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>
133 #include "net-sysfs.h"
135 /* Instead of increasing this, you should create a hash table. */
136 #define MAX_GRO_SKBS 8
138 /* This should be increased if a protocol with a bigger head is added. */
139 #define GRO_MAX_HEAD (MAX_HEADER + 128)
141 static DEFINE_SPINLOCK(ptype_lock);
142 static DEFINE_SPINLOCK(offload_lock);
143 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
144 struct list_head ptype_all __read_mostly; /* Taps */
145 static struct list_head offload_base __read_mostly;
148 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
151 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
153 * Writers must hold the rtnl semaphore while they loop through the
154 * dev_base_head list, and hold dev_base_lock for writing when they do the
155 * actual updates. This allows pure readers to access the list even
156 * while a writer is preparing to update it.
158 * To put it another way, dev_base_lock is held for writing only to
159 * protect against pure readers; the rtnl semaphore provides the
160 * protection against other writers.
162 * See, for example usages, register_netdevice() and
163 * unregister_netdevice(), which must be called with the rtnl
166 DEFINE_RWLOCK(dev_base_lock);
167 EXPORT_SYMBOL(dev_base_lock);
169 seqcount_t devnet_rename_seq;
171 static inline void dev_base_seq_inc(struct net *net)
173 while (++net->dev_base_seq == 0);
176 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
178 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
180 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
183 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
185 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
188 static inline void rps_lock(struct softnet_data *sd)
191 spin_lock(&sd->input_pkt_queue.lock);
195 static inline void rps_unlock(struct softnet_data *sd)
198 spin_unlock(&sd->input_pkt_queue.lock);
202 /* Device list insertion */
203 static void list_netdevice(struct net_device *dev)
205 struct net *net = dev_net(dev);
209 write_lock_bh(&dev_base_lock);
210 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
211 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
212 hlist_add_head_rcu(&dev->index_hlist,
213 dev_index_hash(net, dev->ifindex));
214 write_unlock_bh(&dev_base_lock);
216 dev_base_seq_inc(net);
219 /* Device list removal
220 * caller must respect a RCU grace period before freeing/reusing dev
222 static void unlist_netdevice(struct net_device *dev)
226 /* Unlink dev from the device chain */
227 write_lock_bh(&dev_base_lock);
228 list_del_rcu(&dev->dev_list);
229 hlist_del_rcu(&dev->name_hlist);
230 hlist_del_rcu(&dev->index_hlist);
231 write_unlock_bh(&dev_base_lock);
233 dev_base_seq_inc(dev_net(dev));
240 static RAW_NOTIFIER_HEAD(netdev_chain);
243 * Device drivers call our routines to queue packets here. We empty the
244 * queue in the local softnet handler.
247 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
248 EXPORT_PER_CPU_SYMBOL(softnet_data);
250 #ifdef CONFIG_LOCKDEP
252 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
253 * according to dev->type
255 static const unsigned short netdev_lock_type[] =
256 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
257 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
258 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
259 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
260 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
261 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
262 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
263 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
264 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
265 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
266 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
267 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
268 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
269 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
270 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
272 static const char *const netdev_lock_name[] =
273 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
274 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
275 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
276 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
277 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
278 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
279 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
280 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
281 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
282 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
283 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
284 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
285 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
286 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
287 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
289 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
290 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
292 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
296 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
297 if (netdev_lock_type[i] == dev_type)
299 /* the last key is used by default */
300 return ARRAY_SIZE(netdev_lock_type) - 1;
303 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
304 unsigned short dev_type)
308 i = netdev_lock_pos(dev_type);
309 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
310 netdev_lock_name[i]);
313 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
317 i = netdev_lock_pos(dev->type);
318 lockdep_set_class_and_name(&dev->addr_list_lock,
319 &netdev_addr_lock_key[i],
320 netdev_lock_name[i]);
323 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
324 unsigned short dev_type)
327 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
332 /*******************************************************************************
334 Protocol management and registration routines
336 *******************************************************************************/
339 * Add a protocol ID to the list. Now that the input handler is
340 * smarter we can dispense with all the messy stuff that used to be
343 * BEWARE!!! Protocol handlers, mangling input packets,
344 * MUST BE last in hash buckets and checking protocol handlers
345 * MUST start from promiscuous ptype_all chain in net_bh.
346 * It is true now, do not change it.
347 * Explanation follows: if protocol handler, mangling packet, will
348 * be the first on list, it is not able to sense, that packet
349 * is cloned and should be copied-on-write, so that it will
350 * change it and subsequent readers will get broken packet.
354 static inline struct list_head *ptype_head(const struct packet_type *pt)
356 if (pt->type == htons(ETH_P_ALL))
359 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
363 * dev_add_pack - add packet handler
364 * @pt: packet type declaration
366 * Add a protocol handler to the networking stack. The passed &packet_type
367 * is linked into kernel lists and may not be freed until it has been
368 * removed from the kernel lists.
370 * This call does not sleep therefore it can not
371 * guarantee all CPU's that are in middle of receiving packets
372 * will see the new packet type (until the next received packet).
375 void dev_add_pack(struct packet_type *pt)
377 struct list_head *head = ptype_head(pt);
379 spin_lock(&ptype_lock);
380 list_add_rcu(&pt->list, head);
381 spin_unlock(&ptype_lock);
383 EXPORT_SYMBOL(dev_add_pack);
386 * __dev_remove_pack - remove packet handler
387 * @pt: packet type declaration
389 * Remove a protocol handler that was previously added to the kernel
390 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
391 * from the kernel lists and can be freed or reused once this function
394 * The packet type might still be in use by receivers
395 * and must not be freed until after all the CPU's have gone
396 * through a quiescent state.
398 void __dev_remove_pack(struct packet_type *pt)
400 struct list_head *head = ptype_head(pt);
401 struct packet_type *pt1;
403 spin_lock(&ptype_lock);
405 list_for_each_entry(pt1, head, list) {
407 list_del_rcu(&pt->list);
412 pr_warn("dev_remove_pack: %p not found\n", pt);
414 spin_unlock(&ptype_lock);
416 EXPORT_SYMBOL(__dev_remove_pack);
419 * dev_remove_pack - remove packet handler
420 * @pt: packet type declaration
422 * Remove a protocol handler that was previously added to the kernel
423 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
424 * from the kernel lists and can be freed or reused once this function
427 * This call sleeps to guarantee that no CPU is looking at the packet
430 void dev_remove_pack(struct packet_type *pt)
432 __dev_remove_pack(pt);
436 EXPORT_SYMBOL(dev_remove_pack);
440 * dev_add_offload - register offload handlers
441 * @po: protocol offload declaration
443 * Add protocol offload handlers to the networking stack. The passed
444 * &proto_offload is linked into kernel lists and may not be freed until
445 * it has been removed from the kernel lists.
447 * This call does not sleep therefore it can not
448 * guarantee all CPU's that are in middle of receiving packets
449 * will see the new offload handlers (until the next received packet).
451 void dev_add_offload(struct packet_offload *po)
453 struct list_head *head = &offload_base;
455 spin_lock(&offload_lock);
456 list_add_rcu(&po->list, head);
457 spin_unlock(&offload_lock);
459 EXPORT_SYMBOL(dev_add_offload);
462 * __dev_remove_offload - remove offload handler
463 * @po: packet offload declaration
465 * Remove a protocol offload handler that was previously added to the
466 * kernel offload handlers by dev_add_offload(). The passed &offload_type
467 * is removed from the kernel lists and can be freed or reused once this
470 * The packet type might still be in use by receivers
471 * and must not be freed until after all the CPU's have gone
472 * through a quiescent state.
474 void __dev_remove_offload(struct packet_offload *po)
476 struct list_head *head = &offload_base;
477 struct packet_offload *po1;
479 spin_lock(&offload_lock);
481 list_for_each_entry(po1, head, list) {
483 list_del_rcu(&po->list);
488 pr_warn("dev_remove_offload: %p not found\n", po);
490 spin_unlock(&offload_lock);
492 EXPORT_SYMBOL(__dev_remove_offload);
495 * dev_remove_offload - remove packet offload handler
496 * @po: packet offload declaration
498 * Remove a packet offload handler that was previously added to the kernel
499 * offload handlers by dev_add_offload(). The passed &offload_type is
500 * removed from the kernel lists and can be freed or reused once this
503 * This call sleeps to guarantee that no CPU is looking at the packet
506 void dev_remove_offload(struct packet_offload *po)
508 __dev_remove_offload(po);
512 EXPORT_SYMBOL(dev_remove_offload);
514 /******************************************************************************
516 Device Boot-time Settings Routines
518 *******************************************************************************/
520 /* Boot time configuration table */
521 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
524 * netdev_boot_setup_add - add new setup entry
525 * @name: name of the device
526 * @map: configured settings for the device
528 * Adds new setup entry to the dev_boot_setup list. The function
529 * returns 0 on error and 1 on success. This is a generic routine to
532 static int netdev_boot_setup_add(char *name, struct ifmap *map)
534 struct netdev_boot_setup *s;
538 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
539 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
540 memset(s[i].name, 0, sizeof(s[i].name));
541 strlcpy(s[i].name, name, IFNAMSIZ);
542 memcpy(&s[i].map, map, sizeof(s[i].map));
547 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
551 * netdev_boot_setup_check - check boot time settings
552 * @dev: the netdevice
554 * Check boot time settings for the device.
555 * The found settings are set for the device to be used
556 * later in the device probing.
557 * Returns 0 if no settings found, 1 if they are.
559 int netdev_boot_setup_check(struct net_device *dev)
561 struct netdev_boot_setup *s = dev_boot_setup;
564 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
565 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
566 !strcmp(dev->name, s[i].name)) {
567 dev->irq = s[i].map.irq;
568 dev->base_addr = s[i].map.base_addr;
569 dev->mem_start = s[i].map.mem_start;
570 dev->mem_end = s[i].map.mem_end;
576 EXPORT_SYMBOL(netdev_boot_setup_check);
580 * netdev_boot_base - get address from boot time settings
581 * @prefix: prefix for network device
582 * @unit: id for network device
584 * Check boot time settings for the base address of device.
585 * The found settings are set for the device to be used
586 * later in the device probing.
587 * Returns 0 if no settings found.
589 unsigned long netdev_boot_base(const char *prefix, int unit)
591 const struct netdev_boot_setup *s = dev_boot_setup;
595 sprintf(name, "%s%d", prefix, unit);
598 * If device already registered then return base of 1
599 * to indicate not to probe for this interface
601 if (__dev_get_by_name(&init_net, name))
604 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
605 if (!strcmp(name, s[i].name))
606 return s[i].map.base_addr;
611 * Saves at boot time configured settings for any netdevice.
613 int __init netdev_boot_setup(char *str)
618 str = get_options(str, ARRAY_SIZE(ints), ints);
623 memset(&map, 0, sizeof(map));
627 map.base_addr = ints[2];
629 map.mem_start = ints[3];
631 map.mem_end = ints[4];
633 /* Add new entry to the list */
634 return netdev_boot_setup_add(str, &map);
637 __setup("netdev=", netdev_boot_setup);
639 /*******************************************************************************
641 Device Interface Subroutines
643 *******************************************************************************/
646 * __dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. Must be called under RTNL semaphore
651 * or @dev_base_lock. If the name is found a pointer to the device
652 * is returned. If the name is not found then %NULL is returned. The
653 * reference counters are not incremented so the caller must be
654 * careful with locks.
657 struct net_device *__dev_get_by_name(struct net *net, const char *name)
659 struct net_device *dev;
660 struct hlist_head *head = dev_name_hash(net, name);
662 hlist_for_each_entry(dev, head, name_hlist)
663 if (!strncmp(dev->name, name, IFNAMSIZ))
668 EXPORT_SYMBOL(__dev_get_by_name);
671 * dev_get_by_name_rcu - find a device by its name
672 * @net: the applicable net namespace
673 * @name: name to find
675 * Find an interface by name.
676 * If the name is found a pointer to the device is returned.
677 * If the name is not found then %NULL is returned.
678 * The reference counters are not incremented so the caller must be
679 * careful with locks. The caller must hold RCU lock.
682 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
684 struct net_device *dev;
685 struct hlist_head *head = dev_name_hash(net, name);
687 hlist_for_each_entry_rcu(dev, head, name_hlist)
688 if (!strncmp(dev->name, name, IFNAMSIZ))
693 EXPORT_SYMBOL(dev_get_by_name_rcu);
696 * dev_get_by_name - find a device by its name
697 * @net: the applicable net namespace
698 * @name: name to find
700 * Find an interface by name. This can be called from any
701 * context and does its own locking. The returned handle has
702 * the usage count incremented and the caller must use dev_put() to
703 * release it when it is no longer needed. %NULL is returned if no
704 * matching device is found.
707 struct net_device *dev_get_by_name(struct net *net, const char *name)
709 struct net_device *dev;
712 dev = dev_get_by_name_rcu(net, name);
718 EXPORT_SYMBOL(dev_get_by_name);
721 * __dev_get_by_index - find a device by its ifindex
722 * @net: the applicable net namespace
723 * @ifindex: index of device
725 * Search for an interface by index. Returns %NULL if the device
726 * is not found or a pointer to the device. The device has not
727 * had its reference counter increased so the caller must be careful
728 * about locking. The caller must hold either the RTNL semaphore
732 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
734 struct net_device *dev;
735 struct hlist_head *head = dev_index_hash(net, ifindex);
737 hlist_for_each_entry(dev, head, index_hlist)
738 if (dev->ifindex == ifindex)
743 EXPORT_SYMBOL(__dev_get_by_index);
746 * dev_get_by_index_rcu - find a device by its ifindex
747 * @net: the applicable net namespace
748 * @ifindex: index of device
750 * Search for an interface by index. Returns %NULL if the device
751 * is not found or a pointer to the device. The device has not
752 * had its reference counter increased so the caller must be careful
753 * about locking. The caller must hold RCU lock.
756 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
758 struct net_device *dev;
759 struct hlist_head *head = dev_index_hash(net, ifindex);
761 hlist_for_each_entry_rcu(dev, head, index_hlist)
762 if (dev->ifindex == ifindex)
767 EXPORT_SYMBOL(dev_get_by_index_rcu);
771 * dev_get_by_index - find a device by its ifindex
772 * @net: the applicable net namespace
773 * @ifindex: index of device
775 * Search for an interface by index. Returns NULL if the device
776 * is not found or a pointer to the device. The device returned has
777 * had a reference added and the pointer is safe until the user calls
778 * dev_put to indicate they have finished with it.
781 struct net_device *dev_get_by_index(struct net *net, int ifindex)
783 struct net_device *dev;
786 dev = dev_get_by_index_rcu(net, ifindex);
792 EXPORT_SYMBOL(dev_get_by_index);
795 * dev_getbyhwaddr_rcu - find a device by its hardware address
796 * @net: the applicable net namespace
797 * @type: media type of device
798 * @ha: hardware address
800 * Search for an interface by MAC address. Returns NULL if the device
801 * is not found or a pointer to the device.
802 * The caller must hold RCU or RTNL.
803 * The returned device has not had its ref count increased
804 * and the caller must therefore be careful about locking
808 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
811 struct net_device *dev;
813 for_each_netdev_rcu(net, dev)
814 if (dev->type == type &&
815 !memcmp(dev->dev_addr, ha, dev->addr_len))
820 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
822 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
824 struct net_device *dev;
827 for_each_netdev(net, dev)
828 if (dev->type == type)
833 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
835 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
837 struct net_device *dev, *ret = NULL;
840 for_each_netdev_rcu(net, dev)
841 if (dev->type == type) {
849 EXPORT_SYMBOL(dev_getfirstbyhwtype);
852 * dev_get_by_flags_rcu - find any device with given flags
853 * @net: the applicable net namespace
854 * @if_flags: IFF_* values
855 * @mask: bitmask of bits in if_flags to check
857 * Search for any interface with the given flags. Returns NULL if a device
858 * is not found or a pointer to the device. Must be called inside
859 * rcu_read_lock(), and result refcount is unchanged.
862 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
865 struct net_device *dev, *ret;
868 for_each_netdev_rcu(net, dev) {
869 if (((dev->flags ^ if_flags) & mask) == 0) {
876 EXPORT_SYMBOL(dev_get_by_flags_rcu);
879 * dev_valid_name - check if name is okay for network device
882 * Network device names need to be valid file names to
883 * to allow sysfs to work. We also disallow any kind of
886 bool dev_valid_name(const char *name)
890 if (strlen(name) >= IFNAMSIZ)
892 if (!strcmp(name, ".") || !strcmp(name, ".."))
