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 <net/busy_poll.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/stat.h>
103 #include <net/dst_metadata.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/module.h>
110 #include <linux/netpoll.h>
111 #include <linux/rcupdate.h>
112 #include <linux/delay.h>
113 #include <net/iw_handler.h>
114 #include <asm/current.h>
115 #include <linux/audit.h>
116 #include <linux/dmaengine.h>
117 #include <linux/err.h>
118 #include <linux/ctype.h>
119 #include <linux/if_arp.h>
120 #include <linux/if_vlan.h>
121 #include <linux/ip.h>
123 #include <net/mpls.h>
124 #include <linux/ipv6.h>
125 #include <linux/in.h>
126 #include <linux/jhash.h>
127 #include <linux/random.h>
128 #include <trace/events/napi.h>
129 #include <trace/events/net.h>
130 #include <trace/events/skb.h>
131 #include <linux/pci.h>
132 #include <linux/inetdevice.h>
133 #include <linux/cpu_rmap.h>
134 #include <linux/static_key.h>
135 #include <linux/hashtable.h>
136 #include <linux/vmalloc.h>
137 #include <linux/if_macvlan.h>
138 #include <linux/errqueue.h>
139 #include <linux/hrtimer.h>
140 #include <linux/netfilter_ingress.h>
141 #include <linux/sctp.h>
143 #include "net-sysfs.h"
145 /* Instead of increasing this, you should create a hash table. */
146 #define MAX_GRO_SKBS 8
148 /* This should be increased if a protocol with a bigger head is added. */
149 #define GRO_MAX_HEAD (MAX_HEADER + 128)
151 static DEFINE_SPINLOCK(ptype_lock);
152 static DEFINE_SPINLOCK(offload_lock);
153 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
154 struct list_head ptype_all __read_mostly; /* Taps */
155 static struct list_head offload_base __read_mostly;
157 static int netif_rx_internal(struct sk_buff *skb);
158 static int call_netdevice_notifiers_info(unsigned long val,
159 struct net_device *dev,
160 struct netdev_notifier_info *info);
163 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
166 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
168 * Writers must hold the rtnl semaphore while they loop through the
169 * dev_base_head list, and hold dev_base_lock for writing when they do the
170 * actual updates. This allows pure readers to access the list even
171 * while a writer is preparing to update it.
173 * To put it another way, dev_base_lock is held for writing only to
174 * protect against pure readers; the rtnl semaphore provides the
175 * protection against other writers.
177 * See, for example usages, register_netdevice() and
178 * unregister_netdevice(), which must be called with the rtnl
181 DEFINE_RWLOCK(dev_base_lock);
182 EXPORT_SYMBOL(dev_base_lock);
184 /* protects napi_hash addition/deletion and napi_gen_id */
185 static DEFINE_SPINLOCK(napi_hash_lock);
187 static unsigned int napi_gen_id = NR_CPUS;
188 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
190 static seqcount_t devnet_rename_seq;
192 static inline void dev_base_seq_inc(struct net *net)
194 while (++net->dev_base_seq == 0);
197 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
199 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
201 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
204 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
206 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
209 static inline void rps_lock(struct softnet_data *sd)
212 spin_lock(&sd->input_pkt_queue.lock);
216 static inline void rps_unlock(struct softnet_data *sd)
219 spin_unlock(&sd->input_pkt_queue.lock);
223 /* Device list insertion */
224 static void list_netdevice(struct net_device *dev)
226 struct net *net = dev_net(dev);
230 write_lock_bh(&dev_base_lock);
231 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
232 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
233 hlist_add_head_rcu(&dev->index_hlist,
234 dev_index_hash(net, dev->ifindex));
235 write_unlock_bh(&dev_base_lock);
237 dev_base_seq_inc(net);
240 /* Device list removal
241 * caller must respect a RCU grace period before freeing/reusing dev
243 static void unlist_netdevice(struct net_device *dev)
247 /* Unlink dev from the device chain */
248 write_lock_bh(&dev_base_lock);
249 list_del_rcu(&dev->dev_list);
250 hlist_del_rcu(&dev->name_hlist);
251 hlist_del_rcu(&dev->index_hlist);
252 write_unlock_bh(&dev_base_lock);
254 dev_base_seq_inc(dev_net(dev));
261 static RAW_NOTIFIER_HEAD(netdev_chain);
264 * Device drivers call our routines to queue packets here. We empty the
265 * queue in the local softnet handler.
