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 <net/mpls.h>
122 #include <linux/ipv6.h>
123 #include <linux/in.h>
124 #include <linux/jhash.h>
125 #include <linux/random.h>
126 #include <trace/events/napi.h>
127 #include <trace/events/net.h>
128 #include <trace/events/skb.h>
129 #include <linux/pci.h>
130 #include <linux/inetdevice.h>
131 #include <linux/cpu_rmap.h>
132 #include <linux/static_key.h>
133 #include <linux/hashtable.h>
134 #include <linux/vmalloc.h>
135 #include <linux/if_macvlan.h>
136 #include <linux/errqueue.h>
137 #include <linux/hrtimer.h>
139 #include "net-sysfs.h"
141 /* Instead of increasing this, you should create a hash table. */
142 #define MAX_GRO_SKBS 8
144 /* This should be increased if a protocol with a bigger head is added. */
145 #define GRO_MAX_HEAD (MAX_HEADER + 128)
147 static DEFINE_SPINLOCK(ptype_lock);
148 static DEFINE_SPINLOCK(offload_lock);
149 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
150 struct list_head ptype_all __read_mostly; /* Taps */
151 static struct list_head offload_base __read_mostly;
153 static int netif_rx_internal(struct sk_buff *skb);
154 static int call_netdevice_notifiers_info(unsigned long val,
155 struct net_device *dev,
156 struct netdev_notifier_info *info);
159 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
162 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
164 * Writers must hold the rtnl semaphore while they loop through the
165 * dev_base_head list, and hold dev_base_lock for writing when they do the
166 * actual updates. This allows pure readers to access the list even
167 * while a writer is preparing to update it.
169 * To put it another way, dev_base_lock is held for writing only to
170 * protect against pure readers; the rtnl semaphore provides the
171 * protection against other writers.
173 * See, for example usages, register_netdevice() and
174 * unregister_netdevice(), which must be called with the rtnl
177 DEFINE_RWLOCK(dev_base_lock);
178 EXPORT_SYMBOL(dev_base_lock);
180 /* protects napi_hash addition/deletion and napi_gen_id */
181 static DEFINE_SPINLOCK(napi_hash_lock);
183 static unsigned int napi_gen_id;
184 static DEFINE_HASHTABLE(napi_hash, 8);
186 static seqcount_t devnet_rename_seq;
188 static inline void dev_base_seq_inc(struct net *net)
190 while (++net->dev_base_seq == 0);
193 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
195 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
197 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
200 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
202 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
205 static inline void rps_lock(struct softnet_data *sd)
208 spin_lock(&sd->input_pkt_queue.lock);
212 static inline void rps_unlock(struct softnet_data *sd)
215 spin_unlock(&sd->input_pkt_queue.lock);
219 /* Device list insertion */
220 static void list_netdevice(struct net_device *dev)
222 struct net *net = dev_net(dev);
226 write_lock_bh(&dev_base_lock);
227 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
228 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
229 hlist_add_head_rcu(&dev->index_hlist,
230 dev_index_hash(net, dev->ifindex));
231 write_unlock_bh(&dev_base_lock);
233 dev_base_seq_inc(net);
236 /* Device list removal
237 * caller must respect a RCU grace period before freeing/reusing dev
239 static void unlist_netdevice(struct net_device *dev)
243 /* Unlink dev from the device chain */
244 write_lock_bh(&dev_base_lock);
245 list_del_rcu(&dev->dev_list);
246 hlist_del_rcu(&dev->name_hlist);
247 hlist_del_rcu(&dev->index_hlist);
248 write_unlock_bh(&dev_base_lock);
250 dev_base_seq_inc(dev_net(dev));
257 static RAW_NOTIFIER_HEAD(netdev_chain);
260 * Device drivers call our routines to queue packets here. We empty the
261 * queue in the local softnet handler.
264 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
265 EXPORT_PER_CPU_SYMBOL(softnet_data);
267 #ifdef CONFIG_LOCKDEP
269 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
270 * according to dev->type
272 static const unsigned short netdev_lock_type[] =
273 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
274 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
275 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
276 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
277 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
278 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
279 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
280 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
281 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
282 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
283 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
284 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
285 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
286 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
287 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
289 static const char *const netdev_lock_name[] =
290 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
291 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
292 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
293 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
294 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
295 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
296 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
297 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
298 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
299 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
300 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
301 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
302 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
303 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
304 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
306 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
307 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
309 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
313 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
314 if (netdev_lock_type[i] == dev_type)
316 /* the last key is used by default */
317 return ARRAY_SIZE(netdev_lock_type) - 1;
320 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
321 unsigned short dev_type)
325 i = netdev_lock_pos(dev_type);
326 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
327 netdev_lock_name[i]);
330 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
334 i = netdev_lock_pos(dev->type);
335 lockdep_set_class_and_name(&dev->addr_list_lock,
336 &netdev_addr_lock_key[i],
337 netdev_lock_name[i]);
340 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
341 unsigned short dev_type)
344 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
349 /*******************************************************************************
351 Protocol management and registration routines
353 *******************************************************************************/
356 * Add a protocol ID to the list. Now that the input handler is
357 * smarter we can dispense with all the messy stuff that used to be
360 * BEWARE!!! Protocol handlers, mangling input packets,
361 * MUST BE last in hash buckets and checking protocol handlers
362 * MUST start from promiscuous ptype_all chain in net_bh.
363 * It is true now, do not change it.
364 * Explanation follows: if protocol handler, mangling packet, will
365 * be the first on list, it is not able to sense, that packet
366 * is cloned and should be copied-on-write, so that it will
367 * change it and subsequent readers will get broken packet.
371 static inline struct list_head *ptype_head(const struct packet_type *pt)
373 if (pt->type == htons(ETH_P_ALL))
374 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
376 return pt->dev ? &pt->dev->ptype_specific :
377 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
381 * dev_add_pack - add packet handler
382 * @pt: packet type declaration
384 * Add a protocol handler to the networking stack. The passed &packet_type
385 * is linked into kernel lists and may not be freed until it has been
386 * removed from the kernel lists.
388 * This call does not sleep therefore it can not
389 * guarantee all CPU's that are in middle of receiving packets
390 * will see the new packet type (until the next received packet).
393 void dev_add_pack(struct packet_type *pt)
395 struct list_head *head = ptype_head(pt);
397 spin_lock(&ptype_lock);
398 list_add_rcu(&pt->list, head);
399 spin_unlock(&ptype_lock);
401 EXPORT_SYMBOL(dev_add_pack);
404 * __dev_remove_pack - remove packet handler
405 * @pt: packet type declaration
407 * Remove a protocol handler that was previously added to the kernel
408 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
409 * from the kernel lists and can be freed or reused once this function
412 * The packet type might still be in use by receivers
413 * and must not be freed until after all the CPU's have gone
414 * through a quiescent state.
416 void __dev_remove_pack(struct packet_type *pt)
418 struct list_head *head = ptype_head(pt);
419 struct packet_type *pt1;
421 spin_lock(&ptype_lock);
423 list_for_each_entry(pt1, head, list) {
425 list_del_rcu(&pt->list);
430 pr_warn("dev_remove_pack: %p not found\n", pt);
432 spin_unlock(&ptype_lock);
434 EXPORT_SYMBOL(__dev_remove_pack);
437 * dev_remove_pack - remove packet handler
438 * @pt: packet type declaration
440 * Remove a protocol handler that was previously added to the kernel
441 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
442 * from the kernel lists and can be freed or reused once this function
445 * This call sleeps to guarantee that no CPU is looking at the packet
448 void dev_remove_pack(struct packet_type *pt)
450 __dev_remove_pack(pt);
454 EXPORT_SYMBOL(dev_remove_pack);
458 * dev_add_offload - register offload handlers
459 * @po: protocol offload declaration
461 * Add protocol offload handlers to the networking stack. The passed
462 * &proto_offload is linked into kernel lists and may not be freed until
463 * it has been removed from the kernel lists.
465 * This call does not sleep therefore it can not
466 * guarantee all CPU's that are in middle of receiving packets
467 * will see the new offload handlers (until the next received packet).
469 void dev_add_offload(struct packet_offload *po)
471 struct list_head *head = &offload_base;
473 spin_lock(&offload_lock);
474 list_add_rcu(&po->list, head);
475 spin_unlock(&offload_lock);
477 EXPORT_SYMBOL(dev_add_offload);
480 * __dev_remove_offload - remove offload handler
481 * @po: packet offload declaration
483 * Remove a protocol offload handler that was previously added to the
484 * kernel offload handlers by dev_add_offload(). The passed &offload_type
485 * is removed from the kernel lists and can be freed or reused once this
488 * The packet type might still be in use by receivers
489 * and must not be freed until after all the CPU's have gone
490 * through a quiescent state.
492 static void __dev_remove_offload(struct packet_offload *po)
494 struct list_head *head = &offload_base;
495 struct packet_offload *po1;
497 spin_lock(&offload_lock);
499 list_for_each_entry(po1, head, list) {
501 list_del_rcu(&po->list);
506 pr_warn("dev_remove_offload: %p not found\n", po);
508 spin_unlock(&offload_lock);
512 * dev_remove_offload - remove packet offload handler
513 * @po: packet offload declaration
515 * Remove a packet offload handler that was previously added to the kernel
516 * offload handlers by dev_add_offload(). The passed &offload_type is
517 * removed from the kernel lists and can be freed or reused once this
520 * This call sleeps to guarantee that no CPU is looking at the packet
523 void dev_remove_offload(struct packet_offload *po)
525 __dev_remove_offload(po);
529 EXPORT_SYMBOL(dev_remove_offload);
531 /******************************************************************************
533 Device Boot-time Settings Routines
535 *******************************************************************************/
537 /* Boot time configuration table */
538 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
541 * netdev_boot_setup_add - add new setup entry
542 * @name: name of the device
543 * @map: configured settings for the device
545 * Adds new setup entry to the dev_boot_setup list. The function
546 * returns 0 on error and 1 on success. This is a generic routine to
549 static int netdev_boot_setup_add(char *name, struct ifmap *map)
551 struct netdev_boot_setup *s;
555 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
556 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
557 memset(s[i].name, 0, sizeof(s[i].name));
558 strlcpy(s[i].name, name, IFNAMSIZ);
559 memcpy(&s[i].map, map, sizeof(s[i].map));
564 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
568 * netdev_boot_setup_check - check boot time settings
569 * @dev: the netdevice
571 * Check boot time settings for the device.
572 * The found settings are set for the device to be used
573 * later in the device probing.
574 * Returns 0 if no settings found, 1 if they are.
576 int netdev_boot_setup_check(struct net_device *dev)
578 struct netdev_boot_setup *s = dev_boot_setup;
581 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
582 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
583 !strcmp(dev->name, s[i].name)) {
584 dev->irq = s[i].map.irq;
585 dev->base_addr = s[i].map.base_addr;
586 dev->mem_start = s[i].map.mem_start;
587 dev->mem_end = s[i].map.mem_end;
593 EXPORT_SYMBOL(netdev_boot_setup_check);
597 * netdev_boot_base - get address from boot time settings
598 * @prefix: prefix for network device
599 * @unit: id for network device
601 * Check boot time settings for the base address of device.
602 * The found settings are set for the device to be used
603 * later in the device probing.
604 * Returns 0 if no settings found.
606 unsigned long netdev_boot_base(const char *prefix, int unit)
608 const struct netdev_boot_setup *s = dev_boot_setup;
612 sprintf(name, "%s%d", prefix, unit);
615 * If device already registered then return base of 1
616 * to indicate not to probe for this interface
618 if (__dev_get_by_name(&init_net, name))
621 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
622 if (!strcmp(name, s[i].name))
623 return s[i].map.base_addr;
628 * Saves at boot time configured settings for any netdevice.
630 int __init netdev_boot_setup(char *str)
635 str = get_options(str, ARRAY_SIZE(ints), ints);
640 memset(&map, 0, sizeof(map));
644 map.base_addr = ints[2];
646 map.mem_start = ints[3];
648 map.mem_end = ints[4];
650 /* Add new entry to the list */
651 return netdev_boot_setup_add(str, &map);
654 __setup("netdev=", netdev_boot_setup);
656 /*******************************************************************************
658 Device Interface Subroutines
660 *******************************************************************************/
663 * __dev_get_by_name - find a device by its name
664 * @net: the applicable net namespace
665 * @name: name to find
667 * Find an interface by name. Must be called under RTNL semaphore
668 * or @dev_base_lock. If the name is found a pointer to the device
669 * is returned. If the name is not found then %NULL is returned. The
670 * reference counters are not incremented so the caller must be
671 * careful with locks.
674 struct net_device *__dev_get_by_name(struct net *net, const char *name)
676 struct net_device *dev;
677 struct hlist_head *head = dev_name_hash(net, name);
679 hlist_for_each_entry(dev, head, name_hlist)
680 if (!strncmp(dev->name, name, IFNAMSIZ))
685 EXPORT_SYMBOL(__dev_get_by_name);
688 * dev_get_by_name_rcu - find a device by its name
689 * @net: the applicable net namespace
690 * @name: name to find
692 * Find an interface by name.
693 * If the name is found a pointer to the device is returned.
694 * If the name is not found then %NULL is returned.
695 * The reference counters are not incremented so the caller must be
696 * careful with locks. The caller must hold RCU lock.
699 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
701 struct net_device *dev;
702 struct hlist_head *head = dev_name_hash(net, name);
704 hlist_for_each_entry_rcu(dev, head, name_hlist)
705 if (!strncmp(dev->name, name, IFNAMSIZ))
710 EXPORT_SYMBOL(dev_get_by_name_rcu);
713 * dev_get_by_name - find a device by its name
714 * @net: the applicable net namespace
715 * @name: name to find
717 * Find an interface by name. This can be called from any
718 * context and does its own locking. The returned handle has
719 * the usage count incremented and the caller must use dev_put() to
720 * release it when it is no longer needed. %NULL is returned if no
721 * matching device is found.
724 struct net_device *dev_get_by_name(struct net *net, const char *name)
726 struct net_device *dev;
729 dev = dev_get_by_name_rcu(net, name);
735 EXPORT_SYMBOL(dev_get_by_name);
738 * __dev_get_by_index - find a device by its ifindex
739 * @net: the applicable net namespace
740 * @ifindex: index of device
742 * Search for an interface by index. Returns %NULL if the device
743 * is not found or a pointer to the device. The device has not
744 * had its reference counter increased so the caller must be careful
745 * about locking. The caller must hold either the RTNL semaphore
749 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
751 struct net_device *dev;
752 struct hlist_head *head = dev_index_hash(net, ifindex);
754 hlist_for_each_entry(dev, head, index_hlist)
755 if (dev->ifindex == ifindex)
760 EXPORT_SYMBOL(__dev_get_by_index);
763 * dev_get_by_index_rcu - find a device by its ifindex
764 * @net: the applicable net namespace
765 * @ifindex: index of device
767 * Search for an interface by index. Returns %NULL if the device
768 * is not found or a pointer to the device. The device has not
769 * had its reference counter increased so the caller must be careful
770 * about locking. The caller must hold RCU lock.
773 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
775 struct net_device *dev;
776 struct hlist_head *head = dev_index_hash(net, ifindex);
778 hlist_for_each_entry_rcu(dev, head, index_hlist)
779 if (dev->ifindex == ifindex)
784 EXPORT_SYMBOL(dev_get_by_index_rcu);
788 * dev_get_by_index - find a device by its ifindex
789 * @net: the applicable net namespace
790 * @ifindex: index of device
792 * Search for an interface by index. Returns NULL if the device
793 * is not found or a pointer to the device. The device returned has
794 * had a reference added and the pointer is safe until the user calls
795 * dev_put to indicate they have finished with it.
798 struct net_device *dev_get_by_index(struct net *net, int ifindex)
800 struct net_device *dev;
803 dev = dev_get_by_index_rcu(net, ifindex);
809 EXPORT_SYMBOL(dev_get_by_index);
812 * netdev_get_name - get a netdevice name, knowing its ifindex.
813 * @net: network namespace
814 * @name: a pointer to the buffer where the name will be stored.
815 * @ifindex: the ifindex of the interface to get the name from.
817 * The use of raw_seqcount_begin() and cond_resched() before
818 * retrying is required as we want to give the writers a chance
819 * to complete when CONFIG_PREEMPT is not set.
821 int netdev_get_name(struct net *net, char *name, int ifindex)
823 struct net_device *dev;
827 seq = raw_seqcount_begin(&devnet_rename_seq);
829 dev = dev_get_by_index_rcu(net, ifindex);
835 strcpy(name, dev->name);
837 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
846 * dev_getbyhwaddr_rcu - find a device by its hardware address
847 * @net: the applicable net namespace
848 * @type: media type of device
849 * @ha: hardware address
851 * Search for an interface by MAC address. Returns NULL if the device
852 * is not found or a pointer to the device.
853 * The caller must hold RCU or RTNL.
854 * The returned device has not had its ref count increased
855 * and the caller must therefore be careful about locking
859 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
862 struct net_device *dev;
864 for_each_netdev_rcu(net, dev)
865 if (dev->type == type &&
866 !memcmp(dev->dev_addr, ha, dev->addr_len))
871 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
873 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
875 struct net_device *dev;
878 for_each_netdev(net, dev)
879 if (dev->type == type)
884 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
886 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
888 struct net_device *dev, *ret = NULL;
891 for_each_netdev_rcu(net, dev)
892 if (dev->type == type) {
900 EXPORT_SYMBOL(dev_getfirstbyhwtype);
903 * __dev_get_by_flags - find any device with given flags
904 * @net: the applicable net namespace
905 * @if_flags: IFF_* values
906 * @mask: bitmask of bits in if_flags to check
908 * Search for any interface with the given flags. Returns NULL if a device
909 * is not found or a pointer to the device. Must be called inside
910 * rtnl_lock(), and result refcount is unchanged.
913 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
916 struct net_device *dev, *ret;
921 for_each_netdev(net, dev) {
922 if (((dev->flags ^ if_flags) & mask) == 0) {
929 EXPORT_SYMBOL(__dev_get_by_flags);
932 * dev_valid_name - check if name is okay for network device
935 * Network device names need to be valid file names to
936 * to allow sysfs to work. We also disallow any kind of
939 bool dev_valid_name(const char *name)
943 if (strlen(name) >= IFNAMSIZ)
945 if (!strcmp(name, ".") || !strcmp(name, ".."))
949 if (*name == '/' || isspace(*name))
955 EXPORT_SYMBOL(dev_valid_name);
958 * __dev_alloc_name - allocate a name for a device
959 * @net: network namespace to allocate the device name in
960 * @name: name format string
961 * @buf: scratch buffer and result name string
963 * Passed a format string - eg "lt%d" it will try and find a suitable
964 * id. It scans list of devices to build up a free map, then chooses
965 * the first empty slot. The caller must hold the dev_base or rtnl lock
966 * while allocating the name and adding the device in order to avoid
968 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
969 * Returns the number of the unit assigned or a negative errno code.