896 if (*name == '/' || isspace(*name))
902 EXPORT_SYMBOL(dev_valid_name);
905 * __dev_alloc_name - allocate a name for a device
906 * @net: network namespace to allocate the device name in
907 * @name: name format string
908 * @buf: scratch buffer and result name string
910 * Passed a format string - eg "lt%d" it will try and find a suitable
911 * id. It scans list of devices to build up a free map, then chooses
912 * the first empty slot. The caller must hold the dev_base or rtnl lock
913 * while allocating the name and adding the device in order to avoid
915 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
916 * Returns the number of the unit assigned or a negative errno code.
919 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
923 const int max_netdevices = 8*PAGE_SIZE;
924 unsigned long *inuse;
925 struct net_device *d;
927 p = strnchr(name, IFNAMSIZ-1, '%');
930 * Verify the string as this thing may have come from
931 * the user. There must be either one "%d" and no other "%"
934 if (p[1] != 'd' || strchr(p + 2, '%'))
937 /* Use one page as a bit array of possible slots */
938 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
942 for_each_netdev(net, d) {
943 if (!sscanf(d->name, name, &i))
945 if (i < 0 || i >= max_netdevices)
948 /* avoid cases where sscanf is not exact inverse of printf */
949 snprintf(buf, IFNAMSIZ, name, i);
950 if (!strncmp(buf, d->name, IFNAMSIZ))
954 i = find_first_zero_bit(inuse, max_netdevices);
955 free_page((unsigned long) inuse);
959 snprintf(buf, IFNAMSIZ, name, i);
960 if (!__dev_get_by_name(net, buf))
963 /* It is possible to run out of possible slots
964 * when the name is long and there isn't enough space left
965 * for the digits, or if all bits are used.
971 * dev_alloc_name - allocate a name for a device
973 * @name: name format string
975 * Passed a format string - eg "lt%d" it will try and find a suitable
976 * id. It scans list of devices to build up a free map, then chooses
977 * the first empty slot. The caller must hold the dev_base or rtnl lock
978 * while allocating the name and adding the device in order to avoid
980 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
981 * Returns the number of the unit assigned or a negative errno code.
984 int dev_alloc_name(struct net_device *dev, const char *name)
990 BUG_ON(!dev_net(dev));
992 ret = __dev_alloc_name(net, name, buf);
994 strlcpy(dev->name, buf, IFNAMSIZ);
997 EXPORT_SYMBOL(dev_alloc_name);
999 static int dev_alloc_name_ns(struct net *net,
1000 struct net_device *dev,
1006 ret = __dev_alloc_name(net, name, buf);
1008 strlcpy(dev->name, buf, IFNAMSIZ);
1012 static int dev_get_valid_name(struct net *net,
1013 struct net_device *dev,
1018 if (!dev_valid_name(name))
1021 if (strchr(name, '%'))
1022 return dev_alloc_name_ns(net, dev, name);
1023 else if (__dev_get_by_name(net, name))
1025 else if (dev->name != name)
1026 strlcpy(dev->name, name, IFNAMSIZ);
1032 * dev_change_name - change name of a device
1034 * @newname: name (or format string) must be at least IFNAMSIZ
1036 * Change name of a device, can pass format strings "eth%d".
1039 int dev_change_name(struct net_device *dev, const char *newname)
1041 char oldname[IFNAMSIZ];
1047 BUG_ON(!dev_net(dev));
1050 if (dev->flags & IFF_UP)
1053 write_seqcount_begin(&devnet_rename_seq);
1055 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1056 write_seqcount_end(&devnet_rename_seq);
1060 memcpy(oldname, dev->name, IFNAMSIZ);
1062 err = dev_get_valid_name(net, dev, newname);
1064 write_seqcount_end(&devnet_rename_seq);
1069 ret = device_rename(&dev->dev, dev->name);
1071 memcpy(dev->name, oldname, IFNAMSIZ);
1072 write_seqcount_end(&devnet_rename_seq);
1076 write_seqcount_end(&devnet_rename_seq);
1078 write_lock_bh(&dev_base_lock);
1079 hlist_del_rcu(&dev->name_hlist);
1080 write_unlock_bh(&dev_base_lock);
1084 write_lock_bh(&dev_base_lock);
1085 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1086 write_unlock_bh(&dev_base_lock);
1088 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1089 ret = notifier_to_errno(ret);
1092 /* err >= 0 after dev_alloc_name() or stores the first errno */
1095 write_seqcount_begin(&devnet_rename_seq);
1096 memcpy(dev->name, oldname, IFNAMSIZ);
1099 pr_err("%s: name change rollback failed: %d\n",
1108 * dev_set_alias - change ifalias of a device
1110 * @alias: name up to IFALIASZ
1111 * @len: limit of bytes to copy from info
1113 * Set ifalias for a device,
1115 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1121 if (len >= IFALIASZ)
1125 kfree(dev->ifalias);
1126 dev->ifalias = NULL;
1130 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1133 dev->ifalias = new_ifalias;
1135 strlcpy(dev->ifalias, alias, len+1);
1141 * netdev_features_change - device changes features
1142 * @dev: device to cause notification
1144 * Called to indicate a device has changed features.
1146 void netdev_features_change(struct net_device *dev)
1148 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1150 EXPORT_SYMBOL(netdev_features_change);
1153 * netdev_state_change - device changes state
1154 * @dev: device to cause notification
1156 * Called to indicate a device has changed state. This function calls
1157 * the notifier chains for netdev_chain and sends a NEWLINK message
1158 * to the routing socket.
1160 void netdev_state_change(struct net_device *dev)
1162 if (dev->flags & IFF_UP) {
1163 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1164 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1167 EXPORT_SYMBOL(netdev_state_change);
1170 * netdev_notify_peers - notify network peers about existence of @dev
1171 * @dev: network device
1173 * Generate traffic such that interested network peers are aware of
1174 * @dev, such as by generating a gratuitous ARP. This may be used when
1175 * a device wants to inform the rest of the network about some sort of
1176 * reconfiguration such as a failover event or virtual machine
1179 void netdev_notify_peers(struct net_device *dev)
1182 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1185 EXPORT_SYMBOL(netdev_notify_peers);
1187 static int __dev_open(struct net_device *dev)
1189 const struct net_device_ops *ops = dev->netdev_ops;
1194 if (!netif_device_present(dev))
1197 /* Block netpoll from trying to do any rx path servicing.
1198 * If we don't do this there is a chance ndo_poll_controller
1199 * or ndo_poll may be running while we open the device
1201 netpoll_rx_disable(dev);
1203 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1204 ret = notifier_to_errno(ret);
1208 set_bit(__LINK_STATE_START, &dev->state);
1210 if (ops->ndo_validate_addr)
1211 ret = ops->ndo_validate_addr(dev);
1213 if (!ret && ops->ndo_open)
1214 ret = ops->ndo_open(dev);
1216 netpoll_rx_enable(dev);
1219 clear_bit(__LINK_STATE_START, &dev->state);
1221 dev->flags |= IFF_UP;
1222 net_dmaengine_get();
1223 dev_set_rx_mode(dev);
1225 add_device_randomness(dev->dev_addr, dev->addr_len);
1232 * dev_open - prepare an interface for use.
1233 * @dev: device to open
1235 * Takes a device from down to up state. The device's private open
1236 * function is invoked and then the multicast lists are loaded. Finally
1237 * the device is moved into the up state and a %NETDEV_UP message is
1238 * sent to the netdev notifier chain.
1240 * Calling this function on an active interface is a nop. On a failure
1241 * a negative errno code is returned.
1243 int dev_open(struct net_device *dev)
1247 if (dev->flags & IFF_UP)
1250 ret = __dev_open(dev);
1254 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1255 call_netdevice_notifiers(NETDEV_UP, dev);
1259 EXPORT_SYMBOL(dev_open);
1261 static int __dev_close_many(struct list_head *head)
1263 struct net_device *dev;
1268 list_for_each_entry(dev, head, unreg_list) {
1269 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1271 clear_bit(__LINK_STATE_START, &dev->state);
1273 /* Synchronize to scheduled poll. We cannot touch poll list, it
1274 * can be even on different cpu. So just clear netif_running().
1276 * dev->stop() will invoke napi_disable() on all of it's
1277 * napi_struct instances on this device.
1279 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1282 dev_deactivate_many(head);
1284 list_for_each_entry(dev, head, unreg_list) {
1285 const struct net_device_ops *ops = dev->netdev_ops;
1288 * Call the device specific close. This cannot fail.
1289 * Only if device is UP
1291 * We allow it to be called even after a DETACH hot-plug
1297 dev->flags &= ~IFF_UP;
1298 net_dmaengine_put();
1304 static int __dev_close(struct net_device *dev)
1309 /* Temporarily disable netpoll until the interface is down */
1310 netpoll_rx_disable(dev);
1312 list_add(&dev->unreg_list, &single);
1313 retval = __dev_close_many(&single);
1316 netpoll_rx_enable(dev);
1320 static int dev_close_many(struct list_head *head)
1322 struct net_device *dev, *tmp;
1323 LIST_HEAD(tmp_list);
1325 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1326 if (!(dev->flags & IFF_UP))
1327 list_move(&dev->unreg_list, &tmp_list);
1329 __dev_close_many(head);
1331 list_for_each_entry(dev, head, unreg_list) {
1332 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1333 call_netdevice_notifiers(NETDEV_DOWN, dev);
1336 /* rollback_registered_many needs the complete original list */
1337 list_splice(&tmp_list, head);
1342 * dev_close - shutdown an interface.
1343 * @dev: device to shutdown
1345 * This function moves an active device into down state. A
1346 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1347 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1350 int dev_close(struct net_device *dev)
1352 if (dev->flags & IFF_UP) {
1355 /* Block netpoll rx while the interface is going down */
1356 netpoll_rx_disable(dev);
1358 list_add(&dev->unreg_list, &single);
1359 dev_close_many(&single);
1362 netpoll_rx_enable(dev);
1366 EXPORT_SYMBOL(dev_close);
1370 * dev_disable_lro - disable Large Receive Offload on a device
1373 * Disable Large Receive Offload (LRO) on a net device. Must be
1374 * called under RTNL. This is needed if received packets may be
1375 * forwarded to another interface.
1377 void dev_disable_lro(struct net_device *dev)
1380 * If we're trying to disable lro on a vlan device
1381 * use the underlying physical device instead
1383 if (is_vlan_dev(dev))
1384 dev = vlan_dev_real_dev(dev);
1386 dev->wanted_features &= ~NETIF_F_LRO;
1387 netdev_update_features(dev);
1389 if (unlikely(dev->features & NETIF_F_LRO))
1390 netdev_WARN(dev, "failed to disable LRO!\n");
1392 EXPORT_SYMBOL(dev_disable_lro);
1394 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1395 struct net_device *dev)
1397 struct netdev_notifier_info info;
1399 netdev_notifier_info_init(&info, dev);
1400 return nb->notifier_call(nb, val, &info);
1403 static int dev_boot_phase = 1;
1406 * register_netdevice_notifier - register a network notifier block
1409 * Register a notifier to be called when network device events occur.
1410 * The notifier passed is linked into the kernel structures and must
1411 * not be reused until it has been unregistered. A negative errno code
1412 * is returned on a failure.
1414 * When registered all registration and up events are replayed
1415 * to the new notifier to allow device to have a race free
1416 * view of the network device list.
1419 int register_netdevice_notifier(struct notifier_block *nb)
1421 struct net_device *dev;
1422 struct net_device *last;
1427 err = raw_notifier_chain_register(&netdev_chain, nb);
1433 for_each_netdev(net, dev) {
1434 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1435 err = notifier_to_errno(err);
1439 if (!(dev->flags & IFF_UP))
1442 call_netdevice_notifier(nb, NETDEV_UP, dev);
1453 for_each_netdev(net, dev) {
1457 if (dev->flags & IFF_UP) {
1458 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1460 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1462 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1467 raw_notifier_chain_unregister(&netdev_chain, nb);
1470 EXPORT_SYMBOL(register_netdevice_notifier);
1473 * unregister_netdevice_notifier - unregister a network notifier block
1476 * Unregister a notifier previously registered by
1477 * register_netdevice_notifier(). The notifier is unlinked into the
1478 * kernel structures and may then be reused. A negative errno code
1479 * is returned on a failure.
1481 * After unregistering unregister and down device events are synthesized
1482 * for all devices on the device list to the removed notifier to remove
1483 * the need for special case cleanup code.
1486 int unregister_netdevice_notifier(struct notifier_block *nb)
1488 struct net_device *dev;
1493 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1498 for_each_netdev(net, dev) {
1499 if (dev->flags & IFF_UP) {
1500 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1502 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1504 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1511 EXPORT_SYMBOL(unregister_netdevice_notifier);
1514 * call_netdevice_notifiers_info - call all network notifier blocks
1515 * @val: value passed unmodified to notifier function
1516 * @dev: net_device pointer passed unmodified to notifier function
1517 * @info: notifier information data
1519 * Call all network notifier blocks. Parameters and return value
1520 * are as for raw_notifier_call_chain().
1523 int call_netdevice_notifiers_info(unsigned long val, struct net_device *dev,
1524 struct netdev_notifier_info *info)
1527 netdev_notifier_info_init(info, dev);
1528 return raw_notifier_call_chain(&netdev_chain, val, info);
1530 EXPORT_SYMBOL(call_netdevice_notifiers_info);
1533 * call_netdevice_notifiers - call all network notifier blocks
1534 * @val: value passed unmodified to notifier function
1535 * @dev: net_device pointer passed unmodified to notifier function
1537 * Call all network notifier blocks. Parameters and return value
1538 * are as for raw_notifier_call_chain().
1541 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1543 struct netdev_notifier_info info;
1545 return call_netdevice_notifiers_info(val, dev, &info);
1547 EXPORT_SYMBOL(call_netdevice_notifiers);
1549 static struct static_key netstamp_needed __read_mostly;
1550 #ifdef HAVE_JUMP_LABEL
1551 /* We are not allowed to call static_key_slow_dec() from irq context
1552 * If net_disable_timestamp() is called from irq context, defer the
1553 * static_key_slow_dec() calls.
1555 static atomic_t netstamp_needed_deferred;
1558 void net_enable_timestamp(void)
1560 #ifdef HAVE_JUMP_LABEL
1561 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1565 static_key_slow_dec(&netstamp_needed);
1569 static_key_slow_inc(&netstamp_needed);
1571 EXPORT_SYMBOL(net_enable_timestamp);
1573 void net_disable_timestamp(void)
1575 #ifdef HAVE_JUMP_LABEL
1576 if (in_interrupt()) {
1577 atomic_inc(&netstamp_needed_deferred);
1581 static_key_slow_dec(&netstamp_needed);
1583 EXPORT_SYMBOL(net_disable_timestamp);
1585 static inline void net_timestamp_set(struct sk_buff *skb)
1587 skb->tstamp.tv64 = 0;
1588 if (static_key_false(&netstamp_needed))
1589 __net_timestamp(skb);
1592 #define net_timestamp_check(COND, SKB) \
1593 if (static_key_false(&netstamp_needed)) { \
1594 if ((COND) && !(SKB)->tstamp.tv64) \
1595 __net_timestamp(SKB); \
1598 static inline bool is_skb_forwardable(struct net_device *dev,
1599 struct sk_buff *skb)
1603 if (!(dev->flags & IFF_UP))
1606 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1607 if (skb->len <= len)
1610 /* if TSO is enabled, we don't care about the length as the packet
1611 * could be forwarded without being segmented before
1613 if (skb_is_gso(skb))
1620 * dev_forward_skb - loopback an skb to another netif
1622 * @dev: destination network device
1623 * @skb: buffer to forward
1626 * NET_RX_SUCCESS (no congestion)
1627 * NET_RX_DROP (packet was dropped, but freed)
1629 * dev_forward_skb can be used for injecting an skb from the
1630 * start_xmit function of one device into the receive queue
1631 * of another device.
1633 * The receiving device may be in another namespace, so
1634 * we have to clear all information in the skb that could
1635 * impact namespace isolation.
1637 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1639 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1640 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1641 atomic_long_inc(&dev->rx_dropped);
1649 if (unlikely(!is_skb_forwardable(dev, skb))) {
1650 atomic_long_inc(&dev->rx_dropped);
1656 skb->tstamp.tv64 = 0;
1657 skb->pkt_type = PACKET_HOST;
1658 skb->protocol = eth_type_trans(skb, dev);
1662 nf_reset_trace(skb);
1663 return netif_rx(skb);
1665 EXPORT_SYMBOL_GPL(dev_forward_skb);
1667 static inline int deliver_skb(struct sk_buff *skb,
1668 struct packet_type *pt_prev,
1669 struct net_device *orig_dev)
1671 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1673 atomic_inc(&skb->users);
1674 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1677 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1679 if (!ptype->af_packet_priv || !skb->sk)
1682 if (ptype->id_match)
1683 return ptype->id_match(ptype, skb->sk);
1684 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1691 * Support routine. Sends outgoing frames to any network
1692 * taps currently in use.
