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);
1395 static int dev_boot_phase = 1;
1398 * register_netdevice_notifier - register a network notifier block
1401 * Register a notifier to be called when network device events occur.
1402 * The notifier passed is linked into the kernel structures and must
1403 * not be reused until it has been unregistered. A negative errno code
1404 * is returned on a failure.
1406 * When registered all registration and up events are replayed
1407 * to the new notifier to allow device to have a race free
1408 * view of the network device list.
1411 int register_netdevice_notifier(struct notifier_block *nb)
1413 struct net_device *dev;
1414 struct net_device *last;
1419 err = raw_notifier_chain_register(&netdev_chain, nb);
1425 for_each_netdev(net, dev) {
1426 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1427 err = notifier_to_errno(err);
1431 if (!(dev->flags & IFF_UP))
1434 nb->notifier_call(nb, NETDEV_UP, dev);
1445 for_each_netdev(net, dev) {
1449 if (dev->flags & IFF_UP) {
1450 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1451 nb->notifier_call(nb, NETDEV_DOWN, dev);
1453 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1458 raw_notifier_chain_unregister(&netdev_chain, nb);
1461 EXPORT_SYMBOL(register_netdevice_notifier);
1464 * unregister_netdevice_notifier - unregister a network notifier block
1467 * Unregister a notifier previously registered by
1468 * register_netdevice_notifier(). The notifier is unlinked into the
1469 * kernel structures and may then be reused. A negative errno code
1470 * is returned on a failure.
1472 * After unregistering unregister and down device events are synthesized
1473 * for all devices on the device list to the removed notifier to remove
1474 * the need for special case cleanup code.
1477 int unregister_netdevice_notifier(struct notifier_block *nb)
1479 struct net_device *dev;
1484 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1489 for_each_netdev(net, dev) {
1490 if (dev->flags & IFF_UP) {
1491 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1492 nb->notifier_call(nb, NETDEV_DOWN, dev);
1494 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1501 EXPORT_SYMBOL(unregister_netdevice_notifier);
1504 * call_netdevice_notifiers - call all network notifier blocks
1505 * @val: value passed unmodified to notifier function
1506 * @dev: net_device pointer passed unmodified to notifier function
1508 * Call all network notifier blocks. Parameters and return value
1509 * are as for raw_notifier_call_chain().
1512 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1515 return raw_notifier_call_chain(&netdev_chain, val, dev);
1517 EXPORT_SYMBOL(call_netdevice_notifiers);
1519 static struct static_key netstamp_needed __read_mostly;
1520 #ifdef HAVE_JUMP_LABEL
1521 /* We are not allowed to call static_key_slow_dec() from irq context
1522 * If net_disable_timestamp() is called from irq context, defer the
1523 * static_key_slow_dec() calls.
1525 static atomic_t netstamp_needed_deferred;
1528 void net_enable_timestamp(void)
1530 #ifdef HAVE_JUMP_LABEL
1531 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1535 static_key_slow_dec(&netstamp_needed);
1539 static_key_slow_inc(&netstamp_needed);
1541 EXPORT_SYMBOL(net_enable_timestamp);
1543 void net_disable_timestamp(void)
1545 #ifdef HAVE_JUMP_LABEL
1546 if (in_interrupt()) {
1547 atomic_inc(&netstamp_needed_deferred);
1551 static_key_slow_dec(&netstamp_needed);
1553 EXPORT_SYMBOL(net_disable_timestamp);
1555 static inline void net_timestamp_set(struct sk_buff *skb)
1557 skb->tstamp.tv64 = 0;
1558 if (static_key_false(&netstamp_needed))
1559 __net_timestamp(skb);
1562 #define net_timestamp_check(COND, SKB) \
1563 if (static_key_false(&netstamp_needed)) { \
1564 if ((COND) && !(SKB)->tstamp.tv64) \
1565 __net_timestamp(SKB); \
1568 static inline bool is_skb_forwardable(struct net_device *dev,
1569 struct sk_buff *skb)
1573 if (!(dev->flags & IFF_UP))
1576 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1577 if (skb->len <= len)
1580 /* if TSO is enabled, we don't care about the length as the packet
1581 * could be forwarded without being segmented before
1583 if (skb_is_gso(skb))
1590 * dev_forward_skb - loopback an skb to another netif
1592 * @dev: destination network device
1593 * @skb: buffer to forward
1596 * NET_RX_SUCCESS (no congestion)
1597 * NET_RX_DROP (packet was dropped, but freed)
1599 * dev_forward_skb can be used for injecting an skb from the
1600 * start_xmit function of one device into the receive queue
1601 * of another device.
1603 * The receiving device may be in another namespace, so
1604 * we have to clear all information in the skb that could
1605 * impact namespace isolation.
1607 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1609 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1610 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1611 atomic_long_inc(&dev->rx_dropped);
1619 if (unlikely(!is_skb_forwardable(dev, skb))) {
1620 atomic_long_inc(&dev->rx_dropped);
1626 skb->tstamp.tv64 = 0;
1627 skb->pkt_type = PACKET_HOST;
1628 skb->protocol = eth_type_trans(skb, dev);
1632 nf_reset_trace(skb);
1633 return netif_rx(skb);
1635 EXPORT_SYMBOL_GPL(dev_forward_skb);
1637 static inline int deliver_skb(struct sk_buff *skb,
1638 struct packet_type *pt_prev,
1639 struct net_device *orig_dev)
1641 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1643 atomic_inc(&skb->users);
1644 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1647 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1649 if (!ptype->af_packet_priv || !skb->sk)
1652 if (ptype->id_match)
1653 return ptype->id_match(ptype, skb->sk);
1654 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1661 * Support routine. Sends outgoing frames to any network
1662 * taps currently in use.
1665 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1667 struct packet_type *ptype;
1668 struct sk_buff *skb2 = NULL;
1669 struct packet_type *pt_prev = NULL;
1672 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1673 /* Never send packets back to the socket
1674 * they originated from - MvS (miquels@drinkel.ow.org)
1676 if ((ptype->dev == dev || !ptype->dev) &&
1677 (!skb_loop_sk(ptype, skb))) {
1679 deliver_skb(skb2, pt_prev, skb->dev);
1684 skb2 = skb_clone(skb, GFP_ATOMIC);
1688 net_timestamp_set(skb2);
1690 /* skb->nh should be correctly
1691 set by sender, so that the second statement is
1692 just protection against buggy protocols.
1694 skb_reset_mac_header(skb2);
1696 if (skb_network_header(skb2) < skb2->data ||
1697 skb2->network_header > skb2->tail) {
1698 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1699 ntohs(skb2->protocol),
1701 skb_reset_network_header(skb2);
1704 skb2->transport_header = skb2->network_header;
1705 skb2->pkt_type = PACKET_OUTGOING;
1710 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1715 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1716 * @dev: Network device
1717 * @txq: number of queues available
1719 * If real_num_tx_queues is changed the tc mappings may no longer be
1720 * valid. To resolve this verify the tc mapping remains valid and if
1721 * not NULL the mapping. With no priorities mapping to this
1722 * offset/count pair it will no longer be used. In the worst case TC0
1723 * is invalid nothing can be done so disable priority mappings. If is
1724 * expected that drivers will fix this mapping if they can before
1725 * calling netif_set_real_num_tx_queues.
1727 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1730 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1732 /* If TC0 is invalidated disable TC mapping */
1733 if (tc->offset + tc->count > txq) {
1734 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1739 /* Invalidated prio to tc mappings set to TC0 */
1740 for (i = 1; i < TC_BITMASK + 1; i++) {
1741 int q = netdev_get_prio_tc_map(dev, i);
1743 tc = &dev->tc_to_txq[q];
1744 if (tc->offset + tc->count > txq) {
1745 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1747 netdev_set_prio_tc_map(dev, i, 0);
1753 static DEFINE_MUTEX(xps_map_mutex);
1754 #define xmap_dereference(P) \
1755 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1757 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1760 struct xps_map *map = NULL;
1764 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1766 for (pos = 0; map && pos < map->len; pos++) {
1767 if (map->queues[pos] == index) {
1769 map->queues[pos] = map->queues[--map->len];
1771 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1772 kfree_rcu(map, rcu);
1782 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1784 struct xps_dev_maps *dev_maps;
1786 bool active = false;
1788 mutex_lock(&xps_map_mutex);
1789 dev_maps = xmap_dereference(dev->xps_maps);
1794 for_each_possible_cpu(cpu) {
1795 for (i = index; i < dev->num_tx_queues; i++) {
1796 if (!remove_xps_queue(dev_maps, cpu, i))
1799 if (i == dev->num_tx_queues)
1804 RCU_INIT_POINTER(dev->xps_maps, NULL);
1805 kfree_rcu(dev_maps, rcu);
1808 for (i = index; i < dev->num_tx_queues; i++)
1809 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1813 mutex_unlock(&xps_map_mutex);
1816 static struct xps_map *expand_xps_map(struct xps_map *map,
1819 struct xps_map *new_map;
1820 int alloc_len = XPS_MIN_MAP_ALLOC;
1823 for (pos = 0; map && pos < map->len; pos++) {
1824 if (map->queues[pos] != index)
1829 /* Need to add queue to this CPU's existing map */
1831 if (pos < map->alloc_len)
1834 alloc_len = map->alloc_len * 2;
1837 /* Need to allocate new map to store queue on this CPU's map */
1838 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1843 for (i = 0; i < pos; i++)
1844 new_map->queues[i] = map->queues[i];
1845 new_map->alloc_len = alloc_len;
1851 int netif_set_xps_queue(struct net_device *dev, struct cpumask *mask, u16 index)
1853 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1854 struct xps_map *map, *new_map;
1855 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1856 int cpu, numa_node_id = -2;
1857 bool active = false;
1859 mutex_lock(&xps_map_mutex);
1861 dev_maps = xmap_dereference(dev->xps_maps);
1863 /* allocate memory for queue storage */
1864 for_each_online_cpu(cpu) {
1865 if (!cpumask_test_cpu(cpu, mask))
1869 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1870 if (!new_dev_maps) {
1871 mutex_unlock(&xps_map_mutex);
1875 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1878 map = expand_xps_map(map, cpu, index);
1882 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1886 goto out_no_new_maps;
1888 for_each_possible_cpu(cpu) {
1889 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1890 /* add queue to CPU maps */
1893 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1894 while ((pos < map->len) && (map->queues[pos] != index))
1897 if (pos == map->len)
1898 map->queues[map->len++] = index;
1900 if (numa_node_id == -2)
1901 numa_node_id = cpu_to_node(cpu);
1902 else if (numa_node_id != cpu_to_node(cpu))
1905 } else if (dev_maps) {
1906 /* fill in the new device map from the old device map */
1907 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1908 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1913 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1915 /* Cleanup old maps */
1917 for_each_possible_cpu(cpu) {
1918 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1919 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1920 if (map && map != new_map)
1921 kfree_rcu(map, rcu);
1924 kfree_rcu(dev_maps, rcu);
1927 dev_maps = new_dev_maps;
1931 /* update Tx queue numa node */
1932 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
1933 (numa_node_id >= 0) ? numa_node_id :
1939 /* removes queue from unused CPUs */
1940 for_each_possible_cpu(cpu) {
1941 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
1944 if (remove_xps_queue(dev_maps, cpu, index))
1948 /* free map if not active */
1950 RCU_INIT_POINTER(dev->xps_maps, NULL);
1951 kfree_rcu(dev_maps, rcu);
1955 mutex_unlock(&xps_map_mutex);
1959 /* remove any maps that we added */
1960 for_each_possible_cpu(cpu) {
1961 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1962 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1964 if (new_map && new_map != map)
1968 mutex_unlock(&xps_map_mutex);
1970 kfree(new_dev_maps);
1973 EXPORT_SYMBOL(netif_set_xps_queue);
1977 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1978 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1980 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1984 if (txq < 1 || txq > dev->num_tx_queues)
1987 if (dev->reg_state == NETREG_REGISTERED ||
1988 dev->reg_state == NETREG_UNREGISTERING) {
1991 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1997 netif_setup_tc(dev, txq);
1999 if (txq < dev->real_num_tx_queues) {
2000 qdisc_reset_all_tx_gt(dev, txq);
2002 netif_reset_xps_queues_gt(dev, txq);
2007 dev->real_num_tx_queues = txq;
2010 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2014 * netif_set_real_num_rx_queues - set actual number of RX queues used
2015 * @dev: Network device
2016 * @rxq: Actual number of RX queues
2018 * This must be called either with the rtnl_lock held or before
2019 * registration of the net device. Returns 0 on success, or a
2020 * negative error code. If called before registration, it always
2023 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2027 if (rxq < 1 || rxq > dev->num_rx_queues)
2030 if (dev->reg_state == NETREG_REGISTERED) {
2033 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2039 dev->real_num_rx_queues = rxq;
2042 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2046 * netif_get_num_default_rss_queues - default number of RSS queues
2048 * This routine should set an upper limit on the number of RSS queues
2049 * used by default by multiqueue devices.
