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 void netdev_notifier_info_init(struct netdev_notifier_info *info,
1395 struct net_device *dev)
1400 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1401 struct net_device *dev)
1403 struct netdev_notifier_info info;
1405 netdev_notifier_info_init(&info, dev);
1406 return nb->notifier_call(nb, val, &info);
1409 static int dev_boot_phase = 1;
1412 * register_netdevice_notifier - register a network notifier block
1415 * Register a notifier to be called when network device events occur.
1416 * The notifier passed is linked into the kernel structures and must
1417 * not be reused until it has been unregistered. A negative errno code
1418 * is returned on a failure.
1420 * When registered all registration and up events are replayed
1421 * to the new notifier to allow device to have a race free
1422 * view of the network device list.
1425 int register_netdevice_notifier(struct notifier_block *nb)
1427 struct net_device *dev;
1428 struct net_device *last;
1433 err = raw_notifier_chain_register(&netdev_chain, nb);
1439 for_each_netdev(net, dev) {
1440 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1441 err = notifier_to_errno(err);
1445 if (!(dev->flags & IFF_UP))
1448 call_netdevice_notifier(nb, NETDEV_UP, dev);
1459 for_each_netdev(net, dev) {
1463 if (dev->flags & IFF_UP) {
1464 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1466 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1468 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1473 raw_notifier_chain_unregister(&netdev_chain, nb);
1476 EXPORT_SYMBOL(register_netdevice_notifier);
1479 * unregister_netdevice_notifier - unregister a network notifier block
1482 * Unregister a notifier previously registered by
1483 * register_netdevice_notifier(). The notifier is unlinked into the
1484 * kernel structures and may then be reused. A negative errno code
1485 * is returned on a failure.
1487 * After unregistering unregister and down device events are synthesized
1488 * for all devices on the device list to the removed notifier to remove
1489 * the need for special case cleanup code.
1492 int unregister_netdevice_notifier(struct notifier_block *nb)
1494 struct net_device *dev;
1499 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1504 for_each_netdev(net, dev) {
1505 if (dev->flags & IFF_UP) {
1506 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1508 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1510 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1517 EXPORT_SYMBOL(unregister_netdevice_notifier);
1520 * call_netdevice_notifiers_info - call all network notifier blocks
1521 * @val: value passed unmodified to notifier function
1522 * @dev: net_device pointer passed unmodified to notifier function
1523 * @info: notifier information data
1525 * Call all network notifier blocks. Parameters and return value
1526 * are as for raw_notifier_call_chain().
1529 int call_netdevice_notifiers_info(unsigned long val, struct net_device *dev,
1530 struct netdev_notifier_info *info)
1533 netdev_notifier_info_init(info, dev);
1534 return raw_notifier_call_chain(&netdev_chain, val, info);
1536 EXPORT_SYMBOL(call_netdevice_notifiers_info);
1539 * call_netdevice_notifiers - call all network notifier blocks
1540 * @val: value passed unmodified to notifier function
1541 * @dev: net_device pointer passed unmodified to notifier function
1543 * Call all network notifier blocks. Parameters and return value
1544 * are as for raw_notifier_call_chain().
1547 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1549 struct netdev_notifier_info info;
1551 return call_netdevice_notifiers_info(val, dev, &info);
1553 EXPORT_SYMBOL(call_netdevice_notifiers);
1555 static struct static_key netstamp_needed __read_mostly;
1556 #ifdef HAVE_JUMP_LABEL
1557 /* We are not allowed to call static_key_slow_dec() from irq context
1558 * If net_disable_timestamp() is called from irq context, defer the
1559 * static_key_slow_dec() calls.
1561 static atomic_t netstamp_needed_deferred;
1564 void net_enable_timestamp(void)
1566 #ifdef HAVE_JUMP_LABEL
1567 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1571 static_key_slow_dec(&netstamp_needed);
1575 static_key_slow_inc(&netstamp_needed);
1577 EXPORT_SYMBOL(net_enable_timestamp);
1579 void net_disable_timestamp(void)
1581 #ifdef HAVE_JUMP_LABEL
1582 if (in_interrupt()) {
1583 atomic_inc(&netstamp_needed_deferred);
1587 static_key_slow_dec(&netstamp_needed);
1589 EXPORT_SYMBOL(net_disable_timestamp);
1591 static inline void net_timestamp_set(struct sk_buff *skb)
1593 skb->tstamp.tv64 = 0;
1594 if (static_key_false(&netstamp_needed))
1595 __net_timestamp(skb);
1598 #define net_timestamp_check(COND, SKB) \
1599 if (static_key_false(&netstamp_needed)) { \
1600 if ((COND) && !(SKB)->tstamp.tv64) \
1601 __net_timestamp(SKB); \
1604 static inline bool is_skb_forwardable(struct net_device *dev,
1605 struct sk_buff *skb)
1609 if (!(dev->flags & IFF_UP))
1612 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1613 if (skb->len <= len)
1616 /* if TSO is enabled, we don't care about the length as the packet
1617 * could be forwarded without being segmented before
1619 if (skb_is_gso(skb))
1626 * dev_forward_skb - loopback an skb to another netif
1628 * @dev: destination network device
1629 * @skb: buffer to forward
1632 * NET_RX_SUCCESS (no congestion)
1633 * NET_RX_DROP (packet was dropped, but freed)
1635 * dev_forward_skb can be used for injecting an skb from the
1636 * start_xmit function of one device into the receive queue
1637 * of another device.
1639 * The receiving device may be in another namespace, so
1640 * we have to clear all information in the skb that could
1641 * impact namespace isolation.
1643 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1645 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1646 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1647 atomic_long_inc(&dev->rx_dropped);
1655 if (unlikely(!is_skb_forwardable(dev, skb))) {
1656 atomic_long_inc(&dev->rx_dropped);
1662 skb->tstamp.tv64 = 0;
1663 skb->pkt_type = PACKET_HOST;
1664 skb->protocol = eth_type_trans(skb, dev);
1668 nf_reset_trace(skb);
1669 return netif_rx(skb);
1671 EXPORT_SYMBOL_GPL(dev_forward_skb);
1673 static inline int deliver_skb(struct sk_buff *skb,
1674 struct packet_type *pt_prev,
1675 struct net_device *orig_dev)
1677 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1679 atomic_inc(&skb->users);
1680 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1683 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1685 if (!ptype->af_packet_priv || !skb->sk)
1688 if (ptype->id_match)
1689 return ptype->id_match(ptype, skb->sk);
1690 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1697 * Support routine. Sends outgoing frames to any network
1698 * taps currently in use.
1701 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1703 struct packet_type *ptype;
1704 struct sk_buff *skb2 = NULL;
1705 struct packet_type *pt_prev = NULL;
1708 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1709 /* Never send packets back to the socket
1710 * they originated from - MvS (miquels@drinkel.ow.org)
1712 if ((ptype->dev == dev || !ptype->dev) &&
1713 (!skb_loop_sk(ptype, skb))) {
1715 deliver_skb(skb2, pt_prev, skb->dev);
1720 skb2 = skb_clone(skb, GFP_ATOMIC);
1724 net_timestamp_set(skb2);
1726 /* skb->nh should be correctly
1727 set by sender, so that the second statement is
1728 just protection against buggy protocols.
1730 skb_reset_mac_header(skb2);
1732 if (skb_network_header(skb2) < skb2->data ||
1733 skb2->network_header > skb2->tail) {
1734 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1735 ntohs(skb2->protocol),
1737 skb_reset_network_header(skb2);
1740 skb2->transport_header = skb2->network_header;
1741 skb2->pkt_type = PACKET_OUTGOING;
1746 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1751 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1752 * @dev: Network device
1753 * @txq: number of queues available
1755 * If real_num_tx_queues is changed the tc mappings may no longer be
1756 * valid. To resolve this verify the tc mapping remains valid and if
1757 * not NULL the mapping. With no priorities mapping to this
1758 * offset/count pair it will no longer be used. In the worst case TC0
1759 * is invalid nothing can be done so disable priority mappings. If is
1760 * expected that drivers will fix this mapping if they can before
1761 * calling netif_set_real_num_tx_queues.
1763 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1766 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1768 /* If TC0 is invalidated disable TC mapping */
1769 if (tc->offset + tc->count > txq) {
1770 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1775 /* Invalidated prio to tc mappings set to TC0 */
1776 for (i = 1; i < TC_BITMASK + 1; i++) {
1777 int q = netdev_get_prio_tc_map(dev, i);
1779 tc = &dev->tc_to_txq[q];
1780 if (tc->offset + tc->count > txq) {
1781 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1783 netdev_set_prio_tc_map(dev, i, 0);
1789 static DEFINE_MUTEX(xps_map_mutex);
1790 #define xmap_dereference(P) \
1791 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1793 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1796 struct xps_map *map = NULL;
1800 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1802 for (pos = 0; map && pos < map->len; pos++) {
1803 if (map->queues[pos] == index) {
1805 map->queues[pos] = map->queues[--map->len];
1807 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1808 kfree_rcu(map, rcu);
1818 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1820 struct xps_dev_maps *dev_maps;
1822 bool active = false;
1824 mutex_lock(&xps_map_mutex);
1825 dev_maps = xmap_dereference(dev->xps_maps);
1830 for_each_possible_cpu(cpu) {
1831 for (i = index; i < dev->num_tx_queues; i++) {
1832 if (!remove_xps_queue(dev_maps, cpu, i))
1835 if (i == dev->num_tx_queues)
1840 RCU_INIT_POINTER(dev->xps_maps, NULL);
1841 kfree_rcu(dev_maps, rcu);
1844 for (i = index; i < dev->num_tx_queues; i++)
1845 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1849 mutex_unlock(&xps_map_mutex);
1852 static struct xps_map *expand_xps_map(struct xps_map *map,
1855 struct xps_map *new_map;
1856 int alloc_len = XPS_MIN_MAP_ALLOC;
1859 for (pos = 0; map && pos < map->len; pos++) {
1860 if (map->queues[pos] != index)
1865 /* Need to add queue to this CPU's existing map */
1867 if (pos < map->alloc_len)
1870 alloc_len = map->alloc_len * 2;
1873 /* Need to allocate new map to store queue on this CPU's map */
1874 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1879 for (i = 0; i < pos; i++)
1880 new_map->queues[i] = map->queues[i];
1881 new_map->alloc_len = alloc_len;
1887 int netif_set_xps_queue(struct net_device *dev, struct cpumask *mask, u16 index)
1889 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1890 struct xps_map *map, *new_map;
1891 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1892 int cpu, numa_node_id = -2;
1893 bool active = false;
1895 mutex_lock(&xps_map_mutex);
1897 dev_maps = xmap_dereference(dev->xps_maps);
1899 /* allocate memory for queue storage */
1900 for_each_online_cpu(cpu) {
1901 if (!cpumask_test_cpu(cpu, mask))
1905 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1906 if (!new_dev_maps) {
1907 mutex_unlock(&xps_map_mutex);
1911 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1914 map = expand_xps_map(map, cpu, index);
1918 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1922 goto out_no_new_maps;
1924 for_each_possible_cpu(cpu) {
1925 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1926 /* add queue to CPU maps */
1929 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1930 while ((pos < map->len) && (map->queues[pos] != index))
1933 if (pos == map->len)
1934 map->queues[map->len++] = index;
1936 if (numa_node_id == -2)
1937 numa_node_id = cpu_to_node(cpu);
1938 else if (numa_node_id != cpu_to_node(cpu))
1941 } else if (dev_maps) {
1942 /* fill in the new device map from the old device map */
1943 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1944 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1949 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1951 /* Cleanup old maps */
1953 for_each_possible_cpu(cpu) {
1954 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1955 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1956 if (map && map != new_map)
1957 kfree_rcu(map, rcu);
1960 kfree_rcu(dev_maps, rcu);
1963 dev_maps = new_dev_maps;
1967 /* update Tx queue numa node */
1968 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
1969 (numa_node_id >= 0) ? numa_node_id :
1975 /* removes queue from unused CPUs */
1976 for_each_possible_cpu(cpu) {
1977 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
1980 if (remove_xps_queue(dev_maps, cpu, index))
1984 /* free map if not active */
1986 RCU_INIT_POINTER(dev->xps_maps, NULL);
1987 kfree_rcu(dev_maps, rcu);
1991 mutex_unlock(&xps_map_mutex);
1995 /* remove any maps that we added */
1996 for_each_possible_cpu(cpu) {
1997 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1998 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2000 if (new_map && new_map != map)
2004 mutex_unlock(&xps_map_mutex);
2006 kfree(new_dev_maps);
2009 EXPORT_SYMBOL(netif_set_xps_queue);
2013 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2014 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2016 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2020 if (txq < 1 || txq > dev->num_tx_queues)
2023 if (dev->reg_state == NETREG_REGISTERED ||
2024 dev->reg_state == NETREG_UNREGISTERING) {
2027 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2033 netif_setup_tc(dev, txq);
2035 if (txq < dev->real_num_tx_queues) {
2036 qdisc_reset_all_tx_gt(dev, txq);
2038 netif_reset_xps_queues_gt(dev, txq);
2043 dev->real_num_tx_queues = txq;
2046 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2050 * netif_set_real_num_rx_queues - set actual number of RX queues used
2051 * @dev: Network device
2052 * @rxq: Actual number of RX queues
2054 * This must be called either with the rtnl_lock held or before
2055 * registration of the net device. Returns 0 on success, or a
2056 * negative error code. If called before registration, it always
2059 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2063 if (rxq < 1 || rxq > dev->num_rx_queues)
2066 if (dev->reg_state == NETREG_REGISTERED) {
2069 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2075 dev->real_num_rx_queues = rxq;
2078 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2082 * netif_get_num_default_rss_queues - default number of RSS queues
2084 * This routine should set an upper limit on the number of RSS queues
2085 * used by default by multiqueue devices.
