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);
3033 u32 rps_cpu_mask __read_mostly;
3034 EXPORT_SYMBOL(rps_cpu_mask);
3036 struct static_key rps_needed __read_mostly;
3038 static struct rps_dev_flow *
3039 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3040 struct rps_dev_flow *rflow, u16 next_cpu)
3042 if (next_cpu != RPS_NO_CPU) {
3043 #ifdef CONFIG_RFS_ACCEL
3044 struct netdev_rx_queue *rxqueue;
3045 struct rps_dev_flow_table *flow_table;
3046 struct rps_dev_flow *old_rflow;
3051 /* Should we steer this flow to a different hardware queue? */
3052 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3053 !(dev->features & NETIF_F_NTUPLE))
3055 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3056 if (rxq_index == skb_get_rx_queue(skb))
3059 rxqueue = dev->_rx + rxq_index;
3060 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3063 flow_id = skb_get_hash(skb) & flow_table->mask;
3064 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3065 rxq_index, flow_id);
3069 rflow = &flow_table->flows[flow_id];
3071 if (old_rflow->filter == rflow->filter)
3072 old_rflow->filter = RPS_NO_FILTER;
3076 per_cpu(softnet_data, next_cpu).input_queue_head;
3079 rflow->cpu = next_cpu;
3084 * get_rps_cpu is called from netif_receive_skb and returns the target
3085 * CPU from the RPS map of the receiving queue for a given skb.
3086 * rcu_read_lock must be held on entry.
3088 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3089 struct rps_dev_flow **rflowp)
3091 const struct rps_sock_flow_table *sock_flow_table;
3092 struct netdev_rx_queue *rxqueue = dev->_rx;
3093 struct rps_dev_flow_table *flow_table;
3094 struct rps_map *map;
3099 if (skb_rx_queue_recorded(skb)) {
3100 u16 index = skb_get_rx_queue(skb);
3102 if (unlikely(index >= dev->real_num_rx_queues)) {
3103 WARN_ONCE(dev->real_num_rx_queues > 1,
3104 "%s received packet on queue %u, but number "
3105 "of RX queues is %u\n",
3106 dev->name, index, dev->real_num_rx_queues);
3112 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3114 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3115 map = rcu_dereference(rxqueue->rps_map);
3116 if (!flow_table && !map)
3119 skb_reset_network_header(skb);
3120 hash = skb_get_hash(skb);
3124 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3125 if (flow_table && sock_flow_table) {
3126 struct rps_dev_flow *rflow;
3130 /* First check into global flow table if there is a match */
3131 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3132 if ((ident ^ hash) & ~rps_cpu_mask)
3135 next_cpu = ident & rps_cpu_mask;
3137 /* OK, now we know there is a match,
3138 * we can look at the local (per receive queue) flow table
3140 rflow = &flow_table->flows[hash & flow_table->mask];
3144 * If the desired CPU (where last recvmsg was done) is
3145 * different from current CPU (one in the rx-queue flow
3146 * table entry), switch if one of the following holds:
3147 * - Current CPU is unset (equal to RPS_NO_CPU).
3148 * - Current CPU is offline.
3149 * - The current CPU's queue tail has advanced beyond the
3150 * last packet that was enqueued using this table entry.
3151 * This guarantees that all previous packets for the flow
3152 * have been dequeued, thus preserving in order delivery.
3154 if (unlikely(tcpu != next_cpu) &&
3155 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3156 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3157 rflow->last_qtail)) >= 0)) {
3159 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3162 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3172 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3173 if (cpu_online(tcpu)) {
3183 #ifdef CONFIG_RFS_ACCEL
3186 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3187 * @dev: Device on which the filter was set
3188 * @rxq_index: RX queue index
3189 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3190 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3192 * Drivers that implement ndo_rx_flow_steer() should periodically call
3193 * this function for each installed filter and remove the filters for
3194 * which it returns %true.
3196 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3197 u32 flow_id, u16 filter_id)
3199 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3200 struct rps_dev_flow_table *flow_table;
3201 struct rps_dev_flow *rflow;
3206 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3207 if (flow_table && flow_id <= flow_table->mask) {
3208 rflow = &flow_table->flows[flow_id];
3209 cpu = ACCESS_ONCE(rflow->cpu);
3210 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3211 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3212 rflow->last_qtail) <
3213 (int)(10 * flow_table->mask)))
3219 EXPORT_SYMBOL(rps_may_expire_flow);
3221 #endif /* CONFIG_RFS_ACCEL */
3223 /* Called from hardirq (IPI) context */
3224 static void rps_trigger_softirq(void *data)
3226 struct softnet_data *sd = data;
3228 ____napi_schedule(sd, &sd->backlog);
3232 #endif /* CONFIG_RPS */
3235 * Check if this softnet_data structure is another cpu one
3236 * If yes, queue it to our IPI list and return 1
3239 static int rps_ipi_queued(struct softnet_data *sd)
3242 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3245 sd->rps_ipi_next = mysd->rps_ipi_list;
3246 mysd->rps_ipi_list = sd;
3248 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3251 #endif /* CONFIG_RPS */
3255 #ifdef CONFIG_NET_FLOW_LIMIT
3256 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3259 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3261 #ifdef CONFIG_NET_FLOW_LIMIT
3262 struct sd_flow_limit *fl;
3263 struct softnet_data *sd;
3264 unsigned int old_flow, new_flow;
3266 if (qlen < (netdev_max_backlog >> 1))
3269 sd = this_cpu_ptr(&softnet_data);
3272 fl = rcu_dereference(sd->flow_limit);
3274 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3275 old_flow = fl->history[fl->history_head];
3276 fl->history[fl->history_head] = new_flow;
3279 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3281 if (likely(fl->buckets[old_flow]))
3282 fl->buckets[old_flow]--;
3284 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3296 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3297 * queue (may be a remote CPU queue).
3299 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3300 unsigned int *qtail)
3302 struct softnet_data *sd;
3303 unsigned long flags;
3306 sd = &per_cpu(softnet_data, cpu);
3308 local_irq_save(flags);
3311 qlen = skb_queue_len(&sd->input_pkt_queue);
3312 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3315 __skb_queue_tail(&sd->input_pkt_queue, skb);
3316 input_queue_tail_incr_save(sd, qtail);
3318 local_irq_restore(flags);
3319 return NET_RX_SUCCESS;
3322 /* Schedule NAPI for backlog device
3323 * We can use non atomic operation since we own the queue lock
3325 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3326 if (!rps_ipi_queued(sd))
3327 ____napi_schedule(sd, &sd->backlog);
3335 local_irq_restore(flags);
3337 atomic_long_inc(&skb->dev->rx_dropped);
3342 static int netif_rx_internal(struct sk_buff *skb)
3346 net_timestamp_check(netdev_tstamp_prequeue, skb);
3348 trace_netif_rx(skb);
3350 if (static_key_false(&rps_needed)) {
3351 struct rps_dev_flow voidflow, *rflow = &voidflow;
3357 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3359 cpu = smp_processor_id();
3361 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3369 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3376 * netif_rx - post buffer to the network code
3377 * @skb: buffer to post
3379 * This function receives a packet from a device driver and queues it for
3380 * the upper (protocol) levels to process. It always succeeds. The buffer
3381 * may be dropped during processing for congestion control or by the
3385 * NET_RX_SUCCESS (no congestion)
3386 * NET_RX_DROP (packet was dropped)
3390 int netif_rx(struct sk_buff *skb)
3392 trace_netif_rx_entry(skb);
3394 return netif_rx_internal(skb);
3396 EXPORT_SYMBOL(netif_rx);
3398 int netif_rx_ni(struct sk_buff *skb)
3402 trace_netif_rx_ni_entry(skb);
3405 err = netif_rx_internal(skb);
3406 if (local_softirq_pending())
3412 EXPORT_SYMBOL(netif_rx_ni);
3414 static void net_tx_action(struct softirq_action *h)
3416 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3418 if (sd->completion_queue) {
3419 struct sk_buff *clist;
3421 local_irq_disable();
3422 clist = sd->completion_queue;
3423 sd->completion_queue = NULL;
3427 struct sk_buff *skb = clist;
3428 clist = clist->next;
3430 WARN_ON(atomic_read(&skb->users));
3431 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3432 trace_consume_skb(skb);
3434 trace_kfree_skb(skb, net_tx_action);
3439 if (sd->output_queue) {
3442 local_irq_disable();
3443 head = sd->output_queue;
3444 sd->output_queue = NULL;
3445 sd->output_queue_tailp = &sd->output_queue;
3449 struct Qdisc *q = head;
3450 spinlock_t *root_lock;
3452 head = head->next_sched;
3454 root_lock = qdisc_lock(q);
3455 if (spin_trylock(root_lock)) {
3456 smp_mb__before_atomic();
3457 clear_bit(__QDISC_STATE_SCHED,
3460 spin_unlock(root_lock);
3462 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3464 __netif_reschedule(q);
3466 smp_mb__before_atomic();
3467 clear_bit(__QDISC_STATE_SCHED,
3475 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3476 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3477 /* This hook is defined here for ATM LANE */
3478 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3479 unsigned char *addr) __read_mostly;
3480 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3483 #ifdef CONFIG_NET_CLS_ACT
3484 /* TODO: Maybe we should just force sch_ingress to be compiled in
3485 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3486 * a compare and 2 stores extra right now if we dont have it on
3487 * but have CONFIG_NET_CLS_ACT
3488 * NOTE: This doesn't stop any functionality; if you dont have
3489 * the ingress scheduler, you just can't add policies on ingress.
3492 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3494 struct net_device *dev = skb->dev;
3495 u32 ttl = G_TC_RTTL(skb->tc_verd);
3496 int result = TC_ACT_OK;
3499 if (unlikely(MAX_RED_LOOP < ttl++)) {
3500 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3501 skb->skb_iif, dev->ifindex);
3505 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3506 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3508 q = rcu_dereference(rxq->qdisc);
3509 if (q != &noop_qdisc) {
3510 spin_lock(qdisc_lock(q));
3511 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3512 result = qdisc_enqueue_root(skb, q);
3513 spin_unlock(qdisc_lock(q));
3519 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3520 struct packet_type **pt_prev,
3521 int *ret, struct net_device *orig_dev)
3523 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3525 if (!rxq || rcu_access_pointer(rxq->qdisc) == &noop_qdisc)
3529 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3533 switch (ing_filter(skb, rxq)) {
3547 * netdev_rx_handler_register - register receive handler
3548 * @dev: device to register a handler for
3549 * @rx_handler: receive handler to register
3550 * @rx_handler_data: data pointer that is used by rx handler
3552 * Register a receive handler for a device. This handler will then be
3553 * called from __netif_receive_skb. A negative errno code is returned
3556 * The caller must hold the rtnl_mutex.
3558 * For a general description of rx_handler, see enum rx_handler_result.
3560 int netdev_rx_handler_register(struct net_device *dev,
3561 rx_handler_func_t *rx_handler,
3562 void *rx_handler_data)
3566 if (dev->rx_handler)
3569 /* Note: rx_handler_data must be set before rx_handler */
3570 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3571 rcu_assign_pointer(dev->rx_handler, rx_handler);
3575 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3578 * netdev_rx_handler_unregister - unregister receive handler
3579 * @dev: device to unregister a handler from
3581 * Unregister a receive handler from a device.
3583 * The caller must hold the rtnl_mutex.
3585 void netdev_rx_handler_unregister(struct net_device *dev)
3589 RCU_INIT_POINTER(dev->rx_handler, NULL);
3590 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3591 * section has a guarantee to see a non NULL rx_handler_data
3595 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3597 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3600 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3601 * the special handling of PFMEMALLOC skbs.
