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;
187 static DEFINE_MUTEX(devnet_rename_mutex);
189 static inline void dev_base_seq_inc(struct net *net)
191 while (++net->dev_base_seq == 0);
194 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
196 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
198 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
201 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
203 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
206 static inline void rps_lock(struct softnet_data *sd)
209 raw_spin_lock(&sd->input_pkt_queue.raw_lock);
213 static inline void rps_unlock(struct softnet_data *sd)
216 raw_spin_unlock(&sd->input_pkt_queue.raw_lock);
220 /* Device list insertion */
221 static void list_netdevice(struct net_device *dev)
223 struct net *net = dev_net(dev);
227 write_lock_bh(&dev_base_lock);
228 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
229 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
230 hlist_add_head_rcu(&dev->index_hlist,
231 dev_index_hash(net, dev->ifindex));
232 write_unlock_bh(&dev_base_lock);
234 dev_base_seq_inc(net);
237 /* Device list removal
238 * caller must respect a RCU grace period before freeing/reusing dev
240 static void unlist_netdevice(struct net_device *dev)
244 /* Unlink dev from the device chain */
245 write_lock_bh(&dev_base_lock);
246 list_del_rcu(&dev->dev_list);
247 hlist_del_rcu(&dev->name_hlist);
248 hlist_del_rcu(&dev->index_hlist);
249 write_unlock_bh(&dev_base_lock);
251 dev_base_seq_inc(dev_net(dev));
258 static RAW_NOTIFIER_HEAD(netdev_chain);
261 * Device drivers call our routines to queue packets here. We empty the
262 * queue in the local softnet handler.
265 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
266 EXPORT_PER_CPU_SYMBOL(softnet_data);
268 #ifdef CONFIG_LOCKDEP
270 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
271 * according to dev->type
273 static const unsigned short netdev_lock_type[] =
274 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
275 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
276 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
277 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
278 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
279 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
280 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
281 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
282 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
283 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
284 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
285 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
286 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
287 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
288 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
290 static const char *const netdev_lock_name[] =
291 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
292 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
293 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
294 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
295 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
296 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
297 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
298 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
299 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
300 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
301 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
302 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
303 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
304 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
305 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
307 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
308 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
310 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
314 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
315 if (netdev_lock_type[i] == dev_type)
317 /* the last key is used by default */
318 return ARRAY_SIZE(netdev_lock_type) - 1;
321 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
322 unsigned short dev_type)
326 i = netdev_lock_pos(dev_type);
327 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
328 netdev_lock_name[i]);
331 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
335 i = netdev_lock_pos(dev->type);
336 lockdep_set_class_and_name(&dev->addr_list_lock,
337 &netdev_addr_lock_key[i],
338 netdev_lock_name[i]);
341 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
342 unsigned short dev_type)
345 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
350 /*******************************************************************************
352 Protocol management and registration routines
354 *******************************************************************************/
357 * Add a protocol ID to the list. Now that the input handler is
358 * smarter we can dispense with all the messy stuff that used to be
361 * BEWARE!!! Protocol handlers, mangling input packets,
362 * MUST BE last in hash buckets and checking protocol handlers
363 * MUST start from promiscuous ptype_all chain in net_bh.
364 * It is true now, do not change it.
365 * Explanation follows: if protocol handler, mangling packet, will
366 * be the first on list, it is not able to sense, that packet
367 * is cloned and should be copied-on-write, so that it will
368 * change it and subsequent readers will get broken packet.
372 static inline struct list_head *ptype_head(const struct packet_type *pt)
374 if (pt->type == htons(ETH_P_ALL))
375 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
377 return pt->dev ? &pt->dev->ptype_specific :
378 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
382 * dev_add_pack - add packet handler
383 * @pt: packet type declaration
385 * Add a protocol handler to the networking stack. The passed &packet_type
386 * is linked into kernel lists and may not be freed until it has been
387 * removed from the kernel lists.
389 * This call does not sleep therefore it can not
390 * guarantee all CPU's that are in middle of receiving packets
391 * will see the new packet type (until the next received packet).
394 void dev_add_pack(struct packet_type *pt)
396 struct list_head *head = ptype_head(pt);
398 spin_lock(&ptype_lock);
399 list_add_rcu(&pt->list, head);
400 spin_unlock(&ptype_lock);
402 EXPORT_SYMBOL(dev_add_pack);
405 * __dev_remove_pack - remove packet handler
406 * @pt: packet type declaration
408 * Remove a protocol handler that was previously added to the kernel
409 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
410 * from the kernel lists and can be freed or reused once this function
413 * The packet type might still be in use by receivers
414 * and must not be freed until after all the CPU's have gone
415 * through a quiescent state.
417 void __dev_remove_pack(struct packet_type *pt)
419 struct list_head *head = ptype_head(pt);
420 struct packet_type *pt1;
422 spin_lock(&ptype_lock);
424 list_for_each_entry(pt1, head, list) {
426 list_del_rcu(&pt->list);
431 pr_warn("dev_remove_pack: %p not found\n", pt);
433 spin_unlock(&ptype_lock);
435 EXPORT_SYMBOL(__dev_remove_pack);
438 * dev_remove_pack - remove packet handler
439 * @pt: packet type declaration
441 * Remove a protocol handler that was previously added to the kernel
442 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
443 * from the kernel lists and can be freed or reused once this function
446 * This call sleeps to guarantee that no CPU is looking at the packet
449 void dev_remove_pack(struct packet_type *pt)
451 __dev_remove_pack(pt);
455 EXPORT_SYMBOL(dev_remove_pack);
459 * dev_add_offload - register offload handlers
460 * @po: protocol offload declaration
462 * Add protocol offload handlers to the networking stack. The passed
463 * &proto_offload is linked into kernel lists and may not be freed until
464 * it has been removed from the kernel lists.
466 * This call does not sleep therefore it can not
467 * guarantee all CPU's that are in middle of receiving packets
468 * will see the new offload handlers (until the next received packet).
470 void dev_add_offload(struct packet_offload *po)
472 struct list_head *head = &offload_base;
474 spin_lock(&offload_lock);
475 list_add_rcu(&po->list, head);
476 spin_unlock(&offload_lock);
478 EXPORT_SYMBOL(dev_add_offload);
481 * __dev_remove_offload - remove offload handler
482 * @po: packet offload declaration
484 * Remove a protocol offload handler that was previously added to the
485 * kernel offload handlers by dev_add_offload(). The passed &offload_type
486 * is removed from the kernel lists and can be freed or reused once this
489 * The packet type might still be in use by receivers
490 * and must not be freed until after all the CPU's have gone
491 * through a quiescent state.
493 static void __dev_remove_offload(struct packet_offload *po)
495 struct list_head *head = &offload_base;
496 struct packet_offload *po1;
498 spin_lock(&offload_lock);
500 list_for_each_entry(po1, head, list) {
502 list_del_rcu(&po->list);
507 pr_warn("dev_remove_offload: %p not found\n", po);
509 spin_unlock(&offload_lock);
513 * dev_remove_offload - remove packet offload handler
514 * @po: packet offload declaration
516 * Remove a packet offload handler that was previously added to the kernel
517 * offload handlers by dev_add_offload(). The passed &offload_type is
518 * removed from the kernel lists and can be freed or reused once this
521 * This call sleeps to guarantee that no CPU is looking at the packet
524 void dev_remove_offload(struct packet_offload *po)
526 __dev_remove_offload(po);
530 EXPORT_SYMBOL(dev_remove_offload);
532 /******************************************************************************
534 Device Boot-time Settings Routines
536 *******************************************************************************/
538 /* Boot time configuration table */
539 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
542 * netdev_boot_setup_add - add new setup entry
543 * @name: name of the device
544 * @map: configured settings for the device
546 * Adds new setup entry to the dev_boot_setup list. The function
547 * returns 0 on error and 1 on success. This is a generic routine to
550 static int netdev_boot_setup_add(char *name, struct ifmap *map)
552 struct netdev_boot_setup *s;
556 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
557 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
558 memset(s[i].name, 0, sizeof(s[i].name));
559 strlcpy(s[i].name, name, IFNAMSIZ);
560 memcpy(&s[i].map, map, sizeof(s[i].map));
565 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
569 * netdev_boot_setup_check - check boot time settings
570 * @dev: the netdevice
572 * Check boot time settings for the device.
573 * The found settings are set for the device to be used
574 * later in the device probing.
575 * Returns 0 if no settings found, 1 if they are.
577 int netdev_boot_setup_check(struct net_device *dev)
579 struct netdev_boot_setup *s = dev_boot_setup;
582 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
583 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
584 !strcmp(dev->name, s[i].name)) {
585 dev->irq = s[i].map.irq;
586 dev->base_addr = s[i].map.base_addr;
587 dev->mem_start = s[i].map.mem_start;
588 dev->mem_end = s[i].map.mem_end;
594 EXPORT_SYMBOL(netdev_boot_setup_check);
598 * netdev_boot_base - get address from boot time settings
599 * @prefix: prefix for network device
600 * @unit: id for network device
602 * Check boot time settings for the base address of device.
603 * The found settings are set for the device to be used
604 * later in the device probing.
605 * Returns 0 if no settings found.
607 unsigned long netdev_boot_base(const char *prefix, int unit)
609 const struct netdev_boot_setup *s = dev_boot_setup;
613 sprintf(name, "%s%d", prefix, unit);
616 * If device already registered then return base of 1
617 * to indicate not to probe for this interface
619 if (__dev_get_by_name(&init_net, name))
622 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
623 if (!strcmp(name, s[i].name))
624 return s[i].map.base_addr;
629 * Saves at boot time configured settings for any netdevice.
631 int __init netdev_boot_setup(char *str)
636 str = get_options(str, ARRAY_SIZE(ints), ints);
641 memset(&map, 0, sizeof(map));
645 map.base_addr = ints[2];
647 map.mem_start = ints[3];
649 map.mem_end = ints[4];
651 /* Add new entry to the list */
652 return netdev_boot_setup_add(str, &map);
655 __setup("netdev=", netdev_boot_setup);
657 /*******************************************************************************
659 Device Interface Subroutines
661 *******************************************************************************/
664 * dev_get_iflink - get 'iflink' value of a interface
665 * @dev: targeted interface
667 * Indicates the ifindex the interface is linked to.
668 * Physical interfaces have the same 'ifindex' and 'iflink' values.
671 int dev_get_iflink(const struct net_device *dev)
673 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
674 return dev->netdev_ops->ndo_get_iflink(dev);
676 /* If dev->rtnl_link_ops is set, it's a virtual interface. */
677 if (dev->rtnl_link_ops)
682 EXPORT_SYMBOL(dev_get_iflink);
685 * __dev_get_by_name - find a device by its name
686 * @net: the applicable net namespace
687 * @name: name to find
689 * Find an interface by name. Must be called under RTNL semaphore
690 * or @dev_base_lock. If the name is found a pointer to the device
691 * is returned. If the name is not found then %NULL is returned. The
692 * reference counters are not incremented so the caller must be
693 * careful with locks.
696 struct net_device *__dev_get_by_name(struct net *net, const char *name)
698 struct net_device *dev;
699 struct hlist_head *head = dev_name_hash(net, name);
701 hlist_for_each_entry(dev, head, name_hlist)
702 if (!strncmp(dev->name, name, IFNAMSIZ))
707 EXPORT_SYMBOL(__dev_get_by_name);
710 * dev_get_by_name_rcu - find a device by its name
711 * @net: the applicable net namespace
712 * @name: name to find
714 * Find an interface by name.
715 * If the name is found a pointer to the device is returned.
716 * If the name is not found then %NULL is returned.
717 * The reference counters are not incremented so the caller must be
718 * careful with locks. The caller must hold RCU lock.
721 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
723 struct net_device *dev;
724 struct hlist_head *head = dev_name_hash(net, name);
726 hlist_for_each_entry_rcu(dev, head, name_hlist)
727 if (!strncmp(dev->name, name, IFNAMSIZ))
732 EXPORT_SYMBOL(dev_get_by_name_rcu);
735 * dev_get_by_name - find a device by its name
736 * @net: the applicable net namespace
737 * @name: name to find
739 * Find an interface by name. This can be called from any
740 * context and does its own locking. The returned handle has
741 * the usage count incremented and the caller must use dev_put() to
742 * release it when it is no longer needed. %NULL is returned if no
743 * matching device is found.
746 struct net_device *dev_get_by_name(struct net *net, const char *name)
748 struct net_device *dev;
751 dev = dev_get_by_name_rcu(net, name);
757 EXPORT_SYMBOL(dev_get_by_name);
760 * __dev_get_by_index - find a device by its ifindex
761 * @net: the applicable net namespace
762 * @ifindex: index of device
764 * Search for an interface by index. Returns %NULL if the device
765 * is not found or a pointer to the device. The device has not
766 * had its reference counter increased so the caller must be careful
767 * about locking. The caller must hold either the RTNL semaphore
771 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
773 struct net_device *dev;
774 struct hlist_head *head = dev_index_hash(net, ifindex);
776 hlist_for_each_entry(dev, head, index_hlist)
777 if (dev->ifindex == ifindex)
782 EXPORT_SYMBOL(__dev_get_by_index);
785 * dev_get_by_index_rcu - find a device by its ifindex
786 * @net: the applicable net namespace
787 * @ifindex: index of device
789 * Search for an interface by index. Returns %NULL if the device
790 * is not found or a pointer to the device. The device has not
791 * had its reference counter increased so the caller must be careful
792 * about locking. The caller must hold RCU lock.
795 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
797 struct net_device *dev;
798 struct hlist_head *head = dev_index_hash(net, ifindex);
800 hlist_for_each_entry_rcu(dev, head, index_hlist)
801 if (dev->ifindex == ifindex)
806 EXPORT_SYMBOL(dev_get_by_index_rcu);
810 * dev_get_by_index - find a device by its ifindex
811 * @net: the applicable net namespace
812 * @ifindex: index of device
814 * Search for an interface by index. Returns NULL if the device
815 * is not found or a pointer to the device. The device returned has
816 * had a reference added and the pointer is safe until the user calls
817 * dev_put to indicate they have finished with it.
820 struct net_device *dev_get_by_index(struct net *net, int ifindex)
822 struct net_device *dev;
825 dev = dev_get_by_index_rcu(net, ifindex);
831 EXPORT_SYMBOL(dev_get_by_index);
834 * netdev_get_name - get a netdevice name, knowing its ifindex.
835 * @net: network namespace
836 * @name: a pointer to the buffer where the name will be stored.
837 * @ifindex: the ifindex of the interface to get the name from.
839 * The use of raw_seqcount_begin() and cond_resched() before
840 * retrying is required as we want to give the writers a chance
841 * to complete when CONFIG_PREEMPT is not set.
843 int netdev_get_name(struct net *net, char *name, int ifindex)
845 struct net_device *dev;
849 seq = raw_seqcount_begin(&devnet_rename_seq);
851 dev = dev_get_by_index_rcu(net, ifindex);
857 strcpy(name, dev->name);
859 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
860 mutex_lock(&devnet_rename_mutex);
861 mutex_unlock(&devnet_rename_mutex);
869 * dev_getbyhwaddr_rcu - find a device by its hardware address
870 * @net: the applicable net namespace
871 * @type: media type of device
872 * @ha: hardware address
874 * Search for an interface by MAC address. Returns NULL if the device
875 * is not found or a pointer to the device.
876 * The caller must hold RCU or RTNL.
877 * The returned device has not had its ref count increased
878 * and the caller must therefore be careful about locking
882 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
885 struct net_device *dev;
887 for_each_netdev_rcu(net, dev)
888 if (dev->type == type &&
889 !memcmp(dev->dev_addr, ha, dev->addr_len))
894 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
896 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
898 struct net_device *dev;
901 for_each_netdev(net, dev)
902 if (dev->type == type)
907 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
909 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
911 struct net_device *dev, *ret = NULL;
914 for_each_netdev_rcu(net, dev)
915 if (dev->type == type) {
923 EXPORT_SYMBOL(dev_getfirstbyhwtype);
926 * __dev_get_by_flags - find any device with given flags
927 * @net: the applicable net namespace
928 * @if_flags: IFF_* values
929 * @mask: bitmask of bits in if_flags to check
931 * Search for any interface with the given flags. Returns NULL if a device
932 * is not found or a pointer to the device. Must be called inside
933 * rtnl_lock(), and result refcount is unchanged.
936 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
939 struct net_device *dev, *ret;
944 for_each_netdev(net, dev) {
945 if (((dev->flags ^ if_flags) & mask) == 0) {
952 EXPORT_SYMBOL(__dev_get_by_flags);
955 * dev_valid_name - check if name is okay for network device
958 * Network device names need to be valid file names to
959 * to allow sysfs to work. We also disallow any kind of
962 bool dev_valid_name(const char *name)
966 if (strlen(name) >= IFNAMSIZ)
968 if (!strcmp(name, ".") || !strcmp(name, ".."))
972 if (*name == '/' || *name == ':' || isspace(*name))
978 EXPORT_SYMBOL(dev_valid_name);
981 * __dev_alloc_name - allocate a name for a device
982 * @net: network namespace to allocate the device name in
983 * @name: name format string
984 * @buf: scratch buffer and result name string
986 * Passed a format string - eg "lt%d" it will try and find a suitable
987 * id. It scans list of devices to build up a free map, then chooses
988 * the first empty slot. The caller must hold the dev_base or rtnl lock
989 * while allocating the name and adding the device in order to avoid
991 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
992 * Returns the number of the unit assigned or a negative errno code.
995 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
999 const int max_netdevices = 8*PAGE_SIZE;
1000 unsigned long *inuse;
1001 struct net_device *d;
1003 p = strnchr(name, IFNAMSIZ-1, '%');
1006 * Verify the string as this thing may have come from
1007 * the user. There must be either one "%d" and no other "%"
1010 if (p[1] != 'd' || strchr(p + 2, '%'))
1013 /* Use one page as a bit array of possible slots */
1014 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1018 for_each_netdev(net, d) {
1019 if (!sscanf(d->name, name, &i))
1021 if (i < 0 || i >= max_netdevices)
1024 /* avoid cases where sscanf is not exact inverse of printf */
1025 snprintf(buf, IFNAMSIZ, name, i);
1026 if (!strncmp(buf, d->name, IFNAMSIZ))
1030 i = find_first_zero_bit(inuse, max_netdevices);
1031 free_page((unsigned long) inuse);
1035 snprintf(buf, IFNAMSIZ, name, i);
1036 if (!__dev_get_by_name(net, buf))
1039 /* It is possible to run out of possible slots
1040 * when the name is long and there isn't enough space left
1041 * for the digits, or if all bits are used.
1047 * dev_alloc_name - allocate a name for a device
1049 * @name: name format string
1051 * Passed a format string - eg "lt%d" it will try and find a suitable
1052 * id. It scans list of devices to build up a free map, then chooses
1053 * the first empty slot. The caller must hold the dev_base or rtnl lock
1054 * while allocating the name and adding the device in order to avoid
1056 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1057 * Returns the number of the unit assigned or a negative errno code.
1060 int dev_alloc_name(struct net_device *dev, const char *name)
1066 BUG_ON(!dev_net(dev));
1068 ret = __dev_alloc_name(net, name, buf);
1070 strlcpy(dev->name, buf, IFNAMSIZ);
1073 EXPORT_SYMBOL(dev_alloc_name);
1075 static int dev_alloc_name_ns(struct net *net,
1076 struct net_device *dev,
1082 ret = __dev_alloc_name(net, name, buf);
1084 strlcpy(dev->name, buf, IFNAMSIZ);
1088 static int dev_get_valid_name(struct net *net,
1089 struct net_device *dev,
1094 if (!dev_valid_name(name))
1097 if (strchr(name, '%'))
1098 return dev_alloc_name_ns(net, dev, name);
1099 else if (__dev_get_by_name(net, name))
1101 else if (dev->name != name)
1102 strlcpy(dev->name, name, IFNAMSIZ);
1108 * dev_change_name - change name of a device
1110 * @newname: name (or format string) must be at least IFNAMSIZ
1112 * Change name of a device, can pass format strings "eth%d".