2087 int netif_get_num_default_rss_queues(void)
2089 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2091 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2093 static inline void __netif_reschedule(struct Qdisc *q)
2095 struct softnet_data *sd;
2096 unsigned long flags;
2098 local_irq_save(flags);
2099 sd = &__get_cpu_var(softnet_data);
2100 q->next_sched = NULL;
2101 *sd->output_queue_tailp = q;
2102 sd->output_queue_tailp = &q->next_sched;
2103 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2104 local_irq_restore(flags);
2107 void __netif_schedule(struct Qdisc *q)
2109 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2110 __netif_reschedule(q);
2112 EXPORT_SYMBOL(__netif_schedule);
2114 void dev_kfree_skb_irq(struct sk_buff *skb)
2116 if (atomic_dec_and_test(&skb->users)) {
2117 struct softnet_data *sd;
2118 unsigned long flags;
2120 local_irq_save(flags);
2121 sd = &__get_cpu_var(softnet_data);
2122 skb->next = sd->completion_queue;
2123 sd->completion_queue = skb;
2124 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2125 local_irq_restore(flags);
2128 EXPORT_SYMBOL(dev_kfree_skb_irq);
2130 void dev_kfree_skb_any(struct sk_buff *skb)
2132 if (in_irq() || irqs_disabled())
2133 dev_kfree_skb_irq(skb);
2137 EXPORT_SYMBOL(dev_kfree_skb_any);
2141 * netif_device_detach - mark device as removed
2142 * @dev: network device
2144 * Mark device as removed from system and therefore no longer available.
2146 void netif_device_detach(struct net_device *dev)
2148 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2149 netif_running(dev)) {
2150 netif_tx_stop_all_queues(dev);
2153 EXPORT_SYMBOL(netif_device_detach);
2156 * netif_device_attach - mark device as attached
2157 * @dev: network device
2159 * Mark device as attached from system and restart if needed.
2161 void netif_device_attach(struct net_device *dev)
2163 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2164 netif_running(dev)) {
2165 netif_tx_wake_all_queues(dev);
2166 __netdev_watchdog_up(dev);
2169 EXPORT_SYMBOL(netif_device_attach);
2171 static void skb_warn_bad_offload(const struct sk_buff *skb)
2173 static const netdev_features_t null_features = 0;
2174 struct net_device *dev = skb->dev;
2175 const char *driver = "";
2177 if (!net_ratelimit())
2180 if (dev && dev->dev.parent)
2181 driver = dev_driver_string(dev->dev.parent);
2183 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2184 "gso_type=%d ip_summed=%d\n",
2185 driver, dev ? &dev->features : &null_features,
2186 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2187 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2188 skb_shinfo(skb)->gso_type, skb->ip_summed);
2192 * Invalidate hardware checksum when packet is to be mangled, and
2193 * complete checksum manually on outgoing path.
2195 int skb_checksum_help(struct sk_buff *skb)
2198 int ret = 0, offset;
2200 if (skb->ip_summed == CHECKSUM_COMPLETE)
2201 goto out_set_summed;
2203 if (unlikely(skb_shinfo(skb)->gso_size)) {
2204 skb_warn_bad_offload(skb);
2208 /* Before computing a checksum, we should make sure no frag could
2209 * be modified by an external entity : checksum could be wrong.
2211 if (skb_has_shared_frag(skb)) {
2212 ret = __skb_linearize(skb);
2217 offset = skb_checksum_start_offset(skb);
2218 BUG_ON(offset >= skb_headlen(skb));
2219 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2221 offset += skb->csum_offset;
2222 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2224 if (skb_cloned(skb) &&
2225 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2226 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2231 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2233 skb->ip_summed = CHECKSUM_NONE;
2237 EXPORT_SYMBOL(skb_checksum_help);
2239 __be16 skb_network_protocol(struct sk_buff *skb)
2241 __be16 type = skb->protocol;
2242 int vlan_depth = ETH_HLEN;
2244 /* Tunnel gso handlers can set protocol to ethernet. */
2245 if (type == htons(ETH_P_TEB)) {
2248 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2251 eth = (struct ethhdr *)skb_mac_header(skb);
2252 type = eth->h_proto;
2255 while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2256 struct vlan_hdr *vh;
2258 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2261 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2262 type = vh->h_vlan_encapsulated_proto;
2263 vlan_depth += VLAN_HLEN;
2270 * skb_mac_gso_segment - mac layer segmentation handler.
2271 * @skb: buffer to segment
2272 * @features: features for the output path (see dev->features)
2274 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2275 netdev_features_t features)
2277 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2278 struct packet_offload *ptype;
2279 __be16 type = skb_network_protocol(skb);
2281 if (unlikely(!type))
2282 return ERR_PTR(-EINVAL);
2284 __skb_pull(skb, skb->mac_len);
2287 list_for_each_entry_rcu(ptype, &offload_base, list) {
2288 if (ptype->type == type && ptype->callbacks.gso_segment) {
2289 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2292 err = ptype->callbacks.gso_send_check(skb);
2293 segs = ERR_PTR(err);
2294 if (err || skb_gso_ok(skb, features))
2296 __skb_push(skb, (skb->data -
2297 skb_network_header(skb)));
2299 segs = ptype->callbacks.gso_segment(skb, features);
2305 __skb_push(skb, skb->data - skb_mac_header(skb));
2309 EXPORT_SYMBOL(skb_mac_gso_segment);
2312 /* openvswitch calls this on rx path, so we need a different check.
2314 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2317 return skb->ip_summed != CHECKSUM_PARTIAL;
2319 return skb->ip_summed == CHECKSUM_NONE;
2323 * __skb_gso_segment - Perform segmentation on skb.
2324 * @skb: buffer to segment
2325 * @features: features for the output path (see dev->features)
2326 * @tx_path: whether it is called in TX path
2328 * This function segments the given skb and returns a list of segments.
2330 * It may return NULL if the skb requires no segmentation. This is
2331 * only possible when GSO is used for verifying header integrity.
2333 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2334 netdev_features_t features, bool tx_path)
2336 if (unlikely(skb_needs_check(skb, tx_path))) {
2339 skb_warn_bad_offload(skb);
2341 if (skb_header_cloned(skb) &&
2342 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2343 return ERR_PTR(err);
2346 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2347 skb_reset_mac_header(skb);
2348 skb_reset_mac_len(skb);
2350 return skb_mac_gso_segment(skb, features);
2352 EXPORT_SYMBOL(__skb_gso_segment);
2354 /* Take action when hardware reception checksum errors are detected. */
2356 void netdev_rx_csum_fault(struct net_device *dev)
2358 if (net_ratelimit()) {
2359 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2363 EXPORT_SYMBOL(netdev_rx_csum_fault);
2366 /* Actually, we should eliminate this check as soon as we know, that:
2367 * 1. IOMMU is present and allows to map all the memory.
2368 * 2. No high memory really exists on this machine.
2371 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2373 #ifdef CONFIG_HIGHMEM
2375 if (!(dev->features & NETIF_F_HIGHDMA)) {
2376 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2377 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2378 if (PageHighMem(skb_frag_page(frag)))
2383 if (PCI_DMA_BUS_IS_PHYS) {
2384 struct device *pdev = dev->dev.parent;
2388 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2389 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2390 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2391 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2400 void (*destructor)(struct sk_buff *skb);
2403 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2405 static void dev_gso_skb_destructor(struct sk_buff *skb)
2407 struct dev_gso_cb *cb;
2410 struct sk_buff *nskb = skb->next;
2412 skb->next = nskb->next;
2415 } while (skb->next);
2417 cb = DEV_GSO_CB(skb);
2419 cb->destructor(skb);
2423 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2424 * @skb: buffer to segment
2425 * @features: device features as applicable to this skb
2427 * This function segments the given skb and stores the list of segments
2430 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2432 struct sk_buff *segs;
2434 segs = skb_gso_segment(skb, features);
2436 /* Verifying header integrity only. */
2441 return PTR_ERR(segs);
2444 DEV_GSO_CB(skb)->destructor = skb->destructor;
2445 skb->destructor = dev_gso_skb_destructor;
2450 static netdev_features_t harmonize_features(struct sk_buff *skb,
2451 __be16 protocol, netdev_features_t features)
2453 if (skb->ip_summed != CHECKSUM_NONE &&
2454 !can_checksum_protocol(features, protocol)) {
2455 features &= ~NETIF_F_ALL_CSUM;
2456 } else if (illegal_highdma(skb->dev, skb)) {
2457 features &= ~NETIF_F_SG;
2463 netdev_features_t netif_skb_features(struct sk_buff *skb)
2465 __be16 protocol = skb->protocol;
2466 netdev_features_t features = skb->dev->features;
2468 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2469 features &= ~NETIF_F_GSO_MASK;
2471 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2472 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2473 protocol = veh->h_vlan_encapsulated_proto;
2474 } else if (!vlan_tx_tag_present(skb)) {
2475 return harmonize_features(skb, protocol, features);
2478 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2479 NETIF_F_HW_VLAN_STAG_TX);
2481 if (protocol != htons(ETH_P_8021Q) && protocol != htons(ETH_P_8021AD)) {
2482 return harmonize_features(skb, protocol, features);
2484 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2485 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2486 NETIF_F_HW_VLAN_STAG_TX;
2487 return harmonize_features(skb, protocol, features);
2490 EXPORT_SYMBOL(netif_skb_features);
2493 * Returns true if either:
2494 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2495 * 2. skb is fragmented and the device does not support SG.
2497 static inline int skb_needs_linearize(struct sk_buff *skb,
2498 netdev_features_t features)
2500 return skb_is_nonlinear(skb) &&
2501 ((skb_has_frag_list(skb) &&
2502 !(features & NETIF_F_FRAGLIST)) ||
2503 (skb_shinfo(skb)->nr_frags &&
2504 !(features & NETIF_F_SG)));
2507 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2508 struct netdev_queue *txq)
2510 const struct net_device_ops *ops = dev->netdev_ops;
2511 int rc = NETDEV_TX_OK;
2512 unsigned int skb_len;
2514 if (likely(!skb->next)) {
2515 netdev_features_t features;
2518 * If device doesn't need skb->dst, release it right now while
2519 * its hot in this cpu cache
2521 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2524 features = netif_skb_features(skb);
2526 if (vlan_tx_tag_present(skb) &&
2527 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2528 skb = __vlan_put_tag(skb, skb->vlan_proto,
2529 vlan_tx_tag_get(skb));
2536 /* If encapsulation offload request, verify we are testing
2537 * hardware encapsulation features instead of standard
2538 * features for the netdev
2540 if (skb->encapsulation)
2541 features &= dev->hw_enc_features;
2543 if (netif_needs_gso(skb, features)) {
2544 if (unlikely(dev_gso_segment(skb, features)))
2549 if (skb_needs_linearize(skb, features) &&
2550 __skb_linearize(skb))
2553 /* If packet is not checksummed and device does not
2554 * support checksumming for this protocol, complete
2555 * checksumming here.
2557 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2558 if (skb->encapsulation)
2559 skb_set_inner_transport_header(skb,
2560 skb_checksum_start_offset(skb));
2562 skb_set_transport_header(skb,
2563 skb_checksum_start_offset(skb));
2564 if (!(features & NETIF_F_ALL_CSUM) &&
2565 skb_checksum_help(skb))
2570 if (!list_empty(&ptype_all))
2571 dev_queue_xmit_nit(skb, dev);
2574 rc = ops->ndo_start_xmit(skb, dev);
2575 trace_net_dev_xmit(skb, rc, dev, skb_len);
2576 if (rc == NETDEV_TX_OK)
2577 txq_trans_update(txq);
2583 struct sk_buff *nskb = skb->next;
2585 skb->next = nskb->next;
2588 if (!list_empty(&ptype_all))
2589 dev_queue_xmit_nit(nskb, dev);
2591 skb_len = nskb->len;
2592 rc = ops->ndo_start_xmit(nskb, dev);
2593 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2594 if (unlikely(rc != NETDEV_TX_OK)) {
2595 if (rc & ~NETDEV_TX_MASK)
2596 goto out_kfree_gso_skb;
2597 nskb->next = skb->next;
2601 txq_trans_update(txq);
2602 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2603 return NETDEV_TX_BUSY;
2604 } while (skb->next);
2607 if (likely(skb->next == NULL)) {
2608 skb->destructor = DEV_GSO_CB(skb)->destructor;
2618 static void qdisc_pkt_len_init(struct sk_buff *skb)
2620 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2622 qdisc_skb_cb(skb)->pkt_len = skb->len;
2624 /* To get more precise estimation of bytes sent on wire,
2625 * we add to pkt_len the headers size of all segments
2627 if (shinfo->gso_size) {
2628 unsigned int hdr_len;
2629 u16 gso_segs = shinfo->gso_segs;
2631 /* mac layer + network layer */
2632 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2634 /* + transport layer */
2635 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2636 hdr_len += tcp_hdrlen(skb);
2638 hdr_len += sizeof(struct udphdr);
2640 if (shinfo->gso_type & SKB_GSO_DODGY)
2641 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2644 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2648 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2649 struct net_device *dev,
2650 struct netdev_queue *txq)
2652 spinlock_t *root_lock = qdisc_lock(q);
2656 qdisc_pkt_len_init(skb);
2657 qdisc_calculate_pkt_len(skb, q);
2659 * Heuristic to force contended enqueues to serialize on a
2660 * separate lock before trying to get qdisc main lock.
