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 == '/' || *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 DEFINE_PER_CPU(int, xmit_recursion);
2852 EXPORT_SYMBOL(xmit_recursion);
2854 #define RECURSION_LIMIT 10
2857 * dev_loopback_xmit - loop back @skb
2858 * @skb: buffer to transmit
2860 int dev_loopback_xmit(struct sk_buff *skb)
2862 skb_reset_mac_header(skb);
2863 __skb_pull(skb, skb_network_offset(skb));
2864 skb->pkt_type = PACKET_LOOPBACK;
2865 skb->ip_summed = CHECKSUM_UNNECESSARY;
2866 WARN_ON(!skb_dst(skb));
2871 EXPORT_SYMBOL(dev_loopback_xmit);
2874 * __dev_queue_xmit - transmit a buffer
2875 * @skb: buffer to transmit
2876 * @accel_priv: private data used for L2 forwarding offload
2878 * Queue a buffer for transmission to a network device. The caller must
2879 * have set the device and priority and built the buffer before calling
2880 * this function. The function can be called from an interrupt.
2882 * A negative errno code is returned on a failure. A success does not
2883 * guarantee the frame will be transmitted as it may be dropped due
2884 * to congestion or traffic shaping.
2886 * -----------------------------------------------------------------------------------
2887 * I notice this method can also return errors from the queue disciplines,
2888 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2891 * Regardless of the return value, the skb is consumed, so it is currently
2892 * difficult to retry a send to this method. (You can bump the ref count
2893 * before sending to hold a reference for retry if you are careful.)
2895 * When calling this method, interrupts MUST be enabled. This is because
2896 * the BH enable code must have IRQs enabled so that it will not deadlock.
2899 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2901 struct net_device *dev = skb->dev;
2902 struct netdev_queue *txq;
2906 skb_reset_mac_header(skb);
2908 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
2909 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
2911 /* Disable soft irqs for various locks below. Also
2912 * stops preemption for RCU.
2916 skb_update_prio(skb);
2918 /* If device/qdisc don't need skb->dst, release it right now while
2919 * its hot in this cpu cache.
2921 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2926 txq = netdev_pick_tx(dev, skb, accel_priv);
2927 q = rcu_dereference_bh(txq->qdisc);
2929 #ifdef CONFIG_NET_CLS_ACT
2930 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2932 trace_net_dev_queue(skb);
2934 rc = __dev_xmit_skb(skb, q, dev, txq);
2938 /* The device has no queue. Common case for software devices:
2939 loopback, all the sorts of tunnels...
2941 Really, it is unlikely that netif_tx_lock protection is necessary
2942 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2944 However, it is possible, that they rely on protection
2947 Check this and shot the lock. It is not prone from deadlocks.
2948 Either shot noqueue qdisc, it is even simpler 8)
2950 if (dev->flags & IFF_UP) {
2951 int cpu = smp_processor_id(); /* ok because BHs are off */
2953 if (txq->xmit_lock_owner != cpu) {
2955 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2956 goto recursion_alert;
2958 skb = validate_xmit_skb(skb, dev);
2962 HARD_TX_LOCK(dev, txq, cpu);
2964 if (!netif_xmit_stopped(txq)) {
2965 __this_cpu_inc(xmit_recursion);
2966 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
2967 __this_cpu_dec(xmit_recursion);
2968 if (dev_xmit_complete(rc)) {
2969 HARD_TX_UNLOCK(dev, txq);
2973 HARD_TX_UNLOCK(dev, txq);
2974 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2977 /* Recursion is detected! It is possible,
2981 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2988 rcu_read_unlock_bh();
2990 atomic_long_inc(&dev->tx_dropped);
2991 kfree_skb_list(skb);
2994 rcu_read_unlock_bh();
2998 int dev_queue_xmit(struct sk_buff *skb)
3000 return __dev_queue_xmit(skb, NULL);
3002 EXPORT_SYMBOL(dev_queue_xmit);
3004 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3006 return __dev_queue_xmit(skb, accel_priv);
3008 EXPORT_SYMBOL(dev_queue_xmit_accel);
3011 /*=======================================================================
3013 =======================================================================*/
3015 int netdev_max_backlog __read_mostly = 1000;
3016 EXPORT_SYMBOL(netdev_max_backlog);
3018 int netdev_tstamp_prequeue __read_mostly = 1;
3019 int netdev_budget __read_mostly = 300;
3020 int weight_p __read_mostly = 64; /* old backlog weight */
3022 /* Called with irq disabled */
3023 static inline void ____napi_schedule(struct softnet_data *sd,
3024 struct napi_struct *napi)
3026 list_add_tail(&napi->poll_list, &sd->poll_list);
3027 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3032 /* One global table that all flow-based protocols share. */
3033 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3034 EXPORT_SYMBOL(rps_sock_flow_table);
3035 u32 rps_cpu_mask __read_mostly;
3036 EXPORT_SYMBOL(rps_cpu_mask);
3038 struct static_key rps_needed __read_mostly;
3040 static struct rps_dev_flow *
3041 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3042 struct rps_dev_flow *rflow, u16 next_cpu)
3044 if (next_cpu != RPS_NO_CPU) {
3045 #ifdef CONFIG_RFS_ACCEL
3046 struct netdev_rx_queue *rxqueue;
3047 struct rps_dev_flow_table *flow_table;
3048 struct rps_dev_flow *old_rflow;
3053 /* Should we steer this flow to a different hardware queue? */
3054 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3055 !(dev->features & NETIF_F_NTUPLE))
3057 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3058 if (rxq_index == skb_get_rx_queue(skb))
3061 rxqueue = dev->_rx + rxq_index;
3062 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3065 flow_id = skb_get_hash(skb) & flow_table->mask;
3066 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3067 rxq_index, flow_id);
3071 rflow = &flow_table->flows[flow_id];
3073 if (old_rflow->filter == rflow->filter)
3074 old_rflow->filter = RPS_NO_FILTER;
3078 per_cpu(softnet_data, next_cpu).input_queue_head;
3081 rflow->cpu = next_cpu;
3086 * get_rps_cpu is called from netif_receive_skb and returns the target
3087 * CPU from the RPS map of the receiving queue for a given skb.
3088 * rcu_read_lock must be held on entry.
3090 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3091 struct rps_dev_flow **rflowp)
3093 const struct rps_sock_flow_table *sock_flow_table;
3094 struct netdev_rx_queue *rxqueue = dev->_rx;
3095 struct rps_dev_flow_table *flow_table;
3096 struct rps_map *map;
3101 if (skb_rx_queue_recorded(skb)) {
3102 u16 index = skb_get_rx_queue(skb);
3104 if (unlikely(index >= dev->real_num_rx_queues)) {
3105 WARN_ONCE(dev->real_num_rx_queues > 1,
3106 "%s received packet on queue %u, but number "
3107 "of RX queues is %u\n",
3108 dev->name, index, dev->real_num_rx_queues);
3114 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3116 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3117 map = rcu_dereference(rxqueue->rps_map);
3118 if (!flow_table && !map)
3121 skb_reset_network_header(skb);
3122 hash = skb_get_hash(skb);
3126 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3127 if (flow_table && sock_flow_table) {
3128 struct rps_dev_flow *rflow;
3132 /* First check into global flow table if there is a match */
3133 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3134 if ((ident ^ hash) & ~rps_cpu_mask)
3137 next_cpu = ident & rps_cpu_mask;
3139 /* OK, now we know there is a match,
3140 * we can look at the local (per receive queue) flow table
3142 rflow = &flow_table->flows[hash & flow_table->mask];
3146 * If the desired CPU (where last recvmsg was done) is
3147 * different from current CPU (one in the rx-queue flow
3148 * table entry), switch if one of the following holds:
3149 * - Current CPU is unset (equal to RPS_NO_CPU).
3150 * - Current CPU is offline.
3151 * - The current CPU's queue tail has advanced beyond the
3152 * last packet that was enqueued using this table entry.
3153 * This guarantees that all previous packets for the flow
3154 * have been dequeued, thus preserving in order delivery.
3156 if (unlikely(tcpu != next_cpu) &&
3157 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3158 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3159 rflow->last_qtail)) >= 0)) {
3161 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3164 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3174 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3175 if (cpu_online(tcpu)) {
3185 #ifdef CONFIG_RFS_ACCEL
3188 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3189 * @dev: Device on which the filter was set
3190 * @rxq_index: RX queue index
3191 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3192 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3194 * Drivers that implement ndo_rx_flow_steer() should periodically call
3195 * this function for each installed filter and remove the filters for
3196 * which it returns %true.
3198 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3199 u32 flow_id, u16 filter_id)
3201 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3202 struct rps_dev_flow_table *flow_table;
3203 struct rps_dev_flow *rflow;
3208 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3209 if (flow_table && flow_id <= flow_table->mask) {
3210 rflow = &flow_table->flows[flow_id];
3211 cpu = ACCESS_ONCE(rflow->cpu);
3212 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3213 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3214 rflow->last_qtail) <
3215 (int)(10 * flow_table->mask)))
3221 EXPORT_SYMBOL(rps_may_expire_flow);
3223 #endif /* CONFIG_RFS_ACCEL */
3225 /* Called from hardirq (IPI) context */
3226 static void rps_trigger_softirq(void *data)
3228 struct softnet_data *sd = data;
3230 ____napi_schedule(sd, &sd->backlog);
3234 #endif /* CONFIG_RPS */
3237 * Check if this softnet_data structure is another cpu one
3238 * If yes, queue it to our IPI list and return 1
3241 static int rps_ipi_queued(struct softnet_data *sd)
3244 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3247 sd->rps_ipi_next = mysd->rps_ipi_list;
3248 mysd->rps_ipi_list = sd;
3250 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3253 #endif /* CONFIG_RPS */
3257 #ifdef CONFIG_NET_FLOW_LIMIT
3258 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3261 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3263 #ifdef CONFIG_NET_FLOW_LIMIT
3264 struct sd_flow_limit *fl;
3265 struct softnet_data *sd;
3266 unsigned int old_flow, new_flow;
3268 if (qlen < (netdev_max_backlog >> 1))
3271 sd = this_cpu_ptr(&softnet_data);
3274 fl = rcu_dereference(sd->flow_limit);
3276 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3277 old_flow = fl->history[fl->history_head];
3278 fl->history[fl->history_head] = new_flow;
3281 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3283 if (likely(fl->buckets[old_flow]))
3284 fl->buckets[old_flow]--;
3286 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3298 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3299 * queue (may be a remote CPU queue).