268 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
269 EXPORT_PER_CPU_SYMBOL(softnet_data);
271 #ifdef CONFIG_LOCKDEP
273 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
274 * according to dev->type
276 static const unsigned short netdev_lock_type[] =
277 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
278 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
279 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
280 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
281 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
282 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
283 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
284 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
285 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
286 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
287 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
288 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
289 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
290 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
291 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
293 static const char *const netdev_lock_name[] =
294 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
295 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
296 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
297 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
298 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
299 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
300 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
301 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
302 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
303 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
304 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
305 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
306 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
307 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
308 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
310 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
311 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
313 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
317 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
318 if (netdev_lock_type[i] == dev_type)
320 /* the last key is used by default */
321 return ARRAY_SIZE(netdev_lock_type) - 1;
324 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
325 unsigned short dev_type)
329 i = netdev_lock_pos(dev_type);
330 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
331 netdev_lock_name[i]);
334 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
338 i = netdev_lock_pos(dev->type);
339 lockdep_set_class_and_name(&dev->addr_list_lock,
340 &netdev_addr_lock_key[i],
341 netdev_lock_name[i]);
344 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
345 unsigned short dev_type)
348 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
353 /*******************************************************************************
355 Protocol management and registration routines
357 *******************************************************************************/
360 * Add a protocol ID to the list. Now that the input handler is
361 * smarter we can dispense with all the messy stuff that used to be
364 * BEWARE!!! Protocol handlers, mangling input packets,
365 * MUST BE last in hash buckets and checking protocol handlers
366 * MUST start from promiscuous ptype_all chain in net_bh.
367 * It is true now, do not change it.
368 * Explanation follows: if protocol handler, mangling packet, will
369 * be the first on list, it is not able to sense, that packet
370 * is cloned and should be copied-on-write, so that it will
371 * change it and subsequent readers will get broken packet.
375 static inline struct list_head *ptype_head(const struct packet_type *pt)
377 if (pt->type == htons(ETH_P_ALL))
378 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
380 return pt->dev ? &pt->dev->ptype_specific :
381 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
385 * dev_add_pack - add packet handler
386 * @pt: packet type declaration
388 * Add a protocol handler to the networking stack. The passed &packet_type
389 * is linked into kernel lists and may not be freed until it has been
390 * removed from the kernel lists.
392 * This call does not sleep therefore it can not
393 * guarantee all CPU's that are in middle of receiving packets
394 * will see the new packet type (until the next received packet).
397 void dev_add_pack(struct packet_type *pt)
399 struct list_head *head = ptype_head(pt);
401 spin_lock(&ptype_lock);
402 list_add_rcu(&pt->list, head);
403 spin_unlock(&ptype_lock);
405 EXPORT_SYMBOL(dev_add_pack);
408 * __dev_remove_pack - remove packet handler
409 * @pt: packet type declaration
411 * Remove a protocol handler that was previously added to the kernel
412 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
413 * from the kernel lists and can be freed or reused once this function
416 * The packet type might still be in use by receivers
417 * and must not be freed until after all the CPU's have gone
418 * through a quiescent state.
420 void __dev_remove_pack(struct packet_type *pt)
422 struct list_head *head = ptype_head(pt);
423 struct packet_type *pt1;
425 spin_lock(&ptype_lock);
427 list_for_each_entry(pt1, head, list) {
429 list_del_rcu(&pt->list);
434 pr_warn("dev_remove_pack: %p not found\n", pt);
436 spin_unlock(&ptype_lock);
438 EXPORT_SYMBOL(__dev_remove_pack);
441 * dev_remove_pack - remove packet handler
442 * @pt: packet type declaration
444 * Remove a protocol handler that was previously added to the kernel
445 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
446 * from the kernel lists and can be freed or reused once this function
449 * This call sleeps to guarantee that no CPU is looking at the packet
452 void dev_remove_pack(struct packet_type *pt)
454 __dev_remove_pack(pt);
458 EXPORT_SYMBOL(dev_remove_pack);
462 * dev_add_offload - register offload handlers
463 * @po: protocol offload declaration
465 * Add protocol offload handlers to the networking stack. The passed
466 * &proto_offload is linked into kernel lists and may not be freed until
467 * it has been removed from the kernel lists.
469 * This call does not sleep therefore it can not
470 * guarantee all CPU's that are in middle of receiving packets
471 * will see the new offload handlers (until the next received packet).