1695 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1697 struct packet_type *ptype;
1698 struct sk_buff *skb2 = NULL;
1699 struct packet_type *pt_prev = NULL;
1702 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1703 /* Never send packets back to the socket
1704 * they originated from - MvS (miquels@drinkel.ow.org)
1706 if ((ptype->dev == dev || !ptype->dev) &&
1707 (!skb_loop_sk(ptype, skb))) {
1709 deliver_skb(skb2, pt_prev, skb->dev);
1714 skb2 = skb_clone(skb, GFP_ATOMIC);
1718 net_timestamp_set(skb2);
1720 /* skb->nh should be correctly
1721 set by sender, so that the second statement is
1722 just protection against buggy protocols.
1724 skb_reset_mac_header(skb2);
1726 if (skb_network_header(skb2) < skb2->data ||
1727 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1728 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1729 ntohs(skb2->protocol),
1731 skb_reset_network_header(skb2);
1734 skb2->transport_header = skb2->network_header;
1735 skb2->pkt_type = PACKET_OUTGOING;
1740 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1745 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1746 * @dev: Network device
1747 * @txq: number of queues available
1749 * If real_num_tx_queues is changed the tc mappings may no longer be
1750 * valid. To resolve this verify the tc mapping remains valid and if
1751 * not NULL the mapping. With no priorities mapping to this
1752 * offset/count pair it will no longer be used. In the worst case TC0
1753 * is invalid nothing can be done so disable priority mappings. If is
1754 * expected that drivers will fix this mapping if they can before
1755 * calling netif_set_real_num_tx_queues.
1757 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1760 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1762 /* If TC0 is invalidated disable TC mapping */
1763 if (tc->offset + tc->count > txq) {
1764 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1769 /* Invalidated prio to tc mappings set to TC0 */
1770 for (i = 1; i < TC_BITMASK + 1; i++) {
1771 int q = netdev_get_prio_tc_map(dev, i);
1773 tc = &dev->tc_to_txq[q];
1774 if (tc->offset + tc->count > txq) {
1775 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1777 netdev_set_prio_tc_map(dev, i, 0);
1783 static DEFINE_MUTEX(xps_map_mutex);
1784 #define xmap_dereference(P) \
1785 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1787 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1790 struct xps_map *map = NULL;
1794 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1796 for (pos = 0; map && pos < map->len; pos++) {
1797 if (map->queues[pos] == index) {
1799 map->queues[pos] = map->queues[--map->len];
1801 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1802 kfree_rcu(map, rcu);
1812 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1814 struct xps_dev_maps *dev_maps;
1816 bool active = false;
1818 mutex_lock(&xps_map_mutex);
1819 dev_maps = xmap_dereference(dev->xps_maps);
1824 for_each_possible_cpu(cpu) {
1825 for (i = index; i < dev->num_tx_queues; i++) {
1826 if (!remove_xps_queue(dev_maps, cpu, i))
1829 if (i == dev->num_tx_queues)
1834 RCU_INIT_POINTER(dev->xps_maps, NULL);
1835 kfree_rcu(dev_maps, rcu);
1838 for (i = index; i < dev->num_tx_queues; i++)
1839 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1843 mutex_unlock(&xps_map_mutex);
1846 static struct xps_map *expand_xps_map(struct xps_map *map,
1849 struct xps_map *new_map;
1850 int alloc_len = XPS_MIN_MAP_ALLOC;
1853 for (pos = 0; map && pos < map->len; pos++) {
1854 if (map->queues[pos] != index)
1859 /* Need to add queue to this CPU's existing map */
1861 if (pos < map->alloc_len)
1864 alloc_len = map->alloc_len * 2;
1867 /* Need to allocate new map to store queue on this CPU's map */
1868 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1873 for (i = 0; i < pos; i++)
1874 new_map->queues[i] = map->queues[i];
1875 new_map->alloc_len = alloc_len;
1881 int netif_set_xps_queue(struct net_device *dev, struct cpumask *mask, u16 index)
1883 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1884 struct xps_map *map, *new_map;
1885 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1886 int cpu, numa_node_id = -2;
1887 bool active = false;
1889 mutex_lock(&xps_map_mutex);
1891 dev_maps = xmap_dereference(dev->xps_maps);
1893 /* allocate memory for queue storage */
1894 for_each_online_cpu(cpu) {
1895 if (!cpumask_test_cpu(cpu, mask))
1899 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1900 if (!new_dev_maps) {
1901 mutex_unlock(&xps_map_mutex);
1905 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1908 map = expand_xps_map(map, cpu, index);
1912 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1916 goto out_no_new_maps;
1918 for_each_possible_cpu(cpu) {
1919 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1920 /* add queue to CPU maps */
1923 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1924 while ((pos < map->len) && (map->queues[pos] != index))
1927 if (pos == map->len)
1928 map->queues[map->len++] = index;
1930 if (numa_node_id == -2)
1931 numa_node_id = cpu_to_node(cpu);
1932 else if (numa_node_id != cpu_to_node(cpu))
1935 } else if (dev_maps) {
1936 /* fill in the new device map from the old device map */
1937 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1938 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1943 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1945 /* Cleanup old maps */
1947 for_each_possible_cpu(cpu) {
1948 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1949 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1950 if (map && map != new_map)
1951 kfree_rcu(map, rcu);
1954 kfree_rcu(dev_maps, rcu);
1957 dev_maps = new_dev_maps;
1961 /* update Tx queue numa node */
1962 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
1963 (numa_node_id >= 0) ? numa_node_id :
1969 /* removes queue from unused CPUs */
1970 for_each_possible_cpu(cpu) {
1971 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
1974 if (remove_xps_queue(dev_maps, cpu, index))
1978 /* free map if not active */
1980 RCU_INIT_POINTER(dev->xps_maps, NULL);
1981 kfree_rcu(dev_maps, rcu);
1985 mutex_unlock(&xps_map_mutex);
1989 /* remove any maps that we added */
1990 for_each_possible_cpu(cpu) {
1991 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1992 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1994 if (new_map && new_map != map)
1998 mutex_unlock(&xps_map_mutex);
2000 kfree(new_dev_maps);
2003 EXPORT_SYMBOL(netif_set_xps_queue);
2007 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2008 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2010 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2014 if (txq < 1 || txq > dev->num_tx_queues)
2017 if (dev->reg_state == NETREG_REGISTERED ||
2018 dev->reg_state == NETREG_UNREGISTERING) {
2021 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2027 netif_setup_tc(dev, txq);
2029 if (txq < dev->real_num_tx_queues) {
2030 qdisc_reset_all_tx_gt(dev, txq);
2032 netif_reset_xps_queues_gt(dev, txq);
2037 dev->real_num_tx_queues = txq;
2040 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2044 * netif_set_real_num_rx_queues - set actual number of RX queues used
2045 * @dev: Network device
2046 * @rxq: Actual number of RX queues
2048 * This must be called either with the rtnl_lock held or before
2049 * registration of the net device. Returns 0 on success, or a
2050 * negative error code. If called before registration, it always
2053 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2057 if (rxq < 1 || rxq > dev->num_rx_queues)
2060 if (dev->reg_state == NETREG_REGISTERED) {
2063 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2069 dev->real_num_rx_queues = rxq;
2072 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2076 * netif_get_num_default_rss_queues - default number of RSS queues
2078 * This routine should set an upper limit on the number of RSS queues
2079 * used by default by multiqueue devices.
2081 int netif_get_num_default_rss_queues(void)
2083 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2085 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2087 static inline void __netif_reschedule(struct Qdisc *q)
2089 struct softnet_data *sd;
2090 unsigned long flags;
2092 local_irq_save(flags);
2093 sd = &__get_cpu_var(softnet_data);
2094 q->next_sched = NULL;
2095 *sd->output_queue_tailp = q;
2096 sd->output_queue_tailp = &q->next_sched;
2097 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2098 local_irq_restore(flags);
2101 void __netif_schedule(struct Qdisc *q)
2103 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2104 __netif_reschedule(q);
2106 EXPORT_SYMBOL(__netif_schedule);
2108 void dev_kfree_skb_irq(struct sk_buff *skb)
2110 if (atomic_dec_and_test(&skb->users)) {
2111 struct softnet_data *sd;
2112 unsigned long flags;
2114 local_irq_save(flags);
2115 sd = &__get_cpu_var(softnet_data);
2116 skb->next = sd->completion_queue;
2117 sd->completion_queue = skb;
2118 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2119 local_irq_restore(flags);
2122 EXPORT_SYMBOL(dev_kfree_skb_irq);
2124 void dev_kfree_skb_any(struct sk_buff *skb)
2126 if (in_irq() || irqs_disabled())
2127 dev_kfree_skb_irq(skb);
2131 EXPORT_SYMBOL(dev_kfree_skb_any);
2135 * netif_device_detach - mark device as removed
2136 * @dev: network device
2138 * Mark device as removed from system and therefore no longer available.
2140 void netif_device_detach(struct net_device *dev)
2142 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2143 netif_running(dev)) {
2144 netif_tx_stop_all_queues(dev);
2147 EXPORT_SYMBOL(netif_device_detach);
2150 * netif_device_attach - mark device as attached
2151 * @dev: network device
2153 * Mark device as attached from system and restart if needed.
2155 void netif_device_attach(struct net_device *dev)
2157 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2158 netif_running(dev)) {
2159 netif_tx_wake_all_queues(dev);
2160 __netdev_watchdog_up(dev);
2163 EXPORT_SYMBOL(netif_device_attach);
2165 static void skb_warn_bad_offload(const struct sk_buff *skb)
2167 static const netdev_features_t null_features = 0;
2168 struct net_device *dev = skb->dev;
2169 const char *driver = "";
2171 if (!net_ratelimit())
2174 if (dev && dev->dev.parent)
2175 driver = dev_driver_string(dev->dev.parent);
2177 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2178 "gso_type=%d ip_summed=%d\n",
2179 driver, dev ? &dev->features : &null_features,
2180 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2181 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2182 skb_shinfo(skb)->gso_type, skb->ip_summed);
2186 * Invalidate hardware checksum when packet is to be mangled, and
2187 * complete checksum manually on outgoing path.
2189 int skb_checksum_help(struct sk_buff *skb)
2192 int ret = 0, offset;
2194 if (skb->ip_summed == CHECKSUM_COMPLETE)
2195 goto out_set_summed;
2197 if (unlikely(skb_shinfo(skb)->gso_size)) {
2198 skb_warn_bad_offload(skb);
2202 /* Before computing a checksum, we should make sure no frag could
2203 * be modified by an external entity : checksum could be wrong.
2205 if (skb_has_shared_frag(skb)) {
2206 ret = __skb_linearize(skb);
2211 offset = skb_checksum_start_offset(skb);
2212 BUG_ON(offset >= skb_headlen(skb));
2213 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2215 offset += skb->csum_offset;
2216 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2218 if (skb_cloned(skb) &&
2219 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2220 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2225 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2227 skb->ip_summed = CHECKSUM_NONE;
2231 EXPORT_SYMBOL(skb_checksum_help);
2233 __be16 skb_network_protocol(struct sk_buff *skb)
2235 __be16 type = skb->protocol;
2236 int vlan_depth = ETH_HLEN;
2238 /* Tunnel gso handlers can set protocol to ethernet. */
2239 if (type == htons(ETH_P_TEB)) {
2242 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2245 eth = (struct ethhdr *)skb_mac_header(skb);
2246 type = eth->h_proto;
2249 while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2250 struct vlan_hdr *vh;
2252 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2255 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2256 type = vh->h_vlan_encapsulated_proto;
2257 vlan_depth += VLAN_HLEN;
2264 * skb_mac_gso_segment - mac layer segmentation handler.
2265 * @skb: buffer to segment
2266 * @features: features for the output path (see dev->features)
2268 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2269 netdev_features_t features)
2271 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2272 struct packet_offload *ptype;
2273 __be16 type = skb_network_protocol(skb);
2275 if (unlikely(!type))
2276 return ERR_PTR(-EINVAL);
2278 __skb_pull(skb, skb->mac_len);
2281 list_for_each_entry_rcu(ptype, &offload_base, list) {
2282 if (ptype->type == type && ptype->callbacks.gso_segment) {
2283 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2286 err = ptype->callbacks.gso_send_check(skb);
2287 segs = ERR_PTR(err);
2288 if (err || skb_gso_ok(skb, features))
2290 __skb_push(skb, (skb->data -
2291 skb_network_header(skb)));
2293 segs = ptype->callbacks.gso_segment(skb, features);
2299 __skb_push(skb, skb->data - skb_mac_header(skb));
2303 EXPORT_SYMBOL(skb_mac_gso_segment);
2306 /* openvswitch calls this on rx path, so we need a different check.
2308 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2311 return skb->ip_summed != CHECKSUM_PARTIAL;
2313 return skb->ip_summed == CHECKSUM_NONE;
2317 * __skb_gso_segment - Perform segmentation on skb.
2318 * @skb: buffer to segment
2319 * @features: features for the output path (see dev->features)
2320 * @tx_path: whether it is called in TX path
2322 * This function segments the given skb and returns a list of segments.
2324 * It may return NULL if the skb requires no segmentation. This is
2325 * only possible when GSO is used for verifying header integrity.
2327 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2328 netdev_features_t features, bool tx_path)
2330 if (unlikely(skb_needs_check(skb, tx_path))) {
2333 skb_warn_bad_offload(skb);
2335 if (skb_header_cloned(skb) &&
2336 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2337 return ERR_PTR(err);
2340 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2341 skb_reset_mac_header(skb);
2342 skb_reset_mac_len(skb);
2344 return skb_mac_gso_segment(skb, features);
2346 EXPORT_SYMBOL(__skb_gso_segment);
2348 /* Take action when hardware reception checksum errors are detected. */
2350 void netdev_rx_csum_fault(struct net_device *dev)
2352 if (net_ratelimit()) {
2353 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2357 EXPORT_SYMBOL(netdev_rx_csum_fault);
2360 /* Actually, we should eliminate this check as soon as we know, that:
2361 * 1. IOMMU is present and allows to map all the memory.
2362 * 2. No high memory really exists on this machine.
2365 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2367 #ifdef CONFIG_HIGHMEM
2369 if (!(dev->features & NETIF_F_HIGHDMA)) {
2370 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2371 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2372 if (PageHighMem(skb_frag_page(frag)))
2377 if (PCI_DMA_BUS_IS_PHYS) {
2378 struct device *pdev = dev->dev.parent;
2382 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2383 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2384 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2385 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2394 void (*destructor)(struct sk_buff *skb);
2397 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2399 static void dev_gso_skb_destructor(struct sk_buff *skb)
2401 struct dev_gso_cb *cb;
2404 struct sk_buff *nskb = skb->next;
2406 skb->next = nskb->next;
2409 } while (skb->next);
2411 cb = DEV_GSO_CB(skb);
2413 cb->destructor(skb);
2417 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2418 * @skb: buffer to segment
2419 * @features: device features as applicable to this skb
2421 * This function segments the given skb and stores the list of segments
2424 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2426 struct sk_buff *segs;
2428 segs = skb_gso_segment(skb, features);
2430 /* Verifying header integrity only. */
2435 return PTR_ERR(segs);
2438 DEV_GSO_CB(skb)->destructor = skb->destructor;
2439 skb->destructor = dev_gso_skb_destructor;
2444 static netdev_features_t harmonize_features(struct sk_buff *skb,
2445 __be16 protocol, netdev_features_t features)
2447 if (skb->ip_summed != CHECKSUM_NONE &&
2448 !can_checksum_protocol(features, protocol)) {
2449 features &= ~NETIF_F_ALL_CSUM;
2450 } else if (illegal_highdma(skb->dev, skb)) {
2451 features &= ~NETIF_F_SG;
2457 netdev_features_t netif_skb_features(struct sk_buff *skb)
2459 __be16 protocol = skb->protocol;
2460 netdev_features_t features = skb->dev->features;
2462 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2463 features &= ~NETIF_F_GSO_MASK;
2465 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2466 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2467 protocol = veh->h_vlan_encapsulated_proto;
2468 } else if (!vlan_tx_tag_present(skb)) {
2469 return harmonize_features(skb, protocol, features);
2472 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2473 NETIF_F_HW_VLAN_STAG_TX);
2475 if (protocol != htons(ETH_P_8021Q) && protocol != htons(ETH_P_8021AD)) {
2476 return harmonize_features(skb, protocol, features);
2478 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2479 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2480 NETIF_F_HW_VLAN_STAG_TX;
2481 return harmonize_features(skb, protocol, features);
2484 EXPORT_SYMBOL(netif_skb_features);
2487 * Returns true if either:
2488 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2489 * 2. skb is fragmented and the device does not support SG.