3301 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3302 unsigned int *qtail)
3304 struct softnet_data *sd;
3305 unsigned long flags;
3308 sd = &per_cpu(softnet_data, cpu);
3310 local_irq_save(flags);
3313 qlen = skb_queue_len(&sd->input_pkt_queue);
3314 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3317 __skb_queue_tail(&sd->input_pkt_queue, skb);
3318 input_queue_tail_incr_save(sd, qtail);
3320 local_irq_restore(flags);
3321 return NET_RX_SUCCESS;
3324 /* Schedule NAPI for backlog device
3325 * We can use non atomic operation since we own the queue lock
3327 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3328 if (!rps_ipi_queued(sd))
3329 ____napi_schedule(sd, &sd->backlog);
3337 local_irq_restore(flags);
3339 atomic_long_inc(&skb->dev->rx_dropped);
3344 static int netif_rx_internal(struct sk_buff *skb)
3348 net_timestamp_check(netdev_tstamp_prequeue, skb);
3350 trace_netif_rx(skb);
3352 if (static_key_false(&rps_needed)) {
3353 struct rps_dev_flow voidflow, *rflow = &voidflow;
3359 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3361 cpu = smp_processor_id();
3363 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3371 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3378 * netif_rx - post buffer to the network code
3379 * @skb: buffer to post
3381 * This function receives a packet from a device driver and queues it for
3382 * the upper (protocol) levels to process. It always succeeds. The buffer
3383 * may be dropped during processing for congestion control or by the
3387 * NET_RX_SUCCESS (no congestion)
3388 * NET_RX_DROP (packet was dropped)
3392 int netif_rx(struct sk_buff *skb)
3394 trace_netif_rx_entry(skb);
3396 return netif_rx_internal(skb);
3398 EXPORT_SYMBOL(netif_rx);
3400 int netif_rx_ni(struct sk_buff *skb)
3404 trace_netif_rx_ni_entry(skb);
3407 err = netif_rx_internal(skb);
3408 if (local_softirq_pending())
3414 EXPORT_SYMBOL(netif_rx_ni);
3416 static void net_tx_action(struct softirq_action *h)
3418 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3420 if (sd->completion_queue) {
3421 struct sk_buff *clist;
3423 local_irq_disable();
3424 clist = sd->completion_queue;
3425 sd->completion_queue = NULL;
3429 struct sk_buff *skb = clist;
3430 clist = clist->next;
3432 WARN_ON(atomic_read(&skb->users));
3433 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3434 trace_consume_skb(skb);
3436 trace_kfree_skb(skb, net_tx_action);
3441 if (sd->output_queue) {
3444 local_irq_disable();
3445 head = sd->output_queue;
3446 sd->output_queue = NULL;
3447 sd->output_queue_tailp = &sd->output_queue;
3451 struct Qdisc *q = head;
3452 spinlock_t *root_lock;
3454 head = head->next_sched;
3456 root_lock = qdisc_lock(q);
3457 if (spin_trylock(root_lock)) {
3458 smp_mb__before_atomic();
3459 clear_bit(__QDISC_STATE_SCHED,
3462 spin_unlock(root_lock);
3464 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3466 __netif_reschedule(q);
3468 smp_mb__before_atomic();
3469 clear_bit(__QDISC_STATE_SCHED,
3477 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3478 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3479 /* This hook is defined here for ATM LANE */
3480 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3481 unsigned char *addr) __read_mostly;
3482 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3485 #ifdef CONFIG_NET_CLS_ACT
3486 /* TODO: Maybe we should just force sch_ingress to be compiled in
3487 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3488 * a compare and 2 stores extra right now if we dont have it on
3489 * but have CONFIG_NET_CLS_ACT
3490 * NOTE: This doesn't stop any functionality; if you dont have
3491 * the ingress scheduler, you just can't add policies on ingress.
3494 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3496 struct net_device *dev = skb->dev;
3497 u32 ttl = G_TC_RTTL(skb->tc_verd);
3498 int result = TC_ACT_OK;
3501 if (unlikely(MAX_RED_LOOP < ttl++)) {
3502 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3503 skb->skb_iif, dev->ifindex);
3507 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3508 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3510 q = rcu_dereference(rxq->qdisc);
3511 if (q != &noop_qdisc) {
3512 spin_lock(qdisc_lock(q));
3513 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3514 result = qdisc_enqueue_root(skb, q);
3515 spin_unlock(qdisc_lock(q));
3521 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3522 struct packet_type **pt_prev,
3523 int *ret, struct net_device *orig_dev)
3525 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3527 if (!rxq || rcu_access_pointer(rxq->qdisc) == &noop_qdisc)
3531 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3535 switch (ing_filter(skb, rxq)) {
3549 * netdev_rx_handler_register - register receive handler
3550 * @dev: device to register a handler for
3551 * @rx_handler: receive handler to register
3552 * @rx_handler_data: data pointer that is used by rx handler
3554 * Register a receive handler for a device. This handler will then be
3555 * called from __netif_receive_skb. A negative errno code is returned
3558 * The caller must hold the rtnl_mutex.
3560 * For a general description of rx_handler, see enum rx_handler_result.
3562 int netdev_rx_handler_register(struct net_device *dev,
3563 rx_handler_func_t *rx_handler,
3564 void *rx_handler_data)
3568 if (dev->rx_handler)
3571 /* Note: rx_handler_data must be set before rx_handler */
3572 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3573 rcu_assign_pointer(dev->rx_handler, rx_handler);
3577 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3580 * netdev_rx_handler_unregister - unregister receive handler
3581 * @dev: device to unregister a handler from
3583 * Unregister a receive handler from a device.
3585 * The caller must hold the rtnl_mutex.
3587 void netdev_rx_handler_unregister(struct net_device *dev)
3591 RCU_INIT_POINTER(dev->rx_handler, NULL);
3592 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3593 * section has a guarantee to see a non NULL rx_handler_data
3597 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3599 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3602 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3603 * the special handling of PFMEMALLOC skbs.
3605 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3607 switch (skb->protocol) {
3608 case htons(ETH_P_ARP):
3609 case htons(ETH_P_IP):
3610 case htons(ETH_P_IPV6):
3611 case htons(ETH_P_8021Q):
3612 case htons(ETH_P_8021AD):
3619 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3621 struct packet_type *ptype, *pt_prev;
3622 rx_handler_func_t *rx_handler;
3623 struct net_device *orig_dev;
3624 bool deliver_exact = false;
3625 int ret = NET_RX_DROP;
3628 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3630 trace_netif_receive_skb(skb);
3632 orig_dev = skb->dev;
3634 skb_reset_network_header(skb);
3635 if (!skb_transport_header_was_set(skb))
3636 skb_reset_transport_header(skb);
3637 skb_reset_mac_len(skb);
3644 skb->skb_iif = skb->dev->ifindex;
3646 __this_cpu_inc(softnet_data.processed);
3648 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3649 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3650 skb = skb_vlan_untag(skb);
3655 #ifdef CONFIG_NET_CLS_ACT
3656 if (skb->tc_verd & TC_NCLS) {
3657 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3665 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3667 ret = deliver_skb(skb, pt_prev, orig_dev);
3671 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3673 ret = deliver_skb(skb, pt_prev, orig_dev);
3678 #ifdef CONFIG_NET_CLS_ACT
3679 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3685 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3688 if (skb_vlan_tag_present(skb)) {
3690 ret = deliver_skb(skb, pt_prev, orig_dev);
3693 if (vlan_do_receive(&skb))
3695 else if (unlikely(!skb))
3699 rx_handler = rcu_dereference(skb->dev->rx_handler);
3702 ret = deliver_skb(skb, pt_prev, orig_dev);
3705 switch (rx_handler(&skb)) {
3706 case RX_HANDLER_CONSUMED:
3707 ret = NET_RX_SUCCESS;
3709 case RX_HANDLER_ANOTHER:
3711 case RX_HANDLER_EXACT:
3712 deliver_exact = true;
3713 case RX_HANDLER_PASS:
3720 if (unlikely(skb_vlan_tag_present(skb))) {
3721 if (skb_vlan_tag_get_id(skb))
3722 skb->pkt_type = PACKET_OTHERHOST;
3723 /* Note: we might in the future use prio bits
3724 * and set skb->priority like in vlan_do_receive()
3725 * For the time being, just ignore Priority Code Point
3730 type = skb->protocol;
3732 /* deliver only exact match when indicated */
3733 if (likely(!deliver_exact)) {
3734 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3735 &ptype_base[ntohs(type) &
3739 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3740 &orig_dev->ptype_specific);
3742 if (unlikely(skb->dev != orig_dev)) {
3743 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3744 &skb->dev->ptype_specific);
3748 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3751 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3754 atomic_long_inc(&skb->dev->rx_dropped);
3756 /* Jamal, now you will not able to escape explaining
3757 * me how you were going to use this. :-)
3767 static int __netif_receive_skb(struct sk_buff *skb)
3771 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3772 unsigned long pflags = current->flags;
3775 * PFMEMALLOC skbs are special, they should
3776 * - be delivered to SOCK_MEMALLOC sockets only
3777 * - stay away from userspace
3778 * - have bounded memory usage
3780 * Use PF_MEMALLOC as this saves us from propagating the allocation
3781 * context down to all allocation sites.
3783 current->flags |= PF_MEMALLOC;
3784 ret = __netif_receive_skb_core(skb, true);
3785 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3787 ret = __netif_receive_skb_core(skb, false);
3792 static int netif_receive_skb_internal(struct sk_buff *skb)
3794 net_timestamp_check(netdev_tstamp_prequeue, skb);
3796 if (skb_defer_rx_timestamp(skb))
3797 return NET_RX_SUCCESS;
3800 if (static_key_false(&rps_needed)) {
3801 struct rps_dev_flow voidflow, *rflow = &voidflow;
3806 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3809 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3816 return __netif_receive_skb(skb);
3820 * netif_receive_skb - process receive buffer from network
3821 * @skb: buffer to process
3823 * netif_receive_skb() is the main receive data processing function.
3824 * It always succeeds. The buffer may be dropped during processing
3825 * for congestion control or by the protocol layers.
3827 * This function may only be called from softirq context and interrupts
3828 * should be enabled.
3830 * Return values (usually ignored):
3831 * NET_RX_SUCCESS: no congestion
3832 * NET_RX_DROP: packet was dropped
3834 int netif_receive_skb(struct sk_buff *skb)
3836 trace_netif_receive_skb_entry(skb);
3838 return netif_receive_skb_internal(skb);
3840 EXPORT_SYMBOL(netif_receive_skb);
3842 /* Network device is going away, flush any packets still pending
3843 * Called with irqs disabled.
3845 static void flush_backlog(void *arg)
3847 struct net_device *dev = arg;
3848 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3849 struct sk_buff *skb, *tmp;
3852 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3853 if (skb->dev == dev) {
3854 __skb_unlink(skb, &sd->input_pkt_queue);
3856 input_queue_head_incr(sd);
3861 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3862 if (skb->dev == dev) {
3863 __skb_unlink(skb, &sd->process_queue);
3865 input_queue_head_incr(sd);
3870 static int napi_gro_complete(struct sk_buff *skb)
3872 struct packet_offload *ptype;
3873 __be16 type = skb->protocol;
3874 struct list_head *head = &offload_base;
3877 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3879 if (NAPI_GRO_CB(skb)->count == 1) {
3880 skb_shinfo(skb)->gso_size = 0;
3885 list_for_each_entry_rcu(ptype, head, list) {
3886 if (ptype->type != type || !ptype->callbacks.gro_complete)
3889 err = ptype->callbacks.gro_complete(skb, 0);
3895 WARN_ON(&ptype->list == head);
3897 return NET_RX_SUCCESS;
3901 return netif_receive_skb_internal(skb);
3904 /* napi->gro_list contains packets ordered by age.
3905 * youngest packets at the head of it.
3906 * Complete skbs in reverse order to reduce latencies.