473 void dev_add_offload(struct packet_offload *po)
475 struct packet_offload *elem;
477 spin_lock(&offload_lock);
478 list_for_each_entry(elem, &offload_base, list) {
479 if (po->priority < elem->priority)
482 list_add_rcu(&po->list, elem->list.prev);
483 spin_unlock(&offload_lock);
485 EXPORT_SYMBOL(dev_add_offload);
488 * __dev_remove_offload - remove offload handler
489 * @po: packet offload declaration
491 * Remove a protocol offload handler that was previously added to the
492 * kernel offload handlers by dev_add_offload(). The passed &offload_type
493 * is removed from the kernel lists and can be freed or reused once this
496 * The packet type might still be in use by receivers
497 * and must not be freed until after all the CPU's have gone
498 * through a quiescent state.
500 static void __dev_remove_offload(struct packet_offload *po)
502 struct list_head *head = &offload_base;
503 struct packet_offload *po1;
505 spin_lock(&offload_lock);
507 list_for_each_entry(po1, head, list) {
509 list_del_rcu(&po->list);
514 pr_warn("dev_remove_offload: %p not found\n", po);
516 spin_unlock(&offload_lock);
520 * dev_remove_offload - remove packet offload handler
521 * @po: packet offload declaration
523 * Remove a packet offload handler that was previously added to the kernel
524 * offload handlers by dev_add_offload(). The passed &offload_type is
525 * removed from the kernel lists and can be freed or reused once this
528 * This call sleeps to guarantee that no CPU is looking at the packet
531 void dev_remove_offload(struct packet_offload *po)
533 __dev_remove_offload(po);
537 EXPORT_SYMBOL(dev_remove_offload);
539 /******************************************************************************
541 Device Boot-time Settings Routines
543 *******************************************************************************/
545 /* Boot time configuration table */
546 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
549 * netdev_boot_setup_add - add new setup entry
550 * @name: name of the device
551 * @map: configured settings for the device
553 * Adds new setup entry to the dev_boot_setup list. The function
554 * returns 0 on error and 1 on success. This is a generic routine to
557 static int netdev_boot_setup_add(char *name, struct ifmap *map)
559 struct netdev_boot_setup *s;
563 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
564 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
565 memset(s[i].name, 0, sizeof(s[i].name));
566 strlcpy(s[i].name, name, IFNAMSIZ);
567 memcpy(&s[i].map, map, sizeof(s[i].map));
572 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
576 * netdev_boot_setup_check - check boot time settings
577 * @dev: the netdevice
579 * Check boot time settings for the device.
580 * The found settings are set for the device to be used
581 * later in the device probing.
582 * Returns 0 if no settings found, 1 if they are.
584 int netdev_boot_setup_check(struct net_device *dev)
586 struct netdev_boot_setup *s = dev_boot_setup;
589 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
590 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
591 !strcmp(dev->name, s[i].name)) {
592 dev->irq = s[i].map.irq;
593 dev->base_addr = s[i].map.base_addr;
594 dev->mem_start = s[i].map.mem_start;
595 dev->mem_end = s[i].map.mem_end;
601 EXPORT_SYMBOL(netdev_boot_setup_check);
605 * netdev_boot_base - get address from boot time settings
606 * @prefix: prefix for network device
607 * @unit: id for network device
609 * Check boot time settings for the base address of device.
610 * The found settings are set for the device to be used
611 * later in the device probing.
612 * Returns 0 if no settings found.
614 unsigned long netdev_boot_base(const char *prefix, int unit)
616 const struct netdev_boot_setup *s = dev_boot_setup;
620 sprintf(name, "%s%d", prefix, unit);
623 * If device already registered then return base of 1
624 * to indicate not to probe for this interface
626 if (__dev_get_by_name(&init_net, name))
629 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
630 if (!strcmp(name, s[i].name))
631 return s[i].map.base_addr;
636 * Saves at boot time configured settings for any netdevice.
638 int __init netdev_boot_setup(char *str)
643 str = get_options(str, ARRAY_SIZE(ints), ints);
648 memset(&map, 0, sizeof(map));
652 map.base_addr = ints[2];
654 map.mem_start = ints[3];
656 map.mem_end = ints[4];
658 /* Add new entry to the list */
659 return netdev_boot_setup_add(str, &map);
662 __setup("netdev=", netdev_boot_setup);
664 /*******************************************************************************
666 Device Interface Subroutines
668 *******************************************************************************/
671 * dev_get_iflink - get 'iflink' value of a interface
672 * @dev: targeted interface
674 * Indicates the ifindex the interface is linked to.
