frag0_len = 0;
3852 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3854 !PageHighMem(skb_frag_page(frag0))) {
3855 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3856 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3860 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
3862 struct skb_shared_info *pinfo = skb_shinfo(skb);
3864 BUG_ON(skb->end - skb->tail < grow);
3866 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3868 skb->data_len -= grow;
3871 pinfo->frags[0].page_offset += grow;
3872 skb_frag_size_sub(&pinfo->frags[0], grow);
3874 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
3875 skb_frag_unref(skb, 0);
3876 memmove(pinfo->frags, pinfo->frags + 1,
3877 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
3881 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3883 struct sk_buff **pp = NULL;
3884 struct packet_offload *ptype;
3885 __be16 type = skb->protocol;
3886 struct list_head *head = &offload_base;
3888 enum gro_result ret;
3891 if (!(skb->dev->features & NETIF_F_GRO))
3894 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3897 gro_list_prepare(napi, skb);
3898 NAPI_GRO_CB(skb)->csum = skb->csum; /* Needed for CHECKSUM_COMPLETE */
3901 list_for_each_entry_rcu(ptype, head, list) {
3902 if (ptype->type != type || !ptype->callbacks.gro_receive)
3905 skb_set_network_header(skb, skb_gro_offset(skb));
3906 skb_reset_mac_len(skb);
3907 NAPI_GRO_CB(skb)->same_flow = 0;
3908 NAPI_GRO_CB(skb)->flush = 0;
3909 NAPI_GRO_CB(skb)->free = 0;
3910 NAPI_GRO_CB(skb)->udp_mark = 0;
3912 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3917 if (&ptype->list == head)
3920 same_flow = NAPI_GRO_CB(skb)->same_flow;
3921 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3924 struct sk_buff *nskb = *pp;
3928 napi_gro_complete(nskb);
3935 if (NAPI_GRO_CB(skb)->flush)
3938 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
3939 struct sk_buff *nskb = napi->gro_list;
3941 /* locate the end of the list to select the 'oldest' flow */
3942 while (nskb->next) {
3948 napi_gro_complete(nskb);
3952 NAPI_GRO_CB(skb)->count = 1;
3953 NAPI_GRO_CB(skb)->age = jiffies;
3954 NAPI_GRO_CB(skb)->last = skb;
3955 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3956 skb->next = napi->gro_list;
3957 napi->gro_list = skb;
3961 grow = skb_gro_offset(skb) - skb_headlen(skb);
3963 gro_pull_from_frag0(skb, grow);
3972 struct packet_offload *gro_find_receive_by_type(__be16 type)
3974 struct list_head *offload_head = &offload_base;
3975 struct packet_offload *ptype;
3977 list_for_each_entry_rcu(ptype, offload_head, list) {
3978 if (ptype->type != type || !ptype->callbacks.gro_receive)
3984 EXPORT_SYMBOL(gro_find_receive_by_type);
3986 struct packet_offload *gro_find_complete_by_type(__be16 type)
3988 struct list_head *offload_head = &offload_base;
3989 struct packet_offload *ptype;
3991 list_for_each_entry_rcu(ptype, offload_head, list) {
3992 if (ptype->type != type || !ptype->callbacks.gro_complete)
3998 EXPORT_SYMBOL(gro_find_complete_by_type);
4000 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4004 if (netif_receive_skb_internal(skb))
4012 case GRO_MERGED_FREE:
4013 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4014 kmem_cache_free(skbuff_head_cache, skb);
4027 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4029 trace_napi_gro_receive_entry(skb);
4031 skb_gro_reset_offset(skb);
4033 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4035 EXPORT_SYMBOL(napi_gro_receive);
4037 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4039 __skb_pull(skb, skb_headlen(skb));
4040 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4041 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4043 skb->dev = napi->dev;
4045 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4050 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4052 struct sk_buff *skb = napi->skb;
4055 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4060 EXPORT_SYMBOL(napi_get_frags);
4062 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4063 struct sk_buff *skb,
4069 __skb_push(skb, ETH_HLEN);
4070 skb->protocol = eth_type_trans(skb, skb->dev);
4071 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4076 case GRO_MERGED_FREE:
4077 napi_reuse_skb(napi, skb);
4087 /* Upper GRO stack assumes network header starts at gro_offset=0
4088 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4089 * We copy ethernet header into skb->data to have a common layout.
4091 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4093 struct sk_buff *skb = napi->skb;
4094 const struct ethhdr *eth;
4095 unsigned int hlen = sizeof(*eth);
4099 skb_reset_mac_header(skb);
4100 skb_gro_reset_offset(skb);
4102 eth = skb_gro_header_fast(skb, 0);
4103 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4104 eth = skb_gro_header_slow(skb, hlen, 0);
4105 if (unlikely(!eth)) {
4106 napi_reuse_skb(napi, skb);
4110 gro_pull_from_frag0(skb, hlen);
4111 NAPI_GRO_CB(skb)->frag0 += hlen;
4112 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4114 __skb_pull(skb, hlen);
4117 * This works because the only protocols we care about don't require
4119 * We'll fix it up properly in napi_frags_finish()
4121 skb->protocol = eth->h_proto;
4126 gro_result_t napi_gro_frags(struct napi_struct *napi)
4128 struct sk_buff *skb = napi_frags_skb(napi);
4133 trace_napi_gro_frags_entry(skb);
4135 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4137 EXPORT_SYMBOL(napi_gro_frags);
4140 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4141 * Note: called with local irq disabled, but exits with local irq enabled.
4143 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4146 struct softnet_data *remsd = sd->rps_ipi_list;
4149 sd->rps_ipi_list = NULL;
4153 /* Send pending IPI's to kick RPS processing on remote cpus. */
4155 struct softnet_data *next = remsd->rps_ipi_next;
4157 if (cpu_online(remsd->cpu))
4158 smp_call_function_single_async(remsd->cpu,
4167 static int process_backlog(struct napi_struct *napi, int quota)
4170 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4173 /* Check if we have pending ipi, its better to send them now,
4174 * not waiting net_rx_action() end.
4176 if (sd->rps_ipi_list) {
4177 local_irq_disable();
4178 net_rps_action_and_irq_enable(sd);
4181 napi->weight = weight_p;
4182 local_irq_disable();
4183 while (work < quota) {
4184 struct sk_buff *skb;
4187 while ((skb = __skb_dequeue(&sd->process_queue))) {
4189 __netif_receive_skb(skb);
4190 local_irq_disable();
4191 input_queue_head_incr(sd);
4192 if (++work >= quota) {
4199 qlen = skb_queue_len(&sd->input_pkt_queue);
4201 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4202 &sd->process_queue);
4204 if (qlen < quota - work) {
4206 * Inline a custom version of __napi_complete().
4207 * only current cpu owns and manipulates this napi,
4208 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4209 * we can use a plain write instead of clear_bit(),
4210 * and we dont need an smp_mb() memory barrier.
4212 list_del(&napi->poll_list);
4215 quota = work + qlen;
4225 * __napi_schedule - schedule for receive
4226 * @n: entry to schedule
4228 * The entry's receive function will be scheduled to run
4230 void __napi_schedule(struct napi_struct *n)
4232 unsigned long flags;
4234 local_irq_save(flags);
4235 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4236 local_irq_restore(flags);
4238 EXPORT_SYMBOL(__napi_schedule);
4240 void __napi_complete(struct napi_struct *n)
4242 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4243 BUG_ON(n->gro_list);
4245 list_del(&n->poll_list);
4246 smp_mb__before_atomic();
4247 clear_bit(NAPI_STATE_SCHED, &n->state);
4249 EXPORT_SYMBOL(__napi_complete);
4251 void napi_complete(struct napi_struct *n)
4253 unsigned long flags;
4256 * don't let napi dequeue from the cpu poll list
4257 * just in case its running on a different cpu
4259 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4262 napi_gro_flush(n, false);
4263 local_irq_save(flags);
4265 local_irq_restore(flags);
4267 EXPORT_SYMBOL(napi_complete);
4269 /* must be called under rcu_read_lock(), as we dont take a reference */
4270 struct napi_struct *napi_by_id(unsigned int napi_id)
4272 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4273 struct napi_struct *napi;
4275 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4276 if (napi->napi_id == napi_id)
4281 EXPORT_SYMBOL_GPL(napi_by_id);
4283 void napi_hash_add(struct napi_struct *napi)
4285 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4287 spin_lock(&napi_hash_lock);
4289 /* 0 is not a valid id, we also skip an id that is taken
4290 * we expect both events to be extremely rare
4293 while (!napi->napi_id) {
4294 napi->napi_id = ++napi_gen_id;
4295 if (napi_by_id(napi->napi_id))
4299 hlist_add_head_rcu(&napi->napi_hash_node,
4300 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4302 spin_unlock(&napi_hash_lock);
4305 EXPORT_SYMBOL_GPL(napi_hash_add);
4307 /* Warning : caller is responsible to make sure rcu grace period
4308 * is respected before freeing memory containing @napi
4310 void napi_hash_del(struct napi_struct *napi)
4312 spin_lock(&napi_hash_lock);
4314 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4315 hlist_del_rcu(&napi->napi_hash_node);
4317 spin_unlock(&napi_hash_lock);
4319 EXPORT_SYMBOL_GPL(napi_hash_del);
4321 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4322 int (*poll)(struct napi_struct *, int), int weight)
4324 INIT_LIST_HEAD(&napi->poll_list);
4325 napi->gro_count = 0;
4326 napi->gro_list = NULL;
4329 if (weight > NAPI_POLL_WEIGHT)
4330 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4332 napi->weight = weight;
4333 list_add(&napi->dev_list, &dev->napi_list);
4335 #ifdef CONFIG_NETPOLL
4336 spin_lock_init(&napi->poll_lock);
4337 napi->poll_owner = -1;
4339 set_bit(NAPI_STATE_SCHED, &napi->state);
4341 EXPORT_SYMBOL(netif_napi_add);
4343 void netif_napi_del(struct napi_struct *napi)
4345 list_del_init(&napi->dev_list);
4346 napi_free_frags(napi);
4348 kfree_skb_list(napi->gro_list);
4349 napi->gro_list = NULL;
4350 napi->gro_count = 0;
4352 EXPORT_SYMBOL(netif_napi_del);
4354 static void net_rx_action(struct softirq_action *h)
4356 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4357 unsigned long time_limit = jiffies + 2;
4358 int budget = netdev_budget;
4361 local_irq_disable();
4363 while (!list_empty(&sd->poll_list)) {
4364 struct napi_struct *n;
4367 /* If softirq window is exhuasted then punt.