3908 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3910 struct sk_buff *skb, *prev = NULL;
3912 /* scan list and build reverse chain */
3913 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3918 for (skb = prev; skb; skb = prev) {
3921 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3925 napi_gro_complete(skb);
3929 napi->gro_list = NULL;
3931 EXPORT_SYMBOL(napi_gro_flush);
3933 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3936 unsigned int maclen = skb->dev->hard_header_len;
3937 u32 hash = skb_get_hash_raw(skb);
3939 for (p = napi->gro_list; p; p = p->next) {
3940 unsigned long diffs;
3942 NAPI_GRO_CB(p)->flush = 0;
3944 if (hash != skb_get_hash_raw(p)) {
3945 NAPI_GRO_CB(p)->same_flow = 0;
3949 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3950 diffs |= p->vlan_tci ^ skb->vlan_tci;
3951 if (maclen == ETH_HLEN)
3952 diffs |= compare_ether_header(skb_mac_header(p),
3953 skb_mac_header(skb));
3955 diffs = memcmp(skb_mac_header(p),
3956 skb_mac_header(skb),
3958 NAPI_GRO_CB(p)->same_flow = !diffs;
3962 static void skb_gro_reset_offset(struct sk_buff *skb)
3964 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3965 const skb_frag_t *frag0 = &pinfo->frags[0];
3967 NAPI_GRO_CB(skb)->data_offset = 0;
3968 NAPI_GRO_CB(skb)->frag0 = NULL;
3969 NAPI_GRO_CB(skb)->frag0_len = 0;
3971 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3973 !PageHighMem(skb_frag_page(frag0))) {
3974 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3975 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3979 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
3981 struct skb_shared_info *pinfo = skb_shinfo(skb);
3983 BUG_ON(skb->end - skb->tail < grow);
3985 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3987 skb->data_len -= grow;
3990 pinfo->frags[0].page_offset += grow;
3991 skb_frag_size_sub(&pinfo->frags[0], grow);
3993 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
3994 skb_frag_unref(skb, 0);
3995 memmove(pinfo->frags, pinfo->frags + 1,
3996 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4000 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4002 struct sk_buff **pp = NULL;
4003 struct packet_offload *ptype;
4004 __be16 type = skb->protocol;
4005 struct list_head *head = &offload_base;
4007 enum gro_result ret;
4010 if (!(skb->dev->features & NETIF_F_GRO))
4013 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4016 gro_list_prepare(napi, skb);
4019 list_for_each_entry_rcu(ptype, head, list) {
4020 if (ptype->type != type || !ptype->callbacks.gro_receive)
4023 skb_set_network_header(skb, skb_gro_offset(skb));
4024 skb_reset_mac_len(skb);
4025 NAPI_GRO_CB(skb)->same_flow = 0;
4026 NAPI_GRO_CB(skb)->flush = 0;
4027 NAPI_GRO_CB(skb)->free = 0;
4028 NAPI_GRO_CB(skb)->udp_mark = 0;
4029 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4031 /* Setup for GRO checksum validation */
4032 switch (skb->ip_summed) {
4033 case CHECKSUM_COMPLETE:
4034 NAPI_GRO_CB(skb)->csum = skb->csum;
4035 NAPI_GRO_CB(skb)->csum_valid = 1;
4036 NAPI_GRO_CB(skb)->csum_cnt = 0;
4038 case CHECKSUM_UNNECESSARY:
4039 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4040 NAPI_GRO_CB(skb)->csum_valid = 0;
4043 NAPI_GRO_CB(skb)->csum_cnt = 0;
4044 NAPI_GRO_CB(skb)->csum_valid = 0;
4047 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4052 if (&ptype->list == head)
4055 same_flow = NAPI_GRO_CB(skb)->same_flow;
4056 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4059 struct sk_buff *nskb = *pp;
4063 napi_gro_complete(nskb);
4070 if (NAPI_GRO_CB(skb)->flush)
4073 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4074 struct sk_buff *nskb = napi->gro_list;
4076 /* locate the end of the list to select the 'oldest' flow */
4077 while (nskb->next) {
4083 napi_gro_complete(nskb);
4087 NAPI_GRO_CB(skb)->count = 1;
4088 NAPI_GRO_CB(skb)->age = jiffies;
4089 NAPI_GRO_CB(skb)->last = skb;
4090 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4091 skb->next = napi->gro_list;
4092 napi->gro_list = skb;
4096 grow = skb_gro_offset(skb) - skb_headlen(skb);
4098 gro_pull_from_frag0(skb, grow);
4107 struct packet_offload *gro_find_receive_by_type(__be16 type)
4109 struct list_head *offload_head = &offload_base;
4110 struct packet_offload *ptype;
4112 list_for_each_entry_rcu(ptype, offload_head, list) {
4113 if (ptype->type != type || !ptype->callbacks.gro_receive)
4119 EXPORT_SYMBOL(gro_find_receive_by_type);
4121 struct packet_offload *gro_find_complete_by_type(__be16 type)
4123 struct list_head *offload_head = &offload_base;
4124 struct packet_offload *ptype;
4126 list_for_each_entry_rcu(ptype, offload_head, list) {
4127 if (ptype->type != type || !ptype->callbacks.gro_complete)
4133 EXPORT_SYMBOL(gro_find_complete_by_type);
4135 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4139 if (netif_receive_skb_internal(skb))
4147 case GRO_MERGED_FREE:
4148 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4149 kmem_cache_free(skbuff_head_cache, skb);
4162 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4164 trace_napi_gro_receive_entry(skb);
4166 skb_gro_reset_offset(skb);
4168 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4170 EXPORT_SYMBOL(napi_gro_receive);
4172 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4174 if (unlikely(skb->pfmemalloc)) {
4178 __skb_pull(skb, skb_headlen(skb));
4179 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4180 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4182 skb->dev = napi->dev;
4184 skb->encapsulation = 0;
4185 skb_shinfo(skb)->gso_type = 0;
4186 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4191 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4193 struct sk_buff *skb = napi->skb;
4196 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4201 EXPORT_SYMBOL(napi_get_frags);
4203 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4204 struct sk_buff *skb,
4210 __skb_push(skb, ETH_HLEN);
4211 skb->protocol = eth_type_trans(skb, skb->dev);
4212 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4217 case GRO_MERGED_FREE:
4218 napi_reuse_skb(napi, skb);
4228 /* Upper GRO stack assumes network header starts at gro_offset=0
4229 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4230 * We copy ethernet header into skb->data to have a common layout.
4232 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4234 struct sk_buff *skb = napi->skb;
4235 const struct ethhdr *eth;
4236 unsigned int hlen = sizeof(*eth);
4240 skb_reset_mac_header(skb);
4241 skb_gro_reset_offset(skb);
4243 eth = skb_gro_header_fast(skb, 0);
4244 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4245 eth = skb_gro_header_slow(skb, hlen, 0);
4246 if (unlikely(!eth)) {
4247 napi_reuse_skb(napi, skb);
4251 gro_pull_from_frag0(skb, hlen);
4252 NAPI_GRO_CB(skb)->frag0 += hlen;
4253 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4255 __skb_pull(skb, hlen);
4258 * This works because the only protocols we care about don't require
4260 * We'll fix it up properly in napi_frags_finish()
4262 skb->protocol = eth->h_proto;
4267 gro_result_t napi_gro_frags(struct napi_struct *napi)
4269 struct sk_buff *skb = napi_frags_skb(napi);
4274 trace_napi_gro_frags_entry(skb);
4276 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4278 EXPORT_SYMBOL(napi_gro_frags);
4280 /* Compute the checksum from gro_offset and return the folded value
4281 * after adding in any pseudo checksum.
4283 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4288 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4290 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4291 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4293 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4294 !skb->csum_complete_sw)
4295 netdev_rx_csum_fault(skb->dev);
4298 NAPI_GRO_CB(skb)->csum = wsum;
4299 NAPI_GRO_CB(skb)->csum_valid = 1;
4303 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4306 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4307 * Note: called with local irq disabled, but exits with local irq enabled.
4309 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4312 struct softnet_data *remsd = sd->rps_ipi_list;
4315 sd->rps_ipi_list = NULL;
4319 /* Send pending IPI's to kick RPS processing on remote cpus. */
4321 struct softnet_data *next = remsd->rps_ipi_next;
4323 if (cpu_online(remsd->cpu))
4324 smp_call_function_single_async(remsd->cpu,
4333 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4336 return sd->rps_ipi_list != NULL;
4342 static int process_backlog(struct napi_struct *napi, int quota)
4345 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4347 /* Check if we have pending ipi, its better to send them now,
4348 * not waiting net_rx_action() end.
4350 if (sd_has_rps_ipi_waiting(sd)) {
4351 local_irq_disable();
4352 net_rps_action_and_irq_enable(sd);
4355 napi->weight = weight_p;
4356 local_irq_disable();
4358 struct sk_buff *skb;
4360 while ((skb = __skb_dequeue(&sd->process_queue))) {
4362 __netif_receive_skb(skb);
4363 local_irq_disable();
4364 input_queue_head_incr(sd);
4365 if (++work >= quota) {
4372 if (skb_queue_empty(&sd->input_pkt_queue)) {
4374 * Inline a custom version of __napi_complete().
4375 * only current cpu owns and manipulates this napi,
4376 * and NAPI_STATE_SCHED is the only possible flag set
4378 * We can use a plain write instead of clear_bit(),
4379 * and we dont need an smp_mb() memory barrier.
4387 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4388 &sd->process_queue);
4397 * __napi_schedule - schedule for receive
4398 * @n: entry to schedule
4400 * The entry's receive function will be scheduled to run.
4401 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4403 void __napi_schedule(struct napi_struct *n)
4405 unsigned long flags;
4407 local_irq_save(flags);
4408 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4409 local_irq_restore(flags);
4411 EXPORT_SYMBOL(__napi_schedule);
4414 * __napi_schedule_irqoff - schedule for receive
4415 * @n: entry to schedule
4417 * Variant of __napi_schedule() assuming hard irqs are masked
4419 void __napi_schedule_irqoff(struct napi_struct *n)
4421 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4423 EXPORT_SYMBOL(__napi_schedule_irqoff);
4425 void __napi_complete(struct napi_struct *n)
4427 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4429 list_del_init(&n->poll_list);
4430 smp_mb__before_atomic();
4431 clear_bit(NAPI_STATE_SCHED, &n->state);
4433 EXPORT_SYMBOL(__napi_complete);
4435 void napi_complete_done(struct napi_struct *n, int work_done)
4437 unsigned long flags;
4440 * don't let napi dequeue from the cpu poll list
4441 * just in case its running on a different cpu
4443 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4447 unsigned long timeout = 0;
4450 timeout = n->dev->gro_flush_timeout;
4453 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4454 HRTIMER_MODE_REL_PINNED);
4456 napi_gro_flush(n, false);
4458 if (likely(list_empty(&n->poll_list))) {
4459 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4461 /* If n->poll_list is not empty, we need to mask irqs */
4462 local_irq_save(flags);
4464 local_irq_restore(flags);
4467 EXPORT_SYMBOL(napi_complete_done);
4469 /* must be called under rcu_read_lock(), as we dont take a reference */
4470 struct napi_struct *napi_by_id(unsigned int napi_id)
4472 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4473 struct napi_struct *napi;
4475 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4476 if (napi->napi_id == napi_id)
4481 EXPORT_SYMBOL_GPL(napi_by_id);
4483 void napi_hash_add(struct napi_struct *napi)
4485 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4487 spin_lock(&napi_hash_lock);
4489 /* 0 is not a valid id, we also skip an id that is taken
4490 * we expect both events to be extremely rare
4493 while (!napi->napi_id) {
4494 napi->napi_id = ++napi_gen_id;
4495 if (napi_by_id(napi->napi_id))
4499 hlist_add_head_rcu(&napi->napi_hash_node,
4500 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4502 spin_unlock(&napi_hash_lock);
4505 EXPORT_SYMBOL_GPL(napi_hash_add);
4507 /* Warning : caller is responsible to make sure rcu grace period
4508 * is respected before freeing memory containing @napi
4510 void napi_hash_del(struct napi_struct *napi)
4512 spin_lock(&napi_hash_lock);
4514 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4515 hlist_del_rcu(&napi->napi_hash_node);
4517 spin_unlock(&napi_hash_lock);
4519 EXPORT_SYMBOL_GPL(napi_hash_del);
4521 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4523 struct napi_struct *napi;
4525 napi = container_of(timer, struct napi_struct, timer);
4527 napi_schedule(napi);
4529 return HRTIMER_NORESTART;
4532 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4533 int (*poll)(struct napi_struct *, int), int weight)
4535 INIT_LIST_HEAD(&napi->poll_list);
4536 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4537 napi->timer.function = napi_watchdog;
4538 napi->gro_count = 0;
4539 napi->gro_list = NULL;
4542 if (weight > NAPI_POLL_WEIGHT)
4543 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4545 napi->weight = weight;
4546 list_add(&napi->dev_list, &dev->napi_list);
4548 #ifdef CONFIG_NETPOLL
4549 spin_lock_init(&napi->poll_lock);
4550 napi->poll_owner = -1;
4552 set_bit(NAPI_STATE_SCHED, &napi->state);
4554 EXPORT_SYMBOL(netif_napi_add);
4556 void napi_disable(struct napi_struct *n)
4559 set_bit(NAPI_STATE_DISABLE, &n->state);
4561 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4564 hrtimer_cancel(&n->timer);
4566 clear_bit(NAPI_STATE_DISABLE, &n->state);
4568 EXPORT_SYMBOL(napi_disable);
4570 void netif_napi_del(struct napi_struct *napi)
4572 list_del_init(&napi->dev_list);
4573 napi_free_frags(napi);
4575 kfree_skb_list(napi->gro_list);
4576 napi->gro_list = NULL;
4577 napi->gro_count = 0;
4579 EXPORT_SYMBOL(netif_napi_del);
4581 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4586 list_del_init(&n->poll_list);
4588 have = netpoll_poll_lock(n);
4592 /* This NAPI_STATE_SCHED test is for avoiding a race
4593 * with netpoll's poll_napi(). Only the entity which
4594 * obtains the lock and sees NAPI_STATE_SCHED set will
4595 * actually make the ->poll() call. Therefore we avoid
4596 * accidentally calling ->poll() when NAPI is not scheduled.
4599 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4600 work = n->poll(n, weight);
4604 WARN_ON_ONCE(work > weight);
4606 if (likely(work < weight))
4609 /* Drivers must not modify the NAPI state if they
4610 * consume the entire weight. In such cases this code
4611 * still "owns" the NAPI instance and therefore can
4612 * move the instance around on the list at-will.