972 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
976 const int max_netdevices = 8*PAGE_SIZE;
977 unsigned long *inuse;
978 struct net_device *d;
980 p = strnchr(name, IFNAMSIZ-1, '%');
983 * Verify the string as this thing may have come from
984 * the user. There must be either one "%d" and no other "%"
987 if (p[1] != 'd' || strchr(p + 2, '%'))
990 /* Use one page as a bit array of possible slots */
991 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
995 for_each_netdev(net, d) {
996 if (!sscanf(d->name, name, &i))
998 if (i < 0 || i >= max_netdevices)
1001 /* avoid cases where sscanf is not exact inverse of printf */
1002 snprintf(buf, IFNAMSIZ, name, i);
1003 if (!strncmp(buf, d->name, IFNAMSIZ))
1007 i = find_first_zero_bit(inuse, max_netdevices);
1008 free_page((unsigned long) inuse);
1012 snprintf(buf, IFNAMSIZ, name, i);
1013 if (!__dev_get_by_name(net, buf))
1016 /* It is possible to run out of possible slots
1017 * when the name is long and there isn't enough space left
1018 * for the digits, or if all bits are used.
1024 * dev_alloc_name - allocate a name for a device
1026 * @name: name format string
1028 * Passed a format string - eg "lt%d" it will try and find a suitable
1029 * id. It scans list of devices to build up a free map, then chooses
1030 * the first empty slot. The caller must hold the dev_base or rtnl lock
1031 * while allocating the name and adding the device in order to avoid
1033 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1034 * Returns the number of the unit assigned or a negative errno code.
1037 int dev_alloc_name(struct net_device *dev, const char *name)
1043 BUG_ON(!dev_net(dev));
1045 ret = __dev_alloc_name(net, name, buf);
1047 strlcpy(dev->name, buf, IFNAMSIZ);
1050 EXPORT_SYMBOL(dev_alloc_name);
1052 static int dev_alloc_name_ns(struct net *net,
1053 struct net_device *dev,
1059 ret = __dev_alloc_name(net, name, buf);
1061 strlcpy(dev->name, buf, IFNAMSIZ);
1065 static int dev_get_valid_name(struct net *net,
1066 struct net_device *dev,
1071 if (!dev_valid_name(name))
1074 if (strchr(name, '%'))
1075 return dev_alloc_name_ns(net, dev, name);
1076 else if (__dev_get_by_name(net, name))
1078 else if (dev->name != name)
1079 strlcpy(dev->name, name, IFNAMSIZ);
1085 * dev_change_name - change name of a device
1087 * @newname: name (or format string) must be at least IFNAMSIZ
1089 * Change name of a device, can pass format strings "eth%d".
1092 int dev_change_name(struct net_device *dev, const char *newname)
1094 unsigned char old_assign_type;
1095 char oldname[IFNAMSIZ];
1101 BUG_ON(!dev_net(dev));
1104 if (dev->flags & IFF_UP)
1107 write_seqcount_begin(&devnet_rename_seq);
1109 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1110 write_seqcount_end(&devnet_rename_seq);
1114 memcpy(oldname, dev->name, IFNAMSIZ);
1116 err = dev_get_valid_name(net, dev, newname);
1118 write_seqcount_end(&devnet_rename_seq);
1122 if (oldname[0] && !strchr(oldname, '%'))
1123 netdev_info(dev, "renamed from %s\n", oldname);
1125 old_assign_type = dev->name_assign_type;
1126 dev->name_assign_type = NET_NAME_RENAMED;
1129 ret = device_rename(&dev->dev, dev->name);
1131 memcpy(dev->name, oldname, IFNAMSIZ);
1132 dev->name_assign_type = old_assign_type;
1133 write_seqcount_end(&devnet_rename_seq);
1137 write_seqcount_end(&devnet_rename_seq);
1139 netdev_adjacent_rename_links(dev, oldname);
1141 write_lock_bh(&dev_base_lock);
1142 hlist_del_rcu(&dev->name_hlist);
1143 write_unlock_bh(&dev_base_lock);
1147 write_lock_bh(&dev_base_lock);
1148 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1149 write_unlock_bh(&dev_base_lock);
1151 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1152 ret = notifier_to_errno(ret);
1155 /* err >= 0 after dev_alloc_name() or stores the first errno */
1158 write_seqcount_begin(&devnet_rename_seq);
1159 memcpy(dev->name, oldname, IFNAMSIZ);
1160 memcpy(oldname, newname, IFNAMSIZ);
1161 dev->name_assign_type = old_assign_type;
1162 old_assign_type = NET_NAME_RENAMED;
1165 pr_err("%s: name change rollback failed: %d\n",
1174 * dev_set_alias - change ifalias of a device
1176 * @alias: name up to IFALIASZ
1177 * @len: limit of bytes to copy from info
1179 * Set ifalias for a device,
1181 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1187 if (len >= IFALIASZ)
1191 kfree(dev->ifalias);
1192 dev->ifalias = NULL;
1196 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1199 dev->ifalias = new_ifalias;
1201 strlcpy(dev->ifalias, alias, len+1);
1207 * netdev_features_change - device changes features
1208 * @dev: device to cause notification
1210 * Called to indicate a device has changed features.
1212 void netdev_features_change(struct net_device *dev)
1214 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1216 EXPORT_SYMBOL(netdev_features_change);
1219 * netdev_state_change - device changes state
1220 * @dev: device to cause notification
1222 * Called to indicate a device has changed state. This function calls
1223 * the notifier chains for netdev_chain and sends a NEWLINK message
1224 * to the routing socket.
1226 void netdev_state_change(struct net_device *dev)
1228 if (dev->flags & IFF_UP) {
1229 struct netdev_notifier_change_info change_info;
1231 change_info.flags_changed = 0;
1232 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1234 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1237 EXPORT_SYMBOL(netdev_state_change);
1240 * netdev_notify_peers - notify network peers about existence of @dev
1241 * @dev: network device
1243 * Generate traffic such that interested network peers are aware of
1244 * @dev, such as by generating a gratuitous ARP. This may be used when
1245 * a device wants to inform the rest of the network about some sort of
1246 * reconfiguration such as a failover event or virtual machine
1249 void netdev_notify_peers(struct net_device *dev)
1252 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1255 EXPORT_SYMBOL(netdev_notify_peers);
1257 static int __dev_open(struct net_device *dev)
1259 const struct net_device_ops *ops = dev->netdev_ops;
1264 if (!netif_device_present(dev))
1267 /* Block netpoll from trying to do any rx path servicing.
1268 * If we don't do this there is a chance ndo_poll_controller
1269 * or ndo_poll may be running while we open the device
1271 netpoll_poll_disable(dev);
1273 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1274 ret = notifier_to_errno(ret);
1278 set_bit(__LINK_STATE_START, &dev->state);
1280 if (ops->ndo_validate_addr)
1281 ret = ops->ndo_validate_addr(dev);
1283 if (!ret && ops->ndo_open)
1284 ret = ops->ndo_open(dev);
1286 netpoll_poll_enable(dev);
1289 clear_bit(__LINK_STATE_START, &dev->state);
1291 dev->flags |= IFF_UP;
1292 dev_set_rx_mode(dev);
1294 add_device_randomness(dev->dev_addr, dev->addr_len);
1301 * dev_open - prepare an interface for use.
1302 * @dev: device to open
1304 * Takes a device from down to up state. The device's private open
1305 * function is invoked and then the multicast lists are loaded. Finally
1306 * the device is moved into the up state and a %NETDEV_UP message is
1307 * sent to the netdev notifier chain.
1309 * Calling this function on an active interface is a nop. On a failure
1310 * a negative errno code is returned.
1312 int dev_open(struct net_device *dev)
1316 if (dev->flags & IFF_UP)
1319 ret = __dev_open(dev);
1323 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1324 call_netdevice_notifiers(NETDEV_UP, dev);
1328 EXPORT_SYMBOL(dev_open);
1330 static int __dev_close_many(struct list_head *head)
1332 struct net_device *dev;
1337 list_for_each_entry(dev, head, close_list) {
1338 /* Temporarily disable netpoll until the interface is down */
1339 netpoll_poll_disable(dev);
1341 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1343 clear_bit(__LINK_STATE_START, &dev->state);
1345 /* Synchronize to scheduled poll. We cannot touch poll list, it
1346 * can be even on different cpu. So just clear netif_running().
1348 * dev->stop() will invoke napi_disable() on all of it's
1349 * napi_struct instances on this device.
1351 smp_mb__after_atomic(); /* Commit netif_running(). */
1354 dev_deactivate_many(head);
1356 list_for_each_entry(dev, head, close_list) {
1357 const struct net_device_ops *ops = dev->netdev_ops;
1360 * Call the device specific close. This cannot fail.
1361 * Only if device is UP
1363 * We allow it to be called even after a DETACH hot-plug
1369 dev->flags &= ~IFF_UP;
1370 netpoll_poll_enable(dev);
1376 static int __dev_close(struct net_device *dev)
1381 list_add(&dev->close_list, &single);
1382 retval = __dev_close_many(&single);
1388 static int dev_close_many(struct list_head *head)
1390 struct net_device *dev, *tmp;
1392 /* Remove the devices that don't need to be closed */
1393 list_for_each_entry_safe(dev, tmp, head, close_list)
1394 if (!(dev->flags & IFF_UP))
1395 list_del_init(&dev->close_list);
1397 __dev_close_many(head);
1399 list_for_each_entry_safe(dev, tmp, head, close_list) {
1400 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1401 call_netdevice_notifiers(NETDEV_DOWN, dev);
1402 list_del_init(&dev->close_list);
1409 * dev_close - shutdown an interface.
1410 * @dev: device to shutdown
1412 * This function moves an active device into down state. A
1413 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1414 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1417 int dev_close(struct net_device *dev)
1419 if (dev->flags & IFF_UP) {
1422 list_add(&dev->close_list, &single);
1423 dev_close_many(&single);
1428 EXPORT_SYMBOL(dev_close);
1432 * dev_disable_lro - disable Large Receive Offload on a device
1435 * Disable Large Receive Offload (LRO) on a net device. Must be
1436 * called under RTNL. This is needed if received packets may be
1437 * forwarded to another interface.
1439 void dev_disable_lro(struct net_device *dev)
1441 struct net_device *lower_dev;
1442 struct list_head *iter;
1444 dev->wanted_features &= ~NETIF_F_LRO;
1445 netdev_update_features(dev);
1447 if (unlikely(dev->features & NETIF_F_LRO))
1448 netdev_WARN(dev, "failed to disable LRO!\n");
1450 netdev_for_each_lower_dev(dev, lower_dev, iter)
1451 dev_disable_lro(lower_dev);
1453 EXPORT_SYMBOL(dev_disable_lro);
1455 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1456 struct net_device *dev)
1458 struct netdev_notifier_info info;
1460 netdev_notifier_info_init(&info, dev);
1461 return nb->notifier_call(nb, val, &info);
1464 static int dev_boot_phase = 1;
1467 * register_netdevice_notifier - register a network notifier block
1470 * Register a notifier to be called when network device events occur.
1471 * The notifier passed is linked into the kernel structures and must
1472 * not be reused until it has been unregistered. A negative errno code
1473 * is returned on a failure.
1475 * When registered all registration and up events are replayed
1476 * to the new notifier to allow device to have a race free
1477 * view of the network device list.
1480 int register_netdevice_notifier(struct notifier_block *nb)
1482 struct net_device *dev;
1483 struct net_device *last;
1488 err = raw_notifier_chain_register(&netdev_chain, nb);
1494 for_each_netdev(net, dev) {
1495 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1496 err = notifier_to_errno(err);
1500 if (!(dev->flags & IFF_UP))
1503 call_netdevice_notifier(nb, NETDEV_UP, dev);
1514 for_each_netdev(net, dev) {
1518 if (dev->flags & IFF_UP) {
1519 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1521 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1523 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1528 raw_notifier_chain_unregister(&netdev_chain, nb);
1531 EXPORT_SYMBOL(register_netdevice_notifier);
1534 * unregister_netdevice_notifier - unregister a network notifier block
1537 * Unregister a notifier previously registered by
1538 * register_netdevice_notifier(). The notifier is unlinked into the
1539 * kernel structures and may then be reused. A negative errno code
1540 * is returned on a failure.
1542 * After unregistering unregister and down device events are synthesized
1543 * for all devices on the device list to the removed notifier to remove
1544 * the need for special case cleanup code.
1547 int unregister_netdevice_notifier(struct notifier_block *nb)
1549 struct net_device *dev;
1554 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1559 for_each_netdev(net, dev) {
1560 if (dev->flags & IFF_UP) {
1561 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1563 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1565 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1572 EXPORT_SYMBOL(unregister_netdevice_notifier);
1575 * call_netdevice_notifiers_info - call all network notifier blocks
1576 * @val: value passed unmodified to notifier function
1577 * @dev: net_device pointer passed unmodified to notifier function
1578 * @info: notifier information data
1580 * Call all network notifier blocks. Parameters and return value
1581 * are as for raw_notifier_call_chain().
1584 static int call_netdevice_notifiers_info(unsigned long val,
1585 struct net_device *dev,
1586 struct netdev_notifier_info *info)
1589 netdev_notifier_info_init(info, dev);
1590 return raw_notifier_call_chain(&netdev_chain, val, info);
1594 * call_netdevice_notifiers - call all network notifier blocks
1595 * @val: value passed unmodified to notifier function
1596 * @dev: net_device pointer passed unmodified to notifier function
1598 * Call all network notifier blocks. Parameters and return value
1599 * are as for raw_notifier_call_chain().
1602 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1604 struct netdev_notifier_info info;
1606 return call_netdevice_notifiers_info(val, dev, &info);
1608 EXPORT_SYMBOL(call_netdevice_notifiers);
1610 static struct static_key netstamp_needed __read_mostly;
1611 #ifdef HAVE_JUMP_LABEL
1612 /* We are not allowed to call static_key_slow_dec() from irq context
1613 * If net_disable_timestamp() is called from irq context, defer the
1614 * static_key_slow_dec() calls.
1616 static atomic_t netstamp_needed_deferred;
1619 void net_enable_timestamp(void)
1621 #ifdef HAVE_JUMP_LABEL
1622 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1626 static_key_slow_dec(&netstamp_needed);
1630 static_key_slow_inc(&netstamp_needed);
1632 EXPORT_SYMBOL(net_enable_timestamp);
1634 void net_disable_timestamp(void)
1636 #ifdef HAVE_JUMP_LABEL
1637 if (in_interrupt()) {
1638 atomic_inc(&netstamp_needed_deferred);
1642 static_key_slow_dec(&netstamp_needed);
1644 EXPORT_SYMBOL(net_disable_timestamp);
1646 static inline void net_timestamp_set(struct sk_buff *skb)
1648 skb->tstamp.tv64 = 0;
1649 if (static_key_false(&netstamp_needed))
1650 __net_timestamp(skb);
1653 #define net_timestamp_check(COND, SKB) \
1654 if (static_key_false(&netstamp_needed)) { \
1655 if ((COND) && !(SKB)->tstamp.tv64) \
1656 __net_timestamp(SKB); \
1659 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1663 if (!(dev->flags & IFF_UP))
1666 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1667 if (skb->len <= len)
1670 /* if TSO is enabled, we don't care about the length as the packet
1671 * could be forwarded without being segmented before
1673 if (skb_is_gso(skb))
1678 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1680 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1682 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1683 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1684 atomic_long_inc(&dev->rx_dropped);
1690 if (unlikely(!is_skb_forwardable(dev, skb))) {
1691 atomic_long_inc(&dev->rx_dropped);
1696 skb_scrub_packet(skb, true);
1697 skb->protocol = eth_type_trans(skb, dev);
1698 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1702 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1705 * dev_forward_skb - loopback an skb to another netif
1707 * @dev: destination network device
1708 * @skb: buffer to forward
1711 * NET_RX_SUCCESS (no congestion)
1712 * NET_RX_DROP (packet was dropped, but freed)
1714 * dev_forward_skb can be used for injecting an skb from the
1715 * start_xmit function of one device into the receive queue
1716 * of another device.
1718 * The receiving device may be in another namespace, so
1719 * we have to clear all information in the skb that could
1720 * impact namespace isolation.
1722 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1724 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1726 EXPORT_SYMBOL_GPL(dev_forward_skb);
1728 static inline int deliver_skb(struct sk_buff *skb,
1729 struct packet_type *pt_prev,
1730 struct net_device *orig_dev)
1732 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1734 atomic_inc(&skb->users);
1735 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1738 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1739 struct packet_type **pt,
1740 struct net_device *dev, __be16 type,
1741 struct list_head *ptype_list)
1743 struct packet_type *ptype, *pt_prev = *pt;
1745 list_for_each_entry_rcu(ptype, ptype_list, list) {
1746 if (ptype->type != type)
1749 deliver_skb(skb, pt_prev, dev);
1755 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1757 if (!ptype->af_packet_priv || !skb->sk)
1760 if (ptype->id_match)
1761 return ptype->id_match(ptype, skb->sk);
1762 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1769 * Support routine. Sends outgoing frames to any network
1770 * taps currently in use.
1773 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1775 struct packet_type *ptype;
1776 struct sk_buff *skb2 = NULL;
1777 struct packet_type *pt_prev = NULL;
1778 struct list_head *ptype_list = &ptype_all;
1782 list_for_each_entry_rcu(ptype, ptype_list, list) {
1783 /* Never send packets back to the socket
1784 * they originated from - MvS (miquels@drinkel.ow.org)
1786 if (skb_loop_sk(ptype, skb))
1790 deliver_skb(skb2, pt_prev, skb->dev);
1795 /* need to clone skb, done only once */
1796 skb2 = skb_clone(skb, GFP_ATOMIC);
1800 net_timestamp_set(skb2);
1802 /* skb->nh should be correctly
1803 * set by sender, so that the second statement is
1804 * just protection against buggy protocols.
1806 skb_reset_mac_header(skb2);
1808 if (skb_network_header(skb2) < skb2->data ||
1809 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1810 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1811 ntohs(skb2->protocol),
1813 skb_reset_network_header(skb2);
1816 skb2->transport_header = skb2->network_header;
1817 skb2->pkt_type = PACKET_OUTGOING;
1821 if (ptype_list == &ptype_all) {
1822 ptype_list = &dev->ptype_all;
1827 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1832 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1833 * @dev: Network device
1834 * @txq: number of queues available
1836 * If real_num_tx_queues is changed the tc mappings may no longer be
1837 * valid. To resolve this verify the tc mapping remains valid and if
1838 * not NULL the mapping. With no priorities mapping to this
1839 * offset/count pair it will no longer be used. In the worst case TC0
1840 * is invalid nothing can be done so disable priority mappings. If is
1841 * expected that drivers will fix this mapping if they can before
1842 * calling netif_set_real_num_tx_queues.