2491 static inline int skb_needs_linearize(struct sk_buff *skb,
2492 netdev_features_t features)
2494 return skb_is_nonlinear(skb) &&
2495 ((skb_has_frag_list(skb) &&
2496 !(features & NETIF_F_FRAGLIST)) ||
2497 (skb_shinfo(skb)->nr_frags &&
2498 !(features & NETIF_F_SG)));
2501 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2502 struct netdev_queue *txq)
2504 const struct net_device_ops *ops = dev->netdev_ops;
2505 int rc = NETDEV_TX_OK;
2506 unsigned int skb_len;
2508 if (likely(!skb->next)) {
2509 netdev_features_t features;
2512 * If device doesn't need skb->dst, release it right now while
2513 * its hot in this cpu cache
2515 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2518 features = netif_skb_features(skb);
2520 if (vlan_tx_tag_present(skb) &&
2521 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2522 skb = __vlan_put_tag(skb, skb->vlan_proto,
2523 vlan_tx_tag_get(skb));
2530 /* If encapsulation offload request, verify we are testing
2531 * hardware encapsulation features instead of standard
2532 * features for the netdev
2534 if (skb->encapsulation)
2535 features &= dev->hw_enc_features;
2537 if (netif_needs_gso(skb, features)) {
2538 if (unlikely(dev_gso_segment(skb, features)))
2543 if (skb_needs_linearize(skb, features) &&
2544 __skb_linearize(skb))
2547 /* If packet is not checksummed and device does not
2548 * support checksumming for this protocol, complete
2549 * checksumming here.
2551 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2552 if (skb->encapsulation)
2553 skb_set_inner_transport_header(skb,
2554 skb_checksum_start_offset(skb));
2556 skb_set_transport_header(skb,
2557 skb_checksum_start_offset(skb));
2558 if (!(features & NETIF_F_ALL_CSUM) &&
2559 skb_checksum_help(skb))
2564 if (!list_empty(&ptype_all))
2565 dev_queue_xmit_nit(skb, dev);
2568 rc = ops->ndo_start_xmit(skb, dev);
2569 trace_net_dev_xmit(skb, rc, dev, skb_len);
2570 if (rc == NETDEV_TX_OK)
2571 txq_trans_update(txq);
2577 struct sk_buff *nskb = skb->next;
2579 skb->next = nskb->next;
2582 if (!list_empty(&ptype_all))
2583 dev_queue_xmit_nit(nskb, dev);
2585 skb_len = nskb->len;
2586 rc = ops->ndo_start_xmit(nskb, dev);
2587 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2588 if (unlikely(rc != NETDEV_TX_OK)) {
2589 if (rc & ~NETDEV_TX_MASK)
2590 goto out_kfree_gso_skb;
2591 nskb->next = skb->next;
2595 txq_trans_update(txq);
2596 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2597 return NETDEV_TX_BUSY;
2598 } while (skb->next);
2601 if (likely(skb->next == NULL)) {
2602 skb->destructor = DEV_GSO_CB(skb)->destructor;
2612 static void qdisc_pkt_len_init(struct sk_buff *skb)
2614 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2616 qdisc_skb_cb(skb)->pkt_len = skb->len;
2618 /* To get more precise estimation of bytes sent on wire,
2619 * we add to pkt_len the headers size of all segments
2621 if (shinfo->gso_size) {
2622 unsigned int hdr_len;
2623 u16 gso_segs = shinfo->gso_segs;
2625 /* mac layer + network layer */
2626 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2628 /* + transport layer */
2629 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2630 hdr_len += tcp_hdrlen(skb);
2632 hdr_len += sizeof(struct udphdr);
2634 if (shinfo->gso_type & SKB_GSO_DODGY)
2635 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2638 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2642 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2643 struct net_device *dev,
2644 struct netdev_queue *txq)
2646 spinlock_t *root_lock = qdisc_lock(q);
2650 qdisc_pkt_len_init(skb);
2651 qdisc_calculate_pkt_len(skb, q);
2653 * Heuristic to force contended enqueues to serialize on a
2654 * separate lock before trying to get qdisc main lock.
2655 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2656 * and dequeue packets faster.
2658 contended = qdisc_is_running(q);
2659 if (unlikely(contended))
2660 spin_lock(&q->busylock);
2662 spin_lock(root_lock);
2663 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2666 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2667 qdisc_run_begin(q)) {
2669 * This is a work-conserving queue; there are no old skbs
2670 * waiting to be sent out; and the qdisc is not running -
2671 * xmit the skb directly.
2673 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2676 qdisc_bstats_update(q, skb);
2678 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2679 if (unlikely(contended)) {
2680 spin_unlock(&q->busylock);
2687 rc = NET_XMIT_SUCCESS;
2690 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2691 if (qdisc_run_begin(q)) {
2692 if (unlikely(contended)) {
2693 spin_unlock(&q->busylock);
2699 spin_unlock(root_lock);
2700 if (unlikely(contended))
2701 spin_unlock(&q->busylock);
2705 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2706 static void skb_update_prio(struct sk_buff *skb)
2708 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2710 if (!skb->priority && skb->sk && map) {
2711 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2713 if (prioidx < map->priomap_len)
2714 skb->priority = map->priomap[prioidx];
2718 #define skb_update_prio(skb)
2721 static DEFINE_PER_CPU(int, xmit_recursion);
2722 #define RECURSION_LIMIT 10
2725 * dev_loopback_xmit - loop back @skb
2726 * @skb: buffer to transmit
2728 int dev_loopback_xmit(struct sk_buff *skb)
2730 skb_reset_mac_header(skb);
2731 __skb_pull(skb, skb_network_offset(skb));
2732 skb->pkt_type = PACKET_LOOPBACK;
2733 skb->ip_summed = CHECKSUM_UNNECESSARY;
2734 WARN_ON(!skb_dst(skb));
2739 EXPORT_SYMBOL(dev_loopback_xmit);
2742 * dev_queue_xmit - transmit a buffer
2743 * @skb: buffer to transmit
2745 * Queue a buffer for transmission to a network device. The caller must
2746 * have set the device and priority and built the buffer before calling
2747 * this function. The function can be called from an interrupt.
2749 * A negative errno code is returned on a failure. A success does not
2750 * guarantee the frame will be transmitted as it may be dropped due
2751 * to congestion or traffic shaping.
2753 * -----------------------------------------------------------------------------------
2754 * I notice this method can also return errors from the queue disciplines,
2755 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2758 * Regardless of the return value, the skb is consumed, so it is currently
2759 * difficult to retry a send to this method. (You can bump the ref count
2760 * before sending to hold a reference for retry if you are careful.)
2762 * When calling this method, interrupts MUST be enabled. This is because
2763 * the BH enable code must have IRQs enabled so that it will not deadlock.
2766 int dev_queue_xmit(struct sk_buff *skb)
2768 struct net_device *dev = skb->dev;
2769 struct netdev_queue *txq;
2773 skb_reset_mac_header(skb);
2775 /* Disable soft irqs for various locks below. Also
2776 * stops preemption for RCU.
2780 skb_update_prio(skb);
2782 txq = netdev_pick_tx(dev, skb);
2783 q = rcu_dereference_bh(txq->qdisc);
2785 #ifdef CONFIG_NET_CLS_ACT
2786 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2788 trace_net_dev_queue(skb);
2790 rc = __dev_xmit_skb(skb, q, dev, txq);
2794 /* The device has no queue. Common case for software devices:
2795 loopback, all the sorts of tunnels...
2797 Really, it is unlikely that netif_tx_lock protection is necessary
2798 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2800 However, it is possible, that they rely on protection
2803 Check this and shot the lock. It is not prone from deadlocks.
2804 Either shot noqueue qdisc, it is even simpler 8)
2806 if (dev->flags & IFF_UP) {
2807 int cpu = smp_processor_id(); /* ok because BHs are off */
2809 if (txq->xmit_lock_owner != cpu) {
2811 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2812 goto recursion_alert;
2814 HARD_TX_LOCK(dev, txq, cpu);
2816 if (!netif_xmit_stopped(txq)) {
2817 __this_cpu_inc(xmit_recursion);
2818 rc = dev_hard_start_xmit(skb, dev, txq);
2819 __this_cpu_dec(xmit_recursion);
2820 if (dev_xmit_complete(rc)) {
2821 HARD_TX_UNLOCK(dev, txq);
2825 HARD_TX_UNLOCK(dev, txq);
2826 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2829 /* Recursion is detected! It is possible,
2833 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2839 rcu_read_unlock_bh();
2844 rcu_read_unlock_bh();
2847 EXPORT_SYMBOL(dev_queue_xmit);
2850 /*=======================================================================
2852 =======================================================================*/
2854 int netdev_max_backlog __read_mostly = 1000;
2855 EXPORT_SYMBOL(netdev_max_backlog);
2857 int netdev_tstamp_prequeue __read_mostly = 1;
2858 int netdev_budget __read_mostly = 300;
2859 int weight_p __read_mostly = 64; /* old backlog weight */
2861 /* Called with irq disabled */
2862 static inline void ____napi_schedule(struct softnet_data *sd,
2863 struct napi_struct *napi)
2865 list_add_tail(&napi->poll_list, &sd->poll_list);
2866 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2871 /* One global table that all flow-based protocols share. */
2872 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2873 EXPORT_SYMBOL(rps_sock_flow_table);
2875 struct static_key rps_needed __read_mostly;
2877 static struct rps_dev_flow *
2878 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2879 struct rps_dev_flow *rflow, u16 next_cpu)
2881 if (next_cpu != RPS_NO_CPU) {
2882 #ifdef CONFIG_RFS_ACCEL
2883 struct netdev_rx_queue *rxqueue;
2884 struct rps_dev_flow_table *flow_table;
2885 struct rps_dev_flow *old_rflow;
2890 /* Should we steer this flow to a different hardware queue? */
2891 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2892 !(dev->features & NETIF_F_NTUPLE))
2894 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2895 if (rxq_index == skb_get_rx_queue(skb))
2898 rxqueue = dev->_rx + rxq_index;
2899 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2902 flow_id = skb->rxhash & flow_table->mask;
2903 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2904 rxq_index, flow_id);
2908 rflow = &flow_table->flows[flow_id];
2910 if (old_rflow->filter == rflow->filter)
2911 old_rflow->filter = RPS_NO_FILTER;
2915 per_cpu(softnet_data, next_cpu).input_queue_head;
2918 rflow->cpu = next_cpu;
2923 * get_rps_cpu is called from netif_receive_skb and returns the target
2924 * CPU from the RPS map of the receiving queue for a given skb.
2925 * rcu_read_lock must be held on entry.
2927 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2928 struct rps_dev_flow **rflowp)
2930 struct netdev_rx_queue *rxqueue;
2931 struct rps_map *map;
2932 struct rps_dev_flow_table *flow_table;
2933 struct rps_sock_flow_table *sock_flow_table;
2937 if (skb_rx_queue_recorded(skb)) {
2938 u16 index = skb_get_rx_queue(skb);
2939 if (unlikely(index >= dev->real_num_rx_queues)) {
2940 WARN_ONCE(dev->real_num_rx_queues > 1,
2941 "%s received packet on queue %u, but number "
2942 "of RX queues is %u\n",
2943 dev->name, index, dev->real_num_rx_queues);
2946 rxqueue = dev->_rx + index;
2950 map = rcu_dereference(rxqueue->rps_map);
2952 if (map->len == 1 &&
2953 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2954 tcpu = map->cpus[0];
2955 if (cpu_online(tcpu))
2959 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2963 skb_reset_network_header(skb);
2964 if (!skb_get_rxhash(skb))
2967 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2968 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2969 if (flow_table && sock_flow_table) {
2971 struct rps_dev_flow *rflow;
2973 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2976 next_cpu = sock_flow_table->ents[skb->rxhash &
2977 sock_flow_table->mask];
2980 * If the desired CPU (where last recvmsg was done) is
2981 * different from current CPU (one in the rx-queue flow
2982 * table entry), switch if one of the following holds:
2983 * - Current CPU is unset (equal to RPS_NO_CPU).
2984 * - Current CPU is offline.
2985 * - The current CPU's queue tail has advanced beyond the
2986 * last packet that was enqueued using this table entry.
2987 * This guarantees that all previous packets for the flow
2988 * have been dequeued, thus preserving in order delivery.
2990 if (unlikely(tcpu != next_cpu) &&
2991 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2992 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2993 rflow->last_qtail)) >= 0)) {
2995 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2998 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3006 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
3008 if (cpu_online(tcpu)) {
3018 #ifdef CONFIG_RFS_ACCEL
3021 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3022 * @dev: Device on which the filter was set
3023 * @rxq_index: RX queue index
3024 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3025 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3027 * Drivers that implement ndo_rx_flow_steer() should periodically call
3028 * this function for each installed filter and remove the filters for
3029 * which it returns %true.
3031 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3032 u32 flow_id, u16 filter_id)
3034 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3035 struct rps_dev_flow_table *flow_table;
3036 struct rps_dev_flow *rflow;
3041 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3042 if (flow_table && flow_id <= flow_table->mask) {
3043 rflow = &flow_table->flows[flow_id];
3044 cpu = ACCESS_ONCE(rflow->cpu);
3045 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3046 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3047 rflow->last_qtail) <
3048 (int)(10 * flow_table->mask)))
3054 EXPORT_SYMBOL(rps_may_expire_flow);
3056 #endif /* CONFIG_RFS_ACCEL */
3058 /* Called from hardirq (IPI) context */
3059 static void rps_trigger_softirq(void *data)
3061 struct softnet_data *sd = data;
3063 ____napi_schedule(sd, &sd->backlog);
3067 #endif /* CONFIG_RPS */
3070 * Check if this softnet_data structure is another cpu one
3071 * If yes, queue it to our IPI list and return 1
3074 static int rps_ipi_queued(struct softnet_data *sd)
3077 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3080 sd->rps_ipi_next = mysd->rps_ipi_list;
3081 mysd->rps_ipi_list = sd;
3083 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3086 #endif /* CONFIG_RPS */
3090 #ifdef CONFIG_NET_FLOW_LIMIT
3091 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3094 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3096 #ifdef CONFIG_NET_FLOW_LIMIT
3097 struct sd_flow_limit *fl;
3098 struct softnet_data *sd;
3099 unsigned int old_flow, new_flow;
3101 if (qlen < (netdev_max_backlog >> 1))
3104 sd = &__get_cpu_var(softnet_data);
3107 fl = rcu_dereference(sd->flow_limit);
3109 new_flow = skb_get_rxhash(skb) & (fl->num_buckets - 1);
3110 old_flow = fl->history[fl->history_head];
3111 fl->history[fl->history_head] = new_flow;
3114 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3116 if (likely(fl->buckets[old_flow]))
3117 fl->buckets[old_flow]--;
3119 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3131 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3132 * queue (may be a remote CPU queue).
3134 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3135 unsigned int *qtail)
3137 struct softnet_data *sd;
3138 unsigned long flags;
3141 sd = &per_cpu(softnet_data, cpu);
3143 local_irq_save(flags);
3146 qlen = skb_queue_len(&sd->input_pkt_queue);
3147 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3148 if (skb_queue_len(&sd->input_pkt_queue)) {
3150 __skb_queue_tail(&sd->input_pkt_queue, skb);
3151 input_queue_tail_incr_save(sd, qtail);
3153 local_irq_restore(flags);
3154 return NET_RX_SUCCESS;
3157 /* Schedule NAPI for backlog device
3158 * We can use non atomic operation since we own the queue lock
3160 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3161 if (!rps_ipi_queued(sd))
3162 ____napi_schedule(sd, &sd->backlog);
3170 local_irq_restore(flags);
3172 atomic_long_inc(&skb->dev->rx_dropped);
3178 * netif_rx - post buffer to the network code
3179 * @skb: buffer to post
3181 * This function receives a packet from a device driver and queues it for
3182 * the upper (protocol) levels to process. It always succeeds. The buffer
3183 * may be dropped during processing for congestion control or by the
3187 * NET_RX_SUCCESS (no congestion)
3188 * NET_RX_DROP (packet was dropped)
3192 int netif_rx(struct sk_buff *skb)
3196 /* if netpoll wants it, pretend we never saw it */
3197 if (netpoll_rx(skb))
3200 net_timestamp_check(netdev_tstamp_prequeue, skb);
3202 trace_netif_rx(skb);
3204 if (static_key_false(&rps_needed)) {
3205 struct rps_dev_flow voidflow, *rflow = &voidflow;
3211 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3213 cpu = smp_processor_id();
3215 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3223 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3228 EXPORT_SYMBOL(netif_rx);
3230 int netif_rx_ni(struct sk_buff *skb)
3235 err = netif_rx(skb);
3236 if (local_softirq_pending())
3242 EXPORT_SYMBOL(netif_rx_ni);
3244 static void net_tx_action(struct softirq_action *h)
3246 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3248 if (sd->completion_queue) {
3249 struct sk_buff *clist;
3251 local_irq_disable();
3252 clist = sd->completion_queue;
3253 sd->completion_queue = NULL;
3257 struct sk_buff *skb = clist;
3258 clist = clist->next;
3260 WARN_ON(atomic_read(&skb->users));
3261 trace_kfree_skb(skb, net_tx_action);
3266 if (sd->output_queue) {
3269 local_irq_disable();
3270 head = sd->output_queue;
3271 sd->output_queue = NULL;
3272 sd->output_queue_tailp = &sd->output_queue;
3276 struct Qdisc *q = head;
3277 spinlock_t *root_lock;
3279 head = head->next_sched;
3281 root_lock = qdisc_lock(q);
3282 if (spin_trylock(root_lock)) {
3283 smp_mb__before_clear_bit();
3284 clear_bit(__QDISC_STATE_SCHED,
3287 spin_unlock(root_lock);
3289 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3291 __netif_reschedule(q);
3293 smp_mb__before_clear_bit();
3294 clear_bit(__QDISC_STATE_SCHED,
3302 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3303 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3304 /* This hook is defined here for ATM LANE */
3305 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3306 unsigned char *addr) __read_mostly;
3307 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3310 #ifdef CONFIG_NET_CLS_ACT
3311 /* TODO: Maybe we should just force sch_ingress to be compiled in
3312 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3313 * a compare and 2 stores extra right now if we dont have it on
3314 * but have CONFIG_NET_CLS_ACT
3315 * NOTE: This doesn't stop any functionality; if you dont have
3316 * the ingress scheduler, you just can't add policies on ingress.