2661 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2662 * and dequeue packets faster.
2664 contended = qdisc_is_running(q);
2665 if (unlikely(contended))
2666 spin_lock(&q->busylock);
2668 spin_lock(root_lock);
2669 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2672 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2673 qdisc_run_begin(q)) {
2675 * This is a work-conserving queue; there are no old skbs
2676 * waiting to be sent out; and the qdisc is not running -
2677 * xmit the skb directly.
2679 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2682 qdisc_bstats_update(q, skb);
2684 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2685 if (unlikely(contended)) {
2686 spin_unlock(&q->busylock);
2693 rc = NET_XMIT_SUCCESS;
2696 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2697 if (qdisc_run_begin(q)) {
2698 if (unlikely(contended)) {
2699 spin_unlock(&q->busylock);
2705 spin_unlock(root_lock);
2706 if (unlikely(contended))
2707 spin_unlock(&q->busylock);
2711 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2712 static void skb_update_prio(struct sk_buff *skb)
2714 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2716 if (!skb->priority && skb->sk && map) {
2717 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2719 if (prioidx < map->priomap_len)
2720 skb->priority = map->priomap[prioidx];
2724 #define skb_update_prio(skb)
2727 static DEFINE_PER_CPU(int, xmit_recursion);
2728 #define RECURSION_LIMIT 10
2731 * dev_loopback_xmit - loop back @skb
2732 * @skb: buffer to transmit
2734 int dev_loopback_xmit(struct sk_buff *skb)
2736 skb_reset_mac_header(skb);
2737 __skb_pull(skb, skb_network_offset(skb));
2738 skb->pkt_type = PACKET_LOOPBACK;
2739 skb->ip_summed = CHECKSUM_UNNECESSARY;
2740 WARN_ON(!skb_dst(skb));
2745 EXPORT_SYMBOL(dev_loopback_xmit);
2748 * dev_queue_xmit - transmit a buffer
2749 * @skb: buffer to transmit
2751 * Queue a buffer for transmission to a network device. The caller must
2752 * have set the device and priority and built the buffer before calling
2753 * this function. The function can be called from an interrupt.
2755 * A negative errno code is returned on a failure. A success does not
2756 * guarantee the frame will be transmitted as it may be dropped due
2757 * to congestion or traffic shaping.
2759 * -----------------------------------------------------------------------------------
2760 * I notice this method can also return errors from the queue disciplines,
2761 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2764 * Regardless of the return value, the skb is consumed, so it is currently
2765 * difficult to retry a send to this method. (You can bump the ref count
2766 * before sending to hold a reference for retry if you are careful.)
2768 * When calling this method, interrupts MUST be enabled. This is because
2769 * the BH enable code must have IRQs enabled so that it will not deadlock.
2772 int dev_queue_xmit(struct sk_buff *skb)
2774 struct net_device *dev = skb->dev;
2775 struct netdev_queue *txq;
2779 skb_reset_mac_header(skb);
2781 /* Disable soft irqs for various locks below. Also
2782 * stops preemption for RCU.
2786 skb_update_prio(skb);
2788 txq = netdev_pick_tx(dev, skb);
2789 q = rcu_dereference_bh(txq->qdisc);
2791 #ifdef CONFIG_NET_CLS_ACT
2792 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2794 trace_net_dev_queue(skb);
2796 rc = __dev_xmit_skb(skb, q, dev, txq);
2800 /* The device has no queue. Common case for software devices:
2801 loopback, all the sorts of tunnels...
2803 Really, it is unlikely that netif_tx_lock protection is necessary
2804 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2806 However, it is possible, that they rely on protection
2809 Check this and shot the lock. It is not prone from deadlocks.
2810 Either shot noqueue qdisc, it is even simpler 8)
2812 if (dev->flags & IFF_UP) {
2813 int cpu = smp_processor_id(); /* ok because BHs are off */
2815 if (txq->xmit_lock_owner != cpu) {
2817 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2818 goto recursion_alert;
2820 HARD_TX_LOCK(dev, txq, cpu);
2822 if (!netif_xmit_stopped(txq)) {
2823 __this_cpu_inc(xmit_recursion);
2824 rc = dev_hard_start_xmit(skb, dev, txq);
2825 __this_cpu_dec(xmit_recursion);
2826 if (dev_xmit_complete(rc)) {
2827 HARD_TX_UNLOCK(dev, txq);
2831 HARD_TX_UNLOCK(dev, txq);
2832 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2835 /* Recursion is detected! It is possible,
2839 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2845 rcu_read_unlock_bh();
2850 rcu_read_unlock_bh();
2853 EXPORT_SYMBOL(dev_queue_xmit);
2856 /*=======================================================================
2858 =======================================================================*/
2860 int netdev_max_backlog __read_mostly = 1000;
2861 EXPORT_SYMBOL(netdev_max_backlog);
2863 int netdev_tstamp_prequeue __read_mostly = 1;
2864 int netdev_budget __read_mostly = 300;
2865 int weight_p __read_mostly = 64; /* old backlog weight */
2867 /* Called with irq disabled */
2868 static inline void ____napi_schedule(struct softnet_data *sd,
2869 struct napi_struct *napi)
2871 list_add_tail(&napi->poll_list, &sd->poll_list);
2872 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2877 /* One global table that all flow-based protocols share. */
2878 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2879 EXPORT_SYMBOL(rps_sock_flow_table);
2881 struct static_key rps_needed __read_mostly;
2883 static struct rps_dev_flow *
2884 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2885 struct rps_dev_flow *rflow, u16 next_cpu)
2887 if (next_cpu != RPS_NO_CPU) {
2888 #ifdef CONFIG_RFS_ACCEL
2889 struct netdev_rx_queue *rxqueue;
2890 struct rps_dev_flow_table *flow_table;
2891 struct rps_dev_flow *old_rflow;
2896 /* Should we steer this flow to a different hardware queue? */
2897 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2898 !(dev->features & NETIF_F_NTUPLE))
2900 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2901 if (rxq_index == skb_get_rx_queue(skb))
2904 rxqueue = dev->_rx + rxq_index;
2905 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2908 flow_id = skb->rxhash & flow_table->mask;
2909 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2910 rxq_index, flow_id);
2914 rflow = &flow_table->flows[flow_id];
2916 if (old_rflow->filter == rflow->filter)
2917 old_rflow->filter = RPS_NO_FILTER;
2921 per_cpu(softnet_data, next_cpu).input_queue_head;
2924 rflow->cpu = next_cpu;
2929 * get_rps_cpu is called from netif_receive_skb and returns the target
2930 * CPU from the RPS map of the receiving queue for a given skb.
2931 * rcu_read_lock must be held on entry.
2933 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2934 struct rps_dev_flow **rflowp)
2936 struct netdev_rx_queue *rxqueue;
2937 struct rps_map *map;
2938 struct rps_dev_flow_table *flow_table;
2939 struct rps_sock_flow_table *sock_flow_table;
2943 if (skb_rx_queue_recorded(skb)) {
2944 u16 index = skb_get_rx_queue(skb);
2945 if (unlikely(index >= dev->real_num_rx_queues)) {
2946 WARN_ONCE(dev->real_num_rx_queues > 1,
2947 "%s received packet on queue %u, but number "
2948 "of RX queues is %u\n",
2949 dev->name, index, dev->real_num_rx_queues);
2952 rxqueue = dev->_rx + index;
2956 map = rcu_dereference(rxqueue->rps_map);
2958 if (map->len == 1 &&
2959 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2960 tcpu = map->cpus[0];
2961 if (cpu_online(tcpu))
2965 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2969 skb_reset_network_header(skb);
2970 if (!skb_get_rxhash(skb))
2973 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2974 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2975 if (flow_table && sock_flow_table) {
2977 struct rps_dev_flow *rflow;
2979 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2982 next_cpu = sock_flow_table->ents[skb->rxhash &
2983 sock_flow_table->mask];
2986 * If the desired CPU (where last recvmsg was done) is
2987 * different from current CPU (one in the rx-queue flow
2988 * table entry), switch if one of the following holds:
2989 * - Current CPU is unset (equal to RPS_NO_CPU).
2990 * - Current CPU is offline.
2991 * - The current CPU's queue tail has advanced beyond the
2992 * last packet that was enqueued using this table entry.
2993 * This guarantees that all previous packets for the flow
2994 * have been dequeued, thus preserving in order delivery.
2996 if (unlikely(tcpu != next_cpu) &&
2997 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2998 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2999 rflow->last_qtail)) >= 0)) {
3001 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3004 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3012 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
3014 if (cpu_online(tcpu)) {
3024 #ifdef CONFIG_RFS_ACCEL
3027 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3028 * @dev: Device on which the filter was set
3029 * @rxq_index: RX queue index
3030 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3031 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3033 * Drivers that implement ndo_rx_flow_steer() should periodically call
3034 * this function for each installed filter and remove the filters for
3035 * which it returns %true.
3037 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3038 u32 flow_id, u16 filter_id)
3040 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3041 struct rps_dev_flow_table *flow_table;
3042 struct rps_dev_flow *rflow;
3047 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3048 if (flow_table && flow_id <= flow_table->mask) {
3049 rflow = &flow_table->flows[flow_id];
3050 cpu = ACCESS_ONCE(rflow->cpu);
3051 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3052 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3053 rflow->last_qtail) <
3054 (int)(10 * flow_table->mask)))
3060 EXPORT_SYMBOL(rps_may_expire_flow);
3062 #endif /* CONFIG_RFS_ACCEL */
3064 /* Called from hardirq (IPI) context */
3065 static void rps_trigger_softirq(void *data)
3067 struct softnet_data *sd = data;
3069 ____napi_schedule(sd, &sd->backlog);
3073 #endif /* CONFIG_RPS */
3076 * Check if this softnet_data structure is another cpu one
3077 * If yes, queue it to our IPI list and return 1
3080 static int rps_ipi_queued(struct softnet_data *sd)
3083 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3086 sd->rps_ipi_next = mysd->rps_ipi_list;
3087 mysd->rps_ipi_list = sd;
3089 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3092 #endif /* CONFIG_RPS */
3096 #ifdef CONFIG_NET_FLOW_LIMIT
3097 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3100 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3102 #ifdef CONFIG_NET_FLOW_LIMIT
3103 struct sd_flow_limit *fl;
3104 struct softnet_data *sd;
3105 unsigned int old_flow, new_flow;
3107 if (qlen < (netdev_max_backlog >> 1))
3110 sd = &__get_cpu_var(softnet_data);
3113 fl = rcu_dereference(sd->flow_limit);
3115 new_flow = skb_get_rxhash(skb) & (fl->num_buckets - 1);
3116 old_flow = fl->history[fl->history_head];
3117 fl->history[fl->history_head] = new_flow;
3120 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3122 if (likely(fl->buckets[old_flow]))
3123 fl->buckets[old_flow]--;
3125 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3137 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3138 * queue (may be a remote CPU queue).
3140 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3141 unsigned int *qtail)
3143 struct softnet_data *sd;
3144 unsigned long flags;
3147 sd = &per_cpu(softnet_data, cpu);
3149 local_irq_save(flags);
3152 qlen = skb_queue_len(&sd->input_pkt_queue);
3153 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3154 if (skb_queue_len(&sd->input_pkt_queue)) {
3156 __skb_queue_tail(&sd->input_pkt_queue, skb);
3157 input_queue_tail_incr_save(sd, qtail);
3159 local_irq_restore(flags);
3160 return NET_RX_SUCCESS;
3163 /* Schedule NAPI for backlog device
3164 * We can use non atomic operation since we own the queue lock
3166 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3167 if (!rps_ipi_queued(sd))
3168 ____napi_schedule(sd, &sd->backlog);
3176 local_irq_restore(flags);
3178 atomic_long_inc(&skb->dev->rx_dropped);
3184 * netif_rx - post buffer to the network code
3185 * @skb: buffer to post
3187 * This function receives a packet from a device driver and queues it for
3188 * the upper (protocol) levels to process. It always succeeds. The buffer
3189 * may be dropped during processing for congestion control or by the
3193 * NET_RX_SUCCESS (no congestion)
3194 * NET_RX_DROP (packet was dropped)
3198 int netif_rx(struct sk_buff *skb)
3202 /* if netpoll wants it, pretend we never saw it */
3203 if (netpoll_rx(skb))
3206 net_timestamp_check(netdev_tstamp_prequeue, skb);
3208 trace_netif_rx(skb);
3210 if (static_key_false(&rps_needed)) {
3211 struct rps_dev_flow voidflow, *rflow = &voidflow;
3217 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3219 cpu = smp_processor_id();
3221 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3229 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3234 EXPORT_SYMBOL(netif_rx);
3236 int netif_rx_ni(struct sk_buff *skb)
3241 err = netif_rx(skb);
3242 if (local_softirq_pending())
3248 EXPORT_SYMBOL(netif_rx_ni);
3250 static void net_tx_action(struct softirq_action *h)
3252 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3254 if (sd->completion_queue) {
3255 struct sk_buff *clist;
3257 local_irq_disable();
3258 clist = sd->completion_queue;
3259 sd->completion_queue = NULL;
3263 struct sk_buff *skb = clist;
3264 clist = clist->next;
3266 WARN_ON(atomic_read(&skb->users));
3267 trace_kfree_skb(skb, net_tx_action);
3272 if (sd->output_queue) {
3275 local_irq_disable();
3276 head = sd->output_queue;
3277 sd->output_queue = NULL;
3278 sd->output_queue_tailp = &sd->output_queue;
3282 struct Qdisc *q = head;
3283 spinlock_t *root_lock;
3285 head = head->next_sched;
3287 root_lock = qdisc_lock(q);
3288 if (spin_trylock(root_lock)) {
3289 smp_mb__before_clear_bit();
3290 clear_bit(__QDISC_STATE_SCHED,
3293 spin_unlock(root_lock);
3295 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3297 __netif_reschedule(q);
3299 smp_mb__before_clear_bit();
3300 clear_bit(__QDISC_STATE_SCHED,
3308 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3309 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3310 /* This hook is defined here for ATM LANE */
3311 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3312 unsigned char *addr) __read_mostly;
3313 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3316 #ifdef CONFIG_NET_CLS_ACT
3317 /* TODO: Maybe we should just force sch_ingress to be compiled in
3318 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3319 * a compare and 2 stores extra right now if we dont have it on
3320 * but have CONFIG_NET_CLS_ACT
3321 * NOTE: This doesn't stop any functionality; if you dont have
3322 * the ingress scheduler, you just can't add policies on ingress.