3603 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3605 switch (skb->protocol) {
3606 case htons(ETH_P_ARP):
3607 case htons(ETH_P_IP):
3608 case htons(ETH_P_IPV6):
3609 case htons(ETH_P_8021Q):
3610 case htons(ETH_P_8021AD):
3617 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3619 struct packet_type *ptype, *pt_prev;
3620 rx_handler_func_t *rx_handler;
3621 struct net_device *orig_dev;
3622 bool deliver_exact = false;
3623 int ret = NET_RX_DROP;
3626 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3628 trace_netif_receive_skb(skb);
3630 orig_dev = skb->dev;
3632 skb_reset_network_header(skb);
3633 if (!skb_transport_header_was_set(skb))
3634 skb_reset_transport_header(skb);
3635 skb_reset_mac_len(skb);
3642 skb->skb_iif = skb->dev->ifindex;
3644 __this_cpu_inc(softnet_data.processed);
3646 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3647 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3648 skb = skb_vlan_untag(skb);
3653 #ifdef CONFIG_NET_CLS_ACT
3654 if (skb->tc_verd & TC_NCLS) {
3655 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3663 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3665 ret = deliver_skb(skb, pt_prev, orig_dev);
3669 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3671 ret = deliver_skb(skb, pt_prev, orig_dev);
3676 #ifdef CONFIG_NET_CLS_ACT
3677 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3683 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3686 if (skb_vlan_tag_present(skb)) {
3688 ret = deliver_skb(skb, pt_prev, orig_dev);
3691 if (vlan_do_receive(&skb))
3693 else if (unlikely(!skb))
3697 rx_handler = rcu_dereference(skb->dev->rx_handler);
3700 ret = deliver_skb(skb, pt_prev, orig_dev);
3703 switch (rx_handler(&skb)) {
3704 case RX_HANDLER_CONSUMED:
3705 ret = NET_RX_SUCCESS;
3707 case RX_HANDLER_ANOTHER:
3709 case RX_HANDLER_EXACT:
3710 deliver_exact = true;
3711 case RX_HANDLER_PASS:
3718 if (unlikely(skb_vlan_tag_present(skb))) {
3719 if (skb_vlan_tag_get_id(skb))
3720 skb->pkt_type = PACKET_OTHERHOST;
3721 /* Note: we might in the future use prio bits
3722 * and set skb->priority like in vlan_do_receive()
3723 * For the time being, just ignore Priority Code Point
3728 type = skb->protocol;
3730 /* deliver only exact match when indicated */
3731 if (likely(!deliver_exact)) {
3732 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3733 &ptype_base[ntohs(type) &
3737 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3738 &orig_dev->ptype_specific);
3740 if (unlikely(skb->dev != orig_dev)) {
3741 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3742 &skb->dev->ptype_specific);
3746 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3749 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3752 atomic_long_inc(&skb->dev->rx_dropped);
3754 /* Jamal, now you will not able to escape explaining
3755 * me how you were going to use this. :-)
3765 static int __netif_receive_skb(struct sk_buff *skb)
3769 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3770 unsigned long pflags = current->flags;
3773 * PFMEMALLOC skbs are special, they should
3774 * - be delivered to SOCK_MEMALLOC sockets only
3775 * - stay away from userspace
3776 * - have bounded memory usage
3778 * Use PF_MEMALLOC as this saves us from propagating the allocation
3779 * context down to all allocation sites.
3781 current->flags |= PF_MEMALLOC;
3782 ret = __netif_receive_skb_core(skb, true);
3783 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3785 ret = __netif_receive_skb_core(skb, false);
3790 static int netif_receive_skb_internal(struct sk_buff *skb)
3792 net_timestamp_check(netdev_tstamp_prequeue, skb);
3794 if (skb_defer_rx_timestamp(skb))
3795 return NET_RX_SUCCESS;
3798 if (static_key_false(&rps_needed)) {
3799 struct rps_dev_flow voidflow, *rflow = &voidflow;
3804 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3807 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3814 return __netif_receive_skb(skb);
3818 * netif_receive_skb - process receive buffer from network
3819 * @skb: buffer to process
3821 * netif_receive_skb() is the main receive data processing function.
3822 * It always succeeds. The buffer may be dropped during processing
3823 * for congestion control or by the protocol layers.
3825 * This function may only be called from softirq context and interrupts
3826 * should be enabled.
3828 * Return values (usually ignored):
3829 * NET_RX_SUCCESS: no congestion
3830 * NET_RX_DROP: packet was dropped
3832 int netif_receive_skb(struct sk_buff *skb)
3834 trace_netif_receive_skb_entry(skb);
3836 return netif_receive_skb_internal(skb);
3838 EXPORT_SYMBOL(netif_receive_skb);
3840 /* Network device is going away, flush any packets still pending
3841 * Called with irqs disabled.
3843 static void flush_backlog(void *arg)
3845 struct net_device *dev = arg;
3846 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3847 struct sk_buff *skb, *tmp;
3850 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3851 if (skb->dev == dev) {
3852 __skb_unlink(skb, &sd->input_pkt_queue);
3854 input_queue_head_incr(sd);
3859 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3860 if (skb->dev == dev) {
3861 __skb_unlink(skb, &sd->process_queue);
3863 input_queue_head_incr(sd);
3868 static int napi_gro_complete(struct sk_buff *skb)
3870 struct packet_offload *ptype;
3871 __be16 type = skb->protocol;
3872 struct list_head *head = &offload_base;
3875 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3877 if (NAPI_GRO_CB(skb)->count == 1) {
3878 skb_shinfo(skb)->gso_size = 0;
3883 list_for_each_entry_rcu(ptype, head, list) {
3884 if (ptype->type != type || !ptype->callbacks.gro_complete)
3887 err = ptype->callbacks.gro_complete(skb, 0);
3893 WARN_ON(&ptype->list == head);
3895 return NET_RX_SUCCESS;
3899 return netif_receive_skb_internal(skb);
3902 /* napi->gro_list contains packets ordered by age.
3903 * youngest packets at the head of it.
3904 * Complete skbs in reverse order to reduce latencies.
3906 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3908 struct sk_buff *skb, *prev = NULL;
3910 /* scan list and build reverse chain */
3911 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3916 for (skb = prev; skb; skb = prev) {
3919 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3923 napi_gro_complete(skb);
3927 napi->gro_list = NULL;
3929 EXPORT_SYMBOL(napi_gro_flush);
3931 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3934 unsigned int maclen = skb->dev->hard_header_len;
3935 u32 hash = skb_get_hash_raw(skb);
3937 for (p = napi->gro_list; p; p = p->next) {
3938 unsigned long diffs;
3940 NAPI_GRO_CB(p)->flush = 0;
3942 if (hash != skb_get_hash_raw(p)) {
3943 NAPI_GRO_CB(p)->same_flow = 0;
3947 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3948 diffs |= p->vlan_tci ^ skb->vlan_tci;
3949 if (maclen == ETH_HLEN)
3950 diffs |= compare_ether_header(skb_mac_header(p),
3951 skb_mac_header(skb));
3953 diffs = memcmp(skb_mac_header(p),
3954 skb_mac_header(skb),
3956 NAPI_GRO_CB(p)->same_flow = !diffs;
3960 static void skb_gro_reset_offset(struct sk_buff *skb)
3962 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3963 const skb_frag_t *frag0 = &pinfo->frags[0];
3965 NAPI_GRO_CB(skb)->data_offset = 0;
3966 NAPI_GRO_CB(skb)->frag0 = NULL;
3967 NAPI_GRO_CB(skb)->frag0_len = 0;
3969 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3971 !PageHighMem(skb_frag_page(frag0))) {
3972 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3973 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3977 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
3979 struct skb_shared_info *pinfo = skb_shinfo(skb);
3981 BUG_ON(skb->end - skb->tail < grow);
3983 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3985 skb->data_len -= grow;
3988 pinfo->frags[0].page_offset += grow;
3989 skb_frag_size_sub(&pinfo->frags[0], grow);
3991 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
3992 skb_frag_unref(skb, 0);
3993 memmove(pinfo->frags, pinfo->frags + 1,
3994 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
3998 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4000 struct sk_buff **pp = NULL;
4001 struct packet_offload *ptype;
4002 __be16 type = skb->protocol;
4003 struct list_head *head = &offload_base;
4005 enum gro_result ret;
4008 if (!(skb->dev->features & NETIF_F_GRO))
4011 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4014 gro_list_prepare(napi, skb);
4017 list_for_each_entry_rcu(ptype, head, list) {
4018 if (ptype->type != type || !ptype->callbacks.gro_receive)
4021 skb_set_network_header(skb, skb_gro_offset(skb));
4022 skb_reset_mac_len(skb);
4023 NAPI_GRO_CB(skb)->same_flow = 0;
4024 NAPI_GRO_CB(skb)->flush = 0;
4025 NAPI_GRO_CB(skb)->free = 0;
4026 NAPI_GRO_CB(skb)->udp_mark = 0;
4028 /* Setup for GRO checksum validation */
4029 switch (skb->ip_summed) {
4030 case CHECKSUM_COMPLETE:
4031 NAPI_GRO_CB(skb)->csum = skb->csum;
4032 NAPI_GRO_CB(skb)->csum_valid = 1;
4033 NAPI_GRO_CB(skb)->csum_cnt = 0;
4035 case CHECKSUM_UNNECESSARY:
4036 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4037 NAPI_GRO_CB(skb)->csum_valid = 0;
4040 NAPI_GRO_CB(skb)->csum_cnt = 0;
4041 NAPI_GRO_CB(skb)->csum_valid = 0;
4044 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4049 if (&ptype->list == head)
4052 same_flow = NAPI_GRO_CB(skb)->same_flow;
4053 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4056 struct sk_buff *nskb = *pp;
4060 napi_gro_complete(nskb);
4067 if (NAPI_GRO_CB(skb)->flush)
4070 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4071 struct sk_buff *nskb = napi->gro_list;
4073 /* locate the end of the list to select the 'oldest' flow */
4074 while (nskb->next) {
4080 napi_gro_complete(nskb);
4084 NAPI_GRO_CB(skb)->count = 1;
4085 NAPI_GRO_CB(skb)->age = jiffies;
4086 NAPI_GRO_CB(skb)->last = skb;
4087 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4088 skb->next = napi->gro_list;
4089 napi->gro_list = skb;
4093 grow = skb_gro_offset(skb) - skb_headlen(skb);
4095 gro_pull_from_frag0(skb, grow);
4104 struct packet_offload *gro_find_receive_by_type(__be16 type)
4106 struct list_head *offload_head = &offload_base;
4107 struct packet_offload *ptype;
4109 list_for_each_entry_rcu(ptype, offload_head, list) {
4110 if (ptype->type != type || !ptype->callbacks.gro_receive)
4116 EXPORT_SYMBOL(gro_find_receive_by_type);
4118 struct packet_offload *gro_find_complete_by_type(__be16 type)
4120 struct list_head *offload_head = &offload_base;
4121 struct packet_offload *ptype;
4123 list_for_each_entry_rcu(ptype, offload_head, list) {
4124 if (ptype->type != type || !ptype->callbacks.gro_complete)
4130 EXPORT_SYMBOL(gro_find_complete_by_type);
4132 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4136 if (netif_receive_skb_internal(skb))
4144 case GRO_MERGED_FREE:
4145 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4146 kmem_cache_free(skbuff_head_cache, skb);
4159 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4161 trace_napi_gro_receive_entry(skb);
4163 skb_gro_reset_offset(skb);
4165 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4167 EXPORT_SYMBOL(napi_gro_receive);
4169 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4171 if (unlikely(skb->pfmemalloc)) {
4175 __skb_pull(skb, skb_headlen(skb));
4176 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4177 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4179 skb->dev = napi->dev;
4181 skb->encapsulation = 0;
4182 skb_shinfo(skb)->gso_type = 0;
4183 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4188 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4190 struct sk_buff *skb = napi->skb;
4193 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4198 EXPORT_SYMBOL(napi_get_frags);
4200 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4201 struct sk_buff *skb,
4207 __skb_push(skb, ETH_HLEN);
4208 skb->protocol = eth_type_trans(skb, skb->dev);
4209 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4214 case GRO_MERGED_FREE:
4215 napi_reuse_skb(napi, skb);
4225 /* Upper GRO stack assumes network header starts at gro_offset=0
4226 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4227 * We copy ethernet header into skb->data to have a common layout.