1115 int dev_change_name(struct net_device *dev, const char *newname)
1117 unsigned char old_assign_type;
1118 char oldname[IFNAMSIZ];
1124 BUG_ON(!dev_net(dev));
1127 if (dev->flags & IFF_UP)
1130 mutex_lock(&devnet_rename_mutex);
1131 __raw_write_seqcount_begin(&devnet_rename_seq);
1133 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1136 memcpy(oldname, dev->name, IFNAMSIZ);
1138 err = dev_get_valid_name(net, dev, newname);
1142 if (oldname[0] && !strchr(oldname, '%'))
1143 netdev_info(dev, "renamed from %s\n", oldname);
1145 old_assign_type = dev->name_assign_type;
1146 dev->name_assign_type = NET_NAME_RENAMED;
1149 ret = device_rename(&dev->dev, dev->name);
1151 memcpy(dev->name, oldname, IFNAMSIZ);
1152 dev->name_assign_type = old_assign_type;
1157 __raw_write_seqcount_end(&devnet_rename_seq);
1158 mutex_unlock(&devnet_rename_mutex);
1160 netdev_adjacent_rename_links(dev, oldname);
1162 write_lock_bh(&dev_base_lock);
1163 hlist_del_rcu(&dev->name_hlist);
1164 write_unlock_bh(&dev_base_lock);
1168 write_lock_bh(&dev_base_lock);
1169 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1170 write_unlock_bh(&dev_base_lock);
1172 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1173 ret = notifier_to_errno(ret);
1176 /* err >= 0 after dev_alloc_name() or stores the first errno */
1179 mutex_lock(&devnet_rename_mutex);
1180 __raw_write_seqcount_begin(&devnet_rename_seq);
1181 memcpy(dev->name, oldname, IFNAMSIZ);
1182 memcpy(oldname, newname, IFNAMSIZ);
1183 dev->name_assign_type = old_assign_type;
1184 old_assign_type = NET_NAME_RENAMED;
1187 pr_err("%s: name change rollback failed: %d\n",
1195 __raw_write_seqcount_end(&devnet_rename_seq);
1196 mutex_unlock(&devnet_rename_mutex);
1201 * dev_set_alias - change ifalias of a device
1203 * @alias: name up to IFALIASZ
1204 * @len: limit of bytes to copy from info
1206 * Set ifalias for a device,
1208 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1214 if (len >= IFALIASZ)
1218 kfree(dev->ifalias);
1219 dev->ifalias = NULL;
1223 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1226 dev->ifalias = new_ifalias;
1228 strlcpy(dev->ifalias, alias, len+1);
1234 * netdev_features_change - device changes features
1235 * @dev: device to cause notification
1237 * Called to indicate a device has changed features.
1239 void netdev_features_change(struct net_device *dev)
1241 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1243 EXPORT_SYMBOL(netdev_features_change);
1246 * netdev_state_change - device changes state
1247 * @dev: device to cause notification
1249 * Called to indicate a device has changed state. This function calls
1250 * the notifier chains for netdev_chain and sends a NEWLINK message
1251 * to the routing socket.
1253 void netdev_state_change(struct net_device *dev)
1255 if (dev->flags & IFF_UP) {
1256 struct netdev_notifier_change_info change_info;
1258 change_info.flags_changed = 0;
1259 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1261 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1264 EXPORT_SYMBOL(netdev_state_change);
1267 * netdev_notify_peers - notify network peers about existence of @dev
1268 * @dev: network device
1270 * Generate traffic such that interested network peers are aware of
1271 * @dev, such as by generating a gratuitous ARP. This may be used when
1272 * a device wants to inform the rest of the network about some sort of
1273 * reconfiguration such as a failover event or virtual machine
1276 void netdev_notify_peers(struct net_device *dev)
1279 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1282 EXPORT_SYMBOL(netdev_notify_peers);
1284 static int __dev_open(struct net_device *dev)
1286 const struct net_device_ops *ops = dev->netdev_ops;
1291 if (!netif_device_present(dev))
1294 /* Block netpoll from trying to do any rx path servicing.
1295 * If we don't do this there is a chance ndo_poll_controller
1296 * or ndo_poll may be running while we open the device
1298 netpoll_poll_disable(dev);
1300 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1301 ret = notifier_to_errno(ret);
1305 set_bit(__LINK_STATE_START, &dev->state);
1307 if (ops->ndo_validate_addr)
1308 ret = ops->ndo_validate_addr(dev);
1310 if (!ret && ops->ndo_open)
1311 ret = ops->ndo_open(dev);
1313 netpoll_poll_enable(dev);
1316 clear_bit(__LINK_STATE_START, &dev->state);
1318 dev->flags |= IFF_UP;
1319 dev_set_rx_mode(dev);
1321 add_device_randomness(dev->dev_addr, dev->addr_len);
1328 * dev_open - prepare an interface for use.
1329 * @dev: device to open
1331 * Takes a device from down to up state. The device's private open
1332 * function is invoked and then the multicast lists are loaded. Finally
1333 * the device is moved into the up state and a %NETDEV_UP message is
1334 * sent to the netdev notifier chain.
1336 * Calling this function on an active interface is a nop. On a failure
1337 * a negative errno code is returned.
1339 int dev_open(struct net_device *dev)
1343 if (dev->flags & IFF_UP)
1346 ret = __dev_open(dev);
1350 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1351 call_netdevice_notifiers(NETDEV_UP, dev);
1355 EXPORT_SYMBOL(dev_open);
1357 static int __dev_close_many(struct list_head *head)
1359 struct net_device *dev;
1364 list_for_each_entry(dev, head, close_list) {
1365 /* Temporarily disable netpoll until the interface is down */
1366 netpoll_poll_disable(dev);
1368 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1370 clear_bit(__LINK_STATE_START, &dev->state);
1372 /* Synchronize to scheduled poll. We cannot touch poll list, it
1373 * can be even on different cpu. So just clear netif_running().
1375 * dev->stop() will invoke napi_disable() on all of it's
1376 * napi_struct instances on this device.
1378 smp_mb__after_atomic(); /* Commit netif_running(). */
1381 dev_deactivate_many(head);
1383 list_for_each_entry(dev, head, close_list) {
1384 const struct net_device_ops *ops = dev->netdev_ops;
1387 * Call the device specific close. This cannot fail.
1388 * Only if device is UP
1390 * We allow it to be called even after a DETACH hot-plug
1396 dev->flags &= ~IFF_UP;
1397 netpoll_poll_enable(dev);
1403 static int __dev_close(struct net_device *dev)
1408 list_add(&dev->close_list, &single);
1409 retval = __dev_close_many(&single);
1415 int dev_close_many(struct list_head *head, bool unlink)
1417 struct net_device *dev, *tmp;
1419 /* Remove the devices that don't need to be closed */
1420 list_for_each_entry_safe(dev, tmp, head, close_list)
1421 if (!(dev->flags & IFF_UP))
1422 list_del_init(&dev->close_list);
1424 __dev_close_many(head);
1426 list_for_each_entry_safe(dev, tmp, head, close_list) {
1427 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1428 call_netdevice_notifiers(NETDEV_DOWN, dev);
1430 list_del_init(&dev->close_list);
1435 EXPORT_SYMBOL(dev_close_many);
1438 * dev_close - shutdown an interface.
1439 * @dev: device to shutdown
1441 * This function moves an active device into down state. A
1442 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1443 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1446 int dev_close(struct net_device *dev)
1448 if (dev->flags & IFF_UP) {
1451 list_add(&dev->close_list, &single);
1452 dev_close_many(&single, true);
1457 EXPORT_SYMBOL(dev_close);
1461 * dev_disable_lro - disable Large Receive Offload on a device
1464 * Disable Large Receive Offload (LRO) on a net device. Must be
1465 * called under RTNL. This is needed if received packets may be
1466 * forwarded to another interface.
1468 void dev_disable_lro(struct net_device *dev)
1470 struct net_device *lower_dev;
1471 struct list_head *iter;
1473 dev->wanted_features &= ~NETIF_F_LRO;
1474 netdev_update_features(dev);
1476 if (unlikely(dev->features & NETIF_F_LRO))
1477 netdev_WARN(dev, "failed to disable LRO!\n");
1479 netdev_for_each_lower_dev(dev, lower_dev, iter)
1480 dev_disable_lro(lower_dev);
1482 EXPORT_SYMBOL(dev_disable_lro);
1484 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1485 struct net_device *dev)
1487 struct netdev_notifier_info info;
1489 netdev_notifier_info_init(&info, dev);
1490 return nb->notifier_call(nb, val, &info);
1493 static int dev_boot_phase = 1;
1496 * register_netdevice_notifier - register a network notifier block
1499 * Register a notifier to be called when network device events occur.
1500 * The notifier passed is linked into the kernel structures and must
1501 * not be reused until it has been unregistered. A negative errno code
1502 * is returned on a failure.
1504 * When registered all registration and up events are replayed
1505 * to the new notifier to allow device to have a race free
1506 * view of the network device list.
1509 int register_netdevice_notifier(struct notifier_block *nb)
1511 struct net_device *dev;
1512 struct net_device *last;
1517 err = raw_notifier_chain_register(&netdev_chain, nb);
1523 for_each_netdev(net, dev) {
1524 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1525 err = notifier_to_errno(err);
1529 if (!(dev->flags & IFF_UP))
1532 call_netdevice_notifier(nb, NETDEV_UP, dev);
1543 for_each_netdev(net, dev) {
1547 if (dev->flags & IFF_UP) {
1548 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1550 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1552 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1557 raw_notifier_chain_unregister(&netdev_chain, nb);
1560 EXPORT_SYMBOL(register_netdevice_notifier);
1563 * unregister_netdevice_notifier - unregister a network notifier block
1566 * Unregister a notifier previously registered by
1567 * register_netdevice_notifier(). The notifier is unlinked into the
1568 * kernel structures and may then be reused. A negative errno code
1569 * is returned on a failure.
1571 * After unregistering unregister and down device events are synthesized
1572 * for all devices on the device list to the removed notifier to remove
1573 * the need for special case cleanup code.
1576 int unregister_netdevice_notifier(struct notifier_block *nb)
1578 struct net_device *dev;
1583 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1588 for_each_netdev(net, dev) {
1589 if (dev->flags & IFF_UP) {
1590 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1592 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1594 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1601 EXPORT_SYMBOL(unregister_netdevice_notifier);
1604 * call_netdevice_notifiers_info - call all network notifier blocks
1605 * @val: value passed unmodified to notifier function
1606 * @dev: net_device pointer passed unmodified to notifier function
1607 * @info: notifier information data
1609 * Call all network notifier blocks. Parameters and return value
1610 * are as for raw_notifier_call_chain().
1613 static int call_netdevice_notifiers_info(unsigned long val,
1614 struct net_device *dev,
1615 struct netdev_notifier_info *info)
1618 netdev_notifier_info_init(info, dev);
1619 return raw_notifier_call_chain(&netdev_chain, val, info);
1623 * call_netdevice_notifiers - call all network notifier blocks
1624 * @val: value passed unmodified to notifier function
1625 * @dev: net_device pointer passed unmodified to notifier function
1627 * Call all network notifier blocks. Parameters and return value
1628 * are as for raw_notifier_call_chain().
1631 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1633 struct netdev_notifier_info info;
1635 return call_netdevice_notifiers_info(val, dev, &info);
1637 EXPORT_SYMBOL(call_netdevice_notifiers);
1639 #ifdef CONFIG_NET_CLS_ACT
1640 static struct static_key ingress_needed __read_mostly;
1642 void net_inc_ingress_queue(void)
1644 static_key_slow_inc(&ingress_needed);
1646 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1648 void net_dec_ingress_queue(void)
1650 static_key_slow_dec(&ingress_needed);
1652 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1655 static struct static_key netstamp_needed __read_mostly;
1656 #ifdef HAVE_JUMP_LABEL
1657 /* We are not allowed to call static_key_slow_dec() from irq context
1658 * If net_disable_timestamp() is called from irq context, defer the
1659 * static_key_slow_dec() calls.
1661 static atomic_t netstamp_needed_deferred;
1664 void net_enable_timestamp(void)
1666 #ifdef HAVE_JUMP_LABEL
1667 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1671 static_key_slow_dec(&netstamp_needed);
1675 static_key_slow_inc(&netstamp_needed);
1677 EXPORT_SYMBOL(net_enable_timestamp);
1679 void net_disable_timestamp(void)
1681 #ifdef HAVE_JUMP_LABEL
1682 if (in_interrupt()) {
1683 atomic_inc(&netstamp_needed_deferred);
1687 static_key_slow_dec(&netstamp_needed);
1689 EXPORT_SYMBOL(net_disable_timestamp);
1691 static inline void net_timestamp_set(struct sk_buff *skb)
1693 skb->tstamp.tv64 = 0;
1694 if (static_key_false(&netstamp_needed))
1695 __net_timestamp(skb);
1698 #define net_timestamp_check(COND, SKB) \
1699 if (static_key_false(&netstamp_needed)) { \
1700 if ((COND) && !(SKB)->tstamp.tv64) \
1701 __net_timestamp(SKB); \
1704 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1708 if (!(dev->flags & IFF_UP))
1711 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1712 if (skb->len <= len)
1715 /* if TSO is enabled, we don't care about the length as the packet
1716 * could be forwarded without being segmented before
1718 if (skb_is_gso(skb))
1723 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1725 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1727 if (skb_orphan_frags(skb, GFP_ATOMIC) ||
1728 unlikely(!is_skb_forwardable(dev, skb))) {
1729 atomic_long_inc(&dev->rx_dropped);
1734 skb_scrub_packet(skb, true);
1736 skb->protocol = eth_type_trans(skb, dev);
1737 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1741 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1744 * dev_forward_skb - loopback an skb to another netif
1746 * @dev: destination network device
1747 * @skb: buffer to forward
1750 * NET_RX_SUCCESS (no congestion)
1751 * NET_RX_DROP (packet was dropped, but freed)
1753 * dev_forward_skb can be used for injecting an skb from the
1754 * start_xmit function of one device into the receive queue
1755 * of another device.
1757 * The receiving device may be in another namespace, so
1758 * we have to clear all information in the skb that could
1759 * impact namespace isolation.
1761 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1763 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1765 EXPORT_SYMBOL_GPL(dev_forward_skb);
1767 static inline int deliver_skb(struct sk_buff *skb,
1768 struct packet_type *pt_prev,
1769 struct net_device *orig_dev)
1771 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1773 atomic_inc(&skb->users);
1774 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1777 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1778 struct packet_type **pt,
1779 struct net_device *orig_dev,
1781 struct list_head *ptype_list)
1783 struct packet_type *ptype, *pt_prev = *pt;
1785 list_for_each_entry_rcu(ptype, ptype_list, list) {
1786 if (ptype->type != type)
1789 deliver_skb(skb, pt_prev, orig_dev);
1795 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1797 if (!ptype->af_packet_priv || !skb->sk)
1800 if (ptype->id_match)
1801 return ptype->id_match(ptype, skb->sk);
1802 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1809 * Support routine. Sends outgoing frames to any network
1810 * taps currently in use.
1813 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1815 struct packet_type *ptype;
1816 struct sk_buff *skb2 = NULL;
1817 struct packet_type *pt_prev = NULL;
1818 struct list_head *ptype_list = &ptype_all;
1822 list_for_each_entry_rcu(ptype, ptype_list, list) {
1823 /* Never send packets back to the socket
1824 * they originated from - MvS (miquels@drinkel.ow.org)
1826 if (skb_loop_sk(ptype, skb))
1830 deliver_skb(skb2, pt_prev, skb->dev);
1835 /* need to clone skb, done only once */
1836 skb2 = skb_clone(skb, GFP_ATOMIC);
1840 net_timestamp_set(skb2);
1842 /* skb->nh should be correctly
1843 * set by sender, so that the second statement is
1844 * just protection against buggy protocols.
1846 skb_reset_mac_header(skb2);
1848 if (skb_network_header(skb2) < skb2->data ||
1849 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1850 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1851 ntohs(skb2->protocol),
1853 skb_reset_network_header(skb2);
1856 skb2->transport_header = skb2->network_header;
1857 skb2->pkt_type = PACKET_OUTGOING;
1861 if (ptype_list == &ptype_all) {
1862 ptype_list = &dev->ptype_all;
1867 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1872 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1873 * @dev: Network device
1874 * @txq: number of queues available
1876 * If real_num_tx_queues is changed the tc mappings may no longer be
1877 * valid. To resolve this verify the tc mapping remains valid and if
1878 * not NULL the mapping. With no priorities mapping to this
1879 * offset/count pair it will no longer be used. In the worst case TC0
1880 * is invalid nothing can be done so disable priority mappings. If is
1881 * expected that drivers will fix this mapping if they can before
1882 * calling netif_set_real_num_tx_queues.