675 * Physical interfaces have the same 'ifindex' and 'iflink' values.
678 int dev_get_iflink(const struct net_device *dev)
680 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
681 return dev->netdev_ops->ndo_get_iflink(dev);
685 EXPORT_SYMBOL(dev_get_iflink);
688 * dev_fill_metadata_dst - Retrieve tunnel egress information.
689 * @dev: targeted interface
692 * For better visibility of tunnel traffic OVS needs to retrieve
693 * egress tunnel information for a packet. Following API allows
694 * user to get this info.
696 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
698 struct ip_tunnel_info *info;
700 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
703 info = skb_tunnel_info_unclone(skb);
706 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
709 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
711 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
714 * __dev_get_by_name - find a device by its name
715 * @net: the applicable net namespace
716 * @name: name to find
718 * Find an interface by name. Must be called under RTNL semaphore
719 * or @dev_base_lock. If the name is found a pointer to the device
720 * is returned. If the name is not found then %NULL is returned. The
721 * reference counters are not incremented so the caller must be
722 * careful with locks.
725 struct net_device *__dev_get_by_name(struct net *net, const char *name)
727 struct net_device *dev;
728 struct hlist_head *head = dev_name_hash(net, name);
730 hlist_for_each_entry(dev, head, name_hlist)
731 if (!strncmp(dev->name, name, IFNAMSIZ))
736 EXPORT_SYMBOL(__dev_get_by_name);
739 * dev_get_by_name_rcu - find a device by its name
740 * @net: the applicable net namespace
741 * @name: name to find
743 * Find an interface by name.
744 * If the name is found a pointer to the device is returned.
745 * If the name is not found then %NULL is returned.
746 * The reference counters are not incremented so the caller must be
747 * careful with locks. The caller must hold RCU lock.
750 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
752 struct net_device *dev;
753 struct hlist_head *head = dev_name_hash(net, name);
755 hlist_for_each_entry_rcu(dev, head, name_hlist)
756 if (!strncmp(dev->name, name, IFNAMSIZ))
761 EXPORT_SYMBOL(dev_get_by_name_rcu);
764 * dev_get_by_name - find a device by its name
765 * @net: the applicable net namespace
766 * @name: name to find
768 * Find an interface by name. This can be called from any
769 * context and does its own locking. The returned handle has
770 * the usage count incremented and the caller must use dev_put() to
771 * release it when it is no longer needed. %NULL is returned if no
772 * matching device is found.
775 struct net_device *dev_get_by_name(struct net *net, const char *name)
777 struct net_device *dev;
780 dev = dev_get_by_name_rcu(net, name);
786 EXPORT_SYMBOL(dev_get_by_name);
789 * __dev_get_by_index - find a device by its ifindex
790 * @net: the applicable net namespace
791 * @ifindex: index of device
793 * Search for an interface by index. Returns %NULL if the device
794 * is not found or a pointer to the device. The device has not
795 * had its reference counter increased so the caller must be careful
796 * about locking. The caller must hold either the RTNL semaphore
800 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
802 struct net_device *dev;
803 struct hlist_head *head = dev_index_hash(net, ifindex);
805 hlist_for_each_entry(dev, head, index_hlist)
806 if (dev->ifindex == ifindex)
811 EXPORT_SYMBOL(__dev_get_by_index);
814 * dev_get_by_index_rcu - find a device by its ifindex
815 * @net: the applicable net namespace
816 * @ifindex: index of device
818 * Search for an interface by index. Returns %NULL if the device
819 * is not found or a pointer to the device. The device has not
820 * had its reference counter increased so the caller must be careful
821 * about locking. The caller must hold RCU lock.
824 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
826 struct net_device *dev;
827 struct hlist_head *head = dev_index_hash(net, ifindex);
829 hlist_for_each_entry_rcu(dev, head, index_hlist)
830 if (dev->ifindex == ifindex)
835 EXPORT_SYMBOL(dev_get_by_index_rcu);
839 * dev_get_by_index - find a device by its ifindex
840 * @net: the applicable net namespace
841 * @ifindex: index of device
843 * Search for an interface by index. Returns NULL if the device
844 * is not found or a pointer to the device. The device returned has
845 * had a reference added and the pointer is safe until the user calls
846 * dev_put to indicate they have finished with it.