4229 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4231 struct sk_buff *skb = napi->skb;
4232 const struct ethhdr *eth;
4233 unsigned int hlen = sizeof(*eth);
4237 skb_reset_mac_header(skb);
4238 skb_gro_reset_offset(skb);
4240 eth = skb_gro_header_fast(skb, 0);
4241 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4242 eth = skb_gro_header_slow(skb, hlen, 0);
4243 if (unlikely(!eth)) {
4244 napi_reuse_skb(napi, skb);
4248 gro_pull_from_frag0(skb, hlen);
4249 NAPI_GRO_CB(skb)->frag0 += hlen;
4250 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4252 __skb_pull(skb, hlen);
4255 * This works because the only protocols we care about don't require
4257 * We'll fix it up properly in napi_frags_finish()
4259 skb->protocol = eth->h_proto;
4264 gro_result_t napi_gro_frags(struct napi_struct *napi)
4266 struct sk_buff *skb = napi_frags_skb(napi);
4271 trace_napi_gro_frags_entry(skb);
4273 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4275 EXPORT_SYMBOL(napi_gro_frags);
4277 /* Compute the checksum from gro_offset and return the folded value
4278 * after adding in any pseudo checksum.
4280 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4285 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4287 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4288 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4290 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4291 !skb->csum_complete_sw)
4292 netdev_rx_csum_fault(skb->dev);
4295 NAPI_GRO_CB(skb)->csum = wsum;
4296 NAPI_GRO_CB(skb)->csum_valid = 1;
4300 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4303 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4304 * Note: called with local irq disabled, but exits with local irq enabled.
4306 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4309 struct softnet_data *remsd = sd->rps_ipi_list;
4312 sd->rps_ipi_list = NULL;
4316 /* Send pending IPI's to kick RPS processing on remote cpus. */
4318 struct softnet_data *next = remsd->rps_ipi_next;
4320 if (cpu_online(remsd->cpu))
4321 smp_call_function_single_async(remsd->cpu,
4330 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4333 return sd->rps_ipi_list != NULL;
4339 static int process_backlog(struct napi_struct *napi, int quota)
4342 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4344 /* Check if we have pending ipi, its better to send them now,
4345 * not waiting net_rx_action() end.
4347 if (sd_has_rps_ipi_waiting(sd)) {
4348 local_irq_disable();
4349 net_rps_action_and_irq_enable(sd);
4352 napi->weight = weight_p;
4353 local_irq_disable();
4355 struct sk_buff *skb;
4357 while ((skb = __skb_dequeue(&sd->process_queue))) {
4359 __netif_receive_skb(skb);
4360 local_irq_disable();
4361 input_queue_head_incr(sd);
4362 if (++work >= quota) {
4369 if (skb_queue_empty(&sd->input_pkt_queue)) {
4371 * Inline a custom version of __napi_complete().
4372 * only current cpu owns and manipulates this napi,
4373 * and NAPI_STATE_SCHED is the only possible flag set
4375 * We can use a plain write instead of clear_bit(),
4376 * and we dont need an smp_mb() memory barrier.
4384 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4385 &sd->process_queue);
4394 * __napi_schedule - schedule for receive
4395 * @n: entry to schedule
4397 * The entry's receive function will be scheduled to run.
4398 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4400 void __napi_schedule(struct napi_struct *n)
4402 unsigned long flags;
4404 local_irq_save(flags);
4405 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4406 local_irq_restore(flags);
4408 EXPORT_SYMBOL(__napi_schedule);
4411 * __napi_schedule_irqoff - schedule for receive
4412 * @n: entry to schedule
4414 * Variant of __napi_schedule() assuming hard irqs are masked
4416 void __napi_schedule_irqoff(struct napi_struct *n)
4418 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4420 EXPORT_SYMBOL(__napi_schedule_irqoff);
4422 void __napi_complete(struct napi_struct *n)
4424 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4426 list_del_init(&n->poll_list);
4427 smp_mb__before_atomic();
4428 clear_bit(NAPI_STATE_SCHED, &n->state);
4430 EXPORT_SYMBOL(__napi_complete);
4432 void napi_complete_done(struct napi_struct *n, int work_done)
4434 unsigned long flags;
4437 * don't let napi dequeue from the cpu poll list
4438 * just in case its running on a different cpu
4440 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4444 unsigned long timeout = 0;
4447 timeout = n->dev->gro_flush_timeout;
4450 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4451 HRTIMER_MODE_REL_PINNED);
4453 napi_gro_flush(n, false);
4455 if (likely(list_empty(&n->poll_list))) {
4456 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4458 /* If n->poll_list is not empty, we need to mask irqs */
4459 local_irq_save(flags);
4461 local_irq_restore(flags);
4464 EXPORT_SYMBOL(napi_complete_done);
4466 /* must be called under rcu_read_lock(), as we dont take a reference */
4467 struct napi_struct *napi_by_id(unsigned int napi_id)
4469 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4470 struct napi_struct *napi;
4472 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4473 if (napi->napi_id == napi_id)
4478 EXPORT_SYMBOL_GPL(napi_by_id);
4480 void napi_hash_add(struct napi_struct *napi)
4482 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4484 spin_lock(&napi_hash_lock);
4486 /* 0 is not a valid id, we also skip an id that is taken
4487 * we expect both events to be extremely rare
4490 while (!napi->napi_id) {
4491 napi->napi_id = ++napi_gen_id;
4492 if (napi_by_id(napi->napi_id))
4496 hlist_add_head_rcu(&napi->napi_hash_node,
4497 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4499 spin_unlock(&napi_hash_lock);
4502 EXPORT_SYMBOL_GPL(napi_hash_add);
4504 /* Warning : caller is responsible to make sure rcu grace period
4505 * is respected before freeing memory containing @napi
4507 void napi_hash_del(struct napi_struct *napi)
4509 spin_lock(&napi_hash_lock);
4511 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4512 hlist_del_rcu(&napi->napi_hash_node);
4514 spin_unlock(&napi_hash_lock);
4516 EXPORT_SYMBOL_GPL(napi_hash_del);
4518 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4520 struct napi_struct *napi;
4522 napi = container_of(timer, struct napi_struct, timer);
4524 napi_schedule(napi);
4526 return HRTIMER_NORESTART;
4529 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4530 int (*poll)(struct napi_struct *, int), int weight)
4532 INIT_LIST_HEAD(&napi->poll_list);
4533 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4534 napi->timer.function = napi_watchdog;
4535 napi->gro_count = 0;
4536 napi->gro_list = NULL;
4539 if (weight > NAPI_POLL_WEIGHT)
4540 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4542 napi->weight = weight;
4543 list_add(&napi->dev_list, &dev->napi_list);
4545 #ifdef CONFIG_NETPOLL
4546 spin_lock_init(&napi->poll_lock);
4547 napi->poll_owner = -1;
4549 set_bit(NAPI_STATE_SCHED, &napi->state);
4551 EXPORT_SYMBOL(netif_napi_add);
4553 void napi_disable(struct napi_struct *n)
4556 set_bit(NAPI_STATE_DISABLE, &n->state);
4558 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4561 hrtimer_cancel(&n->timer);
4563 clear_bit(NAPI_STATE_DISABLE, &n->state);
4565 EXPORT_SYMBOL(napi_disable);
4567 void netif_napi_del(struct napi_struct *napi)
4569 list_del_init(&napi->dev_list);
4570 napi_free_frags(napi);
4572 kfree_skb_list(napi->gro_list);
4573 napi->gro_list = NULL;
4574 napi->gro_count = 0;
4576 EXPORT_SYMBOL(netif_napi_del);
4578 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4583 list_del_init(&n->poll_list);
4585 have = netpoll_poll_lock(n);
4589 /* This NAPI_STATE_SCHED test is for avoiding a race
4590 * with netpoll's poll_napi(). Only the entity which
4591 * obtains the lock and sees NAPI_STATE_SCHED set will
4592 * actually make the ->poll() call. Therefore we avoid
4593 * accidentally calling ->poll() when NAPI is not scheduled.
4596 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4597 work = n->poll(n, weight);
4601 WARN_ON_ONCE(work > weight);
4603 if (likely(work < weight))
4606 /* Drivers must not modify the NAPI state if they
4607 * consume the entire weight. In such cases this code
4608 * still "owns" the NAPI instance and therefore can
4609 * move the instance around on the list at-will.
4611 if (unlikely(napi_disable_pending(n))) {
4617 /* flush too old packets
4618 * If HZ < 1000, flush all packets.
4620 napi_gro_flush(n, HZ >= 1000);
4623 /* Some drivers may have called napi_schedule
4624 * prior to exhausting their budget.
4626 if (unlikely(!list_empty(&n->poll_list))) {
4627 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4628 n->dev ? n->dev->name : "backlog");
4632 list_add_tail(&n->poll_list, repoll);
4635 netpoll_poll_unlock(have);
4640 static void net_rx_action(struct softirq_action *h)
4642 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4643 unsigned long time_limit = jiffies + 2;
4644 int budget = netdev_budget;
4648 local_irq_disable();
4649 list_splice_init(&sd->poll_list, &list);
4653 struct napi_struct *n;
4655 if (list_empty(&list)) {
4656 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4661 n = list_first_entry(&list, struct napi_struct, poll_list);
4662 budget -= napi_poll(n, &repoll);
4664 /* If softirq window is exhausted then punt.
4665 * Allow this to run for 2 jiffies since which will allow
4666 * an average latency of 1.5/HZ.
4668 if (unlikely(budget <= 0 ||
4669 time_after_eq(jiffies, time_limit))) {
4675 local_irq_disable();
4677 list_splice_tail_init(&sd->poll_list, &list);
4678 list_splice_tail(&repoll, &list);
4679 list_splice(&list, &sd->poll_list);
4680 if (!list_empty(&sd->poll_list))
4681 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4683 net_rps_action_and_irq_enable(sd);
4686 struct netdev_adjacent {
4687 struct net_device *dev;
4689 /* upper master flag, there can only be one master device per list */
4692 /* counter for the number of times this device was added to us */
4695 /* private field for the users */
4698 struct list_head list;
4699 struct rcu_head rcu;
4702 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4703 struct net_device *adj_dev,
4704 struct list_head *adj_list)
4706 struct netdev_adjacent *adj;
4708 list_for_each_entry(adj, adj_list, list) {
4709 if (adj->dev == adj_dev)
4716 * netdev_has_upper_dev - Check if device is linked to an upper device
4718 * @upper_dev: upper device to check
4720 * Find out if a device is linked to specified upper device and return true
4721 * in case it is. Note that this checks only immediate upper device,
4722 * not through a complete stack of devices. The caller must hold the RTNL lock.
4724 bool netdev_has_upper_dev(struct net_device *dev,
4725 struct net_device *upper_dev)
4729 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4731 EXPORT_SYMBOL(netdev_has_upper_dev);
4734 * netdev_has_any_upper_dev - Check if device is linked to some device
4737 * Find out if a device is linked to an upper device and return true in case
4738 * it is. The caller must hold the RTNL lock.