1844 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1847 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1849 /* If TC0 is invalidated disable TC mapping */
1850 if (tc->offset + tc->count > txq) {
1851 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1856 /* Invalidated prio to tc mappings set to TC0 */
1857 for (i = 1; i < TC_BITMASK + 1; i++) {
1858 int q = netdev_get_prio_tc_map(dev, i);
1860 tc = &dev->tc_to_txq[q];
1861 if (tc->offset + tc->count > txq) {
1862 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1864 netdev_set_prio_tc_map(dev, i, 0);
1870 static DEFINE_MUTEX(xps_map_mutex);
1871 #define xmap_dereference(P) \
1872 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1874 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1877 struct xps_map *map = NULL;
1881 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1883 for (pos = 0; map && pos < map->len; pos++) {
1884 if (map->queues[pos] == index) {
1886 map->queues[pos] = map->queues[--map->len];
1888 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1889 kfree_rcu(map, rcu);
1899 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1901 struct xps_dev_maps *dev_maps;
1903 bool active = false;
1905 mutex_lock(&xps_map_mutex);
1906 dev_maps = xmap_dereference(dev->xps_maps);
1911 for_each_possible_cpu(cpu) {
1912 for (i = index; i < dev->num_tx_queues; i++) {
1913 if (!remove_xps_queue(dev_maps, cpu, i))
1916 if (i == dev->num_tx_queues)
1921 RCU_INIT_POINTER(dev->xps_maps, NULL);
1922 kfree_rcu(dev_maps, rcu);
1925 for (i = index; i < dev->num_tx_queues; i++)
1926 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1930 mutex_unlock(&xps_map_mutex);
1933 static struct xps_map *expand_xps_map(struct xps_map *map,
1936 struct xps_map *new_map;
1937 int alloc_len = XPS_MIN_MAP_ALLOC;
1940 for (pos = 0; map && pos < map->len; pos++) {
1941 if (map->queues[pos] != index)
1946 /* Need to add queue to this CPU's existing map */
1948 if (pos < map->alloc_len)
1951 alloc_len = map->alloc_len * 2;
1954 /* Need to allocate new map to store queue on this CPU's map */
1955 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1960 for (i = 0; i < pos; i++)
1961 new_map->queues[i] = map->queues[i];
1962 new_map->alloc_len = alloc_len;
1968 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1971 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1972 struct xps_map *map, *new_map;
1973 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1974 int cpu, numa_node_id = -2;
1975 bool active = false;
1977 mutex_lock(&xps_map_mutex);
1979 dev_maps = xmap_dereference(dev->xps_maps);
1981 /* allocate memory for queue storage */
1982 for_each_online_cpu(cpu) {
1983 if (!cpumask_test_cpu(cpu, mask))
1987 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1988 if (!new_dev_maps) {
1989 mutex_unlock(&xps_map_mutex);
1993 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1996 map = expand_xps_map(map, cpu, index);
2000 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2004 goto out_no_new_maps;
2006 for_each_possible_cpu(cpu) {
2007 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2008 /* add queue to CPU maps */
2011 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2012 while ((pos < map->len) && (map->queues[pos] != index))
2015 if (pos == map->len)
2016 map->queues[map->len++] = index;
2018 if (numa_node_id == -2)
2019 numa_node_id = cpu_to_node(cpu);
2020 else if (numa_node_id != cpu_to_node(cpu))
2023 } else if (dev_maps) {
2024 /* fill in the new device map from the old device map */
2025 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2026 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2031 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2033 /* Cleanup old maps */
2035 for_each_possible_cpu(cpu) {
2036 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2037 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2038 if (map && map != new_map)
2039 kfree_rcu(map, rcu);
2042 kfree_rcu(dev_maps, rcu);
2045 dev_maps = new_dev_maps;
2049 /* update Tx queue numa node */
2050 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2051 (numa_node_id >= 0) ? numa_node_id :
2057 /* removes queue from unused CPUs */
2058 for_each_possible_cpu(cpu) {
2059 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2062 if (remove_xps_queue(dev_maps, cpu, index))
2066 /* free map if not active */
2068 RCU_INIT_POINTER(dev->xps_maps, NULL);
2069 kfree_rcu(dev_maps, rcu);
2073 mutex_unlock(&xps_map_mutex);
2077 /* remove any maps that we added */
2078 for_each_possible_cpu(cpu) {
2079 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2080 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2082 if (new_map && new_map != map)
2086 mutex_unlock(&xps_map_mutex);
2088 kfree(new_dev_maps);
2091 EXPORT_SYMBOL(netif_set_xps_queue);
2095 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2096 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2098 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2102 if (txq < 1 || txq > dev->num_tx_queues)
2105 if (dev->reg_state == NETREG_REGISTERED ||
2106 dev->reg_state == NETREG_UNREGISTERING) {
2109 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2115 netif_setup_tc(dev, txq);
2117 if (txq < dev->real_num_tx_queues) {
2118 qdisc_reset_all_tx_gt(dev, txq);
2120 netif_reset_xps_queues_gt(dev, txq);
2125 dev->real_num_tx_queues = txq;
2128 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2132 * netif_set_real_num_rx_queues - set actual number of RX queues used
2133 * @dev: Network device
2134 * @rxq: Actual number of RX queues
2136 * This must be called either with the rtnl_lock held or before
2137 * registration of the net device. Returns 0 on success, or a
2138 * negative error code. If called before registration, it always
2141 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2145 if (rxq < 1 || rxq > dev->num_rx_queues)
2148 if (dev->reg_state == NETREG_REGISTERED) {
2151 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2157 dev->real_num_rx_queues = rxq;
2160 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2164 * netif_get_num_default_rss_queues - default number of RSS queues
2166 * This routine should set an upper limit on the number of RSS queues
2167 * used by default by multiqueue devices.
2169 int netif_get_num_default_rss_queues(void)
2171 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2173 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2175 static inline void __netif_reschedule(struct Qdisc *q)
2177 struct softnet_data *sd;
2178 unsigned long flags;
2180 local_irq_save(flags);
2181 sd = this_cpu_ptr(&softnet_data);
2182 q->next_sched = NULL;
2183 *sd->output_queue_tailp = q;
2184 sd->output_queue_tailp = &q->next_sched;
2185 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2186 local_irq_restore(flags);
2189 void __netif_schedule(struct Qdisc *q)
2191 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2192 __netif_reschedule(q);
2194 EXPORT_SYMBOL(__netif_schedule);
2196 struct dev_kfree_skb_cb {
2197 enum skb_free_reason reason;
2200 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2202 return (struct dev_kfree_skb_cb *)skb->cb;
2205 void netif_schedule_queue(struct netdev_queue *txq)
2208 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2209 struct Qdisc *q = rcu_dereference(txq->qdisc);
2211 __netif_schedule(q);
2215 EXPORT_SYMBOL(netif_schedule_queue);
2218 * netif_wake_subqueue - allow sending packets on subqueue
2219 * @dev: network device
2220 * @queue_index: sub queue index
2222 * Resume individual transmit queue of a device with multiple transmit queues.
2224 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2226 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2228 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2232 q = rcu_dereference(txq->qdisc);
2233 __netif_schedule(q);
2237 EXPORT_SYMBOL(netif_wake_subqueue);
2239 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2241 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2245 q = rcu_dereference(dev_queue->qdisc);
2246 __netif_schedule(q);
2250 EXPORT_SYMBOL(netif_tx_wake_queue);
2252 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2254 unsigned long flags;
2256 if (likely(atomic_read(&skb->users) == 1)) {
2258 atomic_set(&skb->users, 0);
2259 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2262 get_kfree_skb_cb(skb)->reason = reason;
2263 local_irq_save(flags);
2264 skb->next = __this_cpu_read(softnet_data.completion_queue);
2265 __this_cpu_write(softnet_data.completion_queue, skb);
2266 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2267 local_irq_restore(flags);
2269 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2271 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2273 if (in_irq() || irqs_disabled())
2274 __dev_kfree_skb_irq(skb, reason);
2278 EXPORT_SYMBOL(__dev_kfree_skb_any);
2282 * netif_device_detach - mark device as removed
2283 * @dev: network device
2285 * Mark device as removed from system and therefore no longer available.
2287 void netif_device_detach(struct net_device *dev)
2289 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2290 netif_running(dev)) {
2291 netif_tx_stop_all_queues(dev);
2294 EXPORT_SYMBOL(netif_device_detach);
2297 * netif_device_attach - mark device as attached
2298 * @dev: network device
2300 * Mark device as attached from system and restart if needed.
2302 void netif_device_attach(struct net_device *dev)
2304 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2305 netif_running(dev)) {
2306 netif_tx_wake_all_queues(dev);
2307 __netdev_watchdog_up(dev);
2310 EXPORT_SYMBOL(netif_device_attach);
2312 static void skb_warn_bad_offload(const struct sk_buff *skb)
2314 static const netdev_features_t null_features = 0;
2315 struct net_device *dev = skb->dev;
2316 const char *driver = "";
2318 if (!net_ratelimit())
2321 if (dev && dev->dev.parent)
2322 driver = dev_driver_string(dev->dev.parent);
2324 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2325 "gso_type=%d ip_summed=%d\n",
2326 driver, dev ? &dev->features : &null_features,
2327 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2328 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2329 skb_shinfo(skb)->gso_type, skb->ip_summed);
2333 * Invalidate hardware checksum when packet is to be mangled, and
2334 * complete checksum manually on outgoing path.
2336 int skb_checksum_help(struct sk_buff *skb)
2339 int ret = 0, offset;
2341 if (skb->ip_summed == CHECKSUM_COMPLETE)
2342 goto out_set_summed;
2344 if (unlikely(skb_shinfo(skb)->gso_size)) {
2345 skb_warn_bad_offload(skb);
2349 /* Before computing a checksum, we should make sure no frag could
2350 * be modified by an external entity : checksum could be wrong.
2352 if (skb_has_shared_frag(skb)) {
2353 ret = __skb_linearize(skb);
2358 offset = skb_checksum_start_offset(skb);
2359 BUG_ON(offset >= skb_headlen(skb));
2360 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2362 offset += skb->csum_offset;
2363 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2365 if (skb_cloned(skb) &&
2366 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2367 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2372 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2374 skb->ip_summed = CHECKSUM_NONE;
2378 EXPORT_SYMBOL(skb_checksum_help);
2380 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2382 __be16 type = skb->protocol;
2384 /* Tunnel gso handlers can set protocol to ethernet. */
2385 if (type == htons(ETH_P_TEB)) {
2388 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2391 eth = (struct ethhdr *)skb_mac_header(skb);
2392 type = eth->h_proto;
2395 return __vlan_get_protocol(skb, type, depth);
2399 * skb_mac_gso_segment - mac layer segmentation handler.
2400 * @skb: buffer to segment
2401 * @features: features for the output path (see dev->features)
2403 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2404 netdev_features_t features)
2406 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2407 struct packet_offload *ptype;
2408 int vlan_depth = skb->mac_len;
2409 __be16 type = skb_network_protocol(skb, &vlan_depth);
2411 if (unlikely(!type))
2412 return ERR_PTR(-EINVAL);
2414 __skb_pull(skb, vlan_depth);
2417 list_for_each_entry_rcu(ptype, &offload_base, list) {
2418 if (ptype->type == type && ptype->callbacks.gso_segment) {
2419 segs = ptype->callbacks.gso_segment(skb, features);
2425 __skb_push(skb, skb->data - skb_mac_header(skb));
2429 EXPORT_SYMBOL(skb_mac_gso_segment);
2432 /* openvswitch calls this on rx path, so we need a different check.
2434 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2437 return skb->ip_summed != CHECKSUM_PARTIAL;
2439 return skb->ip_summed == CHECKSUM_NONE;
2443 * __skb_gso_segment - Perform segmentation on skb.
2444 * @skb: buffer to segment
2445 * @features: features for the output path (see dev->features)
2446 * @tx_path: whether it is called in TX path
2448 * This function segments the given skb and returns a list of segments.
2450 * It may return NULL if the skb requires no segmentation. This is
2451 * only possible when GSO is used for verifying header integrity.
2453 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2454 netdev_features_t features, bool tx_path)
2456 if (unlikely(skb_needs_check(skb, tx_path))) {
2459 skb_warn_bad_offload(skb);
2461 err = skb_cow_head(skb, 0);
2463 return ERR_PTR(err);
2466 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2467 SKB_GSO_CB(skb)->encap_level = 0;
2469 skb_reset_mac_header(skb);
2470 skb_reset_mac_len(skb);
2472 return skb_mac_gso_segment(skb, features);
2474 EXPORT_SYMBOL(__skb_gso_segment);
2476 /* Take action when hardware reception checksum errors are detected. */
2478 void netdev_rx_csum_fault(struct net_device *dev)
2480 if (net_ratelimit()) {
2481 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2485 EXPORT_SYMBOL(netdev_rx_csum_fault);
2488 /* Actually, we should eliminate this check as soon as we know, that:
2489 * 1. IOMMU is present and allows to map all the memory.
2490 * 2. No high memory really exists on this machine.
2493 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2495 #ifdef CONFIG_HIGHMEM
2497 if (!(dev->features & NETIF_F_HIGHDMA)) {
2498 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2499 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2500 if (PageHighMem(skb_frag_page(frag)))
2505 if (PCI_DMA_BUS_IS_PHYS) {
2506 struct device *pdev = dev->dev.parent;
2510 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2511 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2512 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2513 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2521 /* If MPLS offload request, verify we are testing hardware MPLS features
2522 * instead of standard features for the netdev.
2524 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2525 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2526 netdev_features_t features,
2529 if (eth_p_mpls(type))
2530 features &= skb->dev->mpls_features;
2535 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2536 netdev_features_t features,
2543 static netdev_features_t harmonize_features(struct sk_buff *skb,
2544 netdev_features_t features)
2549 type = skb_network_protocol(skb, &tmp);
2550 features = net_mpls_features(skb, features, type);
2552 if (skb->ip_summed != CHECKSUM_NONE &&
2553 !can_checksum_protocol(features, type)) {
2554 features &= ~NETIF_F_ALL_CSUM;
2555 } else if (illegal_highdma(skb->dev, skb)) {
2556 features &= ~NETIF_F_SG;
2562 netdev_features_t netif_skb_features(struct sk_buff *skb)
2564 struct net_device *dev = skb->dev;
2565 netdev_features_t features = dev->features;
2566 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2567 __be16 protocol = skb->protocol;
2569 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2570 features &= ~NETIF_F_GSO_MASK;
2572 /* If encapsulation offload request, verify we are testing
2573 * hardware encapsulation features instead of standard
2574 * features for the netdev
2576 if (skb->encapsulation)
2577 features &= dev->hw_enc_features;
2579 if (!skb_vlan_tag_present(skb)) {
2580 if (unlikely(protocol == htons(ETH_P_8021Q) ||
2581 protocol == htons(ETH_P_8021AD))) {
2582 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2583 protocol = veh->h_vlan_encapsulated_proto;
2589 features = netdev_intersect_features(features,
2590 dev->vlan_features |
2591 NETIF_F_HW_VLAN_CTAG_TX |
2592 NETIF_F_HW_VLAN_STAG_TX);
2594 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2595 features = netdev_intersect_features(features,
2600 NETIF_F_HW_VLAN_CTAG_TX |
2601 NETIF_F_HW_VLAN_STAG_TX);
2604 if (dev->netdev_ops->ndo_features_check)
2605 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2608 return harmonize_features(skb, features);
2610 EXPORT_SYMBOL(netif_skb_features);
2612 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2613 struct netdev_queue *txq, bool more)
2618 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2619 dev_queue_xmit_nit(skb, dev);
2622 trace_net_dev_start_xmit(skb, dev);
2623 rc = netdev_start_xmit(skb, dev, txq, more);
2624 trace_net_dev_xmit(skb, rc, dev, len);
2629 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2630 struct netdev_queue *txq, int *ret)
2632 struct sk_buff *skb = first;
2633 int rc = NETDEV_TX_OK;
2636 struct sk_buff *next = skb->next;
2639 rc = xmit_one(skb, dev, txq, next != NULL);
2640 if (unlikely(!dev_xmit_complete(rc))) {
2646 if (netif_xmit_stopped(txq) && skb) {
2647 rc = NETDEV_TX_BUSY;
2657 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2658 netdev_features_t features)
2660 if (skb_vlan_tag_present(skb) &&
2661 !vlan_hw_offload_capable(features, skb->vlan_proto))
2662 skb = __vlan_hwaccel_push_inside(skb);
2666 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2668 netdev_features_t features;
2673 features = netif_skb_features(skb);
2674 skb = validate_xmit_vlan(skb, features);
2678 if (netif_needs_gso(dev, skb, features)) {
2679 struct sk_buff *segs;
2681 segs = skb_gso_segment(skb, features);
2689 if (skb_needs_linearize(skb, features) &&
2690 __skb_linearize(skb))
2693 /* If packet is not checksummed and device does not
2694 * support checksumming for this protocol, complete
2695 * checksumming here.
2697 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2698 if (skb->encapsulation)
2699 skb_set_inner_transport_header(skb,
2700 skb_checksum_start_offset(skb));
2702 skb_set_transport_header(skb,
2703 skb_checksum_start_offset(skb));
2704 if (!(features & NETIF_F_ALL_CSUM) &&
2705 skb_checksum_help(skb))
2718 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2720 struct sk_buff *next, *head = NULL, *tail;
2722 for (; skb != NULL; skb = next) {
2726 /* in case skb wont be segmented, point to itself */
2729 skb = validate_xmit_skb(skb, dev);
2737 /* If skb was segmented, skb->prev points to
2738 * the last segment. If not, it still contains skb.
2745 static void qdisc_pkt_len_init(struct sk_buff *skb)
2747 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2749 qdisc_skb_cb(skb)->pkt_len = skb->len;
2751 /* To get more precise estimation of bytes sent on wire,
2752 * we add to pkt_len the headers size of all segments
2754 if (shinfo->gso_size) {
2755 unsigned int hdr_len;
2756 u16 gso_segs = shinfo->gso_segs;
2758 /* mac layer + network layer */
2759 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2761 /* + transport layer */
2762 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2763 hdr_len += tcp_hdrlen(skb);
2765 hdr_len += sizeof(struct udphdr);
2767 if (shinfo->gso_type & SKB_GSO_DODGY)
2768 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2771 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2775 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2776 struct net_device *dev,
2777 struct netdev_queue *txq)
2779 spinlock_t *root_lock = qdisc_lock(q);
2783 qdisc_pkt_len_init(skb);
2784 qdisc_calculate_pkt_len(skb, q);
2786 * Heuristic to force contended enqueues to serialize on a
2787 * separate lock before trying to get qdisc main lock.
2788 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2789 * often and dequeue packets faster.
2791 contended = qdisc_is_running(q);
2792 if (unlikely(contended))
2793 spin_lock(&q->busylock);
2795 spin_lock(root_lock);
2796 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2799 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2800 qdisc_run_begin(q)) {
2802 * This is a work-conserving queue; there are no old skbs
2803 * waiting to be sent out; and the qdisc is not running -
2804 * xmit the skb directly.
2807 qdisc_bstats_update(q, skb);
2809 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2810 if (unlikely(contended)) {
2811 spin_unlock(&q->busylock);
2818 rc = NET_XMIT_SUCCESS;
2820 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2821 if (qdisc_run_begin(q)) {
2822 if (unlikely(contended)) {
2823 spin_unlock(&q->busylock);
2829 spin_unlock(root_lock);
2830 if (unlikely(contended))
2831 spin_unlock(&q->busylock);
2835 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2836 static void skb_update_prio(struct sk_buff *skb)
2838 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2840 if (!skb->priority && skb->sk && map) {
2841 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2843 if (prioidx < map->priomap_len)
2844 skb->priority = map->priomap[prioidx];
2848 #define skb_update_prio(skb)
2851 static DEFINE_PER_CPU(int, xmit_recursion);
2852 #define RECURSION_LIMIT 10
2855 * dev_loopback_xmit - loop back @skb
2856 * @skb: buffer to transmit
2858 int dev_loopback_xmit(struct sk_buff *skb)
2860 skb_reset_mac_header(skb);
2861 __skb_pull(skb, skb_network_offset(skb));
2862 skb->pkt_type = PACKET_LOOPBACK;
2863 skb->ip_summed = CHECKSUM_UNNECESSARY;
2864 WARN_ON(!skb_dst(skb));
2869 EXPORT_SYMBOL(dev_loopback_xmit);
2872 * __dev_queue_xmit - transmit a buffer
2873 * @skb: buffer to transmit
2874 * @accel_priv: private data used for L2 forwarding offload
2876 * Queue a buffer for transmission to a network device. The caller must
2877 * have set the device and priority and built the buffer before calling
2878 * this function. The function can be called from an interrupt.