2051 int netif_get_num_default_rss_queues(void)
2053 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2055 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2057 static inline void __netif_reschedule(struct Qdisc *q)
2059 struct softnet_data *sd;
2060 unsigned long flags;
2062 local_irq_save(flags);
2063 sd = &__get_cpu_var(softnet_data);
2064 q->next_sched = NULL;
2065 *sd->output_queue_tailp = q;
2066 sd->output_queue_tailp = &q->next_sched;
2067 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2068 local_irq_restore(flags);
2071 void __netif_schedule(struct Qdisc *q)
2073 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2074 __netif_reschedule(q);
2076 EXPORT_SYMBOL(__netif_schedule);
2078 void dev_kfree_skb_irq(struct sk_buff *skb)
2080 if (atomic_dec_and_test(&skb->users)) {
2081 struct softnet_data *sd;
2082 unsigned long flags;
2084 local_irq_save(flags);
2085 sd = &__get_cpu_var(softnet_data);
2086 skb->next = sd->completion_queue;
2087 sd->completion_queue = skb;
2088 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2089 local_irq_restore(flags);
2092 EXPORT_SYMBOL(dev_kfree_skb_irq);
2094 void dev_kfree_skb_any(struct sk_buff *skb)
2096 if (in_irq() || irqs_disabled())
2097 dev_kfree_skb_irq(skb);
2101 EXPORT_SYMBOL(dev_kfree_skb_any);
2105 * netif_device_detach - mark device as removed
2106 * @dev: network device
2108 * Mark device as removed from system and therefore no longer available.
2110 void netif_device_detach(struct net_device *dev)
2112 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2113 netif_running(dev)) {
2114 netif_tx_stop_all_queues(dev);
2117 EXPORT_SYMBOL(netif_device_detach);
2120 * netif_device_attach - mark device as attached
2121 * @dev: network device
2123 * Mark device as attached from system and restart if needed.
2125 void netif_device_attach(struct net_device *dev)
2127 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2128 netif_running(dev)) {
2129 netif_tx_wake_all_queues(dev);
2130 __netdev_watchdog_up(dev);
2133 EXPORT_SYMBOL(netif_device_attach);
2135 static void skb_warn_bad_offload(const struct sk_buff *skb)
2137 static const netdev_features_t null_features = 0;
2138 struct net_device *dev = skb->dev;
2139 const char *driver = "";
2141 if (!net_ratelimit())
2144 if (dev && dev->dev.parent)
2145 driver = dev_driver_string(dev->dev.parent);
2147 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2148 "gso_type=%d ip_summed=%d\n",
2149 driver, dev ? &dev->features : &null_features,
2150 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2151 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2152 skb_shinfo(skb)->gso_type, skb->ip_summed);
2156 * Invalidate hardware checksum when packet is to be mangled, and
2157 * complete checksum manually on outgoing path.
2159 int skb_checksum_help(struct sk_buff *skb)
2162 int ret = 0, offset;
2164 if (skb->ip_summed == CHECKSUM_COMPLETE)
2165 goto out_set_summed;
2167 if (unlikely(skb_shinfo(skb)->gso_size)) {
2168 skb_warn_bad_offload(skb);
2172 /* Before computing a checksum, we should make sure no frag could
2173 * be modified by an external entity : checksum could be wrong.
2175 if (skb_has_shared_frag(skb)) {
2176 ret = __skb_linearize(skb);
2181 offset = skb_checksum_start_offset(skb);
2182 BUG_ON(offset >= skb_headlen(skb));
2183 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2185 offset += skb->csum_offset;
2186 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2188 if (skb_cloned(skb) &&
2189 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2190 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2195 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2197 skb->ip_summed = CHECKSUM_NONE;
2201 EXPORT_SYMBOL(skb_checksum_help);
2203 __be16 skb_network_protocol(struct sk_buff *skb)
2205 __be16 type = skb->protocol;
2206 int vlan_depth = ETH_HLEN;
2208 /* Tunnel gso handlers can set protocol to ethernet. */
2209 if (type == htons(ETH_P_TEB)) {
2212 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2215 eth = (struct ethhdr *)skb_mac_header(skb);
2216 type = eth->h_proto;
2219 while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2220 struct vlan_hdr *vh;
2222 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2225 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2226 type = vh->h_vlan_encapsulated_proto;
2227 vlan_depth += VLAN_HLEN;
2234 * skb_mac_gso_segment - mac layer segmentation handler.
2235 * @skb: buffer to segment
2236 * @features: features for the output path (see dev->features)
2238 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2239 netdev_features_t features)
2241 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2242 struct packet_offload *ptype;
2243 __be16 type = skb_network_protocol(skb);
2245 if (unlikely(!type))
2246 return ERR_PTR(-EINVAL);
2248 __skb_pull(skb, skb->mac_len);
2251 list_for_each_entry_rcu(ptype, &offload_base, list) {
2252 if (ptype->type == type && ptype->callbacks.gso_segment) {
2253 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2256 err = ptype->callbacks.gso_send_check(skb);
2257 segs = ERR_PTR(err);
2258 if (err || skb_gso_ok(skb, features))
2260 __skb_push(skb, (skb->data -
2261 skb_network_header(skb)));
2263 segs = ptype->callbacks.gso_segment(skb, features);
2269 __skb_push(skb, skb->data - skb_mac_header(skb));
2273 EXPORT_SYMBOL(skb_mac_gso_segment);
2276 /* openvswitch calls this on rx path, so we need a different check.
2278 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2281 return skb->ip_summed != CHECKSUM_PARTIAL;
2283 return skb->ip_summed == CHECKSUM_NONE;
2287 * __skb_gso_segment - Perform segmentation on skb.
2288 * @skb: buffer to segment
2289 * @features: features for the output path (see dev->features)
2290 * @tx_path: whether it is called in TX path
2292 * This function segments the given skb and returns a list of segments.
2294 * It may return NULL if the skb requires no segmentation. This is
2295 * only possible when GSO is used for verifying header integrity.
2297 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2298 netdev_features_t features, bool tx_path)
2300 if (unlikely(skb_needs_check(skb, tx_path))) {
2303 skb_warn_bad_offload(skb);
2305 if (skb_header_cloned(skb) &&
2306 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2307 return ERR_PTR(err);
2310 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2311 skb_reset_mac_header(skb);
2312 skb_reset_mac_len(skb);
2314 return skb_mac_gso_segment(skb, features);
2316 EXPORT_SYMBOL(__skb_gso_segment);
2318 /* Take action when hardware reception checksum errors are detected. */
2320 void netdev_rx_csum_fault(struct net_device *dev)
2322 if (net_ratelimit()) {
2323 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2327 EXPORT_SYMBOL(netdev_rx_csum_fault);
2330 /* Actually, we should eliminate this check as soon as we know, that:
2331 * 1. IOMMU is present and allows to map all the memory.
2332 * 2. No high memory really exists on this machine.
2335 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2337 #ifdef CONFIG_HIGHMEM
2339 if (!(dev->features & NETIF_F_HIGHDMA)) {
2340 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2341 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2342 if (PageHighMem(skb_frag_page(frag)))
2347 if (PCI_DMA_BUS_IS_PHYS) {
2348 struct device *pdev = dev->dev.parent;
2352 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2353 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2354 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2355 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2364 void (*destructor)(struct sk_buff *skb);
2367 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2369 static void dev_gso_skb_destructor(struct sk_buff *skb)
2371 struct dev_gso_cb *cb;
2374 struct sk_buff *nskb = skb->next;
2376 skb->next = nskb->next;
2379 } while (skb->next);
2381 cb = DEV_GSO_CB(skb);
2383 cb->destructor(skb);
2387 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2388 * @skb: buffer to segment
2389 * @features: device features as applicable to this skb
2391 * This function segments the given skb and stores the list of segments
2394 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2396 struct sk_buff *segs;
2398 segs = skb_gso_segment(skb, features);
2400 /* Verifying header integrity only. */
2405 return PTR_ERR(segs);
2408 DEV_GSO_CB(skb)->destructor = skb->destructor;
2409 skb->destructor = dev_gso_skb_destructor;
2414 static netdev_features_t harmonize_features(struct sk_buff *skb,
2415 __be16 protocol, netdev_features_t features)
2417 if (skb->ip_summed != CHECKSUM_NONE &&
2418 !can_checksum_protocol(features, protocol)) {
2419 features &= ~NETIF_F_ALL_CSUM;
2420 } else if (illegal_highdma(skb->dev, skb)) {
2421 features &= ~NETIF_F_SG;
2427 netdev_features_t netif_skb_features(struct sk_buff *skb)
2429 __be16 protocol = skb->protocol;
2430 netdev_features_t features = skb->dev->features;
2432 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2433 features &= ~NETIF_F_GSO_MASK;
2435 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2436 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2437 protocol = veh->h_vlan_encapsulated_proto;
2438 } else if (!vlan_tx_tag_present(skb)) {
2439 return harmonize_features(skb, protocol, features);
2442 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2443 NETIF_F_HW_VLAN_STAG_TX);
2445 if (protocol != htons(ETH_P_8021Q) && protocol != htons(ETH_P_8021AD)) {
2446 return harmonize_features(skb, protocol, features);
2448 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2449 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2450 NETIF_F_HW_VLAN_STAG_TX;
2451 return harmonize_features(skb, protocol, features);
2454 EXPORT_SYMBOL(netif_skb_features);
2457 * Returns true if either:
2458 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2459 * 2. skb is fragmented and the device does not support SG.
2461 static inline int skb_needs_linearize(struct sk_buff *skb,
2462 netdev_features_t features)
2464 return skb_is_nonlinear(skb) &&
2465 ((skb_has_frag_list(skb) &&
2466 !(features & NETIF_F_FRAGLIST)) ||
2467 (skb_shinfo(skb)->nr_frags &&
2468 !(features & NETIF_F_SG)));
2471 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2472 struct netdev_queue *txq)
2474 const struct net_device_ops *ops = dev->netdev_ops;
2475 int rc = NETDEV_TX_OK;
2476 unsigned int skb_len;
2478 if (likely(!skb->next)) {
2479 netdev_features_t features;
2482 * If device doesn't need skb->dst, release it right now while
2483 * its hot in this cpu cache
2485 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2488 features = netif_skb_features(skb);
2490 if (vlan_tx_tag_present(skb) &&
2491 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2492 skb = __vlan_put_tag(skb, skb->vlan_proto,
2493 vlan_tx_tag_get(skb));
2500 /* If encapsulation offload request, verify we are testing
2501 * hardware encapsulation features instead of standard
2502 * features for the netdev
2504 if (skb->encapsulation)
2505 features &= dev->hw_enc_features;
2507 if (netif_needs_gso(skb, features)) {
2508 if (unlikely(dev_gso_segment(skb, features)))
2513 if (skb_needs_linearize(skb, features) &&
2514 __skb_linearize(skb))
2517 /* If packet is not checksummed and device does not
2518 * support checksumming for this protocol, complete
2519 * checksumming here.
2521 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2522 if (skb->encapsulation)
2523 skb_set_inner_transport_header(skb,
2524 skb_checksum_start_offset(skb));
2526 skb_set_transport_header(skb,
2527 skb_checksum_start_offset(skb));
2528 if (!(features & NETIF_F_ALL_CSUM) &&
2529 skb_checksum_help(skb))
2534 if (!list_empty(&ptype_all))
2535 dev_queue_xmit_nit(skb, dev);
2538 rc = ops->ndo_start_xmit(skb, dev);
2539 trace_net_dev_xmit(skb, rc, dev, skb_len);
2540 if (rc == NETDEV_TX_OK)
2541 txq_trans_update(txq);
2547 struct sk_buff *nskb = skb->next;
2549 skb->next = nskb->next;
2552 if (!list_empty(&ptype_all))
2553 dev_queue_xmit_nit(nskb, dev);
2555 skb_len = nskb->len;
2556 rc = ops->ndo_start_xmit(nskb, dev);
2557 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2558 if (unlikely(rc != NETDEV_TX_OK)) {
2559 if (rc & ~NETDEV_TX_MASK)
2560 goto out_kfree_gso_skb;
2561 nskb->next = skb->next;
2565 txq_trans_update(txq);
2566 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2567 return NETDEV_TX_BUSY;
2568 } while (skb->next);
2571 if (likely(skb->next == NULL)) {
2572 skb->destructor = DEV_GSO_CB(skb)->destructor;
2582 static void qdisc_pkt_len_init(struct sk_buff *skb)
2584 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2586 qdisc_skb_cb(skb)->pkt_len = skb->len;
2588 /* To get more precise estimation of bytes sent on wire,
2589 * we add to pkt_len the headers size of all segments
2591 if (shinfo->gso_size) {
2592 unsigned int hdr_len;
2593 u16 gso_segs = shinfo->gso_segs;
2595 /* mac layer + network layer */
2596 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2598 /* + transport layer */
2599 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2600 hdr_len += tcp_hdrlen(skb);
2602 hdr_len += sizeof(struct udphdr);
2604 if (shinfo->gso_type & SKB_GSO_DODGY)
2605 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2608 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2612 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2613 struct net_device *dev,
2614 struct netdev_queue *txq)
2616 spinlock_t *root_lock = qdisc_lock(q);
2620 qdisc_pkt_len_init(skb);
2621 qdisc_calculate_pkt_len(skb, q);
2623 * Heuristic to force contended enqueues to serialize on a
2624 * separate lock before trying to get qdisc main lock.
2625 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2626 * and dequeue packets faster.
2628 contended = qdisc_is_running(q);
2629 if (unlikely(contended))
2630 spin_lock(&q->busylock);
2632 spin_lock(root_lock);
2633 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2636 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2637 qdisc_run_begin(q)) {
2639 * This is a work-conserving queue; there are no old skbs
2640 * waiting to be sent out; and the qdisc is not running -
2641 * xmit the skb directly.