4368 * Allow this to run for 2 jiffies since which will allow
4369 * an average latency of 1.5/HZ.
4371 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4376 /* Even though interrupts have been re-enabled, this
4377 * access is safe because interrupts can only add new
4378 * entries to the tail of this list, and only ->poll()
4379 * calls can remove this head entry from the list.
4381 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4383 have = netpoll_poll_lock(n);
4387 /* This NAPI_STATE_SCHED test is for avoiding a race
4388 * with netpoll's poll_napi(). Only the entity which
4389 * obtains the lock and sees NAPI_STATE_SCHED set will
4390 * actually make the ->poll() call. Therefore we avoid
4391 * accidentally calling ->poll() when NAPI is not scheduled.
4394 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4395 work = n->poll(n, weight);
4399 WARN_ON_ONCE(work > weight);
4403 local_irq_disable();
4405 /* Drivers must not modify the NAPI state if they
4406 * consume the entire weight. In such cases this code
4407 * still "owns" the NAPI instance and therefore can
4408 * move the instance around on the list at-will.
4410 if (unlikely(work == weight)) {
4411 if (unlikely(napi_disable_pending(n))) {
4414 local_irq_disable();
4417 /* flush too old packets
4418 * If HZ < 1000, flush all packets.
4421 napi_gro_flush(n, HZ >= 1000);
4422 local_irq_disable();
4424 list_move_tail(&n->poll_list, &sd->poll_list);
4428 netpoll_poll_unlock(have);
4431 net_rps_action_and_irq_enable(sd);
4433 #ifdef CONFIG_NET_DMA
4435 * There may not be any more sk_buffs coming right now, so push
4436 * any pending DMA copies to hardware
4438 dma_issue_pending_all();
4445 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4449 struct netdev_adjacent {
4450 struct net_device *dev;
4452 /* upper master flag, there can only be one master device per list */
4455 /* counter for the number of times this device was added to us */
4458 /* private field for the users */
4461 struct list_head list;
4462 struct rcu_head rcu;
4465 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4466 struct net_device *adj_dev,
4467 struct list_head *adj_list)
4469 struct netdev_adjacent *adj;
4471 list_for_each_entry(adj, adj_list, list) {
4472 if (adj->dev == adj_dev)
4479 * netdev_has_upper_dev - Check if device is linked to an upper device
4481 * @upper_dev: upper device to check
4483 * Find out if a device is linked to specified upper device and return true
4484 * in case it is. Note that this checks only immediate upper device,
4485 * not through a complete stack of devices. The caller must hold the RTNL lock.
4487 bool netdev_has_upper_dev(struct net_device *dev,
4488 struct net_device *upper_dev)
4492 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4494 EXPORT_SYMBOL(netdev_has_upper_dev);
4497 * netdev_has_any_upper_dev - Check if device is linked to some device
4500 * Find out if a device is linked to an upper device and return true in case
4501 * it is. The caller must hold the RTNL lock.
4503 static bool netdev_has_any_upper_dev(struct net_device *dev)
4507 return !list_empty(&dev->all_adj_list.upper);
4511 * netdev_master_upper_dev_get - Get master upper device
4514 * Find a master upper device and return pointer to it or NULL in case
4515 * it's not there. The caller must hold the RTNL lock.
4517 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4519 struct netdev_adjacent *upper;
4523 if (list_empty(&dev->adj_list.upper))
4526 upper = list_first_entry(&dev->adj_list.upper,
4527 struct netdev_adjacent, list);
4528 if (likely(upper->master))
4532 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4534 void *netdev_adjacent_get_private(struct list_head *adj_list)
4536 struct netdev_adjacent *adj;
4538 adj = list_entry(adj_list, struct netdev_adjacent, list);
4540 return adj->private;
4542 EXPORT_SYMBOL(netdev_adjacent_get_private);
4545 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4547 * @iter: list_head ** of the current position
4549 * Gets the next device from the dev's upper list, starting from iter
4550 * position. The caller must hold RCU read lock.
4552 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4553 struct list_head **iter)
4555 struct netdev_adjacent *upper;
4557 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4559 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4561 if (&upper->list == &dev->adj_list.upper)
4564 *iter = &upper->list;
4568 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4571 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4573 * @iter: list_head ** of the current position
4575 * Gets the next device from the dev's upper list, starting from iter
4576 * position. The caller must hold RCU read lock.
4578 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4579 struct list_head **iter)
4581 struct netdev_adjacent *upper;
4583 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4585 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4587 if (&upper->list == &dev->all_adj_list.upper)
4590 *iter = &upper->list;
4594 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4597 * netdev_lower_get_next_private - Get the next ->private from the
4598 * lower neighbour list
4600 * @iter: list_head ** of the current position
4602 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4603 * list, starting from iter position. The caller must hold either hold the
4604 * RTNL lock or its own locking that guarantees that the neighbour lower
4605 * list will remain unchainged.
4607 void *netdev_lower_get_next_private(struct net_device *dev,
4608 struct list_head **iter)
4610 struct netdev_adjacent *lower;
4612 lower = list_entry(*iter, struct netdev_adjacent, list);
4614 if (&lower->list == &dev->adj_list.lower)
4617 *iter = lower->list.next;
4619 return lower->private;
4621 EXPORT_SYMBOL(netdev_lower_get_next_private);
4624 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4625 * lower neighbour list, RCU
4628 * @iter: list_head ** of the current position
4630 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4631 * list, starting from iter position. The caller must hold RCU read lock.
4633 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4634 struct list_head **iter)
4636 struct netdev_adjacent *lower;
4638 WARN_ON_ONCE(!rcu_read_lock_held());
4640 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4642 if (&lower->list == &dev->adj_list.lower)
4645 *iter = &lower->list;
4647 return lower->private;
4649 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4652 * netdev_lower_get_next - Get the next device from the lower neighbour
4655 * @iter: list_head ** of the current position
4657 * Gets the next netdev_adjacent from the dev's lower neighbour
4658 * list, starting from iter position. The caller must hold RTNL lock or
4659 * its own locking that guarantees that the neighbour lower
4660 * list will remain unchainged.
4662 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4664 struct netdev_adjacent *lower;
4666 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4668 if (&lower->list == &dev->adj_list.lower)
4671 *iter = &lower->list;
4675 EXPORT_SYMBOL(netdev_lower_get_next);
4678 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4679 * lower neighbour list, RCU
4683 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4684 * list. The caller must hold RCU read lock.
4686 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4688 struct netdev_adjacent *lower;
4690 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4691 struct netdev_adjacent, list);
4693 return lower->private;
4696 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4699 * netdev_master_upper_dev_get_rcu - Get master upper device
4702 * Find a master upper device and return pointer to it or NULL in case
4703 * it's not there. The caller must hold the RCU read lock.