4740 static bool netdev_has_any_upper_dev(struct net_device *dev)
4744 return !list_empty(&dev->all_adj_list.upper);
4748 * netdev_master_upper_dev_get - Get master upper device
4751 * Find a master upper device and return pointer to it or NULL in case
4752 * it's not there. The caller must hold the RTNL lock.
4754 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4756 struct netdev_adjacent *upper;
4760 if (list_empty(&dev->adj_list.upper))
4763 upper = list_first_entry(&dev->adj_list.upper,
4764 struct netdev_adjacent, list);
4765 if (likely(upper->master))
4769 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4771 void *netdev_adjacent_get_private(struct list_head *adj_list)
4773 struct netdev_adjacent *adj;
4775 adj = list_entry(adj_list, struct netdev_adjacent, list);
4777 return adj->private;
4779 EXPORT_SYMBOL(netdev_adjacent_get_private);
4782 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4784 * @iter: list_head ** of the current position
4786 * Gets the next device from the dev's upper list, starting from iter
4787 * position. The caller must hold RCU read lock.
4789 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4790 struct list_head **iter)
4792 struct netdev_adjacent *upper;
4794 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4796 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4798 if (&upper->list == &dev->adj_list.upper)
4801 *iter = &upper->list;
4805 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4808 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4810 * @iter: list_head ** of the current position
4812 * Gets the next device from the dev's upper list, starting from iter
4813 * position. The caller must hold RCU read lock.
4815 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4816 struct list_head **iter)
4818 struct netdev_adjacent *upper;
4820 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4822 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4824 if (&upper->list == &dev->all_adj_list.upper)
4827 *iter = &upper->list;
4831 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4834 * netdev_lower_get_next_private - Get the next ->private from the
4835 * lower neighbour list
4837 * @iter: list_head ** of the current position
4839 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4840 * list, starting from iter position. The caller must hold either hold the
4841 * RTNL lock or its own locking that guarantees that the neighbour lower
4842 * list will remain unchainged.
4844 void *netdev_lower_get_next_private(struct net_device *dev,
4845 struct list_head **iter)
4847 struct netdev_adjacent *lower;
4849 lower = list_entry(*iter, struct netdev_adjacent, list);
4851 if (&lower->list == &dev->adj_list.lower)
4854 *iter = lower->list.next;
4856 return lower->private;
4858 EXPORT_SYMBOL(netdev_lower_get_next_private);
4861 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4862 * lower neighbour list, RCU
4865 * @iter: list_head ** of the current position
4867 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4868 * list, starting from iter position. The caller must hold RCU read lock.
4870 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4871 struct list_head **iter)
4873 struct netdev_adjacent *lower;
4875 WARN_ON_ONCE(!rcu_read_lock_held());
4877 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4879 if (&lower->list == &dev->adj_list.lower)
4882 *iter = &lower->list;
4884 return lower->private;
4886 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4889 * netdev_lower_get_next - Get the next device from the lower neighbour
4892 * @iter: list_head ** of the current position
4894 * Gets the next netdev_adjacent from the dev's lower neighbour
4895 * list, starting from iter position. The caller must hold RTNL lock or
4896 * its own locking that guarantees that the neighbour lower
4897 * list will remain unchainged.
4899 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4901 struct netdev_adjacent *lower;
4903 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4905 if (&lower->list == &dev->adj_list.lower)
4908 *iter = &lower->list;
4912 EXPORT_SYMBOL(netdev_lower_get_next);
4915 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4916 * lower neighbour list, RCU
4920 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4921 * list. The caller must hold RCU read lock.
4923 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4925 struct netdev_adjacent *lower;
4927 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4928 struct netdev_adjacent, list);
4930 return lower->private;
4933 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4936 * netdev_master_upper_dev_get_rcu - Get master upper device
4939 * Find a master upper device and return pointer to it or NULL in case
4940 * it's not there. The caller must hold the RCU read lock.
4942 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4944 struct netdev_adjacent *upper;
4946 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4947 struct netdev_adjacent, list);
4948 if (upper && likely(upper->master))
4952 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4954 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4955 struct net_device *adj_dev,
4956 struct list_head *dev_list)
4958 char linkname[IFNAMSIZ+7];
4959 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4960 "upper_%s" : "lower_%s", adj_dev->name);
4961 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4964 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4966 struct list_head *dev_list)
4968 char linkname[IFNAMSIZ+7];
4969 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4970 "upper_%s" : "lower_%s", name);
4971 sysfs_remove_link(&(dev->dev.kobj), linkname);
4974 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
4975 struct net_device *adj_dev,
4976 struct list_head *dev_list)
4978 return (dev_list == &dev->adj_list.upper ||
4979 dev_list == &dev->adj_list.lower) &&
4980 net_eq(dev_net(dev), dev_net(adj_dev));
4983 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4984 struct net_device *adj_dev,
4985 struct list_head *dev_list,
4986 void *private, bool master)
4988 struct netdev_adjacent *adj;
4991 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4998 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5003 adj->master = master;
5005 adj->private = private;
5008 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5009 adj_dev->name, dev->name, adj_dev->name);
5011 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5012 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5017 /* Ensure that master link is always the first item in list. */
5019 ret = sysfs_create_link(&(dev->dev.kobj),
5020 &(adj_dev->dev.kobj), "master");
5022 goto remove_symlinks;
5024 list_add_rcu(&adj->list, dev_list);
5026 list_add_tail_rcu(&adj->list, dev_list);
5032 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5033 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5041 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5042 struct net_device *adj_dev,
5043 struct list_head *dev_list)
5045 struct netdev_adjacent *adj;
5047 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5050 pr_err("tried to remove device %s from %s\n",
5051 dev->name, adj_dev->name);
5055 if (adj->ref_nr > 1) {
5056 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5063 sysfs_remove_link(&(dev->dev.kobj), "master");
5065 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5066 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5068 list_del_rcu(&adj->list);
5069 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5070 adj_dev->name, dev->name, adj_dev->name);
5072 kfree_rcu(adj, rcu);
5075 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5076 struct net_device *upper_dev,
5077 struct list_head *up_list,
5078 struct list_head *down_list,
5079 void *private, bool master)
5083 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5088 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5091 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5098 static int __netdev_adjacent_dev_link(struct net_device *dev,
5099 struct net_device *upper_dev)
5101 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5102 &dev->all_adj_list.upper,
5103 &upper_dev->all_adj_list.lower,
5107 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5108 struct net_device *upper_dev,
5109 struct list_head *up_list,
5110 struct list_head *down_list)
5112 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5113 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5116 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5117 struct net_device *upper_dev)
5119 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5120 &dev->all_adj_list.upper,
5121 &upper_dev->all_adj_list.lower);
5124 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5125 struct net_device *upper_dev,
5126 void *private, bool master)
5128 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5133 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5134 &dev->adj_list.upper,
5135 &upper_dev->adj_list.lower,
5138 __netdev_adjacent_dev_unlink(dev, upper_dev);
5145 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5146 struct net_device *upper_dev)
5148 __netdev_adjacent_dev_unlink(dev, upper_dev);
5149 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5150 &dev->adj_list.upper,
5151 &upper_dev->adj_list.lower);
5154 static int __netdev_upper_dev_link(struct net_device *dev,
5155 struct net_device *upper_dev, bool master,
5158 struct netdev_adjacent *i, *j, *to_i, *to_j;
5163 if (dev == upper_dev)
5166 /* To prevent loops, check if dev is not upper device to upper_dev. */
5167 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5170 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
5173 if (master && netdev_master_upper_dev_get(dev))
5176 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5181 /* Now that we linked these devs, make all the upper_dev's
5182 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5183 * versa, and don't forget the devices itself. All of these
5184 * links are non-neighbours.
5186 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5187 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5188 pr_debug("Interlinking %s with %s, non-neighbour\n",
5189 i->dev->name, j->dev->name);
5190 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5196 /* add dev to every upper_dev's upper device */
5197 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5198 pr_debug("linking %s's upper device %s with %s\n",
5199 upper_dev->name, i->dev->name, dev->name);
5200 ret = __netdev_adjacent_dev_link(dev, i->dev);
5202 goto rollback_upper_mesh;
5205 /* add upper_dev to every dev's lower device */
5206 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5207 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5208 i->dev->name, upper_dev->name);
5209 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5211 goto rollback_lower_mesh;
5214 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5217 rollback_lower_mesh:
5219 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5222 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5227 rollback_upper_mesh:
5229 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5232 __netdev_adjacent_dev_unlink(dev, i->dev);
5240 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5241 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5242 if (i == to_i && j == to_j)
5244 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5250 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5256 * netdev_upper_dev_link - Add a link to the upper device
5258 * @upper_dev: new upper device
5260 * Adds a link to device which is upper to this one. The caller must hold
5261 * the RTNL lock. On a failure a negative errno code is returned.
5262 * On success the reference counts are adjusted and the function
5265 int netdev_upper_dev_link(struct net_device *dev,
5266 struct net_device *upper_dev)
5268 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5270 EXPORT_SYMBOL(netdev_upper_dev_link);
5273 * netdev_master_upper_dev_link - Add a master link to the upper device
5275 * @upper_dev: new upper device
5277 * Adds a link to device which is upper to this one. In this case, only
5278 * one master upper device can be linked, although other non-master devices
5279 * might be linked as well. The caller must hold the RTNL lock.
5280 * On a failure a negative errno code is returned. On success the reference
5281 * counts are adjusted and the function returns zero.
5283 int netdev_master_upper_dev_link(struct net_device *dev,
5284 struct net_device *upper_dev)
5286 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5288 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5290 int netdev_master_upper_dev_link_private(struct net_device *dev,
5291 struct net_device *upper_dev,
5294 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5296 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5299 * netdev_upper_dev_unlink - Removes a link to upper device
5301 * @upper_dev: new upper device
5303 * Removes a link to device which is upper to this one. The caller must hold
5306 void netdev_upper_dev_unlink(struct net_device *dev,
5307 struct net_device *upper_dev)
5309 struct netdev_adjacent *i, *j;
5312 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5314 /* Here is the tricky part. We must remove all dev's lower
5315 * devices from all upper_dev's upper devices and vice
5316 * versa, to maintain the graph relationship.
5318 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5319 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5320 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5322 /* remove also the devices itself from lower/upper device
5325 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5326 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5328 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5329 __netdev_adjacent_dev_unlink(dev, i->dev);
5331 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5333 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5336 * netdev_bonding_info_change - Dispatch event about slave change
5338 * @netdev_bonding_info: info to dispatch
5340 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5341 * The caller must hold the RTNL lock.