849 struct net_device *dev_get_by_index(struct net *net, int ifindex)
851 struct net_device *dev;
854 dev = dev_get_by_index_rcu(net, ifindex);
860 EXPORT_SYMBOL(dev_get_by_index);
863 * netdev_get_name - get a netdevice name, knowing its ifindex.
864 * @net: network namespace
865 * @name: a pointer to the buffer where the name will be stored.
866 * @ifindex: the ifindex of the interface to get the name from.
868 * The use of raw_seqcount_begin() and cond_resched() before
869 * retrying is required as we want to give the writers a chance
870 * to complete when CONFIG_PREEMPT is not set.
872 int netdev_get_name(struct net *net, char *name, int ifindex)
874 struct net_device *dev;
878 seq = raw_seqcount_begin(&devnet_rename_seq);
880 dev = dev_get_by_index_rcu(net, ifindex);
886 strcpy(name, dev->name);
888 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
897 * dev_getbyhwaddr_rcu - find a device by its hardware address
898 * @net: the applicable net namespace
899 * @type: media type of device
900 * @ha: hardware address
902 * Search for an interface by MAC address. Returns NULL if the device
903 * is not found or a pointer to the device.
904 * The caller must hold RCU or RTNL.
905 * The returned device has not had its ref count increased
906 * and the caller must therefore be careful about locking
910 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
913 struct net_device *dev;
915 for_each_netdev_rcu(net, dev)
916 if (dev->type == type &&
917 !memcmp(dev->dev_addr, ha, dev->addr_len))
922 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
924 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
926 struct net_device *dev;
929 for_each_netdev(net, dev)
930 if (dev->type == type)
935 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
937 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
939 struct net_device *dev, *ret = NULL;
942 for_each_netdev_rcu(net, dev)
943 if (dev->type == type) {
951 EXPORT_SYMBOL(dev_getfirstbyhwtype);
954 * __dev_get_by_flags - find any device with given flags
955 * @net: the applicable net namespace
956 * @if_flags: IFF_* values
957 * @mask: bitmask of bits in if_flags to check
959 * Search for any interface with the given flags. Returns NULL if a device
960 * is not found or a pointer to the device. Must be called inside
961 * rtnl_lock(), and result refcount is unchanged.
964 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
967 struct net_device *dev, *ret;
972 for_each_netdev(net, dev) {
973 if (((dev->flags ^ if_flags) & mask) == 0) {
980 EXPORT_SYMBOL(__dev_get_by_flags);
983 * dev_valid_name - check if name is okay for network device
986 * Network device names need to be valid file names to
987 * to allow sysfs to work. We also disallow any kind of
990 bool dev_valid_name(const char *name)
994 if (strlen(name) >= IFNAMSIZ)
996 if (!strcmp(name, ".") || !strcmp(name, ".."))
1000 if (*name == '/' || *name == ':' || isspace(*name))
1006 EXPORT_SYMBOL(dev_valid_name);
1009 * __dev_alloc_name - allocate a name for a device
1010 * @net: network namespace to allocate the device name in
1011 * @name: name format string
1012 * @buf: scratch buffer and result name string
1014 * Passed a format string - eg "lt%d" it will try and find a suitable
1015 * id. It scans list of devices to build up a free map, then chooses
1016 * the first empty slot. The caller must hold the dev_base or rtnl lock
1017 * while allocating the name and adding the device in order to avoid
1019 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1020 * Returns the number of the unit assigned or a negative errno code.
1023 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1027 const int max_netdevices = 8*PAGE_SIZE;
1028 unsigned long *inuse;
1029 struct net_device *d;
1031 p = strnchr(name, IFNAMSIZ-1, '%');
1034 * Verify the string as this thing may have come from
1035 * the user. There must be either one "%d" and no other "%"
1038 if (p[1] != 'd' || strchr(p + 2, '%'))
1041 /* Use one page as a bit array of possible slots */
1042 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1046 for_each_netdev(net, d) {
1047 if (!sscanf(d->name, name, &i))
1049 if (i < 0 || i >= max_netdevices)
1052 /* avoid cases where sscanf is not exact inverse of printf */
1053 snprintf(buf, IFNAMSIZ, name, i);
1054 if (!strncmp(buf, d->name, IFNAMSIZ))
1058 i = find_first_zero_bit(inuse, max_netdevices);
1059 free_page((unsigned long) inuse);
1063 snprintf(buf, IFNAMSIZ, name, i);
1064 if (!__dev_get_by_name(net, buf))
1067 /* It is possible to run out of possible slots
1068 * when the name is long and there isn't enough space left
1069 * for the digits, or if all bits are used.