3325 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3327 struct net_device *dev = skb->dev;
3328 u32 ttl = G_TC_RTTL(skb->tc_verd);
3329 int result = TC_ACT_OK;
3332 if (unlikely(MAX_RED_LOOP < ttl++)) {
3333 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3334 skb->skb_iif, dev->ifindex);
3338 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3339 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3342 if (q != &noop_qdisc) {
3343 spin_lock(qdisc_lock(q));
3344 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3345 result = qdisc_enqueue_root(skb, q);
3346 spin_unlock(qdisc_lock(q));
3352 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3353 struct packet_type **pt_prev,
3354 int *ret, struct net_device *orig_dev)
3356 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3358 if (!rxq || rxq->qdisc == &noop_qdisc)
3362 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3366 switch (ing_filter(skb, rxq)) {
3380 * netdev_rx_handler_register - register receive handler
3381 * @dev: device to register a handler for
3382 * @rx_handler: receive handler to register
3383 * @rx_handler_data: data pointer that is used by rx handler
3385 * Register a receive hander for a device. This handler will then be
3386 * called from __netif_receive_skb. A negative errno code is returned
3389 * The caller must hold the rtnl_mutex.
3391 * For a general description of rx_handler, see enum rx_handler_result.
3393 int netdev_rx_handler_register(struct net_device *dev,
3394 rx_handler_func_t *rx_handler,
3395 void *rx_handler_data)
3399 if (dev->rx_handler)
3402 /* Note: rx_handler_data must be set before rx_handler */
3403 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3404 rcu_assign_pointer(dev->rx_handler, rx_handler);
3408 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3411 * netdev_rx_handler_unregister - unregister receive handler
3412 * @dev: device to unregister a handler from
3414 * Unregister a receive handler from a device.
3416 * The caller must hold the rtnl_mutex.
3418 void netdev_rx_handler_unregister(struct net_device *dev)
3422 RCU_INIT_POINTER(dev->rx_handler, NULL);
3423 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3424 * section has a guarantee to see a non NULL rx_handler_data
3428 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3430 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3433 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3434 * the special handling of PFMEMALLOC skbs.
3436 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3438 switch (skb->protocol) {
3439 case __constant_htons(ETH_P_ARP):
3440 case __constant_htons(ETH_P_IP):
3441 case __constant_htons(ETH_P_IPV6):
3442 case __constant_htons(ETH_P_8021Q):
3443 case __constant_htons(ETH_P_8021AD):
3450 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3452 struct packet_type *ptype, *pt_prev;
3453 rx_handler_func_t *rx_handler;
3454 struct net_device *orig_dev;
3455 struct net_device *null_or_dev;
3456 bool deliver_exact = false;
3457 int ret = NET_RX_DROP;
3460 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3462 trace_netif_receive_skb(skb);
3464 /* if we've gotten here through NAPI, check netpoll */
3465 if (netpoll_receive_skb(skb))
3468 orig_dev = skb->dev;
3470 skb_reset_network_header(skb);
3471 if (!skb_transport_header_was_set(skb))
3472 skb_reset_transport_header(skb);
3473 skb_reset_mac_len(skb);
3480 skb->skb_iif = skb->dev->ifindex;
3482 __this_cpu_inc(softnet_data.processed);
3484 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3485 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3486 skb = vlan_untag(skb);
3491 #ifdef CONFIG_NET_CLS_ACT
3492 if (skb->tc_verd & TC_NCLS) {
3493 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3501 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3502 if (!ptype->dev || ptype->dev == skb->dev) {
3504 ret = deliver_skb(skb, pt_prev, orig_dev);
3510 #ifdef CONFIG_NET_CLS_ACT
3511 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3517 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3520 if (vlan_tx_tag_present(skb)) {
3522 ret = deliver_skb(skb, pt_prev, orig_dev);
3525 if (vlan_do_receive(&skb))
3527 else if (unlikely(!skb))
3531 rx_handler = rcu_dereference(skb->dev->rx_handler);
3534 ret = deliver_skb(skb, pt_prev, orig_dev);
3537 switch (rx_handler(&skb)) {
3538 case RX_HANDLER_CONSUMED:
3539 ret = NET_RX_SUCCESS;
3541 case RX_HANDLER_ANOTHER:
3543 case RX_HANDLER_EXACT:
3544 deliver_exact = true;
3545 case RX_HANDLER_PASS:
3552 if (vlan_tx_nonzero_tag_present(skb))
3553 skb->pkt_type = PACKET_OTHERHOST;
3555 /* deliver only exact match when indicated */
3556 null_or_dev = deliver_exact ? skb->dev : NULL;
3558 type = skb->protocol;
3559 list_for_each_entry_rcu(ptype,
3560 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3561 if (ptype->type == type &&
3562 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3563 ptype->dev == orig_dev)) {
3565 ret = deliver_skb(skb, pt_prev, orig_dev);
3571 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3574 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3577 atomic_long_inc(&skb->dev->rx_dropped);
3579 /* Jamal, now you will not able to escape explaining
3580 * me how you were going to use this. :-)
3591 static int __netif_receive_skb(struct sk_buff *skb)
3595 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3596 unsigned long pflags = current->flags;
3599 * PFMEMALLOC skbs are special, they should
3600 * - be delivered to SOCK_MEMALLOC sockets only
3601 * - stay away from userspace
3602 * - have bounded memory usage
3604 * Use PF_MEMALLOC as this saves us from propagating the allocation
3605 * context down to all allocation sites.
3607 current->flags |= PF_MEMALLOC;
3608 ret = __netif_receive_skb_core(skb, true);
3609 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3611 ret = __netif_receive_skb_core(skb, false);
3617 * netif_receive_skb - process receive buffer from network
3618 * @skb: buffer to process
3620 * netif_receive_skb() is the main receive data processing function.
3621 * It always succeeds. The buffer may be dropped during processing
3622 * for congestion control or by the protocol layers.
3624 * This function may only be called from softirq context and interrupts
3625 * should be enabled.
3627 * Return values (usually ignored):
3628 * NET_RX_SUCCESS: no congestion
3629 * NET_RX_DROP: packet was dropped
3631 int netif_receive_skb(struct sk_buff *skb)
3633 net_timestamp_check(netdev_tstamp_prequeue, skb);
3635 if (skb_defer_rx_timestamp(skb))
3636 return NET_RX_SUCCESS;
3639 if (static_key_false(&rps_needed)) {
3640 struct rps_dev_flow voidflow, *rflow = &voidflow;
3645 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3648 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3655 return __netif_receive_skb(skb);
3657 EXPORT_SYMBOL(netif_receive_skb);
3659 /* Network device is going away, flush any packets still pending
3660 * Called with irqs disabled.
3662 static void flush_backlog(void *arg)
3664 struct net_device *dev = arg;
3665 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3666 struct sk_buff *skb, *tmp;
3669 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3670 if (skb->dev == dev) {
3671 __skb_unlink(skb, &sd->input_pkt_queue);
3673 input_queue_head_incr(sd);
3678 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3679 if (skb->dev == dev) {
3680 __skb_unlink(skb, &sd->process_queue);
3682 input_queue_head_incr(sd);
3687 static int napi_gro_complete(struct sk_buff *skb)
3689 struct packet_offload *ptype;
3690 __be16 type = skb->protocol;
3691 struct list_head *head = &offload_base;
3694 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3696 if (NAPI_GRO_CB(skb)->count == 1) {
3697 skb_shinfo(skb)->gso_size = 0;
3702 list_for_each_entry_rcu(ptype, head, list) {
3703 if (ptype->type != type || !ptype->callbacks.gro_complete)
3706 err = ptype->callbacks.gro_complete(skb);
3712 WARN_ON(&ptype->list == head);
3714 return NET_RX_SUCCESS;
3718 return netif_receive_skb(skb);
3721 /* napi->gro_list contains packets ordered by age.
3722 * youngest packets at the head of it.
3723 * Complete skbs in reverse order to reduce latencies.
3725 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3727 struct sk_buff *skb, *prev = NULL;
3729 /* scan list and build reverse chain */
3730 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3735 for (skb = prev; skb; skb = prev) {
3738 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3742 napi_gro_complete(skb);
3746 napi->gro_list = NULL;
3748 EXPORT_SYMBOL(napi_gro_flush);
3750 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3753 unsigned int maclen = skb->dev->hard_header_len;
3755 for (p = napi->gro_list; p; p = p->next) {
3756 unsigned long diffs;
3758 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3759 diffs |= p->vlan_tci ^ skb->vlan_tci;
3760 if (maclen == ETH_HLEN)
3761 diffs |= compare_ether_header(skb_mac_header(p),
3762 skb_gro_mac_header(skb));
3764 diffs = memcmp(skb_mac_header(p),
3765 skb_gro_mac_header(skb),
3767 NAPI_GRO_CB(p)->same_flow = !diffs;
3768 NAPI_GRO_CB(p)->flush = 0;
3772 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3774 struct sk_buff **pp = NULL;
3775 struct packet_offload *ptype;
3776 __be16 type = skb->protocol;
3777 struct list_head *head = &offload_base;
3779 enum gro_result ret;
3781 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3784 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3787 gro_list_prepare(napi, skb);
3790 list_for_each_entry_rcu(ptype, head, list) {
3791 if (ptype->type != type || !ptype->callbacks.gro_receive)
3794 skb_set_network_header(skb, skb_gro_offset(skb));
3795 skb_reset_mac_len(skb);
3796 NAPI_GRO_CB(skb)->same_flow = 0;
3797 NAPI_GRO_CB(skb)->flush = 0;
3798 NAPI_GRO_CB(skb)->free = 0;
3800 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3805 if (&ptype->list == head)
3808 same_flow = NAPI_GRO_CB(skb)->same_flow;
3809 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3812 struct sk_buff *nskb = *pp;
3816 napi_gro_complete(nskb);
3823 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3827 NAPI_GRO_CB(skb)->count = 1;
3828 NAPI_GRO_CB(skb)->age = jiffies;
3829 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3830 skb->next = napi->gro_list;
3831 napi->gro_list = skb;
3835 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3836 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3838 BUG_ON(skb->end - skb->tail < grow);
3840 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3843 skb->data_len -= grow;
3845 skb_shinfo(skb)->frags[0].page_offset += grow;
3846 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3848 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3849 skb_frag_unref(skb, 0);
3850 memmove(skb_shinfo(skb)->frags,
3851 skb_shinfo(skb)->frags + 1,
3852 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3865 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3869 if (netif_receive_skb(skb))
3877 case GRO_MERGED_FREE:
3878 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3879 kmem_cache_free(skbuff_head_cache, skb);
3892 static void skb_gro_reset_offset(struct sk_buff *skb)
3894 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3895 const skb_frag_t *frag0 = &pinfo->frags[0];
3897 NAPI_GRO_CB(skb)->data_offset = 0;
3898 NAPI_GRO_CB(skb)->frag0 = NULL;
3899 NAPI_GRO_CB(skb)->frag0_len = 0;
3901 if (skb->mac_header == skb->tail &&
3903 !PageHighMem(skb_frag_page(frag0))) {
3904 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3905 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3909 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3911 skb_gro_reset_offset(skb);
3913 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
3915 EXPORT_SYMBOL(napi_gro_receive);
3917 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3919 __skb_pull(skb, skb_headlen(skb));
3920 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3921 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3923 skb->dev = napi->dev;
3929 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3931 struct sk_buff *skb = napi->skb;
3934 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3940 EXPORT_SYMBOL(napi_get_frags);
3942 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3948 skb->protocol = eth_type_trans(skb, skb->dev);
3950 if (ret == GRO_HELD)
3951 skb_gro_pull(skb, -ETH_HLEN);
3952 else if (netif_receive_skb(skb))
3957 case GRO_MERGED_FREE:
3958 napi_reuse_skb(napi, skb);
3968 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3970 struct sk_buff *skb = napi->skb;
3977 skb_reset_mac_header(skb);
3978 skb_gro_reset_offset(skb);
3980 off = skb_gro_offset(skb);
3981 hlen = off + sizeof(*eth);
3982 eth = skb_gro_header_fast(skb, off);
3983 if (skb_gro_header_hard(skb, hlen)) {
3984 eth = skb_gro_header_slow(skb, hlen, off);
3985 if (unlikely(!eth)) {
3986 napi_reuse_skb(napi, skb);
3992 skb_gro_pull(skb, sizeof(*eth));
3995 * This works because the only protocols we care about don't require
3996 * special handling. We'll fix it up properly at the end.
