2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
121 #include <linux/ipv6.h>
122 #include <linux/in.h>
123 #include <linux/jhash.h>
124 #include <linux/random.h>
125 #include <trace/events/napi.h>
126 #include <trace/events/net.h>
127 #include <trace/events/skb.h>
128 #include <linux/pci.h>
129 #include <linux/inetdevice.h>
130 #include <linux/cpu_rmap.h>
131 #include <linux/static_key.h>
132 #include <linux/hashtable.h>
133 #include <linux/vmalloc.h>
134 #include <linux/if_macvlan.h>
136 #include "net-sysfs.h"
138 /* Instead of increasing this, you should create a hash table. */
139 #define MAX_GRO_SKBS 8
141 /* This should be increased if a protocol with a bigger head is added. */
142 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 static DEFINE_SPINLOCK(ptype_lock);
145 static DEFINE_SPINLOCK(offload_lock);
146 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
147 struct list_head ptype_all __read_mostly; /* Taps */
148 static struct list_head offload_base __read_mostly;
150 static int netif_rx_internal(struct sk_buff *skb);
153 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
156 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
158 * Writers must hold the rtnl semaphore while they loop through the
159 * dev_base_head list, and hold dev_base_lock for writing when they do the
160 * actual updates. This allows pure readers to access the list even
161 * while a writer is preparing to update it.
163 * To put it another way, dev_base_lock is held for writing only to
164 * protect against pure readers; the rtnl semaphore provides the
165 * protection against other writers.
167 * See, for example usages, register_netdevice() and
168 * unregister_netdevice(), which must be called with the rtnl
171 DEFINE_RWLOCK(dev_base_lock);
172 EXPORT_SYMBOL(dev_base_lock);
174 /* protects napi_hash addition/deletion and napi_gen_id */
175 static DEFINE_SPINLOCK(napi_hash_lock);
177 static unsigned int napi_gen_id;
178 static DEFINE_HASHTABLE(napi_hash, 8);
180 static seqcount_t devnet_rename_seq;
182 static inline void dev_base_seq_inc(struct net *net)
184 while (++net->dev_base_seq == 0);
187 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
189 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
191 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
194 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
196 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
199 static inline void rps_lock(struct softnet_data *sd)
202 spin_lock(&sd->input_pkt_queue.lock);
206 static inline void rps_unlock(struct softnet_data *sd)
209 spin_unlock(&sd->input_pkt_queue.lock);
213 /* Device list insertion */
214 static void list_netdevice(struct net_device *dev)
216 struct net *net = dev_net(dev);
220 write_lock_bh(&dev_base_lock);
221 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
222 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
223 hlist_add_head_rcu(&dev->index_hlist,
224 dev_index_hash(net, dev->ifindex));
225 write_unlock_bh(&dev_base_lock);
227 dev_base_seq_inc(net);
230 /* Device list removal
231 * caller must respect a RCU grace period before freeing/reusing dev
233 static void unlist_netdevice(struct net_device *dev)
237 /* Unlink dev from the device chain */
238 write_lock_bh(&dev_base_lock);
239 list_del_rcu(&dev->dev_list);
240 hlist_del_rcu(&dev->name_hlist);
241 hlist_del_rcu(&dev->index_hlist);
242 write_unlock_bh(&dev_base_lock);
244 dev_base_seq_inc(dev_net(dev));
251 static RAW_NOTIFIER_HEAD(netdev_chain);
254 * Device drivers call our routines to queue packets here. We empty the
255 * queue in the local softnet handler.
258 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
259 EXPORT_PER_CPU_SYMBOL(softnet_data);
261 #ifdef CONFIG_LOCKDEP
263 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
264 * according to dev->type
266 static const unsigned short netdev_lock_type[] =
267 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
268 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
269 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
270 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
271 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
272 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
273 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
274 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
275 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
276 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
277 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
278 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
279 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
280 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
281 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
283 static const char *const netdev_lock_name[] =
284 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
285 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
286 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
287 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
288 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
289 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
290 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
291 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
292 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
293 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
294 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
295 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
296 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
297 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
298 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
300 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
301 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
303 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
307 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
308 if (netdev_lock_type[i] == dev_type)
310 /* the last key is used by default */
311 return ARRAY_SIZE(netdev_lock_type) - 1;
314 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
315 unsigned short dev_type)
319 i = netdev_lock_pos(dev_type);
320 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
321 netdev_lock_name[i]);
324 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
328 i = netdev_lock_pos(dev->type);
329 lockdep_set_class_and_name(&dev->addr_list_lock,
330 &netdev_addr_lock_key[i],
331 netdev_lock_name[i]);
334 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
335 unsigned short dev_type)
338 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
343 /*******************************************************************************
345 Protocol management and registration routines
347 *******************************************************************************/
350 * Add a protocol ID to the list. Now that the input handler is
351 * smarter we can dispense with all the messy stuff that used to be
354 * BEWARE!!! Protocol handlers, mangling input packets,
355 * MUST BE last in hash buckets and checking protocol handlers
356 * MUST start from promiscuous ptype_all chain in net_bh.
357 * It is true now, do not change it.
358 * Explanation follows: if protocol handler, mangling packet, will
359 * be the first on list, it is not able to sense, that packet
360 * is cloned and should be copied-on-write, so that it will
361 * change it and subsequent readers will get broken packet.
365 static inline struct list_head *ptype_head(const struct packet_type *pt)
367 if (pt->type == htons(ETH_P_ALL))
370 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
374 * dev_add_pack - add packet handler
375 * @pt: packet type declaration
377 * Add a protocol handler to the networking stack. The passed &packet_type
378 * is linked into kernel lists and may not be freed until it has been
379 * removed from the kernel lists.
381 * This call does not sleep therefore it can not
382 * guarantee all CPU's that are in middle of receiving packets
383 * will see the new packet type (until the next received packet).
386 void dev_add_pack(struct packet_type *pt)
388 struct list_head *head = ptype_head(pt);
390 spin_lock(&ptype_lock);
391 list_add_rcu(&pt->list, head);
392 spin_unlock(&ptype_lock);
394 EXPORT_SYMBOL(dev_add_pack);
397 * __dev_remove_pack - remove packet handler
398 * @pt: packet type declaration
400 * Remove a protocol handler that was previously added to the kernel
401 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
402 * from the kernel lists and can be freed or reused once this function
405 * The packet type might still be in use by receivers
406 * and must not be freed until after all the CPU's have gone
407 * through a quiescent state.
409 void __dev_remove_pack(struct packet_type *pt)
411 struct list_head *head = ptype_head(pt);
412 struct packet_type *pt1;
414 spin_lock(&ptype_lock);
416 list_for_each_entry(pt1, head, list) {
418 list_del_rcu(&pt->list);
423 pr_warn("dev_remove_pack: %p not found\n", pt);
425 spin_unlock(&ptype_lock);
427 EXPORT_SYMBOL(__dev_remove_pack);
430 * dev_remove_pack - remove packet handler
431 * @pt: packet type declaration
433 * Remove a protocol handler that was previously added to the kernel
434 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
435 * from the kernel lists and can be freed or reused once this function
438 * This call sleeps to guarantee that no CPU is looking at the packet
441 void dev_remove_pack(struct packet_type *pt)
443 __dev_remove_pack(pt);
447 EXPORT_SYMBOL(dev_remove_pack);
451 * dev_add_offload - register offload handlers
452 * @po: protocol offload declaration
454 * Add protocol offload handlers to the networking stack. The passed
455 * &proto_offload is linked into kernel lists and may not be freed until
456 * it has been removed from the kernel lists.
458 * This call does not sleep therefore it can not
459 * guarantee all CPU's that are in middle of receiving packets
460 * will see the new offload handlers (until the next received packet).
462 void dev_add_offload(struct packet_offload *po)
464 struct list_head *head = &offload_base;
466 spin_lock(&offload_lock);
467 list_add_rcu(&po->list, head);
468 spin_unlock(&offload_lock);
470 EXPORT_SYMBOL(dev_add_offload);
473 * __dev_remove_offload - remove offload handler
474 * @po: packet offload declaration
476 * Remove a protocol offload handler that was previously added to the
477 * kernel offload handlers by dev_add_offload(). The passed &offload_type
478 * is removed from the kernel lists and can be freed or reused once this
481 * The packet type might still be in use by receivers
482 * and must not be freed until after all the CPU's have gone
483 * through a quiescent state.
485 static void __dev_remove_offload(struct packet_offload *po)
487 struct list_head *head = &offload_base;
488 struct packet_offload *po1;
490 spin_lock(&offload_lock);
492 list_for_each_entry(po1, head, list) {
494 list_del_rcu(&po->list);
499 pr_warn("dev_remove_offload: %p not found\n", po);
501 spin_unlock(&offload_lock);
505 * dev_remove_offload - remove packet offload handler
506 * @po: packet offload declaration
508 * Remove a packet offload handler that was previously added to the kernel
509 * offload handlers by dev_add_offload(). The passed &offload_type is
510 * removed from the kernel lists and can be freed or reused once this
513 * This call sleeps to guarantee that no CPU is looking at the packet
516 void dev_remove_offload(struct packet_offload *po)
518 __dev_remove_offload(po);
522 EXPORT_SYMBOL(dev_remove_offload);
524 /******************************************************************************
526 Device Boot-time Settings Routines
528 *******************************************************************************/
530 /* Boot time configuration table */
531 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
534 * netdev_boot_setup_add - add new setup entry
535 * @name: name of the device
536 * @map: configured settings for the device
538 * Adds new setup entry to the dev_boot_setup list. The function
539 * returns 0 on error and 1 on success. This is a generic routine to
542 static int netdev_boot_setup_add(char *name, struct ifmap *map)
544 struct netdev_boot_setup *s;
548 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
549 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
550 memset(s[i].name, 0, sizeof(s[i].name));
551 strlcpy(s[i].name, name, IFNAMSIZ);
552 memcpy(&s[i].map, map, sizeof(s[i].map));
557 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
561 * netdev_boot_setup_check - check boot time settings
562 * @dev: the netdevice
564 * Check boot time settings for the device.
565 * The found settings are set for the device to be used
566 * later in the device probing.
567 * Returns 0 if no settings found, 1 if they are.
569 int netdev_boot_setup_check(struct net_device *dev)
571 struct netdev_boot_setup *s = dev_boot_setup;
574 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
575 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
576 !strcmp(dev->name, s[i].name)) {
577 dev->irq = s[i].map.irq;
578 dev->base_addr = s[i].map.base_addr;
579 dev->mem_start = s[i].map.mem_start;
580 dev->mem_end = s[i].map.mem_end;
586 EXPORT_SYMBOL(netdev_boot_setup_check);
590 * netdev_boot_base - get address from boot time settings
591 * @prefix: prefix for network device
592 * @unit: id for network device
594 * Check boot time settings for the base address of device.
595 * The found settings are set for the device to be used
596 * later in the device probing.
597 * Returns 0 if no settings found.
599 unsigned long netdev_boot_base(const char *prefix, int unit)
601 const struct netdev_boot_setup *s = dev_boot_setup;
605 sprintf(name, "%s%d", prefix, unit);
608 * If device already registered then return base of 1
609 * to indicate not to probe for this interface
611 if (__dev_get_by_name(&init_net, name))
614 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
615 if (!strcmp(name, s[i].name))
616 return s[i].map.base_addr;
621 * Saves at boot time configured settings for any netdevice.
623 int __init netdev_boot_setup(char *str)
628 str = get_options(str, ARRAY_SIZE(ints), ints);
633 memset(&map, 0, sizeof(map));
637 map.base_addr = ints[2];
639 map.mem_start = ints[3];
641 map.mem_end = ints[4];
643 /* Add new entry to the list */
644 return netdev_boot_setup_add(str, &map);
647 __setup("netdev=", netdev_boot_setup);
649 /*******************************************************************************
651 Device Interface Subroutines
653 *******************************************************************************/
656 * __dev_get_by_name - find a device by its name
657 * @net: the applicable net namespace
658 * @name: name to find
660 * Find an interface by name. Must be called under RTNL semaphore
661 * or @dev_base_lock. If the name is found a pointer to the device
662 * is returned. If the name is not found then %NULL is returned. The
663 * reference counters are not incremented so the caller must be
664 * careful with locks.
667 struct net_device *__dev_get_by_name(struct net *net, const char *name)
669 struct net_device *dev;
670 struct hlist_head *head = dev_name_hash(net, name);
672 hlist_for_each_entry(dev, head, name_hlist)
673 if (!strncmp(dev->name, name, IFNAMSIZ))
678 EXPORT_SYMBOL(__dev_get_by_name);
681 * dev_get_by_name_rcu - find a device by its name
682 * @net: the applicable net namespace
683 * @name: name to find
685 * Find an interface by name.
686 * If the name is found a pointer to the device is returned.
687 * If the name is not found then %NULL is returned.
688 * The reference counters are not incremented so the caller must be
689 * careful with locks. The caller must hold RCU lock.
692 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
694 struct net_device *dev;
695 struct hlist_head *head = dev_name_hash(net, name);
697 hlist_for_each_entry_rcu(dev, head, name_hlist)
698 if (!strncmp(dev->name, name, IFNAMSIZ))
703 EXPORT_SYMBOL(dev_get_by_name_rcu);
706 * dev_get_by_name - find a device by its name
707 * @net: the applicable net namespace
708 * @name: name to find
710 * Find an interface by name. This can be called from any
711 * context and does its own locking. The returned handle has
712 * the usage count incremented and the caller must use dev_put() to
713 * release it when it is no longer needed. %NULL is returned if no
714 * matching device is found.
717 struct net_device *dev_get_by_name(struct net *net, const char *name)
719 struct net_device *dev;
722 dev = dev_get_by_name_rcu(net, name);
728 EXPORT_SYMBOL(dev_get_by_name);
731 * __dev_get_by_index - find a device by its ifindex
732 * @net: the applicable net namespace
733 * @ifindex: index of device
735 * Search for an interface by index. Returns %NULL if the device
736 * is not found or a pointer to the device. The device has not
737 * had its reference counter increased so the caller must be careful
738 * about locking. The caller must hold either the RTNL semaphore
742 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
744 struct net_device *dev;
745 struct hlist_head *head = dev_index_hash(net, ifindex);
747 hlist_for_each_entry(dev, head, index_hlist)
748 if (dev->ifindex == ifindex)
753 EXPORT_SYMBOL(__dev_get_by_index);
756 * dev_get_by_index_rcu - find a device by its ifindex
757 * @net: the applicable net namespace
758 * @ifindex: index of device
760 * Search for an interface by index. Returns %NULL if the device
761 * is not found or a pointer to the device. The device has not
762 * had its reference counter increased so the caller must be careful
763 * about locking. The caller must hold RCU lock.
766 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
768 struct net_device *dev;
769 struct hlist_head *head = dev_index_hash(net, ifindex);
771 hlist_for_each_entry_rcu(dev, head, index_hlist)
772 if (dev->ifindex == ifindex)
777 EXPORT_SYMBOL(dev_get_by_index_rcu);
781 * dev_get_by_index - find a device by its ifindex
782 * @net: the applicable net namespace
783 * @ifindex: index of device
785 * Search for an interface by index. Returns NULL if the device
786 * is not found or a pointer to the device. The device returned has
787 * had a reference added and the pointer is safe until the user calls
788 * dev_put to indicate they have finished with it.
791 struct net_device *dev_get_by_index(struct net *net, int ifindex)
793 struct net_device *dev;
796 dev = dev_get_by_index_rcu(net, ifindex);
802 EXPORT_SYMBOL(dev_get_by_index);
805 * netdev_get_name - get a netdevice name, knowing its ifindex.
806 * @net: network namespace
807 * @name: a pointer to the buffer where the name will be stored.
808 * @ifindex: the ifindex of the interface to get the name from.
810 * The use of raw_seqcount_begin() and cond_resched() before
811 * retrying is required as we want to give the writers a chance
812 * to complete when CONFIG_PREEMPT is not set.
814 int netdev_get_name(struct net *net, char *name, int ifindex)
816 struct net_device *dev;
820 seq = raw_seqcount_begin(&devnet_rename_seq);
822 dev = dev_get_by_index_rcu(net, ifindex);
828 strcpy(name, dev->name);
830 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
839 * dev_getbyhwaddr_rcu - find a device by its hardware address
840 * @net: the applicable net namespace
841 * @type: media type of device
842 * @ha: hardware address
844 * Search for an interface by MAC address. Returns NULL if the device
845 * is not found or a pointer to the device.
846 * The caller must hold RCU or RTNL.
847 * The returned device has not had its ref count increased
848 * and the caller must therefore be careful about locking
852 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
855 struct net_device *dev;
857 for_each_netdev_rcu(net, dev)
858 if (dev->type == type &&
859 !memcmp(dev->dev_addr, ha, dev->addr_len))
864 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
866 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
868 struct net_device *dev;
871 for_each_netdev(net, dev)
872 if (dev->type == type)
877 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
879 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
881 struct net_device *dev, *ret = NULL;
884 for_each_netdev_rcu(net, dev)
885 if (dev->type == type) {
893 EXPORT_SYMBOL(dev_getfirstbyhwtype);
896 * dev_get_by_flags_rcu - find any device with given flags
897 * @net: the applicable net namespace
898 * @if_flags: IFF_* values
899 * @mask: bitmask of bits in if_flags to check
901 * Search for any interface with the given flags. Returns NULL if a device
902 * is not found or a pointer to the device. Must be called inside
903 * rcu_read_lock(), and result refcount is unchanged.
906 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
909 struct net_device *dev, *ret;
912 for_each_netdev_rcu(net, dev) {
913 if (((dev->flags ^ if_flags) & mask) == 0) {
920 EXPORT_SYMBOL(dev_get_by_flags_rcu);
923 * dev_valid_name - check if name is okay for network device
926 * Network device names need to be valid file names to
927 * to allow sysfs to work. We also disallow any kind of
930 bool dev_valid_name(const char *name)
934 if (strlen(name) >= IFNAMSIZ)
936 if (!strcmp(name, ".") || !strcmp(name, ".."))
940 if (*name == '/' || isspace(*name))
946 EXPORT_SYMBOL(dev_valid_name);
949 * __dev_alloc_name - allocate a name for a device
950 * @net: network namespace to allocate the device name in
951 * @name: name format string
952 * @buf: scratch buffer and result name string
954 * Passed a format string - eg "lt%d" it will try and find a suitable
955 * id. It scans list of devices to build up a free map, then chooses
956 * the first empty slot. The caller must hold the dev_base or rtnl lock
957 * while allocating the name and adding the device in order to avoid
959 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
960 * Returns the number of the unit assigned or a negative errno code.