4705 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4707 struct netdev_adjacent *upper;
4709 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4710 struct netdev_adjacent, list);
4711 if (upper && likely(upper->master))
4715 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4717 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4718 struct net_device *adj_dev,
4719 struct list_head *dev_list)
4721 char linkname[IFNAMSIZ+7];
4722 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4723 "upper_%s" : "lower_%s", adj_dev->name);
4724 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4727 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4729 struct list_head *dev_list)
4731 char linkname[IFNAMSIZ+7];
4732 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4733 "upper_%s" : "lower_%s", name);
4734 sysfs_remove_link(&(dev->dev.kobj), linkname);
4737 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4738 (dev_list == &dev->adj_list.upper || \
4739 dev_list == &dev->adj_list.lower)
4741 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4742 struct net_device *adj_dev,
4743 struct list_head *dev_list,
4744 void *private, bool master)
4746 struct netdev_adjacent *adj;
4749 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4756 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4761 adj->master = master;
4763 adj->private = private;
4766 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4767 adj_dev->name, dev->name, adj_dev->name);
4769 if (netdev_adjacent_is_neigh_list(dev, dev_list)) {
4770 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
4775 /* Ensure that master link is always the first item in list. */
4777 ret = sysfs_create_link(&(dev->dev.kobj),
4778 &(adj_dev->dev.kobj), "master");
4780 goto remove_symlinks;
4782 list_add_rcu(&adj->list, dev_list);
4784 list_add_tail_rcu(&adj->list, dev_list);
4790 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4791 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4799 static void __netdev_adjacent_dev_remove(struct net_device *dev,
4800 struct net_device *adj_dev,
4801 struct list_head *dev_list)
4803 struct netdev_adjacent *adj;
4805 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4808 pr_err("tried to remove device %s from %s\n",
4809 dev->name, adj_dev->name);
4813 if (adj->ref_nr > 1) {
4814 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4821 sysfs_remove_link(&(dev->dev.kobj), "master");
4823 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4824 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4826 list_del_rcu(&adj->list);
4827 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4828 adj_dev->name, dev->name, adj_dev->name);
4830 kfree_rcu(adj, rcu);
4833 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4834 struct net_device *upper_dev,
4835 struct list_head *up_list,
4836 struct list_head *down_list,
4837 void *private, bool master)
4841 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4846 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4849 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4856 static int __netdev_adjacent_dev_link(struct net_device *dev,
4857 struct net_device *upper_dev)
4859 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4860 &dev->all_adj_list.upper,
4861 &upper_dev->all_adj_list.lower,
4865 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4866 struct net_device *upper_dev,
4867 struct list_head *up_list,
4868 struct list_head *down_list)
4870 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4871 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4874 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
4875 struct net_device *upper_dev)
4877 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4878 &dev->all_adj_list.upper,
4879 &upper_dev->all_adj_list.lower);
4882 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4883 struct net_device *upper_dev,
4884 void *private, bool master)
4886 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4891 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4892 &dev->adj_list.upper,
4893 &upper_dev->adj_list.lower,
4896 __netdev_adjacent_dev_unlink(dev, upper_dev);
4903 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4904 struct net_device *upper_dev)
4906 __netdev_adjacent_dev_unlink(dev, upper_dev);
4907 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4908 &dev->adj_list.upper,
4909 &upper_dev->adj_list.lower);
4912 static int __netdev_upper_dev_link(struct net_device *dev,
4913 struct net_device *upper_dev, bool master,
4916 struct netdev_adjacent *i, *j, *to_i, *to_j;
4921 if (dev == upper_dev)
4924 /* To prevent loops, check if dev is not upper device to upper_dev. */
4925 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
4928 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
4931 if (master && netdev_master_upper_dev_get(dev))
4934 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
4939 /* Now that we linked these devs, make all the upper_dev's
4940 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4941 * versa, and don't forget the devices itself. All of these
4942 * links are non-neighbours.
4944 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4945 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4946 pr_debug("Interlinking %s with %s, non-neighbour\n",
4947 i->dev->name, j->dev->name);
4948 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
4954 /* add dev to every upper_dev's upper device */
4955 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4956 pr_debug("linking %s's upper device %s with %s\n",
4957 upper_dev->name, i->dev->name, dev->name);
4958 ret = __netdev_adjacent_dev_link(dev, i->dev);
4960 goto rollback_upper_mesh;
4963 /* add upper_dev to every dev's lower device */
4964 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4965 pr_debug("linking %s's lower device %s with %s\n", dev->name,
4966 i->dev->name, upper_dev->name);
4967 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
4969 goto rollback_lower_mesh;
4972 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4975 rollback_lower_mesh:
4977 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4980 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4985 rollback_upper_mesh:
4987 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4990 __netdev_adjacent_dev_unlink(dev, i->dev);
4998 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4999 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5000 if (i == to_i && j == to_j)
5002 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5008 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5014 * netdev_upper_dev_link - Add a link to the upper device
5016 * @upper_dev: new upper device
5018 * Adds a link to device which is upper to this one. The caller must hold
5019 * the RTNL lock. On a failure a negative errno code is returned.
5020 * On success the reference counts are adjusted and the function
5023 int netdev_upper_dev_link(struct net_device *dev,
5024 struct net_device *upper_dev)
5026 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5028 EXPORT_SYMBOL(netdev_upper_dev_link);
5031 * netdev_master_upper_dev_link - Add a master link to the upper device
5033 * @upper_dev: new upper device
5035 * Adds a link to device which is upper to this one. In this case, only
5036 * one master upper device can be linked, although other non-master devices
5037 * might be linked as well. The caller must hold the RTNL lock.
5038 * On a failure a negative errno code is returned. On success the reference
5039 * counts are adjusted and the function returns zero.
5041 int netdev_master_upper_dev_link(struct net_device *dev,
5042 struct net_device *upper_dev)
5044 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5046 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5048 int netdev_master_upper_dev_link_private(struct net_device *dev,
5049 struct net_device *upper_dev,
5052 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5054 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5057 * netdev_upper_dev_unlink - Removes a link to upper device
5059 * @upper_dev: new upper device
5061 * Removes a link to device which is upper to this one. The caller must hold
5064 void netdev_upper_dev_unlink(struct net_device *dev,
5065 struct net_device *upper_dev)
5067 struct netdev_adjacent *i, *j;
5070 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5072 /* Here is the tricky part. We must remove all dev's lower
5073 * devices from all upper_dev's upper devices and vice
5074 * versa, to maintain the graph relationship.
5076 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5077 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5078 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5080 /* remove also the devices itself from lower/upper device
5083 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5084 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5086 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5087 __netdev_adjacent_dev_unlink(dev, i->dev);
5089 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5091 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5093 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5095 struct netdev_adjacent *iter;
5097 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5098 netdev_adjacent_sysfs_del(iter->dev, oldname,
5099 &iter->dev->adj_list.lower);
5100 netdev_adjacent_sysfs_add(iter->dev, dev,
5101 &iter->dev->adj_list.lower);
5104 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5105 netdev_adjacent_sysfs_del(iter->dev, oldname,
5106 &iter->dev->adj_list.upper);
5107 netdev_adjacent_sysfs_add(iter->dev, dev,
5108 &iter->dev->adj_list.upper);
5112 void *netdev_lower_dev_get_private(struct net_device *dev,
5113 struct net_device *lower_dev)
5115 struct netdev_adjacent *lower;
5119 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5123 return lower->private;
5125 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5128 int dev_get_nest_level(struct net_device *dev,
5129 bool (*type_check)(struct net_device *dev))
5131 struct net_device *lower = NULL;
5132 struct list_head *iter;
5138 netdev_for_each_lower_dev(dev, lower, iter) {
5139 nest = dev_get_nest_level(lower, type_check);
5140 if (max_nest < nest)
5144 if (type_check(dev))
5149 EXPORT_SYMBOL(dev_get_nest_level);
5151 static void dev_change_rx_flags(struct net_device *dev, int flags)
5153 const struct net_device_ops *ops = dev->netdev_ops;
5155 if (ops->ndo_change_rx_flags)
5156 ops->ndo_change_rx_flags(dev, flags);
5159 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5161 unsigned int old_flags = dev->flags;
5167 dev->flags |= IFF_PROMISC;
5168 dev->promiscuity += inc;
5169 if (dev->promiscuity == 0) {
5172 * If inc causes overflow, untouch promisc and return error.