4614 if (unlikely(napi_disable_pending(n))) {
4620 /* flush too old packets
4621 * If HZ < 1000, flush all packets.
4623 napi_gro_flush(n, HZ >= 1000);
4626 /* Some drivers may have called napi_schedule
4627 * prior to exhausting their budget.
4629 if (unlikely(!list_empty(&n->poll_list))) {
4630 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4631 n->dev ? n->dev->name : "backlog");
4635 list_add_tail(&n->poll_list, repoll);
4638 netpoll_poll_unlock(have);
4643 static void net_rx_action(struct softirq_action *h)
4645 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4646 unsigned long time_limit = jiffies + 2;
4647 int budget = netdev_budget;
4651 local_irq_disable();
4652 list_splice_init(&sd->poll_list, &list);
4656 struct napi_struct *n;
4658 if (list_empty(&list)) {
4659 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4664 n = list_first_entry(&list, struct napi_struct, poll_list);
4665 budget -= napi_poll(n, &repoll);
4667 /* If softirq window is exhausted then punt.
4668 * Allow this to run for 2 jiffies since which will allow
4669 * an average latency of 1.5/HZ.
4671 if (unlikely(budget <= 0 ||
4672 time_after_eq(jiffies, time_limit))) {
4678 local_irq_disable();
4680 list_splice_tail_init(&sd->poll_list, &list);
4681 list_splice_tail(&repoll, &list);
4682 list_splice(&list, &sd->poll_list);
4683 if (!list_empty(&sd->poll_list))
4684 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4686 net_rps_action_and_irq_enable(sd);
4689 struct netdev_adjacent {
4690 struct net_device *dev;
4692 /* upper master flag, there can only be one master device per list */
4695 /* counter for the number of times this device was added to us */
4698 /* private field for the users */
4701 struct list_head list;
4702 struct rcu_head rcu;
4705 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4706 struct net_device *adj_dev,
4707 struct list_head *adj_list)
4709 struct netdev_adjacent *adj;
4711 list_for_each_entry(adj, adj_list, list) {
4712 if (adj->dev == adj_dev)
4719 * netdev_has_upper_dev - Check if device is linked to an upper device
4721 * @upper_dev: upper device to check
4723 * Find out if a device is linked to specified upper device and return true
4724 * in case it is. Note that this checks only immediate upper device,
4725 * not through a complete stack of devices. The caller must hold the RTNL lock.
4727 bool netdev_has_upper_dev(struct net_device *dev,
4728 struct net_device *upper_dev)
4732 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4734 EXPORT_SYMBOL(netdev_has_upper_dev);
4737 * netdev_has_any_upper_dev - Check if device is linked to some device
4740 * Find out if a device is linked to an upper device and return true in case
4741 * it is. The caller must hold the RTNL lock.
4743 static bool netdev_has_any_upper_dev(struct net_device *dev)
4747 return !list_empty(&dev->all_adj_list.upper);
4751 * netdev_master_upper_dev_get - Get master upper device
4754 * Find a master upper device and return pointer to it or NULL in case
4755 * it's not there. The caller must hold the RTNL lock.
4757 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4759 struct netdev_adjacent *upper;
4763 if (list_empty(&dev->adj_list.upper))
4766 upper = list_first_entry(&dev->adj_list.upper,
4767 struct netdev_adjacent, list);
4768 if (likely(upper->master))
4772 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4774 void *netdev_adjacent_get_private(struct list_head *adj_list)
4776 struct netdev_adjacent *adj;
4778 adj = list_entry(adj_list, struct netdev_adjacent, list);
4780 return adj->private;
4782 EXPORT_SYMBOL(netdev_adjacent_get_private);
4785 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4787 * @iter: list_head ** of the current position
4789 * Gets the next device from the dev's upper list, starting from iter
4790 * position. The caller must hold RCU read lock.
4792 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4793 struct list_head **iter)
4795 struct netdev_adjacent *upper;
4797 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4799 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4801 if (&upper->list == &dev->adj_list.upper)
4804 *iter = &upper->list;
4808 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4811 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4813 * @iter: list_head ** of the current position
4815 * Gets the next device from the dev's upper list, starting from iter
4816 * position. The caller must hold RCU read lock.
4818 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4819 struct list_head **iter)
4821 struct netdev_adjacent *upper;
4823 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4825 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4827 if (&upper->list == &dev->all_adj_list.upper)
4830 *iter = &upper->list;
4834 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4837 * netdev_lower_get_next_private - Get the next ->private from the
4838 * lower neighbour list
4840 * @iter: list_head ** of the current position
4842 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4843 * list, starting from iter position. The caller must hold either hold the
4844 * RTNL lock or its own locking that guarantees that the neighbour lower
4845 * list will remain unchainged.
4847 void *netdev_lower_get_next_private(struct net_device *dev,
4848 struct list_head **iter)
4850 struct netdev_adjacent *lower;
4852 lower = list_entry(*iter, struct netdev_adjacent, list);
4854 if (&lower->list == &dev->adj_list.lower)
4857 *iter = lower->list.next;
4859 return lower->private;
4861 EXPORT_SYMBOL(netdev_lower_get_next_private);
4864 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4865 * lower neighbour list, RCU
4868 * @iter: list_head ** of the current position
4870 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4871 * list, starting from iter position. The caller must hold RCU read lock.
4873 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4874 struct list_head **iter)
4876 struct netdev_adjacent *lower;
4878 WARN_ON_ONCE(!rcu_read_lock_held());
4880 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4882 if (&lower->list == &dev->adj_list.lower)
4885 *iter = &lower->list;
4887 return lower->private;
4889 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4892 * netdev_lower_get_next - Get the next device from the lower neighbour
4895 * @iter: list_head ** of the current position
4897 * Gets the next netdev_adjacent from the dev's lower neighbour
4898 * list, starting from iter position. The caller must hold RTNL lock or
4899 * its own locking that guarantees that the neighbour lower
4900 * list will remain unchainged.
4902 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4904 struct netdev_adjacent *lower;
4906 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4908 if (&lower->list == &dev->adj_list.lower)
4911 *iter = &lower->list;
4915 EXPORT_SYMBOL(netdev_lower_get_next);
4918 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4919 * lower neighbour list, RCU
4923 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4924 * list. The caller must hold RCU read lock.
4926 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4928 struct netdev_adjacent *lower;
4930 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4931 struct netdev_adjacent, list);
4933 return lower->private;
4936 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4939 * netdev_master_upper_dev_get_rcu - Get master upper device
4942 * Find a master upper device and return pointer to it or NULL in case
4943 * it's not there. The caller must hold the RCU read lock.
4945 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4947 struct netdev_adjacent *upper;
4949 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4950 struct netdev_adjacent, list);
4951 if (upper && likely(upper->master))
4955 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4957 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4958 struct net_device *adj_dev,
4959 struct list_head *dev_list)
4961 char linkname[IFNAMSIZ+7];
4962 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4963 "upper_%s" : "lower_%s", adj_dev->name);
4964 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4967 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4969 struct list_head *dev_list)
4971 char linkname[IFNAMSIZ+7];
4972 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4973 "upper_%s" : "lower_%s", name);
4974 sysfs_remove_link(&(dev->dev.kobj), linkname);
4977 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
4978 struct net_device *adj_dev,
4979 struct list_head *dev_list)
4981 return (dev_list == &dev->adj_list.upper ||
4982 dev_list == &dev->adj_list.lower) &&
4983 net_eq(dev_net(dev), dev_net(adj_dev));
4986 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4987 struct net_device *adj_dev,
4988 struct list_head *dev_list,
4989 void *private, bool master)
4991 struct netdev_adjacent *adj;
4994 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5001 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5006 adj->master = master;
5008 adj->private = private;
5011 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5012 adj_dev->name, dev->name, adj_dev->name);
5014 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5015 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5020 /* Ensure that master link is always the first item in list. */
5022 ret = sysfs_create_link(&(dev->dev.kobj),
5023 &(adj_dev->dev.kobj), "master");
5025 goto remove_symlinks;
5027 list_add_rcu(&adj->list, dev_list);
5029 list_add_tail_rcu(&adj->list, dev_list);
5035 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5036 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5044 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5045 struct net_device *adj_dev,
5046 struct list_head *dev_list)
5048 struct netdev_adjacent *adj;
5050 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5053 pr_err("tried to remove device %s from %s\n",
5054 dev->name, adj_dev->name);
5058 if (adj->ref_nr > 1) {
5059 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5066 sysfs_remove_link(&(dev->dev.kobj), "master");
5068 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5069 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5071 list_del_rcu(&adj->list);
5072 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5073 adj_dev->name, dev->name, adj_dev->name);
5075 kfree_rcu(adj, rcu);
5078 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5079 struct net_device *upper_dev,
5080 struct list_head *up_list,
5081 struct list_head *down_list,
5082 void *private, bool master)
5086 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5091 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5094 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5101 static int __netdev_adjacent_dev_link(struct net_device *dev,
5102 struct net_device *upper_dev)
5104 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5105 &dev->all_adj_list.upper,
5106 &upper_dev->all_adj_list.lower,
5110 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5111 struct net_device *upper_dev,
5112 struct list_head *up_list,
5113 struct list_head *down_list)
5115 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5116 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5119 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5120 struct net_device *upper_dev)
5122 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5123 &dev->all_adj_list.upper,
5124 &upper_dev->all_adj_list.lower);
5127 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5128 struct net_device *upper_dev,
5129 void *private, bool master)
5131 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5136 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5137 &dev->adj_list.upper,
5138 &upper_dev->adj_list.lower,
5141 __netdev_adjacent_dev_unlink(dev, upper_dev);
5148 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5149 struct net_device *upper_dev)
5151 __netdev_adjacent_dev_unlink(dev, upper_dev);
5152 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5153 &dev->adj_list.upper,
5154 &upper_dev->adj_list.lower);
5157 static int __netdev_upper_dev_link(struct net_device *dev,
5158 struct net_device *upper_dev, bool master,
5161 struct netdev_adjacent *i, *j, *to_i, *to_j;
5166 if (dev == upper_dev)
5169 /* To prevent loops, check if dev is not upper device to upper_dev. */
5170 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5173 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
5176 if (master && netdev_master_upper_dev_get(dev))
5179 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5184 /* Now that we linked these devs, make all the upper_dev's
5185 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5186 * versa, and don't forget the devices itself. All of these
5187 * links are non-neighbours.
5189 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5190 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5191 pr_debug("Interlinking %s with %s, non-neighbour\n",
5192 i->dev->name, j->dev->name);
5193 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5199 /* add dev to every upper_dev's upper device */
5200 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5201 pr_debug("linking %s's upper device %s with %s\n",
5202 upper_dev->name, i->dev->name, dev->name);
5203 ret = __netdev_adjacent_dev_link(dev, i->dev);
5205 goto rollback_upper_mesh;
5208 /* add upper_dev to every dev's lower device */
5209 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5210 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5211 i->dev->name, upper_dev->name);
5212 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5214 goto rollback_lower_mesh;
5217 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5220 rollback_lower_mesh:
5222 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5225 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5230 rollback_upper_mesh:
5232 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5235 __netdev_adjacent_dev_unlink(dev, i->dev);
5243 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5244 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5245 if (i == to_i && j == to_j)
5247 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5253 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5259 * netdev_upper_dev_link - Add a link to the upper device
5261 * @upper_dev: new upper device
5263 * Adds a link to device which is upper to this one. The caller must hold
5264 * the RTNL lock. On a failure a negative errno code is returned.
5265 * On success the reference counts are adjusted and the function
5268 int netdev_upper_dev_link(struct net_device *dev,
5269 struct net_device *upper_dev)
5271 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5273 EXPORT_SYMBOL(netdev_upper_dev_link);
5276 * netdev_master_upper_dev_link - Add a master link to the upper device
5278 * @upper_dev: new upper device
5280 * Adds a link to device which is upper to this one. In this case, only
5281 * one master upper device can be linked, although other non-master devices
5282 * might be linked as well. The caller must hold the RTNL lock.
5283 * On a failure a negative errno code is returned. On success the reference
5284 * counts are adjusted and the function returns zero.