1884 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1887 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1889 /* If TC0 is invalidated disable TC mapping */
1890 if (tc->offset + tc->count > txq) {
1891 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1896 /* Invalidated prio to tc mappings set to TC0 */
1897 for (i = 1; i < TC_BITMASK + 1; i++) {
1898 int q = netdev_get_prio_tc_map(dev, i);
1900 tc = &dev->tc_to_txq[q];
1901 if (tc->offset + tc->count > txq) {
1902 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1904 netdev_set_prio_tc_map(dev, i, 0);
1910 static DEFINE_MUTEX(xps_map_mutex);
1911 #define xmap_dereference(P) \
1912 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1914 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1917 struct xps_map *map = NULL;
1921 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1923 for (pos = 0; map && pos < map->len; pos++) {
1924 if (map->queues[pos] == index) {
1926 map->queues[pos] = map->queues[--map->len];
1928 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1929 kfree_rcu(map, rcu);
1939 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1941 struct xps_dev_maps *dev_maps;
1943 bool active = false;
1945 mutex_lock(&xps_map_mutex);
1946 dev_maps = xmap_dereference(dev->xps_maps);
1951 for_each_possible_cpu(cpu) {
1952 for (i = index; i < dev->num_tx_queues; i++) {
1953 if (!remove_xps_queue(dev_maps, cpu, i))
1956 if (i == dev->num_tx_queues)
1961 RCU_INIT_POINTER(dev->xps_maps, NULL);
1962 kfree_rcu(dev_maps, rcu);
1965 for (i = index; i < dev->num_tx_queues; i++)
1966 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1970 mutex_unlock(&xps_map_mutex);
1973 static struct xps_map *expand_xps_map(struct xps_map *map,
1976 struct xps_map *new_map;
1977 int alloc_len = XPS_MIN_MAP_ALLOC;
1980 for (pos = 0; map && pos < map->len; pos++) {
1981 if (map->queues[pos] != index)
1986 /* Need to add queue to this CPU's existing map */
1988 if (pos < map->alloc_len)
1991 alloc_len = map->alloc_len * 2;
1994 /* Need to allocate new map to store queue on this CPU's map */
1995 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2000 for (i = 0; i < pos; i++)
2001 new_map->queues[i] = map->queues[i];
2002 new_map->alloc_len = alloc_len;
2008 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2011 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2012 struct xps_map *map, *new_map;
2013 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2014 int cpu, numa_node_id = -2;
2015 bool active = false;
2017 mutex_lock(&xps_map_mutex);
2019 dev_maps = xmap_dereference(dev->xps_maps);
2021 /* allocate memory for queue storage */
2022 for_each_online_cpu(cpu) {
2023 if (!cpumask_test_cpu(cpu, mask))
2027 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2028 if (!new_dev_maps) {
2029 mutex_unlock(&xps_map_mutex);
2033 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2036 map = expand_xps_map(map, cpu, index);
2040 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2044 goto out_no_new_maps;
2046 for_each_possible_cpu(cpu) {
2047 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2048 /* add queue to CPU maps */
2051 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2052 while ((pos < map->len) && (map->queues[pos] != index))
2055 if (pos == map->len)
2056 map->queues[map->len++] = index;
2058 if (numa_node_id == -2)
2059 numa_node_id = cpu_to_node(cpu);
2060 else if (numa_node_id != cpu_to_node(cpu))
2063 } else if (dev_maps) {
2064 /* fill in the new device map from the old device map */
2065 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2066 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2071 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2073 /* Cleanup old maps */
2075 for_each_possible_cpu(cpu) {
2076 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2077 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2078 if (map && map != new_map)
2079 kfree_rcu(map, rcu);
2082 kfree_rcu(dev_maps, rcu);
2085 dev_maps = new_dev_maps;
2089 /* update Tx queue numa node */
2090 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2091 (numa_node_id >= 0) ? numa_node_id :
2097 /* removes queue from unused CPUs */
2098 for_each_possible_cpu(cpu) {
2099 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2102 if (remove_xps_queue(dev_maps, cpu, index))
2106 /* free map if not active */
2108 RCU_INIT_POINTER(dev->xps_maps, NULL);
2109 kfree_rcu(dev_maps, rcu);
2113 mutex_unlock(&xps_map_mutex);
2117 /* remove any maps that we added */
2118 for_each_possible_cpu(cpu) {
2119 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2120 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2122 if (new_map && new_map != map)
2126 mutex_unlock(&xps_map_mutex);
2128 kfree(new_dev_maps);
2131 EXPORT_SYMBOL(netif_set_xps_queue);
2135 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2136 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2138 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2142 if (txq < 1 || txq > dev->num_tx_queues)
2145 if (dev->reg_state == NETREG_REGISTERED ||
2146 dev->reg_state == NETREG_UNREGISTERING) {
2149 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2155 netif_setup_tc(dev, txq);
2157 if (txq < dev->real_num_tx_queues) {
2158 qdisc_reset_all_tx_gt(dev, txq);
2160 netif_reset_xps_queues_gt(dev, txq);
2165 dev->real_num_tx_queues = txq;
2168 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2172 * netif_set_real_num_rx_queues - set actual number of RX queues used
2173 * @dev: Network device
2174 * @rxq: Actual number of RX queues
2176 * This must be called either with the rtnl_lock held or before
2177 * registration of the net device. Returns 0 on success, or a
2178 * negative error code. If called before registration, it always
2181 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2185 if (rxq < 1 || rxq > dev->num_rx_queues)
2188 if (dev->reg_state == NETREG_REGISTERED) {
2191 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2197 dev->real_num_rx_queues = rxq;
2200 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2204 * netif_get_num_default_rss_queues - default number of RSS queues
2206 * This routine should set an upper limit on the number of RSS queues
2207 * used by default by multiqueue devices.
2209 int netif_get_num_default_rss_queues(void)
2211 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2213 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2215 static inline void __netif_reschedule(struct Qdisc *q)
2217 struct softnet_data *sd;
2218 unsigned long flags;
2220 local_irq_save(flags);
2221 sd = this_cpu_ptr(&softnet_data);
2222 q->next_sched = NULL;
2223 *sd->output_queue_tailp = q;
2224 sd->output_queue_tailp = &q->next_sched;
2225 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2226 local_irq_restore(flags);
2227 preempt_check_resched_rt();
2230 void __netif_schedule(struct Qdisc *q)
2232 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2233 __netif_reschedule(q);
2235 EXPORT_SYMBOL(__netif_schedule);
2237 struct dev_kfree_skb_cb {
2238 enum skb_free_reason reason;
2241 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2243 return (struct dev_kfree_skb_cb *)skb->cb;
2246 void netif_schedule_queue(struct netdev_queue *txq)
2249 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2250 struct Qdisc *q = rcu_dereference(txq->qdisc);
2252 __netif_schedule(q);
2256 EXPORT_SYMBOL(netif_schedule_queue);
2259 * netif_wake_subqueue - allow sending packets on subqueue
2260 * @dev: network device
2261 * @queue_index: sub queue index
2263 * Resume individual transmit queue of a device with multiple transmit queues.
2265 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2267 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2269 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2273 q = rcu_dereference(txq->qdisc);
2274 __netif_schedule(q);
2278 EXPORT_SYMBOL(netif_wake_subqueue);
2280 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2282 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2286 q = rcu_dereference(dev_queue->qdisc);
2287 __netif_schedule(q);
2291 EXPORT_SYMBOL(netif_tx_wake_queue);
2293 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2295 unsigned long flags;
2297 if (likely(atomic_read(&skb->users) == 1)) {
2299 atomic_set(&skb->users, 0);
2300 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2303 get_kfree_skb_cb(skb)->reason = reason;
2304 local_irq_save(flags);
2305 skb->next = __this_cpu_read(softnet_data.completion_queue);
2306 __this_cpu_write(softnet_data.completion_queue, skb);
2307 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2308 local_irq_restore(flags);
2309 preempt_check_resched_rt();
2311 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2313 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2315 if (in_irq() || irqs_disabled())
2316 __dev_kfree_skb_irq(skb, reason);
2320 EXPORT_SYMBOL(__dev_kfree_skb_any);
2324 * netif_device_detach - mark device as removed
2325 * @dev: network device
2327 * Mark device as removed from system and therefore no longer available.
2329 void netif_device_detach(struct net_device *dev)
2331 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2332 netif_running(dev)) {
2333 netif_tx_stop_all_queues(dev);
2336 EXPORT_SYMBOL(netif_device_detach);
2339 * netif_device_attach - mark device as attached
2340 * @dev: network device
2342 * Mark device as attached from system and restart if needed.
2344 void netif_device_attach(struct net_device *dev)
2346 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2347 netif_running(dev)) {
2348 netif_tx_wake_all_queues(dev);
2349 __netdev_watchdog_up(dev);
2352 EXPORT_SYMBOL(netif_device_attach);
2354 static void skb_warn_bad_offload(const struct sk_buff *skb)
2356 static const netdev_features_t null_features = 0;
2357 struct net_device *dev = skb->dev;
2358 const char *driver = "";
2360 if (!net_ratelimit())
2363 if (dev && dev->dev.parent)
2364 driver = dev_driver_string(dev->dev.parent);
2366 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2367 "gso_type=%d ip_summed=%d\n",
2368 driver, dev ? &dev->features : &null_features,
2369 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2370 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2371 skb_shinfo(skb)->gso_type, skb->ip_summed);
2375 * Invalidate hardware checksum when packet is to be mangled, and
2376 * complete checksum manually on outgoing path.
2378 int skb_checksum_help(struct sk_buff *skb)
2381 int ret = 0, offset;
2383 if (skb->ip_summed == CHECKSUM_COMPLETE)
2384 goto out_set_summed;
2386 if (unlikely(skb_shinfo(skb)->gso_size)) {
2387 skb_warn_bad_offload(skb);
2391 /* Before computing a checksum, we should make sure no frag could
2392 * be modified by an external entity : checksum could be wrong.
2394 if (skb_has_shared_frag(skb)) {
2395 ret = __skb_linearize(skb);
2400 offset = skb_checksum_start_offset(skb);
2401 BUG_ON(offset >= skb_headlen(skb));
2402 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2404 offset += skb->csum_offset;
2405 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2407 if (skb_cloned(skb) &&
2408 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2409 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2414 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2416 skb->ip_summed = CHECKSUM_NONE;
2420 EXPORT_SYMBOL(skb_checksum_help);
2422 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2424 __be16 type = skb->protocol;
2426 /* Tunnel gso handlers can set protocol to ethernet. */
2427 if (type == htons(ETH_P_TEB)) {
2430 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2433 eth = (struct ethhdr *)skb_mac_header(skb);
2434 type = eth->h_proto;
2437 return __vlan_get_protocol(skb, type, depth);
2441 * skb_mac_gso_segment - mac layer segmentation handler.
2442 * @skb: buffer to segment
2443 * @features: features for the output path (see dev->features)
2445 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2446 netdev_features_t features)
2448 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2449 struct packet_offload *ptype;
2450 int vlan_depth = skb->mac_len;
2451 __be16 type = skb_network_protocol(skb, &vlan_depth);
2453 if (unlikely(!type))
2454 return ERR_PTR(-EINVAL);
2456 __skb_pull(skb, vlan_depth);
2459 list_for_each_entry_rcu(ptype, &offload_base, list) {
2460 if (ptype->type == type && ptype->callbacks.gso_segment) {
2461 segs = ptype->callbacks.gso_segment(skb, features);
2467 __skb_push(skb, skb->data - skb_mac_header(skb));
2471 EXPORT_SYMBOL(skb_mac_gso_segment);
2474 /* openvswitch calls this on rx path, so we need a different check.
2476 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2479 return skb->ip_summed != CHECKSUM_PARTIAL;
2481 return skb->ip_summed == CHECKSUM_NONE;
2485 * __skb_gso_segment - Perform segmentation on skb.
2486 * @skb: buffer to segment
2487 * @features: features for the output path (see dev->features)
2488 * @tx_path: whether it is called in TX path
2490 * This function segments the given skb and returns a list of segments.
2492 * It may return NULL if the skb requires no segmentation. This is
2493 * only possible when GSO is used for verifying header integrity.
2495 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2496 netdev_features_t features, bool tx_path)
2498 if (unlikely(skb_needs_check(skb, tx_path))) {
2501 skb_warn_bad_offload(skb);
2503 err = skb_cow_head(skb, 0);
2505 return ERR_PTR(err);
2508 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2509 SKB_GSO_CB(skb)->encap_level = 0;
2511 skb_reset_mac_header(skb);
2512 skb_reset_mac_len(skb);
2514 return skb_mac_gso_segment(skb, features);
2516 EXPORT_SYMBOL(__skb_gso_segment);
2518 /* Take action when hardware reception checksum errors are detected. */
2520 void netdev_rx_csum_fault(struct net_device *dev)
2522 if (net_ratelimit()) {
2523 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2527 EXPORT_SYMBOL(netdev_rx_csum_fault);
2530 /* Actually, we should eliminate this check as soon as we know, that:
2531 * 1. IOMMU is present and allows to map all the memory.
2532 * 2. No high memory really exists on this machine.
2535 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2537 #ifdef CONFIG_HIGHMEM
2539 if (!(dev->features & NETIF_F_HIGHDMA)) {
2540 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2541 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2542 if (PageHighMem(skb_frag_page(frag)))
2547 if (PCI_DMA_BUS_IS_PHYS) {
2548 struct device *pdev = dev->dev.parent;
2552 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2553 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2554 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2555 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2563 /* If MPLS offload request, verify we are testing hardware MPLS features
2564 * instead of standard features for the netdev.
2566 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2567 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2568 netdev_features_t features,
2571 if (eth_p_mpls(type))
2572 features &= skb->dev->mpls_features;
2577 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2578 netdev_features_t features,
2585 static netdev_features_t harmonize_features(struct sk_buff *skb,
2586 netdev_features_t features)
2591 type = skb_network_protocol(skb, &tmp);
2592 features = net_mpls_features(skb, features, type);
2594 if (skb->ip_summed != CHECKSUM_NONE &&
2595 !can_checksum_protocol(features, type)) {
2596 features &= ~NETIF_F_ALL_CSUM;
2597 } else if (illegal_highdma(skb->dev, skb)) {
2598 features &= ~NETIF_F_SG;
2604 netdev_features_t passthru_features_check(struct sk_buff *skb,
2605 struct net_device *dev,
2606 netdev_features_t features)
2610 EXPORT_SYMBOL(passthru_features_check);
2612 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2613 struct net_device *dev,
2614 netdev_features_t features)
2616 return vlan_features_check(skb, features);
2619 netdev_features_t netif_skb_features(struct sk_buff *skb)
2621 struct net_device *dev = skb->dev;
2622 netdev_features_t features = dev->features;
2623 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2625 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2626 features &= ~NETIF_F_GSO_MASK;
2628 /* If encapsulation offload request, verify we are testing
2629 * hardware encapsulation features instead of standard
2630 * features for the netdev
2632 if (skb->encapsulation)
2633 features &= dev->hw_enc_features;
2635 if (skb_vlan_tagged(skb))
2636 features = netdev_intersect_features(features,
2637 dev->vlan_features |
2638 NETIF_F_HW_VLAN_CTAG_TX |
2639 NETIF_F_HW_VLAN_STAG_TX);
2641 if (dev->netdev_ops->ndo_features_check)
2642 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2645 features &= dflt_features_check(skb, dev, features);
2647 return harmonize_features(skb, features);
2649 EXPORT_SYMBOL(netif_skb_features);
2651 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2652 struct netdev_queue *txq, bool more)
2657 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2658 dev_queue_xmit_nit(skb, dev);
2661 trace_net_dev_start_xmit(skb, dev);
2662 rc = netdev_start_xmit(skb, dev, txq, more);
2663 trace_net_dev_xmit(skb, rc, dev, len);
2668 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2669 struct netdev_queue *txq, int *ret)
2671 struct sk_buff *skb = first;
2672 int rc = NETDEV_TX_OK;
2675 struct sk_buff *next = skb->next;
2678 rc = xmit_one(skb, dev, txq, next != NULL);
2679 if (unlikely(!dev_xmit_complete(rc))) {
2685 if (netif_xmit_stopped(txq) && skb) {
2686 rc = NETDEV_TX_BUSY;
2696 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2697 netdev_features_t features)
2699 if (skb_vlan_tag_present(skb) &&
2700 !vlan_hw_offload_capable(features, skb->vlan_proto))
2701 skb = __vlan_hwaccel_push_inside(skb);
2705 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2707 netdev_features_t features;
2712 features = netif_skb_features(skb);
2713 skb = validate_xmit_vlan(skb, features);
2717 if (netif_needs_gso(skb, features)) {
2718 struct sk_buff *segs;
2720 segs = skb_gso_segment(skb, features);
2728 if (skb_needs_linearize(skb, features) &&
2729 __skb_linearize(skb))
2732 /* If packet is not checksummed and device does not
2733 * support checksumming for this protocol, complete
2734 * checksumming here.
2736 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2737 if (skb->encapsulation)
2738 skb_set_inner_transport_header(skb,
2739 skb_checksum_start_offset(skb));
2741 skb_set_transport_header(skb,
2742 skb_checksum_start_offset(skb));
2743 if (!(features & NETIF_F_ALL_CSUM) &&
2744 skb_checksum_help(skb))
2757 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2759 struct sk_buff *next, *head = NULL, *tail;
2761 for (; skb != NULL; skb = next) {
2765 /* in case skb wont be segmented, point to itself */
2768 skb = validate_xmit_skb(skb, dev);
2776 /* If skb was segmented, skb->prev points to
2777 * the last segment. If not, it still contains skb.
2784 static void qdisc_pkt_len_init(struct sk_buff *skb)
2786 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2788 qdisc_skb_cb(skb)->pkt_len = skb->len;
2790 /* To get more precise estimation of bytes sent on wire,
2791 * we add to pkt_len the headers size of all segments
2793 if (shinfo->gso_size) {
2794 unsigned int hdr_len;
2795 u16 gso_segs = shinfo->gso_segs;
2797 /* mac layer + network layer */
2798 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2800 /* + transport layer */
2801 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2802 hdr_len += tcp_hdrlen(skb);
2804 hdr_len += sizeof(struct udphdr);
2806 if (shinfo->gso_type & SKB_GSO_DODGY)
2807 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2810 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2814 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2815 struct net_device *dev,
2816 struct netdev_queue *txq)
2818 spinlock_t *root_lock = qdisc_lock(q);
2822 qdisc_pkt_len_init(skb);
2823 qdisc_calculate_pkt_len(skb, q);
2825 * Heuristic to force contended enqueues to serialize on a
2826 * separate lock before trying to get qdisc main lock.
2827 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2828 * often and dequeue packets faster.
2830 contended = qdisc_is_running(q);
2831 if (unlikely(contended))
2832 spin_lock(&q->busylock);
2834 spin_lock(root_lock);
2835 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2838 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2839 qdisc_run_begin(q)) {
2841 * This is a work-conserving queue; there are no old skbs
2842 * waiting to be sent out; and the qdisc is not running -
2843 * xmit the skb directly.
2846 qdisc_bstats_update(q, skb);
2848 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2849 if (unlikely(contended)) {
2850 spin_unlock(&q->busylock);
2857 rc = NET_XMIT_SUCCESS;
2859 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2860 if (qdisc_run_begin(q)) {
2861 if (unlikely(contended)) {
2862 spin_unlock(&q->busylock);
2868 spin_unlock(root_lock);
2869 if (unlikely(contended))
2870 spin_unlock(&q->busylock);
2874 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2875 static void skb_update_prio(struct sk_buff *skb)
2877 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2879 if (!skb->priority && skb->sk && map) {
2880 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2882 if (prioidx < map->priomap_len)
2883 skb->priority = map->priomap[prioidx];
2887 #define skb_update_prio(skb)
2890 DEFINE_PER_CPU(int, xmit_recursion);
2891 EXPORT_SYMBOL(xmit_recursion);
2893 #define RECURSION_LIMIT 10
2896 * dev_loopback_xmit - loop back @skb
2897 * @skb: buffer to transmit
2899 int dev_loopback_xmit(struct sock *sk, struct sk_buff *skb)
2901 skb_reset_mac_header(skb);
2902 __skb_pull(skb, skb_network_offset(skb));
2903 skb->pkt_type = PACKET_LOOPBACK;
2904 skb->ip_summed = CHECKSUM_UNNECESSARY;
2905 WARN_ON(!skb_dst(skb));
2910 EXPORT_SYMBOL(dev_loopback_xmit);
2913 * __dev_queue_xmit - transmit a buffer
2914 * @skb: buffer to transmit
2915 * @accel_priv: private data used for L2 forwarding offload
2917 * Queue a buffer for transmission to a network device. The caller must
2918 * have set the device and priority and built the buffer before calling
2919 * this function. The function can be called from an interrupt.
2921 * A negative errno code is returned on a failure. A success does not
2922 * guarantee the frame will be transmitted as it may be dropped due
2923 * to congestion or traffic shaping.
2925 * -----------------------------------------------------------------------------------
2926 * I notice this method can also return errors from the queue disciplines,
2927 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2930 * Regardless of the return value, the skb is consumed, so it is currently
2931 * difficult to retry a send to this method. (You can bump the ref count
2932 * before sending to hold a reference for retry if you are careful.)
2934 * When calling this method, interrupts MUST be enabled. This is because
2935 * the BH enable code must have IRQs enabled so that it will not deadlock.
2938 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2940 struct net_device *dev = skb->dev;
2941 struct netdev_queue *txq;
2945 skb_reset_mac_header(skb);
2947 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
2948 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
2950 /* Disable soft irqs for various locks below. Also
2951 * stops preemption for RCU.
2955 skb_update_prio(skb);
2957 /* If device/qdisc don't need skb->dst, release it right now while
2958 * its hot in this cpu cache.