5175 dev->flags &= ~IFF_PROMISC;
5177 dev->promiscuity -= inc;
5178 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5183 if (dev->flags != old_flags) {
5184 pr_info("device %s %s promiscuous mode\n",
5186 dev->flags & IFF_PROMISC ? "entered" : "left");
5187 if (audit_enabled) {
5188 current_uid_gid(&uid, &gid);
5189 audit_log(current->audit_context, GFP_ATOMIC,
5190 AUDIT_ANOM_PROMISCUOUS,
5191 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5192 dev->name, (dev->flags & IFF_PROMISC),
5193 (old_flags & IFF_PROMISC),
5194 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5195 from_kuid(&init_user_ns, uid),
5196 from_kgid(&init_user_ns, gid),
5197 audit_get_sessionid(current));
5200 dev_change_rx_flags(dev, IFF_PROMISC);
5203 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5208 * dev_set_promiscuity - update promiscuity count on a device
5212 * Add or remove promiscuity from a device. While the count in the device
5213 * remains above zero the interface remains promiscuous. Once it hits zero
5214 * the device reverts back to normal filtering operation. A negative inc
5215 * value is used to drop promiscuity on the device.
5216 * Return 0 if successful or a negative errno code on error.
5218 int dev_set_promiscuity(struct net_device *dev, int inc)
5220 unsigned int old_flags = dev->flags;
5223 err = __dev_set_promiscuity(dev, inc, true);
5226 if (dev->flags != old_flags)
5227 dev_set_rx_mode(dev);
5230 EXPORT_SYMBOL(dev_set_promiscuity);
5232 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5234 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5238 dev->flags |= IFF_ALLMULTI;
5239 dev->allmulti += inc;
5240 if (dev->allmulti == 0) {
5243 * If inc causes overflow, untouch allmulti and return error.
5246 dev->flags &= ~IFF_ALLMULTI;
5248 dev->allmulti -= inc;
5249 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5254 if (dev->flags ^ old_flags) {
5255 dev_change_rx_flags(dev, IFF_ALLMULTI);
5256 dev_set_rx_mode(dev);
5258 __dev_notify_flags(dev, old_flags,
5259 dev->gflags ^ old_gflags);
5265 * dev_set_allmulti - update allmulti count on a device
5269 * Add or remove reception of all multicast frames to a device. While the
5270 * count in the device remains above zero the interface remains listening
5271 * to all interfaces. Once it hits zero the device reverts back to normal
5272 * filtering operation. A negative @inc value is used to drop the counter
5273 * when releasing a resource needing all multicasts.
5274 * Return 0 if successful or a negative errno code on error.
5277 int dev_set_allmulti(struct net_device *dev, int inc)
5279 return __dev_set_allmulti(dev, inc, true);
5281 EXPORT_SYMBOL(dev_set_allmulti);
5284 * Upload unicast and multicast address lists to device and
5285 * configure RX filtering. When the device doesn't support unicast
5286 * filtering it is put in promiscuous mode while unicast addresses
5289 void __dev_set_rx_mode(struct net_device *dev)
5291 const struct net_device_ops *ops = dev->netdev_ops;
5293 /* dev_open will call this function so the list will stay sane. */
5294 if (!(dev->flags&IFF_UP))
5297 if (!netif_device_present(dev))
5300 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5301 /* Unicast addresses changes may only happen under the rtnl,
5302 * therefore calling __dev_set_promiscuity here is safe.
5304 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5305 __dev_set_promiscuity(dev, 1, false);
5306 dev->uc_promisc = true;
5307 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5308 __dev_set_promiscuity(dev, -1, false);
5309 dev->uc_promisc = false;
5313 if (ops->ndo_set_rx_mode)
5314 ops->ndo_set_rx_mode(dev);
5317 void dev_set_rx_mode(struct net_device *dev)
5319 netif_addr_lock_bh(dev);
5320 __dev_set_rx_mode(dev);
5321 netif_addr_unlock_bh(dev);
5325 * dev_get_flags - get flags reported to userspace
5328 * Get the combination of flag bits exported through APIs to userspace.
5330 unsigned int dev_get_flags(const struct net_device *dev)
5334 flags = (dev->flags & ~(IFF_PROMISC |
5339 (dev->gflags & (IFF_PROMISC |
5342 if (netif_running(dev)) {
5343 if (netif_oper_up(dev))
5344 flags |= IFF_RUNNING;
5345 if (netif_carrier_ok(dev))
5346 flags |= IFF_LOWER_UP;
5347 if (netif_dormant(dev))
5348 flags |= IFF_DORMANT;
5353 EXPORT_SYMBOL(dev_get_flags);
5355 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5357 unsigned int old_flags = dev->flags;
5363 * Set the flags on our device.
5366 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5367 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5369 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5373 * Load in the correct multicast list now the flags have changed.
5376 if ((old_flags ^ flags) & IFF_MULTICAST)
5377 dev_change_rx_flags(dev, IFF_MULTICAST);
5379 dev_set_rx_mode(dev);
5382 * Have we downed the interface. We handle IFF_UP ourselves
5383 * according to user attempts to set it, rather than blindly
5388 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
5389 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5392 dev_set_rx_mode(dev);
5395 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5396 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5397 unsigned int old_flags = dev->flags;
5399 dev->gflags ^= IFF_PROMISC;
5401 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5402 if (dev->flags != old_flags)
5403 dev_set_rx_mode(dev);
5406 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5407 is important. Some (broken) drivers set IFF_PROMISC, when
5408 IFF_ALLMULTI is requested not asking us and not reporting.
5410 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5411 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5413 dev->gflags ^= IFF_ALLMULTI;
5414 __dev_set_allmulti(dev, inc, false);
5420 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5421 unsigned int gchanges)
5423 unsigned int changes = dev->flags ^ old_flags;
5426 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5428 if (changes & IFF_UP) {
5429 if (dev->flags & IFF_UP)
5430 call_netdevice_notifiers(NETDEV_UP, dev);
5432 call_netdevice_notifiers(NETDEV_DOWN, dev);
5435 if (dev->flags & IFF_UP &&
5436 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5437 struct netdev_notifier_change_info change_info;
5439 change_info.flags_changed = changes;
5440 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5446 * dev_change_flags - change device settings
5448 * @flags: device state flags
5450 * Change settings on device based state flags. The flags are
5451 * in the userspace exported format.
5453 int dev_change_flags(struct net_device *dev, unsigned int flags)
5456 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5458 ret = __dev_change_flags(dev, flags);
5462 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5463 __dev_notify_flags(dev, old_flags, changes);
5466 EXPORT_SYMBOL(dev_change_flags);
5468 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5470 const struct net_device_ops *ops = dev->netdev_ops;
5472 if (ops->ndo_change_mtu)
5473 return ops->ndo_change_mtu(dev, new_mtu);
5480 * dev_set_mtu - Change maximum transfer unit
5482 * @new_mtu: new transfer unit
5484 * Change the maximum transfer size of the network device.
5486 int dev_set_mtu(struct net_device *dev, int new_mtu)
5490 if (new_mtu == dev->mtu)
5493 /* MTU must be positive. */
5497 if (!netif_device_present(dev))
5500 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5501 err = notifier_to_errno(err);
5505 orig_mtu = dev->mtu;
5506 err = __dev_set_mtu(dev, new_mtu);
5509 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5510 err = notifier_to_errno(err);
5512 /* setting mtu back and notifying everyone again,
5513 * so that they have a chance to revert changes.
5515 __dev_set_mtu(dev, orig_mtu);
5516 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5521 EXPORT_SYMBOL(dev_set_mtu);
5524 * dev_set_group - Change group this device belongs to
5526 * @new_group: group this device should belong to
5528 void dev_set_group(struct net_device *dev, int new_group)
5530 dev->group = new_group;
5532 EXPORT_SYMBOL(dev_set_group);
5535 * dev_set_mac_address - Change Media Access Control Address
5539 * Change the hardware (MAC) address of the device
5541 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5543 const struct net_device_ops *ops = dev->netdev_ops;
5546 if (!ops->ndo_set_mac_address)
5548 if (sa->sa_family != dev->type)
5550 if (!netif_device_present(dev))
5552 err = ops->ndo_set_mac_address(dev, sa);
5555 dev->addr_assign_type = NET_ADDR_SET;
5556 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5557 add_device_randomness(dev->dev_addr, dev->addr_len);
5560 EXPORT_SYMBOL(dev_set_mac_address);
5563 * dev_change_carrier - Change device carrier
5565 * @new_carrier: new value
5567 * Change device carrier
5569 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5571 const struct net_device_ops *ops = dev->netdev_ops;
5573 if (!ops->ndo_change_carrier)
5575 if (!netif_device_present(dev))
5577 return ops->ndo_change_carrier(dev, new_carrier);
5579 EXPORT_SYMBOL(dev_change_carrier);
5582 * dev_get_phys_port_id - Get device physical port ID
5586 * Get device physical port ID
5588 int dev_get_phys_port_id(struct net_device *dev,
5589 struct netdev_phys_port_id *ppid)
5591 const struct net_device_ops *ops = dev->netdev_ops;
5593 if (!ops->ndo_get_phys_port_id)
5595 return ops->ndo_get_phys_port_id(dev, ppid);
5597 EXPORT_SYMBOL(dev_get_phys_port_id);
5600 * dev_new_index - allocate an ifindex
5601 * @net: the applicable net namespace
5603 * Returns a suitable unique value for a new device interface
5604 * number. The caller must hold the rtnl semaphore or the
5605 * dev_base_lock to be sure it remains unique.