5286 int netdev_master_upper_dev_link(struct net_device *dev,
5287 struct net_device *upper_dev)
5289 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5291 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5293 int netdev_master_upper_dev_link_private(struct net_device *dev,
5294 struct net_device *upper_dev,
5297 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5299 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5302 * netdev_upper_dev_unlink - Removes a link to upper device
5304 * @upper_dev: new upper device
5306 * Removes a link to device which is upper to this one. The caller must hold
5309 void netdev_upper_dev_unlink(struct net_device *dev,
5310 struct net_device *upper_dev)
5312 struct netdev_adjacent *i, *j;
5315 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5317 /* Here is the tricky part. We must remove all dev's lower
5318 * devices from all upper_dev's upper devices and vice
5319 * versa, to maintain the graph relationship.
5321 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5322 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5323 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5325 /* remove also the devices itself from lower/upper device
5328 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5329 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5331 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5332 __netdev_adjacent_dev_unlink(dev, i->dev);
5334 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5336 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5339 * netdev_bonding_info_change - Dispatch event about slave change
5341 * @bonding_info: info to dispatch
5343 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5344 * The caller must hold the RTNL lock.
5346 void netdev_bonding_info_change(struct net_device *dev,
5347 struct netdev_bonding_info *bonding_info)
5349 struct netdev_notifier_bonding_info info;
5351 memcpy(&info.bonding_info, bonding_info,
5352 sizeof(struct netdev_bonding_info));
5353 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5356 EXPORT_SYMBOL(netdev_bonding_info_change);
5358 static void netdev_adjacent_add_links(struct net_device *dev)
5360 struct netdev_adjacent *iter;
5362 struct net *net = dev_net(dev);
5364 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5365 if (!net_eq(net,dev_net(iter->dev)))
5367 netdev_adjacent_sysfs_add(iter->dev, dev,
5368 &iter->dev->adj_list.lower);
5369 netdev_adjacent_sysfs_add(dev, iter->dev,
5370 &dev->adj_list.upper);
5373 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5374 if (!net_eq(net,dev_net(iter->dev)))
5376 netdev_adjacent_sysfs_add(iter->dev, dev,
5377 &iter->dev->adj_list.upper);
5378 netdev_adjacent_sysfs_add(dev, iter->dev,
5379 &dev->adj_list.lower);
5383 static void netdev_adjacent_del_links(struct net_device *dev)
5385 struct netdev_adjacent *iter;
5387 struct net *net = dev_net(dev);
5389 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5390 if (!net_eq(net,dev_net(iter->dev)))
5392 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5393 &iter->dev->adj_list.lower);
5394 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5395 &dev->adj_list.upper);
5398 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5399 if (!net_eq(net,dev_net(iter->dev)))
5401 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5402 &iter->dev->adj_list.upper);
5403 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5404 &dev->adj_list.lower);
5408 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5410 struct netdev_adjacent *iter;
5412 struct net *net = dev_net(dev);
5414 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5415 if (!net_eq(net,dev_net(iter->dev)))
5417 netdev_adjacent_sysfs_del(iter->dev, oldname,
5418 &iter->dev->adj_list.lower);
5419 netdev_adjacent_sysfs_add(iter->dev, dev,
5420 &iter->dev->adj_list.lower);
5423 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5424 if (!net_eq(net,dev_net(iter->dev)))
5426 netdev_adjacent_sysfs_del(iter->dev, oldname,
5427 &iter->dev->adj_list.upper);
5428 netdev_adjacent_sysfs_add(iter->dev, dev,
5429 &iter->dev->adj_list.upper);
5433 void *netdev_lower_dev_get_private(struct net_device *dev,
5434 struct net_device *lower_dev)
5436 struct netdev_adjacent *lower;
5440 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5444 return lower->private;
5446 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5449 int dev_get_nest_level(struct net_device *dev,
5450 bool (*type_check)(struct net_device *dev))
5452 struct net_device *lower = NULL;
5453 struct list_head *iter;
5459 netdev_for_each_lower_dev(dev, lower, iter) {
5460 nest = dev_get_nest_level(lower, type_check);
5461 if (max_nest < nest)
5465 if (type_check(dev))
5470 EXPORT_SYMBOL(dev_get_nest_level);
5472 static void dev_change_rx_flags(struct net_device *dev, int flags)
5474 const struct net_device_ops *ops = dev->netdev_ops;
5476 if (ops->ndo_change_rx_flags)
5477 ops->ndo_change_rx_flags(dev, flags);
5480 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5482 unsigned int old_flags = dev->flags;
5488 dev->flags |= IFF_PROMISC;
5489 dev->promiscuity += inc;
5490 if (dev->promiscuity == 0) {
5493 * If inc causes overflow, untouch promisc and return error.
5496 dev->flags &= ~IFF_PROMISC;
5498 dev->promiscuity -= inc;
5499 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5504 if (dev->flags != old_flags) {
5505 pr_info("device %s %s promiscuous mode\n",
5507 dev->flags & IFF_PROMISC ? "entered" : "left");
5508 if (audit_enabled) {
5509 current_uid_gid(&uid, &gid);
5510 audit_log(current->audit_context, GFP_ATOMIC,
5511 AUDIT_ANOM_PROMISCUOUS,
5512 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5513 dev->name, (dev->flags & IFF_PROMISC),
5514 (old_flags & IFF_PROMISC),
5515 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5516 from_kuid(&init_user_ns, uid),
5517 from_kgid(&init_user_ns, gid),
5518 audit_get_sessionid(current));
5521 dev_change_rx_flags(dev, IFF_PROMISC);
5524 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5529 * dev_set_promiscuity - update promiscuity count on a device
5533 * Add or remove promiscuity from a device. While the count in the device
5534 * remains above zero the interface remains promiscuous. Once it hits zero
5535 * the device reverts back to normal filtering operation. A negative inc
5536 * value is used to drop promiscuity on the device.
5537 * Return 0 if successful or a negative errno code on error.
5539 int dev_set_promiscuity(struct net_device *dev, int inc)
5541 unsigned int old_flags = dev->flags;
5544 err = __dev_set_promiscuity(dev, inc, true);
5547 if (dev->flags != old_flags)
5548 dev_set_rx_mode(dev);
5551 EXPORT_SYMBOL(dev_set_promiscuity);
5553 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5555 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5559 dev->flags |= IFF_ALLMULTI;
5560 dev->allmulti += inc;
5561 if (dev->allmulti == 0) {
5564 * If inc causes overflow, untouch allmulti and return error.
5567 dev->flags &= ~IFF_ALLMULTI;
5569 dev->allmulti -= inc;
5570 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5575 if (dev->flags ^ old_flags) {
5576 dev_change_rx_flags(dev, IFF_ALLMULTI);
5577 dev_set_rx_mode(dev);
5579 __dev_notify_flags(dev, old_flags,
5580 dev->gflags ^ old_gflags);
5586 * dev_set_allmulti - update allmulti count on a device
5590 * Add or remove reception of all multicast frames to a device. While the
5591 * count in the device remains above zero the interface remains listening
5592 * to all interfaces. Once it hits zero the device reverts back to normal
5593 * filtering operation. A negative @inc value is used to drop the counter
5594 * when releasing a resource needing all multicasts.
5595 * Return 0 if successful or a negative errno code on error.
5598 int dev_set_allmulti(struct net_device *dev, int inc)
5600 return __dev_set_allmulti(dev, inc, true);
5602 EXPORT_SYMBOL(dev_set_allmulti);
5605 * Upload unicast and multicast address lists to device and
5606 * configure RX filtering. When the device doesn't support unicast
5607 * filtering it is put in promiscuous mode while unicast addresses
5610 void __dev_set_rx_mode(struct net_device *dev)
5612 const struct net_device_ops *ops = dev->netdev_ops;
5614 /* dev_open will call this function so the list will stay sane. */
5615 if (!(dev->flags&IFF_UP))
5618 if (!netif_device_present(dev))
5621 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5622 /* Unicast addresses changes may only happen under the rtnl,
5623 * therefore calling __dev_set_promiscuity here is safe.
5625 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5626 __dev_set_promiscuity(dev, 1, false);
5627 dev->uc_promisc = true;
5628 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5629 __dev_set_promiscuity(dev, -1, false);
5630 dev->uc_promisc = false;
5634 if (ops->ndo_set_rx_mode)
5635 ops->ndo_set_rx_mode(dev);
5638 void dev_set_rx_mode(struct net_device *dev)
5640 netif_addr_lock_bh(dev);
5641 __dev_set_rx_mode(dev);
5642 netif_addr_unlock_bh(dev);
5646 * dev_get_flags - get flags reported to userspace
5649 * Get the combination of flag bits exported through APIs to userspace.
5651 unsigned int dev_get_flags(const struct net_device *dev)
5655 flags = (dev->flags & ~(IFF_PROMISC |
5660 (dev->gflags & (IFF_PROMISC |
5663 if (netif_running(dev)) {
5664 if (netif_oper_up(dev))
5665 flags |= IFF_RUNNING;
5666 if (netif_carrier_ok(dev))
5667 flags |= IFF_LOWER_UP;
5668 if (netif_dormant(dev))
5669 flags |= IFF_DORMANT;
5674 EXPORT_SYMBOL(dev_get_flags);
5676 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5678 unsigned int old_flags = dev->flags;
5684 * Set the flags on our device.
5687 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5688 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5690 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5694 * Load in the correct multicast list now the flags have changed.
5697 if ((old_flags ^ flags) & IFF_MULTICAST)
5698 dev_change_rx_flags(dev, IFF_MULTICAST);
5700 dev_set_rx_mode(dev);
5703 * Have we downed the interface. We handle IFF_UP ourselves
5704 * according to user attempts to set it, rather than blindly
5709 if ((old_flags ^ flags) & IFF_UP)
5710 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5712 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5713 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5714 unsigned int old_flags = dev->flags;
5716 dev->gflags ^= IFF_PROMISC;
5718 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5719 if (dev->flags != old_flags)
5720 dev_set_rx_mode(dev);
5723 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5724 is important. Some (broken) drivers set IFF_PROMISC, when
5725 IFF_ALLMULTI is requested not asking us and not reporting.
5727 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5728 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5730 dev->gflags ^= IFF_ALLMULTI;
5731 __dev_set_allmulti(dev, inc, false);
5737 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5738 unsigned int gchanges)
5740 unsigned int changes = dev->flags ^ old_flags;
5743 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5745 if (changes & IFF_UP) {
5746 if (dev->flags & IFF_UP)
5747 call_netdevice_notifiers(NETDEV_UP, dev);
5749 call_netdevice_notifiers(NETDEV_DOWN, dev);
5752 if (dev->flags & IFF_UP &&
5753 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5754 struct netdev_notifier_change_info change_info;
5756 change_info.flags_changed = changes;
5757 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5763 * dev_change_flags - change device settings
5765 * @flags: device state flags
5767 * Change settings on device based state flags. The flags are
5768 * in the userspace exported format.
5770 int dev_change_flags(struct net_device *dev, unsigned int flags)
5773 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5775 ret = __dev_change_flags(dev, flags);
5779 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5780 __dev_notify_flags(dev, old_flags, changes);
5783 EXPORT_SYMBOL(dev_change_flags);
5785 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5787 const struct net_device_ops *ops = dev->netdev_ops;
5789 if (ops->ndo_change_mtu)
5790 return ops->ndo_change_mtu(dev, new_mtu);
5797 * dev_set_mtu - Change maximum transfer unit
5799 * @new_mtu: new transfer unit
5801 * Change the maximum transfer size of the network device.
5803 int dev_set_mtu(struct net_device *dev, int new_mtu)
5807 if (new_mtu == dev->mtu)
5810 /* MTU must be positive. */
5814 if (!netif_device_present(dev))
5817 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5818 err = notifier_to_errno(err);
5822 orig_mtu = dev->mtu;
5823 err = __dev_set_mtu(dev, new_mtu);
5826 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5827 err = notifier_to_errno(err);
5829 /* setting mtu back and notifying everyone again,
5830 * so that they have a chance to revert changes.