3998 skb->protocol = eth->h_proto;
4004 gro_result_t napi_gro_frags(struct napi_struct *napi)
4006 struct sk_buff *skb = napi_frags_skb(napi);
4011 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4013 EXPORT_SYMBOL(napi_gro_frags);
4016 * net_rps_action sends any pending IPI's for rps.
4017 * Note: called with local irq disabled, but exits with local irq enabled.
4019 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4022 struct softnet_data *remsd = sd->rps_ipi_list;
4025 sd->rps_ipi_list = NULL;
4029 /* Send pending IPI's to kick RPS processing on remote cpus. */
4031 struct softnet_data *next = remsd->rps_ipi_next;
4033 if (cpu_online(remsd->cpu))
4034 __smp_call_function_single(remsd->cpu,
4043 static int process_backlog(struct napi_struct *napi, int quota)
4046 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4049 /* Check if we have pending ipi, its better to send them now,
4050 * not waiting net_rx_action() end.
4052 if (sd->rps_ipi_list) {
4053 local_irq_disable();
4054 net_rps_action_and_irq_enable(sd);
4057 napi->weight = weight_p;
4058 local_irq_disable();
4059 while (work < quota) {
4060 struct sk_buff *skb;
4063 while ((skb = __skb_dequeue(&sd->process_queue))) {
4065 __netif_receive_skb(skb);
4066 local_irq_disable();
4067 input_queue_head_incr(sd);
4068 if (++work >= quota) {
4075 qlen = skb_queue_len(&sd->input_pkt_queue);
4077 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4078 &sd->process_queue);
4080 if (qlen < quota - work) {
4082 * Inline a custom version of __napi_complete().
4083 * only current cpu owns and manipulates this napi,
4084 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4085 * we can use a plain write instead of clear_bit(),
4086 * and we dont need an smp_mb() memory barrier.
4088 list_del(&napi->poll_list);
4091 quota = work + qlen;
4101 * __napi_schedule - schedule for receive
4102 * @n: entry to schedule
4104 * The entry's receive function will be scheduled to run
4106 void __napi_schedule(struct napi_struct *n)
4108 unsigned long flags;
4110 local_irq_save(flags);
4111 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4112 local_irq_restore(flags);
4114 EXPORT_SYMBOL(__napi_schedule);
4116 void __napi_complete(struct napi_struct *n)
4118 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4119 BUG_ON(n->gro_list);
4121 list_del(&n->poll_list);
4122 smp_mb__before_clear_bit();
4123 clear_bit(NAPI_STATE_SCHED, &n->state);
4125 EXPORT_SYMBOL(__napi_complete);
4127 void napi_complete(struct napi_struct *n)
4129 unsigned long flags;
4132 * don't let napi dequeue from the cpu poll list
4133 * just in case its running on a different cpu
4135 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4138 napi_gro_flush(n, false);
4139 local_irq_save(flags);
4141 local_irq_restore(flags);
4143 EXPORT_SYMBOL(napi_complete);
4145 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4146 int (*poll)(struct napi_struct *, int), int weight)
4148 INIT_LIST_HEAD(&napi->poll_list);
4149 napi->gro_count = 0;
4150 napi->gro_list = NULL;
4153 if (weight > NAPI_POLL_WEIGHT)
4154 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4156 napi->weight = weight;
4157 list_add(&napi->dev_list, &dev->napi_list);
4159 #ifdef CONFIG_NETPOLL
4160 spin_lock_init(&napi->poll_lock);
4161 napi->poll_owner = -1;
4163 set_bit(NAPI_STATE_SCHED, &napi->state);
4165 EXPORT_SYMBOL(netif_napi_add);
4167 void netif_napi_del(struct napi_struct *napi)
4169 struct sk_buff *skb, *next;
4171 list_del_init(&napi->dev_list);
4172 napi_free_frags(napi);
4174 for (skb = napi->gro_list; skb; skb = next) {
4180 napi->gro_list = NULL;
4181 napi->gro_count = 0;
4183 EXPORT_SYMBOL(netif_napi_del);
4185 static void net_rx_action(struct softirq_action *h)
4187 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4188 unsigned long time_limit = jiffies + 2;
4189 int budget = netdev_budget;
4192 local_irq_disable();
4194 while (!list_empty(&sd->poll_list)) {
4195 struct napi_struct *n;
4198 /* If softirq window is exhuasted then punt.
4199 * Allow this to run for 2 jiffies since which will allow
4200 * an average latency of 1.5/HZ.
4202 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4207 /* Even though interrupts have been re-enabled, this
4208 * access is safe because interrupts can only add new
4209 * entries to the tail of this list, and only ->poll()
4210 * calls can remove this head entry from the list.
4212 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4214 have = netpoll_poll_lock(n);
4218 /* This NAPI_STATE_SCHED test is for avoiding a race
4219 * with netpoll's poll_napi(). Only the entity which
4220 * obtains the lock and sees NAPI_STATE_SCHED set will
4221 * actually make the ->poll() call. Therefore we avoid
4222 * accidentally calling ->poll() when NAPI is not scheduled.
4225 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4226 work = n->poll(n, weight);
4230 WARN_ON_ONCE(work > weight);
4234 local_irq_disable();
4236 /* Drivers must not modify the NAPI state if they
4237 * consume the entire weight. In such cases this code
4238 * still "owns" the NAPI instance and therefore can
4239 * move the instance around on the list at-will.
4241 if (unlikely(work == weight)) {
4242 if (unlikely(napi_disable_pending(n))) {
4245 local_irq_disable();
4248 /* flush too old packets
4249 * If HZ < 1000, flush all packets.
4252 napi_gro_flush(n, HZ >= 1000);
4253 local_irq_disable();
4255 list_move_tail(&n->poll_list, &sd->poll_list);
4259 netpoll_poll_unlock(have);
4262 net_rps_action_and_irq_enable(sd);
4264 #ifdef CONFIG_NET_DMA
4266 * There may not be any more sk_buffs coming right now, so push
4267 * any pending DMA copies to hardware
4269 dma_issue_pending_all();
4276 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4280 struct netdev_upper {
4281 struct net_device *dev;
4283 struct list_head list;
4284 struct rcu_head rcu;
4285 struct list_head search_list;
4288 static void __append_search_uppers(struct list_head *search_list,
4289 struct net_device *dev)
4291 struct netdev_upper *upper;
4293 list_for_each_entry(upper, &dev->upper_dev_list, list) {
4294 /* check if this upper is not already in search list */
4295 if (list_empty(&upper->search_list))
4296 list_add_tail(&upper->search_list, search_list);
4300 static bool __netdev_search_upper_dev(struct net_device *dev,
4301 struct net_device *upper_dev)
4303 LIST_HEAD(search_list);
4304 struct netdev_upper *upper;
4305 struct netdev_upper *tmp;
4308 __append_search_uppers(&search_list, dev);
4309 list_for_each_entry(upper, &search_list, search_list) {
4310 if (upper->dev == upper_dev) {
4314 __append_search_uppers(&search_list, upper->dev);
4316 list_for_each_entry_safe(upper, tmp, &search_list, search_list)
4317 INIT_LIST_HEAD(&upper->search_list);
4321 static struct netdev_upper *__netdev_find_upper(struct net_device *dev,
4322 struct net_device *upper_dev)
4324 struct netdev_upper *upper;
4326 list_for_each_entry(upper, &dev->upper_dev_list, list) {
4327 if (upper->dev == upper_dev)
4334 * netdev_has_upper_dev - Check if device is linked to an upper device
4336 * @upper_dev: upper device to check
4338 * Find out if a device is linked to specified upper device and return true
4339 * in case it is. Note that this checks only immediate upper device,
4340 * not through a complete stack of devices. The caller must hold the RTNL lock.
4342 bool netdev_has_upper_dev(struct net_device *dev,
4343 struct net_device *upper_dev)
4347 return __netdev_find_upper(dev, upper_dev);
4349 EXPORT_SYMBOL(netdev_has_upper_dev);
4352 * netdev_has_any_upper_dev - Check if device is linked to some device
4355 * Find out if a device is linked to an upper device and return true in case
4356 * it is. The caller must hold the RTNL lock.
4358 bool netdev_has_any_upper_dev(struct net_device *dev)
4362 return !list_empty(&dev->upper_dev_list);
4364 EXPORT_SYMBOL(netdev_has_any_upper_dev);
4367 * netdev_master_upper_dev_get - Get master upper device
4370 * Find a master upper device and return pointer to it or NULL in case
4371 * it's not there. The caller must hold the RTNL lock.
4373 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4375 struct netdev_upper *upper;
4379 if (list_empty(&dev->upper_dev_list))
4382 upper = list_first_entry(&dev->upper_dev_list,
4383 struct netdev_upper, list);
4384 if (likely(upper->master))
4388 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4391 * netdev_master_upper_dev_get_rcu - Get master upper device
4394 * Find a master upper device and return pointer to it or NULL in case
4395 * it's not there. The caller must hold the RCU read lock.
4397 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4399 struct netdev_upper *upper;
4401 upper = list_first_or_null_rcu(&dev->upper_dev_list,
4402 struct netdev_upper, list);
4403 if (upper && likely(upper->master))
4407 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4409 static int __netdev_upper_dev_link(struct net_device *dev,
4410 struct net_device *upper_dev, bool master)
4412 struct netdev_upper *upper;
4416 if (dev == upper_dev)
4419 /* To prevent loops, check if dev is not upper device to upper_dev. */
4420 if (__netdev_search_upper_dev(upper_dev, dev))
4423 if (__netdev_find_upper(dev, upper_dev))
4426 if (master && netdev_master_upper_dev_get(dev))
4429 upper = kmalloc(sizeof(*upper), GFP_KERNEL);
4433 upper->dev = upper_dev;
4434 upper->master = master;
4435 INIT_LIST_HEAD(&upper->search_list);
4437 /* Ensure that master upper link is always the first item in list. */
4439 list_add_rcu(&upper->list, &dev->upper_dev_list);
4441 list_add_tail_rcu(&upper->list, &dev->upper_dev_list);
4442 dev_hold(upper_dev);
4443 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4448 * netdev_upper_dev_link - Add a link to the upper device
4450 * @upper_dev: new upper device
4452 * Adds a link to device which is upper to this one. The caller must hold
4453 * the RTNL lock. On a failure a negative errno code is returned.
4454 * On success the reference counts are adjusted and the function
4457 int netdev_upper_dev_link(struct net_device *dev,
4458 struct net_device *upper_dev)
4460 return __netdev_upper_dev_link(dev, upper_dev, false);
4462 EXPORT_SYMBOL(netdev_upper_dev_link);
4465 * netdev_master_upper_dev_link - Add a master link to the upper device
4467 * @upper_dev: new upper device
4469 * Adds a link to device which is upper to this one. In this case, only
4470 * one master upper device can be linked, although other non-master devices
4471 * might be linked as well. The caller must hold the RTNL lock.
4472 * On a failure a negative errno code is returned. On success the reference
4473 * counts are adjusted and the function returns zero.
4475 int netdev_master_upper_dev_link(struct net_device *dev,
4476 struct net_device *upper_dev)
4478 return __netdev_upper_dev_link(dev, upper_dev, true);
4480 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4483 * netdev_upper_dev_unlink - Removes a link to upper device
4485 * @upper_dev: new upper device
4487 * Removes a link to device which is upper to this one. The caller must hold
4490 void netdev_upper_dev_unlink(struct net_device *dev,
4491 struct net_device *upper_dev)
4493 struct netdev_upper *upper;
4497 upper = __netdev_find_upper(dev, upper_dev);
4500 list_del_rcu(&upper->list);
4502 kfree_rcu(upper, rcu);
4503 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4505 EXPORT_SYMBOL(netdev_upper_dev_unlink);
4507 static void dev_change_rx_flags(struct net_device *dev, int flags)
4509 const struct net_device_ops *ops = dev->netdev_ops;
4511 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4512 ops->ndo_change_rx_flags(dev, flags);
4515 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4517 unsigned int old_flags = dev->flags;
4523 dev->flags |= IFF_PROMISC;
4524 dev->promiscuity += inc;
4525 if (dev->promiscuity == 0) {
4528 * If inc causes overflow, untouch promisc and return error.
4531 dev->flags &= ~IFF_PROMISC;
4533 dev->promiscuity -= inc;
4534 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4539 if (dev->flags != old_flags) {
4540 pr_info("device %s %s promiscuous mode\n",
4542 dev->flags & IFF_PROMISC ? "entered" : "left");
4543 if (audit_enabled) {
4544 current_uid_gid(&uid, &gid);
4545 audit_log(current->audit_context, GFP_ATOMIC,
4546 AUDIT_ANOM_PROMISCUOUS,
4547 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4548 dev->name, (dev->flags & IFF_PROMISC),
4549 (old_flags & IFF_PROMISC),
4550 from_kuid(&init_user_ns, audit_get_loginuid(current)),
4551 from_kuid(&init_user_ns, uid),
4552 from_kgid(&init_user_ns, gid),
4553 audit_get_sessionid(current));
4556 dev_change_rx_flags(dev, IFF_PROMISC);
4562 * dev_set_promiscuity - update promiscuity count on a device
4566 * Add or remove promiscuity from a device. While the count in the device
4567 * remains above zero the interface remains promiscuous. Once it hits zero
4568 * the device reverts back to normal filtering operation. A negative inc
4569 * value is used to drop promiscuity on the device.