2643 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2646 qdisc_bstats_update(q, skb);
2648 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2649 if (unlikely(contended)) {
2650 spin_unlock(&q->busylock);
2657 rc = NET_XMIT_SUCCESS;
2660 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2661 if (qdisc_run_begin(q)) {
2662 if (unlikely(contended)) {
2663 spin_unlock(&q->busylock);
2669 spin_unlock(root_lock);
2670 if (unlikely(contended))
2671 spin_unlock(&q->busylock);
2675 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2676 static void skb_update_prio(struct sk_buff *skb)
2678 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2680 if (!skb->priority && skb->sk && map) {
2681 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2683 if (prioidx < map->priomap_len)
2684 skb->priority = map->priomap[prioidx];
2688 #define skb_update_prio(skb)
2691 static DEFINE_PER_CPU(int, xmit_recursion);
2692 #define RECURSION_LIMIT 10
2695 * dev_loopback_xmit - loop back @skb
2696 * @skb: buffer to transmit
2698 int dev_loopback_xmit(struct sk_buff *skb)
2700 skb_reset_mac_header(skb);
2701 __skb_pull(skb, skb_network_offset(skb));
2702 skb->pkt_type = PACKET_LOOPBACK;
2703 skb->ip_summed = CHECKSUM_UNNECESSARY;
2704 WARN_ON(!skb_dst(skb));
2709 EXPORT_SYMBOL(dev_loopback_xmit);
2712 * dev_queue_xmit - transmit a buffer
2713 * @skb: buffer to transmit
2715 * Queue a buffer for transmission to a network device. The caller must
2716 * have set the device and priority and built the buffer before calling
2717 * this function. The function can be called from an interrupt.
2719 * A negative errno code is returned on a failure. A success does not
2720 * guarantee the frame will be transmitted as it may be dropped due
2721 * to congestion or traffic shaping.
2723 * -----------------------------------------------------------------------------------
2724 * I notice this method can also return errors from the queue disciplines,
2725 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2728 * Regardless of the return value, the skb is consumed, so it is currently
2729 * difficult to retry a send to this method. (You can bump the ref count
2730 * before sending to hold a reference for retry if you are careful.)
2732 * When calling this method, interrupts MUST be enabled. This is because
2733 * the BH enable code must have IRQs enabled so that it will not deadlock.
2736 int dev_queue_xmit(struct sk_buff *skb)
2738 struct net_device *dev = skb->dev;
2739 struct netdev_queue *txq;
2743 skb_reset_mac_header(skb);
2745 /* Disable soft irqs for various locks below. Also
2746 * stops preemption for RCU.
2750 skb_update_prio(skb);
2752 txq = netdev_pick_tx(dev, skb);
2753 q = rcu_dereference_bh(txq->qdisc);
2755 #ifdef CONFIG_NET_CLS_ACT
2756 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2758 trace_net_dev_queue(skb);
2760 rc = __dev_xmit_skb(skb, q, dev, txq);
2764 /* The device has no queue. Common case for software devices:
2765 loopback, all the sorts of tunnels...
2767 Really, it is unlikely that netif_tx_lock protection is necessary
2768 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2770 However, it is possible, that they rely on protection
2773 Check this and shot the lock. It is not prone from deadlocks.
2774 Either shot noqueue qdisc, it is even simpler 8)
2776 if (dev->flags & IFF_UP) {
2777 int cpu = smp_processor_id(); /* ok because BHs are off */
2779 if (txq->xmit_lock_owner != cpu) {
2781 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2782 goto recursion_alert;
2784 HARD_TX_LOCK(dev, txq, cpu);
2786 if (!netif_xmit_stopped(txq)) {
2787 __this_cpu_inc(xmit_recursion);
2788 rc = dev_hard_start_xmit(skb, dev, txq);
2789 __this_cpu_dec(xmit_recursion);
2790 if (dev_xmit_complete(rc)) {
2791 HARD_TX_UNLOCK(dev, txq);
2795 HARD_TX_UNLOCK(dev, txq);
2796 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2799 /* Recursion is detected! It is possible,
2803 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2809 rcu_read_unlock_bh();
2814 rcu_read_unlock_bh();
2817 EXPORT_SYMBOL(dev_queue_xmit);
2820 /*=======================================================================
2822 =======================================================================*/
2824 int netdev_max_backlog __read_mostly = 1000;
2825 EXPORT_SYMBOL(netdev_max_backlog);
2827 int netdev_tstamp_prequeue __read_mostly = 1;
2828 int netdev_budget __read_mostly = 300;
2829 int weight_p __read_mostly = 64; /* old backlog weight */
2831 /* Called with irq disabled */
2832 static inline void ____napi_schedule(struct softnet_data *sd,
2833 struct napi_struct *napi)
2835 list_add_tail(&napi->poll_list, &sd->poll_list);
2836 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2841 /* One global table that all flow-based protocols share. */
2842 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2843 EXPORT_SYMBOL(rps_sock_flow_table);
2845 struct static_key rps_needed __read_mostly;
2847 static struct rps_dev_flow *
2848 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2849 struct rps_dev_flow *rflow, u16 next_cpu)
2851 if (next_cpu != RPS_NO_CPU) {
2852 #ifdef CONFIG_RFS_ACCEL
2853 struct netdev_rx_queue *rxqueue;
2854 struct rps_dev_flow_table *flow_table;
2855 struct rps_dev_flow *old_rflow;
2860 /* Should we steer this flow to a different hardware queue? */
2861 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2862 !(dev->features & NETIF_F_NTUPLE))
2864 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2865 if (rxq_index == skb_get_rx_queue(skb))
2868 rxqueue = dev->_rx + rxq_index;
2869 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2872 flow_id = skb->rxhash & flow_table->mask;
2873 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2874 rxq_index, flow_id);
2878 rflow = &flow_table->flows[flow_id];
2880 if (old_rflow->filter == rflow->filter)
2881 old_rflow->filter = RPS_NO_FILTER;
2885 per_cpu(softnet_data, next_cpu).input_queue_head;
2888 rflow->cpu = next_cpu;
2893 * get_rps_cpu is called from netif_receive_skb and returns the target
2894 * CPU from the RPS map of the receiving queue for a given skb.
2895 * rcu_read_lock must be held on entry.
2897 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2898 struct rps_dev_flow **rflowp)
2900 struct netdev_rx_queue *rxqueue;
2901 struct rps_map *map;
2902 struct rps_dev_flow_table *flow_table;
2903 struct rps_sock_flow_table *sock_flow_table;
2907 if (skb_rx_queue_recorded(skb)) {
2908 u16 index = skb_get_rx_queue(skb);
2909 if (unlikely(index >= dev->real_num_rx_queues)) {
2910 WARN_ONCE(dev->real_num_rx_queues > 1,
2911 "%s received packet on queue %u, but number "
2912 "of RX queues is %u\n",
2913 dev->name, index, dev->real_num_rx_queues);
2916 rxqueue = dev->_rx + index;
2920 map = rcu_dereference(rxqueue->rps_map);
2922 if (map->len == 1 &&
2923 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2924 tcpu = map->cpus[0];
2925 if (cpu_online(tcpu))
2929 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2933 skb_reset_network_header(skb);
2934 if (!skb_get_rxhash(skb))
2937 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2938 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2939 if (flow_table && sock_flow_table) {
2941 struct rps_dev_flow *rflow;
2943 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2946 next_cpu = sock_flow_table->ents[skb->rxhash &
2947 sock_flow_table->mask];
2950 * If the desired CPU (where last recvmsg was done) is
2951 * different from current CPU (one in the rx-queue flow
2952 * table entry), switch if one of the following holds:
2953 * - Current CPU is unset (equal to RPS_NO_CPU).
2954 * - Current CPU is offline.
2955 * - The current CPU's queue tail has advanced beyond the
2956 * last packet that was enqueued using this table entry.
2957 * This guarantees that all previous packets for the flow
2958 * have been dequeued, thus preserving in order delivery.
2960 if (unlikely(tcpu != next_cpu) &&
2961 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2962 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2963 rflow->last_qtail)) >= 0)) {
2965 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2968 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2976 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2978 if (cpu_online(tcpu)) {
2988 #ifdef CONFIG_RFS_ACCEL
2991 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2992 * @dev: Device on which the filter was set
2993 * @rxq_index: RX queue index
2994 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2995 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2997 * Drivers that implement ndo_rx_flow_steer() should periodically call
2998 * this function for each installed filter and remove the filters for
2999 * which it returns %true.
3001 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3002 u32 flow_id, u16 filter_id)
3004 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3005 struct rps_dev_flow_table *flow_table;
3006 struct rps_dev_flow *rflow;
3011 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3012 if (flow_table && flow_id <= flow_table->mask) {
3013 rflow = &flow_table->flows[flow_id];
3014 cpu = ACCESS_ONCE(rflow->cpu);
3015 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3016 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3017 rflow->last_qtail) <
3018 (int)(10 * flow_table->mask)))
3024 EXPORT_SYMBOL(rps_may_expire_flow);
3026 #endif /* CONFIG_RFS_ACCEL */
3028 /* Called from hardirq (IPI) context */
3029 static void rps_trigger_softirq(void *data)
3031 struct softnet_data *sd = data;
3033 ____napi_schedule(sd, &sd->backlog);
3037 #endif /* CONFIG_RPS */
3040 * Check if this softnet_data structure is another cpu one
3041 * If yes, queue it to our IPI list and return 1
3044 static int rps_ipi_queued(struct softnet_data *sd)
3047 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3050 sd->rps_ipi_next = mysd->rps_ipi_list;
3051 mysd->rps_ipi_list = sd;
3053 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3056 #endif /* CONFIG_RPS */
3060 #ifdef CONFIG_NET_FLOW_LIMIT
3061 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3064 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3066 #ifdef CONFIG_NET_FLOW_LIMIT
3067 struct sd_flow_limit *fl;
3068 struct softnet_data *sd;
3069 unsigned int old_flow, new_flow;
3071 if (qlen < (netdev_max_backlog >> 1))
3074 sd = &__get_cpu_var(softnet_data);
3077 fl = rcu_dereference(sd->flow_limit);
3079 new_flow = skb_get_rxhash(skb) & (fl->num_buckets - 1);
3080 old_flow = fl->history[fl->history_head];
3081 fl->history[fl->history_head] = new_flow;
3084 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3086 if (likely(fl->buckets[old_flow]))
3087 fl->buckets[old_flow]--;
3089 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3101 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3102 * queue (may be a remote CPU queue).
3104 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3105 unsigned int *qtail)
3107 struct softnet_data *sd;
3108 unsigned long flags;
3111 sd = &per_cpu(softnet_data, cpu);
3113 local_irq_save(flags);
3116 qlen = skb_queue_len(&sd->input_pkt_queue);
3117 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3118 if (skb_queue_len(&sd->input_pkt_queue)) {
3120 __skb_queue_tail(&sd->input_pkt_queue, skb);
3121 input_queue_tail_incr_save(sd, qtail);
3123 local_irq_restore(flags);
3124 return NET_RX_SUCCESS;
3127 /* Schedule NAPI for backlog device
3128 * We can use non atomic operation since we own the queue lock
3130 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3131 if (!rps_ipi_queued(sd))
3132 ____napi_schedule(sd, &sd->backlog);
3140 local_irq_restore(flags);
3142 atomic_long_inc(&skb->dev->rx_dropped);
3148 * netif_rx - post buffer to the network code
3149 * @skb: buffer to post
3151 * This function receives a packet from a device driver and queues it for
3152 * the upper (protocol) levels to process. It always succeeds. The buffer
3153 * may be dropped during processing for congestion control or by the
3157 * NET_RX_SUCCESS (no congestion)
3158 * NET_RX_DROP (packet was dropped)
3162 int netif_rx(struct sk_buff *skb)
3166 /* if netpoll wants it, pretend we never saw it */
3167 if (netpoll_rx(skb))
3170 net_timestamp_check(netdev_tstamp_prequeue, skb);
3172 trace_netif_rx(skb);
3174 if (static_key_false(&rps_needed)) {
3175 struct rps_dev_flow voidflow, *rflow = &voidflow;
3181 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3183 cpu = smp_processor_id();
3185 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3193 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3198 EXPORT_SYMBOL(netif_rx);
3200 int netif_rx_ni(struct sk_buff *skb)
3205 err = netif_rx(skb);
3206 if (local_softirq_pending())
3212 EXPORT_SYMBOL(netif_rx_ni);
3214 static void net_tx_action(struct softirq_action *h)
3216 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3218 if (sd->completion_queue) {
3219 struct sk_buff *clist;
3221 local_irq_disable();
3222 clist = sd->completion_queue;
3223 sd->completion_queue = NULL;
3227 struct sk_buff *skb = clist;
3228 clist = clist->next;
3230 WARN_ON(atomic_read(&skb->users));
3231 trace_kfree_skb(skb, net_tx_action);
3236 if (sd->output_queue) {
3239 local_irq_disable();
3240 head = sd->output_queue;
3241 sd->output_queue = NULL;
3242 sd->output_queue_tailp = &sd->output_queue;
3246 struct Qdisc *q = head;
3247 spinlock_t *root_lock;
3249 head = head->next_sched;
3251 root_lock = qdisc_lock(q);
3252 if (spin_trylock(root_lock)) {
3253 smp_mb__before_clear_bit();
3254 clear_bit(__QDISC_STATE_SCHED,
3257 spin_unlock(root_lock);
3259 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3261 __netif_reschedule(q);
3263 smp_mb__before_clear_bit();
3264 clear_bit(__QDISC_STATE_SCHED,
3272 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3273 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3274 /* This hook is defined here for ATM LANE */
3275 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3276 unsigned char *addr) __read_mostly;
3277 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3280 #ifdef CONFIG_NET_CLS_ACT
3281 /* TODO: Maybe we should just force sch_ingress to be compiled in
3282 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3283 * a compare and 2 stores extra right now if we dont have it on
3284 * but have CONFIG_NET_CLS_ACT
3285 * NOTE: This doesn't stop any functionality; if you dont have
3286 * the ingress scheduler, you just can't add policies on ingress.