963 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
967 const int max_netdevices = 8*PAGE_SIZE;
968 unsigned long *inuse;
969 struct net_device *d;
971 p = strnchr(name, IFNAMSIZ-1, '%');
974 * Verify the string as this thing may have come from
975 * the user. There must be either one "%d" and no other "%"
978 if (p[1] != 'd' || strchr(p + 2, '%'))
981 /* Use one page as a bit array of possible slots */
982 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
986 for_each_netdev(net, d) {
987 if (!sscanf(d->name, name, &i))
989 if (i < 0 || i >= max_netdevices)
992 /* avoid cases where sscanf is not exact inverse of printf */
993 snprintf(buf, IFNAMSIZ, name, i);
994 if (!strncmp(buf, d->name, IFNAMSIZ))
998 i = find_first_zero_bit(inuse, max_netdevices);
999 free_page((unsigned long) inuse);
1003 snprintf(buf, IFNAMSIZ, name, i);
1004 if (!__dev_get_by_name(net, buf))
1007 /* It is possible to run out of possible slots
1008 * when the name is long and there isn't enough space left
1009 * for the digits, or if all bits are used.
1015 * dev_alloc_name - allocate a name for a device
1017 * @name: name format string
1019 * Passed a format string - eg "lt%d" it will try and find a suitable
1020 * id. It scans list of devices to build up a free map, then chooses
1021 * the first empty slot. The caller must hold the dev_base or rtnl lock
1022 * while allocating the name and adding the device in order to avoid
1024 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1025 * Returns the number of the unit assigned or a negative errno code.
1028 int dev_alloc_name(struct net_device *dev, const char *name)
1034 BUG_ON(!dev_net(dev));
1036 ret = __dev_alloc_name(net, name, buf);
1038 strlcpy(dev->name, buf, IFNAMSIZ);
1041 EXPORT_SYMBOL(dev_alloc_name);
1043 static int dev_alloc_name_ns(struct net *net,
1044 struct net_device *dev,
1050 ret = __dev_alloc_name(net, name, buf);
1052 strlcpy(dev->name, buf, IFNAMSIZ);
1056 static int dev_get_valid_name(struct net *net,
1057 struct net_device *dev,
1062 if (!dev_valid_name(name))
1065 if (strchr(name, '%'))
1066 return dev_alloc_name_ns(net, dev, name);
1067 else if (__dev_get_by_name(net, name))
1069 else if (dev->name != name)
1070 strlcpy(dev->name, name, IFNAMSIZ);
1076 * dev_change_name - change name of a device
1078 * @newname: name (or format string) must be at least IFNAMSIZ
1080 * Change name of a device, can pass format strings "eth%d".
1083 int dev_change_name(struct net_device *dev, const char *newname)
1085 char oldname[IFNAMSIZ];
1091 BUG_ON(!dev_net(dev));
1094 if (dev->flags & IFF_UP)
1097 write_seqcount_begin(&devnet_rename_seq);
1099 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1100 write_seqcount_end(&devnet_rename_seq);
1104 memcpy(oldname, dev->name, IFNAMSIZ);
1106 err = dev_get_valid_name(net, dev, newname);
1108 write_seqcount_end(&devnet_rename_seq);
1113 ret = device_rename(&dev->dev, dev->name);
1115 memcpy(dev->name, oldname, IFNAMSIZ);
1116 write_seqcount_end(&devnet_rename_seq);
1120 write_seqcount_end(&devnet_rename_seq);
1122 netdev_adjacent_rename_links(dev, oldname);
1124 write_lock_bh(&dev_base_lock);
1125 hlist_del_rcu(&dev->name_hlist);
1126 write_unlock_bh(&dev_base_lock);
1130 write_lock_bh(&dev_base_lock);
1131 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1132 write_unlock_bh(&dev_base_lock);
1134 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1135 ret = notifier_to_errno(ret);
1138 /* err >= 0 after dev_alloc_name() or stores the first errno */
1141 write_seqcount_begin(&devnet_rename_seq);
1142 memcpy(dev->name, oldname, IFNAMSIZ);
1143 memcpy(oldname, newname, IFNAMSIZ);
1146 pr_err("%s: name change rollback failed: %d\n",
1155 * dev_set_alias - change ifalias of a device
1157 * @alias: name up to IFALIASZ
1158 * @len: limit of bytes to copy from info
1160 * Set ifalias for a device,
1162 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1168 if (len >= IFALIASZ)
1172 kfree(dev->ifalias);
1173 dev->ifalias = NULL;
1177 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1180 dev->ifalias = new_ifalias;
1182 strlcpy(dev->ifalias, alias, len+1);
1188 * netdev_features_change - device changes features
1189 * @dev: device to cause notification
1191 * Called to indicate a device has changed features.
1193 void netdev_features_change(struct net_device *dev)
1195 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1197 EXPORT_SYMBOL(netdev_features_change);
1200 * netdev_state_change - device changes state
1201 * @dev: device to cause notification
1203 * Called to indicate a device has changed state. This function calls
1204 * the notifier chains for netdev_chain and sends a NEWLINK message
1205 * to the routing socket.
1207 void netdev_state_change(struct net_device *dev)
1209 if (dev->flags & IFF_UP) {
1210 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1211 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1214 EXPORT_SYMBOL(netdev_state_change);
1217 * netdev_notify_peers - notify network peers about existence of @dev
1218 * @dev: network device
1220 * Generate traffic such that interested network peers are aware of
1221 * @dev, such as by generating a gratuitous ARP. This may be used when
1222 * a device wants to inform the rest of the network about some sort of
1223 * reconfiguration such as a failover event or virtual machine
1226 void netdev_notify_peers(struct net_device *dev)
1229 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1232 EXPORT_SYMBOL(netdev_notify_peers);
1234 static int __dev_open(struct net_device *dev)
1236 const struct net_device_ops *ops = dev->netdev_ops;
1241 if (!netif_device_present(dev))
1244 /* Block netpoll from trying to do any rx path servicing.
1245 * If we don't do this there is a chance ndo_poll_controller
1246 * or ndo_poll may be running while we open the device
1248 netpoll_poll_disable(dev);
1250 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1251 ret = notifier_to_errno(ret);
1255 set_bit(__LINK_STATE_START, &dev->state);
1257 if (ops->ndo_validate_addr)
1258 ret = ops->ndo_validate_addr(dev);
1260 if (!ret && ops->ndo_open)
1261 ret = ops->ndo_open(dev);
1263 netpoll_poll_enable(dev);
1266 clear_bit(__LINK_STATE_START, &dev->state);
1268 dev->flags |= IFF_UP;
1269 net_dmaengine_get();
1270 dev_set_rx_mode(dev);
1272 add_device_randomness(dev->dev_addr, dev->addr_len);
1279 * dev_open - prepare an interface for use.
1280 * @dev: device to open
1282 * Takes a device from down to up state. The device's private open
1283 * function is invoked and then the multicast lists are loaded. Finally
1284 * the device is moved into the up state and a %NETDEV_UP message is
1285 * sent to the netdev notifier chain.
1287 * Calling this function on an active interface is a nop. On a failure
1288 * a negative errno code is returned.
1290 int dev_open(struct net_device *dev)
1294 if (dev->flags & IFF_UP)
1297 ret = __dev_open(dev);
1301 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1302 call_netdevice_notifiers(NETDEV_UP, dev);
1306 EXPORT_SYMBOL(dev_open);
1308 static int __dev_close_many(struct list_head *head)
1310 struct net_device *dev;
1315 list_for_each_entry(dev, head, close_list) {
1316 /* Temporarily disable netpoll until the interface is down */
1317 netpoll_poll_disable(dev);
1319 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1321 clear_bit(__LINK_STATE_START, &dev->state);
1323 /* Synchronize to scheduled poll. We cannot touch poll list, it
1324 * can be even on different cpu. So just clear netif_running().
1326 * dev->stop() will invoke napi_disable() on all of it's
1327 * napi_struct instances on this device.
1329 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1332 dev_deactivate_many(head);
1334 list_for_each_entry(dev, head, close_list) {
1335 const struct net_device_ops *ops = dev->netdev_ops;
1338 * Call the device specific close. This cannot fail.
1339 * Only if device is UP
1341 * We allow it to be called even after a DETACH hot-plug
1347 dev->flags &= ~IFF_UP;
1348 net_dmaengine_put();
1349 netpoll_poll_enable(dev);
1355 static int __dev_close(struct net_device *dev)
1360 list_add(&dev->close_list, &single);
1361 retval = __dev_close_many(&single);
1367 static int dev_close_many(struct list_head *head)
1369 struct net_device *dev, *tmp;
1371 /* Remove the devices that don't need to be closed */
1372 list_for_each_entry_safe(dev, tmp, head, close_list)
1373 if (!(dev->flags & IFF_UP))
1374 list_del_init(&dev->close_list);
1376 __dev_close_many(head);
1378 list_for_each_entry_safe(dev, tmp, head, close_list) {
1379 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1380 call_netdevice_notifiers(NETDEV_DOWN, dev);
1381 list_del_init(&dev->close_list);
1388 * dev_close - shutdown an interface.
1389 * @dev: device to shutdown
1391 * This function moves an active device into down state. A
1392 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1393 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1396 int dev_close(struct net_device *dev)
1398 if (dev->flags & IFF_UP) {
1401 list_add(&dev->close_list, &single);
1402 dev_close_many(&single);
1407 EXPORT_SYMBOL(dev_close);
1411 * dev_disable_lro - disable Large Receive Offload on a device
1414 * Disable Large Receive Offload (LRO) on a net device. Must be
1415 * called under RTNL. This is needed if received packets may be
1416 * forwarded to another interface.
1418 void dev_disable_lro(struct net_device *dev)
1421 * If we're trying to disable lro on a vlan device
1422 * use the underlying physical device instead
1424 if (is_vlan_dev(dev))
1425 dev = vlan_dev_real_dev(dev);
1427 /* the same for macvlan devices */
1428 if (netif_is_macvlan(dev))
1429 dev = macvlan_dev_real_dev(dev);
1431 dev->wanted_features &= ~NETIF_F_LRO;
1432 netdev_update_features(dev);
1434 if (unlikely(dev->features & NETIF_F_LRO))
1435 netdev_WARN(dev, "failed to disable LRO!\n");
1437 EXPORT_SYMBOL(dev_disable_lro);
1439 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1440 struct net_device *dev)
1442 struct netdev_notifier_info info;
1444 netdev_notifier_info_init(&info, dev);
1445 return nb->notifier_call(nb, val, &info);
1448 static int dev_boot_phase = 1;
1451 * register_netdevice_notifier - register a network notifier block
1454 * Register a notifier to be called when network device events occur.
1455 * The notifier passed is linked into the kernel structures and must
1456 * not be reused until it has been unregistered. A negative errno code
1457 * is returned on a failure.
1459 * When registered all registration and up events are replayed
1460 * to the new notifier to allow device to have a race free
1461 * view of the network device list.
1464 int register_netdevice_notifier(struct notifier_block *nb)
1466 struct net_device *dev;
1467 struct net_device *last;
1472 err = raw_notifier_chain_register(&netdev_chain, nb);
1478 for_each_netdev(net, dev) {
1479 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1480 err = notifier_to_errno(err);
1484 if (!(dev->flags & IFF_UP))
1487 call_netdevice_notifier(nb, NETDEV_UP, dev);
1498 for_each_netdev(net, dev) {
1502 if (dev->flags & IFF_UP) {
1503 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1505 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1507 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1512 raw_notifier_chain_unregister(&netdev_chain, nb);
1515 EXPORT_SYMBOL(register_netdevice_notifier);
1518 * unregister_netdevice_notifier - unregister a network notifier block
1521 * Unregister a notifier previously registered by
1522 * register_netdevice_notifier(). The notifier is unlinked into the
1523 * kernel structures and may then be reused. A negative errno code
1524 * is returned on a failure.
1526 * After unregistering unregister and down device events are synthesized
1527 * for all devices on the device list to the removed notifier to remove
1528 * the need for special case cleanup code.
1531 int unregister_netdevice_notifier(struct notifier_block *nb)
1533 struct net_device *dev;
1538 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1543 for_each_netdev(net, dev) {
1544 if (dev->flags & IFF_UP) {
1545 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1547 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1549 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1556 EXPORT_SYMBOL(unregister_netdevice_notifier);
1559 * call_netdevice_notifiers_info - call all network notifier blocks
1560 * @val: value passed unmodified to notifier function
1561 * @dev: net_device pointer passed unmodified to notifier function
1562 * @info: notifier information data
1564 * Call all network notifier blocks. Parameters and return value
1565 * are as for raw_notifier_call_chain().
1568 static int call_netdevice_notifiers_info(unsigned long val,
1569 struct net_device *dev,
1570 struct netdev_notifier_info *info)
1573 netdev_notifier_info_init(info, dev);
1574 return raw_notifier_call_chain(&netdev_chain, val, info);
1578 * call_netdevice_notifiers - call all network notifier blocks
1579 * @val: value passed unmodified to notifier function
1580 * @dev: net_device pointer passed unmodified to notifier function
1582 * Call all network notifier blocks. Parameters and return value
1583 * are as for raw_notifier_call_chain().
1586 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1588 struct netdev_notifier_info info;
1590 return call_netdevice_notifiers_info(val, dev, &info);
1592 EXPORT_SYMBOL(call_netdevice_notifiers);
1594 static struct static_key netstamp_needed __read_mostly;
1595 #ifdef HAVE_JUMP_LABEL
1596 /* We are not allowed to call static_key_slow_dec() from irq context
1597 * If net_disable_timestamp() is called from irq context, defer the
1598 * static_key_slow_dec() calls.
1600 static atomic_t netstamp_needed_deferred;
1603 void net_enable_timestamp(void)
1605 #ifdef HAVE_JUMP_LABEL
1606 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1610 static_key_slow_dec(&netstamp_needed);
1614 static_key_slow_inc(&netstamp_needed);
1616 EXPORT_SYMBOL(net_enable_timestamp);
1618 void net_disable_timestamp(void)
1620 #ifdef HAVE_JUMP_LABEL
1621 if (in_interrupt()) {
1622 atomic_inc(&netstamp_needed_deferred);
1626 static_key_slow_dec(&netstamp_needed);
1628 EXPORT_SYMBOL(net_disable_timestamp);
1630 static inline void net_timestamp_set(struct sk_buff *skb)
1632 skb->tstamp.tv64 = 0;
1633 if (static_key_false(&netstamp_needed))
1634 __net_timestamp(skb);
1637 #define net_timestamp_check(COND, SKB) \
1638 if (static_key_false(&netstamp_needed)) { \
1639 if ((COND) && !(SKB)->tstamp.tv64) \
1640 __net_timestamp(SKB); \
1643 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1647 if (!(dev->flags & IFF_UP))
1650 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1651 if (skb->len <= len)
1654 /* if TSO is enabled, we don't care about the length as the packet
1655 * could be forwarded without being segmented before
1657 if (skb_is_gso(skb))
1662 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1664 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1666 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1667 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1668 atomic_long_inc(&dev->rx_dropped);
1674 if (unlikely(!is_skb_forwardable(dev, skb))) {
1675 atomic_long_inc(&dev->rx_dropped);
1680 skb_scrub_packet(skb, true);
1681 skb->protocol = eth_type_trans(skb, dev);
1685 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1688 * dev_forward_skb - loopback an skb to another netif
1690 * @dev: destination network device
1691 * @skb: buffer to forward
1694 * NET_RX_SUCCESS (no congestion)
1695 * NET_RX_DROP (packet was dropped, but freed)
1697 * dev_forward_skb can be used for injecting an skb from the
1698 * start_xmit function of one device into the receive queue
1699 * of another device.
1701 * The receiving device may be in another namespace, so
1702 * we have to clear all information in the skb that could
1703 * impact namespace isolation.
1705 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1707 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1709 EXPORT_SYMBOL_GPL(dev_forward_skb);
1711 static inline int deliver_skb(struct sk_buff *skb,
1712 struct packet_type *pt_prev,
1713 struct net_device *orig_dev)
1715 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1717 atomic_inc(&skb->users);
1718 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1721 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1723 if (!ptype->af_packet_priv || !skb->sk)
1726 if (ptype->id_match)
1727 return ptype->id_match(ptype, skb->sk);
1728 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1735 * Support routine. Sends outgoing frames to any network
1736 * taps currently in use.
1739 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1741 struct packet_type *ptype;
1742 struct sk_buff *skb2 = NULL;
1743 struct packet_type *pt_prev = NULL;
1746 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1747 /* Never send packets back to the socket
1748 * they originated from - MvS (miquels@drinkel.ow.org)
1750 if ((ptype->dev == dev || !ptype->dev) &&
1751 (!skb_loop_sk(ptype, skb))) {
1753 deliver_skb(skb2, pt_prev, skb->dev);
1758 skb2 = skb_clone(skb, GFP_ATOMIC);
1762 net_timestamp_set(skb2);
1764 /* skb->nh should be correctly
1765 set by sender, so that the second statement is
1766 just protection against buggy protocols.
1768 skb_reset_mac_header(skb2);
1770 if (skb_network_header(skb2) < skb2->data ||
1771 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1772 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1773 ntohs(skb2->protocol),
1775 skb_reset_network_header(skb2);
1778 skb2->transport_header = skb2->network_header;
1779 skb2->pkt_type = PACKET_OUTGOING;
1784 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1789 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1790 * @dev: Network device
1791 * @txq: number of queues available
1793 * If real_num_tx_queues is changed the tc mappings may no longer be
1794 * valid. To resolve this verify the tc mapping remains valid and if
1795 * not NULL the mapping. With no priorities mapping to this
1796 * offset/count pair it will no longer be used. In the worst case TC0
1797 * is invalid nothing can be done so disable priority mappings. If is
1798 * expected that drivers will fix this mapping if they can before
1799 * calling netif_set_real_num_tx_queues.