5832 __dev_set_mtu(dev, orig_mtu);
5833 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5838 EXPORT_SYMBOL(dev_set_mtu);
5841 * dev_set_group - Change group this device belongs to
5843 * @new_group: group this device should belong to
5845 void dev_set_group(struct net_device *dev, int new_group)
5847 dev->group = new_group;
5849 EXPORT_SYMBOL(dev_set_group);
5852 * dev_set_mac_address - Change Media Access Control Address
5856 * Change the hardware (MAC) address of the device
5858 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5860 const struct net_device_ops *ops = dev->netdev_ops;
5863 if (!ops->ndo_set_mac_address)
5865 if (sa->sa_family != dev->type)
5867 if (!netif_device_present(dev))
5869 err = ops->ndo_set_mac_address(dev, sa);
5872 dev->addr_assign_type = NET_ADDR_SET;
5873 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5874 add_device_randomness(dev->dev_addr, dev->addr_len);
5877 EXPORT_SYMBOL(dev_set_mac_address);
5880 * dev_change_carrier - Change device carrier
5882 * @new_carrier: new value
5884 * Change device carrier
5886 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5888 const struct net_device_ops *ops = dev->netdev_ops;
5890 if (!ops->ndo_change_carrier)
5892 if (!netif_device_present(dev))
5894 return ops->ndo_change_carrier(dev, new_carrier);
5896 EXPORT_SYMBOL(dev_change_carrier);
5899 * dev_get_phys_port_id - Get device physical port ID
5903 * Get device physical port ID
5905 int dev_get_phys_port_id(struct net_device *dev,
5906 struct netdev_phys_item_id *ppid)
5908 const struct net_device_ops *ops = dev->netdev_ops;
5910 if (!ops->ndo_get_phys_port_id)
5912 return ops->ndo_get_phys_port_id(dev, ppid);
5914 EXPORT_SYMBOL(dev_get_phys_port_id);
5917 * dev_new_index - allocate an ifindex
5918 * @net: the applicable net namespace
5920 * Returns a suitable unique value for a new device interface
5921 * number. The caller must hold the rtnl semaphore or the
5922 * dev_base_lock to be sure it remains unique.
5924 static int dev_new_index(struct net *net)
5926 int ifindex = net->ifindex;
5930 if (!__dev_get_by_index(net, ifindex))
5931 return net->ifindex = ifindex;
5935 /* Delayed registration/unregisteration */
5936 static LIST_HEAD(net_todo_list);
5937 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5939 static void net_set_todo(struct net_device *dev)
5941 list_add_tail(&dev->todo_list, &net_todo_list);
5942 dev_net(dev)->dev_unreg_count++;
5945 static void rollback_registered_many(struct list_head *head)
5947 struct net_device *dev, *tmp;
5948 LIST_HEAD(close_head);
5950 BUG_ON(dev_boot_phase);
5953 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5954 /* Some devices call without registering
5955 * for initialization unwind. Remove those
5956 * devices and proceed with the remaining.
5958 if (dev->reg_state == NETREG_UNINITIALIZED) {
5959 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5963 list_del(&dev->unreg_list);
5966 dev->dismantle = true;
5967 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5970 /* If device is running, close it first. */
5971 list_for_each_entry(dev, head, unreg_list)
5972 list_add_tail(&dev->close_list, &close_head);
5973 dev_close_many(&close_head);
5975 list_for_each_entry(dev, head, unreg_list) {
5976 /* And unlink it from device chain. */
5977 unlist_netdevice(dev);
5979 dev->reg_state = NETREG_UNREGISTERING;
5984 list_for_each_entry(dev, head, unreg_list) {
5985 struct sk_buff *skb = NULL;
5987 /* Shutdown queueing discipline. */
5991 /* Notify protocols, that we are about to destroy
5992 this device. They should clean all the things.
5994 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5996 if (!dev->rtnl_link_ops ||
5997 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5998 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6002 * Flush the unicast and multicast chains
6007 if (dev->netdev_ops->ndo_uninit)
6008 dev->netdev_ops->ndo_uninit(dev);
6011 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6013 /* Notifier chain MUST detach us all upper devices. */
6014 WARN_ON(netdev_has_any_upper_dev(dev));
6016 /* Remove entries from kobject tree */
6017 netdev_unregister_kobject(dev);
6019 /* Remove XPS queueing entries */
6020 netif_reset_xps_queues_gt(dev, 0);
6026 list_for_each_entry(dev, head, unreg_list)
6030 static void rollback_registered(struct net_device *dev)
6034 list_add(&dev->unreg_list, &single);
6035 rollback_registered_many(&single);
6039 static netdev_features_t netdev_fix_features(struct net_device *dev,
6040 netdev_features_t features)
6042 /* Fix illegal checksum combinations */
6043 if ((features & NETIF_F_HW_CSUM) &&
6044 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6045 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6046 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6049 /* TSO requires that SG is present as well. */
6050 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6051 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6052 features &= ~NETIF_F_ALL_TSO;
6055 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6056 !(features & NETIF_F_IP_CSUM)) {
6057 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6058 features &= ~NETIF_F_TSO;
6059 features &= ~NETIF_F_TSO_ECN;
6062 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6063 !(features & NETIF_F_IPV6_CSUM)) {
6064 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6065 features &= ~NETIF_F_TSO6;
6068 /* TSO ECN requires that TSO is present as well. */
6069 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6070 features &= ~NETIF_F_TSO_ECN;
6072 /* Software GSO depends on SG. */
6073 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6074 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6075 features &= ~NETIF_F_GSO;
6078 /* UFO needs SG and checksumming */
6079 if (features & NETIF_F_UFO) {
6080 /* maybe split UFO into V4 and V6? */
6081 if (!((features & NETIF_F_GEN_CSUM) ||
6082 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6083 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6085 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6086 features &= ~NETIF_F_UFO;
6089 if (!(features & NETIF_F_SG)) {
6091 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6092 features &= ~NETIF_F_UFO;
6096 #ifdef CONFIG_NET_RX_BUSY_POLL
6097 if (dev->netdev_ops->ndo_busy_poll)
6098 features |= NETIF_F_BUSY_POLL;
6101 features &= ~NETIF_F_BUSY_POLL;
6106 int __netdev_update_features(struct net_device *dev)
6108 netdev_features_t features;
6113 features = netdev_get_wanted_features(dev);
6115 if (dev->netdev_ops->ndo_fix_features)
6116 features = dev->netdev_ops->ndo_fix_features(dev, features);
6118 /* driver might be less strict about feature dependencies */
6119 features = netdev_fix_features(dev, features);
6121 if (dev->features == features)
6124 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6125 &dev->features, &features);
6127 if (dev->netdev_ops->ndo_set_features)
6128 err = dev->netdev_ops->ndo_set_features(dev, features);
6130 if (unlikely(err < 0)) {
6132 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6133 err, &features, &dev->features);
6138 dev->features = features;
6144 * netdev_update_features - recalculate device features
6145 * @dev: the device to check
6147 * Recalculate dev->features set and send notifications if it
6148 * has changed. Should be called after driver or hardware dependent
6149 * conditions might have changed that influence the features.
6151 void netdev_update_features(struct net_device *dev)
6153 if (__netdev_update_features(dev))
6154 netdev_features_change(dev);
6156 EXPORT_SYMBOL(netdev_update_features);
6159 * netdev_change_features - recalculate device features
6160 * @dev: the device to check
6162 * Recalculate dev->features set and send notifications even
6163 * if they have not changed. Should be called instead of
6164 * netdev_update_features() if also dev->vlan_features might
6165 * have changed to allow the changes to be propagated to stacked
6168 void netdev_change_features(struct net_device *dev)
6170 __netdev_update_features(dev);
6171 netdev_features_change(dev);
6173 EXPORT_SYMBOL(netdev_change_features);
6176 * netif_stacked_transfer_operstate - transfer operstate
6177 * @rootdev: the root or lower level device to transfer state from
6178 * @dev: the device to transfer operstate to
6180 * Transfer operational state from root to device. This is normally
6181 * called when a stacking relationship exists between the root
6182 * device and the device(a leaf device).
6184 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6185 struct net_device *dev)
6187 if (rootdev->operstate == IF_OPER_DORMANT)
6188 netif_dormant_on(dev);
6190 netif_dormant_off(dev);
6192 if (netif_carrier_ok(rootdev)) {
6193 if (!netif_carrier_ok(dev))
6194 netif_carrier_on(dev);
6196 if (netif_carrier_ok(dev))
6197 netif_carrier_off(dev);
6200 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6203 static int netif_alloc_rx_queues(struct net_device *dev)
6205 unsigned int i, count = dev->num_rx_queues;
6206 struct netdev_rx_queue *rx;
6207 size_t sz = count * sizeof(*rx);
6211 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6219 for (i = 0; i < count; i++)
6225 static void netdev_init_one_queue(struct net_device *dev,
6226 struct netdev_queue *queue, void *_unused)
6228 /* Initialize queue lock */
6229 spin_lock_init(&queue->_xmit_lock);
6230 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6231 queue->xmit_lock_owner = -1;
6232 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6235 dql_init(&queue->dql, HZ);
6239 static void netif_free_tx_queues(struct net_device *dev)
6244 static int netif_alloc_netdev_queues(struct net_device *dev)
6246 unsigned int count = dev->num_tx_queues;
6247 struct netdev_queue *tx;
6248 size_t sz = count * sizeof(*tx);
6250 BUG_ON(count < 1 || count > 0xffff);
6252 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6260 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6261 spin_lock_init(&dev->tx_global_lock);
6267 * register_netdevice - register a network device
6268 * @dev: device to register
6270 * Take a completed network device structure and add it to the kernel
6271 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6272 * chain. 0 is returned on success. A negative errno code is returned
6273 * on a failure to set up the device, or if the name is a duplicate.
6275 * Callers must hold the rtnl semaphore. You may want
6276 * register_netdev() instead of this.
6279 * The locking appears insufficient to guarantee two parallel registers
6280 * will not get the same name.
6283 int register_netdevice(struct net_device *dev)
6286 struct net *net = dev_net(dev);
6288 BUG_ON(dev_boot_phase);
6293 /* When net_device's are persistent, this will be fatal. */
6294 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6297 spin_lock_init(&dev->addr_list_lock);
6298 netdev_set_addr_lockdep_class(dev);
6302 ret = dev_get_valid_name(net, dev, dev->name);
6306 /* Init, if this function is available */
6307 if (dev->netdev_ops->ndo_init) {
6308 ret = dev->netdev_ops->ndo_init(dev);
6316 if (((dev->hw_features | dev->features) &
6317 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6318 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6319 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6320 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6327 dev->ifindex = dev_new_index(net);
6328 else if (__dev_get_by_index(net, dev->ifindex))
6331 if (dev->iflink == -1)
6332 dev->iflink = dev->ifindex;
6334 /* Transfer changeable features to wanted_features and enable
6335 * software offloads (GSO and GRO).
6337 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6338 dev->features |= NETIF_F_SOFT_FEATURES;
6339 dev->wanted_features = dev->features & dev->hw_features;
6341 if (!(dev->flags & IFF_LOOPBACK)) {
6342 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6345 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6347 dev->vlan_features |= NETIF_F_HIGHDMA;
6349 /* Make NETIF_F_SG inheritable to tunnel devices.
6351 dev->hw_enc_features |= NETIF_F_SG;
6353 /* Make NETIF_F_SG inheritable to MPLS.
6355 dev->mpls_features |= NETIF_F_SG;
6357 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6358 ret = notifier_to_errno(ret);
6362 ret = netdev_register_kobject(dev);
6365 dev->reg_state = NETREG_REGISTERED;
6367 __netdev_update_features(dev);
6370 * Default initial state at registry is that the
6371 * device is present.
6374 set_bit(__LINK_STATE_PRESENT, &dev->state);
6376 linkwatch_init_dev(dev);
6378 dev_init_scheduler(dev);
6380 list_netdevice(dev);
6381 add_device_randomness(dev->dev_addr, dev->addr_len);
6383 /* If the device has permanent device address, driver should
6384 * set dev_addr and also addr_assign_type should be set to
6385 * NET_ADDR_PERM (default value).
6387 if (dev->addr_assign_type == NET_ADDR_PERM)
6388 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6390 /* Notify protocols, that a new device appeared. */
6391 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6392 ret = notifier_to_errno(ret);
6394 rollback_registered(dev);
6395 dev->reg_state = NETREG_UNREGISTERED;
6398 * Prevent userspace races by waiting until the network
6399 * device is fully setup before sending notifications.