5607 static int dev_new_index(struct net *net)
5609 int ifindex = net->ifindex;
5613 if (!__dev_get_by_index(net, ifindex))
5614 return net->ifindex = ifindex;
5618 /* Delayed registration/unregisteration */
5619 static LIST_HEAD(net_todo_list);
5620 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5622 static void net_set_todo(struct net_device *dev)
5624 list_add_tail(&dev->todo_list, &net_todo_list);
5625 dev_net(dev)->dev_unreg_count++;
5628 static void rollback_registered_many(struct list_head *head)
5630 struct net_device *dev, *tmp;
5631 LIST_HEAD(close_head);
5633 BUG_ON(dev_boot_phase);
5636 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5637 /* Some devices call without registering
5638 * for initialization unwind. Remove those
5639 * devices and proceed with the remaining.
5641 if (dev->reg_state == NETREG_UNINITIALIZED) {
5642 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5646 list_del(&dev->unreg_list);
5649 dev->dismantle = true;
5650 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5653 /* If device is running, close it first. */
5654 list_for_each_entry(dev, head, unreg_list)
5655 list_add_tail(&dev->close_list, &close_head);
5656 dev_close_many(&close_head);
5658 list_for_each_entry(dev, head, unreg_list) {
5659 /* And unlink it from device chain. */
5660 unlist_netdevice(dev);
5662 dev->reg_state = NETREG_UNREGISTERING;
5667 list_for_each_entry(dev, head, unreg_list) {
5668 /* Shutdown queueing discipline. */
5672 /* Notify protocols, that we are about to destroy
5673 this device. They should clean all the things.
5675 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5677 if (!dev->rtnl_link_ops ||
5678 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5679 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5682 * Flush the unicast and multicast chains
5687 if (dev->netdev_ops->ndo_uninit)
5688 dev->netdev_ops->ndo_uninit(dev);
5690 /* Notifier chain MUST detach us all upper devices. */
5691 WARN_ON(netdev_has_any_upper_dev(dev));
5693 /* Remove entries from kobject tree */
5694 netdev_unregister_kobject(dev);
5696 /* Remove XPS queueing entries */
5697 netif_reset_xps_queues_gt(dev, 0);
5703 list_for_each_entry(dev, head, unreg_list)
5707 static void rollback_registered(struct net_device *dev)
5711 list_add(&dev->unreg_list, &single);
5712 rollback_registered_many(&single);
5716 static netdev_features_t netdev_fix_features(struct net_device *dev,
5717 netdev_features_t features)
5719 /* Fix illegal checksum combinations */
5720 if ((features & NETIF_F_HW_CSUM) &&
5721 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5722 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5723 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5726 /* TSO requires that SG is present as well. */
5727 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5728 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5729 features &= ~NETIF_F_ALL_TSO;
5732 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5733 !(features & NETIF_F_IP_CSUM)) {
5734 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5735 features &= ~NETIF_F_TSO;
5736 features &= ~NETIF_F_TSO_ECN;
5739 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5740 !(features & NETIF_F_IPV6_CSUM)) {
5741 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5742 features &= ~NETIF_F_TSO6;
5745 /* TSO ECN requires that TSO is present as well. */
5746 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5747 features &= ~NETIF_F_TSO_ECN;
5749 /* Software GSO depends on SG. */
5750 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5751 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5752 features &= ~NETIF_F_GSO;
5755 /* UFO needs SG and checksumming */
5756 if (features & NETIF_F_UFO) {
5757 /* maybe split UFO into V4 and V6? */
5758 if (!((features & NETIF_F_GEN_CSUM) ||
5759 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5760 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5762 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5763 features &= ~NETIF_F_UFO;
5766 if (!(features & NETIF_F_SG)) {
5768 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5769 features &= ~NETIF_F_UFO;
5773 #ifdef CONFIG_NET_RX_BUSY_POLL
5774 if (dev->netdev_ops->ndo_busy_poll)
5775 features |= NETIF_F_BUSY_POLL;
5778 features &= ~NETIF_F_BUSY_POLL;
5783 int __netdev_update_features(struct net_device *dev)
5785 netdev_features_t features;
5790 features = netdev_get_wanted_features(dev);
5792 if (dev->netdev_ops->ndo_fix_features)
5793 features = dev->netdev_ops->ndo_fix_features(dev, features);
5795 /* driver might be less strict about feature dependencies */
5796 features = netdev_fix_features(dev, features);
5798 if (dev->features == features)
5801 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5802 &dev->features, &features);
5804 if (dev->netdev_ops->ndo_set_features)
5805 err = dev->netdev_ops->ndo_set_features(dev, features);
5807 if (unlikely(err < 0)) {
5809 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5810 err, &features, &dev->features);
5815 dev->features = features;
5821 * netdev_update_features - recalculate device features
5822 * @dev: the device to check
5824 * Recalculate dev->features set and send notifications if it
5825 * has changed. Should be called after driver or hardware dependent
5826 * conditions might have changed that influence the features.
5828 void netdev_update_features(struct net_device *dev)
5830 if (__netdev_update_features(dev))
5831 netdev_features_change(dev);
5833 EXPORT_SYMBOL(netdev_update_features);
5836 * netdev_change_features - recalculate device features
5837 * @dev: the device to check
5839 * Recalculate dev->features set and send notifications even
5840 * if they have not changed. Should be called instead of
5841 * netdev_update_features() if also dev->vlan_features might
5842 * have changed to allow the changes to be propagated to stacked
5845 void netdev_change_features(struct net_device *dev)
5847 __netdev_update_features(dev);
5848 netdev_features_change(dev);
5850 EXPORT_SYMBOL(netdev_change_features);
5853 * netif_stacked_transfer_operstate - transfer operstate
5854 * @rootdev: the root or lower level device to transfer state from
5855 * @dev: the device to transfer operstate to
5857 * Transfer operational state from root to device. This is normally
5858 * called when a stacking relationship exists between the root
5859 * device and the device(a leaf device).
5861 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5862 struct net_device *dev)
5864 if (rootdev->operstate == IF_OPER_DORMANT)
5865 netif_dormant_on(dev);
5867 netif_dormant_off(dev);
5869 if (netif_carrier_ok(rootdev)) {
5870 if (!netif_carrier_ok(dev))
5871 netif_carrier_on(dev);
5873 if (netif_carrier_ok(dev))
5874 netif_carrier_off(dev);
5877 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5880 static int netif_alloc_rx_queues(struct net_device *dev)
5882 unsigned int i, count = dev->num_rx_queues;
5883 struct netdev_rx_queue *rx;
5887 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5893 for (i = 0; i < count; i++)
5899 static void netdev_init_one_queue(struct net_device *dev,
5900 struct netdev_queue *queue, void *_unused)
5902 /* Initialize queue lock */
5903 spin_lock_init(&queue->_xmit_lock);
5904 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5905 queue->xmit_lock_owner = -1;
5906 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5909 dql_init(&queue->dql, HZ);
5913 static void netif_free_tx_queues(struct net_device *dev)
5915 if (is_vmalloc_addr(dev->_tx))
5921 static int netif_alloc_netdev_queues(struct net_device *dev)
5923 unsigned int count = dev->num_tx_queues;
5924 struct netdev_queue *tx;
5925 size_t sz = count * sizeof(*tx);
5927 BUG_ON(count < 1 || count > 0xffff);
5929 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
5937 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5938 spin_lock_init(&dev->tx_global_lock);
5944 * register_netdevice - register a network device
5945 * @dev: device to register
5947 * Take a completed network device structure and add it to the kernel
5948 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5949 * chain. 0 is returned on success. A negative errno code is returned
5950 * on a failure to set up the device, or if the name is a duplicate.
5952 * Callers must hold the rtnl semaphore. You may want
5953 * register_netdev() instead of this.
5956 * The locking appears insufficient to guarantee two parallel registers
5957 * will not get the same name.
5960 int register_netdevice(struct net_device *dev)
5963 struct net *net = dev_net(dev);
5965 BUG_ON(dev_boot_phase);
5970 /* When net_device's are persistent, this will be fatal. */
5971 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5974 spin_lock_init(&dev->addr_list_lock);
5975 netdev_set_addr_lockdep_class(dev);
5979 ret = dev_get_valid_name(net, dev, dev->name);
5983 /* Init, if this function is available */
5984 if (dev->netdev_ops->ndo_init) {
5985 ret = dev->netdev_ops->ndo_init(dev);
5993 if (((dev->hw_features | dev->features) &
5994 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5995 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5996 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5997 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6004 dev->ifindex = dev_new_index(net);
6005 else if (__dev_get_by_index(net, dev->ifindex))
6008 if (dev->iflink == -1)
6009 dev->iflink = dev->ifindex;
6011 /* Transfer changeable features to wanted_features and enable
6012 * software offloads (GSO and GRO).