2880 * A negative errno code is returned on a failure. A success does not
2881 * guarantee the frame will be transmitted as it may be dropped due
2882 * to congestion or traffic shaping.
2884 * -----------------------------------------------------------------------------------
2885 * I notice this method can also return errors from the queue disciplines,
2886 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2889 * Regardless of the return value, the skb is consumed, so it is currently
2890 * difficult to retry a send to this method. (You can bump the ref count
2891 * before sending to hold a reference for retry if you are careful.)
2893 * When calling this method, interrupts MUST be enabled. This is because
2894 * the BH enable code must have IRQs enabled so that it will not deadlock.
2897 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2899 struct net_device *dev = skb->dev;
2900 struct netdev_queue *txq;
2904 skb_reset_mac_header(skb);
2906 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
2907 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
2909 /* Disable soft irqs for various locks below. Also
2910 * stops preemption for RCU.
2914 skb_update_prio(skb);
2916 /* If device/qdisc don't need skb->dst, release it right now while
2917 * its hot in this cpu cache.
2919 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2924 txq = netdev_pick_tx(dev, skb, accel_priv);
2925 q = rcu_dereference_bh(txq->qdisc);
2927 #ifdef CONFIG_NET_CLS_ACT
2928 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2930 trace_net_dev_queue(skb);
2932 rc = __dev_xmit_skb(skb, q, dev, txq);
2936 /* The device has no queue. Common case for software devices:
2937 loopback, all the sorts of tunnels...
2939 Really, it is unlikely that netif_tx_lock protection is necessary
2940 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2942 However, it is possible, that they rely on protection
2945 Check this and shot the lock. It is not prone from deadlocks.
2946 Either shot noqueue qdisc, it is even simpler 8)
2948 if (dev->flags & IFF_UP) {
2949 int cpu = smp_processor_id(); /* ok because BHs are off */
2951 if (txq->xmit_lock_owner != cpu) {
2953 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2954 goto recursion_alert;
2956 skb = validate_xmit_skb(skb, dev);
2960 HARD_TX_LOCK(dev, txq, cpu);
2962 if (!netif_xmit_stopped(txq)) {
2963 __this_cpu_inc(xmit_recursion);
2964 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
2965 __this_cpu_dec(xmit_recursion);
2966 if (dev_xmit_complete(rc)) {
2967 HARD_TX_UNLOCK(dev, txq);
2971 HARD_TX_UNLOCK(dev, txq);
2972 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2975 /* Recursion is detected! It is possible,
2979 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2986 rcu_read_unlock_bh();
2988 atomic_long_inc(&dev->tx_dropped);
2989 kfree_skb_list(skb);
2992 rcu_read_unlock_bh();
2996 int dev_queue_xmit(struct sk_buff *skb)
2998 return __dev_queue_xmit(skb, NULL);
3000 EXPORT_SYMBOL(dev_queue_xmit);
3002 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3004 return __dev_queue_xmit(skb, accel_priv);
3006 EXPORT_SYMBOL(dev_queue_xmit_accel);
3009 /*=======================================================================
3011 =======================================================================*/
3013 int netdev_max_backlog __read_mostly = 1000;
3014 EXPORT_SYMBOL(netdev_max_backlog);
3016 int netdev_tstamp_prequeue __read_mostly = 1;
3017 int netdev_budget __read_mostly = 300;
3018 int weight_p __read_mostly = 64; /* old backlog weight */
3020 /* Called with irq disabled */
3021 static inline void ____napi_schedule(struct softnet_data *sd,
3022 struct napi_struct *napi)
3024 list_add_tail(&napi->poll_list, &sd->poll_list);
3025 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3030 /* One global table that all flow-based protocols share. */
3031 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3032 EXPORT_SYMBOL(rps_sock_flow_table);
3034 struct static_key rps_needed __read_mostly;
3036 static struct rps_dev_flow *
3037 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3038 struct rps_dev_flow *rflow, u16 next_cpu)
3040 if (next_cpu != RPS_NO_CPU) {
3041 #ifdef CONFIG_RFS_ACCEL
3042 struct netdev_rx_queue *rxqueue;
3043 struct rps_dev_flow_table *flow_table;
3044 struct rps_dev_flow *old_rflow;
3049 /* Should we steer this flow to a different hardware queue? */
3050 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3051 !(dev->features & NETIF_F_NTUPLE))
3053 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3054 if (rxq_index == skb_get_rx_queue(skb))
3057 rxqueue = dev->_rx + rxq_index;
3058 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3061 flow_id = skb_get_hash(skb) & flow_table->mask;
3062 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3063 rxq_index, flow_id);
3067 rflow = &flow_table->flows[flow_id];
3069 if (old_rflow->filter == rflow->filter)
3070 old_rflow->filter = RPS_NO_FILTER;
3074 per_cpu(softnet_data, next_cpu).input_queue_head;
3077 rflow->cpu = next_cpu;
3082 * get_rps_cpu is called from netif_receive_skb and returns the target
3083 * CPU from the RPS map of the receiving queue for a given skb.
3084 * rcu_read_lock must be held on entry.
3086 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3087 struct rps_dev_flow **rflowp)
3089 struct netdev_rx_queue *rxqueue;
3090 struct rps_map *map;
3091 struct rps_dev_flow_table *flow_table;
3092 struct rps_sock_flow_table *sock_flow_table;
3097 if (skb_rx_queue_recorded(skb)) {
3098 u16 index = skb_get_rx_queue(skb);
3099 if (unlikely(index >= dev->real_num_rx_queues)) {
3100 WARN_ONCE(dev->real_num_rx_queues > 1,
3101 "%s received packet on queue %u, but number "
3102 "of RX queues is %u\n",
3103 dev->name, index, dev->real_num_rx_queues);
3106 rxqueue = dev->_rx + index;
3110 map = rcu_dereference(rxqueue->rps_map);
3112 if (map->len == 1 &&
3113 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3114 tcpu = map->cpus[0];
3115 if (cpu_online(tcpu))
3119 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3123 skb_reset_network_header(skb);
3124 hash = skb_get_hash(skb);
3128 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3129 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3130 if (flow_table && sock_flow_table) {
3132 struct rps_dev_flow *rflow;
3134 rflow = &flow_table->flows[hash & flow_table->mask];
3137 next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask];
3140 * If the desired CPU (where last recvmsg was done) is
3141 * different from current CPU (one in the rx-queue flow
3142 * table entry), switch if one of the following holds:
3143 * - Current CPU is unset (equal to RPS_NO_CPU).
3144 * - Current CPU is offline.
3145 * - The current CPU's queue tail has advanced beyond the
3146 * last packet that was enqueued using this table entry.
3147 * This guarantees that all previous packets for the flow
3148 * have been dequeued, thus preserving in order delivery.
3150 if (unlikely(tcpu != next_cpu) &&
3151 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3152 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3153 rflow->last_qtail)) >= 0)) {
3155 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3158 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3166 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3167 if (cpu_online(tcpu)) {
3177 #ifdef CONFIG_RFS_ACCEL
3180 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3181 * @dev: Device on which the filter was set
3182 * @rxq_index: RX queue index
3183 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3184 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3186 * Drivers that implement ndo_rx_flow_steer() should periodically call
3187 * this function for each installed filter and remove the filters for
3188 * which it returns %true.
3190 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3191 u32 flow_id, u16 filter_id)
3193 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3194 struct rps_dev_flow_table *flow_table;
3195 struct rps_dev_flow *rflow;
3200 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3201 if (flow_table && flow_id <= flow_table->mask) {
3202 rflow = &flow_table->flows[flow_id];
3203 cpu = ACCESS_ONCE(rflow->cpu);
3204 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3205 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3206 rflow->last_qtail) <
3207 (int)(10 * flow_table->mask)))
3213 EXPORT_SYMBOL(rps_may_expire_flow);
3215 #endif /* CONFIG_RFS_ACCEL */
3217 /* Called from hardirq (IPI) context */
3218 static void rps_trigger_softirq(void *data)
3220 struct softnet_data *sd = data;
3222 ____napi_schedule(sd, &sd->backlog);
3226 #endif /* CONFIG_RPS */
3229 * Check if this softnet_data structure is another cpu one
3230 * If yes, queue it to our IPI list and return 1
3233 static int rps_ipi_queued(struct softnet_data *sd)
3236 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3239 sd->rps_ipi_next = mysd->rps_ipi_list;
3240 mysd->rps_ipi_list = sd;
3242 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3245 #endif /* CONFIG_RPS */
3249 #ifdef CONFIG_NET_FLOW_LIMIT
3250 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3253 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3255 #ifdef CONFIG_NET_FLOW_LIMIT
3256 struct sd_flow_limit *fl;
3257 struct softnet_data *sd;
3258 unsigned int old_flow, new_flow;
3260 if (qlen < (netdev_max_backlog >> 1))
3263 sd = this_cpu_ptr(&softnet_data);
3266 fl = rcu_dereference(sd->flow_limit);
3268 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3269 old_flow = fl->history[fl->history_head];
3270 fl->history[fl->history_head] = new_flow;
3273 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3275 if (likely(fl->buckets[old_flow]))
3276 fl->buckets[old_flow]--;
3278 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3290 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3291 * queue (may be a remote CPU queue).
3293 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3294 unsigned int *qtail)
3296 struct softnet_data *sd;
3297 unsigned long flags;
3300 sd = &per_cpu(softnet_data, cpu);
3302 local_irq_save(flags);
3305 qlen = skb_queue_len(&sd->input_pkt_queue);
3306 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3309 __skb_queue_tail(&sd->input_pkt_queue, skb);
3310 input_queue_tail_incr_save(sd, qtail);
3312 local_irq_restore(flags);
3313 return NET_RX_SUCCESS;
3316 /* Schedule NAPI for backlog device
3317 * We can use non atomic operation since we own the queue lock
3319 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3320 if (!rps_ipi_queued(sd))
3321 ____napi_schedule(sd, &sd->backlog);
3329 local_irq_restore(flags);
3331 atomic_long_inc(&skb->dev->rx_dropped);
3336 static int netif_rx_internal(struct sk_buff *skb)
3340 net_timestamp_check(netdev_tstamp_prequeue, skb);
3342 trace_netif_rx(skb);
3344 if (static_key_false(&rps_needed)) {
3345 struct rps_dev_flow voidflow, *rflow = &voidflow;
3351 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3353 cpu = smp_processor_id();
3355 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3363 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3370 * netif_rx - post buffer to the network code
3371 * @skb: buffer to post
3373 * This function receives a packet from a device driver and queues it for
3374 * the upper (protocol) levels to process. It always succeeds. The buffer
3375 * may be dropped during processing for congestion control or by the
3379 * NET_RX_SUCCESS (no congestion)
3380 * NET_RX_DROP (packet was dropped)
3384 int netif_rx(struct sk_buff *skb)
3386 trace_netif_rx_entry(skb);
3388 return netif_rx_internal(skb);
3390 EXPORT_SYMBOL(netif_rx);
3392 int netif_rx_ni(struct sk_buff *skb)
3396 trace_netif_rx_ni_entry(skb);
3399 err = netif_rx_internal(skb);
3400 if (local_softirq_pending())
3406 EXPORT_SYMBOL(netif_rx_ni);
3408 static void net_tx_action(struct softirq_action *h)
3410 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3412 if (sd->completion_queue) {
3413 struct sk_buff *clist;
3415 local_irq_disable();
3416 clist = sd->completion_queue;
3417 sd->completion_queue = NULL;
3421 struct sk_buff *skb = clist;
3422 clist = clist->next;
3424 WARN_ON(atomic_read(&skb->users));
3425 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3426 trace_consume_skb(skb);
3428 trace_kfree_skb(skb, net_tx_action);
3433 if (sd->output_queue) {
3436 local_irq_disable();
3437 head = sd->output_queue;
3438 sd->output_queue = NULL;
3439 sd->output_queue_tailp = &sd->output_queue;
3443 struct Qdisc *q = head;
3444 spinlock_t *root_lock;
3446 head = head->next_sched;
3448 root_lock = qdisc_lock(q);
3449 if (spin_trylock(root_lock)) {
3450 smp_mb__before_atomic();
3451 clear_bit(__QDISC_STATE_SCHED,
3454 spin_unlock(root_lock);
3456 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3458 __netif_reschedule(q);
3460 smp_mb__before_atomic();
3461 clear_bit(__QDISC_STATE_SCHED,
3469 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3470 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3471 /* This hook is defined here for ATM LANE */
3472 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3473 unsigned char *addr) __read_mostly;
3474 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3477 #ifdef CONFIG_NET_CLS_ACT
3478 /* TODO: Maybe we should just force sch_ingress to be compiled in
3479 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3480 * a compare and 2 stores extra right now if we dont have it on
3481 * but have CONFIG_NET_CLS_ACT
3482 * NOTE: This doesn't stop any functionality; if you dont have
3483 * the ingress scheduler, you just can't add policies on ingress.
3486 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3488 struct net_device *dev = skb->dev;
3489 u32 ttl = G_TC_RTTL(skb->tc_verd);
3490 int result = TC_ACT_OK;
3493 if (unlikely(MAX_RED_LOOP < ttl++)) {
3494 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3495 skb->skb_iif, dev->ifindex);
3499 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3500 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3502 q = rcu_dereference(rxq->qdisc);
3503 if (q != &noop_qdisc) {
3504 spin_lock(qdisc_lock(q));
3505 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3506 result = qdisc_enqueue_root(skb, q);
3507 spin_unlock(qdisc_lock(q));
3513 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3514 struct packet_type **pt_prev,
3515 int *ret, struct net_device *orig_dev)
3517 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3519 if (!rxq || rcu_access_pointer(rxq->qdisc) == &noop_qdisc)
3523 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3527 switch (ing_filter(skb, rxq)) {
3541 * netdev_rx_handler_register - register receive handler
3542 * @dev: device to register a handler for
3543 * @rx_handler: receive handler to register
3544 * @rx_handler_data: data pointer that is used by rx handler
3546 * Register a receive handler for a device. This handler will then be
3547 * called from __netif_receive_skb. A negative errno code is returned
3550 * The caller must hold the rtnl_mutex.
3552 * For a general description of rx_handler, see enum rx_handler_result.
3554 int netdev_rx_handler_register(struct net_device *dev,
3555 rx_handler_func_t *rx_handler,
3556 void *rx_handler_data)
3560 if (dev->rx_handler)
3563 /* Note: rx_handler_data must be set before rx_handler */
3564 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3565 rcu_assign_pointer(dev->rx_handler, rx_handler);
3569 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3572 * netdev_rx_handler_unregister - unregister receive handler
3573 * @dev: device to unregister a handler from
3575 * Unregister a receive handler from a device.
3577 * The caller must hold the rtnl_mutex.
3579 void netdev_rx_handler_unregister(struct net_device *dev)
3583 RCU_INIT_POINTER(dev->rx_handler, NULL);
3584 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3585 * section has a guarantee to see a non NULL rx_handler_data
3589 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3591 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3594 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3595 * the special handling of PFMEMALLOC skbs.
3597 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3599 switch (skb->protocol) {
3600 case htons(ETH_P_ARP):
3601 case htons(ETH_P_IP):
3602 case htons(ETH_P_IPV6):
3603 case htons(ETH_P_8021Q):
3604 case htons(ETH_P_8021AD):
3611 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3613 struct packet_type *ptype, *pt_prev;
3614 rx_handler_func_t *rx_handler;
3615 struct net_device *orig_dev;
3616 bool deliver_exact = false;
3617 int ret = NET_RX_DROP;
3620 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3622 trace_netif_receive_skb(skb);
3624 orig_dev = skb->dev;
3626 skb_reset_network_header(skb);
3627 if (!skb_transport_header_was_set(skb))
3628 skb_reset_transport_header(skb);
3629 skb_reset_mac_len(skb);
3636 skb->skb_iif = skb->dev->ifindex;
3638 __this_cpu_inc(softnet_data.processed);
3640 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3641 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3642 skb = skb_vlan_untag(skb);
3647 #ifdef CONFIG_NET_CLS_ACT
3648 if (skb->tc_verd & TC_NCLS) {
3649 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3657 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3659 ret = deliver_skb(skb, pt_prev, orig_dev);
3663 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3665 ret = deliver_skb(skb, pt_prev, orig_dev);
3670 #ifdef CONFIG_NET_CLS_ACT
3671 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3677 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3680 if (skb_vlan_tag_present(skb)) {
3682 ret = deliver_skb(skb, pt_prev, orig_dev);
3685 if (vlan_do_receive(&skb))
3687 else if (unlikely(!skb))
3691 rx_handler = rcu_dereference(skb->dev->rx_handler);
3694 ret = deliver_skb(skb, pt_prev, orig_dev);
3697 switch (rx_handler(&skb)) {
3698 case RX_HANDLER_CONSUMED:
3699 ret = NET_RX_SUCCESS;
3701 case RX_HANDLER_ANOTHER:
3703 case RX_HANDLER_EXACT:
3704 deliver_exact = true;
3705 case RX_HANDLER_PASS:
3712 if (unlikely(skb_vlan_tag_present(skb))) {
3713 if (skb_vlan_tag_get_id(skb))
3714 skb->pkt_type = PACKET_OTHERHOST;
3715 /* Note: we might in the future use prio bits
3716 * and set skb->priority like in vlan_do_receive()
3717 * For the time being, just ignore Priority Code Point
3722 type = skb->protocol;
3724 /* deliver only exact match when indicated */
3725 if (likely(!deliver_exact)) {
3726 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3727 &ptype_base[ntohs(type) &
3731 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3732 &orig_dev->ptype_specific);
3734 if (unlikely(skb->dev != orig_dev)) {
3735 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3736 &skb->dev->ptype_specific);
3740 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3743 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3746 atomic_long_inc(&skb->dev->rx_dropped);
3748 /* Jamal, now you will not able to escape explaining
3749 * me how you were going to use this. :-)
3759 static int __netif_receive_skb(struct sk_buff *skb)
3763 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3764 unsigned long pflags = current->flags;
3767 * PFMEMALLOC skbs are special, they should
3768 * - be delivered to SOCK_MEMALLOC sockets only
3769 * - stay away from userspace
3770 * - have bounded memory usage
3772 * Use PF_MEMALLOC as this saves us from propagating the allocation
3773 * context down to all allocation sites.
3775 current->flags |= PF_MEMALLOC;
3776 ret = __netif_receive_skb_core(skb, true);
3777 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3779 ret = __netif_receive_skb_core(skb, false);
3784 static int netif_receive_skb_internal(struct sk_buff *skb)
3786 net_timestamp_check(netdev_tstamp_prequeue, skb);
3788 if (skb_defer_rx_timestamp(skb))
3789 return NET_RX_SUCCESS;
3792 if (static_key_false(&rps_needed)) {
3793 struct rps_dev_flow voidflow, *rflow = &voidflow;
3798 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3801 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3808 return __netif_receive_skb(skb);
3812 * netif_receive_skb - process receive buffer from network
3813 * @skb: buffer to process
3815 * netif_receive_skb() is the main receive data processing function.