1801 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1804 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1806 /* If TC0 is invalidated disable TC mapping */
1807 if (tc->offset + tc->count > txq) {
1808 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1813 /* Invalidated prio to tc mappings set to TC0 */
1814 for (i = 1; i < TC_BITMASK + 1; i++) {
1815 int q = netdev_get_prio_tc_map(dev, i);
1817 tc = &dev->tc_to_txq[q];
1818 if (tc->offset + tc->count > txq) {
1819 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1821 netdev_set_prio_tc_map(dev, i, 0);
1827 static DEFINE_MUTEX(xps_map_mutex);
1828 #define xmap_dereference(P) \
1829 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1831 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1834 struct xps_map *map = NULL;
1838 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1840 for (pos = 0; map && pos < map->len; pos++) {
1841 if (map->queues[pos] == index) {
1843 map->queues[pos] = map->queues[--map->len];
1845 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1846 kfree_rcu(map, rcu);
1856 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1858 struct xps_dev_maps *dev_maps;
1860 bool active = false;
1862 mutex_lock(&xps_map_mutex);
1863 dev_maps = xmap_dereference(dev->xps_maps);
1868 for_each_possible_cpu(cpu) {
1869 for (i = index; i < dev->num_tx_queues; i++) {
1870 if (!remove_xps_queue(dev_maps, cpu, i))
1873 if (i == dev->num_tx_queues)
1878 RCU_INIT_POINTER(dev->xps_maps, NULL);
1879 kfree_rcu(dev_maps, rcu);
1882 for (i = index; i < dev->num_tx_queues; i++)
1883 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1887 mutex_unlock(&xps_map_mutex);
1890 static struct xps_map *expand_xps_map(struct xps_map *map,
1893 struct xps_map *new_map;
1894 int alloc_len = XPS_MIN_MAP_ALLOC;
1897 for (pos = 0; map && pos < map->len; pos++) {
1898 if (map->queues[pos] != index)
1903 /* Need to add queue to this CPU's existing map */
1905 if (pos < map->alloc_len)
1908 alloc_len = map->alloc_len * 2;
1911 /* Need to allocate new map to store queue on this CPU's map */
1912 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1917 for (i = 0; i < pos; i++)
1918 new_map->queues[i] = map->queues[i];
1919 new_map->alloc_len = alloc_len;
1925 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1928 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1929 struct xps_map *map, *new_map;
1930 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1931 int cpu, numa_node_id = -2;
1932 bool active = false;
1934 mutex_lock(&xps_map_mutex);
1936 dev_maps = xmap_dereference(dev->xps_maps);
1938 /* allocate memory for queue storage */
1939 for_each_online_cpu(cpu) {
1940 if (!cpumask_test_cpu(cpu, mask))
1944 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1945 if (!new_dev_maps) {
1946 mutex_unlock(&xps_map_mutex);
1950 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1953 map = expand_xps_map(map, cpu, index);
1957 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1961 goto out_no_new_maps;
1963 for_each_possible_cpu(cpu) {
1964 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1965 /* add queue to CPU maps */
1968 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1969 while ((pos < map->len) && (map->queues[pos] != index))
1972 if (pos == map->len)
1973 map->queues[map->len++] = index;
1975 if (numa_node_id == -2)
1976 numa_node_id = cpu_to_node(cpu);
1977 else if (numa_node_id != cpu_to_node(cpu))
1980 } else if (dev_maps) {
1981 /* fill in the new device map from the old device map */
1982 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1983 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1988 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1990 /* Cleanup old maps */
1992 for_each_possible_cpu(cpu) {
1993 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1994 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1995 if (map && map != new_map)
1996 kfree_rcu(map, rcu);
1999 kfree_rcu(dev_maps, rcu);
2002 dev_maps = new_dev_maps;
2006 /* update Tx queue numa node */
2007 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2008 (numa_node_id >= 0) ? numa_node_id :
2014 /* removes queue from unused CPUs */
2015 for_each_possible_cpu(cpu) {
2016 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2019 if (remove_xps_queue(dev_maps, cpu, index))
2023 /* free map if not active */
2025 RCU_INIT_POINTER(dev->xps_maps, NULL);
2026 kfree_rcu(dev_maps, rcu);
2030 mutex_unlock(&xps_map_mutex);
2034 /* remove any maps that we added */
2035 for_each_possible_cpu(cpu) {
2036 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2037 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2039 if (new_map && new_map != map)
2043 mutex_unlock(&xps_map_mutex);
2045 kfree(new_dev_maps);
2048 EXPORT_SYMBOL(netif_set_xps_queue);
2052 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2053 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2055 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2059 if (txq < 1 || txq > dev->num_tx_queues)
2062 if (dev->reg_state == NETREG_REGISTERED ||
2063 dev->reg_state == NETREG_UNREGISTERING) {
2066 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2072 netif_setup_tc(dev, txq);
2074 if (txq < dev->real_num_tx_queues) {
2075 qdisc_reset_all_tx_gt(dev, txq);
2077 netif_reset_xps_queues_gt(dev, txq);
2082 dev->real_num_tx_queues = txq;
2085 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2089 * netif_set_real_num_rx_queues - set actual number of RX queues used
2090 * @dev: Network device
2091 * @rxq: Actual number of RX queues
2093 * This must be called either with the rtnl_lock held or before
2094 * registration of the net device. Returns 0 on success, or a
2095 * negative error code. If called before registration, it always
2098 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2102 if (rxq < 1 || rxq > dev->num_rx_queues)
2105 if (dev->reg_state == NETREG_REGISTERED) {
2108 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2114 dev->real_num_rx_queues = rxq;
2117 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2121 * netif_get_num_default_rss_queues - default number of RSS queues
2123 * This routine should set an upper limit on the number of RSS queues
2124 * used by default by multiqueue devices.
2126 int netif_get_num_default_rss_queues(void)
2128 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2130 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2132 static inline void __netif_reschedule(struct Qdisc *q)
2134 struct softnet_data *sd;
2135 unsigned long flags;
2137 local_irq_save(flags);
2138 sd = &__get_cpu_var(softnet_data);
2139 q->next_sched = NULL;
2140 *sd->output_queue_tailp = q;
2141 sd->output_queue_tailp = &q->next_sched;
2142 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2143 local_irq_restore(flags);
2146 void __netif_schedule(struct Qdisc *q)
2148 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2149 __netif_reschedule(q);
2151 EXPORT_SYMBOL(__netif_schedule);
2153 struct dev_kfree_skb_cb {
2154 enum skb_free_reason reason;
2157 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2159 return (struct dev_kfree_skb_cb *)skb->cb;
2162 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2164 unsigned long flags;
2166 if (likely(atomic_read(&skb->users) == 1)) {
2168 atomic_set(&skb->users, 0);
2169 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2172 get_kfree_skb_cb(skb)->reason = reason;
2173 local_irq_save(flags);
2174 skb->next = __this_cpu_read(softnet_data.completion_queue);
2175 __this_cpu_write(softnet_data.completion_queue, skb);
2176 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2177 local_irq_restore(flags);
2179 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2181 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2183 if (in_irq() || irqs_disabled())
2184 __dev_kfree_skb_irq(skb, reason);
2188 EXPORT_SYMBOL(__dev_kfree_skb_any);
2192 * netif_device_detach - mark device as removed
2193 * @dev: network device
2195 * Mark device as removed from system and therefore no longer available.
2197 void netif_device_detach(struct net_device *dev)
2199 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2200 netif_running(dev)) {
2201 netif_tx_stop_all_queues(dev);
2204 EXPORT_SYMBOL(netif_device_detach);
2207 * netif_device_attach - mark device as attached
2208 * @dev: network device
2210 * Mark device as attached from system and restart if needed.
2212 void netif_device_attach(struct net_device *dev)
2214 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2215 netif_running(dev)) {
2216 netif_tx_wake_all_queues(dev);
2217 __netdev_watchdog_up(dev);
2220 EXPORT_SYMBOL(netif_device_attach);
2222 static void skb_warn_bad_offload(const struct sk_buff *skb)
2224 static const netdev_features_t null_features = 0;
2225 struct net_device *dev = skb->dev;
2226 const char *driver = "";
2228 if (!net_ratelimit())
2231 if (dev && dev->dev.parent)
2232 driver = dev_driver_string(dev->dev.parent);
2234 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2235 "gso_type=%d ip_summed=%d\n",
2236 driver, dev ? &dev->features : &null_features,
2237 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2238 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2239 skb_shinfo(skb)->gso_type, skb->ip_summed);
2243 * Invalidate hardware checksum when packet is to be mangled, and
2244 * complete checksum manually on outgoing path.
2246 int skb_checksum_help(struct sk_buff *skb)
2249 int ret = 0, offset;
2251 if (skb->ip_summed == CHECKSUM_COMPLETE)
2252 goto out_set_summed;
2254 if (unlikely(skb_shinfo(skb)->gso_size)) {
2255 skb_warn_bad_offload(skb);
2259 /* Before computing a checksum, we should make sure no frag could
2260 * be modified by an external entity : checksum could be wrong.
2262 if (skb_has_shared_frag(skb)) {
2263 ret = __skb_linearize(skb);
2268 offset = skb_checksum_start_offset(skb);
2269 BUG_ON(offset >= skb_headlen(skb));
2270 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2272 offset += skb->csum_offset;
2273 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2275 if (skb_cloned(skb) &&
2276 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2277 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2282 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2284 skb->ip_summed = CHECKSUM_NONE;
2288 EXPORT_SYMBOL(skb_checksum_help);
2290 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2292 unsigned int vlan_depth = skb->mac_len;
2293 __be16 type = skb->protocol;
2295 /* Tunnel gso handlers can set protocol to ethernet. */
2296 if (type == htons(ETH_P_TEB)) {
2299 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2302 eth = (struct ethhdr *)skb_mac_header(skb);
2303 type = eth->h_proto;
2306 /* if skb->protocol is 802.1Q/AD then the header should already be
2307 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at
2308 * ETH_HLEN otherwise
2310 if (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2312 if (unlikely(WARN_ON(vlan_depth < VLAN_HLEN)))
2314 vlan_depth -= VLAN_HLEN;
2316 vlan_depth = ETH_HLEN;
2319 struct vlan_hdr *vh;
2321 if (unlikely(!pskb_may_pull(skb,
2322 vlan_depth + VLAN_HLEN)))
2325 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2326 type = vh->h_vlan_encapsulated_proto;
2327 vlan_depth += VLAN_HLEN;
2328 } while (type == htons(ETH_P_8021Q) ||
2329 type == htons(ETH_P_8021AD));
2332 *depth = vlan_depth;
2338 * skb_mac_gso_segment - mac layer segmentation handler.
2339 * @skb: buffer to segment
2340 * @features: features for the output path (see dev->features)
2342 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2343 netdev_features_t features)
2345 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2346 struct packet_offload *ptype;
2347 int vlan_depth = skb->mac_len;
2348 __be16 type = skb_network_protocol(skb, &vlan_depth);
2350 if (unlikely(!type))
2351 return ERR_PTR(-EINVAL);
2353 __skb_pull(skb, vlan_depth);
2356 list_for_each_entry_rcu(ptype, &offload_base, list) {
2357 if (ptype->type == type && ptype->callbacks.gso_segment) {
2358 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2361 err = ptype->callbacks.gso_send_check(skb);
2362 segs = ERR_PTR(err);
2363 if (err || skb_gso_ok(skb, features))
2365 __skb_push(skb, (skb->data -
2366 skb_network_header(skb)));
2368 segs = ptype->callbacks.gso_segment(skb, features);
2374 __skb_push(skb, skb->data - skb_mac_header(skb));
2378 EXPORT_SYMBOL(skb_mac_gso_segment);
2381 /* openvswitch calls this on rx path, so we need a different check.
2383 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2386 return skb->ip_summed != CHECKSUM_PARTIAL;
2388 return skb->ip_summed == CHECKSUM_NONE;
2392 * __skb_gso_segment - Perform segmentation on skb.
2393 * @skb: buffer to segment
2394 * @features: features for the output path (see dev->features)
2395 * @tx_path: whether it is called in TX path
2397 * This function segments the given skb and returns a list of segments.
2399 * It may return NULL if the skb requires no segmentation. This is
2400 * only possible when GSO is used for verifying header integrity.
2402 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2403 netdev_features_t features, bool tx_path)
2405 if (unlikely(skb_needs_check(skb, tx_path))) {
2408 skb_warn_bad_offload(skb);
2410 if (skb_header_cloned(skb) &&
2411 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2412 return ERR_PTR(err);
2415 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2416 SKB_GSO_CB(skb)->encap_level = 0;
2418 skb_reset_mac_header(skb);
2419 skb_reset_mac_len(skb);
2421 return skb_mac_gso_segment(skb, features);
2423 EXPORT_SYMBOL(__skb_gso_segment);
2425 /* Take action when hardware reception checksum errors are detected. */
2427 void netdev_rx_csum_fault(struct net_device *dev)
2429 if (net_ratelimit()) {
2430 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2434 EXPORT_SYMBOL(netdev_rx_csum_fault);
2437 /* Actually, we should eliminate this check as soon as we know, that:
2438 * 1. IOMMU is present and allows to map all the memory.
2439 * 2. No high memory really exists on this machine.
2442 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2444 #ifdef CONFIG_HIGHMEM
2446 if (!(dev->features & NETIF_F_HIGHDMA)) {
2447 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2448 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2449 if (PageHighMem(skb_frag_page(frag)))
2454 if (PCI_DMA_BUS_IS_PHYS) {
2455 struct device *pdev = dev->dev.parent;
2459 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2460 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2461 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2462 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2471 void (*destructor)(struct sk_buff *skb);
2474 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2476 static void dev_gso_skb_destructor(struct sk_buff *skb)
2478 struct dev_gso_cb *cb;
2480 kfree_skb_list(skb->next);
2483 cb = DEV_GSO_CB(skb);
2485 cb->destructor(skb);
2489 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2490 * @skb: buffer to segment
2491 * @features: device features as applicable to this skb
2493 * This function segments the given skb and stores the list of segments
2496 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2498 struct sk_buff *segs;
2500 segs = skb_gso_segment(skb, features);
2502 /* Verifying header integrity only. */
2507 return PTR_ERR(segs);
2510 DEV_GSO_CB(skb)->destructor = skb->destructor;
2511 skb->destructor = dev_gso_skb_destructor;
2516 static netdev_features_t harmonize_features(struct sk_buff *skb,
2517 netdev_features_t features)
2521 if (skb->ip_summed != CHECKSUM_NONE &&
2522 !can_checksum_protocol(features, skb_network_protocol(skb, &tmp))) {
2523 features &= ~NETIF_F_ALL_CSUM;
2524 } else if (illegal_highdma(skb->dev, skb)) {
2525 features &= ~NETIF_F_SG;
2531 netdev_features_t netif_skb_features(struct sk_buff *skb)
2533 __be16 protocol = skb->protocol;
2534 netdev_features_t features = skb->dev->features;
2536 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2537 features &= ~NETIF_F_GSO_MASK;
2539 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2540 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2541 protocol = veh->h_vlan_encapsulated_proto;
2542 } else if (!vlan_tx_tag_present(skb)) {
2543 return harmonize_features(skb, features);
2546 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2547 NETIF_F_HW_VLAN_STAG_TX);
2549 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2550 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2551 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2552 NETIF_F_HW_VLAN_STAG_TX;
2554 return harmonize_features(skb, features);
2556 EXPORT_SYMBOL(netif_skb_features);
2558 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2559 struct netdev_queue *txq)
2561 const struct net_device_ops *ops = dev->netdev_ops;
2562 int rc = NETDEV_TX_OK;
2563 unsigned int skb_len;
2565 if (likely(!skb->next)) {
2566 netdev_features_t features;
2569 * If device doesn't need skb->dst, release it right now while
2570 * its hot in this cpu cache
2572 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2575 features = netif_skb_features(skb);
2577 if (vlan_tx_tag_present(skb) &&
2578 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2579 skb = __vlan_put_tag(skb, skb->vlan_proto,
2580 vlan_tx_tag_get(skb));
2587 /* If encapsulation offload request, verify we are testing
2588 * hardware encapsulation features instead of standard
2589 * features for the netdev
2591 if (skb->encapsulation)
2592 features &= dev->hw_enc_features;
2594 if (netif_needs_gso(skb, features)) {
2595 if (unlikely(dev_gso_segment(skb, features)))
2600 if (skb_needs_linearize(skb, features) &&
2601 __skb_linearize(skb))
2604 /* If packet is not checksummed and device does not
2605 * support checksumming for this protocol, complete
2606 * checksumming here.
2608 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2609 if (skb->encapsulation)
2610 skb_set_inner_transport_header(skb,
2611 skb_checksum_start_offset(skb));
2613 skb_set_transport_header(skb,
2614 skb_checksum_start_offset(skb));
2615 if (!(features & NETIF_F_ALL_CSUM) &&
2616 skb_checksum_help(skb))
2621 if (!list_empty(&ptype_all))
2622 dev_queue_xmit_nit(skb, dev);
2625 trace_net_dev_start_xmit(skb, dev);
2626 rc = ops->ndo_start_xmit(skb, dev);
2627 trace_net_dev_xmit(skb, rc, dev, skb_len);
2628 if (rc == NETDEV_TX_OK)
2629 txq_trans_update(txq);
2635 struct sk_buff *nskb = skb->next;
2637 skb->next = nskb->next;
2640 if (!list_empty(&ptype_all))
2641 dev_queue_xmit_nit(nskb, dev);
2643 skb_len = nskb->len;
2644 trace_net_dev_start_xmit(nskb, dev);
2645 rc = ops->ndo_start_xmit(nskb, dev);
2646 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2647 if (unlikely(rc != NETDEV_TX_OK)) {
2648 if (rc & ~NETDEV_TX_MASK)
2649 goto out_kfree_gso_skb;
2650 nskb->next = skb->next;
2654 txq_trans_update(txq);
2655 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2656 return NETDEV_TX_BUSY;
2657 } while (skb->next);
2660 if (likely(skb->next == NULL)) {
2661 skb->destructor = DEV_GSO_CB(skb)->destructor;
2670 EXPORT_SYMBOL_GPL(dev_hard_start_xmit);
2672 static void qdisc_pkt_len_init(struct sk_buff *skb)
2674 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2676 qdisc_skb_cb(skb)->pkt_len = skb->len;
2678 /* To get more precise estimation of bytes sent on wire,
2679 * we add to pkt_len the headers size of all segments
2681 if (shinfo->gso_size) {
2682 unsigned int hdr_len;
2683 u16 gso_segs = shinfo->gso_segs;
2685 /* mac layer + network layer */
2686 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2688 /* + transport layer */
2689 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2690 hdr_len += tcp_hdrlen(skb);
2692 hdr_len += sizeof(struct udphdr);
2694 if (shinfo->gso_type & SKB_GSO_DODGY)
2695 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2698 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2702 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2703 struct net_device *dev,
2704 struct netdev_queue *txq)
2706 spinlock_t *root_lock = qdisc_lock(q);
2710 qdisc_pkt_len_init(skb);
2711 qdisc_calculate_pkt_len(skb, q);
2713 * Heuristic to force contended enqueues to serialize on a
2714 * separate lock before trying to get qdisc main lock.
2715 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2716 * and dequeue packets faster.