6014 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6015 dev->features |= NETIF_F_SOFT_FEATURES;
6016 dev->wanted_features = dev->features & dev->hw_features;
6018 if (!(dev->flags & IFF_LOOPBACK)) {
6019 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6022 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6024 dev->vlan_features |= NETIF_F_HIGHDMA;
6026 /* Make NETIF_F_SG inheritable to tunnel devices.
6028 dev->hw_enc_features |= NETIF_F_SG;
6030 /* Make NETIF_F_SG inheritable to MPLS.
6032 dev->mpls_features |= NETIF_F_SG;
6034 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6035 ret = notifier_to_errno(ret);
6039 ret = netdev_register_kobject(dev);
6042 dev->reg_state = NETREG_REGISTERED;
6044 __netdev_update_features(dev);
6047 * Default initial state at registry is that the
6048 * device is present.
6051 set_bit(__LINK_STATE_PRESENT, &dev->state);
6053 linkwatch_init_dev(dev);
6055 dev_init_scheduler(dev);
6057 list_netdevice(dev);
6058 add_device_randomness(dev->dev_addr, dev->addr_len);
6060 /* If the device has permanent device address, driver should
6061 * set dev_addr and also addr_assign_type should be set to
6062 * NET_ADDR_PERM (default value).
6064 if (dev->addr_assign_type == NET_ADDR_PERM)
6065 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6067 /* Notify protocols, that a new device appeared. */
6068 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6069 ret = notifier_to_errno(ret);
6071 rollback_registered(dev);
6072 dev->reg_state = NETREG_UNREGISTERED;
6075 * Prevent userspace races by waiting until the network
6076 * device is fully setup before sending notifications.
6078 if (!dev->rtnl_link_ops ||
6079 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6080 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6086 if (dev->netdev_ops->ndo_uninit)
6087 dev->netdev_ops->ndo_uninit(dev);
6090 EXPORT_SYMBOL(register_netdevice);
6093 * init_dummy_netdev - init a dummy network device for NAPI
6094 * @dev: device to init
6096 * This takes a network device structure and initialize the minimum
6097 * amount of fields so it can be used to schedule NAPI polls without
6098 * registering a full blown interface. This is to be used by drivers
6099 * that need to tie several hardware interfaces to a single NAPI
6100 * poll scheduler due to HW limitations.
6102 int init_dummy_netdev(struct net_device *dev)
6104 /* Clear everything. Note we don't initialize spinlocks
6105 * are they aren't supposed to be taken by any of the
6106 * NAPI code and this dummy netdev is supposed to be
6107 * only ever used for NAPI polls
6109 memset(dev, 0, sizeof(struct net_device));
6111 /* make sure we BUG if trying to hit standard
6112 * register/unregister code path
6114 dev->reg_state = NETREG_DUMMY;
6116 /* NAPI wants this */
6117 INIT_LIST_HEAD(&dev->napi_list);
6119 /* a dummy interface is started by default */
6120 set_bit(__LINK_STATE_PRESENT, &dev->state);
6121 set_bit(__LINK_STATE_START, &dev->state);
6123 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6124 * because users of this 'device' dont need to change
6130 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6134 * register_netdev - register a network device
6135 * @dev: device to register
6137 * Take a completed network device structure and add it to the kernel
6138 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6139 * chain. 0 is returned on success. A negative errno code is returned
6140 * on a failure to set up the device, or if the name is a duplicate.
6142 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6143 * and expands the device name if you passed a format string to
6146 int register_netdev(struct net_device *dev)
6151 err = register_netdevice(dev);
6155 EXPORT_SYMBOL(register_netdev);
6157 int netdev_refcnt_read(const struct net_device *dev)
6161 for_each_possible_cpu(i)
6162 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6165 EXPORT_SYMBOL(netdev_refcnt_read);
6168 * netdev_wait_allrefs - wait until all references are gone.
6169 * @dev: target net_device
6171 * This is called when unregistering network devices.
6173 * Any protocol or device that holds a reference should register
6174 * for netdevice notification, and cleanup and put back the
6175 * reference if they receive an UNREGISTER event.
6176 * We can get stuck here if buggy protocols don't correctly
6179 static void netdev_wait_allrefs(struct net_device *dev)
6181 unsigned long rebroadcast_time, warning_time;
6184 linkwatch_forget_dev(dev);
6186 rebroadcast_time = warning_time = jiffies;
6187 refcnt = netdev_refcnt_read(dev);
6189 while (refcnt != 0) {
6190 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6193 /* Rebroadcast unregister notification */
6194 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6200 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6201 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6203 /* We must not have linkwatch events
6204 * pending on unregister. If this
6205 * happens, we simply run the queue
6206 * unscheduled, resulting in a noop
6209 linkwatch_run_queue();
6214 rebroadcast_time = jiffies;
6219 refcnt = netdev_refcnt_read(dev);
6221 if (time_after(jiffies, warning_time + 10 * HZ)) {
6222 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6224 warning_time = jiffies;
6233 * register_netdevice(x1);
6234 * register_netdevice(x2);
6236 * unregister_netdevice(y1);
6237 * unregister_netdevice(y2);
6243 * We are invoked by rtnl_unlock().
6244 * This allows us to deal with problems:
6245 * 1) We can delete sysfs objects which invoke hotplug
6246 * without deadlocking with linkwatch via keventd.
6247 * 2) Since we run with the RTNL semaphore not held, we can sleep
6248 * safely in order to wait for the netdev refcnt to drop to zero.
6250 * We must not return until all unregister events added during
6251 * the interval the lock was held have been completed.
6253 void netdev_run_todo(void)
6255 struct list_head list;
6257 /* Snapshot list, allow later requests */
6258 list_replace_init(&net_todo_list, &list);
6263 /* Wait for rcu callbacks to finish before next phase */
6264 if (!list_empty(&list))
6267 while (!list_empty(&list)) {
6268 struct net_device *dev
6269 = list_first_entry(&list, struct net_device, todo_list);
6270 list_del(&dev->todo_list);
6273 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6276 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6277 pr_err("network todo '%s' but state %d\n",
6278 dev->name, dev->reg_state);
6283 dev->reg_state = NETREG_UNREGISTERED;
6285 on_each_cpu(flush_backlog, dev, 1);
6287 netdev_wait_allrefs(dev);
6290 BUG_ON(netdev_refcnt_read(dev));
6291 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6292 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6293 WARN_ON(dev->dn_ptr);
6295 if (dev->destructor)
6296 dev->destructor(dev);
6298 /* Report a network device has been unregistered */
6300 dev_net(dev)->dev_unreg_count--;
6302 wake_up(&netdev_unregistering_wq);
6304 /* Free network device */
6305 kobject_put(&dev->dev.kobj);
6309 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6310 * fields in the same order, with only the type differing.
6312 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6313 const struct net_device_stats *netdev_stats)
6315 #if BITS_PER_LONG == 64
6316 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6317 memcpy(stats64, netdev_stats, sizeof(*stats64));
6319 size_t i, n = sizeof(*stats64) / sizeof(u64);
6320 const unsigned long *src = (const unsigned long *)netdev_stats;
6321 u64 *dst = (u64 *)stats64;
6323 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6324 sizeof(*stats64) / sizeof(u64));
6325 for (i = 0; i < n; i++)
6329 EXPORT_SYMBOL(netdev_stats_to_stats64);
6332 * dev_get_stats - get network device statistics
6333 * @dev: device to get statistics from
6334 * @storage: place to store stats
6336 * Get network statistics from device. Return @storage.
6337 * The device driver may provide its own method by setting
6338 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6339 * otherwise the internal statistics structure is used.