3816 * It always succeeds. The buffer may be dropped during processing
3817 * for congestion control or by the protocol layers.
3819 * This function may only be called from softirq context and interrupts
3820 * should be enabled.
3822 * Return values (usually ignored):
3823 * NET_RX_SUCCESS: no congestion
3824 * NET_RX_DROP: packet was dropped
3826 int netif_receive_skb(struct sk_buff *skb)
3828 trace_netif_receive_skb_entry(skb);
3830 return netif_receive_skb_internal(skb);
3832 EXPORT_SYMBOL(netif_receive_skb);
3834 /* Network device is going away, flush any packets still pending
3835 * Called with irqs disabled.
3837 static void flush_backlog(void *arg)
3839 struct net_device *dev = arg;
3840 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3841 struct sk_buff *skb, *tmp;
3844 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3845 if (skb->dev == dev) {
3846 __skb_unlink(skb, &sd->input_pkt_queue);
3848 input_queue_head_incr(sd);
3853 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3854 if (skb->dev == dev) {
3855 __skb_unlink(skb, &sd->process_queue);
3857 input_queue_head_incr(sd);
3862 static int napi_gro_complete(struct sk_buff *skb)
3864 struct packet_offload *ptype;
3865 __be16 type = skb->protocol;
3866 struct list_head *head = &offload_base;
3869 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3871 if (NAPI_GRO_CB(skb)->count == 1) {
3872 skb_shinfo(skb)->gso_size = 0;
3877 list_for_each_entry_rcu(ptype, head, list) {
3878 if (ptype->type != type || !ptype->callbacks.gro_complete)
3881 err = ptype->callbacks.gro_complete(skb, 0);
3887 WARN_ON(&ptype->list == head);
3889 return NET_RX_SUCCESS;
3893 return netif_receive_skb_internal(skb);
3896 /* napi->gro_list contains packets ordered by age.
3897 * youngest packets at the head of it.
3898 * Complete skbs in reverse order to reduce latencies.
3900 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3902 struct sk_buff *skb, *prev = NULL;
3904 /* scan list and build reverse chain */
3905 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3910 for (skb = prev; skb; skb = prev) {
3913 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3917 napi_gro_complete(skb);
3921 napi->gro_list = NULL;
3923 EXPORT_SYMBOL(napi_gro_flush);
3925 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3928 unsigned int maclen = skb->dev->hard_header_len;
3929 u32 hash = skb_get_hash_raw(skb);
3931 for (p = napi->gro_list; p; p = p->next) {
3932 unsigned long diffs;
3934 NAPI_GRO_CB(p)->flush = 0;
3936 if (hash != skb_get_hash_raw(p)) {
3937 NAPI_GRO_CB(p)->same_flow = 0;
3941 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3942 diffs |= p->vlan_tci ^ skb->vlan_tci;
3943 if (maclen == ETH_HLEN)
3944 diffs |= compare_ether_header(skb_mac_header(p),
3945 skb_mac_header(skb));
3947 diffs = memcmp(skb_mac_header(p),
3948 skb_mac_header(skb),
3950 NAPI_GRO_CB(p)->same_flow = !diffs;
3954 static void skb_gro_reset_offset(struct sk_buff *skb)
3956 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3957 const skb_frag_t *frag0 = &pinfo->frags[0];
3959 NAPI_GRO_CB(skb)->data_offset = 0;
3960 NAPI_GRO_CB(skb)->frag0 = NULL;
3961 NAPI_GRO_CB(skb)->frag0_len = 0;
3963 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3965 !PageHighMem(skb_frag_page(frag0))) {
3966 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3967 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3971 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
3973 struct skb_shared_info *pinfo = skb_shinfo(skb);
3975 BUG_ON(skb->end - skb->tail < grow);
3977 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3979 skb->data_len -= grow;
3982 pinfo->frags[0].page_offset += grow;
3983 skb_frag_size_sub(&pinfo->frags[0], grow);
3985 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
3986 skb_frag_unref(skb, 0);
3987 memmove(pinfo->frags, pinfo->frags + 1,
3988 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
3992 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3994 struct sk_buff **pp = NULL;
3995 struct packet_offload *ptype;
3996 __be16 type = skb->protocol;
3997 struct list_head *head = &offload_base;
3999 enum gro_result ret;
4002 if (!(skb->dev->features & NETIF_F_GRO))
4005 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4008 gro_list_prepare(napi, skb);
4011 list_for_each_entry_rcu(ptype, head, list) {
4012 if (ptype->type != type || !ptype->callbacks.gro_receive)
4015 skb_set_network_header(skb, skb_gro_offset(skb));
4016 skb_reset_mac_len(skb);
4017 NAPI_GRO_CB(skb)->same_flow = 0;
4018 NAPI_GRO_CB(skb)->flush = 0;
4019 NAPI_GRO_CB(skb)->free = 0;
4020 NAPI_GRO_CB(skb)->udp_mark = 0;
4022 /* Setup for GRO checksum validation */
4023 switch (skb->ip_summed) {
4024 case CHECKSUM_COMPLETE:
4025 NAPI_GRO_CB(skb)->csum = skb->csum;
4026 NAPI_GRO_CB(skb)->csum_valid = 1;
4027 NAPI_GRO_CB(skb)->csum_cnt = 0;
4029 case CHECKSUM_UNNECESSARY:
4030 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4031 NAPI_GRO_CB(skb)->csum_valid = 0;
4034 NAPI_GRO_CB(skb)->csum_cnt = 0;
4035 NAPI_GRO_CB(skb)->csum_valid = 0;
4038 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4043 if (&ptype->list == head)
4046 same_flow = NAPI_GRO_CB(skb)->same_flow;
4047 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4050 struct sk_buff *nskb = *pp;
4054 napi_gro_complete(nskb);
4061 if (NAPI_GRO_CB(skb)->flush)
4064 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4065 struct sk_buff *nskb = napi->gro_list;
4067 /* locate the end of the list to select the 'oldest' flow */
4068 while (nskb->next) {
4074 napi_gro_complete(nskb);
4078 NAPI_GRO_CB(skb)->count = 1;
4079 NAPI_GRO_CB(skb)->age = jiffies;
4080 NAPI_GRO_CB(skb)->last = skb;
4081 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4082 skb->next = napi->gro_list;
4083 napi->gro_list = skb;
4087 grow = skb_gro_offset(skb) - skb_headlen(skb);
4089 gro_pull_from_frag0(skb, grow);
4098 struct packet_offload *gro_find_receive_by_type(__be16 type)
4100 struct list_head *offload_head = &offload_base;
4101 struct packet_offload *ptype;
4103 list_for_each_entry_rcu(ptype, offload_head, list) {
4104 if (ptype->type != type || !ptype->callbacks.gro_receive)
4110 EXPORT_SYMBOL(gro_find_receive_by_type);
4112 struct packet_offload *gro_find_complete_by_type(__be16 type)
4114 struct list_head *offload_head = &offload_base;
4115 struct packet_offload *ptype;
4117 list_for_each_entry_rcu(ptype, offload_head, list) {
4118 if (ptype->type != type || !ptype->callbacks.gro_complete)
4124 EXPORT_SYMBOL(gro_find_complete_by_type);
4126 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4130 if (netif_receive_skb_internal(skb))
4138 case GRO_MERGED_FREE:
4139 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4140 kmem_cache_free(skbuff_head_cache, skb);
4153 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4155 trace_napi_gro_receive_entry(skb);
4157 skb_gro_reset_offset(skb);
4159 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4161 EXPORT_SYMBOL(napi_gro_receive);
4163 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4165 if (unlikely(skb->pfmemalloc)) {
4169 __skb_pull(skb, skb_headlen(skb));
4170 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4171 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4173 skb->dev = napi->dev;
4175 skb->encapsulation = 0;
4176 skb_shinfo(skb)->gso_type = 0;
4177 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4182 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4184 struct sk_buff *skb = napi->skb;
4187 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4192 EXPORT_SYMBOL(napi_get_frags);
4194 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4195 struct sk_buff *skb,
4201 __skb_push(skb, ETH_HLEN);
4202 skb->protocol = eth_type_trans(skb, skb->dev);
4203 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4208 case GRO_MERGED_FREE:
4209 napi_reuse_skb(napi, skb);
4219 /* Upper GRO stack assumes network header starts at gro_offset=0
4220 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4221 * We copy ethernet header into skb->data to have a common layout.
4223 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4225 struct sk_buff *skb = napi->skb;
4226 const struct ethhdr *eth;
4227 unsigned int hlen = sizeof(*eth);
4231 skb_reset_mac_header(skb);
4232 skb_gro_reset_offset(skb);
4234 eth = skb_gro_header_fast(skb, 0);
4235 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4236 eth = skb_gro_header_slow(skb, hlen, 0);
4237 if (unlikely(!eth)) {
4238 napi_reuse_skb(napi, skb);
4242 gro_pull_from_frag0(skb, hlen);
4243 NAPI_GRO_CB(skb)->frag0 += hlen;
4244 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4246 __skb_pull(skb, hlen);
4249 * This works because the only protocols we care about don't require
4251 * We'll fix it up properly in napi_frags_finish()
4253 skb->protocol = eth->h_proto;
4258 gro_result_t napi_gro_frags(struct napi_struct *napi)
4260 struct sk_buff *skb = napi_frags_skb(napi);
4265 trace_napi_gro_frags_entry(skb);
4267 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4269 EXPORT_SYMBOL(napi_gro_frags);
4271 /* Compute the checksum from gro_offset and return the folded value
4272 * after adding in any pseudo checksum.
4274 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4279 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4281 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4282 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4284 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4285 !skb->csum_complete_sw)
4286 netdev_rx_csum_fault(skb->dev);
4289 NAPI_GRO_CB(skb)->csum = wsum;
4290 NAPI_GRO_CB(skb)->csum_valid = 1;
4294 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4297 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4298 * Note: called with local irq disabled, but exits with local irq enabled.
4300 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4303 struct softnet_data *remsd = sd->rps_ipi_list;
4306 sd->rps_ipi_list = NULL;
4310 /* Send pending IPI's to kick RPS processing on remote cpus. */
4312 struct softnet_data *next = remsd->rps_ipi_next;
4314 if (cpu_online(remsd->cpu))
4315 smp_call_function_single_async(remsd->cpu,
4324 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4327 return sd->rps_ipi_list != NULL;
4333 static int process_backlog(struct napi_struct *napi, int quota)
4336 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4338 /* Check if we have pending ipi, its better to send them now,
4339 * not waiting net_rx_action() end.
4341 if (sd_has_rps_ipi_waiting(sd)) {
4342 local_irq_disable();
4343 net_rps_action_and_irq_enable(sd);
4346 napi->weight = weight_p;
4347 local_irq_disable();
4349 struct sk_buff *skb;
4351 while ((skb = __skb_dequeue(&sd->process_queue))) {
4353 __netif_receive_skb(skb);
4354 local_irq_disable();
4355 input_queue_head_incr(sd);
4356 if (++work >= quota) {
4363 if (skb_queue_empty(&sd->input_pkt_queue)) {
4365 * Inline a custom version of __napi_complete().
4366 * only current cpu owns and manipulates this napi,
4367 * and NAPI_STATE_SCHED is the only possible flag set
4369 * We can use a plain write instead of clear_bit(),
4370 * and we dont need an smp_mb() memory barrier.
4378 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4379 &sd->process_queue);
4388 * __napi_schedule - schedule for receive
4389 * @n: entry to schedule
4391 * The entry's receive function will be scheduled to run.
4392 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4394 void __napi_schedule(struct napi_struct *n)
4396 unsigned long flags;
4398 local_irq_save(flags);
4399 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4400 local_irq_restore(flags);
4402 EXPORT_SYMBOL(__napi_schedule);
4405 * __napi_schedule_irqoff - schedule for receive
4406 * @n: entry to schedule
4408 * Variant of __napi_schedule() assuming hard irqs are masked
4410 void __napi_schedule_irqoff(struct napi_struct *n)
4412 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4414 EXPORT_SYMBOL(__napi_schedule_irqoff);
4416 void __napi_complete(struct napi_struct *n)
4418 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4420 list_del_init(&n->poll_list);
4421 smp_mb__before_atomic();
4422 clear_bit(NAPI_STATE_SCHED, &n->state);
4424 EXPORT_SYMBOL(__napi_complete);
4426 void napi_complete_done(struct napi_struct *n, int work_done)
4428 unsigned long flags;
4431 * don't let napi dequeue from the cpu poll list
4432 * just in case its running on a different cpu
4434 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4438 unsigned long timeout = 0;
4441 timeout = n->dev->gro_flush_timeout;
4444 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4445 HRTIMER_MODE_REL_PINNED);
4447 napi_gro_flush(n, false);
4449 if (likely(list_empty(&n->poll_list))) {
4450 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4452 /* If n->poll_list is not empty, we need to mask irqs */
4453 local_irq_save(flags);
4455 local_irq_restore(flags);
4458 EXPORT_SYMBOL(napi_complete_done);
4460 /* must be called under rcu_read_lock(), as we dont take a reference */
4461 struct napi_struct *napi_by_id(unsigned int napi_id)
4463 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4464 struct napi_struct *napi;
4466 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4467 if (napi->napi_id == napi_id)
4472 EXPORT_SYMBOL_GPL(napi_by_id);
4474 void napi_hash_add(struct napi_struct *napi)
4476 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4478 spin_lock(&napi_hash_lock);
4480 /* 0 is not a valid id, we also skip an id that is taken
4481 * we expect both events to be extremely rare
4484 while (!napi->napi_id) {
4485 napi->napi_id = ++napi_gen_id;
4486 if (napi_by_id(napi->napi_id))
4490 hlist_add_head_rcu(&napi->napi_hash_node,
4491 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4493 spin_unlock(&napi_hash_lock);
4496 EXPORT_SYMBOL_GPL(napi_hash_add);
4498 /* Warning : caller is responsible to make sure rcu grace period
4499 * is respected before freeing memory containing @napi
4501 void napi_hash_del(struct napi_struct *napi)
4503 spin_lock(&napi_hash_lock);
4505 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4506 hlist_del_rcu(&napi->napi_hash_node);
4508 spin_unlock(&napi_hash_lock);
4510 EXPORT_SYMBOL_GPL(napi_hash_del);
4512 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4514 struct napi_struct *napi;
4516 napi = container_of(timer, struct napi_struct, timer);
4518 napi_schedule(napi);
4520 return HRTIMER_NORESTART;
4523 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4524 int (*poll)(struct napi_struct *, int), int weight)
4526 INIT_LIST_HEAD(&napi->poll_list);
4527 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4528 napi->timer.function = napi_watchdog;
4529 napi->gro_count = 0;
4530 napi->gro_list = NULL;
4533 if (weight > NAPI_POLL_WEIGHT)
4534 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4536 napi->weight = weight;
4537 list_add(&napi->dev_list, &dev->napi_list);
4539 #ifdef CONFIG_NETPOLL
4540 spin_lock_init(&napi->poll_lock);
4541 napi->poll_owner = -1;
4543 set_bit(NAPI_STATE_SCHED, &napi->state);
4545 EXPORT_SYMBOL(netif_napi_add);
4547 void napi_disable(struct napi_struct *n)
4550 set_bit(NAPI_STATE_DISABLE, &n->state);
4552 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4555 hrtimer_cancel(&n->timer);
4557 clear_bit(NAPI_STATE_DISABLE, &n->state);
4559 EXPORT_SYMBOL(napi_disable);
4561 void netif_napi_del(struct napi_struct *napi)
4563 list_del_init(&napi->dev_list);
4564 napi_free_frags(napi);
4566 kfree_skb_list(napi->gro_list);
4567 napi->gro_list = NULL;
4568 napi->gro_count = 0;
4570 EXPORT_SYMBOL(netif_napi_del);
4572 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4577 list_del_init(&n->poll_list);
4579 have = netpoll_poll_lock(n);
4583 /* This NAPI_STATE_SCHED test is for avoiding a race
4584 * with netpoll's poll_napi(). Only the entity which
4585 * obtains the lock and sees NAPI_STATE_SCHED set will
4586 * actually make the ->poll() call. Therefore we avoid
4587 * accidentally calling ->poll() when NAPI is not scheduled.
4590 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4591 work = n->poll(n, weight);
4595 WARN_ON_ONCE(work > weight);
4597 if (likely(work < weight))
4600 /* Drivers must not modify the NAPI state if they
4601 * consume the entire weight. In such cases this code
4602 * still "owns" the NAPI instance and therefore can
4603 * move the instance around on the list at-will.
4605 if (unlikely(napi_disable_pending(n))) {
4611 /* flush too old packets
4612 * If HZ < 1000, flush all packets.
4614 napi_gro_flush(n, HZ >= 1000);
4617 /* Some drivers may have called napi_schedule
4618 * prior to exhausting their budget.
4620 if (unlikely(!list_empty(&n->poll_list))) {
4621 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4622 n->dev ? n->dev->name : "backlog");
4626 list_add_tail(&n->poll_list, repoll);
4629 netpoll_poll_unlock(have);
4634 static void net_rx_action(struct softirq_action *h)
4636 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4637 unsigned long time_limit = jiffies + 2;
4638 int budget = netdev_budget;
4642 local_irq_disable();
4643 list_splice_init(&sd->poll_list, &list);
4647 struct napi_struct *n;
4649 if (list_empty(&list)) {
4650 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4655 n = list_first_entry(&list, struct napi_struct, poll_list);
4656 budget -= napi_poll(n, &repoll);
4658 /* If softirq window is exhausted then punt.
4659 * Allow this to run for 2 jiffies since which will allow
4660 * an average latency of 1.5/HZ.
4662 if (unlikely(budget <= 0 ||
4663 time_after_eq(jiffies, time_limit))) {
4669 local_irq_disable();
4671 list_splice_tail_init(&sd->poll_list, &list);
4672 list_splice_tail(&repoll, &list);
4673 list_splice(&list, &sd->poll_list);
4674 if (!list_empty(&sd->poll_list))
4675 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4677 net_rps_action_and_irq_enable(sd);
4680 struct netdev_adjacent {
4681 struct net_device *dev;
4683 /* upper master flag, there can only be one master device per list */
4686 /* counter for the number of times this device was added to us */
4689 /* private field for the users */
4692 struct list_head list;
4693 struct rcu_head rcu;
4696 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4697 struct net_device *adj_dev,
4698 struct list_head *adj_list)
4700 struct netdev_adjacent *adj;
4702 list_for_each_entry(adj, adj_list, list) {
4703 if (adj->dev == adj_dev)
4710 * netdev_has_upper_dev - Check if device is linked to an upper device
4712 * @upper_dev: upper device to check
4714 * Find out if a device is linked to specified upper device and return true
4715 * in case it is. Note that this checks only immediate upper device,
4716 * not through a complete stack of devices. The caller must hold the RTNL lock.
4718 bool netdev_has_upper_dev(struct net_device *dev,
4719 struct net_device *upper_dev)
4723 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4725 EXPORT_SYMBOL(netdev_has_upper_dev);
4728 * netdev_has_any_upper_dev - Check if device is linked to some device
4731 * Find out if a device is linked to an upper device and return true in case
4732 * it is. The caller must hold the RTNL lock.
4734 static bool netdev_has_any_upper_dev(struct net_device *dev)
4738 return !list_empty(&dev->all_adj_list.upper);
4742 * netdev_master_upper_dev_get - Get master upper device
4745 * Find a master upper device and return pointer to it or NULL in case
4746 * it's not there. The caller must hold the RTNL lock.