2718 contended = qdisc_is_running(q);
2719 if (unlikely(contended))
2720 spin_lock(&q->busylock);
2722 spin_lock(root_lock);
2723 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2726 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2727 qdisc_run_begin(q)) {
2729 * This is a work-conserving queue; there are no old skbs
2730 * waiting to be sent out; and the qdisc is not running -
2731 * xmit the skb directly.
2733 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2736 qdisc_bstats_update(q, skb);
2738 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2739 if (unlikely(contended)) {
2740 spin_unlock(&q->busylock);
2747 rc = NET_XMIT_SUCCESS;
2750 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2751 if (qdisc_run_begin(q)) {
2752 if (unlikely(contended)) {
2753 spin_unlock(&q->busylock);
2759 spin_unlock(root_lock);
2760 if (unlikely(contended))
2761 spin_unlock(&q->busylock);
2765 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2766 static void skb_update_prio(struct sk_buff *skb)
2768 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2770 if (!skb->priority && skb->sk && map) {
2771 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2773 if (prioidx < map->priomap_len)
2774 skb->priority = map->priomap[prioidx];
2778 #define skb_update_prio(skb)
2781 static DEFINE_PER_CPU(int, xmit_recursion);
2782 #define RECURSION_LIMIT 10
2785 * dev_loopback_xmit - loop back @skb
2786 * @skb: buffer to transmit
2788 int dev_loopback_xmit(struct sk_buff *skb)
2790 skb_reset_mac_header(skb);
2791 __skb_pull(skb, skb_network_offset(skb));
2792 skb->pkt_type = PACKET_LOOPBACK;
2793 skb->ip_summed = CHECKSUM_UNNECESSARY;
2794 WARN_ON(!skb_dst(skb));
2799 EXPORT_SYMBOL(dev_loopback_xmit);
2802 * __dev_queue_xmit - transmit a buffer
2803 * @skb: buffer to transmit
2804 * @accel_priv: private data used for L2 forwarding offload
2806 * Queue a buffer for transmission to a network device. The caller must
2807 * have set the device and priority and built the buffer before calling
2808 * this function. The function can be called from an interrupt.
2810 * A negative errno code is returned on a failure. A success does not
2811 * guarantee the frame will be transmitted as it may be dropped due
2812 * to congestion or traffic shaping.
2814 * -----------------------------------------------------------------------------------
2815 * I notice this method can also return errors from the queue disciplines,
2816 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2819 * Regardless of the return value, the skb is consumed, so it is currently
2820 * difficult to retry a send to this method. (You can bump the ref count
2821 * before sending to hold a reference for retry if you are careful.)
2823 * When calling this method, interrupts MUST be enabled. This is because
2824 * the BH enable code must have IRQs enabled so that it will not deadlock.
2827 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2829 struct net_device *dev = skb->dev;
2830 struct netdev_queue *txq;
2834 skb_reset_mac_header(skb);
2836 /* Disable soft irqs for various locks below. Also
2837 * stops preemption for RCU.
2841 skb_update_prio(skb);
2843 txq = netdev_pick_tx(dev, skb, accel_priv);
2844 q = rcu_dereference_bh(txq->qdisc);
2846 #ifdef CONFIG_NET_CLS_ACT
2847 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2849 trace_net_dev_queue(skb);
2851 rc = __dev_xmit_skb(skb, q, dev, txq);
2855 /* The device has no queue. Common case for software devices:
2856 loopback, all the sorts of tunnels...
2858 Really, it is unlikely that netif_tx_lock protection is necessary
2859 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2861 However, it is possible, that they rely on protection
2864 Check this and shot the lock. It is not prone from deadlocks.
2865 Either shot noqueue qdisc, it is even simpler 8)
2867 if (dev->flags & IFF_UP) {
2868 int cpu = smp_processor_id(); /* ok because BHs are off */
2870 if (txq->xmit_lock_owner != cpu) {
2872 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2873 goto recursion_alert;
2875 HARD_TX_LOCK(dev, txq, cpu);
2877 if (!netif_xmit_stopped(txq)) {
2878 __this_cpu_inc(xmit_recursion);
2879 rc = dev_hard_start_xmit(skb, dev, txq);
2880 __this_cpu_dec(xmit_recursion);
2881 if (dev_xmit_complete(rc)) {
2882 HARD_TX_UNLOCK(dev, txq);
2886 HARD_TX_UNLOCK(dev, txq);
2887 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2890 /* Recursion is detected! It is possible,
2894 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2900 rcu_read_unlock_bh();
2902 atomic_long_inc(&dev->tx_dropped);
2906 rcu_read_unlock_bh();
2910 int dev_queue_xmit(struct sk_buff *skb)
2912 return __dev_queue_xmit(skb, NULL);
2914 EXPORT_SYMBOL(dev_queue_xmit);
2916 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
2918 return __dev_queue_xmit(skb, accel_priv);
2920 EXPORT_SYMBOL(dev_queue_xmit_accel);
2923 /*=======================================================================
2925 =======================================================================*/
2927 int netdev_max_backlog __read_mostly = 1000;
2928 EXPORT_SYMBOL(netdev_max_backlog);
2930 int netdev_tstamp_prequeue __read_mostly = 1;
2931 int netdev_budget __read_mostly = 300;
2932 int weight_p __read_mostly = 64; /* old backlog weight */
2934 /* Called with irq disabled */
2935 static inline void ____napi_schedule(struct softnet_data *sd,
2936 struct napi_struct *napi)
2938 list_add_tail(&napi->poll_list, &sd->poll_list);
2939 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2944 /* One global table that all flow-based protocols share. */
2945 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2946 EXPORT_SYMBOL(rps_sock_flow_table);
2948 struct static_key rps_needed __read_mostly;
2950 static struct rps_dev_flow *
2951 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2952 struct rps_dev_flow *rflow, u16 next_cpu)
2954 if (next_cpu != RPS_NO_CPU) {
2955 #ifdef CONFIG_RFS_ACCEL
2956 struct netdev_rx_queue *rxqueue;
2957 struct rps_dev_flow_table *flow_table;
2958 struct rps_dev_flow *old_rflow;
2963 /* Should we steer this flow to a different hardware queue? */
2964 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2965 !(dev->features & NETIF_F_NTUPLE))
2967 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2968 if (rxq_index == skb_get_rx_queue(skb))
2971 rxqueue = dev->_rx + rxq_index;
2972 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2975 flow_id = skb_get_hash(skb) & flow_table->mask;
2976 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2977 rxq_index, flow_id);
2981 rflow = &flow_table->flows[flow_id];
2983 if (old_rflow->filter == rflow->filter)
2984 old_rflow->filter = RPS_NO_FILTER;
2988 per_cpu(softnet_data, next_cpu).input_queue_head;
2991 rflow->cpu = next_cpu;
2996 * get_rps_cpu is called from netif_receive_skb and returns the target
2997 * CPU from the RPS map of the receiving queue for a given skb.
2998 * rcu_read_lock must be held on entry.
3000 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3001 struct rps_dev_flow **rflowp)
3003 struct netdev_rx_queue *rxqueue;
3004 struct rps_map *map;
3005 struct rps_dev_flow_table *flow_table;
3006 struct rps_sock_flow_table *sock_flow_table;
3011 if (skb_rx_queue_recorded(skb)) {
3012 u16 index = skb_get_rx_queue(skb);
3013 if (unlikely(index >= dev->real_num_rx_queues)) {
3014 WARN_ONCE(dev->real_num_rx_queues > 1,
3015 "%s received packet on queue %u, but number "
3016 "of RX queues is %u\n",
3017 dev->name, index, dev->real_num_rx_queues);
3020 rxqueue = dev->_rx + index;
3024 map = rcu_dereference(rxqueue->rps_map);
3026 if (map->len == 1 &&
3027 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3028 tcpu = map->cpus[0];
3029 if (cpu_online(tcpu))
3033 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3037 skb_reset_network_header(skb);
3038 hash = skb_get_hash(skb);
3042 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3043 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3044 if (flow_table && sock_flow_table) {
3046 struct rps_dev_flow *rflow;
3048 rflow = &flow_table->flows[hash & flow_table->mask];
3051 next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask];
3054 * If the desired CPU (where last recvmsg was done) is
3055 * different from current CPU (one in the rx-queue flow
3056 * table entry), switch if one of the following holds:
3057 * - Current CPU is unset (equal to RPS_NO_CPU).
3058 * - Current CPU is offline.
3059 * - The current CPU's queue tail has advanced beyond the
3060 * last packet that was enqueued using this table entry.
3061 * This guarantees that all previous packets for the flow
3062 * have been dequeued, thus preserving in order delivery.
3064 if (unlikely(tcpu != next_cpu) &&
3065 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3066 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3067 rflow->last_qtail)) >= 0)) {
3069 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3072 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3080 tcpu = map->cpus[((u64) hash * map->len) >> 32];
3082 if (cpu_online(tcpu)) {
3092 #ifdef CONFIG_RFS_ACCEL
3095 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3096 * @dev: Device on which the filter was set
3097 * @rxq_index: RX queue index
3098 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3099 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3101 * Drivers that implement ndo_rx_flow_steer() should periodically call
3102 * this function for each installed filter and remove the filters for
3103 * which it returns %true.
3105 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3106 u32 flow_id, u16 filter_id)
3108 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3109 struct rps_dev_flow_table *flow_table;
3110 struct rps_dev_flow *rflow;
3115 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3116 if (flow_table && flow_id <= flow_table->mask) {
3117 rflow = &flow_table->flows[flow_id];
3118 cpu = ACCESS_ONCE(rflow->cpu);
3119 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3120 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3121 rflow->last_qtail) <
3122 (int)(10 * flow_table->mask)))
3128 EXPORT_SYMBOL(rps_may_expire_flow);
3130 #endif /* CONFIG_RFS_ACCEL */
3132 /* Called from hardirq (IPI) context */
3133 static void rps_trigger_softirq(void *data)
3135 struct softnet_data *sd = data;
3137 ____napi_schedule(sd, &sd->backlog);
3141 #endif /* CONFIG_RPS */
3144 * Check if this softnet_data structure is another cpu one
3145 * If yes, queue it to our IPI list and return 1
3148 static int rps_ipi_queued(struct softnet_data *sd)
3151 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3154 sd->rps_ipi_next = mysd->rps_ipi_list;
3155 mysd->rps_ipi_list = sd;
3157 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3160 #endif /* CONFIG_RPS */
3164 #ifdef CONFIG_NET_FLOW_LIMIT
3165 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3168 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3170 #ifdef CONFIG_NET_FLOW_LIMIT
3171 struct sd_flow_limit *fl;
3172 struct softnet_data *sd;
3173 unsigned int old_flow, new_flow;
3175 if (qlen < (netdev_max_backlog >> 1))
3178 sd = &__get_cpu_var(softnet_data);
3181 fl = rcu_dereference(sd->flow_limit);
3183 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3184 old_flow = fl->history[fl->history_head];
3185 fl->history[fl->history_head] = new_flow;
3188 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3190 if (likely(fl->buckets[old_flow]))
3191 fl->buckets[old_flow]--;
3193 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3205 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3206 * queue (may be a remote CPU queue).
3208 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3209 unsigned int *qtail)
3211 struct softnet_data *sd;
3212 unsigned long flags;
3215 sd = &per_cpu(softnet_data, cpu);
3217 local_irq_save(flags);
3220 qlen = skb_queue_len(&sd->input_pkt_queue);
3221 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3222 if (skb_queue_len(&sd->input_pkt_queue)) {
3224 __skb_queue_tail(&sd->input_pkt_queue, skb);
3225 input_queue_tail_incr_save(sd, qtail);
3227 local_irq_restore(flags);
3228 return NET_RX_SUCCESS;
3231 /* Schedule NAPI for backlog device
3232 * We can use non atomic operation since we own the queue lock
3234 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3235 if (!rps_ipi_queued(sd))
3236 ____napi_schedule(sd, &sd->backlog);
3244 local_irq_restore(flags);
3246 atomic_long_inc(&skb->dev->rx_dropped);
3251 static int netif_rx_internal(struct sk_buff *skb)
3255 net_timestamp_check(netdev_tstamp_prequeue, skb);
3257 trace_netif_rx(skb);
3259 if (static_key_false(&rps_needed)) {
3260 struct rps_dev_flow voidflow, *rflow = &voidflow;
3266 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3268 cpu = smp_processor_id();
3270 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3278 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3285 * netif_rx - post buffer to the network code
3286 * @skb: buffer to post
3288 * This function receives a packet from a device driver and queues it for
3289 * the upper (protocol) levels to process. It always succeeds. The buffer
3290 * may be dropped during processing for congestion control or by the
3294 * NET_RX_SUCCESS (no congestion)
3295 * NET_RX_DROP (packet was dropped)
3299 int netif_rx(struct sk_buff *skb)
3301 trace_netif_rx_entry(skb);
3303 return netif_rx_internal(skb);
3305 EXPORT_SYMBOL(netif_rx);
3307 int netif_rx_ni(struct sk_buff *skb)
3311 trace_netif_rx_ni_entry(skb);
3314 err = netif_rx_internal(skb);
3315 if (local_softirq_pending())
3321 EXPORT_SYMBOL(netif_rx_ni);
3323 static void net_tx_action(struct softirq_action *h)
3325 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3327 if (sd->completion_queue) {
3328 struct sk_buff *clist;
3330 local_irq_disable();
3331 clist = sd->completion_queue;
3332 sd->completion_queue = NULL;
3336 struct sk_buff *skb = clist;
3337 clist = clist->next;
3339 WARN_ON(atomic_read(&skb->users));
3340 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3341 trace_consume_skb(skb);
3343 trace_kfree_skb(skb, net_tx_action);
3348 if (sd->output_queue) {
3351 local_irq_disable();
3352 head = sd->output_queue;
3353 sd->output_queue = NULL;
3354 sd->output_queue_tailp = &sd->output_queue;
3358 struct Qdisc *q = head;
3359 spinlock_t *root_lock;
3361 head = head->next_sched;
3363 root_lock = qdisc_lock(q);
3364 if (spin_trylock(root_lock)) {
3365 smp_mb__before_clear_bit();
3366 clear_bit(__QDISC_STATE_SCHED,
3369 spin_unlock(root_lock);
3371 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3373 __netif_reschedule(q);
3375 smp_mb__before_clear_bit();
3376 clear_bit(__QDISC_STATE_SCHED,
3384 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3385 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3386 /* This hook is defined here for ATM LANE */
3387 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3388 unsigned char *addr) __read_mostly;
3389 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3392 #ifdef CONFIG_NET_CLS_ACT
3393 /* TODO: Maybe we should just force sch_ingress to be compiled in
3394 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3395 * a compare and 2 stores extra right now if we dont have it on
3396 * but have CONFIG_NET_CLS_ACT
3397 * NOTE: This doesn't stop any functionality; if you dont have
3398 * the ingress scheduler, you just can't add policies on ingress.
3401 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3403 struct net_device *dev = skb->dev;
3404 u32 ttl = G_TC_RTTL(skb->tc_verd);
3405 int result = TC_ACT_OK;
3408 if (unlikely(MAX_RED_LOOP < ttl++)) {
3409 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3410 skb->skb_iif, dev->ifindex);
3414 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3415 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3418 if (q != &noop_qdisc) {
3419 spin_lock(qdisc_lock(q));
3420 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3421 result = qdisc_enqueue_root(skb, q);
3422 spin_unlock(qdisc_lock(q));
3428 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3429 struct packet_type **pt_prev,
3430 int *ret, struct net_device *orig_dev)
3432 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3434 if (!rxq || rxq->qdisc == &noop_qdisc)
3438 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3442 switch (ing_filter(skb, rxq)) {
3456 * netdev_rx_handler_register - register receive handler
3457 * @dev: device to register a handler for
3458 * @rx_handler: receive handler to register
3459 * @rx_handler_data: data pointer that is used by rx handler
3461 * Register a receive handler for a device. This handler will then be
3462 * called from __netif_receive_skb. A negative errno code is returned
3465 * The caller must hold the rtnl_mutex.
3467 * For a general description of rx_handler, see enum rx_handler_result.
3469 int netdev_rx_handler_register(struct net_device *dev,
3470 rx_handler_func_t *rx_handler,
3471 void *rx_handler_data)
3475 if (dev->rx_handler)
3478 /* Note: rx_handler_data must be set before rx_handler */
3479 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3480 rcu_assign_pointer(dev->rx_handler, rx_handler);
3484 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3487 * netdev_rx_handler_unregister - unregister receive handler
3488 * @dev: device to unregister a handler from
3490 * Unregister a receive handler from a device.
3492 * The caller must hold the rtnl_mutex.
3494 void netdev_rx_handler_unregister(struct net_device *dev)
3498 RCU_INIT_POINTER(dev->rx_handler, NULL);
3499 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3500 * section has a guarantee to see a non NULL rx_handler_data
3504 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3506 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3509 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3510 * the special handling of PFMEMALLOC skbs.
3512 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3514 switch (skb->protocol) {
3515 case htons(ETH_P_ARP):
3516 case htons(ETH_P_IP):
3517 case htons(ETH_P_IPV6):
3518 case htons(ETH_P_8021Q):
3519 case htons(ETH_P_8021AD):
3526 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3528 struct packet_type *ptype, *pt_prev;
3529 rx_handler_func_t *rx_handler;
3530 struct net_device *orig_dev;
3531 struct net_device *null_or_dev;
3532 bool deliver_exact = false;
3533 int ret = NET_RX_DROP;
3536 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3538 trace_netif_receive_skb(skb);
3540 orig_dev = skb->dev;
3542 skb_reset_network_header(skb);
3543 if (!skb_transport_header_was_set(skb))
3544 skb_reset_transport_header(skb);
3545 skb_reset_mac_len(skb);
3552 skb->skb_iif = skb->dev->ifindex;
3554 __this_cpu_inc(softnet_data.processed);
3556 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3557 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3558 skb = vlan_untag(skb);
3563 #ifdef CONFIG_NET_CLS_ACT
3564 if (skb->tc_verd & TC_NCLS) {
3565 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3573 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3574 if (!ptype->dev || ptype->dev == skb->dev) {
3576 ret = deliver_skb(skb, pt_prev, orig_dev);
3582 #ifdef CONFIG_NET_CLS_ACT
3583 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3589 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3592 if (vlan_tx_tag_present(skb)) {
3594 ret = deliver_skb(skb, pt_prev, orig_dev);
3597 if (vlan_do_receive(&skb))
3599 else if (unlikely(!skb))
3603 rx_handler = rcu_dereference(skb->dev->rx_handler);
3606 ret = deliver_skb(skb, pt_prev, orig_dev);
3609 switch (rx_handler(&skb)) {
3610 case RX_HANDLER_CONSUMED:
3611 ret = NET_RX_SUCCESS;
3613 case RX_HANDLER_ANOTHER:
3615 case RX_HANDLER_EXACT:
3616 deliver_exact = true;
3617 case RX_HANDLER_PASS:
3624 if (unlikely(vlan_tx_tag_present(skb))) {
3625 if (vlan_tx_tag_get_id(skb))
3626 skb->pkt_type = PACKET_OTHERHOST;
3627 /* Note: we might in the future use prio bits
3628 * and set skb->priority like in vlan_do_receive()
3629 * For the time being, just ignore Priority Code Point
3634 /* deliver only exact match when indicated */
3635 null_or_dev = deliver_exact ? skb->dev : NULL;
3637 type = skb->protocol;
3638 list_for_each_entry_rcu(ptype,
3639 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3640 if (ptype->type == type &&
3641 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3642 ptype->dev == orig_dev)) {
3644 ret = deliver_skb(skb, pt_prev, orig_dev);
3650 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3653 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3656 atomic_long_inc(&skb->dev->rx_dropped);
3658 /* Jamal, now you will not able to escape explaining
3659 * me how you were going to use this. :-)
3669 static int __netif_receive_skb(struct sk_buff *skb)
3673 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3674 unsigned long pflags = current->flags;
3677 * PFMEMALLOC skbs are special, they should
3678 * - be delivered to SOCK_MEMALLOC sockets only
3679 * - stay away from userspace
3680 * - have bounded memory usage
3682 * Use PF_MEMALLOC as this saves us from propagating the allocation
3683 * context down to all allocation sites.