1075 * dev_alloc_name - allocate a name for a device
1077 * @name: name format string
1079 * Passed a format string - eg "lt%d" it will try and find a suitable
1080 * id. It scans list of devices to build up a free map, then chooses
1081 * the first empty slot. The caller must hold the dev_base or rtnl lock
1082 * while allocating the name and adding the device in order to avoid
1084 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1085 * Returns the number of the unit assigned or a negative errno code.
1088 int dev_alloc_name(struct net_device *dev, const char *name)
1094 BUG_ON(!dev_net(dev));
1096 ret = __dev_alloc_name(net, name, buf);
1098 strlcpy(dev->name, buf, IFNAMSIZ);
1101 EXPORT_SYMBOL(dev_alloc_name);
1103 static int dev_alloc_name_ns(struct net *net,
1104 struct net_device *dev,
1110 ret = __dev_alloc_name(net, name, buf);
1112 strlcpy(dev->name, buf, IFNAMSIZ);
1116 static int dev_get_valid_name(struct net *net,
1117 struct net_device *dev,
1122 if (!dev_valid_name(name))
1125 if (strchr(name, '%'))
1126 return dev_alloc_name_ns(net, dev, name);
1127 else if (__dev_get_by_name(net, name))
1129 else if (dev->name != name)
1130 strlcpy(dev->name, name, IFNAMSIZ);
1136 * dev_change_name - change name of a device
1138 * @newname: name (or format string) must be at least IFNAMSIZ
1140 * Change name of a device, can pass format strings "eth%d".
1143 int dev_change_name(struct net_device *dev, const char *newname)
1145 unsigned char old_assign_type;
1146 char oldname[IFNAMSIZ];
1152 BUG_ON(!dev_net(dev));
1155 if (dev->flags & IFF_UP)
1158 write_seqcount_begin(&devnet_rename_seq);
1160 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1161 write_seqcount_end(&devnet_rename_seq);
1165 memcpy(oldname, dev->name, IFNAMSIZ);
1167 err = dev_get_valid_name(net, dev, newname);
1169 write_seqcount_end(&devnet_rename_seq);
1173 if (oldname[0] && !strchr(oldname, '%'))
1174 netdev_info(dev, "renamed from %s\n", oldname);
1176 old_assign_type = dev->name_assign_type;
1177 dev->name_assign_type = NET_NAME_RENAMED;
1180 ret = device_rename(&dev->dev, dev->name);
1182 memcpy(dev->name, oldname, IFNAMSIZ);
1183 dev->name_assign_type = old_assign_type;
1184 write_seqcount_end(&devnet_rename_seq);
1188 write_seqcount_end(&devnet_rename_seq);
1190 netdev_adjacent_rename_links(dev, oldname);
1192 write_lock_bh(&dev_base_lock);
1193 hlist_del_rcu(&dev->name_hlist);
1194 write_unlock_bh(&dev_base_lock);
1198 write_lock_bh(&dev_base_lock);
1199 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1200 write_unlock_bh(&dev_base_lock);
1202 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1203 ret = notifier_to_errno(ret);
1206 /* err >= 0 after dev_alloc_name() or stores the first errno */
1209 write_seqcount_begin(&devnet_rename_seq);
1210 memcpy(dev->name, oldname, IFNAMSIZ);
1211 memcpy(oldname, newname, IFNAMSIZ);
1212 dev->name_assign_type = old_assign_type;
1213 old_assign_type = NET_NAME_RENAMED;
1216 pr_err("%s: name change rollback failed: %d\n",
1225 * dev_set_alias - change ifalias of a device
1227 * @alias: name up to IFALIASZ
1228 * @len: limit of bytes to copy from info
1230 * Set ifalias for a device,
1232 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1238 if (len >= IFALIASZ)
1242 kfree(dev->ifalias);
1243 dev->ifalias = NULL;
1247 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1250 dev->ifalias = new_ifalias;
1252 strlcpy(dev->ifalias, alias, len+1);
1258 * netdev_features_change - device changes features
1259 * @dev: device to cause notification
1261 * Called to indicate a device has changed features.