3319 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3321 struct net_device *dev = skb->dev;
3322 u32 ttl = G_TC_RTTL(skb->tc_verd);
3323 int result = TC_ACT_OK;
3326 if (unlikely(MAX_RED_LOOP < ttl++)) {
3327 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3328 skb->skb_iif, dev->ifindex);
3332 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3333 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3336 if (q != &noop_qdisc) {
3337 spin_lock(qdisc_lock(q));
3338 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3339 result = qdisc_enqueue_root(skb, q);
3340 spin_unlock(qdisc_lock(q));
3346 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3347 struct packet_type **pt_prev,
3348 int *ret, struct net_device *orig_dev)
3350 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3352 if (!rxq || rxq->qdisc == &noop_qdisc)
3356 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3360 switch (ing_filter(skb, rxq)) {
3374 * netdev_rx_handler_register - register receive handler
3375 * @dev: device to register a handler for
3376 * @rx_handler: receive handler to register
3377 * @rx_handler_data: data pointer that is used by rx handler
3379 * Register a receive hander for a device. This handler will then be
3380 * called from __netif_receive_skb. A negative errno code is returned
3383 * The caller must hold the rtnl_mutex.
3385 * For a general description of rx_handler, see enum rx_handler_result.
3387 int netdev_rx_handler_register(struct net_device *dev,
3388 rx_handler_func_t *rx_handler,
3389 void *rx_handler_data)
3393 if (dev->rx_handler)
3396 /* Note: rx_handler_data must be set before rx_handler */
3397 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3398 rcu_assign_pointer(dev->rx_handler, rx_handler);
3402 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3405 * netdev_rx_handler_unregister - unregister receive handler
3406 * @dev: device to unregister a handler from
3408 * Unregister a receive handler from a device.
3410 * The caller must hold the rtnl_mutex.
3412 void netdev_rx_handler_unregister(struct net_device *dev)
3416 RCU_INIT_POINTER(dev->rx_handler, NULL);
3417 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3418 * section has a guarantee to see a non NULL rx_handler_data
3422 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3424 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3427 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3428 * the special handling of PFMEMALLOC skbs.
3430 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3432 switch (skb->protocol) {
3433 case __constant_htons(ETH_P_ARP):
3434 case __constant_htons(ETH_P_IP):
3435 case __constant_htons(ETH_P_IPV6):
3436 case __constant_htons(ETH_P_8021Q):
3437 case __constant_htons(ETH_P_8021AD):
3444 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3446 struct packet_type *ptype, *pt_prev;
3447 rx_handler_func_t *rx_handler;
3448 struct net_device *orig_dev;
3449 struct net_device *null_or_dev;
3450 bool deliver_exact = false;
3451 int ret = NET_RX_DROP;
3454 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3456 trace_netif_receive_skb(skb);
3458 /* if we've gotten here through NAPI, check netpoll */
3459 if (netpoll_receive_skb(skb))
3462 orig_dev = skb->dev;
3464 skb_reset_network_header(skb);
3465 if (!skb_transport_header_was_set(skb))
3466 skb_reset_transport_header(skb);
3467 skb_reset_mac_len(skb);
3474 skb->skb_iif = skb->dev->ifindex;
3476 __this_cpu_inc(softnet_data.processed);
3478 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3479 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3480 skb = vlan_untag(skb);
3485 #ifdef CONFIG_NET_CLS_ACT
3486 if (skb->tc_verd & TC_NCLS) {
3487 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3495 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3496 if (!ptype->dev || ptype->dev == skb->dev) {
3498 ret = deliver_skb(skb, pt_prev, orig_dev);
3504 #ifdef CONFIG_NET_CLS_ACT
3505 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3511 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3514 if (vlan_tx_tag_present(skb)) {
3516 ret = deliver_skb(skb, pt_prev, orig_dev);
3519 if (vlan_do_receive(&skb))
3521 else if (unlikely(!skb))
3525 rx_handler = rcu_dereference(skb->dev->rx_handler);
3528 ret = deliver_skb(skb, pt_prev, orig_dev);
3531 switch (rx_handler(&skb)) {
3532 case RX_HANDLER_CONSUMED:
3533 ret = NET_RX_SUCCESS;
3535 case RX_HANDLER_ANOTHER:
3537 case RX_HANDLER_EXACT:
3538 deliver_exact = true;
3539 case RX_HANDLER_PASS:
3546 if (vlan_tx_nonzero_tag_present(skb))
3547 skb->pkt_type = PACKET_OTHERHOST;
3549 /* deliver only exact match when indicated */
3550 null_or_dev = deliver_exact ? skb->dev : NULL;
3552 type = skb->protocol;
3553 list_for_each_entry_rcu(ptype,
3554 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3555 if (ptype->type == type &&
3556 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3557 ptype->dev == orig_dev)) {
3559 ret = deliver_skb(skb, pt_prev, orig_dev);
3565 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3568 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3571 atomic_long_inc(&skb->dev->rx_dropped);
3573 /* Jamal, now you will not able to escape explaining
3574 * me how you were going to use this. :-)
3585 static int __netif_receive_skb(struct sk_buff *skb)
3589 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3590 unsigned long pflags = current->flags;
3593 * PFMEMALLOC skbs are special, they should
3594 * - be delivered to SOCK_MEMALLOC sockets only
3595 * - stay away from userspace
3596 * - have bounded memory usage
3598 * Use PF_MEMALLOC as this saves us from propagating the allocation
3599 * context down to all allocation sites.
3601 current->flags |= PF_MEMALLOC;
3602 ret = __netif_receive_skb_core(skb, true);
3603 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3605 ret = __netif_receive_skb_core(skb, false);
3611 * netif_receive_skb - process receive buffer from network
3612 * @skb: buffer to process
3614 * netif_receive_skb() is the main receive data processing function.
3615 * It always succeeds. The buffer may be dropped during processing
3616 * for congestion control or by the protocol layers.
3618 * This function may only be called from softirq context and interrupts
3619 * should be enabled.
3621 * Return values (usually ignored):
3622 * NET_RX_SUCCESS: no congestion
3623 * NET_RX_DROP: packet was dropped
3625 int netif_receive_skb(struct sk_buff *skb)
3627 net_timestamp_check(netdev_tstamp_prequeue, skb);
3629 if (skb_defer_rx_timestamp(skb))
3630 return NET_RX_SUCCESS;
3633 if (static_key_false(&rps_needed)) {
3634 struct rps_dev_flow voidflow, *rflow = &voidflow;
3639 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3642 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3649 return __netif_receive_skb(skb);
3651 EXPORT_SYMBOL(netif_receive_skb);
3653 /* Network device is going away, flush any packets still pending
3654 * Called with irqs disabled.
3656 static void flush_backlog(void *arg)
3658 struct net_device *dev = arg;
3659 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3660 struct sk_buff *skb, *tmp;
3663 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3664 if (skb->dev == dev) {
3665 __skb_unlink(skb, &sd->input_pkt_queue);
3667 input_queue_head_incr(sd);
3672 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3673 if (skb->dev == dev) {
3674 __skb_unlink(skb, &sd->process_queue);
3676 input_queue_head_incr(sd);
3681 static int napi_gro_complete(struct sk_buff *skb)
3683 struct packet_offload *ptype;
3684 __be16 type = skb->protocol;
3685 struct list_head *head = &offload_base;
3688 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3690 if (NAPI_GRO_CB(skb)->count == 1) {
3691 skb_shinfo(skb)->gso_size = 0;
3696 list_for_each_entry_rcu(ptype, head, list) {
3697 if (ptype->type != type || !ptype->callbacks.gro_complete)
3700 err = ptype->callbacks.gro_complete(skb);
3706 WARN_ON(&ptype->list == head);
3708 return NET_RX_SUCCESS;
3712 return netif_receive_skb(skb);
3715 /* napi->gro_list contains packets ordered by age.
3716 * youngest packets at the head of it.
3717 * Complete skbs in reverse order to reduce latencies.
3719 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3721 struct sk_buff *skb, *prev = NULL;
3723 /* scan list and build reverse chain */
3724 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3729 for (skb = prev; skb; skb = prev) {
3732 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3736 napi_gro_complete(skb);
3740 napi->gro_list = NULL;
3742 EXPORT_SYMBOL(napi_gro_flush);
3744 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3747 unsigned int maclen = skb->dev->hard_header_len;
3749 for (p = napi->gro_list; p; p = p->next) {
3750 unsigned long diffs;
3752 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3753 diffs |= p->vlan_tci ^ skb->vlan_tci;
3754 if (maclen == ETH_HLEN)
3755 diffs |= compare_ether_header(skb_mac_header(p),
3756 skb_gro_mac_header(skb));
3758 diffs = memcmp(skb_mac_header(p),
3759 skb_gro_mac_header(skb),
3761 NAPI_GRO_CB(p)->same_flow = !diffs;
3762 NAPI_GRO_CB(p)->flush = 0;
3766 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3768 struct sk_buff **pp = NULL;
3769 struct packet_offload *ptype;
3770 __be16 type = skb->protocol;
3771 struct list_head *head = &offload_base;
3773 enum gro_result ret;
3775 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3778 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3781 gro_list_prepare(napi, skb);
3784 list_for_each_entry_rcu(ptype, head, list) {
3785 if (ptype->type != type || !ptype->callbacks.gro_receive)
3788 skb_set_network_header(skb, skb_gro_offset(skb));
3789 skb_reset_mac_len(skb);
3790 NAPI_GRO_CB(skb)->same_flow = 0;
3791 NAPI_GRO_CB(skb)->flush = 0;
3792 NAPI_GRO_CB(skb)->free = 0;
3794 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3799 if (&ptype->list == head)
3802 same_flow = NAPI_GRO_CB(skb)->same_flow;
3803 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3806 struct sk_buff *nskb = *pp;
3810 napi_gro_complete(nskb);
3817 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3821 NAPI_GRO_CB(skb)->count = 1;
3822 NAPI_GRO_CB(skb)->age = jiffies;
3823 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3824 skb->next = napi->gro_list;
3825 napi->gro_list = skb;
3829 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3830 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3832 BUG_ON(skb->end - skb->tail < grow);
3834 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3837 skb->data_len -= grow;
3839 skb_shinfo(skb)->frags[0].page_offset += grow;
3840 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3842 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3843 skb_frag_unref(skb, 0);
3844 memmove(skb_shinfo(skb)->frags,
3845 skb_shinfo(skb)->frags + 1,
3846 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3859 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3863 if (netif_receive_skb(skb))
3871 case GRO_MERGED_FREE:
3872 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3873 kmem_cache_free(skbuff_head_cache, skb);
3886 static void skb_gro_reset_offset(struct sk_buff *skb)
3888 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3889 const skb_frag_t *frag0 = &pinfo->frags[0];
3891 NAPI_GRO_CB(skb)->data_offset = 0;
3892 NAPI_GRO_CB(skb)->frag0 = NULL;
3893 NAPI_GRO_CB(skb)->frag0_len = 0;
3895 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3897 !PageHighMem(skb_frag_page(frag0))) {
3898 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3899 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3903 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3905 skb_gro_reset_offset(skb);
3907 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
3909 EXPORT_SYMBOL(napi_gro_receive);
3911 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3913 __skb_pull(skb, skb_headlen(skb));
3914 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3915 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3917 skb->dev = napi->dev;
3923 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3925 struct sk_buff *skb = napi->skb;
3928 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3934 EXPORT_SYMBOL(napi_get_frags);
3936 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3942 skb->protocol = eth_type_trans(skb, skb->dev);
3944 if (ret == GRO_HELD)
3945 skb_gro_pull(skb, -ETH_HLEN);
3946 else if (netif_receive_skb(skb))
3951 case GRO_MERGED_FREE:
3952 napi_reuse_skb(napi, skb);
3962 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3964 struct sk_buff *skb = napi->skb;
3971 skb_reset_mac_header(skb);
3972 skb_gro_reset_offset(skb);
3974 off = skb_gro_offset(skb);
3975 hlen = off + sizeof(*eth);
3976 eth = skb_gro_header_fast(skb, off);
3977 if (skb_gro_header_hard(skb, hlen)) {
3978 eth = skb_gro_header_slow(skb, hlen, off);
3979 if (unlikely(!eth)) {
3980 napi_reuse_skb(napi, skb);
3986 skb_gro_pull(skb, sizeof(*eth));
3989 * This works because the only protocols we care about don't require
3990 * special handling. We'll fix it up properly at the end.
3992 skb->protocol = eth->h_proto;
3998 gro_result_t napi_gro_frags(struct napi_struct *napi)
4000 struct sk_buff *skb = napi_frags_skb(napi);
4005 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4007 EXPORT_SYMBOL(napi_gro_frags);
4010 * net_rps_action sends any pending IPI's for rps.
4011 * Note: called with local irq disabled, but exits with local irq enabled.
4013 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4016 struct softnet_data *remsd = sd->rps_ipi_list;
4019 sd->rps_ipi_list = NULL;
4023 /* Send pending IPI's to kick RPS processing on remote cpus. */
4025 struct softnet_data *next = remsd->rps_ipi_next;
4027 if (cpu_online(remsd->cpu))
4028 __smp_call_function_single(remsd->cpu,
4037 static int process_backlog(struct napi_struct *napi, int quota)
4040 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4043 /* Check if we have pending ipi, its better to send them now,
4044 * not waiting net_rx_action() end.
4046 if (sd->rps_ipi_list) {
4047 local_irq_disable();
4048 net_rps_action_and_irq_enable(sd);
4051 napi->weight = weight_p;
4052 local_irq_disable();
4053 while (work < quota) {
4054 struct sk_buff *skb;
4057 while ((skb = __skb_dequeue(&sd->process_queue))) {
4059 __netif_receive_skb(skb);
4060 local_irq_disable();
4061 input_queue_head_incr(sd);
4062 if (++work >= quota) {
4069 qlen = skb_queue_len(&sd->input_pkt_queue);
4071 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4072 &sd->process_queue);
4074 if (qlen < quota - work) {
4076 * Inline a custom version of __napi_complete().
4077 * only current cpu owns and manipulates this napi,
4078 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4079 * we can use a plain write instead of clear_bit(),
4080 * and we dont need an smp_mb() memory barrier.