3685 current->flags |= PF_MEMALLOC;
3686 ret = __netif_receive_skb_core(skb, true);
3687 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3689 ret = __netif_receive_skb_core(skb, false);
3694 static int netif_receive_skb_internal(struct sk_buff *skb)
3696 net_timestamp_check(netdev_tstamp_prequeue, skb);
3698 if (skb_defer_rx_timestamp(skb))
3699 return NET_RX_SUCCESS;
3702 if (static_key_false(&rps_needed)) {
3703 struct rps_dev_flow voidflow, *rflow = &voidflow;
3708 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3711 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3718 return __netif_receive_skb(skb);
3722 * netif_receive_skb - process receive buffer from network
3723 * @skb: buffer to process
3725 * netif_receive_skb() is the main receive data processing function.
3726 * It always succeeds. The buffer may be dropped during processing
3727 * for congestion control or by the protocol layers.
3729 * This function may only be called from softirq context and interrupts
3730 * should be enabled.
3732 * Return values (usually ignored):
3733 * NET_RX_SUCCESS: no congestion
3734 * NET_RX_DROP: packet was dropped
3736 int netif_receive_skb(struct sk_buff *skb)
3738 trace_netif_receive_skb_entry(skb);
3740 return netif_receive_skb_internal(skb);
3742 EXPORT_SYMBOL(netif_receive_skb);
3744 /* Network device is going away, flush any packets still pending
3745 * Called with irqs disabled.
3747 static void flush_backlog(void *arg)
3749 struct net_device *dev = arg;
3750 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3751 struct sk_buff *skb, *tmp;
3754 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3755 if (skb->dev == dev) {
3756 __skb_unlink(skb, &sd->input_pkt_queue);
3758 input_queue_head_incr(sd);
3763 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3764 if (skb->dev == dev) {
3765 __skb_unlink(skb, &sd->process_queue);
3767 input_queue_head_incr(sd);
3772 static int napi_gro_complete(struct sk_buff *skb)
3774 struct packet_offload *ptype;
3775 __be16 type = skb->protocol;
3776 struct list_head *head = &offload_base;
3779 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3781 if (NAPI_GRO_CB(skb)->count == 1) {
3782 skb_shinfo(skb)->gso_size = 0;
3787 list_for_each_entry_rcu(ptype, head, list) {
3788 if (ptype->type != type || !ptype->callbacks.gro_complete)
3791 err = ptype->callbacks.gro_complete(skb, 0);
3797 WARN_ON(&ptype->list == head);
3799 return NET_RX_SUCCESS;
3803 return netif_receive_skb_internal(skb);
3806 /* napi->gro_list contains packets ordered by age.
3807 * youngest packets at the head of it.
3808 * Complete skbs in reverse order to reduce latencies.
3810 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3812 struct sk_buff *skb, *prev = NULL;
3814 /* scan list and build reverse chain */
3815 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3820 for (skb = prev; skb; skb = prev) {
3823 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3827 napi_gro_complete(skb);
3831 napi->gro_list = NULL;
3833 EXPORT_SYMBOL(napi_gro_flush);
3835 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3838 unsigned int maclen = skb->dev->hard_header_len;
3839 u32 hash = skb_get_hash_raw(skb);
3841 for (p = napi->gro_list; p; p = p->next) {
3842 unsigned long diffs;
3844 NAPI_GRO_CB(p)->flush = 0;
3846 if (hash != skb_get_hash_raw(p)) {
3847 NAPI_GRO_CB(p)->same_flow = 0;
3851 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3852 diffs |= p->vlan_tci ^ skb->vlan_tci;
3853 if (maclen == ETH_HLEN)
3854 diffs |= compare_ether_header(skb_mac_header(p),
3855 skb_mac_header(skb));
3857 diffs = memcmp(skb_mac_header(p),
3858 skb_mac_header(skb),
3860 NAPI_GRO_CB(p)->same_flow = !diffs;
3864 static void skb_gro_reset_offset(struct sk_buff *skb)
3866 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3867 const skb_frag_t *frag0 = &pinfo->frags[0];
3869 NAPI_GRO_CB(skb)->data_offset = 0;
3870 NAPI_GRO_CB(skb)->frag0 = NULL;
3871 NAPI_GRO_CB(skb)->frag0_len = 0;
3873 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3875 !PageHighMem(skb_frag_page(frag0))) {
3876 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3877 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3881 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
3883 struct skb_shared_info *pinfo = skb_shinfo(skb);
3885 BUG_ON(skb->end - skb->tail < grow);
3887 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3889 skb->data_len -= grow;
3892 pinfo->frags[0].page_offset += grow;
3893 skb_frag_size_sub(&pinfo->frags[0], grow);
3895 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
3896 skb_frag_unref(skb, 0);
3897 memmove(pinfo->frags, pinfo->frags + 1,
3898 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
3902 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3904 struct sk_buff **pp = NULL;
3905 struct packet_offload *ptype;
3906 __be16 type = skb->protocol;
3907 struct list_head *head = &offload_base;
3909 enum gro_result ret;
3912 if (!(skb->dev->features & NETIF_F_GRO))
3915 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3918 gro_list_prepare(napi, skb);
3919 NAPI_GRO_CB(skb)->csum = skb->csum; /* Needed for CHECKSUM_COMPLETE */
3922 list_for_each_entry_rcu(ptype, head, list) {
3923 if (ptype->type != type || !ptype->callbacks.gro_receive)
3926 skb_set_network_header(skb, skb_gro_offset(skb));
3927 skb_reset_mac_len(skb);
3928 NAPI_GRO_CB(skb)->same_flow = 0;
3929 NAPI_GRO_CB(skb)->flush = 0;
3930 NAPI_GRO_CB(skb)->free = 0;
3931 NAPI_GRO_CB(skb)->udp_mark = 0;
3933 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3938 if (&ptype->list == head)
3941 same_flow = NAPI_GRO_CB(skb)->same_flow;
3942 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3945 struct sk_buff *nskb = *pp;
3949 napi_gro_complete(nskb);
3956 if (NAPI_GRO_CB(skb)->flush)
3959 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
3960 struct sk_buff *nskb = napi->gro_list;
3962 /* locate the end of the list to select the 'oldest' flow */
3963 while (nskb->next) {
3969 napi_gro_complete(nskb);
3973 NAPI_GRO_CB(skb)->count = 1;
3974 NAPI_GRO_CB(skb)->age = jiffies;
3975 NAPI_GRO_CB(skb)->last = skb;
3976 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3977 skb->next = napi->gro_list;
3978 napi->gro_list = skb;
3982 grow = skb_gro_offset(skb) - skb_headlen(skb);
3984 gro_pull_from_frag0(skb, grow);
3993 struct packet_offload *gro_find_receive_by_type(__be16 type)
3995 struct list_head *offload_head = &offload_base;
3996 struct packet_offload *ptype;
3998 list_for_each_entry_rcu(ptype, offload_head, list) {
3999 if (ptype->type != type || !ptype->callbacks.gro_receive)
4005 EXPORT_SYMBOL(gro_find_receive_by_type);
4007 struct packet_offload *gro_find_complete_by_type(__be16 type)
4009 struct list_head *offload_head = &offload_base;
4010 struct packet_offload *ptype;
4012 list_for_each_entry_rcu(ptype, offload_head, list) {
4013 if (ptype->type != type || !ptype->callbacks.gro_complete)
4019 EXPORT_SYMBOL(gro_find_complete_by_type);
4021 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4025 if (netif_receive_skb_internal(skb))
4033 case GRO_MERGED_FREE:
4034 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4035 kmem_cache_free(skbuff_head_cache, skb);
4048 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4050 trace_napi_gro_receive_entry(skb);
4052 skb_gro_reset_offset(skb);
4054 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4056 EXPORT_SYMBOL(napi_gro_receive);
4058 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4060 __skb_pull(skb, skb_headlen(skb));
4061 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4062 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4064 skb->dev = napi->dev;
4066 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4071 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4073 struct sk_buff *skb = napi->skb;
4076 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4081 EXPORT_SYMBOL(napi_get_frags);
4083 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4084 struct sk_buff *skb,
4090 __skb_push(skb, ETH_HLEN);
4091 skb->protocol = eth_type_trans(skb, skb->dev);
4092 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4097 case GRO_MERGED_FREE:
4098 napi_reuse_skb(napi, skb);
4108 /* Upper GRO stack assumes network header starts at gro_offset=0
4109 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4110 * We copy ethernet header into skb->data to have a common layout.
4112 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4114 struct sk_buff *skb = napi->skb;
4115 const struct ethhdr *eth;
4116 unsigned int hlen = sizeof(*eth);
4120 skb_reset_mac_header(skb);
4121 skb_gro_reset_offset(skb);
4123 eth = skb_gro_header_fast(skb, 0);
4124 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4125 eth = skb_gro_header_slow(skb, hlen, 0);
4126 if (unlikely(!eth)) {
4127 napi_reuse_skb(napi, skb);
4131 gro_pull_from_frag0(skb, hlen);
4132 NAPI_GRO_CB(skb)->frag0 += hlen;
4133 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4135 __skb_pull(skb, hlen);
4138 * This works because the only protocols we care about don't require
4140 * We'll fix it up properly in napi_frags_finish()
4142 skb->protocol = eth->h_proto;
4147 gro_result_t napi_gro_frags(struct napi_struct *napi)
4149 struct sk_buff *skb = napi_frags_skb(napi);
4154 trace_napi_gro_frags_entry(skb);
4156 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4158 EXPORT_SYMBOL(napi_gro_frags);
4161 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4162 * Note: called with local irq disabled, but exits with local irq enabled.
4164 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4167 struct softnet_data *remsd = sd->rps_ipi_list;
4170 sd->rps_ipi_list = NULL;
4174 /* Send pending IPI's to kick RPS processing on remote cpus. */
4176 struct softnet_data *next = remsd->rps_ipi_next;
4178 if (cpu_online(remsd->cpu))
4179 smp_call_function_single_async(remsd->cpu,
4188 static int process_backlog(struct napi_struct *napi, int quota)
4191 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4194 /* Check if we have pending ipi, its better to send them now,
4195 * not waiting net_rx_action() end.
4197 if (sd->rps_ipi_list) {
4198 local_irq_disable();
4199 net_rps_action_and_irq_enable(sd);
4202 napi->weight = weight_p;
4203 local_irq_disable();
4204 while (work < quota) {
4205 struct sk_buff *skb;
4208 while ((skb = __skb_dequeue(&sd->process_queue))) {
4210 __netif_receive_skb(skb);
4211 local_irq_disable();
4212 input_queue_head_incr(sd);
4213 if (++work >= quota) {
4220 qlen = skb_queue_len(&sd->input_pkt_queue);
4222 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4223 &sd->process_queue);
4225 if (qlen < quota - work) {
4227 * Inline a custom version of __napi_complete().
4228 * only current cpu owns and manipulates this napi,
4229 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4230 * we can use a plain write instead of clear_bit(),
4231 * and we dont need an smp_mb() memory barrier.
4233 list_del(&napi->poll_list);
4236 quota = work + qlen;
4246 * __napi_schedule - schedule for receive
4247 * @n: entry to schedule
4249 * The entry's receive function will be scheduled to run
4251 void __napi_schedule(struct napi_struct *n)
4253 unsigned long flags;
4255 local_irq_save(flags);
4256 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4257 local_irq_restore(flags);
4259 EXPORT_SYMBOL(__napi_schedule);
4261 void __napi_complete(struct napi_struct *n)
4263 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4264 BUG_ON(n->gro_list);
4266 list_del(&n->poll_list);
4267 smp_mb__before_clear_bit();
4268 clear_bit(NAPI_STATE_SCHED, &n->state);
4270 EXPORT_SYMBOL(__napi_complete);
4272 void napi_complete(struct napi_struct *n)
4274 unsigned long flags;
4277 * don't let napi dequeue from the cpu poll list
4278 * just in case its running on a different cpu
4280 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4283 napi_gro_flush(n, false);
4284 local_irq_save(flags);
4286 local_irq_restore(flags);
4288 EXPORT_SYMBOL(napi_complete);
4290 /* must be called under rcu_read_lock(), as we dont take a reference */
4291 struct napi_struct *napi_by_id(unsigned int napi_id)
4293 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4294 struct napi_struct *napi;
4296 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4297 if (napi->napi_id == napi_id)
4302 EXPORT_SYMBOL_GPL(napi_by_id);
4304 void napi_hash_add(struct napi_struct *napi)
4306 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4308 spin_lock(&napi_hash_lock);
4310 /* 0 is not a valid id, we also skip an id that is taken
4311 * we expect both events to be extremely rare
4314 while (!napi->napi_id) {
4315 napi->napi_id = ++napi_gen_id;
4316 if (napi_by_id(napi->napi_id))
4320 hlist_add_head_rcu(&napi->napi_hash_node,
4321 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4323 spin_unlock(&napi_hash_lock);
4326 EXPORT_SYMBOL_GPL(napi_hash_add);
4328 /* Warning : caller is responsible to make sure rcu grace period
4329 * is respected before freeing memory containing @napi
4331 void napi_hash_del(struct napi_struct *napi)
4333 spin_lock(&napi_hash_lock);
4335 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4336 hlist_del_rcu(&napi->napi_hash_node);
4338 spin_unlock(&napi_hash_lock);
4340 EXPORT_SYMBOL_GPL(napi_hash_del);
4342 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4343 int (*poll)(struct napi_struct *, int), int weight)
4345 INIT_LIST_HEAD(&napi->poll_list);
4346 napi->gro_count = 0;
4347 napi->gro_list = NULL;
4350 if (weight > NAPI_POLL_WEIGHT)
4351 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4353 napi->weight = weight;
4354 list_add(&napi->dev_list, &dev->napi_list);
4356 #ifdef CONFIG_NETPOLL
4357 spin_lock_init(&napi->poll_lock);
4358 napi->poll_owner = -1;
4360 set_bit(NAPI_STATE_SCHED, &napi->state);
4362 EXPORT_SYMBOL(netif_napi_add);
4364 void netif_napi_del(struct napi_struct *napi)
4366 list_del_init(&napi->dev_list);
4367 napi_free_frags(napi);
4369 kfree_skb_list(napi->gro_list);
4370 napi->gro_list = NULL;
4371 napi->gro_count = 0;
4373 EXPORT_SYMBOL(netif_napi_del);
4375 static void net_rx_action(struct softirq_action *h)
4377 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4378 unsigned long time_limit = jiffies + 2;
4379 int budget = netdev_budget;
4382 local_irq_disable();
4384 while (!list_empty(&sd->poll_list)) {
4385 struct napi_struct *n;
4388 /* If softirq window is exhuasted then punt.
4389 * Allow this to run for 2 jiffies since which will allow
4390 * an average latency of 1.5/HZ.
4392 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4397 /* Even though interrupts have been re-enabled, this
4398 * access is safe because interrupts can only add new
4399 * entries to the tail of this list, and only ->poll()
4400 * calls can remove this head entry from the list.
4402 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4404 have = netpoll_poll_lock(n);
4408 /* This NAPI_STATE_SCHED test is for avoiding a race
4409 * with netpoll's poll_napi(). Only the entity which
4410 * obtains the lock and sees NAPI_STATE_SCHED set will
4411 * actually make the ->poll() call. Therefore we avoid
4412 * accidentally calling ->poll() when NAPI is not scheduled.
4415 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4416 work = n->poll(n, weight);
4420 WARN_ON_ONCE(work > weight);
4424 local_irq_disable();
4426 /* Drivers must not modify the NAPI state if they
4427 * consume the entire weight. In such cases this code
4428 * still "owns" the NAPI instance and therefore can
4429 * move the instance around on the list at-will.
4431 if (unlikely(work == weight)) {
4432 if (unlikely(napi_disable_pending(n))) {
4435 local_irq_disable();
4438 /* flush too old packets
4439 * If HZ < 1000, flush all packets.
4442 napi_gro_flush(n, HZ >= 1000);
4443 local_irq_disable();
4445 list_move_tail(&n->poll_list, &sd->poll_list);
4449 netpoll_poll_unlock(have);
4452 net_rps_action_and_irq_enable(sd);
4454 #ifdef CONFIG_NET_DMA
4456 * There may not be any more sk_buffs coming right now, so push
4457 * any pending DMA copies to hardware
4459 dma_issue_pending_all();
4466 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4470 struct netdev_adjacent {
4471 struct net_device *dev;
4473 /* upper master flag, there can only be one master device per list */
4476 /* counter for the number of times this device was added to us */
4479 /* private field for the users */
4482 struct list_head list;
4483 struct rcu_head rcu;
4486 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4487 struct net_device *adj_dev,
4488 struct list_head *adj_list)
4490 struct netdev_adjacent *adj;
4492 list_for_each_entry(adj, adj_list, list) {
4493 if (adj->dev == adj_dev)
4500 * netdev_has_upper_dev - Check if device is linked to an upper device
4502 * @upper_dev: upper device to check
4504 * Find out if a device is linked to specified upper device and return true
4505 * in case it is. Note that this checks only immediate upper device,
4506 * not through a complete stack of devices. The caller must hold the RTNL lock.