5343 void netdev_bonding_info_change(struct net_device *dev,
5344 struct netdev_bonding_info *bonding_info)
5346 struct netdev_notifier_bonding_info info;
5348 memcpy(&info.bonding_info, bonding_info,
5349 sizeof(struct netdev_bonding_info));
5350 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5353 EXPORT_SYMBOL(netdev_bonding_info_change);
5355 static void netdev_adjacent_add_links(struct net_device *dev)
5357 struct netdev_adjacent *iter;
5359 struct net *net = dev_net(dev);
5361 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5362 if (!net_eq(net,dev_net(iter->dev)))
5364 netdev_adjacent_sysfs_add(iter->dev, dev,
5365 &iter->dev->adj_list.lower);
5366 netdev_adjacent_sysfs_add(dev, iter->dev,
5367 &dev->adj_list.upper);
5370 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5371 if (!net_eq(net,dev_net(iter->dev)))
5373 netdev_adjacent_sysfs_add(iter->dev, dev,
5374 &iter->dev->adj_list.upper);
5375 netdev_adjacent_sysfs_add(dev, iter->dev,
5376 &dev->adj_list.lower);
5380 static void netdev_adjacent_del_links(struct net_device *dev)
5382 struct netdev_adjacent *iter;
5384 struct net *net = dev_net(dev);
5386 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5387 if (!net_eq(net,dev_net(iter->dev)))
5389 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5390 &iter->dev->adj_list.lower);
5391 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5392 &dev->adj_list.upper);
5395 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5396 if (!net_eq(net,dev_net(iter->dev)))
5398 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5399 &iter->dev->adj_list.upper);
5400 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5401 &dev->adj_list.lower);
5405 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5407 struct netdev_adjacent *iter;
5409 struct net *net = dev_net(dev);
5411 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5412 if (!net_eq(net,dev_net(iter->dev)))
5414 netdev_adjacent_sysfs_del(iter->dev, oldname,
5415 &iter->dev->adj_list.lower);
5416 netdev_adjacent_sysfs_add(iter->dev, dev,
5417 &iter->dev->adj_list.lower);
5420 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5421 if (!net_eq(net,dev_net(iter->dev)))
5423 netdev_adjacent_sysfs_del(iter->dev, oldname,
5424 &iter->dev->adj_list.upper);
5425 netdev_adjacent_sysfs_add(iter->dev, dev,
5426 &iter->dev->adj_list.upper);
5430 void *netdev_lower_dev_get_private(struct net_device *dev,
5431 struct net_device *lower_dev)
5433 struct netdev_adjacent *lower;
5437 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5441 return lower->private;
5443 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5446 int dev_get_nest_level(struct net_device *dev,
5447 bool (*type_check)(struct net_device *dev))
5449 struct net_device *lower = NULL;
5450 struct list_head *iter;
5456 netdev_for_each_lower_dev(dev, lower, iter) {
5457 nest = dev_get_nest_level(lower, type_check);
5458 if (max_nest < nest)
5462 if (type_check(dev))
5467 EXPORT_SYMBOL(dev_get_nest_level);
5469 static void dev_change_rx_flags(struct net_device *dev, int flags)
5471 const struct net_device_ops *ops = dev->netdev_ops;
5473 if (ops->ndo_change_rx_flags)
5474 ops->ndo_change_rx_flags(dev, flags);
5477 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5479 unsigned int old_flags = dev->flags;
5485 dev->flags |= IFF_PROMISC;
5486 dev->promiscuity += inc;
5487 if (dev->promiscuity == 0) {
5490 * If inc causes overflow, untouch promisc and return error.
5493 dev->flags &= ~IFF_PROMISC;
5495 dev->promiscuity -= inc;
5496 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5501 if (dev->flags != old_flags) {
5502 pr_info("device %s %s promiscuous mode\n",
5504 dev->flags & IFF_PROMISC ? "entered" : "left");
5505 if (audit_enabled) {
5506 current_uid_gid(&uid, &gid);
5507 audit_log(current->audit_context, GFP_ATOMIC,
5508 AUDIT_ANOM_PROMISCUOUS,
5509 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5510 dev->name, (dev->flags & IFF_PROMISC),
5511 (old_flags & IFF_PROMISC),
5512 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5513 from_kuid(&init_user_ns, uid),
5514 from_kgid(&init_user_ns, gid),
5515 audit_get_sessionid(current));
5518 dev_change_rx_flags(dev, IFF_PROMISC);
5521 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5526 * dev_set_promiscuity - update promiscuity count on a device
5530 * Add or remove promiscuity from a device. While the count in the device
5531 * remains above zero the interface remains promiscuous. Once it hits zero
5532 * the device reverts back to normal filtering operation. A negative inc
5533 * value is used to drop promiscuity on the device.
5534 * Return 0 if successful or a negative errno code on error.
5536 int dev_set_promiscuity(struct net_device *dev, int inc)
5538 unsigned int old_flags = dev->flags;
5541 err = __dev_set_promiscuity(dev, inc, true);
5544 if (dev->flags != old_flags)
5545 dev_set_rx_mode(dev);
5548 EXPORT_SYMBOL(dev_set_promiscuity);
5550 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5552 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5556 dev->flags |= IFF_ALLMULTI;
5557 dev->allmulti += inc;
5558 if (dev->allmulti == 0) {
5561 * If inc causes overflow, untouch allmulti and return error.
5564 dev->flags &= ~IFF_ALLMULTI;
5566 dev->allmulti -= inc;
5567 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5572 if (dev->flags ^ old_flags) {
5573 dev_change_rx_flags(dev, IFF_ALLMULTI);
5574 dev_set_rx_mode(dev);
5576 __dev_notify_flags(dev, old_flags,
5577 dev->gflags ^ old_gflags);
5583 * dev_set_allmulti - update allmulti count on a device
5587 * Add or remove reception of all multicast frames to a device. While the
5588 * count in the device remains above zero the interface remains listening
5589 * to all interfaces. Once it hits zero the device reverts back to normal
5590 * filtering operation. A negative @inc value is used to drop the counter
5591 * when releasing a resource needing all multicasts.
5592 * Return 0 if successful or a negative errno code on error.
5595 int dev_set_allmulti(struct net_device *dev, int inc)
5597 return __dev_set_allmulti(dev, inc, true);
5599 EXPORT_SYMBOL(dev_set_allmulti);
5602 * Upload unicast and multicast address lists to device and
5603 * configure RX filtering. When the device doesn't support unicast
5604 * filtering it is put in promiscuous mode while unicast addresses
5607 void __dev_set_rx_mode(struct net_device *dev)
5609 const struct net_device_ops *ops = dev->netdev_ops;
5611 /* dev_open will call this function so the list will stay sane. */
5612 if (!(dev->flags&IFF_UP))
5615 if (!netif_device_present(dev))
5618 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5619 /* Unicast addresses changes may only happen under the rtnl,
5620 * therefore calling __dev_set_promiscuity here is safe.
5622 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5623 __dev_set_promiscuity(dev, 1, false);
5624 dev->uc_promisc = true;
5625 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5626 __dev_set_promiscuity(dev, -1, false);
5627 dev->uc_promisc = false;
5631 if (ops->ndo_set_rx_mode)
5632 ops->ndo_set_rx_mode(dev);
5635 void dev_set_rx_mode(struct net_device *dev)
5637 netif_addr_lock_bh(dev);
5638 __dev_set_rx_mode(dev);
5639 netif_addr_unlock_bh(dev);
5643 * dev_get_flags - get flags reported to userspace
5646 * Get the combination of flag bits exported through APIs to userspace.
5648 unsigned int dev_get_flags(const struct net_device *dev)
5652 flags = (dev->flags & ~(IFF_PROMISC |
5657 (dev->gflags & (IFF_PROMISC |
5660 if (netif_running(dev)) {
5661 if (netif_oper_up(dev))
5662 flags |= IFF_RUNNING;
5663 if (netif_carrier_ok(dev))
5664 flags |= IFF_LOWER_UP;
5665 if (netif_dormant(dev))
5666 flags |= IFF_DORMANT;
5671 EXPORT_SYMBOL(dev_get_flags);
5673 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5675 unsigned int old_flags = dev->flags;
5681 * Set the flags on our device.
5684 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5685 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5687 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5691 * Load in the correct multicast list now the flags have changed.
5694 if ((old_flags ^ flags) & IFF_MULTICAST)
5695 dev_change_rx_flags(dev, IFF_MULTICAST);
5697 dev_set_rx_mode(dev);
5700 * Have we downed the interface. We handle IFF_UP ourselves
5701 * according to user attempts to set it, rather than blindly
5706 if ((old_flags ^ flags) & IFF_UP)
5707 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5709 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5710 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5711 unsigned int old_flags = dev->flags;
5713 dev->gflags ^= IFF_PROMISC;
5715 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5716 if (dev->flags != old_flags)
5717 dev_set_rx_mode(dev);
5720 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5721 is important. Some (broken) drivers set IFF_PROMISC, when
5722 IFF_ALLMULTI is requested not asking us and not reporting.
5724 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5725 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5727 dev->gflags ^= IFF_ALLMULTI;
5728 __dev_set_allmulti(dev, inc, false);
5734 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5735 unsigned int gchanges)
5737 unsigned int changes = dev->flags ^ old_flags;
5740 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5742 if (changes & IFF_UP) {
5743 if (dev->flags & IFF_UP)
5744 call_netdevice_notifiers(NETDEV_UP, dev);
5746 call_netdevice_notifiers(NETDEV_DOWN, dev);
5749 if (dev->flags & IFF_UP &&
5750 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5751 struct netdev_notifier_change_info change_info;
5753 change_info.flags_changed = changes;
5754 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5760 * dev_change_flags - change device settings
5762 * @flags: device state flags
5764 * Change settings on device based state flags. The flags are
5765 * in the userspace exported format.
5767 int dev_change_flags(struct net_device *dev, unsigned int flags)
5770 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5772 ret = __dev_change_flags(dev, flags);
5776 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5777 __dev_notify_flags(dev, old_flags, changes);
5780 EXPORT_SYMBOL(dev_change_flags);
5782 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5784 const struct net_device_ops *ops = dev->netdev_ops;
5786 if (ops->ndo_change_mtu)
5787 return ops->ndo_change_mtu(dev, new_mtu);
5794 * dev_set_mtu - Change maximum transfer unit
5796 * @new_mtu: new transfer unit
5798 * Change the maximum transfer size of the network device.
5800 int dev_set_mtu(struct net_device *dev, int new_mtu)
5804 if (new_mtu == dev->mtu)
5807 /* MTU must be positive. */
5811 if (!netif_device_present(dev))
5814 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5815 err = notifier_to_errno(err);
5819 orig_mtu = dev->mtu;
5820 err = __dev_set_mtu(dev, new_mtu);
5823 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5824 err = notifier_to_errno(err);
5826 /* setting mtu back and notifying everyone again,
5827 * so that they have a chance to revert changes.
5829 __dev_set_mtu(dev, orig_mtu);
5830 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5835 EXPORT_SYMBOL(dev_set_mtu);
5838 * dev_set_group - Change group this device belongs to
5840 * @new_group: group this device should belong to
5842 void dev_set_group(struct net_device *dev, int new_group)
5844 dev->group = new_group;
5846 EXPORT_SYMBOL(dev_set_group);
5849 * dev_set_mac_address - Change Media Access Control Address
5853 * Change the hardware (MAC) address of the device
5855 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5857 const struct net_device_ops *ops = dev->netdev_ops;
5860 if (!ops->ndo_set_mac_address)
5862 if (sa->sa_family != dev->type)
5864 if (!netif_device_present(dev))
5866 err = ops->ndo_set_mac_address(dev, sa);
5869 dev->addr_assign_type = NET_ADDR_SET;
5870 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5871 add_device_randomness(dev->dev_addr, dev->addr_len);
5874 EXPORT_SYMBOL(dev_set_mac_address);
5877 * dev_change_carrier - Change device carrier
5879 * @new_carrier: new value
5881 * Change device carrier
5883 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5885 const struct net_device_ops *ops = dev->netdev_ops;
5887 if (!ops->ndo_change_carrier)
5889 if (!netif_device_present(dev))
5891 return ops->ndo_change_carrier(dev, new_carrier);
5893 EXPORT_SYMBOL(dev_change_carrier);
5896 * dev_get_phys_port_id - Get device physical port ID
5900 * Get device physical port ID
5902 int dev_get_phys_port_id(struct net_device *dev,
5903 struct netdev_phys_item_id *ppid)
5905 const struct net_device_ops *ops = dev->netdev_ops;
5907 if (!ops->ndo_get_phys_port_id)
5909 return ops->ndo_get_phys_port_id(dev, ppid);
5911 EXPORT_SYMBOL(dev_get_phys_port_id);
5914 * dev_new_index - allocate an ifindex
5915 * @net: the applicable net namespace
5917 * Returns a suitable unique value for a new device interface
5918 * number. The caller must hold the rtnl semaphore or the
5919 * dev_base_lock to be sure it remains unique.