2960 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2965 txq = netdev_pick_tx(dev, skb, accel_priv);
2966 q = rcu_dereference_bh(txq->qdisc);
2968 #ifdef CONFIG_NET_CLS_ACT
2969 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2971 trace_net_dev_queue(skb);
2973 rc = __dev_xmit_skb(skb, q, dev, txq);
2977 /* The device has no queue. Common case for software devices:
2978 loopback, all the sorts of tunnels...
2980 Really, it is unlikely that netif_tx_lock protection is necessary
2981 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2983 However, it is possible, that they rely on protection
2986 Check this and shot the lock. It is not prone from deadlocks.
2987 Either shot noqueue qdisc, it is even simpler 8)
2989 if (dev->flags & IFF_UP) {
2990 int cpu = smp_processor_id(); /* ok because BHs are off */
2992 if (txq->xmit_lock_owner != cpu) {
2994 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2995 goto recursion_alert;
2997 skb = validate_xmit_skb(skb, dev);
3001 HARD_TX_LOCK(dev, txq, cpu);
3003 if (!netif_xmit_stopped(txq)) {
3004 __this_cpu_inc(xmit_recursion);
3005 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3006 __this_cpu_dec(xmit_recursion);
3007 if (dev_xmit_complete(rc)) {
3008 HARD_TX_UNLOCK(dev, txq);
3012 HARD_TX_UNLOCK(dev, txq);
3013 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3016 /* Recursion is detected! It is possible,
3020 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3027 rcu_read_unlock_bh();
3029 atomic_long_inc(&dev->tx_dropped);
3030 kfree_skb_list(skb);
3033 rcu_read_unlock_bh();
3037 int dev_queue_xmit_sk(struct sock *sk, struct sk_buff *skb)
3039 return __dev_queue_xmit(skb, NULL);
3041 EXPORT_SYMBOL(dev_queue_xmit_sk);
3043 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3045 return __dev_queue_xmit(skb, accel_priv);
3047 EXPORT_SYMBOL(dev_queue_xmit_accel);
3050 /*=======================================================================
3052 =======================================================================*/
3054 int netdev_max_backlog __read_mostly = 1000;
3055 EXPORT_SYMBOL(netdev_max_backlog);
3057 int netdev_tstamp_prequeue __read_mostly = 1;
3058 int netdev_budget __read_mostly = 300;
3059 int weight_p __read_mostly = 64; /* old backlog weight */
3061 /* Called with irq disabled */
3062 static inline void ____napi_schedule(struct softnet_data *sd,
3063 struct napi_struct *napi)
3065 list_add_tail(&napi->poll_list, &sd->poll_list);
3066 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3071 /* One global table that all flow-based protocols share. */
3072 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3073 EXPORT_SYMBOL(rps_sock_flow_table);
3074 u32 rps_cpu_mask __read_mostly;
3075 EXPORT_SYMBOL(rps_cpu_mask);
3077 struct static_key rps_needed __read_mostly;
3079 static struct rps_dev_flow *
3080 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3081 struct rps_dev_flow *rflow, u16 next_cpu)
3083 if (next_cpu < nr_cpu_ids) {
3084 #ifdef CONFIG_RFS_ACCEL
3085 struct netdev_rx_queue *rxqueue;
3086 struct rps_dev_flow_table *flow_table;
3087 struct rps_dev_flow *old_rflow;
3092 /* Should we steer this flow to a different hardware queue? */
3093 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3094 !(dev->features & NETIF_F_NTUPLE))
3096 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3097 if (rxq_index == skb_get_rx_queue(skb))
3100 rxqueue = dev->_rx + rxq_index;
3101 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3104 flow_id = skb_get_hash(skb) & flow_table->mask;
3105 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3106 rxq_index, flow_id);
3110 rflow = &flow_table->flows[flow_id];
3112 if (old_rflow->filter == rflow->filter)
3113 old_rflow->filter = RPS_NO_FILTER;
3117 per_cpu(softnet_data, next_cpu).input_queue_head;
3120 rflow->cpu = next_cpu;
3125 * get_rps_cpu is called from netif_receive_skb and returns the target
3126 * CPU from the RPS map of the receiving queue for a given skb.
3127 * rcu_read_lock must be held on entry.
3129 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3130 struct rps_dev_flow **rflowp)
3132 const struct rps_sock_flow_table *sock_flow_table;
3133 struct netdev_rx_queue *rxqueue = dev->_rx;
3134 struct rps_dev_flow_table *flow_table;
3135 struct rps_map *map;
3140 if (skb_rx_queue_recorded(skb)) {
3141 u16 index = skb_get_rx_queue(skb);
3143 if (unlikely(index >= dev->real_num_rx_queues)) {
3144 WARN_ONCE(dev->real_num_rx_queues > 1,
3145 "%s received packet on queue %u, but number "
3146 "of RX queues is %u\n",
3147 dev->name, index, dev->real_num_rx_queues);
3153 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3155 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3156 map = rcu_dereference(rxqueue->rps_map);
3157 if (!flow_table && !map)
3160 skb_reset_network_header(skb);
3161 hash = skb_get_hash(skb);
3165 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3166 if (flow_table && sock_flow_table) {
3167 struct rps_dev_flow *rflow;
3171 /* First check into global flow table if there is a match */
3172 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3173 if ((ident ^ hash) & ~rps_cpu_mask)
3176 next_cpu = ident & rps_cpu_mask;
3178 /* OK, now we know there is a match,
3179 * we can look at the local (per receive queue) flow table
3181 rflow = &flow_table->flows[hash & flow_table->mask];
3185 * If the desired CPU (where last recvmsg was done) is
3186 * different from current CPU (one in the rx-queue flow
3187 * table entry), switch if one of the following holds:
3188 * - Current CPU is unset (>= nr_cpu_ids).
3189 * - Current CPU is offline.
3190 * - The current CPU's queue tail has advanced beyond the
3191 * last packet that was enqueued using this table entry.
3192 * This guarantees that all previous packets for the flow
3193 * have been dequeued, thus preserving in order delivery.
3195 if (unlikely(tcpu != next_cpu) &&
3196 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3197 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3198 rflow->last_qtail)) >= 0)) {
3200 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3203 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3213 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3214 if (cpu_online(tcpu)) {
3224 #ifdef CONFIG_RFS_ACCEL
3227 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3228 * @dev: Device on which the filter was set
3229 * @rxq_index: RX queue index
3230 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3231 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3233 * Drivers that implement ndo_rx_flow_steer() should periodically call
3234 * this function for each installed filter and remove the filters for
3235 * which it returns %true.
3237 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3238 u32 flow_id, u16 filter_id)
3240 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3241 struct rps_dev_flow_table *flow_table;
3242 struct rps_dev_flow *rflow;
3247 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3248 if (flow_table && flow_id <= flow_table->mask) {
3249 rflow = &flow_table->flows[flow_id];
3250 cpu = ACCESS_ONCE(rflow->cpu);
3251 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3252 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3253 rflow->last_qtail) <
3254 (int)(10 * flow_table->mask)))
3260 EXPORT_SYMBOL(rps_may_expire_flow);
3262 #endif /* CONFIG_RFS_ACCEL */
3264 /* Called from hardirq (IPI) context */
3265 static void rps_trigger_softirq(void *data)
3267 struct softnet_data *sd = data;
3269 ____napi_schedule(sd, &sd->backlog);
3273 #endif /* CONFIG_RPS */
3276 * Check if this softnet_data structure is another cpu one
3277 * If yes, queue it to our IPI list and return 1
3280 static int rps_ipi_queued(struct softnet_data *sd)
3283 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3286 sd->rps_ipi_next = mysd->rps_ipi_list;
3287 mysd->rps_ipi_list = sd;
3289 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3292 #endif /* CONFIG_RPS */
3296 #ifdef CONFIG_NET_FLOW_LIMIT
3297 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3300 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3302 #ifdef CONFIG_NET_FLOW_LIMIT
3303 struct sd_flow_limit *fl;
3304 struct softnet_data *sd;
3305 unsigned int old_flow, new_flow;
3307 if (qlen < (netdev_max_backlog >> 1))
3310 sd = this_cpu_ptr(&softnet_data);
3313 fl = rcu_dereference(sd->flow_limit);
3315 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3316 old_flow = fl->history[fl->history_head];
3317 fl->history[fl->history_head] = new_flow;
3320 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3322 if (likely(fl->buckets[old_flow]))
3323 fl->buckets[old_flow]--;
3325 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3337 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3338 * queue (may be a remote CPU queue).
3340 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3341 unsigned int *qtail)
3343 struct softnet_data *sd;
3344 unsigned long flags;
3347 sd = &per_cpu(softnet_data, cpu);
3349 local_irq_save(flags);
3352 qlen = skb_queue_len(&sd->input_pkt_queue);
3353 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3356 __skb_queue_tail(&sd->input_pkt_queue, skb);
3357 input_queue_tail_incr_save(sd, qtail);
3359 local_irq_restore(flags);
3360 return NET_RX_SUCCESS;
3363 /* Schedule NAPI for backlog device
3364 * We can use non atomic operation since we own the queue lock
3366 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3367 if (!rps_ipi_queued(sd))
3368 ____napi_schedule(sd, &sd->backlog);
3376 local_irq_restore(flags);
3377 preempt_check_resched_rt();
3379 atomic_long_inc(&skb->dev->rx_dropped);
3384 static int netif_rx_internal(struct sk_buff *skb)
3388 net_timestamp_check(netdev_tstamp_prequeue, skb);
3390 trace_netif_rx(skb);
3392 if (static_key_false(&rps_needed)) {
3393 struct rps_dev_flow voidflow, *rflow = &voidflow;
3399 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3401 cpu = smp_processor_id();
3403 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3411 ret = enqueue_to_backlog(skb, get_cpu_light(), &qtail);
3418 * netif_rx - post buffer to the network code
3419 * @skb: buffer to post
3421 * This function receives a packet from a device driver and queues it for
3422 * the upper (protocol) levels to process. It always succeeds. The buffer
3423 * may be dropped during processing for congestion control or by the
3427 * NET_RX_SUCCESS (no congestion)
3428 * NET_RX_DROP (packet was dropped)
3432 int netif_rx(struct sk_buff *skb)
3434 trace_netif_rx_entry(skb);
3436 return netif_rx_internal(skb);
3438 EXPORT_SYMBOL(netif_rx);
3440 int netif_rx_ni(struct sk_buff *skb)
3444 trace_netif_rx_ni_entry(skb);
3447 err = netif_rx_internal(skb);
3452 EXPORT_SYMBOL(netif_rx_ni);
3454 #ifdef CONFIG_PREEMPT_RT_FULL
3456 * RT runs ksoftirqd as a real time thread and the root_lock is a
3457 * "sleeping spinlock". If the trylock fails then we can go into an
3458 * infinite loop when ksoftirqd preempted the task which actually
3459 * holds the lock, because we requeue q and raise NET_TX softirq
3460 * causing ksoftirqd to loop forever.
3462 * It's safe to use spin_lock on RT here as softirqs run in thread
3463 * context and cannot deadlock against the thread which is holding
3466 * On !RT the trylock might fail, but there we bail out from the
3467 * softirq loop after 10 attempts which we can't do on RT. And the
3468 * task holding root_lock cannot be preempted, so the only downside of
3469 * that trylock is that we need 10 loops to decide that we should have
3470 * given up in the first one :)
3472 static inline int take_root_lock(spinlock_t *lock)
3478 static inline int take_root_lock(spinlock_t *lock)
3480 return spin_trylock(lock);
3484 static void net_tx_action(struct softirq_action *h)
3486 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3488 if (sd->completion_queue) {
3489 struct sk_buff *clist;
3491 local_irq_disable();
3492 clist = sd->completion_queue;
3493 sd->completion_queue = NULL;
3497 struct sk_buff *skb = clist;
3498 clist = clist->next;
3500 WARN_ON(atomic_read(&skb->users));
3501 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3502 trace_consume_skb(skb);
3504 trace_kfree_skb(skb, net_tx_action);
3509 if (sd->output_queue) {
3512 local_irq_disable();
3513 head = sd->output_queue;
3514 sd->output_queue = NULL;
3515 sd->output_queue_tailp = &sd->output_queue;
3519 struct Qdisc *q = head;
3520 spinlock_t *root_lock;
3522 head = head->next_sched;
3524 root_lock = qdisc_lock(q);
3525 if (take_root_lock(root_lock)) {
3526 smp_mb__before_atomic();
3527 clear_bit(__QDISC_STATE_SCHED,
3530 spin_unlock(root_lock);
3532 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3534 __netif_reschedule(q);
3536 smp_mb__before_atomic();
3537 clear_bit(__QDISC_STATE_SCHED,
3545 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3546 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3547 /* This hook is defined here for ATM LANE */
3548 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3549 unsigned char *addr) __read_mostly;
3550 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3553 #ifdef CONFIG_NET_CLS_ACT
3554 /* TODO: Maybe we should just force sch_ingress to be compiled in
3555 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3556 * a compare and 2 stores extra right now if we dont have it on
3557 * but have CONFIG_NET_CLS_ACT
3558 * NOTE: This doesn't stop any functionality; if you dont have
3559 * the ingress scheduler, you just can't add policies on ingress.
3562 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3564 struct net_device *dev = skb->dev;
3565 u32 ttl = G_TC_RTTL(skb->tc_verd);
3566 int result = TC_ACT_OK;
3569 if (unlikely(MAX_RED_LOOP < ttl++)) {
3570 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3571 skb->skb_iif, dev->ifindex);
3575 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3576 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3578 q = rcu_dereference(rxq->qdisc);
3579 if (q != &noop_qdisc) {
3580 spin_lock(qdisc_lock(q));
3581 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3582 result = qdisc_enqueue_root(skb, q);
3583 spin_unlock(qdisc_lock(q));
3589 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3590 struct packet_type **pt_prev,
3591 int *ret, struct net_device *orig_dev)
3593 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3595 if (!rxq || rcu_access_pointer(rxq->qdisc) == &noop_qdisc)
3599 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3603 switch (ing_filter(skb, rxq)) {
3615 * netdev_rx_handler_register - register receive handler
3616 * @dev: device to register a handler for
3617 * @rx_handler: receive handler to register
3618 * @rx_handler_data: data pointer that is used by rx handler
3620 * Register a receive handler for a device. This handler will then be
3621 * called from __netif_receive_skb. A negative errno code is returned
3624 * The caller must hold the rtnl_mutex.
3626 * For a general description of rx_handler, see enum rx_handler_result.
3628 int netdev_rx_handler_register(struct net_device *dev,
3629 rx_handler_func_t *rx_handler,
3630 void *rx_handler_data)
3634 if (dev->rx_handler)
3637 /* Note: rx_handler_data must be set before rx_handler */
3638 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3639 rcu_assign_pointer(dev->rx_handler, rx_handler);
3643 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3646 * netdev_rx_handler_unregister - unregister receive handler
3647 * @dev: device to unregister a handler from
3649 * Unregister a receive handler from a device.
3651 * The caller must hold the rtnl_mutex.
3653 void netdev_rx_handler_unregister(struct net_device *dev)
3657 RCU_INIT_POINTER(dev->rx_handler, NULL);
3658 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3659 * section has a guarantee to see a non NULL rx_handler_data
3663 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3665 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3668 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3669 * the special handling of PFMEMALLOC skbs.
3671 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3673 switch (skb->protocol) {
3674 case htons(ETH_P_ARP):
3675 case htons(ETH_P_IP):
3676 case htons(ETH_P_IPV6):
3677 case htons(ETH_P_8021Q):
3678 case htons(ETH_P_8021AD):
3685 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3687 struct packet_type *ptype, *pt_prev;
3688 rx_handler_func_t *rx_handler;
3689 struct net_device *orig_dev;
3690 bool deliver_exact = false;
3691 int ret = NET_RX_DROP;
3694 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3696 trace_netif_receive_skb(skb);
3698 orig_dev = skb->dev;
3700 skb_reset_network_header(skb);
3701 if (!skb_transport_header_was_set(skb))
3702 skb_reset_transport_header(skb);
3703 skb_reset_mac_len(skb);
3710 skb->skb_iif = skb->dev->ifindex;
3712 __this_cpu_inc(softnet_data.processed);
3714 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3715 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3716 skb = skb_vlan_untag(skb);
3721 #ifdef CONFIG_NET_CLS_ACT
3722 if (skb->tc_verd & TC_NCLS) {
3723 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3731 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3733 ret = deliver_skb(skb, pt_prev, orig_dev);
3737 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3739 ret = deliver_skb(skb, pt_prev, orig_dev);
3744 #ifdef CONFIG_NET_CLS_ACT
3745 if (static_key_false(&ingress_needed)) {
3746 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3754 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3757 if (skb_vlan_tag_present(skb)) {
3759 ret = deliver_skb(skb, pt_prev, orig_dev);
3762 if (vlan_do_receive(&skb))
3764 else if (unlikely(!skb))
3768 rx_handler = rcu_dereference(skb->dev->rx_handler);
3771 ret = deliver_skb(skb, pt_prev, orig_dev);
3774 switch (rx_handler(&skb)) {
3775 case RX_HANDLER_CONSUMED:
3776 ret = NET_RX_SUCCESS;
3778 case RX_HANDLER_ANOTHER:
3780 case RX_HANDLER_EXACT:
3781 deliver_exact = true;
3782 case RX_HANDLER_PASS:
3789 if (unlikely(skb_vlan_tag_present(skb))) {
3790 if (skb_vlan_tag_get_id(skb))
3791 skb->pkt_type = PACKET_OTHERHOST;
3792 /* Note: we might in the future use prio bits
3793 * and set skb->priority like in vlan_do_receive()
3794 * For the time being, just ignore Priority Code Point
3799 type = skb->protocol;
3801 /* deliver only exact match when indicated */
3802 if (likely(!deliver_exact)) {
3803 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3804 &ptype_base[ntohs(type) &
3808 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3809 &orig_dev->ptype_specific);
3811 if (unlikely(skb->dev != orig_dev)) {
3812 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3813 &skb->dev->ptype_specific);
3817 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3820 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3823 atomic_long_inc(&skb->dev->rx_dropped);
3825 /* Jamal, now you will not able to escape explaining
3826 * me how you were going to use this. :-)
3836 static int __netif_receive_skb(struct sk_buff *skb)
3840 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3841 unsigned long pflags = current->flags;
3844 * PFMEMALLOC skbs are special, they should
3845 * - be delivered to SOCK_MEMALLOC sockets only
3846 * - stay away from userspace
3847 * - have bounded memory usage
3849 * Use PF_MEMALLOC as this saves us from propagating the allocation
3850 * context down to all allocation sites.
3852 current->flags |= PF_MEMALLOC;
3853 ret = __netif_receive_skb_core(skb, true);
3854 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3856 ret = __netif_receive_skb_core(skb, false);
3861 static int netif_receive_skb_internal(struct sk_buff *skb)
3863 net_timestamp_check(netdev_tstamp_prequeue, skb);
3865 if (skb_defer_rx_timestamp(skb))
3866 return NET_RX_SUCCESS;
3869 if (static_key_false(&rps_needed)) {
3870 struct rps_dev_flow voidflow, *rflow = &voidflow;
3875 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3878 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3885 return __netif_receive_skb(skb);
3889 * netif_receive_skb - process receive buffer from network
3890 * @skb: buffer to process
3892 * netif_receive_skb() is the main receive data processing function.
3893 * It always succeeds. The buffer may be dropped during processing
3894 * for congestion control or by the protocol layers.
3896 * This function may only be called from softirq context and interrupts
3897 * should be enabled.