6341 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6342 struct rtnl_link_stats64 *storage)
6344 const struct net_device_ops *ops = dev->netdev_ops;
6346 if (ops->ndo_get_stats64) {
6347 memset(storage, 0, sizeof(*storage));
6348 ops->ndo_get_stats64(dev, storage);
6349 } else if (ops->ndo_get_stats) {
6350 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6352 netdev_stats_to_stats64(storage, &dev->stats);
6354 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6355 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6358 EXPORT_SYMBOL(dev_get_stats);
6360 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6362 struct netdev_queue *queue = dev_ingress_queue(dev);
6364 #ifdef CONFIG_NET_CLS_ACT
6367 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6370 netdev_init_one_queue(dev, queue, NULL);
6371 queue->qdisc = &noop_qdisc;
6372 queue->qdisc_sleeping = &noop_qdisc;
6373 rcu_assign_pointer(dev->ingress_queue, queue);
6378 static const struct ethtool_ops default_ethtool_ops;
6380 void netdev_set_default_ethtool_ops(struct net_device *dev,
6381 const struct ethtool_ops *ops)
6383 if (dev->ethtool_ops == &default_ethtool_ops)
6384 dev->ethtool_ops = ops;
6386 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6388 void netdev_freemem(struct net_device *dev)
6390 char *addr = (char *)dev - dev->padded;
6392 if (is_vmalloc_addr(addr))
6399 * alloc_netdev_mqs - allocate network device
6400 * @sizeof_priv: size of private data to allocate space for
6401 * @name: device name format string
6402 * @setup: callback to initialize device
6403 * @txqs: the number of TX subqueues to allocate
6404 * @rxqs: the number of RX subqueues to allocate
6406 * Allocates a struct net_device with private data area for driver use
6407 * and performs basic initialization. Also allocates subqueue structs
6408 * for each queue on the device.
6410 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6411 void (*setup)(struct net_device *),
6412 unsigned int txqs, unsigned int rxqs)
6414 struct net_device *dev;
6416 struct net_device *p;
6418 BUG_ON(strlen(name) >= sizeof(dev->name));
6421 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6427 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6432 alloc_size = sizeof(struct net_device);
6434 /* ensure 32-byte alignment of private area */
6435 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6436 alloc_size += sizeof_priv;
6438 /* ensure 32-byte alignment of whole construct */
6439 alloc_size += NETDEV_ALIGN - 1;
6441 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6443 p = vzalloc(alloc_size);
6447 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6448 dev->padded = (char *)dev - (char *)p;
6450 dev->pcpu_refcnt = alloc_percpu(int);
6451 if (!dev->pcpu_refcnt)
6454 if (dev_addr_init(dev))
6460 dev_net_set(dev, &init_net);
6462 dev->gso_max_size = GSO_MAX_SIZE;
6463 dev->gso_max_segs = GSO_MAX_SEGS;
6465 INIT_LIST_HEAD(&dev->napi_list);
6466 INIT_LIST_HEAD(&dev->unreg_list);
6467 INIT_LIST_HEAD(&dev->close_list);
6468 INIT_LIST_HEAD(&dev->link_watch_list);
6469 INIT_LIST_HEAD(&dev->adj_list.upper);
6470 INIT_LIST_HEAD(&dev->adj_list.lower);
6471 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6472 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6473 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6476 dev->num_tx_queues = txqs;
6477 dev->real_num_tx_queues = txqs;
6478 if (netif_alloc_netdev_queues(dev))
6482 dev->num_rx_queues = rxqs;
6483 dev->real_num_rx_queues = rxqs;
6484 if (netif_alloc_rx_queues(dev))
6488 strcpy(dev->name, name);
6489 dev->group = INIT_NETDEV_GROUP;
6490 if (!dev->ethtool_ops)
6491 dev->ethtool_ops = &default_ethtool_ops;
6499 free_percpu(dev->pcpu_refcnt);
6500 netif_free_tx_queues(dev);
6506 netdev_freemem(dev);
6509 EXPORT_SYMBOL(alloc_netdev_mqs);
6512 * free_netdev - free network device
6515 * This function does the last stage of destroying an allocated device
6516 * interface. The reference to the device object is released.
6517 * If this is the last reference then it will be freed.
6519 void free_netdev(struct net_device *dev)
6521 struct napi_struct *p, *n;
6523 release_net(dev_net(dev));
6525 netif_free_tx_queues(dev);
6530 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6532 /* Flush device addresses */
6533 dev_addr_flush(dev);
6535 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6538 free_percpu(dev->pcpu_refcnt);
6539 dev->pcpu_refcnt = NULL;
6541 /* Compatibility with error handling in drivers */
6542 if (dev->reg_state == NETREG_UNINITIALIZED) {
6543 netdev_freemem(dev);
6547 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6548 dev->reg_state = NETREG_RELEASED;
6550 /* will free via device release */
6551 put_device(&dev->dev);
6553 EXPORT_SYMBOL(free_netdev);
6556 * synchronize_net - Synchronize with packet receive processing
6558 * Wait for packets currently being received to be done.
6559 * Does not block later packets from starting.
6561 void synchronize_net(void)
6564 if (rtnl_is_locked())
6565 synchronize_rcu_expedited();
6569 EXPORT_SYMBOL(synchronize_net);
6572 * unregister_netdevice_queue - remove device from the kernel
6576 * This function shuts down a device interface and removes it
6577 * from the kernel tables.
6578 * If head not NULL, device is queued to be unregistered later.
6580 * Callers must hold the rtnl semaphore. You may want
6581 * unregister_netdev() instead of this.
6584 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6589 list_move_tail(&dev->unreg_list, head);
6591 rollback_registered(dev);
6592 /* Finish processing unregister after unlock */
6596 EXPORT_SYMBOL(unregister_netdevice_queue);
6599 * unregister_netdevice_many - unregister many devices
6600 * @head: list of devices
6602 void unregister_netdevice_many(struct list_head *head)
6604 struct net_device *dev;
6606 if (!list_empty(head)) {
6607 rollback_registered_many(head);
6608 list_for_each_entry(dev, head, unreg_list)
6612 EXPORT_SYMBOL(unregister_netdevice_many);
6615 * unregister_netdev - remove device from the kernel
6618 * This function shuts down a device interface and removes it
6619 * from the kernel tables.
6621 * This is just a wrapper for unregister_netdevice that takes
6622 * the rtnl semaphore. In general you want to use this and not
6623 * unregister_netdevice.
6625 void unregister_netdev(struct net_device *dev)
6628 unregister_netdevice(dev);
6631 EXPORT_SYMBOL(unregister_netdev);
6634 * dev_change_net_namespace - move device to different nethost namespace
6636 * @net: network namespace
6637 * @pat: If not NULL name pattern to try if the current device name
6638 * is already taken in the destination network namespace.
6640 * This function shuts down a device interface and moves it
6641 * to a new network namespace. On success 0 is returned, on
6642 * a failure a netagive errno code is returned.
6644 * Callers must hold the rtnl semaphore.
6647 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6653 /* Don't allow namespace local devices to be moved. */
6655 if (dev->features & NETIF_F_NETNS_LOCAL)
6658 /* Ensure the device has been registrered */
6659 if (dev->reg_state != NETREG_REGISTERED)
6662 /* Get out if there is nothing todo */
6664 if (net_eq(dev_net(dev), net))
6667 /* Pick the destination device name, and ensure
6668 * we can use it in the destination network namespace.
6671 if (__dev_get_by_name(net, dev->name)) {
6672 /* We get here if we can't use the current device name */
6675 if (dev_get_valid_name(net, dev, pat) < 0)
6680 * And now a mini version of register_netdevice unregister_netdevice.
6683 /* If device is running close it first. */
6686 /* And unlink it from device chain */
6688 unlist_netdevice(dev);
6692 /* Shutdown queueing discipline. */
6695 /* Notify protocols, that we are about to destroy
6696 this device. They should clean all the things.
6698 Note that dev->reg_state stays at NETREG_REGISTERED.
6699 This is wanted because this way 8021q and macvlan know
6700 the device is just moving and can keep their slaves up.
6702 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6704 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6705 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6708 * Flush the unicast and multicast chains
6713 /* Send a netdev-removed uevent to the old namespace */
6714 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6716 /* Actually switch the network namespace */
6717 dev_net_set(dev, net);
6719 /* If there is an ifindex conflict assign a new one */
6720 if (__dev_get_by_index(net, dev->ifindex)) {
6721 int iflink = (dev->iflink == dev->ifindex);
6722 dev->ifindex = dev_new_index(net);
6724 dev->iflink = dev->ifindex;
6727 /* Send a netdev-add uevent to the new namespace */
6728 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6730 /* Fixup kobjects */
6731 err = device_rename(&dev->dev, dev->name);
6734 /* Add the device back in the hashes */
6735 list_netdevice(dev);
6737 /* Notify protocols, that a new device appeared. */
6738 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6741 * Prevent userspace races by waiting until the network
6742 * device is fully setup before sending notifications.