3289 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3291 struct net_device *dev = skb->dev;
3292 u32 ttl = G_TC_RTTL(skb->tc_verd);
3293 int result = TC_ACT_OK;
3296 if (unlikely(MAX_RED_LOOP < ttl++)) {
3297 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3298 skb->skb_iif, dev->ifindex);
3302 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3303 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3306 if (q != &noop_qdisc) {
3307 spin_lock(qdisc_lock(q));
3308 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3309 result = qdisc_enqueue_root(skb, q);
3310 spin_unlock(qdisc_lock(q));
3316 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3317 struct packet_type **pt_prev,
3318 int *ret, struct net_device *orig_dev)
3320 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3322 if (!rxq || rxq->qdisc == &noop_qdisc)
3326 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3330 switch (ing_filter(skb, rxq)) {
3344 * netdev_rx_handler_register - register receive handler
3345 * @dev: device to register a handler for
3346 * @rx_handler: receive handler to register
3347 * @rx_handler_data: data pointer that is used by rx handler
3349 * Register a receive hander for a device. This handler will then be
3350 * called from __netif_receive_skb. A negative errno code is returned
3353 * The caller must hold the rtnl_mutex.
3355 * For a general description of rx_handler, see enum rx_handler_result.
3357 int netdev_rx_handler_register(struct net_device *dev,
3358 rx_handler_func_t *rx_handler,
3359 void *rx_handler_data)
3363 if (dev->rx_handler)
3366 /* Note: rx_handler_data must be set before rx_handler */
3367 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3368 rcu_assign_pointer(dev->rx_handler, rx_handler);
3372 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3375 * netdev_rx_handler_unregister - unregister receive handler
3376 * @dev: device to unregister a handler from
3378 * Unregister a receive handler from a device.
3380 * The caller must hold the rtnl_mutex.
3382 void netdev_rx_handler_unregister(struct net_device *dev)
3386 RCU_INIT_POINTER(dev->rx_handler, NULL);
3387 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3388 * section has a guarantee to see a non NULL rx_handler_data
3392 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3394 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3397 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3398 * the special handling of PFMEMALLOC skbs.
3400 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3402 switch (skb->protocol) {
3403 case __constant_htons(ETH_P_ARP):
3404 case __constant_htons(ETH_P_IP):
3405 case __constant_htons(ETH_P_IPV6):
3406 case __constant_htons(ETH_P_8021Q):
3407 case __constant_htons(ETH_P_8021AD):
3414 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3416 struct packet_type *ptype, *pt_prev;
3417 rx_handler_func_t *rx_handler;
3418 struct net_device *orig_dev;
3419 struct net_device *null_or_dev;
3420 bool deliver_exact = false;
3421 int ret = NET_RX_DROP;
3424 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3426 trace_netif_receive_skb(skb);
3428 /* if we've gotten here through NAPI, check netpoll */
3429 if (netpoll_receive_skb(skb))
3432 orig_dev = skb->dev;
3434 skb_reset_network_header(skb);
3435 if (!skb_transport_header_was_set(skb))
3436 skb_reset_transport_header(skb);
3437 skb_reset_mac_len(skb);
3444 skb->skb_iif = skb->dev->ifindex;
3446 __this_cpu_inc(softnet_data.processed);
3448 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3449 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3450 skb = vlan_untag(skb);
3455 #ifdef CONFIG_NET_CLS_ACT
3456 if (skb->tc_verd & TC_NCLS) {
3457 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3465 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3466 if (!ptype->dev || ptype->dev == skb->dev) {
3468 ret = deliver_skb(skb, pt_prev, orig_dev);
3474 #ifdef CONFIG_NET_CLS_ACT
3475 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3481 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3484 if (vlan_tx_tag_present(skb)) {
3486 ret = deliver_skb(skb, pt_prev, orig_dev);
3489 if (vlan_do_receive(&skb))
3491 else if (unlikely(!skb))
3495 rx_handler = rcu_dereference(skb->dev->rx_handler);
3498 ret = deliver_skb(skb, pt_prev, orig_dev);
3501 switch (rx_handler(&skb)) {
3502 case RX_HANDLER_CONSUMED:
3503 ret = NET_RX_SUCCESS;
3505 case RX_HANDLER_ANOTHER:
3507 case RX_HANDLER_EXACT:
3508 deliver_exact = true;
3509 case RX_HANDLER_PASS:
3516 if (vlan_tx_nonzero_tag_present(skb))
3517 skb->pkt_type = PACKET_OTHERHOST;
3519 /* deliver only exact match when indicated */
3520 null_or_dev = deliver_exact ? skb->dev : NULL;
3522 type = skb->protocol;
3523 list_for_each_entry_rcu(ptype,
3524 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3525 if (ptype->type == type &&
3526 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3527 ptype->dev == orig_dev)) {
3529 ret = deliver_skb(skb, pt_prev, orig_dev);
3535 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3538 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3541 atomic_long_inc(&skb->dev->rx_dropped);
3543 /* Jamal, now you will not able to escape explaining
3544 * me how you were going to use this. :-)
3555 static int __netif_receive_skb(struct sk_buff *skb)
3559 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3560 unsigned long pflags = current->flags;
3563 * PFMEMALLOC skbs are special, they should
3564 * - be delivered to SOCK_MEMALLOC sockets only
3565 * - stay away from userspace
3566 * - have bounded memory usage
3568 * Use PF_MEMALLOC as this saves us from propagating the allocation
3569 * context down to all allocation sites.
3571 current->flags |= PF_MEMALLOC;
3572 ret = __netif_receive_skb_core(skb, true);
3573 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3575 ret = __netif_receive_skb_core(skb, false);
3581 * netif_receive_skb - process receive buffer from network
3582 * @skb: buffer to process
3584 * netif_receive_skb() is the main receive data processing function.
3585 * It always succeeds. The buffer may be dropped during processing
3586 * for congestion control or by the protocol layers.
3588 * This function may only be called from softirq context and interrupts
3589 * should be enabled.
3591 * Return values (usually ignored):
3592 * NET_RX_SUCCESS: no congestion
3593 * NET_RX_DROP: packet was dropped
3595 int netif_receive_skb(struct sk_buff *skb)
3597 net_timestamp_check(netdev_tstamp_prequeue, skb);
3599 if (skb_defer_rx_timestamp(skb))
3600 return NET_RX_SUCCESS;
3603 if (static_key_false(&rps_needed)) {
3604 struct rps_dev_flow voidflow, *rflow = &voidflow;
3609 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3612 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3619 return __netif_receive_skb(skb);
3621 EXPORT_SYMBOL(netif_receive_skb);
3623 /* Network device is going away, flush any packets still pending
3624 * Called with irqs disabled.
3626 static void flush_backlog(void *arg)
3628 struct net_device *dev = arg;
3629 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3630 struct sk_buff *skb, *tmp;
3633 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3634 if (skb->dev == dev) {
3635 __skb_unlink(skb, &sd->input_pkt_queue);
3637 input_queue_head_incr(sd);
3642 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3643 if (skb->dev == dev) {
3644 __skb_unlink(skb, &sd->process_queue);
3646 input_queue_head_incr(sd);
3651 static int napi_gro_complete(struct sk_buff *skb)
3653 struct packet_offload *ptype;
3654 __be16 type = skb->protocol;
3655 struct list_head *head = &offload_base;
3658 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3660 if (NAPI_GRO_CB(skb)->count == 1) {
3661 skb_shinfo(skb)->gso_size = 0;
3666 list_for_each_entry_rcu(ptype, head, list) {
3667 if (ptype->type != type || !ptype->callbacks.gro_complete)
3670 err = ptype->callbacks.gro_complete(skb);
3676 WARN_ON(&ptype->list == head);
3678 return NET_RX_SUCCESS;
3682 return netif_receive_skb(skb);
3685 /* napi->gro_list contains packets ordered by age.
3686 * youngest packets at the head of it.
3687 * Complete skbs in reverse order to reduce latencies.
3689 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3691 struct sk_buff *skb, *prev = NULL;
3693 /* scan list and build reverse chain */
3694 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3699 for (skb = prev; skb; skb = prev) {
3702 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3706 napi_gro_complete(skb);
3710 napi->gro_list = NULL;
3712 EXPORT_SYMBOL(napi_gro_flush);
3714 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3717 unsigned int maclen = skb->dev->hard_header_len;
3719 for (p = napi->gro_list; p; p = p->next) {
3720 unsigned long diffs;
3722 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3723 diffs |= p->vlan_tci ^ skb->vlan_tci;
3724 if (maclen == ETH_HLEN)
3725 diffs |= compare_ether_header(skb_mac_header(p),
3726 skb_gro_mac_header(skb));
3728 diffs = memcmp(skb_mac_header(p),
3729 skb_gro_mac_header(skb),
3731 NAPI_GRO_CB(p)->same_flow = !diffs;
3732 NAPI_GRO_CB(p)->flush = 0;
3736 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3738 struct sk_buff **pp = NULL;
3739 struct packet_offload *ptype;
3740 __be16 type = skb->protocol;
3741 struct list_head *head = &offload_base;
3743 enum gro_result ret;
3745 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3748 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3751 gro_list_prepare(napi, skb);
3754 list_for_each_entry_rcu(ptype, head, list) {
3755 if (ptype->type != type || !ptype->callbacks.gro_receive)
3758 skb_set_network_header(skb, skb_gro_offset(skb));
3759 skb_reset_mac_len(skb);
3760 NAPI_GRO_CB(skb)->same_flow = 0;
3761 NAPI_GRO_CB(skb)->flush = 0;
3762 NAPI_GRO_CB(skb)->free = 0;
3764 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3769 if (&ptype->list == head)
3772 same_flow = NAPI_GRO_CB(skb)->same_flow;
3773 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3776 struct sk_buff *nskb = *pp;
3780 napi_gro_complete(nskb);
3787 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3791 NAPI_GRO_CB(skb)->count = 1;
3792 NAPI_GRO_CB(skb)->age = jiffies;
3793 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3794 skb->next = napi->gro_list;
3795 napi->gro_list = skb;
3799 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3800 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3802 BUG_ON(skb->end - skb->tail < grow);
3804 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3807 skb->data_len -= grow;
3809 skb_shinfo(skb)->frags[0].page_offset += grow;
3810 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3812 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3813 skb_frag_unref(skb, 0);
3814 memmove(skb_shinfo(skb)->frags,
3815 skb_shinfo(skb)->frags + 1,
3816 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3829 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3833 if (netif_receive_skb(skb))
3841 case GRO_MERGED_FREE:
3842 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3843 kmem_cache_free(skbuff_head_cache, skb);
3856 static void skb_gro_reset_offset(struct sk_buff *skb)
3858 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3859 const skb_frag_t *frag0 = &pinfo->frags[0];
3861 NAPI_GRO_CB(skb)->data_offset = 0;
3862 NAPI_GRO_CB(skb)->frag0 = NULL;
3863 NAPI_GRO_CB(skb)->frag0_len = 0;
3865 if (skb->mac_header == skb->tail &&
3867 !PageHighMem(skb_frag_page(frag0))) {
3868 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3869 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3873 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3875 skb_gro_reset_offset(skb);
3877 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
3879 EXPORT_SYMBOL(napi_gro_receive);
3881 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3883 __skb_pull(skb, skb_headlen(skb));
3884 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3885 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3887 skb->dev = napi->dev;
3893 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3895 struct sk_buff *skb = napi->skb;
3898 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3904 EXPORT_SYMBOL(napi_get_frags);
3906 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3912 skb->protocol = eth_type_trans(skb, skb->dev);
3914 if (ret == GRO_HELD)
3915 skb_gro_pull(skb, -ETH_HLEN);
3916 else if (netif_receive_skb(skb))
3921 case GRO_MERGED_FREE:
3922 napi_reuse_skb(napi, skb);
3932 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3934 struct sk_buff *skb = napi->skb;
3941 skb_reset_mac_header(skb);
3942 skb_gro_reset_offset(skb);
3944 off = skb_gro_offset(skb);
3945 hlen = off + sizeof(*eth);
3946 eth = skb_gro_header_fast(skb, off);
3947 if (skb_gro_header_hard(skb, hlen)) {
3948 eth = skb_gro_header_slow(skb, hlen, off);
3949 if (unlikely(!eth)) {
3950 napi_reuse_skb(napi, skb);
3956 skb_gro_pull(skb, sizeof(*eth));
3959 * This works because the only protocols we care about don't require
3960 * special handling. We'll fix it up properly at the end.