3899 * Return values (usually ignored):
3900 * NET_RX_SUCCESS: no congestion
3901 * NET_RX_DROP: packet was dropped
3903 int netif_receive_skb_sk(struct sock *sk, struct sk_buff *skb)
3905 trace_netif_receive_skb_entry(skb);
3907 return netif_receive_skb_internal(skb);
3909 EXPORT_SYMBOL(netif_receive_skb_sk);
3911 /* Network device is going away, flush any packets still pending
3912 * Called with irqs disabled.
3914 static void flush_backlog(void *arg)
3916 struct net_device *dev = arg;
3917 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3918 struct sk_buff *skb, *tmp;
3921 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3922 if (skb->dev == dev) {
3923 __skb_unlink(skb, &sd->input_pkt_queue);
3924 __skb_queue_tail(&sd->tofree_queue, skb);
3925 input_queue_head_incr(sd);
3930 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3931 if (skb->dev == dev) {
3932 __skb_unlink(skb, &sd->process_queue);
3933 __skb_queue_tail(&sd->tofree_queue, skb);
3934 input_queue_head_incr(sd);
3938 if (!skb_queue_empty(&sd->tofree_queue))
3939 raise_softirq_irqoff(NET_RX_SOFTIRQ);
3942 static int napi_gro_complete(struct sk_buff *skb)
3944 struct packet_offload *ptype;
3945 __be16 type = skb->protocol;
3946 struct list_head *head = &offload_base;
3949 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3951 if (NAPI_GRO_CB(skb)->count == 1) {
3952 skb_shinfo(skb)->gso_size = 0;
3957 list_for_each_entry_rcu(ptype, head, list) {
3958 if (ptype->type != type || !ptype->callbacks.gro_complete)
3961 err = ptype->callbacks.gro_complete(skb, 0);
3967 WARN_ON(&ptype->list == head);
3969 return NET_RX_SUCCESS;
3973 return netif_receive_skb_internal(skb);
3976 /* napi->gro_list contains packets ordered by age.
3977 * youngest packets at the head of it.
3978 * Complete skbs in reverse order to reduce latencies.
3980 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3982 struct sk_buff *skb, *prev = NULL;
3984 /* scan list and build reverse chain */
3985 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3990 for (skb = prev; skb; skb = prev) {
3993 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3997 napi_gro_complete(skb);
4001 napi->gro_list = NULL;
4003 EXPORT_SYMBOL(napi_gro_flush);
4005 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4008 unsigned int maclen = skb->dev->hard_header_len;
4009 u32 hash = skb_get_hash_raw(skb);
4011 for (p = napi->gro_list; p; p = p->next) {
4012 unsigned long diffs;
4014 NAPI_GRO_CB(p)->flush = 0;
4016 if (hash != skb_get_hash_raw(p)) {
4017 NAPI_GRO_CB(p)->same_flow = 0;
4021 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4022 diffs |= p->vlan_tci ^ skb->vlan_tci;
4023 if (maclen == ETH_HLEN)
4024 diffs |= compare_ether_header(skb_mac_header(p),
4025 skb_mac_header(skb));
4027 diffs = memcmp(skb_mac_header(p),
4028 skb_mac_header(skb),
4030 NAPI_GRO_CB(p)->same_flow = !diffs;
4034 static void skb_gro_reset_offset(struct sk_buff *skb)
4036 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4037 const skb_frag_t *frag0 = &pinfo->frags[0];
4039 NAPI_GRO_CB(skb)->data_offset = 0;
4040 NAPI_GRO_CB(skb)->frag0 = NULL;
4041 NAPI_GRO_CB(skb)->frag0_len = 0;
4043 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4045 !PageHighMem(skb_frag_page(frag0))) {
4046 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4047 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4051 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4053 struct skb_shared_info *pinfo = skb_shinfo(skb);
4055 BUG_ON(skb->end - skb->tail < grow);
4057 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4059 skb->data_len -= grow;
4062 pinfo->frags[0].page_offset += grow;
4063 skb_frag_size_sub(&pinfo->frags[0], grow);
4065 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4066 skb_frag_unref(skb, 0);
4067 memmove(pinfo->frags, pinfo->frags + 1,
4068 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4072 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4074 struct sk_buff **pp = NULL;
4075 struct packet_offload *ptype;
4076 __be16 type = skb->protocol;
4077 struct list_head *head = &offload_base;
4079 enum gro_result ret;
4082 if (!(skb->dev->features & NETIF_F_GRO))
4085 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4088 gro_list_prepare(napi, skb);
4091 list_for_each_entry_rcu(ptype, head, list) {
4092 if (ptype->type != type || !ptype->callbacks.gro_receive)
4095 skb_set_network_header(skb, skb_gro_offset(skb));
4096 skb_reset_mac_len(skb);
4097 NAPI_GRO_CB(skb)->same_flow = 0;
4098 NAPI_GRO_CB(skb)->flush = 0;
4099 NAPI_GRO_CB(skb)->free = 0;
4100 NAPI_GRO_CB(skb)->udp_mark = 0;
4101 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4103 /* Setup for GRO checksum validation */
4104 switch (skb->ip_summed) {
4105 case CHECKSUM_COMPLETE:
4106 NAPI_GRO_CB(skb)->csum = skb->csum;
4107 NAPI_GRO_CB(skb)->csum_valid = 1;
4108 NAPI_GRO_CB(skb)->csum_cnt = 0;
4110 case CHECKSUM_UNNECESSARY:
4111 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4112 NAPI_GRO_CB(skb)->csum_valid = 0;
4115 NAPI_GRO_CB(skb)->csum_cnt = 0;
4116 NAPI_GRO_CB(skb)->csum_valid = 0;
4119 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4124 if (&ptype->list == head)
4127 same_flow = NAPI_GRO_CB(skb)->same_flow;
4128 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4131 struct sk_buff *nskb = *pp;
4135 napi_gro_complete(nskb);
4142 if (NAPI_GRO_CB(skb)->flush)
4145 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4146 struct sk_buff *nskb = napi->gro_list;
4148 /* locate the end of the list to select the 'oldest' flow */
4149 while (nskb->next) {
4155 napi_gro_complete(nskb);
4159 NAPI_GRO_CB(skb)->count = 1;
4160 NAPI_GRO_CB(skb)->age = jiffies;
4161 NAPI_GRO_CB(skb)->last = skb;
4162 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4163 skb->next = napi->gro_list;
4164 napi->gro_list = skb;
4168 grow = skb_gro_offset(skb) - skb_headlen(skb);
4170 gro_pull_from_frag0(skb, grow);
4179 struct packet_offload *gro_find_receive_by_type(__be16 type)
4181 struct list_head *offload_head = &offload_base;
4182 struct packet_offload *ptype;
4184 list_for_each_entry_rcu(ptype, offload_head, list) {
4185 if (ptype->type != type || !ptype->callbacks.gro_receive)
4191 EXPORT_SYMBOL(gro_find_receive_by_type);
4193 struct packet_offload *gro_find_complete_by_type(__be16 type)
4195 struct list_head *offload_head = &offload_base;
4196 struct packet_offload *ptype;
4198 list_for_each_entry_rcu(ptype, offload_head, list) {
4199 if (ptype->type != type || !ptype->callbacks.gro_complete)
4205 EXPORT_SYMBOL(gro_find_complete_by_type);
4207 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4211 if (netif_receive_skb_internal(skb))
4219 case GRO_MERGED_FREE:
4220 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4221 kmem_cache_free(skbuff_head_cache, skb);
4234 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4236 trace_napi_gro_receive_entry(skb);
4238 skb_gro_reset_offset(skb);
4240 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4242 EXPORT_SYMBOL(napi_gro_receive);
4244 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4246 if (unlikely(skb->pfmemalloc)) {
4250 __skb_pull(skb, skb_headlen(skb));
4251 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4252 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4254 skb->dev = napi->dev;
4256 skb->encapsulation = 0;
4257 skb_shinfo(skb)->gso_type = 0;
4258 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4263 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4265 struct sk_buff *skb = napi->skb;
4268 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4273 EXPORT_SYMBOL(napi_get_frags);
4275 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4276 struct sk_buff *skb,
4282 __skb_push(skb, ETH_HLEN);
4283 skb->protocol = eth_type_trans(skb, skb->dev);
4284 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4289 case GRO_MERGED_FREE:
4290 napi_reuse_skb(napi, skb);
4300 /* Upper GRO stack assumes network header starts at gro_offset=0
4301 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4302 * We copy ethernet header into skb->data to have a common layout.
4304 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4306 struct sk_buff *skb = napi->skb;
4307 const struct ethhdr *eth;
4308 unsigned int hlen = sizeof(*eth);
4312 skb_reset_mac_header(skb);
4313 skb_gro_reset_offset(skb);
4315 eth = skb_gro_header_fast(skb, 0);
4316 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4317 eth = skb_gro_header_slow(skb, hlen, 0);
4318 if (unlikely(!eth)) {
4319 napi_reuse_skb(napi, skb);
4323 gro_pull_from_frag0(skb, hlen);
4324 NAPI_GRO_CB(skb)->frag0 += hlen;
4325 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4327 __skb_pull(skb, hlen);
4330 * This works because the only protocols we care about don't require
4332 * We'll fix it up properly in napi_frags_finish()
4334 skb->protocol = eth->h_proto;
4339 gro_result_t napi_gro_frags(struct napi_struct *napi)
4341 struct sk_buff *skb = napi_frags_skb(napi);
4346 trace_napi_gro_frags_entry(skb);
4348 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4350 EXPORT_SYMBOL(napi_gro_frags);
4352 /* Compute the checksum from gro_offset and return the folded value
4353 * after adding in any pseudo checksum.
4355 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4360 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4362 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4363 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4365 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4366 !skb->csum_complete_sw)
4367 netdev_rx_csum_fault(skb->dev);
4370 NAPI_GRO_CB(skb)->csum = wsum;
4371 NAPI_GRO_CB(skb)->csum_valid = 1;
4375 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4378 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4379 * Note: called with local irq disabled, but exits with local irq enabled.
4381 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4384 struct softnet_data *remsd = sd->rps_ipi_list;
4387 sd->rps_ipi_list = NULL;
4390 preempt_check_resched_rt();
4392 /* Send pending IPI's to kick RPS processing on remote cpus. */
4394 struct softnet_data *next = remsd->rps_ipi_next;
4396 if (cpu_online(remsd->cpu))
4397 smp_call_function_single_async(remsd->cpu,
4404 preempt_check_resched_rt();
4407 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4410 return sd->rps_ipi_list != NULL;
4416 static int process_backlog(struct napi_struct *napi, int quota)
4419 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4421 /* Check if we have pending ipi, its better to send them now,
4422 * not waiting net_rx_action() end.
4424 if (sd_has_rps_ipi_waiting(sd)) {
4425 local_irq_disable();
4426 net_rps_action_and_irq_enable(sd);
4429 napi->weight = weight_p;
4430 local_irq_disable();
4432 struct sk_buff *skb;
4434 while ((skb = __skb_dequeue(&sd->process_queue))) {
4436 __netif_receive_skb(skb);
4437 local_irq_disable();
4438 input_queue_head_incr(sd);
4439 if (++work >= quota) {
4446 if (skb_queue_empty(&sd->input_pkt_queue)) {
4448 * Inline a custom version of __napi_complete().
4449 * only current cpu owns and manipulates this napi,
4450 * and NAPI_STATE_SCHED is the only possible flag set
4452 * We can use a plain write instead of clear_bit(),
4453 * and we dont need an smp_mb() memory barrier.
4461 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4462 &sd->process_queue);
4471 * __napi_schedule - schedule for receive
4472 * @n: entry to schedule
4474 * The entry's receive function will be scheduled to run.
4475 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4477 void __napi_schedule(struct napi_struct *n)
4479 unsigned long flags;
4481 local_irq_save(flags);
4482 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4483 local_irq_restore(flags);
4484 preempt_check_resched_rt();
4486 EXPORT_SYMBOL(__napi_schedule);
4489 * __napi_schedule_irqoff - schedule for receive
4490 * @n: entry to schedule
4492 * Variant of __napi_schedule() assuming hard irqs are masked
4494 void __napi_schedule_irqoff(struct napi_struct *n)
4496 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4498 EXPORT_SYMBOL(__napi_schedule_irqoff);
4500 void __napi_complete(struct napi_struct *n)
4502 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4504 list_del_init(&n->poll_list);
4505 smp_mb__before_atomic();
4506 clear_bit(NAPI_STATE_SCHED, &n->state);
4508 EXPORT_SYMBOL(__napi_complete);
4510 void napi_complete_done(struct napi_struct *n, int work_done)
4512 unsigned long flags;
4515 * don't let napi dequeue from the cpu poll list
4516 * just in case its running on a different cpu
4518 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4522 unsigned long timeout = 0;
4525 timeout = n->dev->gro_flush_timeout;
4528 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4529 HRTIMER_MODE_REL_PINNED);
4531 napi_gro_flush(n, false);
4533 if (likely(list_empty(&n->poll_list))) {
4534 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4536 /* If n->poll_list is not empty, we need to mask irqs */
4537 local_irq_save(flags);
4539 local_irq_restore(flags);
4542 EXPORT_SYMBOL(napi_complete_done);
4544 /* must be called under rcu_read_lock(), as we dont take a reference */
4545 struct napi_struct *napi_by_id(unsigned int napi_id)
4547 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4548 struct napi_struct *napi;
4550 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4551 if (napi->napi_id == napi_id)
4556 EXPORT_SYMBOL_GPL(napi_by_id);
4558 void napi_hash_add(struct napi_struct *napi)
4560 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4562 spin_lock(&napi_hash_lock);
4564 /* 0 is not a valid id, we also skip an id that is taken
4565 * we expect both events to be extremely rare
4568 while (!napi->napi_id) {
4569 napi->napi_id = ++napi_gen_id;
4570 if (napi_by_id(napi->napi_id))
4574 hlist_add_head_rcu(&napi->napi_hash_node,
4575 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4577 spin_unlock(&napi_hash_lock);
4580 EXPORT_SYMBOL_GPL(napi_hash_add);
4582 /* Warning : caller is responsible to make sure rcu grace period
4583 * is respected before freeing memory containing @napi
4585 void napi_hash_del(struct napi_struct *napi)
4587 spin_lock(&napi_hash_lock);
4589 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4590 hlist_del_rcu(&napi->napi_hash_node);
4592 spin_unlock(&napi_hash_lock);
4594 EXPORT_SYMBOL_GPL(napi_hash_del);
4596 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4598 struct napi_struct *napi;
4600 napi = container_of(timer, struct napi_struct, timer);
4602 napi_schedule(napi);
4604 return HRTIMER_NORESTART;
4607 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4608 int (*poll)(struct napi_struct *, int), int weight)
4610 INIT_LIST_HEAD(&napi->poll_list);
4611 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4612 napi->timer.function = napi_watchdog;
4613 napi->gro_count = 0;
4614 napi->gro_list = NULL;
4617 if (weight > NAPI_POLL_WEIGHT)
4618 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4620 napi->weight = weight;
4621 list_add(&napi->dev_list, &dev->napi_list);
4623 #ifdef CONFIG_NETPOLL
4624 spin_lock_init(&napi->poll_lock);
4625 napi->poll_owner = -1;
4627 set_bit(NAPI_STATE_SCHED, &napi->state);
4629 EXPORT_SYMBOL(netif_napi_add);
4631 void napi_disable(struct napi_struct *n)
4634 set_bit(NAPI_STATE_DISABLE, &n->state);
4636 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4639 hrtimer_cancel(&n->timer);
4641 clear_bit(NAPI_STATE_DISABLE, &n->state);
4643 EXPORT_SYMBOL(napi_disable);
4645 void netif_napi_del(struct napi_struct *napi)
4647 list_del_init(&napi->dev_list);
4648 napi_free_frags(napi);
4650 kfree_skb_list(napi->gro_list);
4651 napi->gro_list = NULL;
4652 napi->gro_count = 0;
4654 EXPORT_SYMBOL(netif_napi_del);
4656 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4661 list_del_init(&n->poll_list);
4663 have = netpoll_poll_lock(n);
4667 /* This NAPI_STATE_SCHED test is for avoiding a race
4668 * with netpoll's poll_napi(). Only the entity which
4669 * obtains the lock and sees NAPI_STATE_SCHED set will
4670 * actually make the ->poll() call. Therefore we avoid
4671 * accidentally calling ->poll() when NAPI is not scheduled.
4674 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4675 work = n->poll(n, weight);
4679 WARN_ON_ONCE(work > weight);
4681 if (likely(work < weight))
4684 /* Drivers must not modify the NAPI state if they
4685 * consume the entire weight. In such cases this code
4686 * still "owns" the NAPI instance and therefore can
4687 * move the instance around on the list at-will.
4689 if (unlikely(napi_disable_pending(n))) {
4695 /* flush too old packets
4696 * If HZ < 1000, flush all packets.
4698 napi_gro_flush(n, HZ >= 1000);
4701 /* Some drivers may have called napi_schedule
4702 * prior to exhausting their budget.
4704 if (unlikely(!list_empty(&n->poll_list))) {
4705 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4706 n->dev ? n->dev->name : "backlog");
4710 list_add_tail(&n->poll_list, repoll);
4713 netpoll_poll_unlock(have);
4718 static void net_rx_action(struct softirq_action *h)
4720 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4721 unsigned long time_limit = jiffies + 2;
4722 int budget = netdev_budget;
4726 local_irq_disable();
4727 list_splice_init(&sd->poll_list, &list);
4731 struct napi_struct *n;
4733 if (list_empty(&list)) {
4734 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4739 n = list_first_entry(&list, struct napi_struct, poll_list);
4740 budget -= napi_poll(n, &repoll);
4742 /* If softirq window is exhausted then punt.
4743 * Allow this to run for 2 jiffies since which will allow
4744 * an average latency of 1.5/HZ.
4746 if (unlikely(budget <= 0 ||
4747 time_after_eq(jiffies, time_limit))) {
4753 local_irq_disable();
4755 list_splice_tail_init(&sd->poll_list, &list);
4756 list_splice_tail(&repoll, &list);
4757 list_splice(&list, &sd->poll_list);
4758 if (!list_empty(&sd->poll_list))
4759 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4761 net_rps_action_and_irq_enable(sd);
4764 struct netdev_adjacent {
4765 struct net_device *dev;
4767 /* upper master flag, there can only be one master device per list */
4770 /* counter for the number of times this device was added to us */
4773 /* private field for the users */
4776 struct list_head list;
4777 struct rcu_head rcu;
4780 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4781 struct net_device *adj_dev,
4782 struct list_head *adj_list)
4784 struct netdev_adjacent *adj;
4786 list_for_each_entry(adj, adj_list, list) {
4787 if (adj->dev == adj_dev)
4794 * netdev_has_upper_dev - Check if device is linked to an upper device
4796 * @upper_dev: upper device to check
4798 * Find out if a device is linked to specified upper device and return true
4799 * in case it is. Note that this checks only immediate upper device,
4800 * not through a complete stack of devices. The caller must hold the RTNL lock.
4802 bool netdev_has_upper_dev(struct net_device *dev,
4803 struct net_device *upper_dev)
4807 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4809 EXPORT_SYMBOL(netdev_has_upper_dev);
4812 * netdev_has_any_upper_dev - Check if device is linked to some device
4815 * Find out if a device is linked to an upper device and return true in case
4816 * it is. The caller must hold the RTNL lock.
4818 static bool netdev_has_any_upper_dev(struct net_device *dev)
4822 return !list_empty(&dev->all_adj_list.upper);
4826 * netdev_master_upper_dev_get - Get master upper device
4829 * Find a master upper device and return pointer to it or NULL in case
4830 * it's not there. The caller must hold the RTNL lock.