6744 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6751 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6753 static int dev_cpu_callback(struct notifier_block *nfb,
6754 unsigned long action,
6757 struct sk_buff **list_skb;
6758 struct sk_buff *skb;
6759 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6760 struct softnet_data *sd, *oldsd;
6762 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6765 local_irq_disable();
6766 cpu = smp_processor_id();
6767 sd = &per_cpu(softnet_data, cpu);
6768 oldsd = &per_cpu(softnet_data, oldcpu);
6770 /* Find end of our completion_queue. */
6771 list_skb = &sd->completion_queue;
6773 list_skb = &(*list_skb)->next;
6774 /* Append completion queue from offline CPU. */
6775 *list_skb = oldsd->completion_queue;
6776 oldsd->completion_queue = NULL;
6778 /* Append output queue from offline CPU. */
6779 if (oldsd->output_queue) {
6780 *sd->output_queue_tailp = oldsd->output_queue;
6781 sd->output_queue_tailp = oldsd->output_queue_tailp;
6782 oldsd->output_queue = NULL;
6783 oldsd->output_queue_tailp = &oldsd->output_queue;
6785 /* Append NAPI poll list from offline CPU. */
6786 if (!list_empty(&oldsd->poll_list)) {
6787 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6788 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6791 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6794 /* Process offline CPU's input_pkt_queue */
6795 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6796 netif_rx_internal(skb);
6797 input_queue_head_incr(oldsd);
6799 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6800 netif_rx_internal(skb);
6801 input_queue_head_incr(oldsd);
6809 * netdev_increment_features - increment feature set by one
6810 * @all: current feature set
6811 * @one: new feature set
6812 * @mask: mask feature set
6814 * Computes a new feature set after adding a device with feature set
6815 * @one to the master device with current feature set @all. Will not
6816 * enable anything that is off in @mask. Returns the new feature set.
6818 netdev_features_t netdev_increment_features(netdev_features_t all,
6819 netdev_features_t one, netdev_features_t mask)
6821 if (mask & NETIF_F_GEN_CSUM)
6822 mask |= NETIF_F_ALL_CSUM;
6823 mask |= NETIF_F_VLAN_CHALLENGED;
6825 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6826 all &= one | ~NETIF_F_ALL_FOR_ALL;
6828 /* If one device supports hw checksumming, set for all. */
6829 if (all & NETIF_F_GEN_CSUM)
6830 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6834 EXPORT_SYMBOL(netdev_increment_features);
6836 static struct hlist_head * __net_init netdev_create_hash(void)
6839 struct hlist_head *hash;
6841 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6843 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6844 INIT_HLIST_HEAD(&hash[i]);
6849 /* Initialize per network namespace state */
6850 static int __net_init netdev_init(struct net *net)
6852 if (net != &init_net)
6853 INIT_LIST_HEAD(&net->dev_base_head);
6855 net->dev_name_head = netdev_create_hash();
6856 if (net->dev_name_head == NULL)
6859 net->dev_index_head = netdev_create_hash();
6860 if (net->dev_index_head == NULL)
6866 kfree(net->dev_name_head);
6872 * netdev_drivername - network driver for the device
6873 * @dev: network device
6875 * Determine network driver for device.
6877 const char *netdev_drivername(const struct net_device *dev)
6879 const struct device_driver *driver;
6880 const struct device *parent;
6881 const char *empty = "";
6883 parent = dev->dev.parent;
6887 driver = parent->driver;
6888 if (driver && driver->name)
6889 return driver->name;
6893 static int __netdev_printk(const char *level, const struct net_device *dev,
6894 struct va_format *vaf)
6898 if (dev && dev->dev.parent) {
6899 r = dev_printk_emit(level[1] - '0',
6902 dev_driver_string(dev->dev.parent),
6903 dev_name(dev->dev.parent),
6904 netdev_name(dev), vaf);
6906 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6908 r = printk("%s(NULL net_device): %pV", level, vaf);
6914 int netdev_printk(const char *level, const struct net_device *dev,
6915 const char *format, ...)
6917 struct va_format vaf;
6921 va_start(args, format);
6926 r = __netdev_printk(level, dev, &vaf);
6932 EXPORT_SYMBOL(netdev_printk);
6934 #define define_netdev_printk_level(func, level) \
6935 int func(const struct net_device *dev, const char *fmt, ...) \
6938 struct va_format vaf; \
6941 va_start(args, fmt); \
6946 r = __netdev_printk(level, dev, &vaf); \
6952 EXPORT_SYMBOL(func);
6954 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6955 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6956 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6957 define_netdev_printk_level(netdev_err, KERN_ERR);
6958 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6959 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6960 define_netdev_printk_level(netdev_info, KERN_INFO);
6962 static void __net_exit netdev_exit(struct net *net)
6964 kfree(net->dev_name_head);
6965 kfree(net->dev_index_head);
6968 static struct pernet_operations __net_initdata netdev_net_ops = {
6969 .init = netdev_init,
6970 .exit = netdev_exit,
6973 static void __net_exit default_device_exit(struct net *net)
6975 struct net_device *dev, *aux;
6977 * Push all migratable network devices back to the
6978 * initial network namespace
6981 for_each_netdev_safe(net, dev, aux) {
6983 char fb_name[IFNAMSIZ];
6985 /* Ignore unmoveable devices (i.e. loopback) */
6986 if (dev->features & NETIF_F_NETNS_LOCAL)
6989 /* Leave virtual devices for the generic cleanup */
6990 if (dev->rtnl_link_ops)
6993 /* Push remaining network devices to init_net */
6994 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6995 err = dev_change_net_namespace(dev, &init_net, fb_name);
6997 pr_emerg("%s: failed to move %s to init_net: %d\n",
6998 __func__, dev->name, err);
7005 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7007 /* Return with the rtnl_lock held when there are no network
7008 * devices unregistering in any network namespace in net_list.
7015 prepare_to_wait(&netdev_unregistering_wq, &wait,
7016 TASK_UNINTERRUPTIBLE);
7017 unregistering = false;
7019 list_for_each_entry(net, net_list, exit_list) {
7020 if (net->dev_unreg_count > 0) {
7021 unregistering = true;
7030 finish_wait(&netdev_unregistering_wq, &wait);
7033 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7035 /* At exit all network devices most be removed from a network
7036 * namespace. Do this in the reverse order of registration.
7037 * Do this across as many network namespaces as possible to
7038 * improve batching efficiency.
7040 struct net_device *dev;
7042 LIST_HEAD(dev_kill_list);
7044 /* To prevent network device cleanup code from dereferencing
7045 * loopback devices or network devices that have been freed
7046 * wait here for all pending unregistrations to complete,
7047 * before unregistring the loopback device and allowing the
7048 * network namespace be freed.
7050 * The netdev todo list containing all network devices
7051 * unregistrations that happen in default_device_exit_batch
7052 * will run in the rtnl_unlock() at the end of
7053 * default_device_exit_batch.
7055 rtnl_lock_unregistering(net_list);
7056 list_for_each_entry(net, net_list, exit_list) {
7057 for_each_netdev_reverse(net, dev) {
7058 if (dev->rtnl_link_ops)
7059 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7061 unregister_netdevice_queue(dev, &dev_kill_list);
7064 unregister_netdevice_many(&dev_kill_list);
7065 list_del(&dev_kill_list);
7069 static struct pernet_operations __net_initdata default_device_ops = {
7070 .exit = default_device_exit,
7071 .exit_batch = default_device_exit_batch,
7075 * Initialize the DEV module. At boot time this walks the device list and
7076 * unhooks any devices that fail to initialise (normally hardware not
7077 * present) and leaves us with a valid list of present and active devices.
7082 * This is called single threaded during boot, so no need
7083 * to take the rtnl semaphore.
7085 static int __init net_dev_init(void)
7087 int i, rc = -ENOMEM;
7089 BUG_ON(!dev_boot_phase);
7091 if (dev_proc_init())
7094 if (netdev_kobject_init())
7097 INIT_LIST_HEAD(&ptype_all);
7098 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7099 INIT_LIST_HEAD(&ptype_base[i]);
7101 INIT_LIST_HEAD(&offload_base);
7103 if (register_pernet_subsys(&netdev_net_ops))
7107 * Initialise the packet receive queues.
7110 for_each_possible_cpu(i) {
7111 struct softnet_data *sd = &per_cpu(softnet_data, i);
7113 skb_queue_head_init(&sd->input_pkt_queue);
7114 skb_queue_head_init(&sd->process_queue);
7115 INIT_LIST_HEAD(&sd->poll_list);
7116 sd->output_queue_tailp = &sd->output_queue;
7118 sd->csd.func = rps_trigger_softirq;
7123 sd->backlog.poll = process_backlog;
7124 sd->backlog.weight = weight_p;
7129 /* The loopback device is special if any other network devices
7130 * is present in a network namespace the loopback device must
7131 * be present. Since we now dynamically allocate and free the
7132 * loopback device ensure this invariant is maintained by
7133 * keeping the loopback device as the first device on the
7134 * list of network devices. Ensuring the loopback devices
7135 * is the first device that appears and the last network device
7138 if (register_pernet_device(&loopback_net_ops))
7141 if (register_pernet_device(&default_device_ops))
7144 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7145 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7147 hotcpu_notifier(dev_cpu_callback, 0);
7154 subsys_initcall(net_dev_init);