5921 static int dev_new_index(struct net *net)
5923 int ifindex = net->ifindex;
5927 if (!__dev_get_by_index(net, ifindex))
5928 return net->ifindex = ifindex;
5932 /* Delayed registration/unregisteration */
5933 static LIST_HEAD(net_todo_list);
5934 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5936 static void net_set_todo(struct net_device *dev)
5938 list_add_tail(&dev->todo_list, &net_todo_list);
5939 dev_net(dev)->dev_unreg_count++;
5942 static void rollback_registered_many(struct list_head *head)
5944 struct net_device *dev, *tmp;
5945 LIST_HEAD(close_head);
5947 BUG_ON(dev_boot_phase);
5950 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5951 /* Some devices call without registering
5952 * for initialization unwind. Remove those
5953 * devices and proceed with the remaining.
5955 if (dev->reg_state == NETREG_UNINITIALIZED) {
5956 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5960 list_del(&dev->unreg_list);
5963 dev->dismantle = true;
5964 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5967 /* If device is running, close it first. */
5968 list_for_each_entry(dev, head, unreg_list)
5969 list_add_tail(&dev->close_list, &close_head);
5970 dev_close_many(&close_head);
5972 list_for_each_entry(dev, head, unreg_list) {
5973 /* And unlink it from device chain. */
5974 unlist_netdevice(dev);
5976 dev->reg_state = NETREG_UNREGISTERING;
5981 list_for_each_entry(dev, head, unreg_list) {
5982 struct sk_buff *skb = NULL;
5984 /* Shutdown queueing discipline. */
5988 /* Notify protocols, that we are about to destroy
5989 this device. They should clean all the things.
5991 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5993 if (!dev->rtnl_link_ops ||
5994 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5995 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
5999 * Flush the unicast and multicast chains
6004 if (dev->netdev_ops->ndo_uninit)
6005 dev->netdev_ops->ndo_uninit(dev);
6008 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6010 /* Notifier chain MUST detach us all upper devices. */
6011 WARN_ON(netdev_has_any_upper_dev(dev));
6013 /* Remove entries from kobject tree */
6014 netdev_unregister_kobject(dev);
6016 /* Remove XPS queueing entries */
6017 netif_reset_xps_queues_gt(dev, 0);
6023 list_for_each_entry(dev, head, unreg_list)
6027 static void rollback_registered(struct net_device *dev)
6031 list_add(&dev->unreg_list, &single);
6032 rollback_registered_many(&single);
6036 static netdev_features_t netdev_fix_features(struct net_device *dev,
6037 netdev_features_t features)
6039 /* Fix illegal checksum combinations */
6040 if ((features & NETIF_F_HW_CSUM) &&
6041 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6042 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6043 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6046 /* TSO requires that SG is present as well. */
6047 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6048 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6049 features &= ~NETIF_F_ALL_TSO;
6052 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6053 !(features & NETIF_F_IP_CSUM)) {
6054 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6055 features &= ~NETIF_F_TSO;
6056 features &= ~NETIF_F_TSO_ECN;
6059 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6060 !(features & NETIF_F_IPV6_CSUM)) {
6061 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6062 features &= ~NETIF_F_TSO6;
6065 /* TSO ECN requires that TSO is present as well. */
6066 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6067 features &= ~NETIF_F_TSO_ECN;
6069 /* Software GSO depends on SG. */
6070 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6071 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6072 features &= ~NETIF_F_GSO;
6075 /* UFO needs SG and checksumming */
6076 if (features & NETIF_F_UFO) {
6077 /* maybe split UFO into V4 and V6? */
6078 if (!((features & NETIF_F_GEN_CSUM) ||
6079 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6080 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6082 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6083 features &= ~NETIF_F_UFO;
6086 if (!(features & NETIF_F_SG)) {
6088 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6089 features &= ~NETIF_F_UFO;
6093 #ifdef CONFIG_NET_RX_BUSY_POLL
6094 if (dev->netdev_ops->ndo_busy_poll)
6095 features |= NETIF_F_BUSY_POLL;
6098 features &= ~NETIF_F_BUSY_POLL;
6103 int __netdev_update_features(struct net_device *dev)
6105 netdev_features_t features;
6110 features = netdev_get_wanted_features(dev);
6112 if (dev->netdev_ops->ndo_fix_features)
6113 features = dev->netdev_ops->ndo_fix_features(dev, features);
6115 /* driver might be less strict about feature dependencies */
6116 features = netdev_fix_features(dev, features);
6118 if (dev->features == features)
6121 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6122 &dev->features, &features);
6124 if (dev->netdev_ops->ndo_set_features)
6125 err = dev->netdev_ops->ndo_set_features(dev, features);
6127 if (unlikely(err < 0)) {
6129 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6130 err, &features, &dev->features);
6135 dev->features = features;
6141 * netdev_update_features - recalculate device features
6142 * @dev: the device to check
6144 * Recalculate dev->features set and send notifications if it
6145 * has changed. Should be called after driver or hardware dependent
6146 * conditions might have changed that influence the features.
6148 void netdev_update_features(struct net_device *dev)
6150 if (__netdev_update_features(dev))
6151 netdev_features_change(dev);
6153 EXPORT_SYMBOL(netdev_update_features);
6156 * netdev_change_features - recalculate device features
6157 * @dev: the device to check
6159 * Recalculate dev->features set and send notifications even
6160 * if they have not changed. Should be called instead of
6161 * netdev_update_features() if also dev->vlan_features might
6162 * have changed to allow the changes to be propagated to stacked
6165 void netdev_change_features(struct net_device *dev)
6167 __netdev_update_features(dev);
6168 netdev_features_change(dev);
6170 EXPORT_SYMBOL(netdev_change_features);
6173 * netif_stacked_transfer_operstate - transfer operstate
6174 * @rootdev: the root or lower level device to transfer state from
6175 * @dev: the device to transfer operstate to
6177 * Transfer operational state from root to device. This is normally
6178 * called when a stacking relationship exists between the root
6179 * device and the device(a leaf device).
6181 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6182 struct net_device *dev)
6184 if (rootdev->operstate == IF_OPER_DORMANT)
6185 netif_dormant_on(dev);
6187 netif_dormant_off(dev);
6189 if (netif_carrier_ok(rootdev)) {
6190 if (!netif_carrier_ok(dev))
6191 netif_carrier_on(dev);
6193 if (netif_carrier_ok(dev))
6194 netif_carrier_off(dev);
6197 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6200 static int netif_alloc_rx_queues(struct net_device *dev)
6202 unsigned int i, count = dev->num_rx_queues;
6203 struct netdev_rx_queue *rx;
6204 size_t sz = count * sizeof(*rx);
6208 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6216 for (i = 0; i < count; i++)
6222 static void netdev_init_one_queue(struct net_device *dev,
6223 struct netdev_queue *queue, void *_unused)
6225 /* Initialize queue lock */
6226 spin_lock_init(&queue->_xmit_lock);
6227 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6228 queue->xmit_lock_owner = -1;
6229 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6232 dql_init(&queue->dql, HZ);
6236 static void netif_free_tx_queues(struct net_device *dev)
6241 static int netif_alloc_netdev_queues(struct net_device *dev)
6243 unsigned int count = dev->num_tx_queues;
6244 struct netdev_queue *tx;
6245 size_t sz = count * sizeof(*tx);
6247 BUG_ON(count < 1 || count > 0xffff);
6249 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6257 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6258 spin_lock_init(&dev->tx_global_lock);
6264 * register_netdevice - register a network device
6265 * @dev: device to register
6267 * Take a completed network device structure and add it to the kernel
6268 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6269 * chain. 0 is returned on success. A negative errno code is returned
6270 * on a failure to set up the device, or if the name is a duplicate.
6272 * Callers must hold the rtnl semaphore. You may want
6273 * register_netdev() instead of this.
6276 * The locking appears insufficient to guarantee two parallel registers
6277 * will not get the same name.
6280 int register_netdevice(struct net_device *dev)
6283 struct net *net = dev_net(dev);
6285 BUG_ON(dev_boot_phase);
6290 /* When net_device's are persistent, this will be fatal. */
6291 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6294 spin_lock_init(&dev->addr_list_lock);
6295 netdev_set_addr_lockdep_class(dev);
6299 ret = dev_get_valid_name(net, dev, dev->name);
6303 /* Init, if this function is available */
6304 if (dev->netdev_ops->ndo_init) {
6305 ret = dev->netdev_ops->ndo_init(dev);
6313 if (((dev->hw_features | dev->features) &
6314 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6315 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6316 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6317 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6324 dev->ifindex = dev_new_index(net);
6325 else if (__dev_get_by_index(net, dev->ifindex))
6328 if (dev->iflink == -1)
6329 dev->iflink = dev->ifindex;
6331 /* Transfer changeable features to wanted_features and enable
6332 * software offloads (GSO and GRO).
6334 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6335 dev->features |= NETIF_F_SOFT_FEATURES;
6336 dev->wanted_features = dev->features & dev->hw_features;
6338 if (!(dev->flags & IFF_LOOPBACK)) {
6339 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6342 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6344 dev->vlan_features |= NETIF_F_HIGHDMA;
6346 /* Make NETIF_F_SG inheritable to tunnel devices.
6348 dev->hw_enc_features |= NETIF_F_SG;
6350 /* Make NETIF_F_SG inheritable to MPLS.
6352 dev->mpls_features |= NETIF_F_SG;
6354 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6355 ret = notifier_to_errno(ret);
6359 ret = netdev_register_kobject(dev);
6362 dev->reg_state = NETREG_REGISTERED;
6364 __netdev_update_features(dev);
6367 * Default initial state at registry is that the
6368 * device is present.
6371 set_bit(__LINK_STATE_PRESENT, &dev->state);
6373 linkwatch_init_dev(dev);
6375 dev_init_scheduler(dev);
6377 list_netdevice(dev);
6378 add_device_randomness(dev->dev_addr, dev->addr_len);
6380 /* If the device has permanent device address, driver should
6381 * set dev_addr and also addr_assign_type should be set to
6382 * NET_ADDR_PERM (default value).
6384 if (dev->addr_assign_type == NET_ADDR_PERM)
6385 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6387 /* Notify protocols, that a new device appeared. */
6388 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6389 ret = notifier_to_errno(ret);
6391 rollback_registered(dev);
6392 dev->reg_state = NETREG_UNREGISTERED;
6395 * Prevent userspace races by waiting until the network
6396 * device is fully setup before sending notifications.
6398 if (!dev->rtnl_link_ops ||
6399 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6400 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6406 if (dev->netdev_ops->ndo_uninit)
6407 dev->netdev_ops->ndo_uninit(dev);
6410 EXPORT_SYMBOL(register_netdevice);
6413 * init_dummy_netdev - init a dummy network device for NAPI
6414 * @dev: device to init
6416 * This takes a network device structure and initialize the minimum
6417 * amount of fields so it can be used to schedule NAPI polls without
6418 * registering a full blown interface. This is to be used by drivers
6419 * that need to tie several hardware interfaces to a single NAPI
6420 * poll scheduler due to HW limitations.
6422 int init_dummy_netdev(struct net_device *dev)
6424 /* Clear everything. Note we don't initialize spinlocks
6425 * are they aren't supposed to be taken by any of the
6426 * NAPI code and this dummy netdev is supposed to be
6427 * only ever used for NAPI polls
6429 memset(dev, 0, sizeof(struct net_device));
6431 /* make sure we BUG if trying to hit standard
6432 * register/unregister code path
6434 dev->reg_state = NETREG_DUMMY;
6436 /* NAPI wants this */
6437 INIT_LIST_HEAD(&dev->napi_list);
6439 /* a dummy interface is started by default */
6440 set_bit(__LINK_STATE_PRESENT, &dev->state);
6441 set_bit(__LINK_STATE_START, &dev->state);
6443 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6444 * because users of this 'device' dont need to change
6450 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6454 * register_netdev - register a network device
6455 * @dev: device to register
6457 * Take a completed network device structure and add it to the kernel
6458 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6459 * chain. 0 is returned on success. A negative errno code is returned
6460 * on a failure to set up the device, or if the name is a duplicate.