4832 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4834 struct netdev_adjacent *upper;
4838 if (list_empty(&dev->adj_list.upper))
4841 upper = list_first_entry(&dev->adj_list.upper,
4842 struct netdev_adjacent, list);
4843 if (likely(upper->master))
4847 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4849 void *netdev_adjacent_get_private(struct list_head *adj_list)
4851 struct netdev_adjacent *adj;
4853 adj = list_entry(adj_list, struct netdev_adjacent, list);
4855 return adj->private;
4857 EXPORT_SYMBOL(netdev_adjacent_get_private);
4860 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4862 * @iter: list_head ** of the current position
4864 * Gets the next device from the dev's upper list, starting from iter
4865 * position. The caller must hold RCU read lock.
4867 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4868 struct list_head **iter)
4870 struct netdev_adjacent *upper;
4872 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4874 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4876 if (&upper->list == &dev->adj_list.upper)
4879 *iter = &upper->list;
4883 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4886 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4888 * @iter: list_head ** of the current position
4890 * Gets the next device from the dev's upper list, starting from iter
4891 * position. The caller must hold RCU read lock.
4893 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4894 struct list_head **iter)
4896 struct netdev_adjacent *upper;
4898 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4900 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4902 if (&upper->list == &dev->all_adj_list.upper)
4905 *iter = &upper->list;
4909 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4912 * netdev_lower_get_next_private - Get the next ->private from the
4913 * lower neighbour list
4915 * @iter: list_head ** of the current position
4917 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4918 * list, starting from iter position. The caller must hold either hold the
4919 * RTNL lock or its own locking that guarantees that the neighbour lower
4920 * list will remain unchainged.
4922 void *netdev_lower_get_next_private(struct net_device *dev,
4923 struct list_head **iter)
4925 struct netdev_adjacent *lower;
4927 lower = list_entry(*iter, struct netdev_adjacent, list);
4929 if (&lower->list == &dev->adj_list.lower)
4932 *iter = lower->list.next;
4934 return lower->private;
4936 EXPORT_SYMBOL(netdev_lower_get_next_private);
4939 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4940 * lower neighbour list, RCU
4943 * @iter: list_head ** of the current position
4945 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4946 * list, starting from iter position. The caller must hold RCU read lock.
4948 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4949 struct list_head **iter)
4951 struct netdev_adjacent *lower;
4953 WARN_ON_ONCE(!rcu_read_lock_held());
4955 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4957 if (&lower->list == &dev->adj_list.lower)
4960 *iter = &lower->list;
4962 return lower->private;
4964 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4967 * netdev_lower_get_next - Get the next device from the lower neighbour
4970 * @iter: list_head ** of the current position
4972 * Gets the next netdev_adjacent from the dev's lower neighbour
4973 * list, starting from iter position. The caller must hold RTNL lock or
4974 * its own locking that guarantees that the neighbour lower
4975 * list will remain unchainged.
4977 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4979 struct netdev_adjacent *lower;
4981 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4983 if (&lower->list == &dev->adj_list.lower)
4986 *iter = &lower->list;
4990 EXPORT_SYMBOL(netdev_lower_get_next);
4993 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4994 * lower neighbour list, RCU
4998 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4999 * list. The caller must hold RCU read lock.
5001 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5003 struct netdev_adjacent *lower;
5005 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5006 struct netdev_adjacent, list);
5008 return lower->private;
5011 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5014 * netdev_master_upper_dev_get_rcu - Get master upper device
5017 * Find a master upper device and return pointer to it or NULL in case
5018 * it's not there. The caller must hold the RCU read lock.
5020 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5022 struct netdev_adjacent *upper;
5024 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5025 struct netdev_adjacent, list);
5026 if (upper && likely(upper->master))
5030 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5032 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5033 struct net_device *adj_dev,
5034 struct list_head *dev_list)
5036 char linkname[IFNAMSIZ+7];
5037 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5038 "upper_%s" : "lower_%s", adj_dev->name);
5039 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5042 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5044 struct list_head *dev_list)
5046 char linkname[IFNAMSIZ+7];
5047 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5048 "upper_%s" : "lower_%s", name);
5049 sysfs_remove_link(&(dev->dev.kobj), linkname);
5052 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5053 struct net_device *adj_dev,
5054 struct list_head *dev_list)
5056 return (dev_list == &dev->adj_list.upper ||
5057 dev_list == &dev->adj_list.lower) &&
5058 net_eq(dev_net(dev), dev_net(adj_dev));
5061 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5062 struct net_device *adj_dev,
5063 struct list_head *dev_list,
5064 void *private, bool master)
5066 struct netdev_adjacent *adj;
5069 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5076 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5081 adj->master = master;
5083 adj->private = private;
5086 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5087 adj_dev->name, dev->name, adj_dev->name);
5089 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5090 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5095 /* Ensure that master link is always the first item in list. */
5097 ret = sysfs_create_link(&(dev->dev.kobj),
5098 &(adj_dev->dev.kobj), "master");
5100 goto remove_symlinks;
5102 list_add_rcu(&adj->list, dev_list);
5104 list_add_tail_rcu(&adj->list, dev_list);
5110 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5111 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5119 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5120 struct net_device *adj_dev,
5121 struct list_head *dev_list)
5123 struct netdev_adjacent *adj;
5125 adj = __netdev_find_adj(dev, adj_dev, dev_list);
5128 pr_err("tried to remove device %s from %s\n",
5129 dev->name, adj_dev->name);
5133 if (adj->ref_nr > 1) {
5134 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5141 sysfs_remove_link(&(dev->dev.kobj), "master");
5143 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5144 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5146 list_del_rcu(&adj->list);
5147 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5148 adj_dev->name, dev->name, adj_dev->name);
5150 kfree_rcu(adj, rcu);
5153 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5154 struct net_device *upper_dev,
5155 struct list_head *up_list,
5156 struct list_head *down_list,
5157 void *private, bool master)
5161 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5166 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5169 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5176 static int __netdev_adjacent_dev_link(struct net_device *dev,
5177 struct net_device *upper_dev)
5179 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5180 &dev->all_adj_list.upper,
5181 &upper_dev->all_adj_list.lower,
5185 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5186 struct net_device *upper_dev,
5187 struct list_head *up_list,
5188 struct list_head *down_list)
5190 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5191 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5194 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5195 struct net_device *upper_dev)
5197 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5198 &dev->all_adj_list.upper,
5199 &upper_dev->all_adj_list.lower);
5202 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5203 struct net_device *upper_dev,
5204 void *private, bool master)
5206 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5211 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5212 &dev->adj_list.upper,
5213 &upper_dev->adj_list.lower,
5216 __netdev_adjacent_dev_unlink(dev, upper_dev);
5223 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5224 struct net_device *upper_dev)
5226 __netdev_adjacent_dev_unlink(dev, upper_dev);
5227 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5228 &dev->adj_list.upper,
5229 &upper_dev->adj_list.lower);
5232 static int __netdev_upper_dev_link(struct net_device *dev,
5233 struct net_device *upper_dev, bool master,
5236 struct netdev_adjacent *i, *j, *to_i, *to_j;
5241 if (dev == upper_dev)
5244 /* To prevent loops, check if dev is not upper device to upper_dev. */
5245 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5248 if (__netdev_find_adj(dev, upper_dev, &dev->adj_list.upper))
5251 if (master && netdev_master_upper_dev_get(dev))
5254 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5259 /* Now that we linked these devs, make all the upper_dev's
5260 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5261 * versa, and don't forget the devices itself. All of these
5262 * links are non-neighbours.
5264 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5265 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5266 pr_debug("Interlinking %s with %s, non-neighbour\n",
5267 i->dev->name, j->dev->name);
5268 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5274 /* add dev to every upper_dev's upper device */
5275 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5276 pr_debug("linking %s's upper device %s with %s\n",
5277 upper_dev->name, i->dev->name, dev->name);
5278 ret = __netdev_adjacent_dev_link(dev, i->dev);
5280 goto rollback_upper_mesh;
5283 /* add upper_dev to every dev's lower device */
5284 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5285 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5286 i->dev->name, upper_dev->name);
5287 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5289 goto rollback_lower_mesh;
5292 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5295 rollback_lower_mesh:
5297 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5300 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5305 rollback_upper_mesh:
5307 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5310 __netdev_adjacent_dev_unlink(dev, i->dev);
5318 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5319 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5320 if (i == to_i && j == to_j)
5322 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5328 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5334 * netdev_upper_dev_link - Add a link to the upper device
5336 * @upper_dev: new upper device
5338 * Adds a link to device which is upper to this one. The caller must hold
5339 * the RTNL lock. On a failure a negative errno code is returned.
5340 * On success the reference counts are adjusted and the function
5343 int netdev_upper_dev_link(struct net_device *dev,
5344 struct net_device *upper_dev)
5346 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5348 EXPORT_SYMBOL(netdev_upper_dev_link);
5351 * netdev_master_upper_dev_link - Add a master link to the upper device
5353 * @upper_dev: new upper device
5355 * Adds a link to device which is upper to this one. In this case, only
5356 * one master upper device can be linked, although other non-master devices
5357 * might be linked as well. The caller must hold the RTNL lock.
5358 * On a failure a negative errno code is returned. On success the reference
5359 * counts are adjusted and the function returns zero.
5361 int netdev_master_upper_dev_link(struct net_device *dev,
5362 struct net_device *upper_dev)
5364 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5366 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5368 int netdev_master_upper_dev_link_private(struct net_device *dev,
5369 struct net_device *upper_dev,
5372 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5374 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5377 * netdev_upper_dev_unlink - Removes a link to upper device
5379 * @upper_dev: new upper device
5381 * Removes a link to device which is upper to this one. The caller must hold
5384 void netdev_upper_dev_unlink(struct net_device *dev,
5385 struct net_device *upper_dev)
5387 struct netdev_adjacent *i, *j;
5390 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5392 /* Here is the tricky part. We must remove all dev's lower
5393 * devices from all upper_dev's upper devices and vice
5394 * versa, to maintain the graph relationship.
5396 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5397 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5398 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5400 /* remove also the devices itself from lower/upper device
5403 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5404 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5406 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5407 __netdev_adjacent_dev_unlink(dev, i->dev);
5409 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5411 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5414 * netdev_bonding_info_change - Dispatch event about slave change
5416 * @bonding_info: info to dispatch
5418 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5419 * The caller must hold the RTNL lock.
5421 void netdev_bonding_info_change(struct net_device *dev,
5422 struct netdev_bonding_info *bonding_info)
5424 struct netdev_notifier_bonding_info info;
5426 memcpy(&info.bonding_info, bonding_info,
5427 sizeof(struct netdev_bonding_info));
5428 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5431 EXPORT_SYMBOL(netdev_bonding_info_change);
5433 static void netdev_adjacent_add_links(struct net_device *dev)
5435 struct netdev_adjacent *iter;
5437 struct net *net = dev_net(dev);
5439 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5440 if (!net_eq(net,dev_net(iter->dev)))
5442 netdev_adjacent_sysfs_add(iter->dev, dev,
5443 &iter->dev->adj_list.lower);
5444 netdev_adjacent_sysfs_add(dev, iter->dev,
5445 &dev->adj_list.upper);
5448 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5449 if (!net_eq(net,dev_net(iter->dev)))
5451 netdev_adjacent_sysfs_add(iter->dev, dev,
5452 &iter->dev->adj_list.upper);
5453 netdev_adjacent_sysfs_add(dev, iter->dev,
5454 &dev->adj_list.lower);
5458 static void netdev_adjacent_del_links(struct net_device *dev)
5460 struct netdev_adjacent *iter;
5462 struct net *net = dev_net(dev);
5464 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5465 if (!net_eq(net,dev_net(iter->dev)))
5467 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5468 &iter->dev->adj_list.lower);
5469 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5470 &dev->adj_list.upper);
5473 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5474 if (!net_eq(net,dev_net(iter->dev)))
5476 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5477 &iter->dev->adj_list.upper);
5478 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5479 &dev->adj_list.lower);
5483 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5485 struct netdev_adjacent *iter;
5487 struct net *net = dev_net(dev);
5489 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5490 if (!net_eq(net,dev_net(iter->dev)))
5492 netdev_adjacent_sysfs_del(iter->dev, oldname,
5493 &iter->dev->adj_list.lower);
5494 netdev_adjacent_sysfs_add(iter->dev, dev,
5495 &iter->dev->adj_list.lower);
5498 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5499 if (!net_eq(net,dev_net(iter->dev)))
5501 netdev_adjacent_sysfs_del(iter->dev, oldname,
5502 &iter->dev->adj_list.upper);
5503 netdev_adjacent_sysfs_add(iter->dev, dev,
5504 &iter->dev->adj_list.upper);
5508 void *netdev_lower_dev_get_private(struct net_device *dev,
5509 struct net_device *lower_dev)
5511 struct netdev_adjacent *lower;
5515 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5519 return lower->private;
5521 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5524 int dev_get_nest_level(struct net_device *dev,
5525 bool (*type_check)(struct net_device *dev))
5527 struct net_device *lower = NULL;
5528 struct list_head *iter;
5534 netdev_for_each_lower_dev(dev, lower, iter) {
5535 nest = dev_get_nest_level(lower, type_check);
5536 if (max_nest < nest)
5540 if (type_check(dev))
5545 EXPORT_SYMBOL(dev_get_nest_level);
5547 static void dev_change_rx_flags(struct net_device *dev, int flags)
5549 const struct net_device_ops *ops = dev->netdev_ops;
5551 if (ops->ndo_change_rx_flags)
5552 ops->ndo_change_rx_flags(dev, flags);
5555 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5557 unsigned int old_flags = dev->flags;
5563 dev->flags |= IFF_PROMISC;
5564 dev->promiscuity += inc;
5565 if (dev->promiscuity == 0) {
5568 * If inc causes overflow, untouch promisc and return error.
5571 dev->flags &= ~IFF_PROMISC;
5573 dev->promiscuity -= inc;
5574 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5579 if (dev->flags != old_flags) {
5580 pr_info("device %s %s promiscuous mode\n",
5582 dev->flags & IFF_PROMISC ? "entered" : "left");
5583 if (audit_enabled) {
5584 current_uid_gid(&uid, &gid);
5585 audit_log(current->audit_context, GFP_ATOMIC,
5586 AUDIT_ANOM_PROMISCUOUS,
5587 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5588 dev->name, (dev->flags & IFF_PROMISC),
5589 (old_flags & IFF_PROMISC),
5590 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5591 from_kuid(&init_user_ns, uid),
5592 from_kgid(&init_user_ns, gid),
5593 audit_get_sessionid(current));
5596 dev_change_rx_flags(dev, IFF_PROMISC);
5599 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5604 * dev_set_promiscuity - update promiscuity count on a device
5608 * Add or remove promiscuity from a device. While the count in the device
5609 * remains above zero the interface remains promiscuous. Once it hits zero
5610 * the device reverts back to normal filtering operation. A negative inc
5611 * value is used to drop promiscuity on the device.
5612 * Return 0 if successful or a negative errno code on error.
5614 int dev_set_promiscuity(struct net_device *dev, int inc)
5616 unsigned int old_flags = dev->flags;
5619 err = __dev_set_promiscuity(dev, inc, true);
5622 if (dev->flags != old_flags)
5623 dev_set_rx_mode(dev);
5626 EXPORT_SYMBOL(dev_set_promiscuity);
5628 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5630 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5634 dev->flags |= IFF_ALLMULTI;
5635 dev->allmulti += inc;
5636 if (dev->allmulti == 0) {
5639 * If inc causes overflow, untouch allmulti and return error.
5642 dev->flags &= ~IFF_ALLMULTI;
5644 dev->allmulti -= inc;
5645 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5650 if (dev->flags ^ old_flags) {
5651 dev_change_rx_flags(dev, IFF_ALLMULTI);
5652 dev_set_rx_mode(dev);
5654 __dev_notify_flags(dev, old_flags,
5655 dev->gflags ^ old_gflags);
5661 * dev_set_allmulti - update allmulti count on a device
5665 * Add or remove reception of all multicast frames to a device. While the
5666 * count in the device remains above zero the interface remains listening
5667 * to all interfaces. Once it hits zero the device reverts back to normal
5668 * filtering operation. A negative @inc value is used to drop the counter
5669 * when releasing a resource needing all multicasts.
5670 * Return 0 if successful or a negative errno code on error.
5673 int dev_set_allmulti(struct net_device *dev, int inc)
5675 return __dev_set_allmulti(dev, inc, true);
5677 EXPORT_SYMBOL(dev_set_allmulti);
5680 * Upload unicast and multicast address lists to device and
5681 * configure RX filtering. When the device doesn't support unicast
5682 * filtering it is put in promiscuous mode while unicast addresses
5685 void __dev_set_rx_mode(struct net_device *dev)
5687 const struct net_device_ops *ops = dev->netdev_ops;
5689 /* dev_open will call this function so the list will stay sane. */
5690 if (!(dev->flags&IFF_UP))
5693 if (!netif_device_present(dev))
5696 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5697 /* Unicast addresses changes may only happen under the rtnl,
5698 * therefore calling __dev_set_promiscuity here is safe.
5700 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5701 __dev_set_promiscuity(dev, 1, false);
5702 dev->uc_promisc = true;
5703 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5704 __dev_set_promiscuity(dev, -1, false);
5705 dev->uc_promisc = false;
5709 if (ops->ndo_set_rx_mode)
5710 ops->ndo_set_rx_mode(dev);
5713 void dev_set_rx_mode(struct net_device *dev)
5715 netif_addr_lock_bh(dev);
5716 __dev_set_rx_mode(dev);
5717 netif_addr_unlock_bh(dev);
5721 * dev_get_flags - get flags reported to userspace
5724 * Get the combination of flag bits exported through APIs to userspace.
5726 unsigned int dev_get_flags(const struct net_device *dev)
5730 flags = (dev->flags & ~(IFF_PROMISC |
5735 (dev->gflags & (IFF_PROMISC |
5738 if (netif_running(dev)) {
5739 if (netif_oper_up(dev))
5740 flags |= IFF_RUNNING;
5741 if (netif_carrier_ok(dev))
5742 flags |= IFF_LOWER_UP;
5743 if (netif_dormant(dev))
5744 flags |= IFF_DORMANT;
5749 EXPORT_SYMBOL(dev_get_flags);
5751 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5753 unsigned int old_flags = dev->flags;
5759 * Set the flags on our device.
5762 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5763 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5765 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5769 * Load in the correct multicast list now the flags have changed.
5772 if ((old_flags ^ flags) & IFF_MULTICAST)
5773 dev_change_rx_flags(dev, IFF_MULTICAST);
5775 dev_set_rx_mode(dev);
5778 * Have we downed the interface. We handle IFF_UP ourselves
5779 * according to user attempts to set it, rather than blindly
5784 if ((old_flags ^ flags) & IFF_UP)
5785 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5787 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5788 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5789 unsigned int old_flags = dev->flags;
5791 dev->gflags ^= IFF_PROMISC;
5793 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5794 if (dev->flags != old_flags)
5795 dev_set_rx_mode(dev);
5798 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5799 is important. Some (broken) drivers set IFF_PROMISC, when
5800 IFF_ALLMULTI is requested not asking us and not reporting.
5802 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5803 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5805 dev->gflags ^= IFF_ALLMULTI;
5806 __dev_set_allmulti(dev, inc, false);
5812 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5813 unsigned int gchanges)
5815 unsigned int changes = dev->flags ^ old_flags;
5818 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5820 if (changes & IFF_UP) {
5821 if (dev->flags & IFF_UP)
5822 call_netdevice_notifiers(NETDEV_UP, dev);
5824 call_netdevice_notifiers(NETDEV_DOWN, dev);
5827 if (dev->flags & IFF_UP &&
5828 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5829 struct netdev_notifier_change_info change_info;
5831 change_info.flags_changed = changes;
5832 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5838 * dev_change_flags - change device settings
5840 * @flags: device state flags
5842 * Change settings on device based state flags. The flags are
5843 * in the userspace exported format.