1263 void netdev_features_change(struct net_device *dev)
1265 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1267 EXPORT_SYMBOL(netdev_features_change);
1270 * netdev_state_change - device changes state
1271 * @dev: device to cause notification
1273 * Called to indicate a device has changed state. This function calls
1274 * the notifier chains for netdev_chain and sends a NEWLINK message
1275 * to the routing socket.
1277 void netdev_state_change(struct net_device *dev)
1279 if (dev->flags & IFF_UP) {
1280 struct netdev_notifier_change_info change_info;
1282 change_info.flags_changed = 0;
1283 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1285 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1288 EXPORT_SYMBOL(netdev_state_change);
1291 * netdev_notify_peers - notify network peers about existence of @dev
1292 * @dev: network device
1294 * Generate traffic such that interested network peers are aware of
1295 * @dev, such as by generating a gratuitous ARP. This may be used when
1296 * a device wants to inform the rest of the network about some sort of
1297 * reconfiguration such as a failover event or virtual machine
1300 void netdev_notify_peers(struct net_device *dev)
1303 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1306 EXPORT_SYMBOL(netdev_notify_peers);
1308 static int __dev_open(struct net_device *dev)
1310 const struct net_device_ops *ops = dev->netdev_ops;
1315 if (!netif_device_present(dev))
1318 /* Block netpoll from trying to do any rx path servicing.
1319 * If we don't do this there is a chance ndo_poll_controller
1320 * or ndo_poll may be running while we open the device
1322 netpoll_poll_disable(dev);
1324 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1325 ret = notifier_to_errno(ret);
1329 set_bit(__LINK_STATE_START, &dev->state);
1331 if (ops->ndo_validate_addr)
1332 ret = ops->ndo_validate_addr(dev);
1334 if (!ret && ops->ndo_open)
1335 ret = ops->ndo_open(dev);
1337 netpoll_poll_enable(dev);
1340 clear_bit(__LINK_STATE_START, &dev->state);
1342 dev->flags |= IFF_UP;
1343 dev_set_rx_mode(dev);
1345 add_device_randomness(dev->dev_addr, dev->addr_len);
1352 * dev_open - prepare an interface for use.
1353 * @dev: device to open
1355 * Takes a device from down to up state. The device's private open
1356 * function is invoked and then the multicast lists are loaded. Finally
1357 * the device is moved into the up state and a %NETDEV_UP message is
1358 * sent to the netdev notifier chain.
1360 * Calling this function on an active interface is a nop. On a failure
1361 * a negative errno code is returned.
1363 int dev_open(struct net_device *dev)
1367 if (dev->flags & IFF_UP)
1370 ret = __dev_open(dev);
1374 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1375 call_netdevice_notifiers(NETDEV_UP, dev);
1379 EXPORT_SYMBOL(dev_open);
1381 static int __dev_close_many(struct list_head *head)
1383 struct net_device *dev;
1388 list_for_each_entry(dev, head, close_list) {
1389 /* Temporarily disable netpoll until the interface is down */
1390 netpoll_poll_disable(dev);
1392 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1394 clear_bit(__LINK_STATE_START, &dev->state);
1396 /* Synchronize to scheduled poll. We cannot touch poll list, it
1397 * can be even on different cpu. So just clear netif_running().
1399 * dev->stop() will invoke napi_disable() on all of it's
1400 * napi_struct instances on this device.
1402 smp_mb__after_atomic(); /* Commit netif_running(). */
1405 dev_deactivate_many(head);
1407 list_for_each_entry(dev, head, close_list) {
1408 const struct net_device_ops *ops = dev->netdev_ops;
1411 * Call the device specific close. This cannot fail.
1412 * Only if device is UP
1414 * We allow it to be called even after a DETACH hot-plug
1420 dev->flags &= ~IFF_UP;
1421 netpoll_poll_enable(dev);
1427 static int __dev_close(struct net_device *dev)
1432 list_add(&dev->close_list, &single);
1433 retval = __dev_close_many(&single);
1439 int dev_close_many(struct list_head *head, bool unlink)
1441 struct net_device *dev, *tmp;
1443 /* Remove the devices that don't need to be closed */
1444 list_for_each_entry_safe(dev, tmp, head, close_list)
1445 if (!(dev->flags & IFF_UP))
1446 list_del_init(&dev->close_list);
1448 __dev_close_many(head);
1450 list_for_each_entry_safe(dev, tmp, head, close_list) {
1451 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1452 call_netdevice_notifiers(NETDEV_DOWN, dev);
1454 list_del_init(&dev->close_list);
1459 EXPORT_SYMBOL(dev_close_many);
1462 * dev_close - shutdown an interface.