4082 list_del(&napi->poll_list);
4085 quota = work + qlen;
4095 * __napi_schedule - schedule for receive
4096 * @n: entry to schedule
4098 * The entry's receive function will be scheduled to run
4100 void __napi_schedule(struct napi_struct *n)
4102 unsigned long flags;
4104 local_irq_save(flags);
4105 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4106 local_irq_restore(flags);
4108 EXPORT_SYMBOL(__napi_schedule);
4110 void __napi_complete(struct napi_struct *n)
4112 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4113 BUG_ON(n->gro_list);
4115 list_del(&n->poll_list);
4116 smp_mb__before_clear_bit();
4117 clear_bit(NAPI_STATE_SCHED, &n->state);
4119 EXPORT_SYMBOL(__napi_complete);
4121 void napi_complete(struct napi_struct *n)
4123 unsigned long flags;
4126 * don't let napi dequeue from the cpu poll list
4127 * just in case its running on a different cpu
4129 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4132 napi_gro_flush(n, false);
4133 local_irq_save(flags);
4135 local_irq_restore(flags);
4137 EXPORT_SYMBOL(napi_complete);
4139 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4140 int (*poll)(struct napi_struct *, int), int weight)
4142 INIT_LIST_HEAD(&napi->poll_list);
4143 napi->gro_count = 0;
4144 napi->gro_list = NULL;
4147 if (weight > NAPI_POLL_WEIGHT)
4148 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4150 napi->weight = weight;
4151 list_add(&napi->dev_list, &dev->napi_list);
4153 #ifdef CONFIG_NETPOLL
4154 spin_lock_init(&napi->poll_lock);
4155 napi->poll_owner = -1;
4157 set_bit(NAPI_STATE_SCHED, &napi->state);
4159 EXPORT_SYMBOL(netif_napi_add);
4161 void netif_napi_del(struct napi_struct *napi)
4163 struct sk_buff *skb, *next;
4165 list_del_init(&napi->dev_list);
4166 napi_free_frags(napi);
4168 for (skb = napi->gro_list; skb; skb = next) {
4174 napi->gro_list = NULL;
4175 napi->gro_count = 0;
4177 EXPORT_SYMBOL(netif_napi_del);
4179 static void net_rx_action(struct softirq_action *h)
4181 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4182 unsigned long time_limit = jiffies + 2;
4183 int budget = netdev_budget;
4186 local_irq_disable();
4188 while (!list_empty(&sd->poll_list)) {
4189 struct napi_struct *n;
4192 /* If softirq window is exhuasted then punt.
4193 * Allow this to run for 2 jiffies since which will allow
4194 * an average latency of 1.5/HZ.
4196 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4201 /* Even though interrupts have been re-enabled, this
4202 * access is safe because interrupts can only add new
4203 * entries to the tail of this list, and only ->poll()
4204 * calls can remove this head entry from the list.
4206 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4208 have = netpoll_poll_lock(n);
4212 /* This NAPI_STATE_SCHED test is for avoiding a race
4213 * with netpoll's poll_napi(). Only the entity which
4214 * obtains the lock and sees NAPI_STATE_SCHED set will
4215 * actually make the ->poll() call. Therefore we avoid
4216 * accidentally calling ->poll() when NAPI is not scheduled.
4219 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4220 work = n->poll(n, weight);
4224 WARN_ON_ONCE(work > weight);
4228 local_irq_disable();
4230 /* Drivers must not modify the NAPI state if they
4231 * consume the entire weight. In such cases this code
4232 * still "owns" the NAPI instance and therefore can
4233 * move the instance around on the list at-will.
4235 if (unlikely(work == weight)) {
4236 if (unlikely(napi_disable_pending(n))) {
4239 local_irq_disable();
4242 /* flush too old packets
4243 * If HZ < 1000, flush all packets.
4246 napi_gro_flush(n, HZ >= 1000);
4247 local_irq_disable();
4249 list_move_tail(&n->poll_list, &sd->poll_list);
4253 netpoll_poll_unlock(have);
4256 net_rps_action_and_irq_enable(sd);
4258 #ifdef CONFIG_NET_DMA
4260 * There may not be any more sk_buffs coming right now, so push
4261 * any pending DMA copies to hardware
4263 dma_issue_pending_all();
4270 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4274 struct netdev_upper {
4275 struct net_device *dev;
4277 struct list_head list;
4278 struct rcu_head rcu;
4279 struct list_head search_list;
4282 static void __append_search_uppers(struct list_head *search_list,
4283 struct net_device *dev)
4285 struct netdev_upper *upper;
4287 list_for_each_entry(upper, &dev->upper_dev_list, list) {
4288 /* check if this upper is not already in search list */
4289 if (list_empty(&upper->search_list))
4290 list_add_tail(&upper->search_list, search_list);
4294 static bool __netdev_search_upper_dev(struct net_device *dev,
4295 struct net_device *upper_dev)
4297 LIST_HEAD(search_list);
4298 struct netdev_upper *upper;
4299 struct netdev_upper *tmp;
4302 __append_search_uppers(&search_list, dev);
4303 list_for_each_entry(upper, &search_list, search_list) {
4304 if (upper->dev == upper_dev) {
4308 __append_search_uppers(&search_list, upper->dev);
4310 list_for_each_entry_safe(upper, tmp, &search_list, search_list)
4311 INIT_LIST_HEAD(&upper->search_list);
4315 static struct netdev_upper *__netdev_find_upper(struct net_device *dev,
4316 struct net_device *upper_dev)
4318 struct netdev_upper *upper;
4320 list_for_each_entry(upper, &dev->upper_dev_list, list) {
4321 if (upper->dev == upper_dev)
4328 * netdev_has_upper_dev - Check if device is linked to an upper device
4330 * @upper_dev: upper device to check
4332 * Find out if a device is linked to specified upper device and return true
4333 * in case it is. Note that this checks only immediate upper device,
4334 * not through a complete stack of devices. The caller must hold the RTNL lock.
4336 bool netdev_has_upper_dev(struct net_device *dev,
4337 struct net_device *upper_dev)
4341 return __netdev_find_upper(dev, upper_dev);
4343 EXPORT_SYMBOL(netdev_has_upper_dev);
4346 * netdev_has_any_upper_dev - Check if device is linked to some device
4349 * Find out if a device is linked to an upper device and return true in case
4350 * it is. The caller must hold the RTNL lock.
4352 bool netdev_has_any_upper_dev(struct net_device *dev)
4356 return !list_empty(&dev->upper_dev_list);
4358 EXPORT_SYMBOL(netdev_has_any_upper_dev);
4361 * netdev_master_upper_dev_get - Get master upper device
4364 * Find a master upper device and return pointer to it or NULL in case
4365 * it's not there. The caller must hold the RTNL lock.
4367 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4369 struct netdev_upper *upper;
4373 if (list_empty(&dev->upper_dev_list))
4376 upper = list_first_entry(&dev->upper_dev_list,
4377 struct netdev_upper, list);
4378 if (likely(upper->master))
4382 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4385 * netdev_master_upper_dev_get_rcu - Get master upper device
4388 * Find a master upper device and return pointer to it or NULL in case
4389 * it's not there. The caller must hold the RCU read lock.
4391 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4393 struct netdev_upper *upper;
4395 upper = list_first_or_null_rcu(&dev->upper_dev_list,
4396 struct netdev_upper, list);
4397 if (upper && likely(upper->master))
4401 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4403 static int __netdev_upper_dev_link(struct net_device *dev,
4404 struct net_device *upper_dev, bool master)
4406 struct netdev_upper *upper;
4410 if (dev == upper_dev)
4413 /* To prevent loops, check if dev is not upper device to upper_dev. */
4414 if (__netdev_search_upper_dev(upper_dev, dev))
4417 if (__netdev_find_upper(dev, upper_dev))
4420 if (master && netdev_master_upper_dev_get(dev))
4423 upper = kmalloc(sizeof(*upper), GFP_KERNEL);
4427 upper->dev = upper_dev;
4428 upper->master = master;
4429 INIT_LIST_HEAD(&upper->search_list);
4431 /* Ensure that master upper link is always the first item in list. */
4433 list_add_rcu(&upper->list, &dev->upper_dev_list);
4435 list_add_tail_rcu(&upper->list, &dev->upper_dev_list);
4436 dev_hold(upper_dev);
4437 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4442 * netdev_upper_dev_link - Add a link to the upper device
4444 * @upper_dev: new upper device
4446 * Adds a link to device which is upper to this one. The caller must hold
4447 * the RTNL lock. On a failure a negative errno code is returned.
4448 * On success the reference counts are adjusted and the function
4451 int netdev_upper_dev_link(struct net_device *dev,
4452 struct net_device *upper_dev)
4454 return __netdev_upper_dev_link(dev, upper_dev, false);
4456 EXPORT_SYMBOL(netdev_upper_dev_link);
4459 * netdev_master_upper_dev_link - Add a master link to the upper device
4461 * @upper_dev: new upper device
4463 * Adds a link to device which is upper to this one. In this case, only
4464 * one master upper device can be linked, although other non-master devices
4465 * might be linked as well. The caller must hold the RTNL lock.
4466 * On a failure a negative errno code is returned. On success the reference
4467 * counts are adjusted and the function returns zero.
4469 int netdev_master_upper_dev_link(struct net_device *dev,
4470 struct net_device *upper_dev)
4472 return __netdev_upper_dev_link(dev, upper_dev, true);
4474 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4477 * netdev_upper_dev_unlink - Removes a link to upper device
4479 * @upper_dev: new upper device
4481 * Removes a link to device which is upper to this one. The caller must hold
4484 void netdev_upper_dev_unlink(struct net_device *dev,
4485 struct net_device *upper_dev)
4487 struct netdev_upper *upper;
4491 upper = __netdev_find_upper(dev, upper_dev);
4494 list_del_rcu(&upper->list);
4496 kfree_rcu(upper, rcu);
4497 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4499 EXPORT_SYMBOL(netdev_upper_dev_unlink);
4501 static void dev_change_rx_flags(struct net_device *dev, int flags)
4503 const struct net_device_ops *ops = dev->netdev_ops;
4505 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4506 ops->ndo_change_rx_flags(dev, flags);
4509 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4511 unsigned int old_flags = dev->flags;
4517 dev->flags |= IFF_PROMISC;
4518 dev->promiscuity += inc;
4519 if (dev->promiscuity == 0) {
4522 * If inc causes overflow, untouch promisc and return error.
4525 dev->flags &= ~IFF_PROMISC;
4527 dev->promiscuity -= inc;
4528 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4533 if (dev->flags != old_flags) {
4534 pr_info("device %s %s promiscuous mode\n",
4536 dev->flags & IFF_PROMISC ? "entered" : "left");
4537 if (audit_enabled) {
4538 current_uid_gid(&uid, &gid);
4539 audit_log(current->audit_context, GFP_ATOMIC,
4540 AUDIT_ANOM_PROMISCUOUS,
4541 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4542 dev->name, (dev->flags & IFF_PROMISC),
4543 (old_flags & IFF_PROMISC),
4544 from_kuid(&init_user_ns, audit_get_loginuid(current)),
4545 from_kuid(&init_user_ns, uid),
4546 from_kgid(&init_user_ns, gid),
4547 audit_get_sessionid(current));
4550 dev_change_rx_flags(dev, IFF_PROMISC);
4556 * dev_set_promiscuity - update promiscuity count on a device
4560 * Add or remove promiscuity from a device. While the count in the device
4561 * remains above zero the interface remains promiscuous. Once it hits zero
4562 * the device reverts back to normal filtering operation. A negative inc
4563 * value is used to drop promiscuity on the device.
4564 * Return 0 if successful or a negative errno code on error.
4566 int dev_set_promiscuity(struct net_device *dev, int inc)
4568 unsigned int old_flags = dev->flags;
4571 err = __dev_set_promiscuity(dev, inc);
4574 if (dev->flags != old_flags)
4575 dev_set_rx_mode(dev);
4578 EXPORT_SYMBOL(dev_set_promiscuity);
4581 * dev_set_allmulti - update allmulti count on a device
4585 * Add or remove reception of all multicast frames to a device. While the
4586 * count in the device remains above zero the interface remains listening
4587 * to all interfaces. Once it hits zero the device reverts back to normal
4588 * filtering operation. A negative @inc value is used to drop the counter
4589 * when releasing a resource needing all multicasts.
4590 * Return 0 if successful or a negative errno code on error.
4593 int dev_set_allmulti(struct net_device *dev, int inc)
4595 unsigned int old_flags = dev->flags;
4599 dev->flags |= IFF_ALLMULTI;
4600 dev->allmulti += inc;
4601 if (dev->allmulti == 0) {
4604 * If inc causes overflow, untouch allmulti and return error.
4607 dev->flags &= ~IFF_ALLMULTI;
4609 dev->allmulti -= inc;
4610 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4615 if (dev->flags ^ old_flags) {
4616 dev_change_rx_flags(dev, IFF_ALLMULTI);
4617 dev_set_rx_mode(dev);
4621 EXPORT_SYMBOL(dev_set_allmulti);
4624 * Upload unicast and multicast address lists to device and
4625 * configure RX filtering. When the device doesn't support unicast
4626 * filtering it is put in promiscuous mode while unicast addresses
4629 void __dev_set_rx_mode(struct net_device *dev)
4631 const struct net_device_ops *ops = dev->netdev_ops;
4633 /* dev_open will call this function so the list will stay sane. */
4634 if (!(dev->flags&IFF_UP))
4637 if (!netif_device_present(dev))
4640 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4641 /* Unicast addresses changes may only happen under the rtnl,
4642 * therefore calling __dev_set_promiscuity here is safe.
4644 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4645 __dev_set_promiscuity(dev, 1);
4646 dev->uc_promisc = true;
4647 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4648 __dev_set_promiscuity(dev, -1);
4649 dev->uc_promisc = false;
4653 if (ops->ndo_set_rx_mode)
4654 ops->ndo_set_rx_mode(dev);
4657 void dev_set_rx_mode(struct net_device *dev)
4659 netif_addr_lock_bh(dev);
4660 __dev_set_rx_mode(dev);
4661 netif_addr_unlock_bh(dev);
4665 * dev_get_flags - get flags reported to userspace
4668 * Get the combination of flag bits exported through APIs to userspace.
4670 unsigned int dev_get_flags(const struct net_device *dev)
4674 flags = (dev->flags & ~(IFF_PROMISC |
4679 (dev->gflags & (IFF_PROMISC |
4682 if (netif_running(dev)) {
4683 if (netif_oper_up(dev))
4684 flags |= IFF_RUNNING;
4685 if (netif_carrier_ok(dev))
4686 flags |= IFF_LOWER_UP;
4687 if (netif_dormant(dev))
4688 flags |= IFF_DORMANT;
4693 EXPORT_SYMBOL(dev_get_flags);
4695 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4697 unsigned int old_flags = dev->flags;
4703 * Set the flags on our device.
4706 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4707 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4709 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4713 * Load in the correct multicast list now the flags have changed.
4716 if ((old_flags ^ flags) & IFF_MULTICAST)
4717 dev_change_rx_flags(dev, IFF_MULTICAST);
4719 dev_set_rx_mode(dev);
4722 * Have we downed the interface. We handle IFF_UP ourselves
4723 * according to user attempts to set it, rather than blindly
4728 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4729 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4732 dev_set_rx_mode(dev);
4735 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4736 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4738 dev->gflags ^= IFF_PROMISC;
4739 dev_set_promiscuity(dev, inc);
4742 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4743 is important. Some (broken) drivers set IFF_PROMISC, when
4744 IFF_ALLMULTI is requested not asking us and not reporting.
4746 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4747 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4749 dev->gflags ^= IFF_ALLMULTI;
4750 dev_set_allmulti(dev, inc);
4756 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4758 unsigned int changes = dev->flags ^ old_flags;
4760 if (changes & IFF_UP) {
4761 if (dev->flags & IFF_UP)
4762 call_netdevice_notifiers(NETDEV_UP, dev);
4764 call_netdevice_notifiers(NETDEV_DOWN, dev);
4767 if (dev->flags & IFF_UP &&
4768 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
4769 struct netdev_notifier_change_info change_info;
4771 change_info.flags_changed = changes;
4772 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
4778 * dev_change_flags - change device settings
4780 * @flags: device state flags
4782 * Change settings on device based state flags. The flags are
4783 * in the userspace exported format.
4785 int dev_change_flags(struct net_device *dev, unsigned int flags)
4788 unsigned int changes, old_flags = dev->flags;
4790 ret = __dev_change_flags(dev, flags);
4794 changes = old_flags ^ dev->flags;
4796 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4798 __dev_notify_flags(dev, old_flags);
4801 EXPORT_SYMBOL(dev_change_flags);
4804 * dev_set_mtu - Change maximum transfer unit
4806 * @new_mtu: new transfer unit
4808 * Change the maximum transfer size of the network device.