4748 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4750 struct netdev_adjacent *upper;
4754 if (list_empty(&dev->adj_list.upper))
4757 upper = list_first_entry(&dev->adj_list.upper,
4758 struct netdev_adjacent, list);
4759 if (likely(upper->master))
4763 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4765 void *netdev_adjacent_get_private(struct list_head *adj_list)
4767 struct netdev_adjacent *adj;
4769 adj = list_entry(adj_list, struct netdev_adjacent, list);
4771 return adj->private;
4773 EXPORT_SYMBOL(netdev_adjacent_get_private);
4776 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4778 * @iter: list_head ** of the current position
4780 * Gets the next device from the dev's upper list, starting from iter
4781 * position. The caller must hold RCU read lock.
4783 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4784 struct list_head **iter)
4786 struct netdev_adjacent *upper;
4788 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4790 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4792 if (&upper->list == &dev->adj_list.upper)
4795 *iter = &upper->list;
4799 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4802 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4804 * @iter: list_head ** of the current position
4806 * Gets the next device from the dev's upper list, starting from iter
4807 * position. The caller must hold RCU read lock.
4809 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4810 struct list_head **iter)
4812 struct netdev_adjacent *upper;
4814 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4816 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4818 if (&upper->list == &dev->all_adj_list.upper)
4821 *iter = &upper->list;
4825 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4828 * netdev_lower_get_next_private - Get the next ->private from the
4829 * lower neighbour list
4831 * @iter: list_head ** of the current position
4833 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4834 * list, starting from iter position. The caller must hold either hold the
4835 * RTNL lock or its own locking that guarantees that the neighbour lower
4836 * list will remain unchainged.
4838 void *netdev_lower_get_next_private(struct net_device *dev,
4839 struct list_head **iter)
4841 struct netdev_adjacent *lower;
4843 lower = list_entry(*iter, struct netdev_adjacent, list);
4845 if (&lower->list == &dev->adj_list.lower)
4848 *iter = lower->list.next;
4850 return lower->private;
4852 EXPORT_SYMBOL(netdev_lower_get_next_private);
4855 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4856 * lower neighbour list, RCU
4859 * @iter: list_head ** of the current position
4861 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4862 * list, starting from iter position. The caller must hold RCU read lock.
4864 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4865 struct list_head **iter)
4867 struct netdev_adjacent *lower;
4869 WARN_ON_ONCE(!rcu_read_lock_held());
4871 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4873 if (&lower->list == &dev->adj_list.lower)
4876 *iter = &lower->list;
4878 return lower->private;
4880 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4883 * netdev_lower_get_next - Get the next device from the lower neighbour
4886 * @iter: list_head ** of the current position
4888 * Gets the next netdev_adjacent from the dev's lower neighbour
4889 * list, starting from iter position. The caller must hold RTNL lock or
4890 * its own locking that guarantees that the neighbour lower
4891 * list will remain unchainged.
4893 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4895 struct netdev_adjacent *lower;
4897 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4899 if (&lower->list == &dev->adj_list.lower)
4902 *iter = &lower->list;
4906 EXPORT_SYMBOL(netdev_lower_get_next);
4909 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4910 * lower neighbour list, RCU
4914 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4915 * list. The caller must hold RCU read lock.
4917 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4919 struct netdev_adjacent *lower;
4921 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4922 struct netdev_adjacent, list);
4924 return lower->private;
4927 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4930 * netdev_master_upper_dev_get_rcu - Get master upper device
4933 * Find a master upper device and return pointer to it or NULL in case
4934 * it's not there. The caller must hold the RCU read lock.
4936 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4938 struct netdev_adjacent *upper;
4940 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4941 struct netdev_adjacent, list);
4942 if (upper && likely(upper->master))
4946 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4948 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4949 struct net_device *adj_dev,
4950 struct list_head *dev_list)
4952 char linkname[IFNAMSIZ+7];
4953 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4954 "upper_%s" : "lower_%s", adj_dev->name);
4955 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4958 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4960 struct list_head *dev_list)
4962 char linkname[IFNAMSIZ+7];
4963 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4964 "upper_%s" : "lower_%s", name);
4965 sysfs_remove_link(&(dev->dev.kobj), linkname);
4968 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
4969 struct net_device *adj_dev,
4970 struct list_head *dev_list)
4972 return (dev_list == &dev->adj_list.upper ||
4973 dev_list == &dev->adj_list.lower) &&
4974 net_eq(dev_net(dev), dev_net(adj_dev));
4977 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4978 struct net_device *adj_dev,
4979 struct list_head *dev_list,
4980 void *private, bool master)
4982 struct netdev_adjacent *adj;
4985 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4992 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4997 adj->master = master;
4999 adj->private = private;
5002 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5003 adj_dev->name, dev->name, adj_dev->name);
5005 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5006 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5011 /* Ensure that master link is always the first item in list. */
5013 ret = sysfs_create_link(&(dev->dev.kobj),
5014 &(adj_dev->dev.kobj), "master");
5016 goto remove_symlinks;
5018 list_add_rcu(&adj->list, dev_list);
5020 list_add_tail_rcu(&adj->list, dev_list);
5026 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5027 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5035 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5036 struct net_device *adj_dev,
5037 struct list_head *dev_list)
5039 struct netdev_adjacent *adj;
5041 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5044 pr_err("tried to remove device %s from %s\n",
5045 dev->name, adj_dev->name);
5049 if (adj->ref_nr > 1) {
5050 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5057 sysfs_remove_link(&(dev->dev.kobj), "master");
5059 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5060 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5062 list_del_rcu(&adj->list);
5063 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5064 adj_dev->name, dev->name, adj_dev->name);
5066 kfree_rcu(adj, rcu);
5069 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5070 struct net_device *upper_dev,
5071 struct list_head *up_list,
5072 struct list_head *down_list,
5073 void *private, bool master)
5077 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5082 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5085 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5092 static int __netdev_adjacent_dev_link(struct net_device *dev,
5093 struct net_device *upper_dev)
5095 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5096 &dev->all_adj_list.upper,
5097 &upper_dev->all_adj_list.lower,
5101 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5102 struct net_device *upper_dev,
5103 struct list_head *up_list,
5104 struct list_head *down_list)
5106 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5107 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5110 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5111 struct net_device *upper_dev)
5113 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5114 &dev->all_adj_list.upper,
5115 &upper_dev->all_adj_list.lower);
5118 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5119 struct net_device *upper_dev,
5120 void *private, bool master)
5122 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5127 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5128 &dev->adj_list.upper,
5129 &upper_dev->adj_list.lower,
5132 __netdev_adjacent_dev_unlink(dev, upper_dev);
5139 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5140 struct net_device *upper_dev)
5142 __netdev_adjacent_dev_unlink(dev, upper_dev);
5143 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5144 &dev->adj_list.upper,
5145 &upper_dev->adj_list.lower);
5148 static int __netdev_upper_dev_link(struct net_device *dev,
5149 struct net_device *upper_dev, bool master,
5152 struct netdev_adjacent *i, *j, *to_i, *to_j;
5157 if (dev == upper_dev)
5160 /* To prevent loops, check if dev is not upper device to upper_dev. */
5161 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5164 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
5167 if (master && netdev_master_upper_dev_get(dev))
5170 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5175 /* Now that we linked these devs, make all the upper_dev's
5176 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5177 * versa, and don't forget the devices itself. All of these
5178 * links are non-neighbours.
5180 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5181 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5182 pr_debug("Interlinking %s with %s, non-neighbour\n",
5183 i->dev->name, j->dev->name);
5184 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5190 /* add dev to every upper_dev's upper device */
5191 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5192 pr_debug("linking %s's upper device %s with %s\n",
5193 upper_dev->name, i->dev->name, dev->name);
5194 ret = __netdev_adjacent_dev_link(dev, i->dev);
5196 goto rollback_upper_mesh;
5199 /* add upper_dev to every dev's lower device */
5200 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5201 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5202 i->dev->name, upper_dev->name);
5203 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5205 goto rollback_lower_mesh;
5208 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5211 rollback_lower_mesh:
5213 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5216 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5221 rollback_upper_mesh:
5223 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5226 __netdev_adjacent_dev_unlink(dev, i->dev);
5234 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5235 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5236 if (i == to_i && j == to_j)
5238 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5244 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5250 * netdev_upper_dev_link - Add a link to the upper device
5252 * @upper_dev: new upper device
5254 * Adds a link to device which is upper to this one. The caller must hold
5255 * the RTNL lock. On a failure a negative errno code is returned.
5256 * On success the reference counts are adjusted and the function
5259 int netdev_upper_dev_link(struct net_device *dev,
5260 struct net_device *upper_dev)
5262 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5264 EXPORT_SYMBOL(netdev_upper_dev_link);
5267 * netdev_master_upper_dev_link - Add a master link to the upper device
5269 * @upper_dev: new upper device
5271 * Adds a link to device which is upper to this one. In this case, only
5272 * one master upper device can be linked, although other non-master devices
5273 * might be linked as well. The caller must hold the RTNL lock.
5274 * On a failure a negative errno code is returned. On success the reference
5275 * counts are adjusted and the function returns zero.
5277 int netdev_master_upper_dev_link(struct net_device *dev,
5278 struct net_device *upper_dev)
5280 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5282 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5284 int netdev_master_upper_dev_link_private(struct net_device *dev,
5285 struct net_device *upper_dev,
5288 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5290 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5293 * netdev_upper_dev_unlink - Removes a link to upper device
5295 * @upper_dev: new upper device
5297 * Removes a link to device which is upper to this one. The caller must hold
5300 void netdev_upper_dev_unlink(struct net_device *dev,
5301 struct net_device *upper_dev)
5303 struct netdev_adjacent *i, *j;
5306 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5308 /* Here is the tricky part. We must remove all dev's lower
5309 * devices from all upper_dev's upper devices and vice
5310 * versa, to maintain the graph relationship.
5312 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5313 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5314 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5316 /* remove also the devices itself from lower/upper device
5319 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5320 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5322 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5323 __netdev_adjacent_dev_unlink(dev, i->dev);
5325 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5327 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5330 * netdev_bonding_info_change - Dispatch event about slave change
5332 * @netdev_bonding_info: info to dispatch
5334 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5335 * The caller must hold the RTNL lock.
5337 void netdev_bonding_info_change(struct net_device *dev,
5338 struct netdev_bonding_info *bonding_info)
5340 struct netdev_notifier_bonding_info info;
5342 memcpy(&info.bonding_info, bonding_info,
5343 sizeof(struct netdev_bonding_info));
5344 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5347 EXPORT_SYMBOL(netdev_bonding_info_change);
5349 static void netdev_adjacent_add_links(struct net_device *dev)
5351 struct netdev_adjacent *iter;
5353 struct net *net = dev_net(dev);
5355 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5356 if (!net_eq(net,dev_net(iter->dev)))
5358 netdev_adjacent_sysfs_add(iter->dev, dev,
5359 &iter->dev->adj_list.lower);
5360 netdev_adjacent_sysfs_add(dev, iter->dev,
5361 &dev->adj_list.upper);
5364 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5365 if (!net_eq(net,dev_net(iter->dev)))
5367 netdev_adjacent_sysfs_add(iter->dev, dev,
5368 &iter->dev->adj_list.upper);
5369 netdev_adjacent_sysfs_add(dev, iter->dev,
5370 &dev->adj_list.lower);
5374 static void netdev_adjacent_del_links(struct net_device *dev)
5376 struct netdev_adjacent *iter;
5378 struct net *net = dev_net(dev);
5380 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5381 if (!net_eq(net,dev_net(iter->dev)))
5383 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5384 &iter->dev->adj_list.lower);
5385 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5386 &dev->adj_list.upper);
5389 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5390 if (!net_eq(net,dev_net(iter->dev)))
5392 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5393 &iter->dev->adj_list.upper);
5394 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5395 &dev->adj_list.lower);
5399 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5401 struct netdev_adjacent *iter;
5403 struct net *net = dev_net(dev);
5405 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5406 if (!net_eq(net,dev_net(iter->dev)))
5408 netdev_adjacent_sysfs_del(iter->dev, oldname,
5409 &iter->dev->adj_list.lower);
5410 netdev_adjacent_sysfs_add(iter->dev, dev,
5411 &iter->dev->adj_list.lower);
5414 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5415 if (!net_eq(net,dev_net(iter->dev)))
5417 netdev_adjacent_sysfs_del(iter->dev, oldname,
5418 &iter->dev->adj_list.upper);
5419 netdev_adjacent_sysfs_add(iter->dev, dev,
5420 &iter->dev->adj_list.upper);
5424 void *netdev_lower_dev_get_private(struct net_device *dev,
5425 struct net_device *lower_dev)
5427 struct netdev_adjacent *lower;
5431 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5435 return lower->private;
5437 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5440 int dev_get_nest_level(struct net_device *dev,
5441 bool (*type_check)(struct net_device *dev))
5443 struct net_device *lower = NULL;
5444 struct list_head *iter;
5450 netdev_for_each_lower_dev(dev, lower, iter) {
5451 nest = dev_get_nest_level(lower, type_check);
5452 if (max_nest < nest)
5456 if (type_check(dev))
5461 EXPORT_SYMBOL(dev_get_nest_level);
5463 static void dev_change_rx_flags(struct net_device *dev, int flags)
5465 const struct net_device_ops *ops = dev->netdev_ops;
5467 if (ops->ndo_change_rx_flags)
5468 ops->ndo_change_rx_flags(dev, flags);
5471 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5473 unsigned int old_flags = dev->flags;
5479 dev->flags |= IFF_PROMISC;
5480 dev->promiscuity += inc;
5481 if (dev->promiscuity == 0) {
5484 * If inc causes overflow, untouch promisc and return error.
5487 dev->flags &= ~IFF_PROMISC;
5489 dev->promiscuity -= inc;
5490 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5495 if (dev->flags != old_flags) {
5496 pr_info("device %s %s promiscuous mode\n",
5498 dev->flags & IFF_PROMISC ? "entered" : "left");
5499 if (audit_enabled) {
5500 current_uid_gid(&uid, &gid);
5501 audit_log(current->audit_context, GFP_ATOMIC,
5502 AUDIT_ANOM_PROMISCUOUS,
5503 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5504 dev->name, (dev->flags & IFF_PROMISC),
5505 (old_flags & IFF_PROMISC),
5506 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5507 from_kuid(&init_user_ns, uid),
5508 from_kgid(&init_user_ns, gid),
5509 audit_get_sessionid(current));
5512 dev_change_rx_flags(dev, IFF_PROMISC);
5515 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5520 * dev_set_promiscuity - update promiscuity count on a device
5524 * Add or remove promiscuity from a device. While the count in the device
5525 * remains above zero the interface remains promiscuous. Once it hits zero
5526 * the device reverts back to normal filtering operation. A negative inc
5527 * value is used to drop promiscuity on the device.
5528 * Return 0 if successful or a negative errno code on error.
5530 int dev_set_promiscuity(struct net_device *dev, int inc)
5532 unsigned int old_flags = dev->flags;
5535 err = __dev_set_promiscuity(dev, inc, true);
5538 if (dev->flags != old_flags)
5539 dev_set_rx_mode(dev);
5542 EXPORT_SYMBOL(dev_set_promiscuity);
5544 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5546 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5550 dev->flags |= IFF_ALLMULTI;
5551 dev->allmulti += inc;
5552 if (dev->allmulti == 0) {
5555 * If inc causes overflow, untouch allmulti and return error.
5558 dev->flags &= ~IFF_ALLMULTI;
5560 dev->allmulti -= inc;
5561 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5566 if (dev->flags ^ old_flags) {
5567 dev_change_rx_flags(dev, IFF_ALLMULTI);
5568 dev_set_rx_mode(dev);
5570 __dev_notify_flags(dev, old_flags,
5571 dev->gflags ^ old_gflags);
5577 * dev_set_allmulti - update allmulti count on a device
5581 * Add or remove reception of all multicast frames to a device. While the
5582 * count in the device remains above zero the interface remains listening
5583 * to all interfaces. Once it hits zero the device reverts back to normal
5584 * filtering operation. A negative @inc value is used to drop the counter
5585 * when releasing a resource needing all multicasts.
5586 * Return 0 if successful or a negative errno code on error.
5589 int dev_set_allmulti(struct net_device *dev, int inc)
5591 return __dev_set_allmulti(dev, inc, true);
5593 EXPORT_SYMBOL(dev_set_allmulti);
5596 * Upload unicast and multicast address lists to device and
5597 * configure RX filtering. When the device doesn't support unicast
5598 * filtering it is put in promiscuous mode while unicast addresses
5601 void __dev_set_rx_mode(struct net_device *dev)
5603 const struct net_device_ops *ops = dev->netdev_ops;
5605 /* dev_open will call this function so the list will stay sane. */
5606 if (!(dev->flags&IFF_UP))
5609 if (!netif_device_present(dev))
5612 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5613 /* Unicast addresses changes may only happen under the rtnl,
5614 * therefore calling __dev_set_promiscuity here is safe.
5616 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5617 __dev_set_promiscuity(dev, 1, false);
5618 dev->uc_promisc = true;
5619 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5620 __dev_set_promiscuity(dev, -1, false);
5621 dev->uc_promisc = false;
5625 if (ops->ndo_set_rx_mode)
5626 ops->ndo_set_rx_mode(dev);
5629 void dev_set_rx_mode(struct net_device *dev)
5631 netif_addr_lock_bh(dev);
5632 __dev_set_rx_mode(dev);
5633 netif_addr_unlock_bh(dev);
5637 * dev_get_flags - get flags reported to userspace
5640 * Get the combination of flag bits exported through APIs to userspace.
5642 unsigned int dev_get_flags(const struct net_device *dev)
5646 flags = (dev->flags & ~(IFF_PROMISC |
5651 (dev->gflags & (IFF_PROMISC |
5654 if (netif_running(dev)) {
5655 if (netif_oper_up(dev))
5656 flags |= IFF_RUNNING;
5657 if (netif_carrier_ok(dev))
5658 flags |= IFF_LOWER_UP;
5659 if (netif_dormant(dev))
5660 flags |= IFF_DORMANT;
5665 EXPORT_SYMBOL(dev_get_flags);
5667 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5669 unsigned int old_flags = dev->flags;
5675 * Set the flags on our device.
5678 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5679 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5681 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5685 * Load in the correct multicast list now the flags have changed.
5688 if ((old_flags ^ flags) & IFF_MULTICAST)
5689 dev_change_rx_flags(dev, IFF_MULTICAST);
5691 dev_set_rx_mode(dev);
5694 * Have we downed the interface. We handle IFF_UP ourselves
5695 * according to user attempts to set it, rather than blindly
5700 if ((old_flags ^ flags) & IFF_UP)
5701 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5703 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5704 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5705 unsigned int old_flags = dev->flags;
5707 dev->gflags ^= IFF_PROMISC;
5709 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5710 if (dev->flags != old_flags)
5711 dev_set_rx_mode(dev);
5714 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5715 is important. Some (broken) drivers set IFF_PROMISC, when
5716 IFF_ALLMULTI is requested not asking us and not reporting.
5718 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5719 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5721 dev->gflags ^= IFF_ALLMULTI;
5722 __dev_set_allmulti(dev, inc, false);
5728 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5729 unsigned int gchanges)
5731 unsigned int changes = dev->flags ^ old_flags;
5734 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5736 if (changes & IFF_UP) {
5737 if (dev->flags & IFF_UP)
5738 call_netdevice_notifiers(NETDEV_UP, dev);
5740 call_netdevice_notifiers(NETDEV_DOWN, dev);
5743 if (dev->flags & IFF_UP &&
5744 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5745 struct netdev_notifier_change_info change_info;
5747 change_info.flags_changed = changes;
5748 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5754 * dev_change_flags - change device settings
5756 * @flags: device state flags
5758 * Change settings on device based state flags. The flags are
5759 * in the userspace exported format.