4508 bool netdev_has_upper_dev(struct net_device *dev,
4509 struct net_device *upper_dev)
4513 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4515 EXPORT_SYMBOL(netdev_has_upper_dev);
4518 * netdev_has_any_upper_dev - Check if device is linked to some device
4521 * Find out if a device is linked to an upper device and return true in case
4522 * it is. The caller must hold the RTNL lock.
4524 static bool netdev_has_any_upper_dev(struct net_device *dev)
4528 return !list_empty(&dev->all_adj_list.upper);
4532 * netdev_master_upper_dev_get - Get master upper device
4535 * Find a master upper device and return pointer to it or NULL in case
4536 * it's not there. The caller must hold the RTNL lock.
4538 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4540 struct netdev_adjacent *upper;
4544 if (list_empty(&dev->adj_list.upper))
4547 upper = list_first_entry(&dev->adj_list.upper,
4548 struct netdev_adjacent, list);
4549 if (likely(upper->master))
4553 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4555 void *netdev_adjacent_get_private(struct list_head *adj_list)
4557 struct netdev_adjacent *adj;
4559 adj = list_entry(adj_list, struct netdev_adjacent, list);
4561 return adj->private;
4563 EXPORT_SYMBOL(netdev_adjacent_get_private);
4566 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4568 * @iter: list_head ** of the current position
4570 * Gets the next device from the dev's upper list, starting from iter
4571 * position. The caller must hold RCU read lock.
4573 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4574 struct list_head **iter)
4576 struct netdev_adjacent *upper;
4578 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4580 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4582 if (&upper->list == &dev->adj_list.upper)
4585 *iter = &upper->list;
4589 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4592 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4594 * @iter: list_head ** of the current position
4596 * Gets the next device from the dev's upper list, starting from iter
4597 * position. The caller must hold RCU read lock.
4599 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4600 struct list_head **iter)
4602 struct netdev_adjacent *upper;
4604 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4606 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4608 if (&upper->list == &dev->all_adj_list.upper)
4611 *iter = &upper->list;
4615 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4618 * netdev_lower_get_next_private - Get the next ->private from the
4619 * lower neighbour list
4621 * @iter: list_head ** of the current position
4623 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4624 * list, starting from iter position. The caller must hold either hold the
4625 * RTNL lock or its own locking that guarantees that the neighbour lower
4626 * list will remain unchainged.
4628 void *netdev_lower_get_next_private(struct net_device *dev,
4629 struct list_head **iter)
4631 struct netdev_adjacent *lower;
4633 lower = list_entry(*iter, struct netdev_adjacent, list);
4635 if (&lower->list == &dev->adj_list.lower)
4638 *iter = lower->list.next;
4640 return lower->private;
4642 EXPORT_SYMBOL(netdev_lower_get_next_private);
4645 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4646 * lower neighbour list, RCU
4649 * @iter: list_head ** of the current position
4651 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4652 * list, starting from iter position. The caller must hold RCU read lock.
4654 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4655 struct list_head **iter)
4657 struct netdev_adjacent *lower;
4659 WARN_ON_ONCE(!rcu_read_lock_held());
4661 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4663 if (&lower->list == &dev->adj_list.lower)
4666 *iter = &lower->list;
4668 return lower->private;
4670 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4673 * netdev_lower_get_next - Get the next device from the lower neighbour
4676 * @iter: list_head ** of the current position
4678 * Gets the next netdev_adjacent from the dev's lower neighbour
4679 * list, starting from iter position. The caller must hold RTNL lock or
4680 * its own locking that guarantees that the neighbour lower
4681 * list will remain unchainged.
4683 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4685 struct netdev_adjacent *lower;
4687 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4689 if (&lower->list == &dev->adj_list.lower)
4692 *iter = &lower->list;
4696 EXPORT_SYMBOL(netdev_lower_get_next);
4699 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4700 * lower neighbour list, RCU
4704 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4705 * list. The caller must hold RCU read lock.
4707 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4709 struct netdev_adjacent *lower;
4711 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4712 struct netdev_adjacent, list);
4714 return lower->private;
4717 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4720 * netdev_master_upper_dev_get_rcu - Get master upper device
4723 * Find a master upper device and return pointer to it or NULL in case
4724 * it's not there. The caller must hold the RCU read lock.
4726 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4728 struct netdev_adjacent *upper;
4730 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4731 struct netdev_adjacent, list);
4732 if (upper && likely(upper->master))
4736 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4738 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4739 struct net_device *adj_dev,
4740 struct list_head *dev_list)
4742 char linkname[IFNAMSIZ+7];
4743 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4744 "upper_%s" : "lower_%s", adj_dev->name);
4745 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4748 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4750 struct list_head *dev_list)
4752 char linkname[IFNAMSIZ+7];
4753 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4754 "upper_%s" : "lower_%s", name);
4755 sysfs_remove_link(&(dev->dev.kobj), linkname);
4758 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4759 (dev_list == &dev->adj_list.upper || \
4760 dev_list == &dev->adj_list.lower)
4762 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4763 struct net_device *adj_dev,
4764 struct list_head *dev_list,
4765 void *private, bool master)
4767 struct netdev_adjacent *adj;
4770 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4777 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4782 adj->master = master;
4784 adj->private = private;
4787 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4788 adj_dev->name, dev->name, adj_dev->name);
4790 if (netdev_adjacent_is_neigh_list(dev, dev_list)) {
4791 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
4796 /* Ensure that master link is always the first item in list. */
4798 ret = sysfs_create_link(&(dev->dev.kobj),
4799 &(adj_dev->dev.kobj), "master");
4801 goto remove_symlinks;
4803 list_add_rcu(&adj->list, dev_list);
4805 list_add_tail_rcu(&adj->list, dev_list);
4811 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4812 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4820 static void __netdev_adjacent_dev_remove(struct net_device *dev,
4821 struct net_device *adj_dev,
4822 struct list_head *dev_list)
4824 struct netdev_adjacent *adj;
4826 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4829 pr_err("tried to remove device %s from %s\n",
4830 dev->name, adj_dev->name);
4834 if (adj->ref_nr > 1) {
4835 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4842 sysfs_remove_link(&(dev->dev.kobj), "master");
4844 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4845 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4847 list_del_rcu(&adj->list);
4848 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4849 adj_dev->name, dev->name, adj_dev->name);
4851 kfree_rcu(adj, rcu);
4854 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4855 struct net_device *upper_dev,
4856 struct list_head *up_list,
4857 struct list_head *down_list,
4858 void *private, bool master)
4862 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4867 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4870 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4877 static int __netdev_adjacent_dev_link(struct net_device *dev,
4878 struct net_device *upper_dev)
4880 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4881 &dev->all_adj_list.upper,
4882 &upper_dev->all_adj_list.lower,
4886 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4887 struct net_device *upper_dev,
4888 struct list_head *up_list,
4889 struct list_head *down_list)
4891 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4892 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4895 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
4896 struct net_device *upper_dev)
4898 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4899 &dev->all_adj_list.upper,
4900 &upper_dev->all_adj_list.lower);
4903 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4904 struct net_device *upper_dev,
4905 void *private, bool master)
4907 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4912 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4913 &dev->adj_list.upper,
4914 &upper_dev->adj_list.lower,
4917 __netdev_adjacent_dev_unlink(dev, upper_dev);
4924 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4925 struct net_device *upper_dev)
4927 __netdev_adjacent_dev_unlink(dev, upper_dev);
4928 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4929 &dev->adj_list.upper,
4930 &upper_dev->adj_list.lower);
4933 static int __netdev_upper_dev_link(struct net_device *dev,
4934 struct net_device *upper_dev, bool master,
4937 struct netdev_adjacent *i, *j, *to_i, *to_j;
4942 if (dev == upper_dev)
4945 /* To prevent loops, check if dev is not upper device to upper_dev. */
4946 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
4949 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
4952 if (master && netdev_master_upper_dev_get(dev))
4955 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
4960 /* Now that we linked these devs, make all the upper_dev's
4961 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4962 * versa, and don't forget the devices itself. All of these
4963 * links are non-neighbours.
4965 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4966 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4967 pr_debug("Interlinking %s with %s, non-neighbour\n",
4968 i->dev->name, j->dev->name);
4969 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
4975 /* add dev to every upper_dev's upper device */
4976 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4977 pr_debug("linking %s's upper device %s with %s\n",
4978 upper_dev->name, i->dev->name, dev->name);
4979 ret = __netdev_adjacent_dev_link(dev, i->dev);
4981 goto rollback_upper_mesh;
4984 /* add upper_dev to every dev's lower device */
4985 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4986 pr_debug("linking %s's lower device %s with %s\n", dev->name,
4987 i->dev->name, upper_dev->name);
4988 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
4990 goto rollback_lower_mesh;
4993 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4996 rollback_lower_mesh:
4998 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5001 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5006 rollback_upper_mesh:
5008 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5011 __netdev_adjacent_dev_unlink(dev, i->dev);
5019 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5020 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5021 if (i == to_i && j == to_j)
5023 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5029 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5035 * netdev_upper_dev_link - Add a link to the upper device
5037 * @upper_dev: new upper device
5039 * Adds a link to device which is upper to this one. The caller must hold
5040 * the RTNL lock. On a failure a negative errno code is returned.
5041 * On success the reference counts are adjusted and the function
5044 int netdev_upper_dev_link(struct net_device *dev,
5045 struct net_device *upper_dev)
5047 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5049 EXPORT_SYMBOL(netdev_upper_dev_link);
5052 * netdev_master_upper_dev_link - Add a master link to the upper device
5054 * @upper_dev: new upper device
5056 * Adds a link to device which is upper to this one. In this case, only
5057 * one master upper device can be linked, although other non-master devices
5058 * might be linked as well. The caller must hold the RTNL lock.
5059 * On a failure a negative errno code is returned. On success the reference
5060 * counts are adjusted and the function returns zero.
5062 int netdev_master_upper_dev_link(struct net_device *dev,
5063 struct net_device *upper_dev)
5065 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5067 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5069 int netdev_master_upper_dev_link_private(struct net_device *dev,
5070 struct net_device *upper_dev,
5073 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5075 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5078 * netdev_upper_dev_unlink - Removes a link to upper device
5080 * @upper_dev: new upper device
5082 * Removes a link to device which is upper to this one. The caller must hold
5085 void netdev_upper_dev_unlink(struct net_device *dev,
5086 struct net_device *upper_dev)
5088 struct netdev_adjacent *i, *j;
5091 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5093 /* Here is the tricky part. We must remove all dev's lower
5094 * devices from all upper_dev's upper devices and vice
5095 * versa, to maintain the graph relationship.
5097 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5098 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5099 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5101 /* remove also the devices itself from lower/upper device
5104 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5105 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5107 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5108 __netdev_adjacent_dev_unlink(dev, i->dev);
5110 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5112 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5114 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5116 struct netdev_adjacent *iter;
5118 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5119 netdev_adjacent_sysfs_del(iter->dev, oldname,
5120 &iter->dev->adj_list.lower);
5121 netdev_adjacent_sysfs_add(iter->dev, dev,
5122 &iter->dev->adj_list.lower);
5125 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5126 netdev_adjacent_sysfs_del(iter->dev, oldname,
5127 &iter->dev->adj_list.upper);
5128 netdev_adjacent_sysfs_add(iter->dev, dev,
5129 &iter->dev->adj_list.upper);
5133 void *netdev_lower_dev_get_private(struct net_device *dev,
5134 struct net_device *lower_dev)
5136 struct netdev_adjacent *lower;
5140 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5144 return lower->private;
5146 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5149 int dev_get_nest_level(struct net_device *dev,
5150 bool (*type_check)(struct net_device *dev))
5152 struct net_device *lower = NULL;
5153 struct list_head *iter;
5159 netdev_for_each_lower_dev(dev, lower, iter) {
5160 nest = dev_get_nest_level(lower, type_check);
5161 if (max_nest < nest)
5165 if (type_check(dev))
5170 EXPORT_SYMBOL(dev_get_nest_level);
5172 static void dev_change_rx_flags(struct net_device *dev, int flags)
5174 const struct net_device_ops *ops = dev->netdev_ops;
5176 if (ops->ndo_change_rx_flags)
5177 ops->ndo_change_rx_flags(dev, flags);
5180 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5182 unsigned int old_flags = dev->flags;
5188 dev->flags |= IFF_PROMISC;
5189 dev->promiscuity += inc;
5190 if (dev->promiscuity == 0) {
5193 * If inc causes overflow, untouch promisc and return error.
5196 dev->flags &= ~IFF_PROMISC;
5198 dev->promiscuity -= inc;
5199 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5204 if (dev->flags != old_flags) {
5205 pr_info("device %s %s promiscuous mode\n",
5207 dev->flags & IFF_PROMISC ? "entered" : "left");
5208 if (audit_enabled) {
5209 current_uid_gid(&uid, &gid);
5210 audit_log(current->audit_context, GFP_ATOMIC,
5211 AUDIT_ANOM_PROMISCUOUS,
5212 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5213 dev->name, (dev->flags & IFF_PROMISC),
5214 (old_flags & IFF_PROMISC),
5215 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5216 from_kuid(&init_user_ns, uid),
5217 from_kgid(&init_user_ns, gid),
5218 audit_get_sessionid(current));
5221 dev_change_rx_flags(dev, IFF_PROMISC);
5224 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5229 * dev_set_promiscuity - update promiscuity count on a device
5233 * Add or remove promiscuity from a device. While the count in the device
5234 * remains above zero the interface remains promiscuous. Once it hits zero
5235 * the device reverts back to normal filtering operation. A negative inc
5236 * value is used to drop promiscuity on the device.
5237 * Return 0 if successful or a negative errno code on error.
5239 int dev_set_promiscuity(struct net_device *dev, int inc)
5241 unsigned int old_flags = dev->flags;
5244 err = __dev_set_promiscuity(dev, inc, true);
5247 if (dev->flags != old_flags)
5248 dev_set_rx_mode(dev);
5251 EXPORT_SYMBOL(dev_set_promiscuity);
5253 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5255 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5259 dev->flags |= IFF_ALLMULTI;
5260 dev->allmulti += inc;
5261 if (dev->allmulti == 0) {
5264 * If inc causes overflow, untouch allmulti and return error.
5267 dev->flags &= ~IFF_ALLMULTI;
5269 dev->allmulti -= inc;
5270 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5275 if (dev->flags ^ old_flags) {
5276 dev_change_rx_flags(dev, IFF_ALLMULTI);
5277 dev_set_rx_mode(dev);
5279 __dev_notify_flags(dev, old_flags,
5280 dev->gflags ^ old_gflags);
5286 * dev_set_allmulti - update allmulti count on a device
5290 * Add or remove reception of all multicast frames to a device. While the
5291 * count in the device remains above zero the interface remains listening
5292 * to all interfaces. Once it hits zero the device reverts back to normal
5293 * filtering operation. A negative @inc value is used to drop the counter
5294 * when releasing a resource needing all multicasts.
5295 * Return 0 if successful or a negative errno code on error.
5298 int dev_set_allmulti(struct net_device *dev, int inc)
5300 return __dev_set_allmulti(dev, inc, true);
5302 EXPORT_SYMBOL(dev_set_allmulti);
5305 * Upload unicast and multicast address lists to device and
5306 * configure RX filtering. When the device doesn't support unicast
5307 * filtering it is put in promiscuous mode while unicast addresses
5310 void __dev_set_rx_mode(struct net_device *dev)
5312 const struct net_device_ops *ops = dev->netdev_ops;
5314 /* dev_open will call this function so the list will stay sane. */
5315 if (!(dev->flags&IFF_UP))
5318 if (!netif_device_present(dev))
5321 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5322 /* Unicast addresses changes may only happen under the rtnl,
5323 * therefore calling __dev_set_promiscuity here is safe.
5325 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5326 __dev_set_promiscuity(dev, 1, false);
5327 dev->uc_promisc = true;
5328 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5329 __dev_set_promiscuity(dev, -1, false);
5330 dev->uc_promisc = false;
5334 if (ops->ndo_set_rx_mode)
5335 ops->ndo_set_rx_mode(dev);
5338 void dev_set_rx_mode(struct net_device *dev)
5340 netif_addr_lock_bh(dev);
5341 __dev_set_rx_mode(dev);
5342 netif_addr_unlock_bh(dev);
5346 * dev_get_flags - get flags reported to userspace
5349 * Get the combination of flag bits exported through APIs to userspace.
5351 unsigned int dev_get_flags(const struct net_device *dev)
5355 flags = (dev->flags & ~(IFF_PROMISC |
5360 (dev->gflags & (IFF_PROMISC |
5363 if (netif_running(dev)) {
5364 if (netif_oper_up(dev))
5365 flags |= IFF_RUNNING;
5366 if (netif_carrier_ok(dev))
5367 flags |= IFF_LOWER_UP;
5368 if (netif_dormant(dev))
5369 flags |= IFF_DORMANT;
5374 EXPORT_SYMBOL(dev_get_flags);
5376 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5378 unsigned int old_flags = dev->flags;
5384 * Set the flags on our device.
5387 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5388 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5390 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5394 * Load in the correct multicast list now the flags have changed.
5397 if ((old_flags ^ flags) & IFF_MULTICAST)
5398 dev_change_rx_flags(dev, IFF_MULTICAST);
5400 dev_set_rx_mode(dev);
5403 * Have we downed the interface. We handle IFF_UP ourselves
5404 * according to user attempts to set it, rather than blindly
5409 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
5410 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5413 dev_set_rx_mode(dev);
5416 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5417 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5418 unsigned int old_flags = dev->flags;
5420 dev->gflags ^= IFF_PROMISC;
5422 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5423 if (dev->flags != old_flags)
5424 dev_set_rx_mode(dev);
5427 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5428 is important. Some (broken) drivers set IFF_PROMISC, when
5429 IFF_ALLMULTI is requested not asking us and not reporting.
5431 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5432 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5434 dev->gflags ^= IFF_ALLMULTI;
5435 __dev_set_allmulti(dev, inc, false);
5441 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5442 unsigned int gchanges)
5444 unsigned int changes = dev->flags ^ old_flags;
5447 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5449 if (changes & IFF_UP) {
5450 if (dev->flags & IFF_UP)
5451 call_netdevice_notifiers(NETDEV_UP, dev);
5453 call_netdevice_notifiers(NETDEV_DOWN, dev);
5456 if (dev->flags & IFF_UP &&
5457 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5458 struct netdev_notifier_change_info change_info;
5460 change_info.flags_changed = changes;
5461 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5467 * dev_change_flags - change device settings
5469 * @flags: device state flags
5471 * Change settings on device based state flags. The flags are
5472 * in the userspace exported format.