5845 int dev_change_flags(struct net_device *dev, unsigned int flags)
5848 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5850 ret = __dev_change_flags(dev, flags);
5854 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5855 __dev_notify_flags(dev, old_flags, changes);
5858 EXPORT_SYMBOL(dev_change_flags);
5860 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5862 const struct net_device_ops *ops = dev->netdev_ops;
5864 if (ops->ndo_change_mtu)
5865 return ops->ndo_change_mtu(dev, new_mtu);
5872 * dev_set_mtu - Change maximum transfer unit
5874 * @new_mtu: new transfer unit
5876 * Change the maximum transfer size of the network device.
5878 int dev_set_mtu(struct net_device *dev, int new_mtu)
5882 if (new_mtu == dev->mtu)
5885 /* MTU must be positive. */
5889 if (!netif_device_present(dev))
5892 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5893 err = notifier_to_errno(err);
5897 orig_mtu = dev->mtu;
5898 err = __dev_set_mtu(dev, new_mtu);
5901 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5902 err = notifier_to_errno(err);
5904 /* setting mtu back and notifying everyone again,
5905 * so that they have a chance to revert changes.
5907 __dev_set_mtu(dev, orig_mtu);
5908 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5913 EXPORT_SYMBOL(dev_set_mtu);
5916 * dev_set_group - Change group this device belongs to
5918 * @new_group: group this device should belong to
5920 void dev_set_group(struct net_device *dev, int new_group)
5922 dev->group = new_group;
5924 EXPORT_SYMBOL(dev_set_group);
5927 * dev_set_mac_address - Change Media Access Control Address
5931 * Change the hardware (MAC) address of the device
5933 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5935 const struct net_device_ops *ops = dev->netdev_ops;
5938 if (!ops->ndo_set_mac_address)
5940 if (sa->sa_family != dev->type)
5942 if (!netif_device_present(dev))
5944 err = ops->ndo_set_mac_address(dev, sa);
5947 dev->addr_assign_type = NET_ADDR_SET;
5948 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5949 add_device_randomness(dev->dev_addr, dev->addr_len);
5952 EXPORT_SYMBOL(dev_set_mac_address);
5955 * dev_change_carrier - Change device carrier
5957 * @new_carrier: new value
5959 * Change device carrier
5961 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5963 const struct net_device_ops *ops = dev->netdev_ops;
5965 if (!ops->ndo_change_carrier)
5967 if (!netif_device_present(dev))
5969 return ops->ndo_change_carrier(dev, new_carrier);
5971 EXPORT_SYMBOL(dev_change_carrier);
5974 * dev_get_phys_port_id - Get device physical port ID
5978 * Get device physical port ID
5980 int dev_get_phys_port_id(struct net_device *dev,
5981 struct netdev_phys_item_id *ppid)
5983 const struct net_device_ops *ops = dev->netdev_ops;
5985 if (!ops->ndo_get_phys_port_id)
5987 return ops->ndo_get_phys_port_id(dev, ppid);
5989 EXPORT_SYMBOL(dev_get_phys_port_id);
5992 * dev_get_phys_port_name - Get device physical port name
5996 * Get device physical port name
5998 int dev_get_phys_port_name(struct net_device *dev,
5999 char *name, size_t len)
6001 const struct net_device_ops *ops = dev->netdev_ops;
6003 if (!ops->ndo_get_phys_port_name)
6005 return ops->ndo_get_phys_port_name(dev, name, len);
6007 EXPORT_SYMBOL(dev_get_phys_port_name);
6010 * dev_new_index - allocate an ifindex
6011 * @net: the applicable net namespace
6013 * Returns a suitable unique value for a new device interface
6014 * number. The caller must hold the rtnl semaphore or the
6015 * dev_base_lock to be sure it remains unique.
6017 static int dev_new_index(struct net *net)
6019 int ifindex = net->ifindex;
6023 if (!__dev_get_by_index(net, ifindex))
6024 return net->ifindex = ifindex;
6028 /* Delayed registration/unregisteration */
6029 static LIST_HEAD(net_todo_list);
6030 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6032 static void net_set_todo(struct net_device *dev)
6034 list_add_tail(&dev->todo_list, &net_todo_list);
6035 dev_net(dev)->dev_unreg_count++;
6038 static void rollback_registered_many(struct list_head *head)
6040 struct net_device *dev, *tmp;
6041 LIST_HEAD(close_head);
6043 BUG_ON(dev_boot_phase);
6046 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6047 /* Some devices call without registering
6048 * for initialization unwind. Remove those
6049 * devices and proceed with the remaining.
6051 if (dev->reg_state == NETREG_UNINITIALIZED) {
6052 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6056 list_del(&dev->unreg_list);
6059 dev->dismantle = true;
6060 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6063 /* If device is running, close it first. */
6064 list_for_each_entry(dev, head, unreg_list)
6065 list_add_tail(&dev->close_list, &close_head);
6066 dev_close_many(&close_head, true);
6068 list_for_each_entry(dev, head, unreg_list) {
6069 /* And unlink it from device chain. */
6070 unlist_netdevice(dev);
6072 dev->reg_state = NETREG_UNREGISTERING;
6077 list_for_each_entry(dev, head, unreg_list) {
6078 struct sk_buff *skb = NULL;
6080 /* Shutdown queueing discipline. */
6084 /* Notify protocols, that we are about to destroy
6085 this device. They should clean all the things.
6087 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6089 if (!dev->rtnl_link_ops ||
6090 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6091 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6095 * Flush the unicast and multicast chains
6100 if (dev->netdev_ops->ndo_uninit)
6101 dev->netdev_ops->ndo_uninit(dev);
6104 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6106 /* Notifier chain MUST detach us all upper devices. */
6107 WARN_ON(netdev_has_any_upper_dev(dev));
6109 /* Remove entries from kobject tree */
6110 netdev_unregister_kobject(dev);
6112 /* Remove XPS queueing entries */
6113 netif_reset_xps_queues_gt(dev, 0);
6119 list_for_each_entry(dev, head, unreg_list)
6123 static void rollback_registered(struct net_device *dev)
6127 list_add(&dev->unreg_list, &single);
6128 rollback_registered_many(&single);
6132 static netdev_features_t netdev_fix_features(struct net_device *dev,
6133 netdev_features_t features)
6135 /* Fix illegal checksum combinations */
6136 if ((features & NETIF_F_HW_CSUM) &&
6137 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6138 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6139 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6142 /* TSO requires that SG is present as well. */
6143 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6144 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6145 features &= ~NETIF_F_ALL_TSO;
6148 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6149 !(features & NETIF_F_IP_CSUM)) {
6150 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6151 features &= ~NETIF_F_TSO;
6152 features &= ~NETIF_F_TSO_ECN;
6155 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6156 !(features & NETIF_F_IPV6_CSUM)) {
6157 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6158 features &= ~NETIF_F_TSO6;
6161 /* TSO ECN requires that TSO is present as well. */
6162 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6163 features &= ~NETIF_F_TSO_ECN;
6165 /* Software GSO depends on SG. */
6166 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6167 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6168 features &= ~NETIF_F_GSO;
6171 /* UFO needs SG and checksumming */
6172 if (features & NETIF_F_UFO) {
6173 /* maybe split UFO into V4 and V6? */
6174 if (!((features & NETIF_F_GEN_CSUM) ||
6175 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6176 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6178 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6179 features &= ~NETIF_F_UFO;
6182 if (!(features & NETIF_F_SG)) {
6184 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6185 features &= ~NETIF_F_UFO;
6189 #ifdef CONFIG_NET_RX_BUSY_POLL
6190 if (dev->netdev_ops->ndo_busy_poll)
6191 features |= NETIF_F_BUSY_POLL;
6194 features &= ~NETIF_F_BUSY_POLL;
6199 int __netdev_update_features(struct net_device *dev)
6201 netdev_features_t features;
6206 features = netdev_get_wanted_features(dev);
6208 if (dev->netdev_ops->ndo_fix_features)
6209 features = dev->netdev_ops->ndo_fix_features(dev, features);
6211 /* driver might be less strict about feature dependencies */
6212 features = netdev_fix_features(dev, features);
6214 if (dev->features == features)
6217 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6218 &dev->features, &features);
6220 if (dev->netdev_ops->ndo_set_features)
6221 err = dev->netdev_ops->ndo_set_features(dev, features);
6223 if (unlikely(err < 0)) {
6225 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6226 err, &features, &dev->features);
6231 dev->features = features;
6237 * netdev_update_features - recalculate device features
6238 * @dev: the device to check
6240 * Recalculate dev->features set and send notifications if it
6241 * has changed. Should be called after driver or hardware dependent
6242 * conditions might have changed that influence the features.
6244 void netdev_update_features(struct net_device *dev)
6246 if (__netdev_update_features(dev))
6247 netdev_features_change(dev);
6249 EXPORT_SYMBOL(netdev_update_features);
6252 * netdev_change_features - recalculate device features
6253 * @dev: the device to check
6255 * Recalculate dev->features set and send notifications even
6256 * if they have not changed. Should be called instead of
6257 * netdev_update_features() if also dev->vlan_features might
6258 * have changed to allow the changes to be propagated to stacked
6261 void netdev_change_features(struct net_device *dev)
6263 __netdev_update_features(dev);
6264 netdev_features_change(dev);
6266 EXPORT_SYMBOL(netdev_change_features);
6269 * netif_stacked_transfer_operstate - transfer operstate
6270 * @rootdev: the root or lower level device to transfer state from
6271 * @dev: the device to transfer operstate to
6273 * Transfer operational state from root to device. This is normally
6274 * called when a stacking relationship exists between the root
6275 * device and the device(a leaf device).
6277 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6278 struct net_device *dev)
6280 if (rootdev->operstate == IF_OPER_DORMANT)
6281 netif_dormant_on(dev);
6283 netif_dormant_off(dev);
6285 if (netif_carrier_ok(rootdev)) {
6286 if (!netif_carrier_ok(dev))
6287 netif_carrier_on(dev);
6289 if (netif_carrier_ok(dev))
6290 netif_carrier_off(dev);
6293 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6296 static int netif_alloc_rx_queues(struct net_device *dev)
6298 unsigned int i, count = dev->num_rx_queues;
6299 struct netdev_rx_queue *rx;
6300 size_t sz = count * sizeof(*rx);
6304 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6312 for (i = 0; i < count; i++)
6318 static void netdev_init_one_queue(struct net_device *dev,
6319 struct netdev_queue *queue, void *_unused)
6321 /* Initialize queue lock */
6322 spin_lock_init(&queue->_xmit_lock);
6323 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6324 queue->xmit_lock_owner = -1;
6325 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6328 dql_init(&queue->dql, HZ);
6332 static void netif_free_tx_queues(struct net_device *dev)
6337 static int netif_alloc_netdev_queues(struct net_device *dev)
6339 unsigned int count = dev->num_tx_queues;
6340 struct netdev_queue *tx;
6341 size_t sz = count * sizeof(*tx);
6343 BUG_ON(count < 1 || count > 0xffff);
6345 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6353 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6354 spin_lock_init(&dev->tx_global_lock);
6360 * register_netdevice - register a network device
6361 * @dev: device to register
6363 * Take a completed network device structure and add it to the kernel
6364 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6365 * chain. 0 is returned on success. A negative errno code is returned
6366 * on a failure to set up the device, or if the name is a duplicate.
6368 * Callers must hold the rtnl semaphore. You may want
6369 * register_netdev() instead of this.
6372 * The locking appears insufficient to guarantee two parallel registers
6373 * will not get the same name.
6376 int register_netdevice(struct net_device *dev)
6379 struct net *net = dev_net(dev);
6381 BUG_ON(dev_boot_phase);
6386 /* When net_device's are persistent, this will be fatal. */
6387 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6390 spin_lock_init(&dev->addr_list_lock);
6391 netdev_set_addr_lockdep_class(dev);
6393 ret = dev_get_valid_name(net, dev, dev->name);
6397 /* Init, if this function is available */
6398 if (dev->netdev_ops->ndo_init) {
6399 ret = dev->netdev_ops->ndo_init(dev);
6407 if (((dev->hw_features | dev->features) &
6408 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6409 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6410 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6411 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6418 dev->ifindex = dev_new_index(net);
6419 else if (__dev_get_by_index(net, dev->ifindex))
6422 /* Transfer changeable features to wanted_features and enable
6423 * software offloads (GSO and GRO).
6425 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6426 dev->features |= NETIF_F_SOFT_FEATURES;
6427 dev->wanted_features = dev->features & dev->hw_features;
6429 if (!(dev->flags & IFF_LOOPBACK)) {
6430 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6433 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6435 dev->vlan_features |= NETIF_F_HIGHDMA;
6437 /* Make NETIF_F_SG inheritable to tunnel devices.
6439 dev->hw_enc_features |= NETIF_F_SG;
6441 /* Make NETIF_F_SG inheritable to MPLS.
6443 dev->mpls_features |= NETIF_F_SG;
6445 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6446 ret = notifier_to_errno(ret);
6450 ret = netdev_register_kobject(dev);
6453 dev->reg_state = NETREG_REGISTERED;
6455 __netdev_update_features(dev);
6458 * Default initial state at registry is that the
6459 * device is present.
6462 set_bit(__LINK_STATE_PRESENT, &dev->state);
6464 linkwatch_init_dev(dev);
6466 dev_init_scheduler(dev);
6468 list_netdevice(dev);
6469 add_device_randomness(dev->dev_addr, dev->addr_len);
6471 /* If the device has permanent device address, driver should
6472 * set dev_addr and also addr_assign_type should be set to
6473 * NET_ADDR_PERM (default value).
6475 if (dev->addr_assign_type == NET_ADDR_PERM)
6476 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6478 /* Notify protocols, that a new device appeared. */
6479 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6480 ret = notifier_to_errno(ret);
6482 rollback_registered(dev);
6483 dev->reg_state = NETREG_UNREGISTERED;
6486 * Prevent userspace races by waiting until the network
6487 * device is fully setup before sending notifications.
6489 if (!dev->rtnl_link_ops ||
6490 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6491 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6497 if (dev->netdev_ops->ndo_uninit)
6498 dev->netdev_ops->ndo_uninit(dev);
6501 EXPORT_SYMBOL(register_netdevice);
6504 * init_dummy_netdev - init a dummy network device for NAPI
6505 * @dev: device to init
6507 * This takes a network device structure and initialize the minimum
6508 * amount of fields so it can be used to schedule NAPI polls without
6509 * registering a full blown interface. This is to be used by drivers
6510 * that need to tie several hardware interfaces to a single NAPI
6511 * poll scheduler due to HW limitations.
6513 int init_dummy_netdev(struct net_device *dev)
6515 /* Clear everything. Note we don't initialize spinlocks
6516 * are they aren't supposed to be taken by any of the
6517 * NAPI code and this dummy netdev is supposed to be
6518 * only ever used for NAPI polls
6520 memset(dev, 0, sizeof(struct net_device));
6522 /* make sure we BUG if trying to hit standard
6523 * register/unregister code path
6525 dev->reg_state = NETREG_DUMMY;
6527 /* NAPI wants this */
6528 INIT_LIST_HEAD(&dev->napi_list);
6530 /* a dummy interface is started by default */
6531 set_bit(__LINK_STATE_PRESENT, &dev->state);
6532 set_bit(__LINK_STATE_START, &dev->state);
6534 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6535 * because users of this 'device' dont need to change
6541 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6545 * register_netdev - register a network device
6546 * @dev: device to register
6548 * Take a completed network device structure and add it to the kernel
6549 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6550 * chain. 0 is returned on success. A negative errno code is returned
6551 * on a failure to set up the device, or if the name is a duplicate.
6553 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6554 * and expands the device name if you passed a format string to
6557 int register_netdev(struct net_device *dev)
6562 err = register_netdevice(dev);
6566 EXPORT_SYMBOL(register_netdev);
6568 int netdev_refcnt_read(const struct net_device *dev)
6572 for_each_possible_cpu(i)
6573 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6576 EXPORT_SYMBOL(netdev_refcnt_read);
6579 * netdev_wait_allrefs - wait until all references are gone.
6580 * @dev: target net_device
6582 * This is called when unregistering network devices.
6584 * Any protocol or device that holds a reference should register
6585 * for netdevice notification, and cleanup and put back the
6586 * reference if they receive an UNREGISTER event.
6587 * We can get stuck here if buggy protocols don't correctly
6590 static void netdev_wait_allrefs(struct net_device *dev)
6592 unsigned long rebroadcast_time, warning_time;
6595 linkwatch_forget_dev(dev);
6597 rebroadcast_time = warning_time = jiffies;
6598 refcnt = netdev_refcnt_read(dev);
6600 while (refcnt != 0) {
6601 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6604 /* Rebroadcast unregister notification */
6605 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6611 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6612 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6614 /* We must not have linkwatch events
6615 * pending on unregister. If this
6616 * happens, we simply run the queue
6617 * unscheduled, resulting in a noop
6620 linkwatch_run_queue();
6625 rebroadcast_time = jiffies;
6630 refcnt = netdev_refcnt_read(dev);
6632 if (time_after(jiffies, warning_time + 10 * HZ)) {
6633 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6635 warning_time = jiffies;
6644 * register_netdevice(x1);
6645 * register_netdevice(x2);
6647 * unregister_netdevice(y1);
6648 * unregister_netdevice(y2);
6654 * We are invoked by rtnl_unlock().
6655 * This allows us to deal with problems:
6656 * 1) We can delete sysfs objects which invoke hotplug
6657 * without deadlocking with linkwatch via keventd.
6658 * 2) Since we run with the RTNL semaphore not held, we can sleep
6659 * safely in order to wait for the netdev refcnt to drop to zero.
6661 * We must not return until all unregister events added during
6662 * the interval the lock was held have been completed.
6664 void netdev_run_todo(void)
6666 struct list_head list;
6668 /* Snapshot list, allow later requests */
6669 list_replace_init(&net_todo_list, &list);
6674 /* Wait for rcu callbacks to finish before next phase */
6675 if (!list_empty(&list))
6678 while (!list_empty(&list)) {
6679 struct net_device *dev
6680 = list_first_entry(&list, struct net_device, todo_list);
6681 list_del(&dev->todo_list);
6684 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6687 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6688 pr_err("network todo '%s' but state %d\n",
6689 dev->name, dev->reg_state);
6694 dev->reg_state = NETREG_UNREGISTERED;
6696 on_each_cpu(flush_backlog, dev, 1);
6698 netdev_wait_allrefs(dev);
6701 BUG_ON(netdev_refcnt_read(dev));
6702 BUG_ON(!list_empty(&dev->ptype_all));
6703 BUG_ON(!list_empty(&dev->ptype_specific));
6704 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6705 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6706 WARN_ON(dev->dn_ptr);
6708 if (dev->destructor)
6709 dev->destructor(dev);
6711 /* Report a network device has been unregistered */
6713 dev_net(dev)->dev_unreg_count--;
6715 wake_up(&netdev_unregistering_wq);
6717 /* Free network device */
6718 kobject_put(&dev->dev.kobj);
6722 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6723 * fields in the same order, with only the type differing.