6462 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6463 * and expands the device name if you passed a format string to
6466 int register_netdev(struct net_device *dev)
6471 err = register_netdevice(dev);
6475 EXPORT_SYMBOL(register_netdev);
6477 int netdev_refcnt_read(const struct net_device *dev)
6481 for_each_possible_cpu(i)
6482 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6485 EXPORT_SYMBOL(netdev_refcnt_read);
6488 * netdev_wait_allrefs - wait until all references are gone.
6489 * @dev: target net_device
6491 * This is called when unregistering network devices.
6493 * Any protocol or device that holds a reference should register
6494 * for netdevice notification, and cleanup and put back the
6495 * reference if they receive an UNREGISTER event.
6496 * We can get stuck here if buggy protocols don't correctly
6499 static void netdev_wait_allrefs(struct net_device *dev)
6501 unsigned long rebroadcast_time, warning_time;
6504 linkwatch_forget_dev(dev);
6506 rebroadcast_time = warning_time = jiffies;
6507 refcnt = netdev_refcnt_read(dev);
6509 while (refcnt != 0) {
6510 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6513 /* Rebroadcast unregister notification */
6514 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6520 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6521 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6523 /* We must not have linkwatch events
6524 * pending on unregister. If this
6525 * happens, we simply run the queue
6526 * unscheduled, resulting in a noop
6529 linkwatch_run_queue();
6534 rebroadcast_time = jiffies;
6539 refcnt = netdev_refcnt_read(dev);
6541 if (time_after(jiffies, warning_time + 10 * HZ)) {
6542 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6544 warning_time = jiffies;
6553 * register_netdevice(x1);
6554 * register_netdevice(x2);
6556 * unregister_netdevice(y1);
6557 * unregister_netdevice(y2);
6563 * We are invoked by rtnl_unlock().
6564 * This allows us to deal with problems:
6565 * 1) We can delete sysfs objects which invoke hotplug
6566 * without deadlocking with linkwatch via keventd.
6567 * 2) Since we run with the RTNL semaphore not held, we can sleep
6568 * safely in order to wait for the netdev refcnt to drop to zero.
6570 * We must not return until all unregister events added during
6571 * the interval the lock was held have been completed.
6573 void netdev_run_todo(void)
6575 struct list_head list;
6577 /* Snapshot list, allow later requests */
6578 list_replace_init(&net_todo_list, &list);
6583 /* Wait for rcu callbacks to finish before next phase */
6584 if (!list_empty(&list))
6587 while (!list_empty(&list)) {
6588 struct net_device *dev
6589 = list_first_entry(&list, struct net_device, todo_list);
6590 list_del(&dev->todo_list);
6593 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6596 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6597 pr_err("network todo '%s' but state %d\n",
6598 dev->name, dev->reg_state);
6603 dev->reg_state = NETREG_UNREGISTERED;
6605 on_each_cpu(flush_backlog, dev, 1);
6607 netdev_wait_allrefs(dev);
6610 BUG_ON(netdev_refcnt_read(dev));
6611 BUG_ON(!list_empty(&dev->ptype_all));
6612 BUG_ON(!list_empty(&dev->ptype_specific));
6613 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6614 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6615 WARN_ON(dev->dn_ptr);
6617 if (dev->destructor)
6618 dev->destructor(dev);
6620 /* Report a network device has been unregistered */
6622 dev_net(dev)->dev_unreg_count--;
6624 wake_up(&netdev_unregistering_wq);
6626 /* Free network device */
6627 kobject_put(&dev->dev.kobj);
6631 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6632 * fields in the same order, with only the type differing.
6634 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6635 const struct net_device_stats *netdev_stats)
6637 #if BITS_PER_LONG == 64
6638 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6639 memcpy(stats64, netdev_stats, sizeof(*stats64));
6641 size_t i, n = sizeof(*stats64) / sizeof(u64);
6642 const unsigned long *src = (const unsigned long *)netdev_stats;
6643 u64 *dst = (u64 *)stats64;
6645 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6646 sizeof(*stats64) / sizeof(u64));
6647 for (i = 0; i < n; i++)
6651 EXPORT_SYMBOL(netdev_stats_to_stats64);
6654 * dev_get_stats - get network device statistics
6655 * @dev: device to get statistics from
6656 * @storage: place to store stats
6658 * Get network statistics from device. Return @storage.
6659 * The device driver may provide its own method by setting
6660 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6661 * otherwise the internal statistics structure is used.
6663 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6664 struct rtnl_link_stats64 *storage)
6666 const struct net_device_ops *ops = dev->netdev_ops;
6668 if (ops->ndo_get_stats64) {
6669 memset(storage, 0, sizeof(*storage));
6670 ops->ndo_get_stats64(dev, storage);
6671 } else if (ops->ndo_get_stats) {
6672 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6674 netdev_stats_to_stats64(storage, &dev->stats);
6676 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6677 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6680 EXPORT_SYMBOL(dev_get_stats);
6682 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6684 struct netdev_queue *queue = dev_ingress_queue(dev);
6686 #ifdef CONFIG_NET_CLS_ACT
6689 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6692 netdev_init_one_queue(dev, queue, NULL);
6693 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
6694 queue->qdisc_sleeping = &noop_qdisc;
6695 rcu_assign_pointer(dev->ingress_queue, queue);
6700 static const struct ethtool_ops default_ethtool_ops;
6702 void netdev_set_default_ethtool_ops(struct net_device *dev,
6703 const struct ethtool_ops *ops)
6705 if (dev->ethtool_ops == &default_ethtool_ops)
6706 dev->ethtool_ops = ops;
6708 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6710 void netdev_freemem(struct net_device *dev)
6712 char *addr = (char *)dev - dev->padded;
6718 * alloc_netdev_mqs - allocate network device
6719 * @sizeof_priv: size of private data to allocate space for
6720 * @name: device name format string
6721 * @name_assign_type: origin of device name
6722 * @setup: callback to initialize device
6723 * @txqs: the number of TX subqueues to allocate
6724 * @rxqs: the number of RX subqueues to allocate
6726 * Allocates a struct net_device with private data area for driver use
6727 * and performs basic initialization. Also allocates subqueue structs
6728 * for each queue on the device.
6730 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6731 unsigned char name_assign_type,
6732 void (*setup)(struct net_device *),
6733 unsigned int txqs, unsigned int rxqs)
6735 struct net_device *dev;
6737 struct net_device *p;
6739 BUG_ON(strlen(name) >= sizeof(dev->name));
6742 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6748 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6753 alloc_size = sizeof(struct net_device);
6755 /* ensure 32-byte alignment of private area */
6756 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6757 alloc_size += sizeof_priv;
6759 /* ensure 32-byte alignment of whole construct */
6760 alloc_size += NETDEV_ALIGN - 1;
6762 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6764 p = vzalloc(alloc_size);
6768 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6769 dev->padded = (char *)dev - (char *)p;
6771 dev->pcpu_refcnt = alloc_percpu(int);
6772 if (!dev->pcpu_refcnt)
6775 if (dev_addr_init(dev))
6781 dev_net_set(dev, &init_net);
6783 dev->gso_max_size = GSO_MAX_SIZE;
6784 dev->gso_max_segs = GSO_MAX_SEGS;
6785 dev->gso_min_segs = 0;
6787 INIT_LIST_HEAD(&dev->napi_list);
6788 INIT_LIST_HEAD(&dev->unreg_list);
6789 INIT_LIST_HEAD(&dev->close_list);
6790 INIT_LIST_HEAD(&dev->link_watch_list);
6791 INIT_LIST_HEAD(&dev->adj_list.upper);
6792 INIT_LIST_HEAD(&dev->adj_list.lower);
6793 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6794 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6795 INIT_LIST_HEAD(&dev->ptype_all);
6796 INIT_LIST_HEAD(&dev->ptype_specific);
6797 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
6800 dev->num_tx_queues = txqs;
6801 dev->real_num_tx_queues = txqs;
6802 if (netif_alloc_netdev_queues(dev))
6806 dev->num_rx_queues = rxqs;
6807 dev->real_num_rx_queues = rxqs;
6808 if (netif_alloc_rx_queues(dev))
6812 strcpy(dev->name, name);
6813 dev->name_assign_type = name_assign_type;
6814 dev->group = INIT_NETDEV_GROUP;
6815 if (!dev->ethtool_ops)
6816 dev->ethtool_ops = &default_ethtool_ops;
6824 free_percpu(dev->pcpu_refcnt);
6826 netdev_freemem(dev);
6829 EXPORT_SYMBOL(alloc_netdev_mqs);
6832 * free_netdev - free network device
6835 * This function does the last stage of destroying an allocated device
6836 * interface. The reference to the device object is released.
6837 * If this is the last reference then it will be freed.
6839 void free_netdev(struct net_device *dev)
6841 struct napi_struct *p, *n;
6843 release_net(dev_net(dev));
6845 netif_free_tx_queues(dev);
6850 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6852 /* Flush device addresses */
6853 dev_addr_flush(dev);
6855 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6858 free_percpu(dev->pcpu_refcnt);
6859 dev->pcpu_refcnt = NULL;
6861 /* Compatibility with error handling in drivers */
6862 if (dev->reg_state == NETREG_UNINITIALIZED) {
6863 netdev_freemem(dev);
6867 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6868 dev->reg_state = NETREG_RELEASED;
6870 /* will free via device release */
6871 put_device(&dev->dev);
6873 EXPORT_SYMBOL(free_netdev);
6876 * synchronize_net - Synchronize with packet receive processing
6878 * Wait for packets currently being received to be done.
6879 * Does not block later packets from starting.
6881 void synchronize_net(void)
6884 if (rtnl_is_locked())
6885 synchronize_rcu_expedited();
6889 EXPORT_SYMBOL(synchronize_net);
6892 * unregister_netdevice_queue - remove device from the kernel
6896 * This function shuts down a device interface and removes it
6897 * from the kernel tables.
6898 * If head not NULL, device is queued to be unregistered later.
6900 * Callers must hold the rtnl semaphore. You may want
6901 * unregister_netdev() instead of this.
6904 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6909 list_move_tail(&dev->unreg_list, head);
6911 rollback_registered(dev);
6912 /* Finish processing unregister after unlock */
6916 EXPORT_SYMBOL(unregister_netdevice_queue);
6919 * unregister_netdevice_many - unregister many devices
6920 * @head: list of devices
6922 * Note: As most callers use a stack allocated list_head,
6923 * we force a list_del() to make sure stack wont be corrupted later.
6925 void unregister_netdevice_many(struct list_head *head)
6927 struct net_device *dev;
6929 if (!list_empty(head)) {
6930 rollback_registered_many(head);
6931 list_for_each_entry(dev, head, unreg_list)
6936 EXPORT_SYMBOL(unregister_netdevice_many);
6939 * unregister_netdev - remove device from the kernel
6942 * This function shuts down a device interface and removes it
6943 * from the kernel tables.
6945 * This is just a wrapper for unregister_netdevice that takes
6946 * the rtnl semaphore. In general you want to use this and not
6947 * unregister_netdevice.
6949 void unregister_netdev(struct net_device *dev)
6952 unregister_netdevice(dev);
6955 EXPORT_SYMBOL(unregister_netdev);
6958 * dev_change_net_namespace - move device to different nethost namespace
6960 * @net: network namespace
6961 * @pat: If not NULL name pattern to try if the current device name
6962 * is already taken in the destination network namespace.