5474 int dev_change_flags(struct net_device *dev, unsigned int flags)
5477 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5479 ret = __dev_change_flags(dev, flags);
5483 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5484 __dev_notify_flags(dev, old_flags, changes);
5487 EXPORT_SYMBOL(dev_change_flags);
5489 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5491 const struct net_device_ops *ops = dev->netdev_ops;
5493 if (ops->ndo_change_mtu)
5494 return ops->ndo_change_mtu(dev, new_mtu);
5501 * dev_set_mtu - Change maximum transfer unit
5503 * @new_mtu: new transfer unit
5505 * Change the maximum transfer size of the network device.
5507 int dev_set_mtu(struct net_device *dev, int new_mtu)
5511 if (new_mtu == dev->mtu)
5514 /* MTU must be positive. */
5518 if (!netif_device_present(dev))
5521 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5522 err = notifier_to_errno(err);
5526 orig_mtu = dev->mtu;
5527 err = __dev_set_mtu(dev, new_mtu);
5530 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5531 err = notifier_to_errno(err);
5533 /* setting mtu back and notifying everyone again,
5534 * so that they have a chance to revert changes.
5536 __dev_set_mtu(dev, orig_mtu);
5537 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5542 EXPORT_SYMBOL(dev_set_mtu);
5545 * dev_set_group - Change group this device belongs to
5547 * @new_group: group this device should belong to
5549 void dev_set_group(struct net_device *dev, int new_group)
5551 dev->group = new_group;
5553 EXPORT_SYMBOL(dev_set_group);
5556 * dev_set_mac_address - Change Media Access Control Address
5560 * Change the hardware (MAC) address of the device
5562 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5564 const struct net_device_ops *ops = dev->netdev_ops;
5567 if (!ops->ndo_set_mac_address)
5569 if (sa->sa_family != dev->type)
5571 if (!netif_device_present(dev))
5573 err = ops->ndo_set_mac_address(dev, sa);
5576 dev->addr_assign_type = NET_ADDR_SET;
5577 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5578 add_device_randomness(dev->dev_addr, dev->addr_len);
5581 EXPORT_SYMBOL(dev_set_mac_address);
5584 * dev_change_carrier - Change device carrier
5586 * @new_carrier: new value
5588 * Change device carrier
5590 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5592 const struct net_device_ops *ops = dev->netdev_ops;
5594 if (!ops->ndo_change_carrier)
5596 if (!netif_device_present(dev))
5598 return ops->ndo_change_carrier(dev, new_carrier);
5600 EXPORT_SYMBOL(dev_change_carrier);
5603 * dev_get_phys_port_id - Get device physical port ID
5607 * Get device physical port ID
5609 int dev_get_phys_port_id(struct net_device *dev,
5610 struct netdev_phys_port_id *ppid)
5612 const struct net_device_ops *ops = dev->netdev_ops;
5614 if (!ops->ndo_get_phys_port_id)
5616 return ops->ndo_get_phys_port_id(dev, ppid);
5618 EXPORT_SYMBOL(dev_get_phys_port_id);
5621 * dev_new_index - allocate an ifindex
5622 * @net: the applicable net namespace
5624 * Returns a suitable unique value for a new device interface
5625 * number. The caller must hold the rtnl semaphore or the
5626 * dev_base_lock to be sure it remains unique.
5628 static int dev_new_index(struct net *net)
5630 int ifindex = net->ifindex;
5634 if (!__dev_get_by_index(net, ifindex))
5635 return net->ifindex = ifindex;
5639 /* Delayed registration/unregisteration */
5640 static LIST_HEAD(net_todo_list);
5641 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5643 static void net_set_todo(struct net_device *dev)
5645 list_add_tail(&dev->todo_list, &net_todo_list);
5646 dev_net(dev)->dev_unreg_count++;
5649 static void rollback_registered_many(struct list_head *head)
5651 struct net_device *dev, *tmp;
5652 LIST_HEAD(close_head);
5654 BUG_ON(dev_boot_phase);
5657 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5658 /* Some devices call without registering
5659 * for initialization unwind. Remove those
5660 * devices and proceed with the remaining.
5662 if (dev->reg_state == NETREG_UNINITIALIZED) {
5663 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5667 list_del(&dev->unreg_list);
5670 dev->dismantle = true;
5671 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5674 /* If device is running, close it first. */
5675 list_for_each_entry(dev, head, unreg_list)
5676 list_add_tail(&dev->close_list, &close_head);
5677 dev_close_many(&close_head);
5679 list_for_each_entry(dev, head, unreg_list) {
5680 /* And unlink it from device chain. */
5681 unlist_netdevice(dev);
5683 dev->reg_state = NETREG_UNREGISTERING;
5688 list_for_each_entry(dev, head, unreg_list) {
5689 /* Shutdown queueing discipline. */
5693 /* Notify protocols, that we are about to destroy
5694 this device. They should clean all the things.
5696 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5699 * Flush the unicast and multicast chains
5704 if (dev->netdev_ops->ndo_uninit)
5705 dev->netdev_ops->ndo_uninit(dev);
5707 if (!dev->rtnl_link_ops ||
5708 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5709 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5711 /* Notifier chain MUST detach us all upper devices. */
5712 WARN_ON(netdev_has_any_upper_dev(dev));
5714 /* Remove entries from kobject tree */
5715 netdev_unregister_kobject(dev);
5717 /* Remove XPS queueing entries */
5718 netif_reset_xps_queues_gt(dev, 0);
5724 list_for_each_entry(dev, head, unreg_list)
5728 static void rollback_registered(struct net_device *dev)
5732 list_add(&dev->unreg_list, &single);
5733 rollback_registered_many(&single);
5737 static netdev_features_t netdev_fix_features(struct net_device *dev,
5738 netdev_features_t features)
5740 /* Fix illegal checksum combinations */
5741 if ((features & NETIF_F_HW_CSUM) &&
5742 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5743 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5744 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5747 /* TSO requires that SG is present as well. */
5748 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5749 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5750 features &= ~NETIF_F_ALL_TSO;
5753 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5754 !(features & NETIF_F_IP_CSUM)) {
5755 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5756 features &= ~NETIF_F_TSO;
5757 features &= ~NETIF_F_TSO_ECN;
5760 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5761 !(features & NETIF_F_IPV6_CSUM)) {
5762 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5763 features &= ~NETIF_F_TSO6;
5766 /* TSO ECN requires that TSO is present as well. */
5767 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5768 features &= ~NETIF_F_TSO_ECN;
5770 /* Software GSO depends on SG. */
5771 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5772 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5773 features &= ~NETIF_F_GSO;
5776 /* UFO needs SG and checksumming */
5777 if (features & NETIF_F_UFO) {
5778 /* maybe split UFO into V4 and V6? */
5779 if (!((features & NETIF_F_GEN_CSUM) ||
5780 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5781 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5783 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5784 features &= ~NETIF_F_UFO;
5787 if (!(features & NETIF_F_SG)) {
5789 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5790 features &= ~NETIF_F_UFO;
5794 #ifdef CONFIG_NET_RX_BUSY_POLL
5795 if (dev->netdev_ops->ndo_busy_poll)
5796 features |= NETIF_F_BUSY_POLL;
5799 features &= ~NETIF_F_BUSY_POLL;
5804 int __netdev_update_features(struct net_device *dev)
5806 netdev_features_t features;
5811 features = netdev_get_wanted_features(dev);
5813 if (dev->netdev_ops->ndo_fix_features)
5814 features = dev->netdev_ops->ndo_fix_features(dev, features);
5816 /* driver might be less strict about feature dependencies */
5817 features = netdev_fix_features(dev, features);
5819 if (dev->features == features)
5822 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5823 &dev->features, &features);
5825 if (dev->netdev_ops->ndo_set_features)
5826 err = dev->netdev_ops->ndo_set_features(dev, features);
5828 if (unlikely(err < 0)) {
5830 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5831 err, &features, &dev->features);
5836 dev->features = features;
5842 * netdev_update_features - recalculate device features
5843 * @dev: the device to check
5845 * Recalculate dev->features set and send notifications if it
5846 * has changed. Should be called after driver or hardware dependent
5847 * conditions might have changed that influence the features.
5849 void netdev_update_features(struct net_device *dev)
5851 if (__netdev_update_features(dev))
5852 netdev_features_change(dev);
5854 EXPORT_SYMBOL(netdev_update_features);
5857 * netdev_change_features - recalculate device features
5858 * @dev: the device to check
5860 * Recalculate dev->features set and send notifications even
5861 * if they have not changed. Should be called instead of
5862 * netdev_update_features() if also dev->vlan_features might
5863 * have changed to allow the changes to be propagated to stacked
5866 void netdev_change_features(struct net_device *dev)
5868 __netdev_update_features(dev);
5869 netdev_features_change(dev);
5871 EXPORT_SYMBOL(netdev_change_features);
5874 * netif_stacked_transfer_operstate - transfer operstate
5875 * @rootdev: the root or lower level device to transfer state from
5876 * @dev: the device to transfer operstate to
5878 * Transfer operational state from root to device. This is normally
5879 * called when a stacking relationship exists between the root
5880 * device and the device(a leaf device).
5882 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5883 struct net_device *dev)
5885 if (rootdev->operstate == IF_OPER_DORMANT)
5886 netif_dormant_on(dev);
5888 netif_dormant_off(dev);
5890 if (netif_carrier_ok(rootdev)) {
5891 if (!netif_carrier_ok(dev))
5892 netif_carrier_on(dev);
5894 if (netif_carrier_ok(dev))
5895 netif_carrier_off(dev);
5898 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5901 static int netif_alloc_rx_queues(struct net_device *dev)
5903 unsigned int i, count = dev->num_rx_queues;
5904 struct netdev_rx_queue *rx;
5908 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5914 for (i = 0; i < count; i++)
5920 static void netdev_init_one_queue(struct net_device *dev,
5921 struct netdev_queue *queue, void *_unused)
5923 /* Initialize queue lock */
5924 spin_lock_init(&queue->_xmit_lock);
5925 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5926 queue->xmit_lock_owner = -1;
5927 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5930 dql_init(&queue->dql, HZ);
5934 static void netif_free_tx_queues(struct net_device *dev)
5936 if (is_vmalloc_addr(dev->_tx))
5942 static int netif_alloc_netdev_queues(struct net_device *dev)
5944 unsigned int count = dev->num_tx_queues;
5945 struct netdev_queue *tx;
5946 size_t sz = count * sizeof(*tx);
5948 BUG_ON(count < 1 || count > 0xffff);
5950 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
5958 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5959 spin_lock_init(&dev->tx_global_lock);
5965 * register_netdevice - register a network device
5966 * @dev: device to register
5968 * Take a completed network device structure and add it to the kernel
5969 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5970 * chain. 0 is returned on success. A negative errno code is returned
5971 * on a failure to set up the device, or if the name is a duplicate.
5973 * Callers must hold the rtnl semaphore. You may want
5974 * register_netdev() instead of this.
5977 * The locking appears insufficient to guarantee two parallel registers
5978 * will not get the same name.
5981 int register_netdevice(struct net_device *dev)
5984 struct net *net = dev_net(dev);
5986 BUG_ON(dev_boot_phase);
5991 /* When net_device's are persistent, this will be fatal. */
5992 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5995 spin_lock_init(&dev->addr_list_lock);
5996 netdev_set_addr_lockdep_class(dev);
6000 ret = dev_get_valid_name(net, dev, dev->name);
6004 /* Init, if this function is available */
6005 if (dev->netdev_ops->ndo_init) {
6006 ret = dev->netdev_ops->ndo_init(dev);
6014 if (((dev->hw_features | dev->features) &
6015 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6016 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6017 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6018 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6025 dev->ifindex = dev_new_index(net);
6026 else if (__dev_get_by_index(net, dev->ifindex))
6029 if (dev->iflink == -1)
6030 dev->iflink = dev->ifindex;
6032 /* Transfer changeable features to wanted_features and enable
6033 * software offloads (GSO and GRO).
6035 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6036 dev->features |= NETIF_F_SOFT_FEATURES;
6037 dev->wanted_features = dev->features & dev->hw_features;
6039 if (!(dev->flags & IFF_LOOPBACK)) {
6040 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6043 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6045 dev->vlan_features |= NETIF_F_HIGHDMA;
6047 /* Make NETIF_F_SG inheritable to tunnel devices.
6049 dev->hw_enc_features |= NETIF_F_SG;
6051 /* Make NETIF_F_SG inheritable to MPLS.
6053 dev->mpls_features |= NETIF_F_SG;
6055 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6056 ret = notifier_to_errno(ret);
6060 ret = netdev_register_kobject(dev);
6063 dev->reg_state = NETREG_REGISTERED;
6065 __netdev_update_features(dev);
6068 * Default initial state at registry is that the
6069 * device is present.
6072 set_bit(__LINK_STATE_PRESENT, &dev->state);
6074 linkwatch_init_dev(dev);
6076 dev_init_scheduler(dev);
6078 list_netdevice(dev);
6079 add_device_randomness(dev->dev_addr, dev->addr_len);
6081 /* If the device has permanent device address, driver should
6082 * set dev_addr and also addr_assign_type should be set to
6083 * NET_ADDR_PERM (default value).
6085 if (dev->addr_assign_type == NET_ADDR_PERM)
6086 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6088 /* Notify protocols, that a new device appeared. */
6089 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6090 ret = notifier_to_errno(ret);
6092 rollback_registered(dev);
6093 dev->reg_state = NETREG_UNREGISTERED;
6096 * Prevent userspace races by waiting until the network
6097 * device is fully setup before sending notifications.
6099 if (!dev->rtnl_link_ops ||
6100 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6101 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6107 if (dev->netdev_ops->ndo_uninit)
6108 dev->netdev_ops->ndo_uninit(dev);
6111 EXPORT_SYMBOL(register_netdevice);
6114 * init_dummy_netdev - init a dummy network device for NAPI
6115 * @dev: device to init
6117 * This takes a network device structure and initialize the minimum
6118 * amount of fields so it can be used to schedule NAPI polls without
6119 * registering a full blown interface. This is to be used by drivers
6120 * that need to tie several hardware interfaces to a single NAPI
6121 * poll scheduler due to HW limitations.
6123 int init_dummy_netdev(struct net_device *dev)
6125 /* Clear everything. Note we don't initialize spinlocks
6126 * are they aren't supposed to be taken by any of the
6127 * NAPI code and this dummy netdev is supposed to be
6128 * only ever used for NAPI polls
6130 memset(dev, 0, sizeof(struct net_device));
6132 /* make sure we BUG if trying to hit standard
6133 * register/unregister code path
6135 dev->reg_state = NETREG_DUMMY;
6137 /* NAPI wants this */
6138 INIT_LIST_HEAD(&dev->napi_list);
6140 /* a dummy interface is started by default */
6141 set_bit(__LINK_STATE_PRESENT, &dev->state);
6142 set_bit(__LINK_STATE_START, &dev->state);
6144 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6145 * because users of this 'device' dont need to change
6151 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6155 * register_netdev - register a network device
6156 * @dev: device to register
6158 * Take a completed network device structure and add it to the kernel
6159 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6160 * chain. 0 is returned on success. A negative errno code is returned
6161 * on a failure to set up the device, or if the name is a duplicate.
6163 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6164 * and expands the device name if you passed a format string to
6167 int register_netdev(struct net_device *dev)
6172 err = register_netdevice(dev);
6176 EXPORT_SYMBOL(register_netdev);
6178 int netdev_refcnt_read(const struct net_device *dev)
6182 for_each_possible_cpu(i)
6183 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6186 EXPORT_SYMBOL(netdev_refcnt_read);
6189 * netdev_wait_allrefs - wait until all references are gone.
6190 * @dev: target net_device
6192 * This is called when unregistering network devices.
6194 * Any protocol or device that holds a reference should register
6195 * for netdevice notification, and cleanup and put back the
6196 * reference if they receive an UNREGISTER event.
6197 * We can get stuck here if buggy protocols don't correctly
6200 static void netdev_wait_allrefs(struct net_device *dev)
6202 unsigned long rebroadcast_time, warning_time;
6205 linkwatch_forget_dev(dev);
6207 rebroadcast_time = warning_time = jiffies;
6208 refcnt = netdev_refcnt_read(dev);
6210 while (refcnt != 0) {
6211 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6214 /* Rebroadcast unregister notification */
6215 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6221 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6222 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6224 /* We must not have linkwatch events
6225 * pending on unregister. If this
6226 * happens, we simply run the queue
6227 * unscheduled, resulting in a noop
6230 linkwatch_run_queue();
6235 rebroadcast_time = jiffies;
6240 refcnt = netdev_refcnt_read(dev);
6242 if (time_after(jiffies, warning_time + 10 * HZ)) {
6243 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6245 warning_time = jiffies;
6254 * register_netdevice(x1);
6255 * register_netdevice(x2);
6257 * unregister_netdevice(y1);
6258 * unregister_netdevice(y2);
6264 * We are invoked by rtnl_unlock().
6265 * This allows us to deal with problems:
6266 * 1) We can delete sysfs objects which invoke hotplug
6267 * without deadlocking with linkwatch via keventd.
6268 * 2) Since we run with the RTNL semaphore not held, we can sleep
6269 * safely in order to wait for the netdev refcnt to drop to zero.
6271 * We must not return until all unregister events added during
6272 * the interval the lock was held have been completed.
6274 void netdev_run_todo(void)
6276 struct list_head list;
6278 /* Snapshot list, allow later requests */
6279 list_replace_init(&net_todo_list, &list);
6284 /* Wait for rcu callbacks to finish before next phase */
6285 if (!list_empty(&list))
6288 while (!list_empty(&list)) {
6289 struct net_device *dev
6290 = list_first_entry(&list, struct net_device, todo_list);
6291 list_del(&dev->todo_list);
6294 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6297 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6298 pr_err("network todo '%s' but state %d\n",
6299 dev->name, dev->reg_state);
6304 dev->reg_state = NETREG_UNREGISTERED;
6306 on_each_cpu(flush_backlog, dev, 1);
6308 netdev_wait_allrefs(dev);
6311 BUG_ON(netdev_refcnt_read(dev));
6312 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6313 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6314 WARN_ON(dev->dn_ptr);
6316 if (dev->destructor)
6317 dev->destructor(dev);
6319 /* Report a network device has been unregistered */
6321 dev_net(dev)->dev_unreg_count--;
6323 wake_up(&netdev_unregistering_wq);
6325 /* Free network device */
6326 kobject_put(&dev->dev.kobj);
6330 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6331 * fields in the same order, with only the type differing.