5761 int dev_change_flags(struct net_device *dev, unsigned int flags)
5764 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5766 ret = __dev_change_flags(dev, flags);
5770 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5771 __dev_notify_flags(dev, old_flags, changes);
5774 EXPORT_SYMBOL(dev_change_flags);
5776 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5778 const struct net_device_ops *ops = dev->netdev_ops;
5780 if (ops->ndo_change_mtu)
5781 return ops->ndo_change_mtu(dev, new_mtu);
5788 * dev_set_mtu - Change maximum transfer unit
5790 * @new_mtu: new transfer unit
5792 * Change the maximum transfer size of the network device.
5794 int dev_set_mtu(struct net_device *dev, int new_mtu)
5798 if (new_mtu == dev->mtu)
5801 /* MTU must be positive. */
5805 if (!netif_device_present(dev))
5808 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5809 err = notifier_to_errno(err);
5813 orig_mtu = dev->mtu;
5814 err = __dev_set_mtu(dev, new_mtu);
5817 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5818 err = notifier_to_errno(err);
5820 /* setting mtu back and notifying everyone again,
5821 * so that they have a chance to revert changes.
5823 __dev_set_mtu(dev, orig_mtu);
5824 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5829 EXPORT_SYMBOL(dev_set_mtu);
5832 * dev_set_group - Change group this device belongs to
5834 * @new_group: group this device should belong to
5836 void dev_set_group(struct net_device *dev, int new_group)
5838 dev->group = new_group;
5840 EXPORT_SYMBOL(dev_set_group);
5843 * dev_set_mac_address - Change Media Access Control Address
5847 * Change the hardware (MAC) address of the device
5849 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5851 const struct net_device_ops *ops = dev->netdev_ops;
5854 if (!ops->ndo_set_mac_address)
5856 if (sa->sa_family != dev->type)
5858 if (!netif_device_present(dev))
5860 err = ops->ndo_set_mac_address(dev, sa);
5863 dev->addr_assign_type = NET_ADDR_SET;
5864 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5865 add_device_randomness(dev->dev_addr, dev->addr_len);
5868 EXPORT_SYMBOL(dev_set_mac_address);
5871 * dev_change_carrier - Change device carrier
5873 * @new_carrier: new value
5875 * Change device carrier
5877 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5879 const struct net_device_ops *ops = dev->netdev_ops;
5881 if (!ops->ndo_change_carrier)
5883 if (!netif_device_present(dev))
5885 return ops->ndo_change_carrier(dev, new_carrier);
5887 EXPORT_SYMBOL(dev_change_carrier);
5890 * dev_get_phys_port_id - Get device physical port ID
5894 * Get device physical port ID
5896 int dev_get_phys_port_id(struct net_device *dev,
5897 struct netdev_phys_item_id *ppid)
5899 const struct net_device_ops *ops = dev->netdev_ops;
5901 if (!ops->ndo_get_phys_port_id)
5903 return ops->ndo_get_phys_port_id(dev, ppid);
5905 EXPORT_SYMBOL(dev_get_phys_port_id);
5908 * dev_new_index - allocate an ifindex
5909 * @net: the applicable net namespace
5911 * Returns a suitable unique value for a new device interface
5912 * number. The caller must hold the rtnl semaphore or the
5913 * dev_base_lock to be sure it remains unique.
5915 static int dev_new_index(struct net *net)
5917 int ifindex = net->ifindex;
5921 if (!__dev_get_by_index(net, ifindex))
5922 return net->ifindex = ifindex;
5926 /* Delayed registration/unregisteration */
5927 static LIST_HEAD(net_todo_list);
5928 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5930 static void net_set_todo(struct net_device *dev)
5932 list_add_tail(&dev->todo_list, &net_todo_list);
5933 dev_net(dev)->dev_unreg_count++;
5936 static void rollback_registered_many(struct list_head *head)
5938 struct net_device *dev, *tmp;
5939 LIST_HEAD(close_head);
5941 BUG_ON(dev_boot_phase);
5944 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5945 /* Some devices call without registering
5946 * for initialization unwind. Remove those
5947 * devices and proceed with the remaining.
5949 if (dev->reg_state == NETREG_UNINITIALIZED) {
5950 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5954 list_del(&dev->unreg_list);
5957 dev->dismantle = true;
5958 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5961 /* If device is running, close it first. */
5962 list_for_each_entry(dev, head, unreg_list)
5963 list_add_tail(&dev->close_list, &close_head);
5964 dev_close_many(&close_head);
5966 list_for_each_entry(dev, head, unreg_list) {
5967 /* And unlink it from device chain. */
5968 unlist_netdevice(dev);
5970 dev->reg_state = NETREG_UNREGISTERING;
5975 list_for_each_entry(dev, head, unreg_list) {
5976 struct sk_buff *skb = NULL;
5978 /* Shutdown queueing discipline. */
5982 /* Notify protocols, that we are about to destroy
5983 this device. They should clean all the things.
5985 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5987 if (!dev->rtnl_link_ops ||
5988 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5989 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
5993 * Flush the unicast and multicast chains
5998 if (dev->netdev_ops->ndo_uninit)
5999 dev->netdev_ops->ndo_uninit(dev);
6002 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6004 /* Notifier chain MUST detach us all upper devices. */
6005 WARN_ON(netdev_has_any_upper_dev(dev));
6007 /* Remove entries from kobject tree */
6008 netdev_unregister_kobject(dev);
6010 /* Remove XPS queueing entries */
6011 netif_reset_xps_queues_gt(dev, 0);
6017 list_for_each_entry(dev, head, unreg_list)
6021 static void rollback_registered(struct net_device *dev)
6025 list_add(&dev->unreg_list, &single);
6026 rollback_registered_many(&single);
6030 static netdev_features_t netdev_fix_features(struct net_device *dev,
6031 netdev_features_t features)
6033 /* Fix illegal checksum combinations */
6034 if ((features & NETIF_F_HW_CSUM) &&
6035 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6036 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6037 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6040 /* TSO requires that SG is present as well. */
6041 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6042 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6043 features &= ~NETIF_F_ALL_TSO;
6046 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6047 !(features & NETIF_F_IP_CSUM)) {
6048 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6049 features &= ~NETIF_F_TSO;
6050 features &= ~NETIF_F_TSO_ECN;
6053 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6054 !(features & NETIF_F_IPV6_CSUM)) {
6055 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6056 features &= ~NETIF_F_TSO6;
6059 /* TSO ECN requires that TSO is present as well. */
6060 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6061 features &= ~NETIF_F_TSO_ECN;
6063 /* Software GSO depends on SG. */
6064 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6065 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6066 features &= ~NETIF_F_GSO;
6069 /* UFO needs SG and checksumming */
6070 if (features & NETIF_F_UFO) {
6071 /* maybe split UFO into V4 and V6? */
6072 if (!((features & NETIF_F_GEN_CSUM) ||
6073 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6074 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6076 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6077 features &= ~NETIF_F_UFO;
6080 if (!(features & NETIF_F_SG)) {
6082 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6083 features &= ~NETIF_F_UFO;
6087 #ifdef CONFIG_NET_RX_BUSY_POLL
6088 if (dev->netdev_ops->ndo_busy_poll)
6089 features |= NETIF_F_BUSY_POLL;
6092 features &= ~NETIF_F_BUSY_POLL;
6097 int __netdev_update_features(struct net_device *dev)
6099 netdev_features_t features;
6104 features = netdev_get_wanted_features(dev);
6106 if (dev->netdev_ops->ndo_fix_features)
6107 features = dev->netdev_ops->ndo_fix_features(dev, features);
6109 /* driver might be less strict about feature dependencies */
6110 features = netdev_fix_features(dev, features);
6112 if (dev->features == features)
6115 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6116 &dev->features, &features);
6118 if (dev->netdev_ops->ndo_set_features)
6119 err = dev->netdev_ops->ndo_set_features(dev, features);
6121 if (unlikely(err < 0)) {
6123 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6124 err, &features, &dev->features);
6129 dev->features = features;
6135 * netdev_update_features - recalculate device features
6136 * @dev: the device to check
6138 * Recalculate dev->features set and send notifications if it
6139 * has changed. Should be called after driver or hardware dependent
6140 * conditions might have changed that influence the features.
6142 void netdev_update_features(struct net_device *dev)
6144 if (__netdev_update_features(dev))
6145 netdev_features_change(dev);
6147 EXPORT_SYMBOL(netdev_update_features);
6150 * netdev_change_features - recalculate device features
6151 * @dev: the device to check
6153 * Recalculate dev->features set and send notifications even
6154 * if they have not changed. Should be called instead of
6155 * netdev_update_features() if also dev->vlan_features might
6156 * have changed to allow the changes to be propagated to stacked
6159 void netdev_change_features(struct net_device *dev)
6161 __netdev_update_features(dev);
6162 netdev_features_change(dev);
6164 EXPORT_SYMBOL(netdev_change_features);
6167 * netif_stacked_transfer_operstate - transfer operstate
6168 * @rootdev: the root or lower level device to transfer state from
6169 * @dev: the device to transfer operstate to
6171 * Transfer operational state from root to device. This is normally
6172 * called when a stacking relationship exists between the root
6173 * device and the device(a leaf device).
6175 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6176 struct net_device *dev)
6178 if (rootdev->operstate == IF_OPER_DORMANT)
6179 netif_dormant_on(dev);
6181 netif_dormant_off(dev);
6183 if (netif_carrier_ok(rootdev)) {
6184 if (!netif_carrier_ok(dev))
6185 netif_carrier_on(dev);
6187 if (netif_carrier_ok(dev))
6188 netif_carrier_off(dev);
6191 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6194 static int netif_alloc_rx_queues(struct net_device *dev)
6196 unsigned int i, count = dev->num_rx_queues;
6197 struct netdev_rx_queue *rx;
6198 size_t sz = count * sizeof(*rx);
6202 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6210 for (i = 0; i < count; i++)
6216 static void netdev_init_one_queue(struct net_device *dev,
6217 struct netdev_queue *queue, void *_unused)
6219 /* Initialize queue lock */
6220 spin_lock_init(&queue->_xmit_lock);
6221 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6222 queue->xmit_lock_owner = -1;
6223 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6226 dql_init(&queue->dql, HZ);
6230 static void netif_free_tx_queues(struct net_device *dev)
6235 static int netif_alloc_netdev_queues(struct net_device *dev)
6237 unsigned int count = dev->num_tx_queues;
6238 struct netdev_queue *tx;
6239 size_t sz = count * sizeof(*tx);
6241 BUG_ON(count < 1 || count > 0xffff);
6243 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6251 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6252 spin_lock_init(&dev->tx_global_lock);
6258 * register_netdevice - register a network device
6259 * @dev: device to register
6261 * Take a completed network device structure and add it to the kernel
6262 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6263 * chain. 0 is returned on success. A negative errno code is returned
6264 * on a failure to set up the device, or if the name is a duplicate.
6266 * Callers must hold the rtnl semaphore. You may want
6267 * register_netdev() instead of this.
6270 * The locking appears insufficient to guarantee two parallel registers
6271 * will not get the same name.
6274 int register_netdevice(struct net_device *dev)
6277 struct net *net = dev_net(dev);
6279 BUG_ON(dev_boot_phase);
6284 /* When net_device's are persistent, this will be fatal. */
6285 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6288 spin_lock_init(&dev->addr_list_lock);
6289 netdev_set_addr_lockdep_class(dev);
6293 ret = dev_get_valid_name(net, dev, dev->name);
6297 /* Init, if this function is available */
6298 if (dev->netdev_ops->ndo_init) {
6299 ret = dev->netdev_ops->ndo_init(dev);
6307 if (((dev->hw_features | dev->features) &
6308 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6309 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6310 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6311 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6318 dev->ifindex = dev_new_index(net);
6319 else if (__dev_get_by_index(net, dev->ifindex))
6322 if (dev->iflink == -1)
6323 dev->iflink = dev->ifindex;
6325 /* Transfer changeable features to wanted_features and enable
6326 * software offloads (GSO and GRO).
6328 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6329 dev->features |= NETIF_F_SOFT_FEATURES;
6330 dev->wanted_features = dev->features & dev->hw_features;
6332 if (!(dev->flags & IFF_LOOPBACK)) {
6333 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6336 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6338 dev->vlan_features |= NETIF_F_HIGHDMA;
6340 /* Make NETIF_F_SG inheritable to tunnel devices.
6342 dev->hw_enc_features |= NETIF_F_SG;
6344 /* Make NETIF_F_SG inheritable to MPLS.
6346 dev->mpls_features |= NETIF_F_SG;
6348 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6349 ret = notifier_to_errno(ret);
6353 ret = netdev_register_kobject(dev);
6356 dev->reg_state = NETREG_REGISTERED;
6358 __netdev_update_features(dev);
6361 * Default initial state at registry is that the
6362 * device is present.
6365 set_bit(__LINK_STATE_PRESENT, &dev->state);
6367 linkwatch_init_dev(dev);
6369 dev_init_scheduler(dev);
6371 list_netdevice(dev);
6372 add_device_randomness(dev->dev_addr, dev->addr_len);
6374 /* If the device has permanent device address, driver should
6375 * set dev_addr and also addr_assign_type should be set to
6376 * NET_ADDR_PERM (default value).
6378 if (dev->addr_assign_type == NET_ADDR_PERM)
6379 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6381 /* Notify protocols, that a new device appeared. */
6382 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6383 ret = notifier_to_errno(ret);
6385 rollback_registered(dev);
6386 dev->reg_state = NETREG_UNREGISTERED;
6389 * Prevent userspace races by waiting until the network
6390 * device is fully setup before sending notifications.
6392 if (!dev->rtnl_link_ops ||
6393 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6394 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6400 if (dev->netdev_ops->ndo_uninit)
6401 dev->netdev_ops->ndo_uninit(dev);
6404 EXPORT_SYMBOL(register_netdevice);
6407 * init_dummy_netdev - init a dummy network device for NAPI
6408 * @dev: device to init
6410 * This takes a network device structure and initialize the minimum
6411 * amount of fields so it can be used to schedule NAPI polls without
6412 * registering a full blown interface. This is to be used by drivers
6413 * that need to tie several hardware interfaces to a single NAPI
6414 * poll scheduler due to HW limitations.
6416 int init_dummy_netdev(struct net_device *dev)
6418 /* Clear everything. Note we don't initialize spinlocks
6419 * are they aren't supposed to be taken by any of the
6420 * NAPI code and this dummy netdev is supposed to be
6421 * only ever used for NAPI polls
6423 memset(dev, 0, sizeof(struct net_device));
6425 /* make sure we BUG if trying to hit standard
6426 * register/unregister code path
6428 dev->reg_state = NETREG_DUMMY;
6430 /* NAPI wants this */
6431 INIT_LIST_HEAD(&dev->napi_list);
6433 /* a dummy interface is started by default */
6434 set_bit(__LINK_STATE_PRESENT, &dev->state);
6435 set_bit(__LINK_STATE_START, &dev->state);
6437 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6438 * because users of this 'device' dont need to change
6444 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6448 * register_netdev - register a network device
6449 * @dev: device to register
6451 * Take a completed network device structure and add it to the kernel
6452 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6453 * chain. 0 is returned on success. A negative errno code is returned
6454 * on a failure to set up the device, or if the name is a duplicate.
6456 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6457 * and expands the device name if you passed a format string to
6460 int register_netdev(struct net_device *dev)
6465 err = register_netdevice(dev);
6469 EXPORT_SYMBOL(register_netdev);
6471 int netdev_refcnt_read(const struct net_device *dev)
6475 for_each_possible_cpu(i)
6476 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6479 EXPORT_SYMBOL(netdev_refcnt_read);
6482 * netdev_wait_allrefs - wait until all references are gone.
6483 * @dev: target net_device
6485 * This is called when unregistering network devices.
6487 * Any protocol or device that holds a reference should register
6488 * for netdevice notification, and cleanup and put back the
6489 * reference if they receive an UNREGISTER event.
6490 * We can get stuck here if buggy protocols don't correctly
6493 static void netdev_wait_allrefs(struct net_device *dev)
6495 unsigned long rebroadcast_time, warning_time;
6498 linkwatch_forget_dev(dev);
6500 rebroadcast_time = warning_time = jiffies;
6501 refcnt = netdev_refcnt_read(dev);
6503 while (refcnt != 0) {
6504 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6507 /* Rebroadcast unregister notification */
6508 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6514 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6515 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6517 /* We must not have linkwatch events
6518 * pending on unregister. If this
6519 * happens, we simply run the queue
6520 * unscheduled, resulting in a noop
6523 linkwatch_run_queue();
6528 rebroadcast_time = jiffies;
6533 refcnt = netdev_refcnt_read(dev);
6535 if (time_after(jiffies, warning_time + 10 * HZ)) {
6536 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6538 warning_time = jiffies;
6547 * register_netdevice(x1);
6548 * register_netdevice(x2);
6550 * unregister_netdevice(y1);
6551 * unregister_netdevice(y2);
6557 * We are invoked by rtnl_unlock().
6558 * This allows us to deal with problems:
6559 * 1) We can delete sysfs objects which invoke hotplug
6560 * without deadlocking with linkwatch via keventd.
6561 * 2) Since we run with the RTNL semaphore not held, we can sleep
6562 * safely in order to wait for the netdev refcnt to drop to zero.
6564 * We must not return until all unregister events added during
6565 * the interval the lock was held have been completed.
6567 void netdev_run_todo(void)
6569 struct list_head list;
6571 /* Snapshot list, allow later requests */
6572 list_replace_init(&net_todo_list, &list);
6577 /* Wait for rcu callbacks to finish before next phase */
6578 if (!list_empty(&list))
6581 while (!list_empty(&list)) {
6582 struct net_device *dev
6583 = list_first_entry(&list, struct net_device, todo_list);
6584 list_del(&dev->todo_list);
6587 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6590 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6591 pr_err("network todo '%s' but state %d\n",
6592 dev->name, dev->reg_state);
6597 dev->reg_state = NETREG_UNREGISTERED;
6599 on_each_cpu(flush_backlog, dev, 1);
6601 netdev_wait_allrefs(dev);
6604 BUG_ON(netdev_refcnt_read(dev));
6605 BUG_ON(!list_empty(&dev->ptype_all));
6606 BUG_ON(!list_empty(&dev->ptype_specific));
6607 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6608 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6609 WARN_ON(dev->dn_ptr);
6611 if (dev->destructor)
6612 dev->destructor(dev);
6614 /* Report a network device has been unregistered */
6616 dev_net(dev)->dev_unreg_count--;
6618 wake_up(&netdev_unregistering_wq);
6620 /* Free network device */
6621 kobject_put(&dev->dev.kobj);
6625 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6626 * fields in the same order, with only the type differing.