6964 * This function shuts down a device interface and moves it
6965 * to a new network namespace. On success 0 is returned, on
6966 * a failure a netagive errno code is returned.
6968 * Callers must hold the rtnl semaphore.
6971 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6977 /* Don't allow namespace local devices to be moved. */
6979 if (dev->features & NETIF_F_NETNS_LOCAL)
6982 /* Ensure the device has been registrered */
6983 if (dev->reg_state != NETREG_REGISTERED)
6986 /* Get out if there is nothing todo */
6988 if (net_eq(dev_net(dev), net))
6991 /* Pick the destination device name, and ensure
6992 * we can use it in the destination network namespace.
6995 if (__dev_get_by_name(net, dev->name)) {
6996 /* We get here if we can't use the current device name */
6999 if (dev_get_valid_name(net, dev, pat) < 0)
7004 * And now a mini version of register_netdevice unregister_netdevice.
7007 /* If device is running close it first. */
7010 /* And unlink it from device chain */
7012 unlist_netdevice(dev);
7016 /* Shutdown queueing discipline. */
7019 /* Notify protocols, that we are about to destroy
7020 this device. They should clean all the things.
7022 Note that dev->reg_state stays at NETREG_REGISTERED.
7023 This is wanted because this way 8021q and macvlan know
7024 the device is just moving and can keep their slaves up.
7026 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7028 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7029 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7032 * Flush the unicast and multicast chains
7037 /* Send a netdev-removed uevent to the old namespace */
7038 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7039 netdev_adjacent_del_links(dev);
7041 /* Actually switch the network namespace */
7042 dev_net_set(dev, net);
7044 /* If there is an ifindex conflict assign a new one */
7045 if (__dev_get_by_index(net, dev->ifindex)) {
7046 int iflink = (dev->iflink == dev->ifindex);
7047 dev->ifindex = dev_new_index(net);
7049 dev->iflink = dev->ifindex;
7052 /* Send a netdev-add uevent to the new namespace */
7053 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7054 netdev_adjacent_add_links(dev);
7056 /* Fixup kobjects */
7057 err = device_rename(&dev->dev, dev->name);
7060 /* Add the device back in the hashes */
7061 list_netdevice(dev);
7063 /* Notify protocols, that a new device appeared. */
7064 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7067 * Prevent userspace races by waiting until the network
7068 * device is fully setup before sending notifications.
7070 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7077 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7079 static int dev_cpu_callback(struct notifier_block *nfb,
7080 unsigned long action,
7083 struct sk_buff **list_skb;
7084 struct sk_buff *skb;
7085 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7086 struct softnet_data *sd, *oldsd;
7088 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7091 local_irq_disable();
7092 cpu = smp_processor_id();
7093 sd = &per_cpu(softnet_data, cpu);
7094 oldsd = &per_cpu(softnet_data, oldcpu);
7096 /* Find end of our completion_queue. */
7097 list_skb = &sd->completion_queue;
7099 list_skb = &(*list_skb)->next;
7100 /* Append completion queue from offline CPU. */
7101 *list_skb = oldsd->completion_queue;
7102 oldsd->completion_queue = NULL;
7104 /* Append output queue from offline CPU. */
7105 if (oldsd->output_queue) {
7106 *sd->output_queue_tailp = oldsd->output_queue;
7107 sd->output_queue_tailp = oldsd->output_queue_tailp;
7108 oldsd->output_queue = NULL;
7109 oldsd->output_queue_tailp = &oldsd->output_queue;
7111 /* Append NAPI poll list from offline CPU, with one exception :
7112 * process_backlog() must be called by cpu owning percpu backlog.
7113 * We properly handle process_queue & input_pkt_queue later.
7115 while (!list_empty(&oldsd->poll_list)) {
7116 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7120 list_del_init(&napi->poll_list);
7121 if (napi->poll == process_backlog)
7124 ____napi_schedule(sd, napi);
7127 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7130 /* Process offline CPU's input_pkt_queue */
7131 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7133 input_queue_head_incr(oldsd);
7135 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7137 input_queue_head_incr(oldsd);
7145 * netdev_increment_features - increment feature set by one
7146 * @all: current feature set
7147 * @one: new feature set
7148 * @mask: mask feature set
7150 * Computes a new feature set after adding a device with feature set
7151 * @one to the master device with current feature set @all. Will not
7152 * enable anything that is off in @mask. Returns the new feature set.
7154 netdev_features_t netdev_increment_features(netdev_features_t all,
7155 netdev_features_t one, netdev_features_t mask)
7157 if (mask & NETIF_F_GEN_CSUM)
7158 mask |= NETIF_F_ALL_CSUM;
7159 mask |= NETIF_F_VLAN_CHALLENGED;
7161 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7162 all &= one | ~NETIF_F_ALL_FOR_ALL;
7164 /* If one device supports hw checksumming, set for all. */
7165 if (all & NETIF_F_GEN_CSUM)
7166 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7170 EXPORT_SYMBOL(netdev_increment_features);
7172 static struct hlist_head * __net_init netdev_create_hash(void)
7175 struct hlist_head *hash;
7177 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7179 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7180 INIT_HLIST_HEAD(&hash[i]);
7185 /* Initialize per network namespace state */
7186 static int __net_init netdev_init(struct net *net)
7188 if (net != &init_net)
7189 INIT_LIST_HEAD(&net->dev_base_head);
7191 net->dev_name_head = netdev_create_hash();
7192 if (net->dev_name_head == NULL)
7195 net->dev_index_head = netdev_create_hash();
7196 if (net->dev_index_head == NULL)
7202 kfree(net->dev_name_head);
7208 * netdev_drivername - network driver for the device
7209 * @dev: network device
7211 * Determine network driver for device.
7213 const char *netdev_drivername(const struct net_device *dev)
7215 const struct device_driver *driver;
7216 const struct device *parent;
7217 const char *empty = "";
7219 parent = dev->dev.parent;
7223 driver = parent->driver;
7224 if (driver && driver->name)
7225 return driver->name;
7229 static void __netdev_printk(const char *level, const struct net_device *dev,
7230 struct va_format *vaf)
7232 if (dev && dev->dev.parent) {
7233 dev_printk_emit(level[1] - '0',
7236 dev_driver_string(dev->dev.parent),
7237 dev_name(dev->dev.parent),
7238 netdev_name(dev), netdev_reg_state(dev),
7241 printk("%s%s%s: %pV",
7242 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7244 printk("%s(NULL net_device): %pV", level, vaf);
7248 void netdev_printk(const char *level, const struct net_device *dev,
7249 const char *format, ...)
7251 struct va_format vaf;
7254 va_start(args, format);
7259 __netdev_printk(level, dev, &vaf);
7263 EXPORT_SYMBOL(netdev_printk);
7265 #define define_netdev_printk_level(func, level) \
7266 void func(const struct net_device *dev, const char *fmt, ...) \
7268 struct va_format vaf; \
7271 va_start(args, fmt); \
7276 __netdev_printk(level, dev, &vaf); \
7280 EXPORT_SYMBOL(func);
7282 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7283 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7284 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7285 define_netdev_printk_level(netdev_err, KERN_ERR);
7286 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7287 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7288 define_netdev_printk_level(netdev_info, KERN_INFO);
7290 static void __net_exit netdev_exit(struct net *net)
7292 kfree(net->dev_name_head);
7293 kfree(net->dev_index_head);
7296 static struct pernet_operations __net_initdata netdev_net_ops = {
7297 .init = netdev_init,
7298 .exit = netdev_exit,
7301 static void __net_exit default_device_exit(struct net *net)
7303 struct net_device *dev, *aux;
7305 * Push all migratable network devices back to the
7306 * initial network namespace
7309 for_each_netdev_safe(net, dev, aux) {
7311 char fb_name[IFNAMSIZ];
7313 /* Ignore unmoveable devices (i.e. loopback) */
7314 if (dev->features & NETIF_F_NETNS_LOCAL)
7317 /* Leave virtual devices for the generic cleanup */
7318 if (dev->rtnl_link_ops)
7321 /* Push remaining network devices to init_net */
7322 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7323 err = dev_change_net_namespace(dev, &init_net, fb_name);
7325 pr_emerg("%s: failed to move %s to init_net: %d\n",
7326 __func__, dev->name, err);
7333 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7335 /* Return with the rtnl_lock held when there are no network
7336 * devices unregistering in any network namespace in net_list.
7340 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7342 add_wait_queue(&netdev_unregistering_wq, &wait);
7344 unregistering = false;
7346 list_for_each_entry(net, net_list, exit_list) {
7347 if (net->dev_unreg_count > 0) {
7348 unregistering = true;
7356 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7358 remove_wait_queue(&netdev_unregistering_wq, &wait);
7361 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7363 /* At exit all network devices most be removed from a network
7364 * namespace. Do this in the reverse order of registration.
7365 * Do this across as many network namespaces as possible to
7366 * improve batching efficiency.
7368 struct net_device *dev;
7370 LIST_HEAD(dev_kill_list);
7372 /* To prevent network device cleanup code from dereferencing
7373 * loopback devices or network devices that have been freed
7374 * wait here for all pending unregistrations to complete,
7375 * before unregistring the loopback device and allowing the
7376 * network namespace be freed.
7378 * The netdev todo list containing all network devices
7379 * unregistrations that happen in default_device_exit_batch
7380 * will run in the rtnl_unlock() at the end of
7381 * default_device_exit_batch.
7383 rtnl_lock_unregistering(net_list);
7384 list_for_each_entry(net, net_list, exit_list) {
7385 for_each_netdev_reverse(net, dev) {
7386 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7387 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7389 unregister_netdevice_queue(dev, &dev_kill_list);
7392 unregister_netdevice_many(&dev_kill_list);
7396 static struct pernet_operations __net_initdata default_device_ops = {
7397 .exit = default_device_exit,
7398 .exit_batch = default_device_exit_batch,
7402 * Initialize the DEV module. At boot time this walks the device list and
7403 * unhooks any devices that fail to initialise (normally hardware not
7404 * present) and leaves us with a valid list of present and active devices.
7409 * This is called single threaded during boot, so no need
7410 * to take the rtnl semaphore.
7412 static int __init net_dev_init(void)
7414 int i, rc = -ENOMEM;
7416 BUG_ON(!dev_boot_phase);
7418 if (dev_proc_init())
7421 if (netdev_kobject_init())
7424 INIT_LIST_HEAD(&ptype_all);
7425 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7426 INIT_LIST_HEAD(&ptype_base[i]);
7428 INIT_LIST_HEAD(&offload_base);
7430 if (register_pernet_subsys(&netdev_net_ops))
7434 * Initialise the packet receive queues.
7437 for_each_possible_cpu(i) {
7438 struct softnet_data *sd = &per_cpu(softnet_data, i);
7440 skb_queue_head_init(&sd->input_pkt_queue);
7441 skb_queue_head_init(&sd->process_queue);
7442 INIT_LIST_HEAD(&sd->poll_list);
7443 sd->output_queue_tailp = &sd->output_queue;
7445 sd->csd.func = rps_trigger_softirq;
7450 sd->backlog.poll = process_backlog;
7451 sd->backlog.weight = weight_p;
7456 /* The loopback device is special if any other network devices
7457 * is present in a network namespace the loopback device must
7458 * be present. Since we now dynamically allocate and free the
7459 * loopback device ensure this invariant is maintained by
7460 * keeping the loopback device as the first device on the
7461 * list of network devices. Ensuring the loopback devices
7462 * is the first device that appears and the last network device
7465 if (register_pernet_device(&loopback_net_ops))
7468 if (register_pernet_device(&default_device_ops))
7471 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7472 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7474 hotcpu_notifier(dev_cpu_callback, 0);
7481 subsys_initcall(net_dev_init);