3962 skb->protocol = eth->h_proto;
3968 gro_result_t napi_gro_frags(struct napi_struct *napi)
3970 struct sk_buff *skb = napi_frags_skb(napi);
3975 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
3977 EXPORT_SYMBOL(napi_gro_frags);
3980 * net_rps_action sends any pending IPI's for rps.
3981 * Note: called with local irq disabled, but exits with local irq enabled.
3983 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3986 struct softnet_data *remsd = sd->rps_ipi_list;
3989 sd->rps_ipi_list = NULL;
3993 /* Send pending IPI's to kick RPS processing on remote cpus. */
3995 struct softnet_data *next = remsd->rps_ipi_next;
3997 if (cpu_online(remsd->cpu))
3998 __smp_call_function_single(remsd->cpu,
4007 static int process_backlog(struct napi_struct *napi, int quota)
4010 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4013 /* Check if we have pending ipi, its better to send them now,
4014 * not waiting net_rx_action() end.
4016 if (sd->rps_ipi_list) {
4017 local_irq_disable();
4018 net_rps_action_and_irq_enable(sd);
4021 napi->weight = weight_p;
4022 local_irq_disable();
4023 while (work < quota) {
4024 struct sk_buff *skb;
4027 while ((skb = __skb_dequeue(&sd->process_queue))) {
4029 __netif_receive_skb(skb);
4030 local_irq_disable();
4031 input_queue_head_incr(sd);
4032 if (++work >= quota) {
4039 qlen = skb_queue_len(&sd->input_pkt_queue);
4041 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4042 &sd->process_queue);
4044 if (qlen < quota - work) {
4046 * Inline a custom version of __napi_complete().
4047 * only current cpu owns and manipulates this napi,
4048 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4049 * we can use a plain write instead of clear_bit(),
4050 * and we dont need an smp_mb() memory barrier.
4052 list_del(&napi->poll_list);
4055 quota = work + qlen;
4065 * __napi_schedule - schedule for receive
4066 * @n: entry to schedule
4068 * The entry's receive function will be scheduled to run
4070 void __napi_schedule(struct napi_struct *n)
4072 unsigned long flags;
4074 local_irq_save(flags);
4075 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4076 local_irq_restore(flags);
4078 EXPORT_SYMBOL(__napi_schedule);
4080 void __napi_complete(struct napi_struct *n)
4082 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4083 BUG_ON(n->gro_list);
4085 list_del(&n->poll_list);
4086 smp_mb__before_clear_bit();
4087 clear_bit(NAPI_STATE_SCHED, &n->state);
4089 EXPORT_SYMBOL(__napi_complete);
4091 void napi_complete(struct napi_struct *n)
4093 unsigned long flags;
4096 * don't let napi dequeue from the cpu poll list
4097 * just in case its running on a different cpu
4099 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4102 napi_gro_flush(n, false);
4103 local_irq_save(flags);
4105 local_irq_restore(flags);
4107 EXPORT_SYMBOL(napi_complete);
4109 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4110 int (*poll)(struct napi_struct *, int), int weight)
4112 INIT_LIST_HEAD(&napi->poll_list);
4113 napi->gro_count = 0;
4114 napi->gro_list = NULL;
4117 if (weight > NAPI_POLL_WEIGHT)
4118 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4120 napi->weight = weight;
4121 list_add(&napi->dev_list, &dev->napi_list);
4123 #ifdef CONFIG_NETPOLL
4124 spin_lock_init(&napi->poll_lock);
4125 napi->poll_owner = -1;
4127 set_bit(NAPI_STATE_SCHED, &napi->state);
4129 EXPORT_SYMBOL(netif_napi_add);
4131 void netif_napi_del(struct napi_struct *napi)
4133 struct sk_buff *skb, *next;
4135 list_del_init(&napi->dev_list);
4136 napi_free_frags(napi);
4138 for (skb = napi->gro_list; skb; skb = next) {
4144 napi->gro_list = NULL;
4145 napi->gro_count = 0;
4147 EXPORT_SYMBOL(netif_napi_del);
4149 static void net_rx_action(struct softirq_action *h)
4151 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4152 unsigned long time_limit = jiffies + 2;
4153 int budget = netdev_budget;
4156 local_irq_disable();
4158 while (!list_empty(&sd->poll_list)) {
4159 struct napi_struct *n;
4162 /* If softirq window is exhuasted then punt.
4163 * Allow this to run for 2 jiffies since which will allow
4164 * an average latency of 1.5/HZ.
4166 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4171 /* Even though interrupts have been re-enabled, this
4172 * access is safe because interrupts can only add new
4173 * entries to the tail of this list, and only ->poll()
4174 * calls can remove this head entry from the list.
4176 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4178 have = netpoll_poll_lock(n);
4182 /* This NAPI_STATE_SCHED test is for avoiding a race
4183 * with netpoll's poll_napi(). Only the entity which
4184 * obtains the lock and sees NAPI_STATE_SCHED set will
4185 * actually make the ->poll() call. Therefore we avoid
4186 * accidentally calling ->poll() when NAPI is not scheduled.
4189 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4190 work = n->poll(n, weight);
4194 WARN_ON_ONCE(work > weight);
4198 local_irq_disable();
4200 /* Drivers must not modify the NAPI state if they
4201 * consume the entire weight. In such cases this code
4202 * still "owns" the NAPI instance and therefore can
4203 * move the instance around on the list at-will.
4205 if (unlikely(work == weight)) {
4206 if (unlikely(napi_disable_pending(n))) {
4209 local_irq_disable();
4212 /* flush too old packets
4213 * If HZ < 1000, flush all packets.
4216 napi_gro_flush(n, HZ >= 1000);
4217 local_irq_disable();
4219 list_move_tail(&n->poll_list, &sd->poll_list);
4223 netpoll_poll_unlock(have);
4226 net_rps_action_and_irq_enable(sd);
4228 #ifdef CONFIG_NET_DMA
4230 * There may not be any more sk_buffs coming right now, so push
4231 * any pending DMA copies to hardware
4233 dma_issue_pending_all();
4240 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4244 struct netdev_upper {
4245 struct net_device *dev;
4247 struct list_head list;
4248 struct rcu_head rcu;
4249 struct list_head search_list;
4252 static void __append_search_uppers(struct list_head *search_list,
4253 struct net_device *dev)
4255 struct netdev_upper *upper;
4257 list_for_each_entry(upper, &dev->upper_dev_list, list) {
4258 /* check if this upper is not already in search list */
4259 if (list_empty(&upper->search_list))
4260 list_add_tail(&upper->search_list, search_list);
4264 static bool __netdev_search_upper_dev(struct net_device *dev,
4265 struct net_device *upper_dev)
4267 LIST_HEAD(search_list);
4268 struct netdev_upper *upper;
4269 struct netdev_upper *tmp;
4272 __append_search_uppers(&search_list, dev);
4273 list_for_each_entry(upper, &search_list, search_list) {
4274 if (upper->dev == upper_dev) {
4278 __append_search_uppers(&search_list, upper->dev);
4280 list_for_each_entry_safe(upper, tmp, &search_list, search_list)
4281 INIT_LIST_HEAD(&upper->search_list);
4285 static struct netdev_upper *__netdev_find_upper(struct net_device *dev,
4286 struct net_device *upper_dev)
4288 struct netdev_upper *upper;
4290 list_for_each_entry(upper, &dev->upper_dev_list, list) {
4291 if (upper->dev == upper_dev)
4298 * netdev_has_upper_dev - Check if device is linked to an upper device
4300 * @upper_dev: upper device to check
4302 * Find out if a device is linked to specified upper device and return true
4303 * in case it is. Note that this checks only immediate upper device,
4304 * not through a complete stack of devices. The caller must hold the RTNL lock.
4306 bool netdev_has_upper_dev(struct net_device *dev,
4307 struct net_device *upper_dev)
4311 return __netdev_find_upper(dev, upper_dev);
4313 EXPORT_SYMBOL(netdev_has_upper_dev);
4316 * netdev_has_any_upper_dev - Check if device is linked to some device
4319 * Find out if a device is linked to an upper device and return true in case
4320 * it is. The caller must hold the RTNL lock.
4322 bool netdev_has_any_upper_dev(struct net_device *dev)
4326 return !list_empty(&dev->upper_dev_list);
4328 EXPORT_SYMBOL(netdev_has_any_upper_dev);
4331 * netdev_master_upper_dev_get - Get master upper device
4334 * Find a master upper device and return pointer to it or NULL in case
4335 * it's not there. The caller must hold the RTNL lock.
4337 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4339 struct netdev_upper *upper;
4343 if (list_empty(&dev->upper_dev_list))
4346 upper = list_first_entry(&dev->upper_dev_list,
4347 struct netdev_upper, list);
4348 if (likely(upper->master))
4352 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4355 * netdev_master_upper_dev_get_rcu - Get master upper device
4358 * Find a master upper device and return pointer to it or NULL in case
4359 * it's not there. The caller must hold the RCU read lock.
4361 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4363 struct netdev_upper *upper;
4365 upper = list_first_or_null_rcu(&dev->upper_dev_list,
4366 struct netdev_upper, list);
4367 if (upper && likely(upper->master))
4371 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4373 static int __netdev_upper_dev_link(struct net_device *dev,
4374 struct net_device *upper_dev, bool master)
4376 struct netdev_upper *upper;
4380 if (dev == upper_dev)
4383 /* To prevent loops, check if dev is not upper device to upper_dev. */
4384 if (__netdev_search_upper_dev(upper_dev, dev))
4387 if (__netdev_find_upper(dev, upper_dev))
4390 if (master && netdev_master_upper_dev_get(dev))
4393 upper = kmalloc(sizeof(*upper), GFP_KERNEL);
4397 upper->dev = upper_dev;
4398 upper->master = master;
4399 INIT_LIST_HEAD(&upper->search_list);
4401 /* Ensure that master upper link is always the first item in list. */
4403 list_add_rcu(&upper->list, &dev->upper_dev_list);
4405 list_add_tail_rcu(&upper->list, &dev->upper_dev_list);
4406 dev_hold(upper_dev);
4407 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4412 * netdev_upper_dev_link - Add a link to the upper device
4414 * @upper_dev: new upper device
4416 * Adds a link to device which is upper to this one. The caller must hold
4417 * the RTNL lock. On a failure a negative errno code is returned.
4418 * On success the reference counts are adjusted and the function
4421 int netdev_upper_dev_link(struct net_device *dev,
4422 struct net_device *upper_dev)
4424 return __netdev_upper_dev_link(dev, upper_dev, false);
4426 EXPORT_SYMBOL(netdev_upper_dev_link);
4429 * netdev_master_upper_dev_link - Add a master link to the upper device
4431 * @upper_dev: new upper device
4433 * Adds a link to device which is upper to this one. In this case, only
4434 * one master upper device can be linked, although other non-master devices
4435 * might be linked as well. The caller must hold the RTNL lock.
4436 * On a failure a negative errno code is returned. On success the reference
4437 * counts are adjusted and the function returns zero.
4439 int netdev_master_upper_dev_link(struct net_device *dev,
4440 struct net_device *upper_dev)
4442 return __netdev_upper_dev_link(dev, upper_dev, true);
4444 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4447 * netdev_upper_dev_unlink - Removes a link to upper device
4449 * @upper_dev: new upper device
4451 * Removes a link to device which is upper to this one. The caller must hold
4454 void netdev_upper_dev_unlink(struct net_device *dev,
4455 struct net_device *upper_dev)
4457 struct netdev_upper *upper;
4461 upper = __netdev_find_upper(dev, upper_dev);
4464 list_del_rcu(&upper->list);
4466 kfree_rcu(upper, rcu);
4467 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4469 EXPORT_SYMBOL(netdev_upper_dev_unlink);
4471 static void dev_change_rx_flags(struct net_device *dev, int flags)
4473 const struct net_device_ops *ops = dev->netdev_ops;
4475 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4476 ops->ndo_change_rx_flags(dev, flags);
4479 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4481 unsigned int old_flags = dev->flags;
4487 dev->flags |= IFF_PROMISC;
4488 dev->promiscuity += inc;
4489 if (dev->promiscuity == 0) {
4492 * If inc causes overflow, untouch promisc and return error.
4495 dev->flags &= ~IFF_PROMISC;
4497 dev->promiscuity -= inc;
4498 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4503 if (dev->flags != old_flags) {
4504 pr_info("device %s %s promiscuous mode\n",
4506 dev->flags & IFF_PROMISC ? "entered" : "left");
4507 if (audit_enabled) {
4508 current_uid_gid(&uid, &gid);
4509 audit_log(current->audit_context, GFP_ATOMIC,
4510 AUDIT_ANOM_PROMISCUOUS,
4511 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4512 dev->name, (dev->flags & IFF_PROMISC),
4513 (old_flags & IFF_PROMISC),
4514 from_kuid(&init_user_ns, audit_get_loginuid(current)),
4515 from_kuid(&init_user_ns, uid),
4516 from_kgid(&init_user_ns, gid),
4517 audit_get_sessionid(current));
4520 dev_change_rx_flags(dev, IFF_PROMISC);
4526 * dev_set_promiscuity - update promiscuity count on a device
4530 * Add or remove promiscuity from a device. While the count in the device
4531 * remains above zero the interface remains promiscuous. Once it hits zero
4532 * the device reverts back to normal filtering operation. A negative inc
4533 * value is used to drop promiscuity on the device.