4570 * Return 0 if successful or a negative errno code on error.
4572 int dev_set_promiscuity(struct net_device *dev, int inc)
4574 unsigned int old_flags = dev->flags;
4577 err = __dev_set_promiscuity(dev, inc);
4580 if (dev->flags != old_flags)
4581 dev_set_rx_mode(dev);
4584 EXPORT_SYMBOL(dev_set_promiscuity);
4587 * dev_set_allmulti - update allmulti count on a device
4591 * Add or remove reception of all multicast frames to a device. While the
4592 * count in the device remains above zero the interface remains listening
4593 * to all interfaces. Once it hits zero the device reverts back to normal
4594 * filtering operation. A negative @inc value is used to drop the counter
4595 * when releasing a resource needing all multicasts.
4596 * Return 0 if successful or a negative errno code on error.
4599 int dev_set_allmulti(struct net_device *dev, int inc)
4601 unsigned int old_flags = dev->flags;
4605 dev->flags |= IFF_ALLMULTI;
4606 dev->allmulti += inc;
4607 if (dev->allmulti == 0) {
4610 * If inc causes overflow, untouch allmulti and return error.
4613 dev->flags &= ~IFF_ALLMULTI;
4615 dev->allmulti -= inc;
4616 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4621 if (dev->flags ^ old_flags) {
4622 dev_change_rx_flags(dev, IFF_ALLMULTI);
4623 dev_set_rx_mode(dev);
4627 EXPORT_SYMBOL(dev_set_allmulti);
4630 * Upload unicast and multicast address lists to device and
4631 * configure RX filtering. When the device doesn't support unicast
4632 * filtering it is put in promiscuous mode while unicast addresses
4635 void __dev_set_rx_mode(struct net_device *dev)
4637 const struct net_device_ops *ops = dev->netdev_ops;
4639 /* dev_open will call this function so the list will stay sane. */
4640 if (!(dev->flags&IFF_UP))
4643 if (!netif_device_present(dev))
4646 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4647 /* Unicast addresses changes may only happen under the rtnl,
4648 * therefore calling __dev_set_promiscuity here is safe.
4650 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4651 __dev_set_promiscuity(dev, 1);
4652 dev->uc_promisc = true;
4653 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4654 __dev_set_promiscuity(dev, -1);
4655 dev->uc_promisc = false;
4659 if (ops->ndo_set_rx_mode)
4660 ops->ndo_set_rx_mode(dev);
4663 void dev_set_rx_mode(struct net_device *dev)
4665 netif_addr_lock_bh(dev);
4666 __dev_set_rx_mode(dev);
4667 netif_addr_unlock_bh(dev);
4671 * dev_get_flags - get flags reported to userspace
4674 * Get the combination of flag bits exported through APIs to userspace.
4676 unsigned int dev_get_flags(const struct net_device *dev)
4680 flags = (dev->flags & ~(IFF_PROMISC |
4685 (dev->gflags & (IFF_PROMISC |
4688 if (netif_running(dev)) {
4689 if (netif_oper_up(dev))
4690 flags |= IFF_RUNNING;
4691 if (netif_carrier_ok(dev))
4692 flags |= IFF_LOWER_UP;
4693 if (netif_dormant(dev))
4694 flags |= IFF_DORMANT;
4699 EXPORT_SYMBOL(dev_get_flags);
4701 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4703 unsigned int old_flags = dev->flags;
4709 * Set the flags on our device.
4712 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4713 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4715 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4719 * Load in the correct multicast list now the flags have changed.
4722 if ((old_flags ^ flags) & IFF_MULTICAST)
4723 dev_change_rx_flags(dev, IFF_MULTICAST);
4725 dev_set_rx_mode(dev);
4728 * Have we downed the interface. We handle IFF_UP ourselves
4729 * according to user attempts to set it, rather than blindly
4734 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4735 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4738 dev_set_rx_mode(dev);
4741 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4742 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4744 dev->gflags ^= IFF_PROMISC;
4745 dev_set_promiscuity(dev, inc);
4748 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4749 is important. Some (broken) drivers set IFF_PROMISC, when
4750 IFF_ALLMULTI is requested not asking us and not reporting.
4752 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4753 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4755 dev->gflags ^= IFF_ALLMULTI;
4756 dev_set_allmulti(dev, inc);
4762 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4764 unsigned int changes = dev->flags ^ old_flags;
4766 if (changes & IFF_UP) {
4767 if (dev->flags & IFF_UP)
4768 call_netdevice_notifiers(NETDEV_UP, dev);
4770 call_netdevice_notifiers(NETDEV_DOWN, dev);
4773 if (dev->flags & IFF_UP &&
4774 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
4775 struct netdev_notifier_change_info change_info;
4777 change_info.flags_changed = changes;
4778 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
4784 * dev_change_flags - change device settings
4786 * @flags: device state flags
4788 * Change settings on device based state flags. The flags are
4789 * in the userspace exported format.
4791 int dev_change_flags(struct net_device *dev, unsigned int flags)
4794 unsigned int changes, old_flags = dev->flags;
4796 ret = __dev_change_flags(dev, flags);
4800 changes = old_flags ^ dev->flags;
4802 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4804 __dev_notify_flags(dev, old_flags);
4807 EXPORT_SYMBOL(dev_change_flags);
4810 * dev_set_mtu - Change maximum transfer unit
4812 * @new_mtu: new transfer unit
4814 * Change the maximum transfer size of the network device.
4816 int dev_set_mtu(struct net_device *dev, int new_mtu)
4818 const struct net_device_ops *ops = dev->netdev_ops;
4821 if (new_mtu == dev->mtu)
4824 /* MTU must be positive. */
4828 if (!netif_device_present(dev))
4832 if (ops->ndo_change_mtu)
4833 err = ops->ndo_change_mtu(dev, new_mtu);
4838 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4841 EXPORT_SYMBOL(dev_set_mtu);
4844 * dev_set_group - Change group this device belongs to
4846 * @new_group: group this device should belong to
4848 void dev_set_group(struct net_device *dev, int new_group)
4850 dev->group = new_group;
4852 EXPORT_SYMBOL(dev_set_group);
4855 * dev_set_mac_address - Change Media Access Control Address
4859 * Change the hardware (MAC) address of the device
4861 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4863 const struct net_device_ops *ops = dev->netdev_ops;
4866 if (!ops->ndo_set_mac_address)
4868 if (sa->sa_family != dev->type)
4870 if (!netif_device_present(dev))
4872 err = ops->ndo_set_mac_address(dev, sa);
4875 dev->addr_assign_type = NET_ADDR_SET;
4876 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4877 add_device_randomness(dev->dev_addr, dev->addr_len);
4880 EXPORT_SYMBOL(dev_set_mac_address);
4883 * dev_change_carrier - Change device carrier
4885 * @new_carrier: new value
4887 * Change device carrier
4889 int dev_change_carrier(struct net_device *dev, bool new_carrier)
4891 const struct net_device_ops *ops = dev->netdev_ops;
4893 if (!ops->ndo_change_carrier)
4895 if (!netif_device_present(dev))
4897 return ops->ndo_change_carrier(dev, new_carrier);
4899 EXPORT_SYMBOL(dev_change_carrier);
4902 * dev_new_index - allocate an ifindex
4903 * @net: the applicable net namespace
4905 * Returns a suitable unique value for a new device interface
4906 * number. The caller must hold the rtnl semaphore or the
4907 * dev_base_lock to be sure it remains unique.
4909 static int dev_new_index(struct net *net)
4911 int ifindex = net->ifindex;
4915 if (!__dev_get_by_index(net, ifindex))
4916 return net->ifindex = ifindex;
4920 /* Delayed registration/unregisteration */
4921 static LIST_HEAD(net_todo_list);
4923 static void net_set_todo(struct net_device *dev)
4925 list_add_tail(&dev->todo_list, &net_todo_list);
4928 static void rollback_registered_many(struct list_head *head)
4930 struct net_device *dev, *tmp;
4932 BUG_ON(dev_boot_phase);
4935 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4936 /* Some devices call without registering
4937 * for initialization unwind. Remove those
4938 * devices and proceed with the remaining.
4940 if (dev->reg_state == NETREG_UNINITIALIZED) {
4941 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
4945 list_del(&dev->unreg_list);
4948 dev->dismantle = true;
4949 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4952 /* If device is running, close it first. */
4953 dev_close_many(head);
4955 list_for_each_entry(dev, head, unreg_list) {
4956 /* And unlink it from device chain. */
4957 unlist_netdevice(dev);
4959 dev->reg_state = NETREG_UNREGISTERING;
4964 list_for_each_entry(dev, head, unreg_list) {
4965 /* Shutdown queueing discipline. */
4969 /* Notify protocols, that we are about to destroy
4970 this device. They should clean all the things.
4972 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4974 if (!dev->rtnl_link_ops ||
4975 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4976 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4979 * Flush the unicast and multicast chains
4984 if (dev->netdev_ops->ndo_uninit)
4985 dev->netdev_ops->ndo_uninit(dev);
4987 /* Notifier chain MUST detach us all upper devices. */
4988 WARN_ON(netdev_has_any_upper_dev(dev));
4990 /* Remove entries from kobject tree */
4991 netdev_unregister_kobject(dev);
4993 /* Remove XPS queueing entries */
4994 netif_reset_xps_queues_gt(dev, 0);
5000 list_for_each_entry(dev, head, unreg_list)
5004 static void rollback_registered(struct net_device *dev)
5008 list_add(&dev->unreg_list, &single);
5009 rollback_registered_many(&single);
5013 static netdev_features_t netdev_fix_features(struct net_device *dev,
5014 netdev_features_t features)
5016 /* Fix illegal checksum combinations */
5017 if ((features & NETIF_F_HW_CSUM) &&
5018 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5019 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5020 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5023 /* TSO requires that SG is present as well. */
5024 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5025 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5026 features &= ~NETIF_F_ALL_TSO;
5029 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5030 !(features & NETIF_F_IP_CSUM)) {
5031 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5032 features &= ~NETIF_F_TSO;
5033 features &= ~NETIF_F_TSO_ECN;
5036 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5037 !(features & NETIF_F_IPV6_CSUM)) {
5038 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5039 features &= ~NETIF_F_TSO6;
5042 /* TSO ECN requires that TSO is present as well. */
5043 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5044 features &= ~NETIF_F_TSO_ECN;
5046 /* Software GSO depends on SG. */
5047 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5048 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5049 features &= ~NETIF_F_GSO;
5052 /* UFO needs SG and checksumming */
5053 if (features & NETIF_F_UFO) {
5054 /* maybe split UFO into V4 and V6? */
5055 if (!((features & NETIF_F_GEN_CSUM) ||
5056 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5057 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5059 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5060 features &= ~NETIF_F_UFO;
5063 if (!(features & NETIF_F_SG)) {
5065 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5066 features &= ~NETIF_F_UFO;
5073 int __netdev_update_features(struct net_device *dev)
5075 netdev_features_t features;
5080 features = netdev_get_wanted_features(dev);
5082 if (dev->netdev_ops->ndo_fix_features)
5083 features = dev->netdev_ops->ndo_fix_features(dev, features);
5085 /* driver might be less strict about feature dependencies */
5086 features = netdev_fix_features(dev, features);
5088 if (dev->features == features)
5091 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5092 &dev->features, &features);
5094 if (dev->netdev_ops->ndo_set_features)
5095 err = dev->netdev_ops->ndo_set_features(dev, features);
5097 if (unlikely(err < 0)) {
5099 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5100 err, &features, &dev->features);
5105 dev->features = features;
5111 * netdev_update_features - recalculate device features
5112 * @dev: the device to check
5114 * Recalculate dev->features set and send notifications if it
5115 * has changed. Should be called after driver or hardware dependent
5116 * conditions might have changed that influence the features.
5118 void netdev_update_features(struct net_device *dev)
5120 if (__netdev_update_features(dev))
5121 netdev_features_change(dev);
5123 EXPORT_SYMBOL(netdev_update_features);
5126 * netdev_change_features - recalculate device features
5127 * @dev: the device to check
5129 * Recalculate dev->features set and send notifications even
5130 * if they have not changed. Should be called instead of
5131 * netdev_update_features() if also dev->vlan_features might
5132 * have changed to allow the changes to be propagated to stacked
5135 void netdev_change_features(struct net_device *dev)
5137 __netdev_update_features(dev);
5138 netdev_features_change(dev);
5140 EXPORT_SYMBOL(netdev_change_features);
5143 * netif_stacked_transfer_operstate - transfer operstate
5144 * @rootdev: the root or lower level device to transfer state from
5145 * @dev: the device to transfer operstate to
5147 * Transfer operational state from root to device. This is normally
5148 * called when a stacking relationship exists between the root
5149 * device and the device(a leaf device).