6628 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6629 const struct net_device_stats *netdev_stats)
6631 #if BITS_PER_LONG == 64
6632 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6633 memcpy(stats64, netdev_stats, sizeof(*stats64));
6635 size_t i, n = sizeof(*stats64) / sizeof(u64);
6636 const unsigned long *src = (const unsigned long *)netdev_stats;
6637 u64 *dst = (u64 *)stats64;
6639 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6640 sizeof(*stats64) / sizeof(u64));
6641 for (i = 0; i < n; i++)
6645 EXPORT_SYMBOL(netdev_stats_to_stats64);
6648 * dev_get_stats - get network device statistics
6649 * @dev: device to get statistics from
6650 * @storage: place to store stats
6652 * Get network statistics from device. Return @storage.
6653 * The device driver may provide its own method by setting
6654 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6655 * otherwise the internal statistics structure is used.
6657 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6658 struct rtnl_link_stats64 *storage)
6660 const struct net_device_ops *ops = dev->netdev_ops;
6662 if (ops->ndo_get_stats64) {
6663 memset(storage, 0, sizeof(*storage));
6664 ops->ndo_get_stats64(dev, storage);
6665 } else if (ops->ndo_get_stats) {
6666 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6668 netdev_stats_to_stats64(storage, &dev->stats);
6670 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6671 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6674 EXPORT_SYMBOL(dev_get_stats);
6676 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6678 struct netdev_queue *queue = dev_ingress_queue(dev);
6680 #ifdef CONFIG_NET_CLS_ACT
6683 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6686 netdev_init_one_queue(dev, queue, NULL);
6687 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
6688 queue->qdisc_sleeping = &noop_qdisc;
6689 rcu_assign_pointer(dev->ingress_queue, queue);
6694 static const struct ethtool_ops default_ethtool_ops;
6696 void netdev_set_default_ethtool_ops(struct net_device *dev,
6697 const struct ethtool_ops *ops)
6699 if (dev->ethtool_ops == &default_ethtool_ops)
6700 dev->ethtool_ops = ops;
6702 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6704 void netdev_freemem(struct net_device *dev)
6706 char *addr = (char *)dev - dev->padded;
6712 * alloc_netdev_mqs - allocate network device
6713 * @sizeof_priv: size of private data to allocate space for
6714 * @name: device name format string
6715 * @name_assign_type: origin of device name
6716 * @setup: callback to initialize device
6717 * @txqs: the number of TX subqueues to allocate
6718 * @rxqs: the number of RX subqueues to allocate
6720 * Allocates a struct net_device with private data area for driver use
6721 * and performs basic initialization. Also allocates subqueue structs
6722 * for each queue on the device.
6724 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6725 unsigned char name_assign_type,
6726 void (*setup)(struct net_device *),
6727 unsigned int txqs, unsigned int rxqs)
6729 struct net_device *dev;
6731 struct net_device *p;
6733 BUG_ON(strlen(name) >= sizeof(dev->name));
6736 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6742 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6747 alloc_size = sizeof(struct net_device);
6749 /* ensure 32-byte alignment of private area */
6750 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6751 alloc_size += sizeof_priv;
6753 /* ensure 32-byte alignment of whole construct */
6754 alloc_size += NETDEV_ALIGN - 1;
6756 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6758 p = vzalloc(alloc_size);
6762 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6763 dev->padded = (char *)dev - (char *)p;
6765 dev->pcpu_refcnt = alloc_percpu(int);
6766 if (!dev->pcpu_refcnt)
6769 if (dev_addr_init(dev))
6775 dev_net_set(dev, &init_net);
6777 dev->gso_max_size = GSO_MAX_SIZE;
6778 dev->gso_max_segs = GSO_MAX_SEGS;
6779 dev->gso_min_segs = 0;
6781 INIT_LIST_HEAD(&dev->napi_list);
6782 INIT_LIST_HEAD(&dev->unreg_list);
6783 INIT_LIST_HEAD(&dev->close_list);
6784 INIT_LIST_HEAD(&dev->link_watch_list);
6785 INIT_LIST_HEAD(&dev->adj_list.upper);
6786 INIT_LIST_HEAD(&dev->adj_list.lower);
6787 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6788 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6789 INIT_LIST_HEAD(&dev->ptype_all);
6790 INIT_LIST_HEAD(&dev->ptype_specific);
6791 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
6794 dev->num_tx_queues = txqs;
6795 dev->real_num_tx_queues = txqs;
6796 if (netif_alloc_netdev_queues(dev))
6800 dev->num_rx_queues = rxqs;
6801 dev->real_num_rx_queues = rxqs;
6802 if (netif_alloc_rx_queues(dev))
6806 strcpy(dev->name, name);
6807 dev->name_assign_type = name_assign_type;
6808 dev->group = INIT_NETDEV_GROUP;
6809 if (!dev->ethtool_ops)
6810 dev->ethtool_ops = &default_ethtool_ops;
6818 free_percpu(dev->pcpu_refcnt);
6820 netdev_freemem(dev);
6823 EXPORT_SYMBOL(alloc_netdev_mqs);
6826 * free_netdev - free network device
6829 * This function does the last stage of destroying an allocated device
6830 * interface. The reference to the device object is released.
6831 * If this is the last reference then it will be freed.
6833 void free_netdev(struct net_device *dev)
6835 struct napi_struct *p, *n;
6837 release_net(dev_net(dev));
6839 netif_free_tx_queues(dev);
6844 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6846 /* Flush device addresses */
6847 dev_addr_flush(dev);
6849 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6852 free_percpu(dev->pcpu_refcnt);
6853 dev->pcpu_refcnt = NULL;
6855 /* Compatibility with error handling in drivers */
6856 if (dev->reg_state == NETREG_UNINITIALIZED) {
6857 netdev_freemem(dev);
6861 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6862 dev->reg_state = NETREG_RELEASED;
6864 /* will free via device release */
6865 put_device(&dev->dev);
6867 EXPORT_SYMBOL(free_netdev);
6870 * synchronize_net - Synchronize with packet receive processing
6872 * Wait for packets currently being received to be done.
6873 * Does not block later packets from starting.
6875 void synchronize_net(void)
6878 if (rtnl_is_locked())
6879 synchronize_rcu_expedited();
6883 EXPORT_SYMBOL(synchronize_net);
6886 * unregister_netdevice_queue - remove device from the kernel
6890 * This function shuts down a device interface and removes it
6891 * from the kernel tables.
6892 * If head not NULL, device is queued to be unregistered later.
6894 * Callers must hold the rtnl semaphore. You may want
6895 * unregister_netdev() instead of this.
6898 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6903 list_move_tail(&dev->unreg_list, head);
6905 rollback_registered(dev);
6906 /* Finish processing unregister after unlock */
6910 EXPORT_SYMBOL(unregister_netdevice_queue);
6913 * unregister_netdevice_many - unregister many devices
6914 * @head: list of devices
6916 * Note: As most callers use a stack allocated list_head,
6917 * we force a list_del() to make sure stack wont be corrupted later.
6919 void unregister_netdevice_many(struct list_head *head)
6921 struct net_device *dev;
6923 if (!list_empty(head)) {
6924 rollback_registered_many(head);
6925 list_for_each_entry(dev, head, unreg_list)
6930 EXPORT_SYMBOL(unregister_netdevice_many);
6933 * unregister_netdev - remove device from the kernel
6936 * This function shuts down a device interface and removes it
6937 * from the kernel tables.
6939 * This is just a wrapper for unregister_netdevice that takes
6940 * the rtnl semaphore. In general you want to use this and not
6941 * unregister_netdevice.
6943 void unregister_netdev(struct net_device *dev)
6946 unregister_netdevice(dev);
6949 EXPORT_SYMBOL(unregister_netdev);
6952 * dev_change_net_namespace - move device to different nethost namespace
6954 * @net: network namespace
6955 * @pat: If not NULL name pattern to try if the current device name
6956 * is already taken in the destination network namespace.
6958 * This function shuts down a device interface and moves it
6959 * to a new network namespace. On success 0 is returned, on
6960 * a failure a netagive errno code is returned.
6962 * Callers must hold the rtnl semaphore.
6965 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6971 /* Don't allow namespace local devices to be moved. */
6973 if (dev->features & NETIF_F_NETNS_LOCAL)
6976 /* Ensure the device has been registrered */
6977 if (dev->reg_state != NETREG_REGISTERED)
6980 /* Get out if there is nothing todo */
6982 if (net_eq(dev_net(dev), net))
6985 /* Pick the destination device name, and ensure
6986 * we can use it in the destination network namespace.
6989 if (__dev_get_by_name(net, dev->name)) {
6990 /* We get here if we can't use the current device name */
6993 if (dev_get_valid_name(net, dev, pat) < 0)
6998 * And now a mini version of register_netdevice unregister_netdevice.
7001 /* If device is running close it first. */
7004 /* And unlink it from device chain */
7006 unlist_netdevice(dev);
7010 /* Shutdown queueing discipline. */
7013 /* Notify protocols, that we are about to destroy
7014 this device. They should clean all the things.
7016 Note that dev->reg_state stays at NETREG_REGISTERED.
7017 This is wanted because this way 8021q and macvlan know
7018 the device is just moving and can keep their slaves up.
7020 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7022 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7023 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7026 * Flush the unicast and multicast chains
7031 /* Send a netdev-removed uevent to the old namespace */
7032 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7033 netdev_adjacent_del_links(dev);
7035 /* Actually switch the network namespace */
7036 dev_net_set(dev, net);
7038 /* If there is an ifindex conflict assign a new one */
7039 if (__dev_get_by_index(net, dev->ifindex)) {
7040 int iflink = (dev->iflink == dev->ifindex);
7041 dev->ifindex = dev_new_index(net);
7043 dev->iflink = dev->ifindex;
7046 /* Send a netdev-add uevent to the new namespace */
7047 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7048 netdev_adjacent_add_links(dev);
7050 /* Fixup kobjects */
7051 err = device_rename(&dev->dev, dev->name);
7054 /* Add the device back in the hashes */
7055 list_netdevice(dev);
7057 /* Notify protocols, that a new device appeared. */
7058 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7061 * Prevent userspace races by waiting until the network
7062 * device is fully setup before sending notifications.
7064 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7071 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7073 static int dev_cpu_callback(struct notifier_block *nfb,
7074 unsigned long action,
7077 struct sk_buff **list_skb;
7078 struct sk_buff *skb;
7079 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7080 struct softnet_data *sd, *oldsd;
7082 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7085 local_irq_disable();
7086 cpu = smp_processor_id();
7087 sd = &per_cpu(softnet_data, cpu);
7088 oldsd = &per_cpu(softnet_data, oldcpu);
7090 /* Find end of our completion_queue. */
7091 list_skb = &sd->completion_queue;
7093 list_skb = &(*list_skb)->next;
7094 /* Append completion queue from offline CPU. */
7095 *list_skb = oldsd->completion_queue;
7096 oldsd->completion_queue = NULL;
7098 /* Append output queue from offline CPU. */
7099 if (oldsd->output_queue) {
7100 *sd->output_queue_tailp = oldsd->output_queue;
7101 sd->output_queue_tailp = oldsd->output_queue_tailp;
7102 oldsd->output_queue = NULL;
7103 oldsd->output_queue_tailp = &oldsd->output_queue;
7105 /* Append NAPI poll list from offline CPU, with one exception :
7106 * process_backlog() must be called by cpu owning percpu backlog.
7107 * We properly handle process_queue & input_pkt_queue later.
7109 while (!list_empty(&oldsd->poll_list)) {
7110 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7114 list_del_init(&napi->poll_list);
7115 if (napi->poll == process_backlog)
7118 ____napi_schedule(sd, napi);
7121 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7124 /* Process offline CPU's input_pkt_queue */
7125 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7126 netif_rx_internal(skb);
7127 input_queue_head_incr(oldsd);
7129 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7130 netif_rx_internal(skb);
7131 input_queue_head_incr(oldsd);
7139 * netdev_increment_features - increment feature set by one
7140 * @all: current feature set
7141 * @one: new feature set
7142 * @mask: mask feature set
7144 * Computes a new feature set after adding a device with feature set
7145 * @one to the master device with current feature set @all. Will not
7146 * enable anything that is off in @mask. Returns the new feature set.
7148 netdev_features_t netdev_increment_features(netdev_features_t all,
7149 netdev_features_t one, netdev_features_t mask)
7151 if (mask & NETIF_F_GEN_CSUM)
7152 mask |= NETIF_F_ALL_CSUM;
7153 mask |= NETIF_F_VLAN_CHALLENGED;
7155 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7156 all &= one | ~NETIF_F_ALL_FOR_ALL;
7158 /* If one device supports hw checksumming, set for all. */
7159 if (all & NETIF_F_GEN_CSUM)
7160 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7164 EXPORT_SYMBOL(netdev_increment_features);
7166 static struct hlist_head * __net_init netdev_create_hash(void)
7169 struct hlist_head *hash;
7171 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7173 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7174 INIT_HLIST_HEAD(&hash[i]);
7179 /* Initialize per network namespace state */
7180 static int __net_init netdev_init(struct net *net)
7182 if (net != &init_net)
7183 INIT_LIST_HEAD(&net->dev_base_head);
7185 net->dev_name_head = netdev_create_hash();
7186 if (net->dev_name_head == NULL)
7189 net->dev_index_head = netdev_create_hash();
7190 if (net->dev_index_head == NULL)
7196 kfree(net->dev_name_head);
7202 * netdev_drivername - network driver for the device
7203 * @dev: network device
7205 * Determine network driver for device.
7207 const char *netdev_drivername(const struct net_device *dev)
7209 const struct device_driver *driver;
7210 const struct device *parent;
7211 const char *empty = "";
7213 parent = dev->dev.parent;
7217 driver = parent->driver;
7218 if (driver && driver->name)
7219 return driver->name;
7223 static void __netdev_printk(const char *level, const struct net_device *dev,
7224 struct va_format *vaf)
7226 if (dev && dev->dev.parent) {
7227 dev_printk_emit(level[1] - '0',
7230 dev_driver_string(dev->dev.parent),
7231 dev_name(dev->dev.parent),
7232 netdev_name(dev), netdev_reg_state(dev),
7235 printk("%s%s%s: %pV",
7236 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7238 printk("%s(NULL net_device): %pV", level, vaf);
7242 void netdev_printk(const char *level, const struct net_device *dev,
7243 const char *format, ...)
7245 struct va_format vaf;
7248 va_start(args, format);
7253 __netdev_printk(level, dev, &vaf);
7257 EXPORT_SYMBOL(netdev_printk);
7259 #define define_netdev_printk_level(func, level) \
7260 void func(const struct net_device *dev, const char *fmt, ...) \
7262 struct va_format vaf; \
7265 va_start(args, fmt); \
7270 __netdev_printk(level, dev, &vaf); \
7274 EXPORT_SYMBOL(func);
7276 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7277 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7278 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7279 define_netdev_printk_level(netdev_err, KERN_ERR);
7280 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7281 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7282 define_netdev_printk_level(netdev_info, KERN_INFO);
7284 static void __net_exit netdev_exit(struct net *net)
7286 kfree(net->dev_name_head);
7287 kfree(net->dev_index_head);
7290 static struct pernet_operations __net_initdata netdev_net_ops = {
7291 .init = netdev_init,
7292 .exit = netdev_exit,
7295 static void __net_exit default_device_exit(struct net *net)
7297 struct net_device *dev, *aux;
7299 * Push all migratable network devices back to the
7300 * initial network namespace
7303 for_each_netdev_safe(net, dev, aux) {
7305 char fb_name[IFNAMSIZ];
7307 /* Ignore unmoveable devices (i.e. loopback) */
7308 if (dev->features & NETIF_F_NETNS_LOCAL)
7311 /* Leave virtual devices for the generic cleanup */
7312 if (dev->rtnl_link_ops)
7315 /* Push remaining network devices to init_net */
7316 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7317 err = dev_change_net_namespace(dev, &init_net, fb_name);
7319 pr_emerg("%s: failed to move %s to init_net: %d\n",
7320 __func__, dev->name, err);
7327 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7329 /* Return with the rtnl_lock held when there are no network
7330 * devices unregistering in any network namespace in net_list.
7334 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7336 add_wait_queue(&netdev_unregistering_wq, &wait);
7338 unregistering = false;
7340 list_for_each_entry(net, net_list, exit_list) {
7341 if (net->dev_unreg_count > 0) {
7342 unregistering = true;
7350 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7352 remove_wait_queue(&netdev_unregistering_wq, &wait);
7355 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7357 /* At exit all network devices most be removed from a network
7358 * namespace. Do this in the reverse order of registration.
7359 * Do this across as many network namespaces as possible to
7360 * improve batching efficiency.
7362 struct net_device *dev;
7364 LIST_HEAD(dev_kill_list);
7366 /* To prevent network device cleanup code from dereferencing
7367 * loopback devices or network devices that have been freed
7368 * wait here for all pending unregistrations to complete,
7369 * before unregistring the loopback device and allowing the
7370 * network namespace be freed.
7372 * The netdev todo list containing all network devices
7373 * unregistrations that happen in default_device_exit_batch
7374 * will run in the rtnl_unlock() at the end of
7375 * default_device_exit_batch.
7377 rtnl_lock_unregistering(net_list);
7378 list_for_each_entry(net, net_list, exit_list) {
7379 for_each_netdev_reverse(net, dev) {
7380 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7381 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7383 unregister_netdevice_queue(dev, &dev_kill_list);
7386 unregister_netdevice_many(&dev_kill_list);
7390 static struct pernet_operations __net_initdata default_device_ops = {
7391 .exit = default_device_exit,
7392 .exit_batch = default_device_exit_batch,
7396 * Initialize the DEV module. At boot time this walks the device list and
7397 * unhooks any devices that fail to initialise (normally hardware not
7398 * present) and leaves us with a valid list of present and active devices.
7403 * This is called single threaded during boot, so no need
7404 * to take the rtnl semaphore.
7406 static int __init net_dev_init(void)
7408 int i, rc = -ENOMEM;
7410 BUG_ON(!dev_boot_phase);
7412 if (dev_proc_init())
7415 if (netdev_kobject_init())
7418 INIT_LIST_HEAD(&ptype_all);
7419 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7420 INIT_LIST_HEAD(&ptype_base[i]);
7422 INIT_LIST_HEAD(&offload_base);
7424 if (register_pernet_subsys(&netdev_net_ops))
7428 * Initialise the packet receive queues.
7431 for_each_possible_cpu(i) {
7432 struct softnet_data *sd = &per_cpu(softnet_data, i);
7434 skb_queue_head_init(&sd->input_pkt_queue);
7435 skb_queue_head_init(&sd->process_queue);
7436 INIT_LIST_HEAD(&sd->poll_list);
7437 sd->output_queue_tailp = &sd->output_queue;
7439 sd->csd.func = rps_trigger_softirq;
7444 sd->backlog.poll = process_backlog;
7445 sd->backlog.weight = weight_p;
7450 /* The loopback device is special if any other network devices
7451 * is present in a network namespace the loopback device must
7452 * be present. Since we now dynamically allocate and free the
7453 * loopback device ensure this invariant is maintained by
7454 * keeping the loopback device as the first device on the
7455 * list of network devices. Ensuring the loopback devices
7456 * is the first device that appears and the last network device
7459 if (register_pernet_device(&loopback_net_ops))
7462 if (register_pernet_device(&default_device_ops))
7465 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7466 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7468 hotcpu_notifier(dev_cpu_callback, 0);
7475 subsys_initcall(net_dev_init);