4534 * Return 0 if successful or a negative errno code on error.
4536 int dev_set_promiscuity(struct net_device *dev, int inc)
4538 unsigned int old_flags = dev->flags;
4541 err = __dev_set_promiscuity(dev, inc);
4544 if (dev->flags != old_flags)
4545 dev_set_rx_mode(dev);
4548 EXPORT_SYMBOL(dev_set_promiscuity);
4551 * dev_set_allmulti - update allmulti count on a device
4555 * Add or remove reception of all multicast frames to a device. While the
4556 * count in the device remains above zero the interface remains listening
4557 * to all interfaces. Once it hits zero the device reverts back to normal
4558 * filtering operation. A negative @inc value is used to drop the counter
4559 * when releasing a resource needing all multicasts.
4560 * Return 0 if successful or a negative errno code on error.
4563 int dev_set_allmulti(struct net_device *dev, int inc)
4565 unsigned int old_flags = dev->flags;
4569 dev->flags |= IFF_ALLMULTI;
4570 dev->allmulti += inc;
4571 if (dev->allmulti == 0) {
4574 * If inc causes overflow, untouch allmulti and return error.
4577 dev->flags &= ~IFF_ALLMULTI;
4579 dev->allmulti -= inc;
4580 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4585 if (dev->flags ^ old_flags) {
4586 dev_change_rx_flags(dev, IFF_ALLMULTI);
4587 dev_set_rx_mode(dev);
4591 EXPORT_SYMBOL(dev_set_allmulti);
4594 * Upload unicast and multicast address lists to device and
4595 * configure RX filtering. When the device doesn't support unicast
4596 * filtering it is put in promiscuous mode while unicast addresses
4599 void __dev_set_rx_mode(struct net_device *dev)
4601 const struct net_device_ops *ops = dev->netdev_ops;
4603 /* dev_open will call this function so the list will stay sane. */
4604 if (!(dev->flags&IFF_UP))
4607 if (!netif_device_present(dev))
4610 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4611 /* Unicast addresses changes may only happen under the rtnl,
4612 * therefore calling __dev_set_promiscuity here is safe.
4614 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4615 __dev_set_promiscuity(dev, 1);
4616 dev->uc_promisc = true;
4617 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4618 __dev_set_promiscuity(dev, -1);
4619 dev->uc_promisc = false;
4623 if (ops->ndo_set_rx_mode)
4624 ops->ndo_set_rx_mode(dev);
4627 void dev_set_rx_mode(struct net_device *dev)
4629 netif_addr_lock_bh(dev);
4630 __dev_set_rx_mode(dev);
4631 netif_addr_unlock_bh(dev);
4635 * dev_get_flags - get flags reported to userspace
4638 * Get the combination of flag bits exported through APIs to userspace.
4640 unsigned int dev_get_flags(const struct net_device *dev)
4644 flags = (dev->flags & ~(IFF_PROMISC |
4649 (dev->gflags & (IFF_PROMISC |
4652 if (netif_running(dev)) {
4653 if (netif_oper_up(dev))
4654 flags |= IFF_RUNNING;
4655 if (netif_carrier_ok(dev))
4656 flags |= IFF_LOWER_UP;
4657 if (netif_dormant(dev))
4658 flags |= IFF_DORMANT;
4663 EXPORT_SYMBOL(dev_get_flags);
4665 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4667 unsigned int old_flags = dev->flags;
4673 * Set the flags on our device.
4676 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4677 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4679 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4683 * Load in the correct multicast list now the flags have changed.
4686 if ((old_flags ^ flags) & IFF_MULTICAST)
4687 dev_change_rx_flags(dev, IFF_MULTICAST);
4689 dev_set_rx_mode(dev);
4692 * Have we downed the interface. We handle IFF_UP ourselves
4693 * according to user attempts to set it, rather than blindly
4698 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4699 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4702 dev_set_rx_mode(dev);
4705 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4706 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4708 dev->gflags ^= IFF_PROMISC;
4709 dev_set_promiscuity(dev, inc);
4712 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4713 is important. Some (broken) drivers set IFF_PROMISC, when
4714 IFF_ALLMULTI is requested not asking us and not reporting.
4716 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4717 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4719 dev->gflags ^= IFF_ALLMULTI;
4720 dev_set_allmulti(dev, inc);
4726 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4728 unsigned int changes = dev->flags ^ old_flags;
4730 if (changes & IFF_UP) {
4731 if (dev->flags & IFF_UP)
4732 call_netdevice_notifiers(NETDEV_UP, dev);
4734 call_netdevice_notifiers(NETDEV_DOWN, dev);
4737 if (dev->flags & IFF_UP &&
4738 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4739 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4743 * dev_change_flags - change device settings
4745 * @flags: device state flags
4747 * Change settings on device based state flags. The flags are
4748 * in the userspace exported format.
4750 int dev_change_flags(struct net_device *dev, unsigned int flags)
4753 unsigned int changes, old_flags = dev->flags;
4755 ret = __dev_change_flags(dev, flags);
4759 changes = old_flags ^ dev->flags;
4761 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4763 __dev_notify_flags(dev, old_flags);
4766 EXPORT_SYMBOL(dev_change_flags);
4769 * dev_set_mtu - Change maximum transfer unit
4771 * @new_mtu: new transfer unit
4773 * Change the maximum transfer size of the network device.
4775 int dev_set_mtu(struct net_device *dev, int new_mtu)
4777 const struct net_device_ops *ops = dev->netdev_ops;
4780 if (new_mtu == dev->mtu)
4783 /* MTU must be positive. */
4787 if (!netif_device_present(dev))
4791 if (ops->ndo_change_mtu)
4792 err = ops->ndo_change_mtu(dev, new_mtu);
4797 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4800 EXPORT_SYMBOL(dev_set_mtu);
4803 * dev_set_group - Change group this device belongs to
4805 * @new_group: group this device should belong to
4807 void dev_set_group(struct net_device *dev, int new_group)
4809 dev->group = new_group;
4811 EXPORT_SYMBOL(dev_set_group);
4814 * dev_set_mac_address - Change Media Access Control Address
4818 * Change the hardware (MAC) address of the device
4820 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4822 const struct net_device_ops *ops = dev->netdev_ops;
4825 if (!ops->ndo_set_mac_address)
4827 if (sa->sa_family != dev->type)
4829 if (!netif_device_present(dev))
4831 err = ops->ndo_set_mac_address(dev, sa);
4834 dev->addr_assign_type = NET_ADDR_SET;
4835 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4836 add_device_randomness(dev->dev_addr, dev->addr_len);
4839 EXPORT_SYMBOL(dev_set_mac_address);
4842 * dev_change_carrier - Change device carrier
4844 * @new_carrier: new value
4846 * Change device carrier
4848 int dev_change_carrier(struct net_device *dev, bool new_carrier)
4850 const struct net_device_ops *ops = dev->netdev_ops;
4852 if (!ops->ndo_change_carrier)
4854 if (!netif_device_present(dev))
4856 return ops->ndo_change_carrier(dev, new_carrier);
4858 EXPORT_SYMBOL(dev_change_carrier);
4861 * dev_new_index - allocate an ifindex
4862 * @net: the applicable net namespace
4864 * Returns a suitable unique value for a new device interface
4865 * number. The caller must hold the rtnl semaphore or the
4866 * dev_base_lock to be sure it remains unique.
4868 static int dev_new_index(struct net *net)
4870 int ifindex = net->ifindex;
4874 if (!__dev_get_by_index(net, ifindex))
4875 return net->ifindex = ifindex;
4879 /* Delayed registration/unregisteration */
4880 static LIST_HEAD(net_todo_list);
4882 static void net_set_todo(struct net_device *dev)
4884 list_add_tail(&dev->todo_list, &net_todo_list);
4887 static void rollback_registered_many(struct list_head *head)
4889 struct net_device *dev, *tmp;
4891 BUG_ON(dev_boot_phase);
4894 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4895 /* Some devices call without registering
4896 * for initialization unwind. Remove those
4897 * devices and proceed with the remaining.
4899 if (dev->reg_state == NETREG_UNINITIALIZED) {
4900 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
4904 list_del(&dev->unreg_list);
4907 dev->dismantle = true;
4908 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4911 /* If device is running, close it first. */
4912 dev_close_many(head);
4914 list_for_each_entry(dev, head, unreg_list) {
4915 /* And unlink it from device chain. */
4916 unlist_netdevice(dev);
4918 dev->reg_state = NETREG_UNREGISTERING;
4923 list_for_each_entry(dev, head, unreg_list) {
4924 /* Shutdown queueing discipline. */
4928 /* Notify protocols, that we are about to destroy
4929 this device. They should clean all the things.
4931 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4933 if (!dev->rtnl_link_ops ||
4934 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4935 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4938 * Flush the unicast and multicast chains
4943 if (dev->netdev_ops->ndo_uninit)
4944 dev->netdev_ops->ndo_uninit(dev);
4946 /* Notifier chain MUST detach us all upper devices. */
4947 WARN_ON(netdev_has_any_upper_dev(dev));
4949 /* Remove entries from kobject tree */
4950 netdev_unregister_kobject(dev);
4952 /* Remove XPS queueing entries */
4953 netif_reset_xps_queues_gt(dev, 0);
4959 list_for_each_entry(dev, head, unreg_list)
4963 static void rollback_registered(struct net_device *dev)
4967 list_add(&dev->unreg_list, &single);
4968 rollback_registered_many(&single);
4972 static netdev_features_t netdev_fix_features(struct net_device *dev,
4973 netdev_features_t features)
4975 /* Fix illegal checksum combinations */
4976 if ((features & NETIF_F_HW_CSUM) &&
4977 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4978 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
4979 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4982 /* TSO requires that SG is present as well. */
4983 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
4984 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
4985 features &= ~NETIF_F_ALL_TSO;
4988 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
4989 !(features & NETIF_F_IP_CSUM)) {
4990 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
4991 features &= ~NETIF_F_TSO;
4992 features &= ~NETIF_F_TSO_ECN;
4995 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
4996 !(features & NETIF_F_IPV6_CSUM)) {
4997 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
4998 features &= ~NETIF_F_TSO6;
5001 /* TSO ECN requires that TSO is present as well. */
5002 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5003 features &= ~NETIF_F_TSO_ECN;
5005 /* Software GSO depends on SG. */
5006 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5007 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5008 features &= ~NETIF_F_GSO;
5011 /* UFO needs SG and checksumming */
5012 if (features & NETIF_F_UFO) {
5013 /* maybe split UFO into V4 and V6? */
5014 if (!((features & NETIF_F_GEN_CSUM) ||
5015 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5016 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5018 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5019 features &= ~NETIF_F_UFO;
5022 if (!(features & NETIF_F_SG)) {
5024 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5025 features &= ~NETIF_F_UFO;
5032 int __netdev_update_features(struct net_device *dev)
5034 netdev_features_t features;
5039 features = netdev_get_wanted_features(dev);
5041 if (dev->netdev_ops->ndo_fix_features)
5042 features = dev->netdev_ops->ndo_fix_features(dev, features);
5044 /* driver might be less strict about feature dependencies */
5045 features = netdev_fix_features(dev, features);
5047 if (dev->features == features)
5050 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5051 &dev->features, &features);
5053 if (dev->netdev_ops->ndo_set_features)
5054 err = dev->netdev_ops->ndo_set_features(dev, features);
5056 if (unlikely(err < 0)) {
5058 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5059 err, &features, &dev->features);
5064 dev->features = features;
5070 * netdev_update_features - recalculate device features
5071 * @dev: the device to check
5073 * Recalculate dev->features set and send notifications if it
5074 * has changed. Should be called after driver or hardware dependent
5075 * conditions might have changed that influence the features.
5077 void netdev_update_features(struct net_device *dev)
5079 if (__netdev_update_features(dev))
5080 netdev_features_change(dev);
5082 EXPORT_SYMBOL(netdev_update_features);
5085 * netdev_change_features - recalculate device features
5086 * @dev: the device to check
5088 * Recalculate dev->features set and send notifications even
5089 * if they have not changed. Should be called instead of
5090 * netdev_update_features() if also dev->vlan_features might
5091 * have changed to allow the changes to be propagated to stacked
5094 void netdev_change_features(struct net_device *dev)
5096 __netdev_update_features(dev);
5097 netdev_features_change(dev);
5099 EXPORT_SYMBOL(netdev_change_features);
5102 * netif_stacked_transfer_operstate - transfer operstate
5103 * @rootdev: the root or lower level device to transfer state from
5104 * @dev: the device to transfer operstate to
5106 * Transfer operational state from root to device. This is normally
5107 * called when a stacking relationship exists between the root
5108 * device and the device(a leaf device).