5151 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5152 struct net_device *dev)
5154 if (rootdev->operstate == IF_OPER_DORMANT)
5155 netif_dormant_on(dev);
5157 netif_dormant_off(dev);
5159 if (netif_carrier_ok(rootdev)) {
5160 if (!netif_carrier_ok(dev))
5161 netif_carrier_on(dev);
5163 if (netif_carrier_ok(dev))
5164 netif_carrier_off(dev);
5167 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5170 static int netif_alloc_rx_queues(struct net_device *dev)
5172 unsigned int i, count = dev->num_rx_queues;
5173 struct netdev_rx_queue *rx;
5177 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5183 for (i = 0; i < count; i++)
5189 static void netdev_init_one_queue(struct net_device *dev,
5190 struct netdev_queue *queue, void *_unused)
5192 /* Initialize queue lock */
5193 spin_lock_init(&queue->_xmit_lock);
5194 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5195 queue->xmit_lock_owner = -1;
5196 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5199 dql_init(&queue->dql, HZ);
5203 static int netif_alloc_netdev_queues(struct net_device *dev)
5205 unsigned int count = dev->num_tx_queues;
5206 struct netdev_queue *tx;
5210 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5216 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5217 spin_lock_init(&dev->tx_global_lock);
5223 * register_netdevice - register a network device
5224 * @dev: device to register
5226 * Take a completed network device structure and add it to the kernel
5227 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5228 * chain. 0 is returned on success. A negative errno code is returned
5229 * on a failure to set up the device, or if the name is a duplicate.
5231 * Callers must hold the rtnl semaphore. You may want
5232 * register_netdev() instead of this.
5235 * The locking appears insufficient to guarantee two parallel registers
5236 * will not get the same name.
5239 int register_netdevice(struct net_device *dev)
5242 struct net *net = dev_net(dev);
5244 BUG_ON(dev_boot_phase);
5249 /* When net_device's are persistent, this will be fatal. */
5250 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5253 spin_lock_init(&dev->addr_list_lock);
5254 netdev_set_addr_lockdep_class(dev);
5258 ret = dev_get_valid_name(net, dev, dev->name);
5262 /* Init, if this function is available */
5263 if (dev->netdev_ops->ndo_init) {
5264 ret = dev->netdev_ops->ndo_init(dev);
5272 if (((dev->hw_features | dev->features) &
5273 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5274 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5275 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5276 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5283 dev->ifindex = dev_new_index(net);
5284 else if (__dev_get_by_index(net, dev->ifindex))
5287 if (dev->iflink == -1)
5288 dev->iflink = dev->ifindex;
5290 /* Transfer changeable features to wanted_features and enable
5291 * software offloads (GSO and GRO).
5293 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5294 dev->features |= NETIF_F_SOFT_FEATURES;
5295 dev->wanted_features = dev->features & dev->hw_features;
5297 /* Turn on no cache copy if HW is doing checksum */
5298 if (!(dev->flags & IFF_LOOPBACK)) {
5299 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5300 if (dev->features & NETIF_F_ALL_CSUM) {
5301 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5302 dev->features |= NETIF_F_NOCACHE_COPY;
5306 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5308 dev->vlan_features |= NETIF_F_HIGHDMA;
5310 /* Make NETIF_F_SG inheritable to tunnel devices.
5312 dev->hw_enc_features |= NETIF_F_SG;
5314 /* Make NETIF_F_SG inheritable to MPLS.
5316 dev->mpls_features |= NETIF_F_SG;
5318 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5319 ret = notifier_to_errno(ret);
5323 ret = netdev_register_kobject(dev);
5326 dev->reg_state = NETREG_REGISTERED;
5328 __netdev_update_features(dev);
5331 * Default initial state at registry is that the
5332 * device is present.
5335 set_bit(__LINK_STATE_PRESENT, &dev->state);
5337 linkwatch_init_dev(dev);
5339 dev_init_scheduler(dev);
5341 list_netdevice(dev);
5342 add_device_randomness(dev->dev_addr, dev->addr_len);
5344 /* If the device has permanent device address, driver should
5345 * set dev_addr and also addr_assign_type should be set to
5346 * NET_ADDR_PERM (default value).
5348 if (dev->addr_assign_type == NET_ADDR_PERM)
5349 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5351 /* Notify protocols, that a new device appeared. */
5352 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5353 ret = notifier_to_errno(ret);
5355 rollback_registered(dev);
5356 dev->reg_state = NETREG_UNREGISTERED;
5359 * Prevent userspace races by waiting until the network
5360 * device is fully setup before sending notifications.
5362 if (!dev->rtnl_link_ops ||
5363 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5364 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5370 if (dev->netdev_ops->ndo_uninit)
5371 dev->netdev_ops->ndo_uninit(dev);
5374 EXPORT_SYMBOL(register_netdevice);
5377 * init_dummy_netdev - init a dummy network device for NAPI
5378 * @dev: device to init
5380 * This takes a network device structure and initialize the minimum
5381 * amount of fields so it can be used to schedule NAPI polls without
5382 * registering a full blown interface. This is to be used by drivers
5383 * that need to tie several hardware interfaces to a single NAPI
5384 * poll scheduler due to HW limitations.
5386 int init_dummy_netdev(struct net_device *dev)
5388 /* Clear everything. Note we don't initialize spinlocks
5389 * are they aren't supposed to be taken by any of the
5390 * NAPI code and this dummy netdev is supposed to be
5391 * only ever used for NAPI polls
5393 memset(dev, 0, sizeof(struct net_device));
5395 /* make sure we BUG if trying to hit standard
5396 * register/unregister code path
5398 dev->reg_state = NETREG_DUMMY;
5400 /* NAPI wants this */
5401 INIT_LIST_HEAD(&dev->napi_list);
5403 /* a dummy interface is started by default */
5404 set_bit(__LINK_STATE_PRESENT, &dev->state);
5405 set_bit(__LINK_STATE_START, &dev->state);
5407 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5408 * because users of this 'device' dont need to change
5414 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5418 * register_netdev - register a network device
5419 * @dev: device to register
5421 * Take a completed network device structure and add it to the kernel
5422 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5423 * chain. 0 is returned on success. A negative errno code is returned
5424 * on a failure to set up the device, or if the name is a duplicate.
5426 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5427 * and expands the device name if you passed a format string to
5430 int register_netdev(struct net_device *dev)
5435 err = register_netdevice(dev);
5439 EXPORT_SYMBOL(register_netdev);
5441 int netdev_refcnt_read(const struct net_device *dev)
5445 for_each_possible_cpu(i)
5446 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5449 EXPORT_SYMBOL(netdev_refcnt_read);
5452 * netdev_wait_allrefs - wait until all references are gone.
5453 * @dev: target net_device
5455 * This is called when unregistering network devices.
5457 * Any protocol or device that holds a reference should register
5458 * for netdevice notification, and cleanup and put back the
5459 * reference if they receive an UNREGISTER event.
5460 * We can get stuck here if buggy protocols don't correctly
5463 static void netdev_wait_allrefs(struct net_device *dev)
5465 unsigned long rebroadcast_time, warning_time;
5468 linkwatch_forget_dev(dev);
5470 rebroadcast_time = warning_time = jiffies;
5471 refcnt = netdev_refcnt_read(dev);
5473 while (refcnt != 0) {
5474 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5477 /* Rebroadcast unregister notification */
5478 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5484 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5485 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5487 /* We must not have linkwatch events
5488 * pending on unregister. If this
5489 * happens, we simply run the queue
5490 * unscheduled, resulting in a noop
5493 linkwatch_run_queue();
5498 rebroadcast_time = jiffies;
5503 refcnt = netdev_refcnt_read(dev);
5505 if (time_after(jiffies, warning_time + 10 * HZ)) {
5506 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5508 warning_time = jiffies;
5517 * register_netdevice(x1);
5518 * register_netdevice(x2);
5520 * unregister_netdevice(y1);
5521 * unregister_netdevice(y2);
5527 * We are invoked by rtnl_unlock().
5528 * This allows us to deal with problems:
5529 * 1) We can delete sysfs objects which invoke hotplug
5530 * without deadlocking with linkwatch via keventd.
5531 * 2) Since we run with the RTNL semaphore not held, we can sleep
5532 * safely in order to wait for the netdev refcnt to drop to zero.
5534 * We must not return until all unregister events added during
5535 * the interval the lock was held have been completed.
5537 void netdev_run_todo(void)
5539 struct list_head list;
5541 /* Snapshot list, allow later requests */
5542 list_replace_init(&net_todo_list, &list);
5547 /* Wait for rcu callbacks to finish before next phase */
5548 if (!list_empty(&list))
5551 while (!list_empty(&list)) {
5552 struct net_device *dev
5553 = list_first_entry(&list, struct net_device, todo_list);
5554 list_del(&dev->todo_list);
5557 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5560 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5561 pr_err("network todo '%s' but state %d\n",
5562 dev->name, dev->reg_state);
5567 dev->reg_state = NETREG_UNREGISTERED;
5569 on_each_cpu(flush_backlog, dev, 1);
5571 netdev_wait_allrefs(dev);
5574 BUG_ON(netdev_refcnt_read(dev));
5575 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5576 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5577 WARN_ON(dev->dn_ptr);
5579 if (dev->destructor)
5580 dev->destructor(dev);
5582 /* Free network device */
5583 kobject_put(&dev->dev.kobj);
5587 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5588 * fields in the same order, with only the type differing.
5590 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5591 const struct net_device_stats *netdev_stats)
5593 #if BITS_PER_LONG == 64
5594 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5595 memcpy(stats64, netdev_stats, sizeof(*stats64));
5597 size_t i, n = sizeof(*stats64) / sizeof(u64);
5598 const unsigned long *src = (const unsigned long *)netdev_stats;
5599 u64 *dst = (u64 *)stats64;
5601 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5602 sizeof(*stats64) / sizeof(u64));
5603 for (i = 0; i < n; i++)
5607 EXPORT_SYMBOL(netdev_stats_to_stats64);
5610 * dev_get_stats - get network device statistics
5611 * @dev: device to get statistics from
5612 * @storage: place to store stats
5614 * Get network statistics from device. Return @storage.
5615 * The device driver may provide its own method by setting
5616 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5617 * otherwise the internal statistics structure is used.
5619 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5620 struct rtnl_link_stats64 *storage)
5622 const struct net_device_ops *ops = dev->netdev_ops;
5624 if (ops->ndo_get_stats64) {
5625 memset(storage, 0, sizeof(*storage));
5626 ops->ndo_get_stats64(dev, storage);
5627 } else if (ops->ndo_get_stats) {
5628 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5630 netdev_stats_to_stats64(storage, &dev->stats);
5632 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5635 EXPORT_SYMBOL(dev_get_stats);
5637 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5639 struct netdev_queue *queue = dev_ingress_queue(dev);
5641 #ifdef CONFIG_NET_CLS_ACT
5644 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5647 netdev_init_one_queue(dev, queue, NULL);
5648 queue->qdisc = &noop_qdisc;
5649 queue->qdisc_sleeping = &noop_qdisc;
5650 rcu_assign_pointer(dev->ingress_queue, queue);
5655 static const struct ethtool_ops default_ethtool_ops;
5657 void netdev_set_default_ethtool_ops(struct net_device *dev,
5658 const struct ethtool_ops *ops)
5660 if (dev->ethtool_ops == &default_ethtool_ops)
5661 dev->ethtool_ops = ops;
5663 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
5666 * alloc_netdev_mqs - allocate network device
5667 * @sizeof_priv: size of private data to allocate space for
5668 * @name: device name format string
5669 * @setup: callback to initialize device
5670 * @txqs: the number of TX subqueues to allocate
5671 * @rxqs: the number of RX subqueues to allocate
5673 * Allocates a struct net_device with private data area for driver use
5674 * and performs basic initialization. Also allocates subquue structs
5675 * for each queue on the device.
5677 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5678 void (*setup)(struct net_device *),
5679 unsigned int txqs, unsigned int rxqs)
5681 struct net_device *dev;
5683 struct net_device *p;
5685 BUG_ON(strlen(name) >= sizeof(dev->name));
5688 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5694 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5699 alloc_size = sizeof(struct net_device);
5701 /* ensure 32-byte alignment of private area */
5702 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5703 alloc_size += sizeof_priv;
5705 /* ensure 32-byte alignment of whole construct */
5706 alloc_size += NETDEV_ALIGN - 1;
5708 p = kzalloc(alloc_size, GFP_KERNEL);
5712 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5713 dev->padded = (char *)dev - (char *)p;
5715 dev->pcpu_refcnt = alloc_percpu(int);
5716 if (!dev->pcpu_refcnt)
5719 if (dev_addr_init(dev))
5725 dev_net_set(dev, &init_net);
5727 dev->gso_max_size = GSO_MAX_SIZE;
5728 dev->gso_max_segs = GSO_MAX_SEGS;
5730 INIT_LIST_HEAD(&dev->napi_list);
5731 INIT_LIST_HEAD(&dev->unreg_list);
5732 INIT_LIST_HEAD(&dev->link_watch_list);
5733 INIT_LIST_HEAD(&dev->upper_dev_list);
5734 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5737 dev->num_tx_queues = txqs;
5738 dev->real_num_tx_queues = txqs;
5739 if (netif_alloc_netdev_queues(dev))
5743 dev->num_rx_queues = rxqs;
5744 dev->real_num_rx_queues = rxqs;
5745 if (netif_alloc_rx_queues(dev))
5749 strcpy(dev->name, name);
5750 dev->group = INIT_NETDEV_GROUP;
5751 if (!dev->ethtool_ops)
5752 dev->ethtool_ops = &default_ethtool_ops;
5760 free_percpu(dev->pcpu_refcnt);
5770 EXPORT_SYMBOL(alloc_netdev_mqs);
5773 * free_netdev - free network device
5776 * This function does the last stage of destroying an allocated device
5777 * interface. The reference to the device object is released.