4810 int dev_set_mtu(struct net_device *dev, int new_mtu)
4812 const struct net_device_ops *ops = dev->netdev_ops;
4815 if (new_mtu == dev->mtu)
4818 /* MTU must be positive. */
4822 if (!netif_device_present(dev))
4826 if (ops->ndo_change_mtu)
4827 err = ops->ndo_change_mtu(dev, new_mtu);
4832 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4835 EXPORT_SYMBOL(dev_set_mtu);
4838 * dev_set_group - Change group this device belongs to
4840 * @new_group: group this device should belong to
4842 void dev_set_group(struct net_device *dev, int new_group)
4844 dev->group = new_group;
4846 EXPORT_SYMBOL(dev_set_group);
4849 * dev_set_mac_address - Change Media Access Control Address
4853 * Change the hardware (MAC) address of the device
4855 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4857 const struct net_device_ops *ops = dev->netdev_ops;
4860 if (!ops->ndo_set_mac_address)
4862 if (sa->sa_family != dev->type)
4864 if (!netif_device_present(dev))
4866 err = ops->ndo_set_mac_address(dev, sa);
4869 dev->addr_assign_type = NET_ADDR_SET;
4870 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4871 add_device_randomness(dev->dev_addr, dev->addr_len);
4874 EXPORT_SYMBOL(dev_set_mac_address);
4877 * dev_change_carrier - Change device carrier
4879 * @new_carrier: new value
4881 * Change device carrier
4883 int dev_change_carrier(struct net_device *dev, bool new_carrier)
4885 const struct net_device_ops *ops = dev->netdev_ops;
4887 if (!ops->ndo_change_carrier)
4889 if (!netif_device_present(dev))
4891 return ops->ndo_change_carrier(dev, new_carrier);
4893 EXPORT_SYMBOL(dev_change_carrier);
4896 * dev_new_index - allocate an ifindex
4897 * @net: the applicable net namespace
4899 * Returns a suitable unique value for a new device interface
4900 * number. The caller must hold the rtnl semaphore or the
4901 * dev_base_lock to be sure it remains unique.
4903 static int dev_new_index(struct net *net)
4905 int ifindex = net->ifindex;
4909 if (!__dev_get_by_index(net, ifindex))
4910 return net->ifindex = ifindex;
4914 /* Delayed registration/unregisteration */
4915 static LIST_HEAD(net_todo_list);
4917 static void net_set_todo(struct net_device *dev)
4919 list_add_tail(&dev->todo_list, &net_todo_list);
4922 static void rollback_registered_many(struct list_head *head)
4924 struct net_device *dev, *tmp;
4926 BUG_ON(dev_boot_phase);
4929 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4930 /* Some devices call without registering
4931 * for initialization unwind. Remove those
4932 * devices and proceed with the remaining.
4934 if (dev->reg_state == NETREG_UNINITIALIZED) {
4935 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
4939 list_del(&dev->unreg_list);
4942 dev->dismantle = true;
4943 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4946 /* If device is running, close it first. */
4947 dev_close_many(head);
4949 list_for_each_entry(dev, head, unreg_list) {
4950 /* And unlink it from device chain. */
4951 unlist_netdevice(dev);
4953 dev->reg_state = NETREG_UNREGISTERING;
4958 list_for_each_entry(dev, head, unreg_list) {
4959 /* Shutdown queueing discipline. */
4963 /* Notify protocols, that we are about to destroy
4964 this device. They should clean all the things.
4966 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4968 if (!dev->rtnl_link_ops ||
4969 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4970 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4973 * Flush the unicast and multicast chains
4978 if (dev->netdev_ops->ndo_uninit)
4979 dev->netdev_ops->ndo_uninit(dev);
4981 /* Notifier chain MUST detach us all upper devices. */
4982 WARN_ON(netdev_has_any_upper_dev(dev));
4984 /* Remove entries from kobject tree */
4985 netdev_unregister_kobject(dev);
4987 /* Remove XPS queueing entries */
4988 netif_reset_xps_queues_gt(dev, 0);
4994 list_for_each_entry(dev, head, unreg_list)
4998 static void rollback_registered(struct net_device *dev)
5002 list_add(&dev->unreg_list, &single);
5003 rollback_registered_many(&single);
5007 static netdev_features_t netdev_fix_features(struct net_device *dev,
5008 netdev_features_t features)
5010 /* Fix illegal checksum combinations */
5011 if ((features & NETIF_F_HW_CSUM) &&
5012 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5013 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5014 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5017 /* TSO requires that SG is present as well. */
5018 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5019 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5020 features &= ~NETIF_F_ALL_TSO;
5023 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5024 !(features & NETIF_F_IP_CSUM)) {
5025 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5026 features &= ~NETIF_F_TSO;
5027 features &= ~NETIF_F_TSO_ECN;
5030 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5031 !(features & NETIF_F_IPV6_CSUM)) {
5032 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5033 features &= ~NETIF_F_TSO6;
5036 /* TSO ECN requires that TSO is present as well. */
5037 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5038 features &= ~NETIF_F_TSO_ECN;
5040 /* Software GSO depends on SG. */
5041 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5042 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5043 features &= ~NETIF_F_GSO;
5046 /* UFO needs SG and checksumming */
5047 if (features & NETIF_F_UFO) {
5048 /* maybe split UFO into V4 and V6? */
5049 if (!((features & NETIF_F_GEN_CSUM) ||
5050 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5051 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5053 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5054 features &= ~NETIF_F_UFO;
5057 if (!(features & NETIF_F_SG)) {
5059 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5060 features &= ~NETIF_F_UFO;
5067 int __netdev_update_features(struct net_device *dev)
5069 netdev_features_t features;
5074 features = netdev_get_wanted_features(dev);
5076 if (dev->netdev_ops->ndo_fix_features)
5077 features = dev->netdev_ops->ndo_fix_features(dev, features);
5079 /* driver might be less strict about feature dependencies */
5080 features = netdev_fix_features(dev, features);
5082 if (dev->features == features)
5085 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5086 &dev->features, &features);
5088 if (dev->netdev_ops->ndo_set_features)
5089 err = dev->netdev_ops->ndo_set_features(dev, features);
5091 if (unlikely(err < 0)) {
5093 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5094 err, &features, &dev->features);
5099 dev->features = features;
5105 * netdev_update_features - recalculate device features
5106 * @dev: the device to check
5108 * Recalculate dev->features set and send notifications if it
5109 * has changed. Should be called after driver or hardware dependent
5110 * conditions might have changed that influence the features.
5112 void netdev_update_features(struct net_device *dev)
5114 if (__netdev_update_features(dev))
5115 netdev_features_change(dev);
5117 EXPORT_SYMBOL(netdev_update_features);
5120 * netdev_change_features - recalculate device features
5121 * @dev: the device to check
5123 * Recalculate dev->features set and send notifications even
5124 * if they have not changed. Should be called instead of
5125 * netdev_update_features() if also dev->vlan_features might
5126 * have changed to allow the changes to be propagated to stacked
5129 void netdev_change_features(struct net_device *dev)
5131 __netdev_update_features(dev);
5132 netdev_features_change(dev);
5134 EXPORT_SYMBOL(netdev_change_features);
5137 * netif_stacked_transfer_operstate - transfer operstate
5138 * @rootdev: the root or lower level device to transfer state from
5139 * @dev: the device to transfer operstate to
5141 * Transfer operational state from root to device. This is normally
5142 * called when a stacking relationship exists between the root
5143 * device and the device(a leaf device).
5145 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5146 struct net_device *dev)
5148 if (rootdev->operstate == IF_OPER_DORMANT)
5149 netif_dormant_on(dev);
5151 netif_dormant_off(dev);
5153 if (netif_carrier_ok(rootdev)) {
5154 if (!netif_carrier_ok(dev))
5155 netif_carrier_on(dev);
5157 if (netif_carrier_ok(dev))
5158 netif_carrier_off(dev);
5161 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5164 static int netif_alloc_rx_queues(struct net_device *dev)
5166 unsigned int i, count = dev->num_rx_queues;
5167 struct netdev_rx_queue *rx;
5171 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5177 for (i = 0; i < count; i++)
5183 static void netdev_init_one_queue(struct net_device *dev,
5184 struct netdev_queue *queue, void *_unused)
5186 /* Initialize queue lock */
5187 spin_lock_init(&queue->_xmit_lock);
5188 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5189 queue->xmit_lock_owner = -1;
5190 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5193 dql_init(&queue->dql, HZ);
5197 static int netif_alloc_netdev_queues(struct net_device *dev)
5199 unsigned int count = dev->num_tx_queues;
5200 struct netdev_queue *tx;
5204 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5210 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5211 spin_lock_init(&dev->tx_global_lock);
5217 * register_netdevice - register a network device
5218 * @dev: device to register
5220 * Take a completed network device structure and add it to the kernel
5221 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5222 * chain. 0 is returned on success. A negative errno code is returned
5223 * on a failure to set up the device, or if the name is a duplicate.
5225 * Callers must hold the rtnl semaphore. You may want
5226 * register_netdev() instead of this.
5229 * The locking appears insufficient to guarantee two parallel registers
5230 * will not get the same name.
5233 int register_netdevice(struct net_device *dev)
5236 struct net *net = dev_net(dev);
5238 BUG_ON(dev_boot_phase);
5243 /* When net_device's are persistent, this will be fatal. */
5244 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5247 spin_lock_init(&dev->addr_list_lock);
5248 netdev_set_addr_lockdep_class(dev);
5252 ret = dev_get_valid_name(net, dev, dev->name);
5256 /* Init, if this function is available */
5257 if (dev->netdev_ops->ndo_init) {
5258 ret = dev->netdev_ops->ndo_init(dev);
5266 if (((dev->hw_features | dev->features) &
5267 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5268 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5269 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5270 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5277 dev->ifindex = dev_new_index(net);
5278 else if (__dev_get_by_index(net, dev->ifindex))
5281 if (dev->iflink == -1)
5282 dev->iflink = dev->ifindex;
5284 /* Transfer changeable features to wanted_features and enable
5285 * software offloads (GSO and GRO).
5287 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5288 dev->features |= NETIF_F_SOFT_FEATURES;
5289 dev->wanted_features = dev->features & dev->hw_features;
5291 /* Turn on no cache copy if HW is doing checksum */
5292 if (!(dev->flags & IFF_LOOPBACK)) {
5293 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5294 if (dev->features & NETIF_F_ALL_CSUM) {
5295 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5296 dev->features |= NETIF_F_NOCACHE_COPY;
5300 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5302 dev->vlan_features |= NETIF_F_HIGHDMA;
5304 /* Make NETIF_F_SG inheritable to tunnel devices.
5306 dev->hw_enc_features |= NETIF_F_SG;
5308 /* Make NETIF_F_SG inheritable to MPLS.
5310 dev->mpls_features |= NETIF_F_SG;
5312 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5313 ret = notifier_to_errno(ret);
5317 ret = netdev_register_kobject(dev);
5320 dev->reg_state = NETREG_REGISTERED;
5322 __netdev_update_features(dev);
5325 * Default initial state at registry is that the
5326 * device is present.
5329 set_bit(__LINK_STATE_PRESENT, &dev->state);
5331 linkwatch_init_dev(dev);
5333 dev_init_scheduler(dev);
5335 list_netdevice(dev);
5336 add_device_randomness(dev->dev_addr, dev->addr_len);
5338 /* If the device has permanent device address, driver should
5339 * set dev_addr and also addr_assign_type should be set to
5340 * NET_ADDR_PERM (default value).
5342 if (dev->addr_assign_type == NET_ADDR_PERM)
5343 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5345 /* Notify protocols, that a new device appeared. */
5346 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5347 ret = notifier_to_errno(ret);
5349 rollback_registered(dev);
5350 dev->reg_state = NETREG_UNREGISTERED;
5353 * Prevent userspace races by waiting until the network
5354 * device is fully setup before sending notifications.
5356 if (!dev->rtnl_link_ops ||
5357 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5358 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5364 if (dev->netdev_ops->ndo_uninit)
5365 dev->netdev_ops->ndo_uninit(dev);
5368 EXPORT_SYMBOL(register_netdevice);
5371 * init_dummy_netdev - init a dummy network device for NAPI
5372 * @dev: device to init
5374 * This takes a network device structure and initialize the minimum
5375 * amount of fields so it can be used to schedule NAPI polls without
5376 * registering a full blown interface. This is to be used by drivers
5377 * that need to tie several hardware interfaces to a single NAPI
5378 * poll scheduler due to HW limitations.
5380 int init_dummy_netdev(struct net_device *dev)
5382 /* Clear everything. Note we don't initialize spinlocks
5383 * are they aren't supposed to be taken by any of the
5384 * NAPI code and this dummy netdev is supposed to be
5385 * only ever used for NAPI polls
5387 memset(dev, 0, sizeof(struct net_device));
5389 /* make sure we BUG if trying to hit standard
5390 * register/unregister code path
5392 dev->reg_state = NETREG_DUMMY;
5394 /* NAPI wants this */
5395 INIT_LIST_HEAD(&dev->napi_list);
5397 /* a dummy interface is started by default */
5398 set_bit(__LINK_STATE_PRESENT, &dev->state);
5399 set_bit(__LINK_STATE_START, &dev->state);
5401 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5402 * because users of this 'device' dont need to change
5408 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5412 * register_netdev - register a network device
5413 * @dev: device to register
5415 * Take a completed network device structure and add it to the kernel
5416 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5417 * chain. 0 is returned on success. A negative errno code is returned
5418 * on a failure to set up the device, or if the name is a duplicate.
5420 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5421 * and expands the device name if you passed a format string to
5424 int register_netdev(struct net_device *dev)
5429 err = register_netdevice(dev);
5433 EXPORT_SYMBOL(register_netdev);
5435 int netdev_refcnt_read(const struct net_device *dev)
5439 for_each_possible_cpu(i)
5440 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5443 EXPORT_SYMBOL(netdev_refcnt_read);
5446 * netdev_wait_allrefs - wait until all references are gone.
5447 * @dev: target net_device
5449 * This is called when unregistering network devices.
5451 * Any protocol or device that holds a reference should register
5452 * for netdevice notification, and cleanup and put back the
5453 * reference if they receive an UNREGISTER event.
5454 * We can get stuck here if buggy protocols don't correctly
5457 static void netdev_wait_allrefs(struct net_device *dev)
5459 unsigned long rebroadcast_time, warning_time;
5462 linkwatch_forget_dev(dev);
5464 rebroadcast_time = warning_time = jiffies;
5465 refcnt = netdev_refcnt_read(dev);
5467 while (refcnt != 0) {
5468 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5471 /* Rebroadcast unregister notification */
5472 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5478 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5479 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5481 /* We must not have linkwatch events
5482 * pending on unregister. If this
5483 * happens, we simply run the queue
5484 * unscheduled, resulting in a noop
5487 linkwatch_run_queue();
5492 rebroadcast_time = jiffies;
5497 refcnt = netdev_refcnt_read(dev);
5499 if (time_after(jiffies, warning_time + 10 * HZ)) {
5500 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5502 warning_time = jiffies;
5511 * register_netdevice(x1);
5512 * register_netdevice(x2);
5514 * unregister_netdevice(y1);
5515 * unregister_netdevice(y2);
5521 * We are invoked by rtnl_unlock().
5522 * This allows us to deal with problems:
5523 * 1) We can delete sysfs objects which invoke hotplug
5524 * without deadlocking with linkwatch via keventd.
5525 * 2) Since we run with the RTNL semaphore not held, we can sleep
5526 * safely in order to wait for the netdev refcnt to drop to zero.
5528 * We must not return until all unregister events added during
5529 * the interval the lock was held have been completed.
5531 void netdev_run_todo(void)
5533 struct list_head list;
5535 /* Snapshot list, allow later requests */
5536 list_replace_init(&net_todo_list, &list);
5541 /* Wait for rcu callbacks to finish before next phase */
5542 if (!list_empty(&list))
5545 while (!list_empty(&list)) {
5546 struct net_device *dev
5547 = list_first_entry(&list, struct net_device, todo_list);
5548 list_del(&dev->todo_list);
5551 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5554 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5555 pr_err("network todo '%s' but state %d\n",
5556 dev->name, dev->reg_state);
5561 dev->reg_state = NETREG_UNREGISTERED;
5563 on_each_cpu(flush_backlog, dev, 1);
5565 netdev_wait_allrefs(dev);
5568 BUG_ON(netdev_refcnt_read(dev));
5569 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5570 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5571 WARN_ON(dev->dn_ptr);
5573 if (dev->destructor)
5574 dev->destructor(dev);
5576 /* Free network device */
5577 kobject_put(&dev->dev.kobj);
5581 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5582 * fields in the same order, with only the type differing.
5584 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5585 const struct net_device_stats *netdev_stats)
5587 #if BITS_PER_LONG == 64
5588 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5589 memcpy(stats64, netdev_stats, sizeof(*stats64));
5591 size_t i, n = sizeof(*stats64) / sizeof(u64);
5592 const unsigned long *src = (const unsigned long *)netdev_stats;
5593 u64 *dst = (u64 *)stats64;
5595 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5596 sizeof(*stats64) / sizeof(u64));
5597 for (i = 0; i < n; i++)
5601 EXPORT_SYMBOL(netdev_stats_to_stats64);
5604 * dev_get_stats - get network device statistics
5605 * @dev: device to get statistics from
5606 * @storage: place to store stats
5608 * Get network statistics from device. Return @storage.
5609 * The device driver may provide its own method by setting
5610 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5611 * otherwise the internal statistics structure is used.