5110 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5111 struct net_device *dev)
5113 if (rootdev->operstate == IF_OPER_DORMANT)
5114 netif_dormant_on(dev);
5116 netif_dormant_off(dev);
5118 if (netif_carrier_ok(rootdev)) {
5119 if (!netif_carrier_ok(dev))
5120 netif_carrier_on(dev);
5122 if (netif_carrier_ok(dev))
5123 netif_carrier_off(dev);
5126 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5129 static int netif_alloc_rx_queues(struct net_device *dev)
5131 unsigned int i, count = dev->num_rx_queues;
5132 struct netdev_rx_queue *rx;
5136 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5142 for (i = 0; i < count; i++)
5148 static void netdev_init_one_queue(struct net_device *dev,
5149 struct netdev_queue *queue, void *_unused)
5151 /* Initialize queue lock */
5152 spin_lock_init(&queue->_xmit_lock);
5153 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5154 queue->xmit_lock_owner = -1;
5155 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5158 dql_init(&queue->dql, HZ);
5162 static int netif_alloc_netdev_queues(struct net_device *dev)
5164 unsigned int count = dev->num_tx_queues;
5165 struct netdev_queue *tx;
5169 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5175 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5176 spin_lock_init(&dev->tx_global_lock);
5182 * register_netdevice - register a network device
5183 * @dev: device to register
5185 * Take a completed network device structure and add it to the kernel
5186 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5187 * chain. 0 is returned on success. A negative errno code is returned
5188 * on a failure to set up the device, or if the name is a duplicate.
5190 * Callers must hold the rtnl semaphore. You may want
5191 * register_netdev() instead of this.
5194 * The locking appears insufficient to guarantee two parallel registers
5195 * will not get the same name.
5198 int register_netdevice(struct net_device *dev)
5201 struct net *net = dev_net(dev);
5203 BUG_ON(dev_boot_phase);
5208 /* When net_device's are persistent, this will be fatal. */
5209 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5212 spin_lock_init(&dev->addr_list_lock);
5213 netdev_set_addr_lockdep_class(dev);
5217 ret = dev_get_valid_name(net, dev, dev->name);
5221 /* Init, if this function is available */
5222 if (dev->netdev_ops->ndo_init) {
5223 ret = dev->netdev_ops->ndo_init(dev);
5231 if (((dev->hw_features | dev->features) &
5232 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5233 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5234 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5235 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5242 dev->ifindex = dev_new_index(net);
5243 else if (__dev_get_by_index(net, dev->ifindex))
5246 if (dev->iflink == -1)
5247 dev->iflink = dev->ifindex;
5249 /* Transfer changeable features to wanted_features and enable
5250 * software offloads (GSO and GRO).
5252 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5253 dev->features |= NETIF_F_SOFT_FEATURES;
5254 dev->wanted_features = dev->features & dev->hw_features;
5256 /* Turn on no cache copy if HW is doing checksum */
5257 if (!(dev->flags & IFF_LOOPBACK)) {
5258 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5259 if (dev->features & NETIF_F_ALL_CSUM) {
5260 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5261 dev->features |= NETIF_F_NOCACHE_COPY;
5265 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5267 dev->vlan_features |= NETIF_F_HIGHDMA;
5269 /* Make NETIF_F_SG inheritable to tunnel devices.
5271 dev->hw_enc_features |= NETIF_F_SG;
5273 /* Make NETIF_F_SG inheritable to MPLS.
5275 dev->mpls_features |= NETIF_F_SG;
5277 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5278 ret = notifier_to_errno(ret);
5282 ret = netdev_register_kobject(dev);
5285 dev->reg_state = NETREG_REGISTERED;
5287 __netdev_update_features(dev);
5290 * Default initial state at registry is that the
5291 * device is present.
5294 set_bit(__LINK_STATE_PRESENT, &dev->state);
5296 linkwatch_init_dev(dev);
5298 dev_init_scheduler(dev);
5300 list_netdevice(dev);
5301 add_device_randomness(dev->dev_addr, dev->addr_len);
5303 /* If the device has permanent device address, driver should
5304 * set dev_addr and also addr_assign_type should be set to
5305 * NET_ADDR_PERM (default value).
5307 if (dev->addr_assign_type == NET_ADDR_PERM)
5308 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5310 /* Notify protocols, that a new device appeared. */
5311 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5312 ret = notifier_to_errno(ret);
5314 rollback_registered(dev);
5315 dev->reg_state = NETREG_UNREGISTERED;
5318 * Prevent userspace races by waiting until the network
5319 * device is fully setup before sending notifications.
5321 if (!dev->rtnl_link_ops ||
5322 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5323 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5329 if (dev->netdev_ops->ndo_uninit)
5330 dev->netdev_ops->ndo_uninit(dev);
5333 EXPORT_SYMBOL(register_netdevice);
5336 * init_dummy_netdev - init a dummy network device for NAPI
5337 * @dev: device to init
5339 * This takes a network device structure and initialize the minimum
5340 * amount of fields so it can be used to schedule NAPI polls without
5341 * registering a full blown interface. This is to be used by drivers
5342 * that need to tie several hardware interfaces to a single NAPI
5343 * poll scheduler due to HW limitations.
5345 int init_dummy_netdev(struct net_device *dev)
5347 /* Clear everything. Note we don't initialize spinlocks
5348 * are they aren't supposed to be taken by any of the
5349 * NAPI code and this dummy netdev is supposed to be
5350 * only ever used for NAPI polls
5352 memset(dev, 0, sizeof(struct net_device));
5354 /* make sure we BUG if trying to hit standard
5355 * register/unregister code path
5357 dev->reg_state = NETREG_DUMMY;
5359 /* NAPI wants this */
5360 INIT_LIST_HEAD(&dev->napi_list);
5362 /* a dummy interface is started by default */
5363 set_bit(__LINK_STATE_PRESENT, &dev->state);
5364 set_bit(__LINK_STATE_START, &dev->state);
5366 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5367 * because users of this 'device' dont need to change
5373 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5377 * register_netdev - register a network device
5378 * @dev: device to register
5380 * Take a completed network device structure and add it to the kernel
5381 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5382 * chain. 0 is returned on success. A negative errno code is returned
5383 * on a failure to set up the device, or if the name is a duplicate.
5385 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5386 * and expands the device name if you passed a format string to
5389 int register_netdev(struct net_device *dev)
5394 err = register_netdevice(dev);
5398 EXPORT_SYMBOL(register_netdev);
5400 int netdev_refcnt_read(const struct net_device *dev)
5404 for_each_possible_cpu(i)
5405 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5408 EXPORT_SYMBOL(netdev_refcnt_read);
5411 * netdev_wait_allrefs - wait until all references are gone.
5412 * @dev: target net_device
5414 * This is called when unregistering network devices.
5416 * Any protocol or device that holds a reference should register
5417 * for netdevice notification, and cleanup and put back the
5418 * reference if they receive an UNREGISTER event.
5419 * We can get stuck here if buggy protocols don't correctly
5422 static void netdev_wait_allrefs(struct net_device *dev)
5424 unsigned long rebroadcast_time, warning_time;
5427 linkwatch_forget_dev(dev);
5429 rebroadcast_time = warning_time = jiffies;
5430 refcnt = netdev_refcnt_read(dev);
5432 while (refcnt != 0) {
5433 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5436 /* Rebroadcast unregister notification */
5437 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5443 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5444 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5446 /* We must not have linkwatch events
5447 * pending on unregister. If this
5448 * happens, we simply run the queue
5449 * unscheduled, resulting in a noop
5452 linkwatch_run_queue();
5457 rebroadcast_time = jiffies;
5462 refcnt = netdev_refcnt_read(dev);
5464 if (time_after(jiffies, warning_time + 10 * HZ)) {
5465 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5467 warning_time = jiffies;
5476 * register_netdevice(x1);
5477 * register_netdevice(x2);
5479 * unregister_netdevice(y1);
5480 * unregister_netdevice(y2);
5486 * We are invoked by rtnl_unlock().
5487 * This allows us to deal with problems:
5488 * 1) We can delete sysfs objects which invoke hotplug
5489 * without deadlocking with linkwatch via keventd.
5490 * 2) Since we run with the RTNL semaphore not held, we can sleep
5491 * safely in order to wait for the netdev refcnt to drop to zero.
5493 * We must not return until all unregister events added during
5494 * the interval the lock was held have been completed.
5496 void netdev_run_todo(void)
5498 struct list_head list;
5500 /* Snapshot list, allow later requests */
5501 list_replace_init(&net_todo_list, &list);
5506 /* Wait for rcu callbacks to finish before next phase */
5507 if (!list_empty(&list))
5510 while (!list_empty(&list)) {
5511 struct net_device *dev
5512 = list_first_entry(&list, struct net_device, todo_list);
5513 list_del(&dev->todo_list);
5516 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5519 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5520 pr_err("network todo '%s' but state %d\n",
5521 dev->name, dev->reg_state);
5526 dev->reg_state = NETREG_UNREGISTERED;
5528 on_each_cpu(flush_backlog, dev, 1);
5530 netdev_wait_allrefs(dev);
5533 BUG_ON(netdev_refcnt_read(dev));
5534 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5535 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5536 WARN_ON(dev->dn_ptr);
5538 if (dev->destructor)
5539 dev->destructor(dev);
5541 /* Free network device */
5542 kobject_put(&dev->dev.kobj);
5546 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5547 * fields in the same order, with only the type differing.
5549 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5550 const struct net_device_stats *netdev_stats)
5552 #if BITS_PER_LONG == 64
5553 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5554 memcpy(stats64, netdev_stats, sizeof(*stats64));
5556 size_t i, n = sizeof(*stats64) / sizeof(u64);
5557 const unsigned long *src = (const unsigned long *)netdev_stats;
5558 u64 *dst = (u64 *)stats64;
5560 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5561 sizeof(*stats64) / sizeof(u64));
5562 for (i = 0; i < n; i++)
5566 EXPORT_SYMBOL(netdev_stats_to_stats64);
5569 * dev_get_stats - get network device statistics
5570 * @dev: device to get statistics from
5571 * @storage: place to store stats
5573 * Get network statistics from device. Return @storage.
5574 * The device driver may provide its own method by setting
5575 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5576 * otherwise the internal statistics structure is used.
5578 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5579 struct rtnl_link_stats64 *storage)
5581 const struct net_device_ops *ops = dev->netdev_ops;
5583 if (ops->ndo_get_stats64) {
5584 memset(storage, 0, sizeof(*storage));
5585 ops->ndo_get_stats64(dev, storage);
5586 } else if (ops->ndo_get_stats) {
5587 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5589 netdev_stats_to_stats64(storage, &dev->stats);
5591 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5594 EXPORT_SYMBOL(dev_get_stats);
5596 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5598 struct netdev_queue *queue = dev_ingress_queue(dev);
5600 #ifdef CONFIG_NET_CLS_ACT
5603 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5606 netdev_init_one_queue(dev, queue, NULL);
5607 queue->qdisc = &noop_qdisc;
5608 queue->qdisc_sleeping = &noop_qdisc;
5609 rcu_assign_pointer(dev->ingress_queue, queue);
5614 static const struct ethtool_ops default_ethtool_ops;
5616 void netdev_set_default_ethtool_ops(struct net_device *dev,
5617 const struct ethtool_ops *ops)
5619 if (dev->ethtool_ops == &default_ethtool_ops)
5620 dev->ethtool_ops = ops;
5622 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
5625 * alloc_netdev_mqs - allocate network device
5626 * @sizeof_priv: size of private data to allocate space for
5627 * @name: device name format string
5628 * @setup: callback to initialize device
5629 * @txqs: the number of TX subqueues to allocate
5630 * @rxqs: the number of RX subqueues to allocate
5632 * Allocates a struct net_device with private data area for driver use
5633 * and performs basic initialization. Also allocates subquue structs
5634 * for each queue on the device.
5636 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5637 void (*setup)(struct net_device *),
5638 unsigned int txqs, unsigned int rxqs)
5640 struct net_device *dev;
5642 struct net_device *p;
5644 BUG_ON(strlen(name) >= sizeof(dev->name));
5647 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5653 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5658 alloc_size = sizeof(struct net_device);
5660 /* ensure 32-byte alignment of private area */
5661 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5662 alloc_size += sizeof_priv;
5664 /* ensure 32-byte alignment of whole construct */
5665 alloc_size += NETDEV_ALIGN - 1;
5667 p = kzalloc(alloc_size, GFP_KERNEL);
5671 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5672 dev->padded = (char *)dev - (char *)p;
5674 dev->pcpu_refcnt = alloc_percpu(int);
5675 if (!dev->pcpu_refcnt)
5678 if (dev_addr_init(dev))
5684 dev_net_set(dev, &init_net);
5686 dev->gso_max_size = GSO_MAX_SIZE;
5687 dev->gso_max_segs = GSO_MAX_SEGS;
5689 INIT_LIST_HEAD(&dev->napi_list);
5690 INIT_LIST_HEAD(&dev->unreg_list);
5691 INIT_LIST_HEAD(&dev->link_watch_list);
5692 INIT_LIST_HEAD(&dev->upper_dev_list);
5693 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5696 dev->num_tx_queues = txqs;
5697 dev->real_num_tx_queues = txqs;
5698 if (netif_alloc_netdev_queues(dev))
5702 dev->num_rx_queues = rxqs;
5703 dev->real_num_rx_queues = rxqs;
5704 if (netif_alloc_rx_queues(dev))
5708 strcpy(dev->name, name);
5709 dev->group = INIT_NETDEV_GROUP;
5710 if (!dev->ethtool_ops)
5711 dev->ethtool_ops = &default_ethtool_ops;
5719 free_percpu(dev->pcpu_refcnt);
5729 EXPORT_SYMBOL(alloc_netdev_mqs);
5732 * free_netdev - free network device
5735 * This function does the last stage of destroying an allocated device
5736 * interface. The reference to the device object is released.
5737 * If this is the last reference then it will be freed.