6401 if (!dev->rtnl_link_ops ||
6402 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6403 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6409 if (dev->netdev_ops->ndo_uninit)
6410 dev->netdev_ops->ndo_uninit(dev);
6413 EXPORT_SYMBOL(register_netdevice);
6416 * init_dummy_netdev - init a dummy network device for NAPI
6417 * @dev: device to init
6419 * This takes a network device structure and initialize the minimum
6420 * amount of fields so it can be used to schedule NAPI polls without
6421 * registering a full blown interface. This is to be used by drivers
6422 * that need to tie several hardware interfaces to a single NAPI
6423 * poll scheduler due to HW limitations.
6425 int init_dummy_netdev(struct net_device *dev)
6427 /* Clear everything. Note we don't initialize spinlocks
6428 * are they aren't supposed to be taken by any of the
6429 * NAPI code and this dummy netdev is supposed to be
6430 * only ever used for NAPI polls
6432 memset(dev, 0, sizeof(struct net_device));
6434 /* make sure we BUG if trying to hit standard
6435 * register/unregister code path
6437 dev->reg_state = NETREG_DUMMY;
6439 /* NAPI wants this */
6440 INIT_LIST_HEAD(&dev->napi_list);
6442 /* a dummy interface is started by default */
6443 set_bit(__LINK_STATE_PRESENT, &dev->state);
6444 set_bit(__LINK_STATE_START, &dev->state);
6446 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6447 * because users of this 'device' dont need to change
6453 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6457 * register_netdev - register a network device
6458 * @dev: device to register
6460 * Take a completed network device structure and add it to the kernel
6461 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6462 * chain. 0 is returned on success. A negative errno code is returned
6463 * on a failure to set up the device, or if the name is a duplicate.
6465 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6466 * and expands the device name if you passed a format string to
6469 int register_netdev(struct net_device *dev)
6474 err = register_netdevice(dev);
6478 EXPORT_SYMBOL(register_netdev);
6480 int netdev_refcnt_read(const struct net_device *dev)
6484 for_each_possible_cpu(i)
6485 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6488 EXPORT_SYMBOL(netdev_refcnt_read);
6491 * netdev_wait_allrefs - wait until all references are gone.
6492 * @dev: target net_device
6494 * This is called when unregistering network devices.
6496 * Any protocol or device that holds a reference should register
6497 * for netdevice notification, and cleanup and put back the
6498 * reference if they receive an UNREGISTER event.
6499 * We can get stuck here if buggy protocols don't correctly
6502 static void netdev_wait_allrefs(struct net_device *dev)
6504 unsigned long rebroadcast_time, warning_time;
6507 linkwatch_forget_dev(dev);
6509 rebroadcast_time = warning_time = jiffies;
6510 refcnt = netdev_refcnt_read(dev);
6512 while (refcnt != 0) {
6513 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6516 /* Rebroadcast unregister notification */
6517 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6523 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6524 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6526 /* We must not have linkwatch events
6527 * pending on unregister. If this
6528 * happens, we simply run the queue
6529 * unscheduled, resulting in a noop
6532 linkwatch_run_queue();
6537 rebroadcast_time = jiffies;
6542 refcnt = netdev_refcnt_read(dev);
6544 if (time_after(jiffies, warning_time + 10 * HZ)) {
6545 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6547 warning_time = jiffies;
6556 * register_netdevice(x1);
6557 * register_netdevice(x2);
6559 * unregister_netdevice(y1);
6560 * unregister_netdevice(y2);
6566 * We are invoked by rtnl_unlock().
6567 * This allows us to deal with problems:
6568 * 1) We can delete sysfs objects which invoke hotplug
6569 * without deadlocking with linkwatch via keventd.
6570 * 2) Since we run with the RTNL semaphore not held, we can sleep
6571 * safely in order to wait for the netdev refcnt to drop to zero.
6573 * We must not return until all unregister events added during
6574 * the interval the lock was held have been completed.
6576 void netdev_run_todo(void)
6578 struct list_head list;
6580 /* Snapshot list, allow later requests */
6581 list_replace_init(&net_todo_list, &list);
6586 /* Wait for rcu callbacks to finish before next phase */
6587 if (!list_empty(&list))
6590 while (!list_empty(&list)) {
6591 struct net_device *dev
6592 = list_first_entry(&list, struct net_device, todo_list);
6593 list_del(&dev->todo_list);
6596 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6599 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6600 pr_err("network todo '%s' but state %d\n",
6601 dev->name, dev->reg_state);
6606 dev->reg_state = NETREG_UNREGISTERED;
6608 on_each_cpu(flush_backlog, dev, 1);
6610 netdev_wait_allrefs(dev);
6613 BUG_ON(netdev_refcnt_read(dev));
6614 BUG_ON(!list_empty(&dev->ptype_all));
6615 BUG_ON(!list_empty(&dev->ptype_specific));
6616 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6617 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6618 WARN_ON(dev->dn_ptr);
6620 if (dev->destructor)
6621 dev->destructor(dev);
6623 /* Report a network device has been unregistered */
6625 dev_net(dev)->dev_unreg_count--;
6627 wake_up(&netdev_unregistering_wq);
6629 /* Free network device */
6630 kobject_put(&dev->dev.kobj);
6634 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6635 * fields in the same order, with only the type differing.
6637 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6638 const struct net_device_stats *netdev_stats)
6640 #if BITS_PER_LONG == 64
6641 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6642 memcpy(stats64, netdev_stats, sizeof(*stats64));
6644 size_t i, n = sizeof(*stats64) / sizeof(u64);
6645 const unsigned long *src = (const unsigned long *)netdev_stats;
6646 u64 *dst = (u64 *)stats64;
6648 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6649 sizeof(*stats64) / sizeof(u64));
6650 for (i = 0; i < n; i++)
6654 EXPORT_SYMBOL(netdev_stats_to_stats64);
6657 * dev_get_stats - get network device statistics
6658 * @dev: device to get statistics from
6659 * @storage: place to store stats
6661 * Get network statistics from device. Return @storage.
6662 * The device driver may provide its own method by setting
6663 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6664 * otherwise the internal statistics structure is used.
6666 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6667 struct rtnl_link_stats64 *storage)
6669 const struct net_device_ops *ops = dev->netdev_ops;
6671 if (ops->ndo_get_stats64) {
6672 memset(storage, 0, sizeof(*storage));
6673 ops->ndo_get_stats64(dev, storage);
6674 } else if (ops->ndo_get_stats) {
6675 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6677 netdev_stats_to_stats64(storage, &dev->stats);
6679 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6680 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6683 EXPORT_SYMBOL(dev_get_stats);
6685 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6687 struct netdev_queue *queue = dev_ingress_queue(dev);
6689 #ifdef CONFIG_NET_CLS_ACT
6692 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6695 netdev_init_one_queue(dev, queue, NULL);
6696 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
6697 queue->qdisc_sleeping = &noop_qdisc;
6698 rcu_assign_pointer(dev->ingress_queue, queue);
6703 static const struct ethtool_ops default_ethtool_ops;
6705 void netdev_set_default_ethtool_ops(struct net_device *dev,
6706 const struct ethtool_ops *ops)
6708 if (dev->ethtool_ops == &default_ethtool_ops)
6709 dev->ethtool_ops = ops;
6711 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6713 void netdev_freemem(struct net_device *dev)
6715 char *addr = (char *)dev - dev->padded;
6721 * alloc_netdev_mqs - allocate network device
6722 * @sizeof_priv: size of private data to allocate space for
6723 * @name: device name format string
6724 * @name_assign_type: origin of device name
6725 * @setup: callback to initialize device
6726 * @txqs: the number of TX subqueues to allocate
6727 * @rxqs: the number of RX subqueues to allocate
6729 * Allocates a struct net_device with private data area for driver use
6730 * and performs basic initialization. Also allocates subqueue structs
6731 * for each queue on the device.
6733 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6734 unsigned char name_assign_type,
6735 void (*setup)(struct net_device *),
6736 unsigned int txqs, unsigned int rxqs)
6738 struct net_device *dev;
6740 struct net_device *p;
6742 BUG_ON(strlen(name) >= sizeof(dev->name));
6745 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6751 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6756 alloc_size = sizeof(struct net_device);
6758 /* ensure 32-byte alignment of private area */
6759 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6760 alloc_size += sizeof_priv;
6762 /* ensure 32-byte alignment of whole construct */
6763 alloc_size += NETDEV_ALIGN - 1;
6765 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6767 p = vzalloc(alloc_size);
6771 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6772 dev->padded = (char *)dev - (char *)p;
6774 dev->pcpu_refcnt = alloc_percpu(int);
6775 if (!dev->pcpu_refcnt)
6778 if (dev_addr_init(dev))
6784 dev_net_set(dev, &init_net);
6786 dev->gso_max_size = GSO_MAX_SIZE;
6787 dev->gso_max_segs = GSO_MAX_SEGS;
6788 dev->gso_min_segs = 0;
6790 INIT_LIST_HEAD(&dev->napi_list);
6791 INIT_LIST_HEAD(&dev->unreg_list);
6792 INIT_LIST_HEAD(&dev->close_list);
6793 INIT_LIST_HEAD(&dev->link_watch_list);
6794 INIT_LIST_HEAD(&dev->adj_list.upper);
6795 INIT_LIST_HEAD(&dev->adj_list.lower);
6796 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6797 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6798 INIT_LIST_HEAD(&dev->ptype_all);
6799 INIT_LIST_HEAD(&dev->ptype_specific);
6800 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
6803 dev->num_tx_queues = txqs;
6804 dev->real_num_tx_queues = txqs;
6805 if (netif_alloc_netdev_queues(dev))
6809 dev->num_rx_queues = rxqs;
6810 dev->real_num_rx_queues = rxqs;
6811 if (netif_alloc_rx_queues(dev))
6815 strcpy(dev->name, name);
6816 dev->name_assign_type = name_assign_type;
6817 dev->group = INIT_NETDEV_GROUP;
6818 if (!dev->ethtool_ops)
6819 dev->ethtool_ops = &default_ethtool_ops;
6827 free_percpu(dev->pcpu_refcnt);
6829 netdev_freemem(dev);
6832 EXPORT_SYMBOL(alloc_netdev_mqs);
6835 * free_netdev - free network device
6838 * This function does the last stage of destroying an allocated device
6839 * interface. The reference to the device object is released.
6840 * If this is the last reference then it will be freed.
6842 void free_netdev(struct net_device *dev)
6844 struct napi_struct *p, *n;
6846 release_net(dev_net(dev));
6848 netif_free_tx_queues(dev);
6853 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6855 /* Flush device addresses */
6856 dev_addr_flush(dev);
6858 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6861 free_percpu(dev->pcpu_refcnt);
6862 dev->pcpu_refcnt = NULL;
6864 /* Compatibility with error handling in drivers */
6865 if (dev->reg_state == NETREG_UNINITIALIZED) {
6866 netdev_freemem(dev);
6870 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6871 dev->reg_state = NETREG_RELEASED;
6873 /* will free via device release */
6874 put_device(&dev->dev);
6876 EXPORT_SYMBOL(free_netdev);
6879 * synchronize_net - Synchronize with packet receive processing
6881 * Wait for packets currently being received to be done.
6882 * Does not block later packets from starting.
6884 void synchronize_net(void)
6887 if (rtnl_is_locked())
6888 synchronize_rcu_expedited();
6892 EXPORT_SYMBOL(synchronize_net);
6895 * unregister_netdevice_queue - remove device from the kernel
6899 * This function shuts down a device interface and removes it
6900 * from the kernel tables.
6901 * If head not NULL, device is queued to be unregistered later.
6903 * Callers must hold the rtnl semaphore. You may want
6904 * unregister_netdev() instead of this.
6907 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6912 list_move_tail(&dev->unreg_list, head);
6914 rollback_registered(dev);
6915 /* Finish processing unregister after unlock */
6919 EXPORT_SYMBOL(unregister_netdevice_queue);
6922 * unregister_netdevice_many - unregister many devices
6923 * @head: list of devices
6925 * Note: As most callers use a stack allocated list_head,
6926 * we force a list_del() to make sure stack wont be corrupted later.
6928 void unregister_netdevice_many(struct list_head *head)
6930 struct net_device *dev;
6932 if (!list_empty(head)) {
6933 rollback_registered_many(head);
6934 list_for_each_entry(dev, head, unreg_list)
6939 EXPORT_SYMBOL(unregister_netdevice_many);
6942 * unregister_netdev - remove device from the kernel
6945 * This function shuts down a device interface and removes it
6946 * from the kernel tables.