6725 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6726 const struct net_device_stats *netdev_stats)
6728 #if BITS_PER_LONG == 64
6729 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6730 memcpy(stats64, netdev_stats, sizeof(*stats64));
6732 size_t i, n = sizeof(*stats64) / sizeof(u64);
6733 const unsigned long *src = (const unsigned long *)netdev_stats;
6734 u64 *dst = (u64 *)stats64;
6736 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6737 sizeof(*stats64) / sizeof(u64));
6738 for (i = 0; i < n; i++)
6742 EXPORT_SYMBOL(netdev_stats_to_stats64);
6745 * dev_get_stats - get network device statistics
6746 * @dev: device to get statistics from
6747 * @storage: place to store stats
6749 * Get network statistics from device. Return @storage.
6750 * The device driver may provide its own method by setting
6751 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6752 * otherwise the internal statistics structure is used.
6754 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6755 struct rtnl_link_stats64 *storage)
6757 const struct net_device_ops *ops = dev->netdev_ops;
6759 if (ops->ndo_get_stats64) {
6760 memset(storage, 0, sizeof(*storage));
6761 ops->ndo_get_stats64(dev, storage);
6762 } else if (ops->ndo_get_stats) {
6763 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6765 netdev_stats_to_stats64(storage, &dev->stats);
6767 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6768 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6771 EXPORT_SYMBOL(dev_get_stats);
6773 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6775 struct netdev_queue *queue = dev_ingress_queue(dev);
6777 #ifdef CONFIG_NET_CLS_ACT
6780 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6783 netdev_init_one_queue(dev, queue, NULL);
6784 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
6785 queue->qdisc_sleeping = &noop_qdisc;
6786 rcu_assign_pointer(dev->ingress_queue, queue);
6791 static const struct ethtool_ops default_ethtool_ops;
6793 void netdev_set_default_ethtool_ops(struct net_device *dev,
6794 const struct ethtool_ops *ops)
6796 if (dev->ethtool_ops == &default_ethtool_ops)
6797 dev->ethtool_ops = ops;
6799 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6801 void netdev_freemem(struct net_device *dev)
6803 char *addr = (char *)dev - dev->padded;
6809 * alloc_netdev_mqs - allocate network device
6810 * @sizeof_priv: size of private data to allocate space for
6811 * @name: device name format string
6812 * @name_assign_type: origin of device name
6813 * @setup: callback to initialize device
6814 * @txqs: the number of TX subqueues to allocate
6815 * @rxqs: the number of RX subqueues to allocate
6817 * Allocates a struct net_device with private data area for driver use
6818 * and performs basic initialization. Also allocates subqueue structs
6819 * for each queue on the device.
6821 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6822 unsigned char name_assign_type,
6823 void (*setup)(struct net_device *),
6824 unsigned int txqs, unsigned int rxqs)
6826 struct net_device *dev;
6828 struct net_device *p;
6830 BUG_ON(strlen(name) >= sizeof(dev->name));
6833 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6839 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6844 alloc_size = sizeof(struct net_device);
6846 /* ensure 32-byte alignment of private area */
6847 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6848 alloc_size += sizeof_priv;
6850 /* ensure 32-byte alignment of whole construct */
6851 alloc_size += NETDEV_ALIGN - 1;
6853 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6855 p = vzalloc(alloc_size);
6859 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6860 dev->padded = (char *)dev - (char *)p;
6862 dev->pcpu_refcnt = alloc_percpu(int);
6863 if (!dev->pcpu_refcnt)
6866 if (dev_addr_init(dev))
6872 dev_net_set(dev, &init_net);
6874 dev->gso_max_size = GSO_MAX_SIZE;
6875 dev->gso_max_segs = GSO_MAX_SEGS;
6876 dev->gso_min_segs = 0;
6878 INIT_LIST_HEAD(&dev->napi_list);
6879 INIT_LIST_HEAD(&dev->unreg_list);
6880 INIT_LIST_HEAD(&dev->close_list);
6881 INIT_LIST_HEAD(&dev->link_watch_list);
6882 INIT_LIST_HEAD(&dev->adj_list.upper);
6883 INIT_LIST_HEAD(&dev->adj_list.lower);
6884 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6885 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6886 INIT_LIST_HEAD(&dev->ptype_all);
6887 INIT_LIST_HEAD(&dev->ptype_specific);
6888 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
6891 dev->num_tx_queues = txqs;
6892 dev->real_num_tx_queues = txqs;
6893 if (netif_alloc_netdev_queues(dev))
6897 dev->num_rx_queues = rxqs;
6898 dev->real_num_rx_queues = rxqs;
6899 if (netif_alloc_rx_queues(dev))
6903 strcpy(dev->name, name);
6904 dev->name_assign_type = name_assign_type;
6905 dev->group = INIT_NETDEV_GROUP;
6906 if (!dev->ethtool_ops)
6907 dev->ethtool_ops = &default_ethtool_ops;
6915 free_percpu(dev->pcpu_refcnt);
6917 netdev_freemem(dev);
6920 EXPORT_SYMBOL(alloc_netdev_mqs);
6923 * free_netdev - free network device
6926 * This function does the last stage of destroying an allocated device
6927 * interface. The reference to the device object is released.
6928 * If this is the last reference then it will be freed.
6930 void free_netdev(struct net_device *dev)
6932 struct napi_struct *p, *n;
6934 netif_free_tx_queues(dev);
6939 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6941 /* Flush device addresses */
6942 dev_addr_flush(dev);
6944 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6947 free_percpu(dev->pcpu_refcnt);
6948 dev->pcpu_refcnt = NULL;
6950 /* Compatibility with error handling in drivers */
6951 if (dev->reg_state == NETREG_UNINITIALIZED) {
6952 netdev_freemem(dev);
6956 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6957 dev->reg_state = NETREG_RELEASED;
6959 /* will free via device release */
6960 put_device(&dev->dev);
6962 EXPORT_SYMBOL(free_netdev);
6965 * synchronize_net - Synchronize with packet receive processing
6967 * Wait for packets currently being received to be done.
6968 * Does not block later packets from starting.
6970 void synchronize_net(void)
6973 if (rtnl_is_locked())
6974 synchronize_rcu_expedited();
6978 EXPORT_SYMBOL(synchronize_net);
6981 * unregister_netdevice_queue - remove device from the kernel
6985 * This function shuts down a device interface and removes it
6986 * from the kernel tables.
6987 * If head not NULL, device is queued to be unregistered later.
6989 * Callers must hold the rtnl semaphore. You may want
6990 * unregister_netdev() instead of this.
6993 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6998 list_move_tail(&dev->unreg_list, head);
7000 rollback_registered(dev);
7001 /* Finish processing unregister after unlock */
7005 EXPORT_SYMBOL(unregister_netdevice_queue);
7008 * unregister_netdevice_many - unregister many devices
7009 * @head: list of devices
7011 * Note: As most callers use a stack allocated list_head,
7012 * we force a list_del() to make sure stack wont be corrupted later.
7014 void unregister_netdevice_many(struct list_head *head)
7016 struct net_device *dev;
7018 if (!list_empty(head)) {
7019 rollback_registered_many(head);
7020 list_for_each_entry(dev, head, unreg_list)
7025 EXPORT_SYMBOL(unregister_netdevice_many);
7028 * unregister_netdev - remove device from the kernel
7031 * This function shuts down a device interface and removes it
7032 * from the kernel tables.
7034 * This is just a wrapper for unregister_netdevice that takes
7035 * the rtnl semaphore. In general you want to use this and not
7036 * unregister_netdevice.
7038 void unregister_netdev(struct net_device *dev)
7041 unregister_netdevice(dev);
7044 EXPORT_SYMBOL(unregister_netdev);
7047 * dev_change_net_namespace - move device to different nethost namespace
7049 * @net: network namespace
7050 * @pat: If not NULL name pattern to try if the current device name
7051 * is already taken in the destination network namespace.
7053 * This function shuts down a device interface and moves it
7054 * to a new network namespace. On success 0 is returned, on
7055 * a failure a netagive errno code is returned.
7057 * Callers must hold the rtnl semaphore.
7060 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7066 /* Don't allow namespace local devices to be moved. */
7068 if (dev->features & NETIF_F_NETNS_LOCAL)
7071 /* Ensure the device has been registrered */
7072 if (dev->reg_state != NETREG_REGISTERED)
7075 /* Get out if there is nothing todo */
7077 if (net_eq(dev_net(dev), net))
7080 /* Pick the destination device name, and ensure
7081 * we can use it in the destination network namespace.
7084 if (__dev_get_by_name(net, dev->name)) {
7085 /* We get here if we can't use the current device name */
7088 if (dev_get_valid_name(net, dev, pat) < 0)
7093 * And now a mini version of register_netdevice unregister_netdevice.
7096 /* If device is running close it first. */
7099 /* And unlink it from device chain */
7101 unlist_netdevice(dev);
7105 /* Shutdown queueing discipline. */
7108 /* Notify protocols, that we are about to destroy
7109 this device. They should clean all the things.
7111 Note that dev->reg_state stays at NETREG_REGISTERED.
7112 This is wanted because this way 8021q and macvlan know
7113 the device is just moving and can keep their slaves up.
7115 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7117 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7118 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7121 * Flush the unicast and multicast chains
7126 /* Send a netdev-removed uevent to the old namespace */
7127 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7128 netdev_adjacent_del_links(dev);
7130 /* Actually switch the network namespace */
7131 dev_net_set(dev, net);
7133 /* If there is an ifindex conflict assign a new one */
7134 if (__dev_get_by_index(net, dev->ifindex))
7135 dev->ifindex = dev_new_index(net);
7137 /* Send a netdev-add uevent to the new namespace */
7138 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7139 netdev_adjacent_add_links(dev);
7141 /* Fixup kobjects */
7142 err = device_rename(&dev->dev, dev->name);
7145 /* Add the device back in the hashes */
7146 list_netdevice(dev);
7148 /* Notify protocols, that a new device appeared. */
7149 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7152 * Prevent userspace races by waiting until the network
7153 * device is fully setup before sending notifications.
7155 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7162 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7164 static int dev_cpu_callback(struct notifier_block *nfb,
7165 unsigned long action,
7168 struct sk_buff **list_skb;
7169 struct sk_buff *skb;
7170 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7171 struct softnet_data *sd, *oldsd;
7173 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7176 local_irq_disable();
7177 cpu = smp_processor_id();
7178 sd = &per_cpu(softnet_data, cpu);
7179 oldsd = &per_cpu(softnet_data, oldcpu);
7181 /* Find end of our completion_queue. */
7182 list_skb = &sd->completion_queue;
7184 list_skb = &(*list_skb)->next;
7185 /* Append completion queue from offline CPU. */
7186 *list_skb = oldsd->completion_queue;
7187 oldsd->completion_queue = NULL;
7189 /* Append output queue from offline CPU. */
7190 if (oldsd->output_queue) {
7191 *sd->output_queue_tailp = oldsd->output_queue;
7192 sd->output_queue_tailp = oldsd->output_queue_tailp;
7193 oldsd->output_queue = NULL;
7194 oldsd->output_queue_tailp = &oldsd->output_queue;
7196 /* Append NAPI poll list from offline CPU, with one exception :
7197 * process_backlog() must be called by cpu owning percpu backlog.
7198 * We properly handle process_queue & input_pkt_queue later.
7200 while (!list_empty(&oldsd->poll_list)) {
7201 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7205 list_del_init(&napi->poll_list);
7206 if (napi->poll == process_backlog)
7209 ____napi_schedule(sd, napi);
7212 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7214 preempt_check_resched_rt();
7216 /* Process offline CPU's input_pkt_queue */
7217 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7219 input_queue_head_incr(oldsd);
7221 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7223 input_queue_head_incr(oldsd);
7225 while ((skb = __skb_dequeue(&oldsd->tofree_queue))) {
7234 * netdev_increment_features - increment feature set by one
7235 * @all: current feature set
7236 * @one: new feature set
7237 * @mask: mask feature set
7239 * Computes a new feature set after adding a device with feature set
7240 * @one to the master device with current feature set @all. Will not
7241 * enable anything that is off in @mask. Returns the new feature set.
7243 netdev_features_t netdev_increment_features(netdev_features_t all,
7244 netdev_features_t one, netdev_features_t mask)
7246 if (mask & NETIF_F_GEN_CSUM)
7247 mask |= NETIF_F_ALL_CSUM;
7248 mask |= NETIF_F_VLAN_CHALLENGED;
7250 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7251 all &= one | ~NETIF_F_ALL_FOR_ALL;
7253 /* If one device supports hw checksumming, set for all. */
7254 if (all & NETIF_F_GEN_CSUM)
7255 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7259 EXPORT_SYMBOL(netdev_increment_features);
7261 static struct hlist_head * __net_init netdev_create_hash(void)
7264 struct hlist_head *hash;
7266 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7268 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7269 INIT_HLIST_HEAD(&hash[i]);
7274 /* Initialize per network namespace state */
7275 static int __net_init netdev_init(struct net *net)
7277 if (net != &init_net)
7278 INIT_LIST_HEAD(&net->dev_base_head);
7280 net->dev_name_head = netdev_create_hash();
7281 if (net->dev_name_head == NULL)
7284 net->dev_index_head = netdev_create_hash();
7285 if (net->dev_index_head == NULL)
7291 kfree(net->dev_name_head);
7297 * netdev_drivername - network driver for the device
7298 * @dev: network device
7300 * Determine network driver for device.
7302 const char *netdev_drivername(const struct net_device *dev)
7304 const struct device_driver *driver;
7305 const struct device *parent;
7306 const char *empty = "";
7308 parent = dev->dev.parent;
7312 driver = parent->driver;
7313 if (driver && driver->name)
7314 return driver->name;
7318 static void __netdev_printk(const char *level, const struct net_device *dev,
7319 struct va_format *vaf)
7321 if (dev && dev->dev.parent) {
7322 dev_printk_emit(level[1] - '0',
7325 dev_driver_string(dev->dev.parent),
7326 dev_name(dev->dev.parent),
7327 netdev_name(dev), netdev_reg_state(dev),
7330 printk("%s%s%s: %pV",
7331 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7333 printk("%s(NULL net_device): %pV", level, vaf);
7337 void netdev_printk(const char *level, const struct net_device *dev,
7338 const char *format, ...)
7340 struct va_format vaf;
7343 va_start(args, format);
7348 __netdev_printk(level, dev, &vaf);
7352 EXPORT_SYMBOL(netdev_printk);
7354 #define define_netdev_printk_level(func, level) \
7355 void func(const struct net_device *dev, const char *fmt, ...) \
7357 struct va_format vaf; \
7360 va_start(args, fmt); \
7365 __netdev_printk(level, dev, &vaf); \
7369 EXPORT_SYMBOL(func);
7371 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7372 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7373 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7374 define_netdev_printk_level(netdev_err, KERN_ERR);
7375 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7376 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7377 define_netdev_printk_level(netdev_info, KERN_INFO);
7379 static void __net_exit netdev_exit(struct net *net)
7381 kfree(net->dev_name_head);
7382 kfree(net->dev_index_head);
7385 static struct pernet_operations __net_initdata netdev_net_ops = {
7386 .init = netdev_init,
7387 .exit = netdev_exit,
7390 static void __net_exit default_device_exit(struct net *net)
7392 struct net_device *dev, *aux;
7394 * Push all migratable network devices back to the
7395 * initial network namespace
7398 for_each_netdev_safe(net, dev, aux) {
7400 char fb_name[IFNAMSIZ];
7402 /* Ignore unmoveable devices (i.e. loopback) */
7403 if (dev->features & NETIF_F_NETNS_LOCAL)
7406 /* Leave virtual devices for the generic cleanup */
7407 if (dev->rtnl_link_ops)
7410 /* Push remaining network devices to init_net */
7411 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7412 err = dev_change_net_namespace(dev, &init_net, fb_name);
7414 pr_emerg("%s: failed to move %s to init_net: %d\n",
7415 __func__, dev->name, err);
7422 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7424 /* Return with the rtnl_lock held when there are no network
7425 * devices unregistering in any network namespace in net_list.
7429 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7431 add_wait_queue(&netdev_unregistering_wq, &wait);
7433 unregistering = false;
7435 list_for_each_entry(net, net_list, exit_list) {
7436 if (net->dev_unreg_count > 0) {
7437 unregistering = true;
7445 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7447 remove_wait_queue(&netdev_unregistering_wq, &wait);
7450 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7452 /* At exit all network devices most be removed from a network
7453 * namespace. Do this in the reverse order of registration.
7454 * Do this across as many network namespaces as possible to
7455 * improve batching efficiency.
7457 struct net_device *dev;
7459 LIST_HEAD(dev_kill_list);
7461 /* To prevent network device cleanup code from dereferencing
7462 * loopback devices or network devices that have been freed
7463 * wait here for all pending unregistrations to complete,
7464 * before unregistring the loopback device and allowing the
7465 * network namespace be freed.
7467 * The netdev todo list containing all network devices
7468 * unregistrations that happen in default_device_exit_batch
7469 * will run in the rtnl_unlock() at the end of
7470 * default_device_exit_batch.
7472 rtnl_lock_unregistering(net_list);
7473 list_for_each_entry(net, net_list, exit_list) {
7474 for_each_netdev_reverse(net, dev) {
7475 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7476 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7478 unregister_netdevice_queue(dev, &dev_kill_list);
7481 unregister_netdevice_many(&dev_kill_list);
7485 static struct pernet_operations __net_initdata default_device_ops = {
7486 .exit = default_device_exit,
7487 .exit_batch = default_device_exit_batch,
7491 * Initialize the DEV module. At boot time this walks the device list and
7492 * unhooks any devices that fail to initialise (normally hardware not
7493 * present) and leaves us with a valid list of present and active devices.
7498 * This is called single threaded during boot, so no need
7499 * to take the rtnl semaphore.
7501 static int __init net_dev_init(void)
7503 int i, rc = -ENOMEM;
7505 BUG_ON(!dev_boot_phase);
7507 if (dev_proc_init())
7510 if (netdev_kobject_init())
7513 INIT_LIST_HEAD(&ptype_all);
7514 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7515 INIT_LIST_HEAD(&ptype_base[i]);
7517 INIT_LIST_HEAD(&offload_base);
7519 if (register_pernet_subsys(&netdev_net_ops))
7523 * Initialise the packet receive queues.
7526 for_each_possible_cpu(i) {
7527 struct softnet_data *sd = &per_cpu(softnet_data, i);
7529 skb_queue_head_init_raw(&sd->input_pkt_queue);
7530 skb_queue_head_init_raw(&sd->process_queue);
7531 skb_queue_head_init_raw(&sd->tofree_queue);
7532 INIT_LIST_HEAD(&sd->poll_list);
7533 sd->output_queue_tailp = &sd->output_queue;
7535 sd->csd.func = rps_trigger_softirq;
7540 sd->backlog.poll = process_backlog;
7541 sd->backlog.weight = weight_p;
7546 /* The loopback device is special if any other network devices
7547 * is present in a network namespace the loopback device must
7548 * be present. Since we now dynamically allocate and free the
7549 * loopback device ensure this invariant is maintained by
7550 * keeping the loopback device as the first device on the
7551 * list of network devices. Ensuring the loopback devices
7552 * is the first device that appears and the last network device
7555 if (register_pernet_device(&loopback_net_ops))
7558 if (register_pernet_device(&default_device_ops))
7561 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7562 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7564 hotcpu_notifier(dev_cpu_callback, 0);
7571 subsys_initcall(net_dev_init);