5739 void free_netdev(struct net_device *dev)
5741 struct napi_struct *p, *n;
5743 release_net(dev_net(dev));
5750 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
5752 /* Flush device addresses */
5753 dev_addr_flush(dev);
5755 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5758 free_percpu(dev->pcpu_refcnt);
5759 dev->pcpu_refcnt = NULL;
5761 /* Compatibility with error handling in drivers */
5762 if (dev->reg_state == NETREG_UNINITIALIZED) {
5763 kfree((char *)dev - dev->padded);
5767 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5768 dev->reg_state = NETREG_RELEASED;
5770 /* will free via device release */
5771 put_device(&dev->dev);
5773 EXPORT_SYMBOL(free_netdev);
5776 * synchronize_net - Synchronize with packet receive processing
5778 * Wait for packets currently being received to be done.
5779 * Does not block later packets from starting.
5781 void synchronize_net(void)
5784 if (rtnl_is_locked())
5785 synchronize_rcu_expedited();
5789 EXPORT_SYMBOL(synchronize_net);
5792 * unregister_netdevice_queue - remove device from the kernel
5796 * This function shuts down a device interface and removes it
5797 * from the kernel tables.
5798 * If head not NULL, device is queued to be unregistered later.
5800 * Callers must hold the rtnl semaphore. You may want
5801 * unregister_netdev() instead of this.
5804 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5809 list_move_tail(&dev->unreg_list, head);
5811 rollback_registered(dev);
5812 /* Finish processing unregister after unlock */
5816 EXPORT_SYMBOL(unregister_netdevice_queue);
5819 * unregister_netdevice_many - unregister many devices
5820 * @head: list of devices
5822 void unregister_netdevice_many(struct list_head *head)
5824 struct net_device *dev;
5826 if (!list_empty(head)) {
5827 rollback_registered_many(head);
5828 list_for_each_entry(dev, head, unreg_list)
5832 EXPORT_SYMBOL(unregister_netdevice_many);
5835 * unregister_netdev - remove device from the kernel
5838 * This function shuts down a device interface and removes it
5839 * from the kernel tables.
5841 * This is just a wrapper for unregister_netdevice that takes
5842 * the rtnl semaphore. In general you want to use this and not
5843 * unregister_netdevice.
5845 void unregister_netdev(struct net_device *dev)
5848 unregister_netdevice(dev);
5851 EXPORT_SYMBOL(unregister_netdev);
5854 * dev_change_net_namespace - move device to different nethost namespace
5856 * @net: network namespace
5857 * @pat: If not NULL name pattern to try if the current device name
5858 * is already taken in the destination network namespace.
5860 * This function shuts down a device interface and moves it
5861 * to a new network namespace. On success 0 is returned, on
5862 * a failure a netagive errno code is returned.
5864 * Callers must hold the rtnl semaphore.
5867 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5873 /* Don't allow namespace local devices to be moved. */
5875 if (dev->features & NETIF_F_NETNS_LOCAL)
5878 /* Ensure the device has been registrered */
5879 if (dev->reg_state != NETREG_REGISTERED)
5882 /* Get out if there is nothing todo */
5884 if (net_eq(dev_net(dev), net))
5887 /* Pick the destination device name, and ensure
5888 * we can use it in the destination network namespace.
5891 if (__dev_get_by_name(net, dev->name)) {
5892 /* We get here if we can't use the current device name */
5895 if (dev_get_valid_name(net, dev, pat) < 0)
5900 * And now a mini version of register_netdevice unregister_netdevice.
5903 /* If device is running close it first. */
5906 /* And unlink it from device chain */
5908 unlist_netdevice(dev);
5912 /* Shutdown queueing discipline. */
5915 /* Notify protocols, that we are about to destroy
5916 this device. They should clean all the things.
5918 Note that dev->reg_state stays at NETREG_REGISTERED.
5919 This is wanted because this way 8021q and macvlan know
5920 the device is just moving and can keep their slaves up.
5922 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5924 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5925 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5928 * Flush the unicast and multicast chains
5933 /* Send a netdev-removed uevent to the old namespace */
5934 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
5936 /* Actually switch the network namespace */
5937 dev_net_set(dev, net);
5939 /* If there is an ifindex conflict assign a new one */
5940 if (__dev_get_by_index(net, dev->ifindex)) {
5941 int iflink = (dev->iflink == dev->ifindex);
5942 dev->ifindex = dev_new_index(net);
5944 dev->iflink = dev->ifindex;
5947 /* Send a netdev-add uevent to the new namespace */
5948 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
5950 /* Fixup kobjects */
5951 err = device_rename(&dev->dev, dev->name);
5954 /* Add the device back in the hashes */
5955 list_netdevice(dev);
5957 /* Notify protocols, that a new device appeared. */
5958 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5961 * Prevent userspace races by waiting until the network
5962 * device is fully setup before sending notifications.
5964 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5971 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5973 static int dev_cpu_callback(struct notifier_block *nfb,
5974 unsigned long action,
5977 struct sk_buff **list_skb;
5978 struct sk_buff *skb;
5979 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5980 struct softnet_data *sd, *oldsd;
5982 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5985 local_irq_disable();
5986 cpu = smp_processor_id();
5987 sd = &per_cpu(softnet_data, cpu);
5988 oldsd = &per_cpu(softnet_data, oldcpu);
5990 /* Find end of our completion_queue. */
5991 list_skb = &sd->completion_queue;
5993 list_skb = &(*list_skb)->next;
5994 /* Append completion queue from offline CPU. */
5995 *list_skb = oldsd->completion_queue;
5996 oldsd->completion_queue = NULL;
5998 /* Append output queue from offline CPU. */
5999 if (oldsd->output_queue) {
6000 *sd->output_queue_tailp = oldsd->output_queue;
6001 sd->output_queue_tailp = oldsd->output_queue_tailp;
6002 oldsd->output_queue = NULL;
6003 oldsd->output_queue_tailp = &oldsd->output_queue;
6005 /* Append NAPI poll list from offline CPU. */
6006 if (!list_empty(&oldsd->poll_list)) {
6007 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6008 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6011 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6014 /* Process offline CPU's input_pkt_queue */
6015 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6017 input_queue_head_incr(oldsd);
6019 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6021 input_queue_head_incr(oldsd);
6029 * netdev_increment_features - increment feature set by one
6030 * @all: current feature set
6031 * @one: new feature set
6032 * @mask: mask feature set
6034 * Computes a new feature set after adding a device with feature set
6035 * @one to the master device with current feature set @all. Will not
6036 * enable anything that is off in @mask. Returns the new feature set.
6038 netdev_features_t netdev_increment_features(netdev_features_t all,
6039 netdev_features_t one, netdev_features_t mask)
6041 if (mask & NETIF_F_GEN_CSUM)
6042 mask |= NETIF_F_ALL_CSUM;
6043 mask |= NETIF_F_VLAN_CHALLENGED;
6045 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6046 all &= one | ~NETIF_F_ALL_FOR_ALL;
6048 /* If one device supports hw checksumming, set for all. */
6049 if (all & NETIF_F_GEN_CSUM)
6050 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6054 EXPORT_SYMBOL(netdev_increment_features);
6056 static struct hlist_head *netdev_create_hash(void)
6059 struct hlist_head *hash;
6061 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6063 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6064 INIT_HLIST_HEAD(&hash[i]);
6069 /* Initialize per network namespace state */
6070 static int __net_init netdev_init(struct net *net)
6072 if (net != &init_net)
6073 INIT_LIST_HEAD(&net->dev_base_head);
6075 net->dev_name_head = netdev_create_hash();
6076 if (net->dev_name_head == NULL)
6079 net->dev_index_head = netdev_create_hash();
6080 if (net->dev_index_head == NULL)
6086 kfree(net->dev_name_head);
6092 * netdev_drivername - network driver for the device
6093 * @dev: network device
6095 * Determine network driver for device.
6097 const char *netdev_drivername(const struct net_device *dev)
6099 const struct device_driver *driver;
6100 const struct device *parent;
6101 const char *empty = "";
6103 parent = dev->dev.parent;
6107 driver = parent->driver;
6108 if (driver && driver->name)
6109 return driver->name;
6113 static int __netdev_printk(const char *level, const struct net_device *dev,
6114 struct va_format *vaf)
6118 if (dev && dev->dev.parent) {
6119 r = dev_printk_emit(level[1] - '0',
6122 dev_driver_string(dev->dev.parent),
6123 dev_name(dev->dev.parent),
6124 netdev_name(dev), vaf);
6126 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6128 r = printk("%s(NULL net_device): %pV", level, vaf);
6134 int netdev_printk(const char *level, const struct net_device *dev,
6135 const char *format, ...)
6137 struct va_format vaf;
6141 va_start(args, format);
6146 r = __netdev_printk(level, dev, &vaf);
6152 EXPORT_SYMBOL(netdev_printk);
6154 #define define_netdev_printk_level(func, level) \
6155 int func(const struct net_device *dev, const char *fmt, ...) \
6158 struct va_format vaf; \
6161 va_start(args, fmt); \
6166 r = __netdev_printk(level, dev, &vaf); \
6172 EXPORT_SYMBOL(func);
6174 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6175 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6176 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6177 define_netdev_printk_level(netdev_err, KERN_ERR);
6178 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6179 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6180 define_netdev_printk_level(netdev_info, KERN_INFO);
6182 static void __net_exit netdev_exit(struct net *net)
6184 kfree(net->dev_name_head);
6185 kfree(net->dev_index_head);
6188 static struct pernet_operations __net_initdata netdev_net_ops = {
6189 .init = netdev_init,
6190 .exit = netdev_exit,
6193 static void __net_exit default_device_exit(struct net *net)
6195 struct net_device *dev, *aux;
6197 * Push all migratable network devices back to the
6198 * initial network namespace
6201 for_each_netdev_safe(net, dev, aux) {
6203 char fb_name[IFNAMSIZ];
6205 /* Ignore unmoveable devices (i.e. loopback) */
6206 if (dev->features & NETIF_F_NETNS_LOCAL)
6209 /* Leave virtual devices for the generic cleanup */
6210 if (dev->rtnl_link_ops)
6213 /* Push remaining network devices to init_net */
6214 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6215 err = dev_change_net_namespace(dev, &init_net, fb_name);
6217 pr_emerg("%s: failed to move %s to init_net: %d\n",
6218 __func__, dev->name, err);
6225 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6227 /* At exit all network devices most be removed from a network
6228 * namespace. Do this in the reverse order of registration.
6229 * Do this across as many network namespaces as possible to
6230 * improve batching efficiency.
6232 struct net_device *dev;
6234 LIST_HEAD(dev_kill_list);
6237 list_for_each_entry(net, net_list, exit_list) {
6238 for_each_netdev_reverse(net, dev) {
6239 if (dev->rtnl_link_ops)
6240 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6242 unregister_netdevice_queue(dev, &dev_kill_list);
6245 unregister_netdevice_many(&dev_kill_list);
6246 list_del(&dev_kill_list);
6250 static struct pernet_operations __net_initdata default_device_ops = {
6251 .exit = default_device_exit,
6252 .exit_batch = default_device_exit_batch,
6256 * Initialize the DEV module. At boot time this walks the device list and
6257 * unhooks any devices that fail to initialise (normally hardware not
6258 * present) and leaves us with a valid list of present and active devices.
6263 * This is called single threaded during boot, so no need
6264 * to take the rtnl semaphore.
6266 static int __init net_dev_init(void)
6268 int i, rc = -ENOMEM;
6270 BUG_ON(!dev_boot_phase);
6272 if (dev_proc_init())
6275 if (netdev_kobject_init())
6278 INIT_LIST_HEAD(&ptype_all);
6279 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6280 INIT_LIST_HEAD(&ptype_base[i]);
6282 INIT_LIST_HEAD(&offload_base);
6284 if (register_pernet_subsys(&netdev_net_ops))
6288 * Initialise the packet receive queues.
6291 for_each_possible_cpu(i) {
6292 struct softnet_data *sd = &per_cpu(softnet_data, i);
6294 memset(sd, 0, sizeof(*sd));
6295 skb_queue_head_init(&sd->input_pkt_queue);
6296 skb_queue_head_init(&sd->process_queue);
6297 sd->completion_queue = NULL;
6298 INIT_LIST_HEAD(&sd->poll_list);
6299 sd->output_queue = NULL;
6300 sd->output_queue_tailp = &sd->output_queue;
6302 sd->csd.func = rps_trigger_softirq;
6308 sd->backlog.poll = process_backlog;
6309 sd->backlog.weight = weight_p;
6310 sd->backlog.gro_list = NULL;
6311 sd->backlog.gro_count = 0;
6313 #ifdef CONFIG_NET_FLOW_LIMIT
6314 sd->flow_limit = NULL;
6320 /* The loopback device is special if any other network devices
6321 * is present in a network namespace the loopback device must
6322 * be present. Since we now dynamically allocate and free the
6323 * loopback device ensure this invariant is maintained by
6324 * keeping the loopback device as the first device on the
6325 * list of network devices. Ensuring the loopback devices
6326 * is the first device that appears and the last network device
6329 if (register_pernet_device(&loopback_net_ops))
6332 if (register_pernet_device(&default_device_ops))
6335 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6336 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6338 hotcpu_notifier(dev_cpu_callback, 0);
6345 subsys_initcall(net_dev_init);