6333 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6334 const struct net_device_stats *netdev_stats)
6336 #if BITS_PER_LONG == 64
6337 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6338 memcpy(stats64, netdev_stats, sizeof(*stats64));
6340 size_t i, n = sizeof(*stats64) / sizeof(u64);
6341 const unsigned long *src = (const unsigned long *)netdev_stats;
6342 u64 *dst = (u64 *)stats64;
6344 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6345 sizeof(*stats64) / sizeof(u64));
6346 for (i = 0; i < n; i++)
6350 EXPORT_SYMBOL(netdev_stats_to_stats64);
6353 * dev_get_stats - get network device statistics
6354 * @dev: device to get statistics from
6355 * @storage: place to store stats
6357 * Get network statistics from device. Return @storage.
6358 * The device driver may provide its own method by setting
6359 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6360 * otherwise the internal statistics structure is used.
6362 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6363 struct rtnl_link_stats64 *storage)
6365 const struct net_device_ops *ops = dev->netdev_ops;
6367 if (ops->ndo_get_stats64) {
6368 memset(storage, 0, sizeof(*storage));
6369 ops->ndo_get_stats64(dev, storage);
6370 } else if (ops->ndo_get_stats) {
6371 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6373 netdev_stats_to_stats64(storage, &dev->stats);
6375 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6376 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6379 EXPORT_SYMBOL(dev_get_stats);
6381 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6383 struct netdev_queue *queue = dev_ingress_queue(dev);
6385 #ifdef CONFIG_NET_CLS_ACT
6388 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6391 netdev_init_one_queue(dev, queue, NULL);
6392 queue->qdisc = &noop_qdisc;
6393 queue->qdisc_sleeping = &noop_qdisc;
6394 rcu_assign_pointer(dev->ingress_queue, queue);
6399 static const struct ethtool_ops default_ethtool_ops;
6401 void netdev_set_default_ethtool_ops(struct net_device *dev,
6402 const struct ethtool_ops *ops)
6404 if (dev->ethtool_ops == &default_ethtool_ops)
6405 dev->ethtool_ops = ops;
6407 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6409 void netdev_freemem(struct net_device *dev)
6411 char *addr = (char *)dev - dev->padded;
6413 if (is_vmalloc_addr(addr))
6420 * alloc_netdev_mqs - allocate network device
6421 * @sizeof_priv: size of private data to allocate space for
6422 * @name: device name format string
6423 * @setup: callback to initialize device
6424 * @txqs: the number of TX subqueues to allocate
6425 * @rxqs: the number of RX subqueues to allocate
6427 * Allocates a struct net_device with private data area for driver use
6428 * and performs basic initialization. Also allocates subqueue structs
6429 * for each queue on the device.
6431 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6432 void (*setup)(struct net_device *),
6433 unsigned int txqs, unsigned int rxqs)
6435 struct net_device *dev;
6437 struct net_device *p;
6439 BUG_ON(strlen(name) >= sizeof(dev->name));
6442 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6448 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6453 alloc_size = sizeof(struct net_device);
6455 /* ensure 32-byte alignment of private area */
6456 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6457 alloc_size += sizeof_priv;
6459 /* ensure 32-byte alignment of whole construct */
6460 alloc_size += NETDEV_ALIGN - 1;
6462 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6464 p = vzalloc(alloc_size);
6468 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6469 dev->padded = (char *)dev - (char *)p;
6471 dev->pcpu_refcnt = alloc_percpu(int);
6472 if (!dev->pcpu_refcnt)
6475 if (dev_addr_init(dev))
6481 dev_net_set(dev, &init_net);
6483 dev->gso_max_size = GSO_MAX_SIZE;
6484 dev->gso_max_segs = GSO_MAX_SEGS;
6486 INIT_LIST_HEAD(&dev->napi_list);
6487 INIT_LIST_HEAD(&dev->unreg_list);
6488 INIT_LIST_HEAD(&dev->close_list);
6489 INIT_LIST_HEAD(&dev->link_watch_list);
6490 INIT_LIST_HEAD(&dev->adj_list.upper);
6491 INIT_LIST_HEAD(&dev->adj_list.lower);
6492 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6493 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6494 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6497 dev->num_tx_queues = txqs;
6498 dev->real_num_tx_queues = txqs;
6499 if (netif_alloc_netdev_queues(dev))
6503 dev->num_rx_queues = rxqs;
6504 dev->real_num_rx_queues = rxqs;
6505 if (netif_alloc_rx_queues(dev))
6509 strcpy(dev->name, name);
6510 dev->group = INIT_NETDEV_GROUP;
6511 if (!dev->ethtool_ops)
6512 dev->ethtool_ops = &default_ethtool_ops;
6520 free_percpu(dev->pcpu_refcnt);
6522 netdev_freemem(dev);
6525 EXPORT_SYMBOL(alloc_netdev_mqs);
6528 * free_netdev - free network device
6531 * This function does the last stage of destroying an allocated device
6532 * interface. The reference to the device object is released.
6533 * If this is the last reference then it will be freed.
6535 void free_netdev(struct net_device *dev)
6537 struct napi_struct *p, *n;
6539 release_net(dev_net(dev));
6541 netif_free_tx_queues(dev);
6546 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6548 /* Flush device addresses */
6549 dev_addr_flush(dev);
6551 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6554 free_percpu(dev->pcpu_refcnt);
6555 dev->pcpu_refcnt = NULL;
6557 /* Compatibility with error handling in drivers */
6558 if (dev->reg_state == NETREG_UNINITIALIZED) {
6559 netdev_freemem(dev);
6563 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6564 dev->reg_state = NETREG_RELEASED;
6566 /* will free via device release */
6567 put_device(&dev->dev);
6569 EXPORT_SYMBOL(free_netdev);
6572 * synchronize_net - Synchronize with packet receive processing
6574 * Wait for packets currently being received to be done.
6575 * Does not block later packets from starting.
6577 void synchronize_net(void)
6580 if (rtnl_is_locked())
6581 synchronize_rcu_expedited();
6585 EXPORT_SYMBOL(synchronize_net);
6588 * unregister_netdevice_queue - remove device from the kernel
6592 * This function shuts down a device interface and removes it
6593 * from the kernel tables.
6594 * If head not NULL, device is queued to be unregistered later.
6596 * Callers must hold the rtnl semaphore. You may want
6597 * unregister_netdev() instead of this.
6600 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6605 list_move_tail(&dev->unreg_list, head);
6607 rollback_registered(dev);
6608 /* Finish processing unregister after unlock */
6612 EXPORT_SYMBOL(unregister_netdevice_queue);
6615 * unregister_netdevice_many - unregister many devices
6616 * @head: list of devices
6618 void unregister_netdevice_many(struct list_head *head)
6620 struct net_device *dev;
6622 if (!list_empty(head)) {
6623 rollback_registered_many(head);
6624 list_for_each_entry(dev, head, unreg_list)
6628 EXPORT_SYMBOL(unregister_netdevice_many);
6631 * unregister_netdev - remove device from the kernel
6634 * This function shuts down a device interface and removes it
6635 * from the kernel tables.
6637 * This is just a wrapper for unregister_netdevice that takes
6638 * the rtnl semaphore. In general you want to use this and not
6639 * unregister_netdevice.
6641 void unregister_netdev(struct net_device *dev)
6644 unregister_netdevice(dev);
6647 EXPORT_SYMBOL(unregister_netdev);
6650 * dev_change_net_namespace - move device to different nethost namespace
6652 * @net: network namespace
6653 * @pat: If not NULL name pattern to try if the current device name
6654 * is already taken in the destination network namespace.
6656 * This function shuts down a device interface and moves it
6657 * to a new network namespace. On success 0 is returned, on
6658 * a failure a netagive errno code is returned.
6660 * Callers must hold the rtnl semaphore.
6663 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6669 /* Don't allow namespace local devices to be moved. */
6671 if (dev->features & NETIF_F_NETNS_LOCAL)
6674 /* Ensure the device has been registrered */
6675 if (dev->reg_state != NETREG_REGISTERED)
6678 /* Get out if there is nothing todo */
6680 if (net_eq(dev_net(dev), net))
6683 /* Pick the destination device name, and ensure
6684 * we can use it in the destination network namespace.
6687 if (__dev_get_by_name(net, dev->name)) {
6688 /* We get here if we can't use the current device name */
6691 if (dev_get_valid_name(net, dev, pat) < 0)
6696 * And now a mini version of register_netdevice unregister_netdevice.
6699 /* If device is running close it first. */
6702 /* And unlink it from device chain */
6704 unlist_netdevice(dev);
6708 /* Shutdown queueing discipline. */
6711 /* Notify protocols, that we are about to destroy
6712 this device. They should clean all the things.
6714 Note that dev->reg_state stays at NETREG_REGISTERED.
6715 This is wanted because this way 8021q and macvlan know
6716 the device is just moving and can keep their slaves up.
6718 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6720 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6721 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6724 * Flush the unicast and multicast chains
6729 /* Send a netdev-removed uevent to the old namespace */
6730 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6732 /* Actually switch the network namespace */
6733 dev_net_set(dev, net);
6735 /* If there is an ifindex conflict assign a new one */
6736 if (__dev_get_by_index(net, dev->ifindex)) {
6737 int iflink = (dev->iflink == dev->ifindex);
6738 dev->ifindex = dev_new_index(net);
6740 dev->iflink = dev->ifindex;
6743 /* Send a netdev-add uevent to the new namespace */
6744 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6746 /* Fixup kobjects */
6747 err = device_rename(&dev->dev, dev->name);
6750 /* Add the device back in the hashes */
6751 list_netdevice(dev);
6753 /* Notify protocols, that a new device appeared. */
6754 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6757 * Prevent userspace races by waiting until the network
6758 * device is fully setup before sending notifications.
6760 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6767 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6769 static int dev_cpu_callback(struct notifier_block *nfb,
6770 unsigned long action,
6773 struct sk_buff **list_skb;
6774 struct sk_buff *skb;
6775 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6776 struct softnet_data *sd, *oldsd;
6778 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6781 local_irq_disable();
6782 cpu = smp_processor_id();
6783 sd = &per_cpu(softnet_data, cpu);
6784 oldsd = &per_cpu(softnet_data, oldcpu);
6786 /* Find end of our completion_queue. */
6787 list_skb = &sd->completion_queue;
6789 list_skb = &(*list_skb)->next;
6790 /* Append completion queue from offline CPU. */
6791 *list_skb = oldsd->completion_queue;
6792 oldsd->completion_queue = NULL;
6794 /* Append output queue from offline CPU. */
6795 if (oldsd->output_queue) {
6796 *sd->output_queue_tailp = oldsd->output_queue;
6797 sd->output_queue_tailp = oldsd->output_queue_tailp;
6798 oldsd->output_queue = NULL;
6799 oldsd->output_queue_tailp = &oldsd->output_queue;
6801 /* Append NAPI poll list from offline CPU. */
6802 if (!list_empty(&oldsd->poll_list)) {
6803 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6804 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6807 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6810 /* Process offline CPU's input_pkt_queue */
6811 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6812 netif_rx_internal(skb);
6813 input_queue_head_incr(oldsd);
6815 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6816 netif_rx_internal(skb);
6817 input_queue_head_incr(oldsd);
6825 * netdev_increment_features - increment feature set by one
6826 * @all: current feature set
6827 * @one: new feature set
6828 * @mask: mask feature set
6830 * Computes a new feature set after adding a device with feature set
6831 * @one to the master device with current feature set @all. Will not
6832 * enable anything that is off in @mask. Returns the new feature set.
6834 netdev_features_t netdev_increment_features(netdev_features_t all,
6835 netdev_features_t one, netdev_features_t mask)
6837 if (mask & NETIF_F_GEN_CSUM)
6838 mask |= NETIF_F_ALL_CSUM;
6839 mask |= NETIF_F_VLAN_CHALLENGED;
6841 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6842 all &= one | ~NETIF_F_ALL_FOR_ALL;
6844 /* If one device supports hw checksumming, set for all. */
6845 if (all & NETIF_F_GEN_CSUM)
6846 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6850 EXPORT_SYMBOL(netdev_increment_features);
6852 static struct hlist_head * __net_init netdev_create_hash(void)
6855 struct hlist_head *hash;
6857 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6859 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6860 INIT_HLIST_HEAD(&hash[i]);
6865 /* Initialize per network namespace state */
6866 static int __net_init netdev_init(struct net *net)
6868 if (net != &init_net)
6869 INIT_LIST_HEAD(&net->dev_base_head);
6871 net->dev_name_head = netdev_create_hash();
6872 if (net->dev_name_head == NULL)
6875 net->dev_index_head = netdev_create_hash();
6876 if (net->dev_index_head == NULL)
6882 kfree(net->dev_name_head);
6888 * netdev_drivername - network driver for the device
6889 * @dev: network device
6891 * Determine network driver for device.
6893 const char *netdev_drivername(const struct net_device *dev)
6895 const struct device_driver *driver;
6896 const struct device *parent;
6897 const char *empty = "";
6899 parent = dev->dev.parent;
6903 driver = parent->driver;
6904 if (driver && driver->name)
6905 return driver->name;
6909 static int __netdev_printk(const char *level, const struct net_device *dev,
6910 struct va_format *vaf)
6914 if (dev && dev->dev.parent) {
6915 r = dev_printk_emit(level[1] - '0',
6918 dev_driver_string(dev->dev.parent),
6919 dev_name(dev->dev.parent),
6920 netdev_name(dev), vaf);
6922 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6924 r = printk("%s(NULL net_device): %pV", level, vaf);
6930 int netdev_printk(const char *level, const struct net_device *dev,
6931 const char *format, ...)
6933 struct va_format vaf;
6937 va_start(args, format);
6942 r = __netdev_printk(level, dev, &vaf);
6948 EXPORT_SYMBOL(netdev_printk);
6950 #define define_netdev_printk_level(func, level) \
6951 int func(const struct net_device *dev, const char *fmt, ...) \
6954 struct va_format vaf; \
6957 va_start(args, fmt); \
6962 r = __netdev_printk(level, dev, &vaf); \
6968 EXPORT_SYMBOL(func);
6970 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6971 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6972 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6973 define_netdev_printk_level(netdev_err, KERN_ERR);
6974 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6975 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6976 define_netdev_printk_level(netdev_info, KERN_INFO);
6978 static void __net_exit netdev_exit(struct net *net)
6980 kfree(net->dev_name_head);
6981 kfree(net->dev_index_head);
6984 static struct pernet_operations __net_initdata netdev_net_ops = {
6985 .init = netdev_init,
6986 .exit = netdev_exit,
6989 static void __net_exit default_device_exit(struct net *net)
6991 struct net_device *dev, *aux;
6993 * Push all migratable network devices back to the
6994 * initial network namespace
6997 for_each_netdev_safe(net, dev, aux) {
6999 char fb_name[IFNAMSIZ];
7001 /* Ignore unmoveable devices (i.e. loopback) */
7002 if (dev->features & NETIF_F_NETNS_LOCAL)
7005 /* Leave virtual devices for the generic cleanup */
7006 if (dev->rtnl_link_ops)
7009 /* Push remaining network devices to init_net */
7010 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7011 err = dev_change_net_namespace(dev, &init_net, fb_name);
7013 pr_emerg("%s: failed to move %s to init_net: %d\n",
7014 __func__, dev->name, err);
7021 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7023 /* Return with the rtnl_lock held when there are no network
7024 * devices unregistering in any network namespace in net_list.
7031 prepare_to_wait(&netdev_unregistering_wq, &wait,
7032 TASK_UNINTERRUPTIBLE);
7033 unregistering = false;
7035 list_for_each_entry(net, net_list, exit_list) {
7036 if (net->dev_unreg_count > 0) {
7037 unregistering = true;
7046 finish_wait(&netdev_unregistering_wq, &wait);
7049 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7051 /* At exit all network devices most be removed from a network
7052 * namespace. Do this in the reverse order of registration.
7053 * Do this across as many network namespaces as possible to
7054 * improve batching efficiency.
7056 struct net_device *dev;
7058 LIST_HEAD(dev_kill_list);
7060 /* To prevent network device cleanup code from dereferencing
7061 * loopback devices or network devices that have been freed
7062 * wait here for all pending unregistrations to complete,
7063 * before unregistring the loopback device and allowing the
7064 * network namespace be freed.
7066 * The netdev todo list containing all network devices
7067 * unregistrations that happen in default_device_exit_batch
7068 * will run in the rtnl_unlock() at the end of
7069 * default_device_exit_batch.
7071 rtnl_lock_unregistering(net_list);
7072 list_for_each_entry(net, net_list, exit_list) {
7073 for_each_netdev_reverse(net, dev) {
7074 if (dev->rtnl_link_ops)
7075 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7077 unregister_netdevice_queue(dev, &dev_kill_list);
7080 unregister_netdevice_many(&dev_kill_list);
7081 list_del(&dev_kill_list);
7085 static struct pernet_operations __net_initdata default_device_ops = {
7086 .exit = default_device_exit,
7087 .exit_batch = default_device_exit_batch,
7091 * Initialize the DEV module. At boot time this walks the device list and
7092 * unhooks any devices that fail to initialise (normally hardware not
7093 * present) and leaves us with a valid list of present and active devices.
7098 * This is called single threaded during boot, so no need
7099 * to take the rtnl semaphore.
7101 static int __init net_dev_init(void)
7103 int i, rc = -ENOMEM;
7105 BUG_ON(!dev_boot_phase);
7107 if (dev_proc_init())
7110 if (netdev_kobject_init())
7113 INIT_LIST_HEAD(&ptype_all);
7114 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7115 INIT_LIST_HEAD(&ptype_base[i]);
7117 INIT_LIST_HEAD(&offload_base);
7119 if (register_pernet_subsys(&netdev_net_ops))
7123 * Initialise the packet receive queues.
7126 for_each_possible_cpu(i) {
7127 struct softnet_data *sd = &per_cpu(softnet_data, i);
7129 skb_queue_head_init(&sd->input_pkt_queue);
7130 skb_queue_head_init(&sd->process_queue);
7131 INIT_LIST_HEAD(&sd->poll_list);
7132 sd->output_queue_tailp = &sd->output_queue;
7134 sd->csd.func = rps_trigger_softirq;
7139 sd->backlog.poll = process_backlog;
7140 sd->backlog.weight = weight_p;
7145 /* The loopback device is special if any other network devices
7146 * is present in a network namespace the loopback device must
7147 * be present. Since we now dynamically allocate and free the
7148 * loopback device ensure this invariant is maintained by
7149 * keeping the loopback device as the first device on the
7150 * list of network devices. Ensuring the loopback devices
7151 * is the first device that appears and the last network device
7154 if (register_pernet_device(&loopback_net_ops))
7157 if (register_pernet_device(&default_device_ops))
7160 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7161 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7163 hotcpu_notifier(dev_cpu_callback, 0);
7170 subsys_initcall(net_dev_init);