5613 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5614 struct rtnl_link_stats64 *storage)
5616 const struct net_device_ops *ops = dev->netdev_ops;
5618 if (ops->ndo_get_stats64) {
5619 memset(storage, 0, sizeof(*storage));
5620 ops->ndo_get_stats64(dev, storage);
5621 } else if (ops->ndo_get_stats) {
5622 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5624 netdev_stats_to_stats64(storage, &dev->stats);
5626 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5629 EXPORT_SYMBOL(dev_get_stats);
5631 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5633 struct netdev_queue *queue = dev_ingress_queue(dev);
5635 #ifdef CONFIG_NET_CLS_ACT
5638 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5641 netdev_init_one_queue(dev, queue, NULL);
5642 queue->qdisc = &noop_qdisc;
5643 queue->qdisc_sleeping = &noop_qdisc;
5644 rcu_assign_pointer(dev->ingress_queue, queue);
5649 static const struct ethtool_ops default_ethtool_ops;
5651 void netdev_set_default_ethtool_ops(struct net_device *dev,
5652 const struct ethtool_ops *ops)
5654 if (dev->ethtool_ops == &default_ethtool_ops)
5655 dev->ethtool_ops = ops;
5657 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
5660 * alloc_netdev_mqs - allocate network device
5661 * @sizeof_priv: size of private data to allocate space for
5662 * @name: device name format string
5663 * @setup: callback to initialize device
5664 * @txqs: the number of TX subqueues to allocate
5665 * @rxqs: the number of RX subqueues to allocate
5667 * Allocates a struct net_device with private data area for driver use
5668 * and performs basic initialization. Also allocates subquue structs
5669 * for each queue on the device.
5671 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5672 void (*setup)(struct net_device *),
5673 unsigned int txqs, unsigned int rxqs)
5675 struct net_device *dev;
5677 struct net_device *p;
5679 BUG_ON(strlen(name) >= sizeof(dev->name));
5682 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5688 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5693 alloc_size = sizeof(struct net_device);
5695 /* ensure 32-byte alignment of private area */
5696 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5697 alloc_size += sizeof_priv;
5699 /* ensure 32-byte alignment of whole construct */
5700 alloc_size += NETDEV_ALIGN - 1;
5702 p = kzalloc(alloc_size, GFP_KERNEL);
5706 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5707 dev->padded = (char *)dev - (char *)p;
5709 dev->pcpu_refcnt = alloc_percpu(int);
5710 if (!dev->pcpu_refcnt)
5713 if (dev_addr_init(dev))
5719 dev_net_set(dev, &init_net);
5721 dev->gso_max_size = GSO_MAX_SIZE;
5722 dev->gso_max_segs = GSO_MAX_SEGS;
5724 INIT_LIST_HEAD(&dev->napi_list);
5725 INIT_LIST_HEAD(&dev->unreg_list);
5726 INIT_LIST_HEAD(&dev->link_watch_list);
5727 INIT_LIST_HEAD(&dev->upper_dev_list);
5728 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5731 dev->num_tx_queues = txqs;
5732 dev->real_num_tx_queues = txqs;
5733 if (netif_alloc_netdev_queues(dev))
5737 dev->num_rx_queues = rxqs;
5738 dev->real_num_rx_queues = rxqs;
5739 if (netif_alloc_rx_queues(dev))
5743 strcpy(dev->name, name);
5744 dev->group = INIT_NETDEV_GROUP;
5745 if (!dev->ethtool_ops)
5746 dev->ethtool_ops = &default_ethtool_ops;
5754 free_percpu(dev->pcpu_refcnt);
5764 EXPORT_SYMBOL(alloc_netdev_mqs);
5767 * free_netdev - free network device
5770 * This function does the last stage of destroying an allocated device
5771 * interface. The reference to the device object is released.
5772 * If this is the last reference then it will be freed.
5774 void free_netdev(struct net_device *dev)
5776 struct napi_struct *p, *n;
5778 release_net(dev_net(dev));
5785 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
5787 /* Flush device addresses */
5788 dev_addr_flush(dev);
5790 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5793 free_percpu(dev->pcpu_refcnt);
5794 dev->pcpu_refcnt = NULL;
5796 /* Compatibility with error handling in drivers */
5797 if (dev->reg_state == NETREG_UNINITIALIZED) {
5798 kfree((char *)dev - dev->padded);
5802 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5803 dev->reg_state = NETREG_RELEASED;
5805 /* will free via device release */
5806 put_device(&dev->dev);
5808 EXPORT_SYMBOL(free_netdev);
5811 * synchronize_net - Synchronize with packet receive processing
5813 * Wait for packets currently being received to be done.
5814 * Does not block later packets from starting.
5816 void synchronize_net(void)
5819 if (rtnl_is_locked())
5820 synchronize_rcu_expedited();
5824 EXPORT_SYMBOL(synchronize_net);
5827 * unregister_netdevice_queue - remove device from the kernel
5831 * This function shuts down a device interface and removes it
5832 * from the kernel tables.
5833 * If head not NULL, device is queued to be unregistered later.
5835 * Callers must hold the rtnl semaphore. You may want
5836 * unregister_netdev() instead of this.
5839 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5844 list_move_tail(&dev->unreg_list, head);
5846 rollback_registered(dev);
5847 /* Finish processing unregister after unlock */
5851 EXPORT_SYMBOL(unregister_netdevice_queue);
5854 * unregister_netdevice_many - unregister many devices
5855 * @head: list of devices
5857 void unregister_netdevice_many(struct list_head *head)
5859 struct net_device *dev;
5861 if (!list_empty(head)) {
5862 rollback_registered_many(head);
5863 list_for_each_entry(dev, head, unreg_list)
5867 EXPORT_SYMBOL(unregister_netdevice_many);
5870 * unregister_netdev - remove device from the kernel
5873 * This function shuts down a device interface and removes it
5874 * from the kernel tables.
5876 * This is just a wrapper for unregister_netdevice that takes
5877 * the rtnl semaphore. In general you want to use this and not
5878 * unregister_netdevice.
5880 void unregister_netdev(struct net_device *dev)
5883 unregister_netdevice(dev);
5886 EXPORT_SYMBOL(unregister_netdev);
5889 * dev_change_net_namespace - move device to different nethost namespace
5891 * @net: network namespace
5892 * @pat: If not NULL name pattern to try if the current device name
5893 * is already taken in the destination network namespace.
5895 * This function shuts down a device interface and moves it
5896 * to a new network namespace. On success 0 is returned, on
5897 * a failure a netagive errno code is returned.
5899 * Callers must hold the rtnl semaphore.
5902 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5908 /* Don't allow namespace local devices to be moved. */
5910 if (dev->features & NETIF_F_NETNS_LOCAL)
5913 /* Ensure the device has been registrered */
5914 if (dev->reg_state != NETREG_REGISTERED)
5917 /* Get out if there is nothing todo */
5919 if (net_eq(dev_net(dev), net))
5922 /* Pick the destination device name, and ensure
5923 * we can use it in the destination network namespace.
5926 if (__dev_get_by_name(net, dev->name)) {
5927 /* We get here if we can't use the current device name */
5930 if (dev_get_valid_name(net, dev, pat) < 0)
5935 * And now a mini version of register_netdevice unregister_netdevice.
5938 /* If device is running close it first. */
5941 /* And unlink it from device chain */
5943 unlist_netdevice(dev);
5947 /* Shutdown queueing discipline. */
5950 /* Notify protocols, that we are about to destroy
5951 this device. They should clean all the things.
5953 Note that dev->reg_state stays at NETREG_REGISTERED.
5954 This is wanted because this way 8021q and macvlan know
5955 the device is just moving and can keep their slaves up.
5957 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5959 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5960 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5963 * Flush the unicast and multicast chains
5968 /* Send a netdev-removed uevent to the old namespace */
5969 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
5971 /* Actually switch the network namespace */
5972 dev_net_set(dev, net);
5974 /* If there is an ifindex conflict assign a new one */
5975 if (__dev_get_by_index(net, dev->ifindex)) {
5976 int iflink = (dev->iflink == dev->ifindex);
5977 dev->ifindex = dev_new_index(net);
5979 dev->iflink = dev->ifindex;
5982 /* Send a netdev-add uevent to the new namespace */
5983 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
5985 /* Fixup kobjects */
5986 err = device_rename(&dev->dev, dev->name);
5989 /* Add the device back in the hashes */
5990 list_netdevice(dev);
5992 /* Notify protocols, that a new device appeared. */
5993 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5996 * Prevent userspace races by waiting until the network
5997 * device is fully setup before sending notifications.
5999 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6006 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6008 static int dev_cpu_callback(struct notifier_block *nfb,
6009 unsigned long action,
6012 struct sk_buff **list_skb;
6013 struct sk_buff *skb;
6014 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6015 struct softnet_data *sd, *oldsd;
6017 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6020 local_irq_disable();
6021 cpu = smp_processor_id();
6022 sd = &per_cpu(softnet_data, cpu);
6023 oldsd = &per_cpu(softnet_data, oldcpu);
6025 /* Find end of our completion_queue. */
6026 list_skb = &sd->completion_queue;
6028 list_skb = &(*list_skb)->next;
6029 /* Append completion queue from offline CPU. */
6030 *list_skb = oldsd->completion_queue;
6031 oldsd->completion_queue = NULL;
6033 /* Append output queue from offline CPU. */
6034 if (oldsd->output_queue) {
6035 *sd->output_queue_tailp = oldsd->output_queue;
6036 sd->output_queue_tailp = oldsd->output_queue_tailp;
6037 oldsd->output_queue = NULL;
6038 oldsd->output_queue_tailp = &oldsd->output_queue;
6040 /* Append NAPI poll list from offline CPU. */
6041 if (!list_empty(&oldsd->poll_list)) {
6042 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6043 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6046 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6049 /* Process offline CPU's input_pkt_queue */
6050 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6052 input_queue_head_incr(oldsd);
6054 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6056 input_queue_head_incr(oldsd);
6064 * netdev_increment_features - increment feature set by one
6065 * @all: current feature set
6066 * @one: new feature set
6067 * @mask: mask feature set
6069 * Computes a new feature set after adding a device with feature set
6070 * @one to the master device with current feature set @all. Will not
6071 * enable anything that is off in @mask. Returns the new feature set.
6073 netdev_features_t netdev_increment_features(netdev_features_t all,
6074 netdev_features_t one, netdev_features_t mask)
6076 if (mask & NETIF_F_GEN_CSUM)
6077 mask |= NETIF_F_ALL_CSUM;
6078 mask |= NETIF_F_VLAN_CHALLENGED;
6080 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6081 all &= one | ~NETIF_F_ALL_FOR_ALL;
6083 /* If one device supports hw checksumming, set for all. */
6084 if (all & NETIF_F_GEN_CSUM)
6085 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6089 EXPORT_SYMBOL(netdev_increment_features);
6091 static struct hlist_head * __net_init netdev_create_hash(void)
6094 struct hlist_head *hash;
6096 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6098 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6099 INIT_HLIST_HEAD(&hash[i]);
6104 /* Initialize per network namespace state */
6105 static int __net_init netdev_init(struct net *net)
6107 if (net != &init_net)
6108 INIT_LIST_HEAD(&net->dev_base_head);
6110 net->dev_name_head = netdev_create_hash();
6111 if (net->dev_name_head == NULL)
6114 net->dev_index_head = netdev_create_hash();
6115 if (net->dev_index_head == NULL)
6121 kfree(net->dev_name_head);
6127 * netdev_drivername - network driver for the device
6128 * @dev: network device
6130 * Determine network driver for device.
6132 const char *netdev_drivername(const struct net_device *dev)
6134 const struct device_driver *driver;
6135 const struct device *parent;
6136 const char *empty = "";
6138 parent = dev->dev.parent;
6142 driver = parent->driver;
6143 if (driver && driver->name)
6144 return driver->name;
6148 static int __netdev_printk(const char *level, const struct net_device *dev,
6149 struct va_format *vaf)
6153 if (dev && dev->dev.parent) {
6154 r = dev_printk_emit(level[1] - '0',
6157 dev_driver_string(dev->dev.parent),
6158 dev_name(dev->dev.parent),
6159 netdev_name(dev), vaf);
6161 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6163 r = printk("%s(NULL net_device): %pV", level, vaf);
6169 int netdev_printk(const char *level, const struct net_device *dev,
6170 const char *format, ...)
6172 struct va_format vaf;
6176 va_start(args, format);
6181 r = __netdev_printk(level, dev, &vaf);
6187 EXPORT_SYMBOL(netdev_printk);
6189 #define define_netdev_printk_level(func, level) \
6190 int func(const struct net_device *dev, const char *fmt, ...) \
6193 struct va_format vaf; \
6196 va_start(args, fmt); \
6201 r = __netdev_printk(level, dev, &vaf); \
6207 EXPORT_SYMBOL(func);
6209 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6210 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6211 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6212 define_netdev_printk_level(netdev_err, KERN_ERR);
6213 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6214 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6215 define_netdev_printk_level(netdev_info, KERN_INFO);
6217 static void __net_exit netdev_exit(struct net *net)
6219 kfree(net->dev_name_head);
6220 kfree(net->dev_index_head);
6223 static struct pernet_operations __net_initdata netdev_net_ops = {
6224 .init = netdev_init,
6225 .exit = netdev_exit,
6228 static void __net_exit default_device_exit(struct net *net)
6230 struct net_device *dev, *aux;
6232 * Push all migratable network devices back to the
6233 * initial network namespace
6236 for_each_netdev_safe(net, dev, aux) {
6238 char fb_name[IFNAMSIZ];
6240 /* Ignore unmoveable devices (i.e. loopback) */
6241 if (dev->features & NETIF_F_NETNS_LOCAL)
6244 /* Leave virtual devices for the generic cleanup */
6245 if (dev->rtnl_link_ops)
6248 /* Push remaining network devices to init_net */
6249 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6250 err = dev_change_net_namespace(dev, &init_net, fb_name);
6252 pr_emerg("%s: failed to move %s to init_net: %d\n",
6253 __func__, dev->name, err);
6260 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6262 /* At exit all network devices most be removed from a network
6263 * namespace. Do this in the reverse order of registration.
6264 * Do this across as many network namespaces as possible to
6265 * improve batching efficiency.
6267 struct net_device *dev;
6269 LIST_HEAD(dev_kill_list);
6272 list_for_each_entry(net, net_list, exit_list) {
6273 for_each_netdev_reverse(net, dev) {
6274 if (dev->rtnl_link_ops)
6275 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6277 unregister_netdevice_queue(dev, &dev_kill_list);
6280 unregister_netdevice_many(&dev_kill_list);
6281 list_del(&dev_kill_list);
6285 static struct pernet_operations __net_initdata default_device_ops = {
6286 .exit = default_device_exit,
6287 .exit_batch = default_device_exit_batch,
6291 * Initialize the DEV module. At boot time this walks the device list and
6292 * unhooks any devices that fail to initialise (normally hardware not
6293 * present) and leaves us with a valid list of present and active devices.
6298 * This is called single threaded during boot, so no need
6299 * to take the rtnl semaphore.
6301 static int __init net_dev_init(void)
6303 int i, rc = -ENOMEM;
6305 BUG_ON(!dev_boot_phase);
6307 if (dev_proc_init())
6310 if (netdev_kobject_init())
6313 INIT_LIST_HEAD(&ptype_all);
6314 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6315 INIT_LIST_HEAD(&ptype_base[i]);
6317 INIT_LIST_HEAD(&offload_base);
6319 if (register_pernet_subsys(&netdev_net_ops))
6323 * Initialise the packet receive queues.
6326 for_each_possible_cpu(i) {
6327 struct softnet_data *sd = &per_cpu(softnet_data, i);
6329 memset(sd, 0, sizeof(*sd));
6330 skb_queue_head_init(&sd->input_pkt_queue);
6331 skb_queue_head_init(&sd->process_queue);
6332 sd->completion_queue = NULL;
6333 INIT_LIST_HEAD(&sd->poll_list);
6334 sd->output_queue = NULL;
6335 sd->output_queue_tailp = &sd->output_queue;
6337 sd->csd.func = rps_trigger_softirq;
6343 sd->backlog.poll = process_backlog;
6344 sd->backlog.weight = weight_p;
6345 sd->backlog.gro_list = NULL;
6346 sd->backlog.gro_count = 0;
6348 #ifdef CONFIG_NET_FLOW_LIMIT
6349 sd->flow_limit = NULL;
6355 /* The loopback device is special if any other network devices
6356 * is present in a network namespace the loopback device must
6357 * be present. Since we now dynamically allocate and free the
6358 * loopback device ensure this invariant is maintained by
6359 * keeping the loopback device as the first device on the
6360 * list of network devices. Ensuring the loopback devices
6361 * is the first device that appears and the last network device
6364 if (register_pernet_device(&loopback_net_ops))
6367 if (register_pernet_device(&default_device_ops))
6370 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6371 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6373 hotcpu_notifier(dev_cpu_callback, 0);
6380 subsys_initcall(net_dev_init);