6948 * This is just a wrapper for unregister_netdevice that takes
6949 * the rtnl semaphore. In general you want to use this and not
6950 * unregister_netdevice.
6952 void unregister_netdev(struct net_device *dev)
6955 unregister_netdevice(dev);
6958 EXPORT_SYMBOL(unregister_netdev);
6961 * dev_change_net_namespace - move device to different nethost namespace
6963 * @net: network namespace
6964 * @pat: If not NULL name pattern to try if the current device name
6965 * is already taken in the destination network namespace.
6967 * This function shuts down a device interface and moves it
6968 * to a new network namespace. On success 0 is returned, on
6969 * a failure a netagive errno code is returned.
6971 * Callers must hold the rtnl semaphore.
6974 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6980 /* Don't allow namespace local devices to be moved. */
6982 if (dev->features & NETIF_F_NETNS_LOCAL)
6985 /* Ensure the device has been registrered */
6986 if (dev->reg_state != NETREG_REGISTERED)
6989 /* Get out if there is nothing todo */
6991 if (net_eq(dev_net(dev), net))
6994 /* Pick the destination device name, and ensure
6995 * we can use it in the destination network namespace.
6998 if (__dev_get_by_name(net, dev->name)) {
6999 /* We get here if we can't use the current device name */
7002 if (dev_get_valid_name(net, dev, pat) < 0)
7007 * And now a mini version of register_netdevice unregister_netdevice.
7010 /* If device is running close it first. */
7013 /* And unlink it from device chain */
7015 unlist_netdevice(dev);
7019 /* Shutdown queueing discipline. */
7022 /* Notify protocols, that we are about to destroy
7023 this device. They should clean all the things.
7025 Note that dev->reg_state stays at NETREG_REGISTERED.
7026 This is wanted because this way 8021q and macvlan know
7027 the device is just moving and can keep their slaves up.
7029 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7031 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7032 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7035 * Flush the unicast and multicast chains
7040 /* Send a netdev-removed uevent to the old namespace */
7041 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7042 netdev_adjacent_del_links(dev);
7044 /* Actually switch the network namespace */
7045 dev_net_set(dev, net);
7047 /* If there is an ifindex conflict assign a new one */
7048 if (__dev_get_by_index(net, dev->ifindex)) {
7049 int iflink = (dev->iflink == dev->ifindex);
7050 dev->ifindex = dev_new_index(net);
7052 dev->iflink = dev->ifindex;
7055 /* Send a netdev-add uevent to the new namespace */
7056 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7057 netdev_adjacent_add_links(dev);
7059 /* Fixup kobjects */
7060 err = device_rename(&dev->dev, dev->name);
7063 /* Add the device back in the hashes */
7064 list_netdevice(dev);
7066 /* Notify protocols, that a new device appeared. */
7067 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7070 * Prevent userspace races by waiting until the network
7071 * device is fully setup before sending notifications.
7073 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7080 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7082 static int dev_cpu_callback(struct notifier_block *nfb,
7083 unsigned long action,
7086 struct sk_buff **list_skb;
7087 struct sk_buff *skb;
7088 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7089 struct softnet_data *sd, *oldsd;
7091 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7094 local_irq_disable();
7095 cpu = smp_processor_id();
7096 sd = &per_cpu(softnet_data, cpu);
7097 oldsd = &per_cpu(softnet_data, oldcpu);
7099 /* Find end of our completion_queue. */
7100 list_skb = &sd->completion_queue;
7102 list_skb = &(*list_skb)->next;
7103 /* Append completion queue from offline CPU. */
7104 *list_skb = oldsd->completion_queue;
7105 oldsd->completion_queue = NULL;
7107 /* Append output queue from offline CPU. */
7108 if (oldsd->output_queue) {
7109 *sd->output_queue_tailp = oldsd->output_queue;
7110 sd->output_queue_tailp = oldsd->output_queue_tailp;
7111 oldsd->output_queue = NULL;
7112 oldsd->output_queue_tailp = &oldsd->output_queue;
7114 /* Append NAPI poll list from offline CPU, with one exception :
7115 * process_backlog() must be called by cpu owning percpu backlog.
7116 * We properly handle process_queue & input_pkt_queue later.
7118 while (!list_empty(&oldsd->poll_list)) {
7119 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7123 list_del_init(&napi->poll_list);
7124 if (napi->poll == process_backlog)
7127 ____napi_schedule(sd, napi);
7130 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7133 /* Process offline CPU's input_pkt_queue */
7134 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7136 input_queue_head_incr(oldsd);
7138 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7140 input_queue_head_incr(oldsd);
7148 * netdev_increment_features - increment feature set by one
7149 * @all: current feature set
7150 * @one: new feature set
7151 * @mask: mask feature set
7153 * Computes a new feature set after adding a device with feature set
7154 * @one to the master device with current feature set @all. Will not
7155 * enable anything that is off in @mask. Returns the new feature set.
7157 netdev_features_t netdev_increment_features(netdev_features_t all,
7158 netdev_features_t one, netdev_features_t mask)
7160 if (mask & NETIF_F_GEN_CSUM)
7161 mask |= NETIF_F_ALL_CSUM;
7162 mask |= NETIF_F_VLAN_CHALLENGED;
7164 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7165 all &= one | ~NETIF_F_ALL_FOR_ALL;
7167 /* If one device supports hw checksumming, set for all. */
7168 if (all & NETIF_F_GEN_CSUM)
7169 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7173 EXPORT_SYMBOL(netdev_increment_features);
7175 static struct hlist_head * __net_init netdev_create_hash(void)
7178 struct hlist_head *hash;
7180 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7182 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7183 INIT_HLIST_HEAD(&hash[i]);
7188 /* Initialize per network namespace state */
7189 static int __net_init netdev_init(struct net *net)
7191 if (net != &init_net)
7192 INIT_LIST_HEAD(&net->dev_base_head);
7194 net->dev_name_head = netdev_create_hash();
7195 if (net->dev_name_head == NULL)
7198 net->dev_index_head = netdev_create_hash();
7199 if (net->dev_index_head == NULL)
7205 kfree(net->dev_name_head);
7211 * netdev_drivername - network driver for the device
7212 * @dev: network device
7214 * Determine network driver for device.
7216 const char *netdev_drivername(const struct net_device *dev)
7218 const struct device_driver *driver;
7219 const struct device *parent;
7220 const char *empty = "";
7222 parent = dev->dev.parent;
7226 driver = parent->driver;
7227 if (driver && driver->name)
7228 return driver->name;
7232 static void __netdev_printk(const char *level, const struct net_device *dev,
7233 struct va_format *vaf)
7235 if (dev && dev->dev.parent) {
7236 dev_printk_emit(level[1] - '0',
7239 dev_driver_string(dev->dev.parent),
7240 dev_name(dev->dev.parent),
7241 netdev_name(dev), netdev_reg_state(dev),
7244 printk("%s%s%s: %pV",
7245 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7247 printk("%s(NULL net_device): %pV", level, vaf);
7251 void netdev_printk(const char *level, const struct net_device *dev,
7252 const char *format, ...)
7254 struct va_format vaf;
7257 va_start(args, format);
7262 __netdev_printk(level, dev, &vaf);
7266 EXPORT_SYMBOL(netdev_printk);
7268 #define define_netdev_printk_level(func, level) \
7269 void func(const struct net_device *dev, const char *fmt, ...) \
7271 struct va_format vaf; \
7274 va_start(args, fmt); \
7279 __netdev_printk(level, dev, &vaf); \
7283 EXPORT_SYMBOL(func);
7285 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7286 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7287 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7288 define_netdev_printk_level(netdev_err, KERN_ERR);
7289 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7290 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7291 define_netdev_printk_level(netdev_info, KERN_INFO);
7293 static void __net_exit netdev_exit(struct net *net)
7295 kfree(net->dev_name_head);
7296 kfree(net->dev_index_head);
7299 static struct pernet_operations __net_initdata netdev_net_ops = {
7300 .init = netdev_init,
7301 .exit = netdev_exit,
7304 static void __net_exit default_device_exit(struct net *net)
7306 struct net_device *dev, *aux;
7308 * Push all migratable network devices back to the
7309 * initial network namespace
7312 for_each_netdev_safe(net, dev, aux) {
7314 char fb_name[IFNAMSIZ];
7316 /* Ignore unmoveable devices (i.e. loopback) */
7317 if (dev->features & NETIF_F_NETNS_LOCAL)
7320 /* Leave virtual devices for the generic cleanup */
7321 if (dev->rtnl_link_ops)
7324 /* Push remaining network devices to init_net */
7325 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7326 err = dev_change_net_namespace(dev, &init_net, fb_name);
7328 pr_emerg("%s: failed to move %s to init_net: %d\n",
7329 __func__, dev->name, err);
7336 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7338 /* Return with the rtnl_lock held when there are no network
7339 * devices unregistering in any network namespace in net_list.
7343 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7345 add_wait_queue(&netdev_unregistering_wq, &wait);
7347 unregistering = false;
7349 list_for_each_entry(net, net_list, exit_list) {
7350 if (net->dev_unreg_count > 0) {
7351 unregistering = true;
7359 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7361 remove_wait_queue(&netdev_unregistering_wq, &wait);
7364 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7366 /* At exit all network devices most be removed from a network
7367 * namespace. Do this in the reverse order of registration.
7368 * Do this across as many network namespaces as possible to
7369 * improve batching efficiency.
7371 struct net_device *dev;
7373 LIST_HEAD(dev_kill_list);
7375 /* To prevent network device cleanup code from dereferencing
7376 * loopback devices or network devices that have been freed
7377 * wait here for all pending unregistrations to complete,
7378 * before unregistring the loopback device and allowing the
7379 * network namespace be freed.
7381 * The netdev todo list containing all network devices
7382 * unregistrations that happen in default_device_exit_batch
7383 * will run in the rtnl_unlock() at the end of
7384 * default_device_exit_batch.
7386 rtnl_lock_unregistering(net_list);
7387 list_for_each_entry(net, net_list, exit_list) {
7388 for_each_netdev_reverse(net, dev) {
7389 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7390 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7392 unregister_netdevice_queue(dev, &dev_kill_list);
7395 unregister_netdevice_many(&dev_kill_list);
7399 static struct pernet_operations __net_initdata default_device_ops = {
7400 .exit = default_device_exit,
7401 .exit_batch = default_device_exit_batch,
7405 * Initialize the DEV module. At boot time this walks the device list and
7406 * unhooks any devices that fail to initialise (normally hardware not
7407 * present) and leaves us with a valid list of present and active devices.
7412 * This is called single threaded during boot, so no need
7413 * to take the rtnl semaphore.
7415 static int __init net_dev_init(void)
7417 int i, rc = -ENOMEM;
7419 BUG_ON(!dev_boot_phase);
7421 if (dev_proc_init())
7424 if (netdev_kobject_init())
7427 INIT_LIST_HEAD(&ptype_all);
7428 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7429 INIT_LIST_HEAD(&ptype_base[i]);
7431 INIT_LIST_HEAD(&offload_base);
7433 if (register_pernet_subsys(&netdev_net_ops))
7437 * Initialise the packet receive queues.
7440 for_each_possible_cpu(i) {
7441 struct softnet_data *sd = &per_cpu(softnet_data, i);
7443 skb_queue_head_init(&sd->input_pkt_queue);
7444 skb_queue_head_init(&sd->process_queue);
7445 INIT_LIST_HEAD(&sd->poll_list);
7446 sd->output_queue_tailp = &sd->output_queue;
7448 sd->csd.func = rps_trigger_softirq;
7453 sd->backlog.poll = process_backlog;
7454 sd->backlog.weight = weight_p;
7459 /* The loopback device is special if any other network devices
7460 * is present in a network namespace the loopback device must
7461 * be present. Since we now dynamically allocate and free the
7462 * loopback device ensure this invariant is maintained by
7463 * keeping the loopback device as the first device on the
7464 * list of network devices. Ensuring the loopback devices
7465 * is the first device that appears and the last network device
7468 if (register_pernet_device(&loopback_net_ops))
7471 if (register_pernet_device(&default_device_ops))
7474 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7475 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7477 hotcpu_notifier(dev_cpu_callback, 0);
7484 subsys_initcall(net_dev_init);