5778 * If this is the last reference then it will be freed.
5780 void free_netdev(struct net_device *dev)
5782 struct napi_struct *p, *n;
5784 release_net(dev_net(dev));
5791 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
5793 /* Flush device addresses */
5794 dev_addr_flush(dev);
5796 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5799 free_percpu(dev->pcpu_refcnt);
5800 dev->pcpu_refcnt = NULL;
5802 /* Compatibility with error handling in drivers */
5803 if (dev->reg_state == NETREG_UNINITIALIZED) {
5804 kfree((char *)dev - dev->padded);
5808 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5809 dev->reg_state = NETREG_RELEASED;
5811 /* will free via device release */
5812 put_device(&dev->dev);
5814 EXPORT_SYMBOL(free_netdev);
5817 * synchronize_net - Synchronize with packet receive processing
5819 * Wait for packets currently being received to be done.
5820 * Does not block later packets from starting.
5822 void synchronize_net(void)
5825 if (rtnl_is_locked())
5826 synchronize_rcu_expedited();
5830 EXPORT_SYMBOL(synchronize_net);
5833 * unregister_netdevice_queue - remove device from the kernel
5837 * This function shuts down a device interface and removes it
5838 * from the kernel tables.
5839 * If head not NULL, device is queued to be unregistered later.
5841 * Callers must hold the rtnl semaphore. You may want
5842 * unregister_netdev() instead of this.
5845 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5850 list_move_tail(&dev->unreg_list, head);
5852 rollback_registered(dev);
5853 /* Finish processing unregister after unlock */
5857 EXPORT_SYMBOL(unregister_netdevice_queue);
5860 * unregister_netdevice_many - unregister many devices
5861 * @head: list of devices
5863 void unregister_netdevice_many(struct list_head *head)
5865 struct net_device *dev;
5867 if (!list_empty(head)) {
5868 rollback_registered_many(head);
5869 list_for_each_entry(dev, head, unreg_list)
5873 EXPORT_SYMBOL(unregister_netdevice_many);
5876 * unregister_netdev - remove device from the kernel
5879 * This function shuts down a device interface and removes it
5880 * from the kernel tables.
5882 * This is just a wrapper for unregister_netdevice that takes
5883 * the rtnl semaphore. In general you want to use this and not
5884 * unregister_netdevice.
5886 void unregister_netdev(struct net_device *dev)
5889 unregister_netdevice(dev);
5892 EXPORT_SYMBOL(unregister_netdev);
5895 * dev_change_net_namespace - move device to different nethost namespace
5897 * @net: network namespace
5898 * @pat: If not NULL name pattern to try if the current device name
5899 * is already taken in the destination network namespace.
5901 * This function shuts down a device interface and moves it
5902 * to a new network namespace. On success 0 is returned, on
5903 * a failure a netagive errno code is returned.
5905 * Callers must hold the rtnl semaphore.
5908 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5914 /* Don't allow namespace local devices to be moved. */
5916 if (dev->features & NETIF_F_NETNS_LOCAL)
5919 /* Ensure the device has been registrered */
5920 if (dev->reg_state != NETREG_REGISTERED)
5923 /* Get out if there is nothing todo */
5925 if (net_eq(dev_net(dev), net))
5928 /* Pick the destination device name, and ensure
5929 * we can use it in the destination network namespace.
5932 if (__dev_get_by_name(net, dev->name)) {
5933 /* We get here if we can't use the current device name */
5936 if (dev_get_valid_name(net, dev, pat) < 0)
5941 * And now a mini version of register_netdevice unregister_netdevice.
5944 /* If device is running close it first. */
5947 /* And unlink it from device chain */
5949 unlist_netdevice(dev);
5953 /* Shutdown queueing discipline. */
5956 /* Notify protocols, that we are about to destroy
5957 this device. They should clean all the things.
5959 Note that dev->reg_state stays at NETREG_REGISTERED.
5960 This is wanted because this way 8021q and macvlan know
5961 the device is just moving and can keep their slaves up.
5963 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5965 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5966 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5969 * Flush the unicast and multicast chains
5974 /* Send a netdev-removed uevent to the old namespace */
5975 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
5977 /* Actually switch the network namespace */
5978 dev_net_set(dev, net);
5980 /* If there is an ifindex conflict assign a new one */
5981 if (__dev_get_by_index(net, dev->ifindex)) {
5982 int iflink = (dev->iflink == dev->ifindex);
5983 dev->ifindex = dev_new_index(net);
5985 dev->iflink = dev->ifindex;
5988 /* Send a netdev-add uevent to the new namespace */
5989 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
5991 /* Fixup kobjects */
5992 err = device_rename(&dev->dev, dev->name);
5995 /* Add the device back in the hashes */
5996 list_netdevice(dev);
5998 /* Notify protocols, that a new device appeared. */
5999 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6002 * Prevent userspace races by waiting until the network
6003 * device is fully setup before sending notifications.
6005 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6012 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6014 static int dev_cpu_callback(struct notifier_block *nfb,
6015 unsigned long action,
6018 struct sk_buff **list_skb;
6019 struct sk_buff *skb;
6020 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6021 struct softnet_data *sd, *oldsd;
6023 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6026 local_irq_disable();
6027 cpu = smp_processor_id();
6028 sd = &per_cpu(softnet_data, cpu);
6029 oldsd = &per_cpu(softnet_data, oldcpu);
6031 /* Find end of our completion_queue. */
6032 list_skb = &sd->completion_queue;
6034 list_skb = &(*list_skb)->next;
6035 /* Append completion queue from offline CPU. */
6036 *list_skb = oldsd->completion_queue;
6037 oldsd->completion_queue = NULL;
6039 /* Append output queue from offline CPU. */
6040 if (oldsd->output_queue) {
6041 *sd->output_queue_tailp = oldsd->output_queue;
6042 sd->output_queue_tailp = oldsd->output_queue_tailp;
6043 oldsd->output_queue = NULL;
6044 oldsd->output_queue_tailp = &oldsd->output_queue;
6046 /* Append NAPI poll list from offline CPU. */
6047 if (!list_empty(&oldsd->poll_list)) {
6048 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6049 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6052 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6055 /* Process offline CPU's input_pkt_queue */
6056 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6058 input_queue_head_incr(oldsd);
6060 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6062 input_queue_head_incr(oldsd);
6070 * netdev_increment_features - increment feature set by one
6071 * @all: current feature set
6072 * @one: new feature set
6073 * @mask: mask feature set
6075 * Computes a new feature set after adding a device with feature set
6076 * @one to the master device with current feature set @all. Will not
6077 * enable anything that is off in @mask. Returns the new feature set.
6079 netdev_features_t netdev_increment_features(netdev_features_t all,
6080 netdev_features_t one, netdev_features_t mask)
6082 if (mask & NETIF_F_GEN_CSUM)
6083 mask |= NETIF_F_ALL_CSUM;
6084 mask |= NETIF_F_VLAN_CHALLENGED;
6086 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6087 all &= one | ~NETIF_F_ALL_FOR_ALL;
6089 /* If one device supports hw checksumming, set for all. */
6090 if (all & NETIF_F_GEN_CSUM)
6091 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6095 EXPORT_SYMBOL(netdev_increment_features);
6097 static struct hlist_head *netdev_create_hash(void)
6100 struct hlist_head *hash;
6102 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6104 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6105 INIT_HLIST_HEAD(&hash[i]);
6110 /* Initialize per network namespace state */
6111 static int __net_init netdev_init(struct net *net)
6113 if (net != &init_net)
6114 INIT_LIST_HEAD(&net->dev_base_head);
6116 net->dev_name_head = netdev_create_hash();
6117 if (net->dev_name_head == NULL)
6120 net->dev_index_head = netdev_create_hash();
6121 if (net->dev_index_head == NULL)
6127 kfree(net->dev_name_head);
6133 * netdev_drivername - network driver for the device
6134 * @dev: network device
6136 * Determine network driver for device.
6138 const char *netdev_drivername(const struct net_device *dev)
6140 const struct device_driver *driver;
6141 const struct device *parent;
6142 const char *empty = "";
6144 parent = dev->dev.parent;
6148 driver = parent->driver;
6149 if (driver && driver->name)
6150 return driver->name;
6154 static int __netdev_printk(const char *level, const struct net_device *dev,
6155 struct va_format *vaf)
6159 if (dev && dev->dev.parent) {
6160 r = dev_printk_emit(level[1] - '0',
6163 dev_driver_string(dev->dev.parent),
6164 dev_name(dev->dev.parent),
6165 netdev_name(dev), vaf);
6167 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6169 r = printk("%s(NULL net_device): %pV", level, vaf);
6175 int netdev_printk(const char *level, const struct net_device *dev,
6176 const char *format, ...)
6178 struct va_format vaf;
6182 va_start(args, format);
6187 r = __netdev_printk(level, dev, &vaf);
6193 EXPORT_SYMBOL(netdev_printk);
6195 #define define_netdev_printk_level(func, level) \
6196 int func(const struct net_device *dev, const char *fmt, ...) \
6199 struct va_format vaf; \
6202 va_start(args, fmt); \
6207 r = __netdev_printk(level, dev, &vaf); \
6213 EXPORT_SYMBOL(func);
6215 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6216 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6217 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6218 define_netdev_printk_level(netdev_err, KERN_ERR);
6219 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6220 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6221 define_netdev_printk_level(netdev_info, KERN_INFO);
6223 static void __net_exit netdev_exit(struct net *net)
6225 kfree(net->dev_name_head);
6226 kfree(net->dev_index_head);
6229 static struct pernet_operations __net_initdata netdev_net_ops = {
6230 .init = netdev_init,
6231 .exit = netdev_exit,
6234 static void __net_exit default_device_exit(struct net *net)
6236 struct net_device *dev, *aux;
6238 * Push all migratable network devices back to the
6239 * initial network namespace
6242 for_each_netdev_safe(net, dev, aux) {
6244 char fb_name[IFNAMSIZ];
6246 /* Ignore unmoveable devices (i.e. loopback) */
6247 if (dev->features & NETIF_F_NETNS_LOCAL)
6250 /* Leave virtual devices for the generic cleanup */
6251 if (dev->rtnl_link_ops)
6254 /* Push remaining network devices to init_net */
6255 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6256 err = dev_change_net_namespace(dev, &init_net, fb_name);
6258 pr_emerg("%s: failed to move %s to init_net: %d\n",
6259 __func__, dev->name, err);
6266 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6268 /* At exit all network devices most be removed from a network
6269 * namespace. Do this in the reverse order of registration.
6270 * Do this across as many network namespaces as possible to
6271 * improve batching efficiency.
6273 struct net_device *dev;
6275 LIST_HEAD(dev_kill_list);
6278 list_for_each_entry(net, net_list, exit_list) {
6279 for_each_netdev_reverse(net, dev) {
6280 if (dev->rtnl_link_ops)
6281 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6283 unregister_netdevice_queue(dev, &dev_kill_list);
6286 unregister_netdevice_many(&dev_kill_list);
6287 list_del(&dev_kill_list);
6291 static struct pernet_operations __net_initdata default_device_ops = {
6292 .exit = default_device_exit,
6293 .exit_batch = default_device_exit_batch,
6297 * Initialize the DEV module. At boot time this walks the device list and
6298 * unhooks any devices that fail to initialise (normally hardware not
6299 * present) and leaves us with a valid list of present and active devices.
6304 * This is called single threaded during boot, so no need
6305 * to take the rtnl semaphore.
6307 static int __init net_dev_init(void)
6309 int i, rc = -ENOMEM;
6311 BUG_ON(!dev_boot_phase);
6313 if (dev_proc_init())
6316 if (netdev_kobject_init())
6319 INIT_LIST_HEAD(&ptype_all);
6320 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6321 INIT_LIST_HEAD(&ptype_base[i]);
6323 INIT_LIST_HEAD(&offload_base);
6325 if (register_pernet_subsys(&netdev_net_ops))
6329 * Initialise the packet receive queues.
6332 for_each_possible_cpu(i) {
6333 struct softnet_data *sd = &per_cpu(softnet_data, i);
6335 memset(sd, 0, sizeof(*sd));
6336 skb_queue_head_init(&sd->input_pkt_queue);
6337 skb_queue_head_init(&sd->process_queue);
6338 sd->completion_queue = NULL;
6339 INIT_LIST_HEAD(&sd->poll_list);
6340 sd->output_queue = NULL;
6341 sd->output_queue_tailp = &sd->output_queue;
6343 sd->csd.func = rps_trigger_softirq;
6349 sd->backlog.poll = process_backlog;
6350 sd->backlog.weight = weight_p;
6351 sd->backlog.gro_list = NULL;
6352 sd->backlog.gro_count = 0;
6354 #ifdef CONFIG_NET_FLOW_LIMIT
6355 sd->flow_limit = NULL;
6361 /* The loopback device is special if any other network devices
6362 * is present in a network namespace the loopback device must
6363 * be present. Since we now dynamically allocate and free the
6364 * loopback device ensure this invariant is maintained by
6365 * keeping the loopback device as the first device on the
6366 * list of network devices. Ensuring the loopback devices
6367 * is the first device that appears and the last network device
6370 if (register_pernet_device(&loopback_net_ops))
6373 if (register_pernet_device(&default_device_ops))
6376 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6377 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6379 hotcpu_notifier(dev_cpu_callback, 0);
6386 subsys_initcall(net_dev_init);