1463 * @dev: device to shutdown
1465 * This function moves an active device into down state. A
1466 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1467 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1470 int dev_close(struct net_device *dev)
1472 if (dev->flags & IFF_UP) {
1475 list_add(&dev->close_list, &single);
1476 dev_close_many(&single, true);
1481 EXPORT_SYMBOL(dev_close);
1485 * dev_disable_lro - disable Large Receive Offload on a device
1488 * Disable Large Receive Offload (LRO) on a net device. Must be
1489 * called under RTNL. This is needed if received packets may be
1490 * forwarded to another interface.
1492 void dev_disable_lro(struct net_device *dev)
1494 struct net_device *lower_dev;
1495 struct list_head *iter;
1497 dev->wanted_features &= ~NETIF_F_LRO;
1498 netdev_update_features(dev);
1500 if (unlikely(dev->features & NETIF_F_LRO))
1501 netdev_WARN(dev, "failed to disable LRO!\n");
1503 netdev_for_each_lower_dev(dev, lower_dev, iter)
1504 dev_disable_lro(lower_dev);
1506 EXPORT_SYMBOL(dev_disable_lro);
1508 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1509 struct net_device *dev)
1511 struct netdev_notifier_info info;
1513 netdev_notifier_info_init(&info, dev);
1514 return nb->notifier_call(nb, val, &info);
1517 static int dev_boot_phase = 1;
1520 * register_netdevice_notifier - register a network notifier block
1523 * Register a notifier to be called when network device events occur.
1524 * The notifier passed is linked into the kernel structures and must
1525 * not be reused until it has been unregistered. A negative errno code
1526 * is returned on a failure.
1528 * When registered all registration and up events are replayed
1529 * to the new notifier to allow device to have a race free
1530 * view of the network device list.
1533 int register_netdevice_notifier(struct notifier_block *nb)
1535 struct net_device *dev;
1536 struct net_device *last;
1541 err = raw_notifier_chain_register(&netdev_chain, nb);
1547 for_each_netdev(net, dev) {
1548 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1549 err = notifier_to_errno(err);
1553 if (!(dev->flags & IFF_UP))
1556 call_netdevice_notifier(nb, NETDEV_UP, dev);
1567 for_each_netdev(net, dev) {
1571 if (dev->flags & IFF_UP) {
1572 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1574 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1576 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1581 raw_notifier_chain_unregister(&netdev_chain, nb);
1584 EXPORT_SYMBOL(register_netdevice_notifier);
1587 * unregister_netdevice_notifier - unregister a network notifier block
1590 * Unregister a notifier previously registered by
1591 * register_netdevice_notifier(). The notifier is unlinked into the
1592 * kernel structures and may then be reused. A negative errno code
1593 * is returned on a failure.
1595 * After unregistering unregister and down device events are synthesized
1596 * for all devices on the device list to the removed notifier to remove
1597 * the need for special case cleanup code.
1600 int unregister_netdevice_notifier(struct notifier_block *nb)
1602 struct net_device *dev;
1607 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1612 for_each_netdev(net, dev) {
1613 if (dev->flags & IFF_UP) {
1614 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1616 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1618 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1625 EXPORT_SYMBOL(unregister_netdevice_notifier);
1628 * call_netdevice_notifiers_info - call all network notifier blocks
1629 * @val: value passed unmodified to notifier function
1630 * @dev: net_device pointer passed unmodified to notifier function
1631 * @info: notifier information data
1633 * Call all network notifier blocks. Parameters and return value
1634 * are as for raw_notifier_call_chain().
1637 static int call_netdevice_notifiers_info(unsigned long val,
1638 struct net_device *dev,
1639 struct netdev_notifier_info *info)
1642 netdev_notifier_info_init(info, dev);
1643 return raw_notifier_call_chain(&netdev_chain, val, info);
1647 * call_netdevice_notifiers - call all network notifier blocks
1648 * @val: value passed unmodified to notifier function
1649 * @dev: net_device pointer passed unmodified to notifier function
1651 * Call all network notifier blocks. Parameters and return value
1652 * are as for raw_notifier_call_chain().
1655 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1657 struct netdev_notifier_info info;
1659 return call_netdevice_notifiers_info(val, dev, &info);
1661 EXPORT_SYMBOL(call_netdevice_notifiers);
1663 #ifdef CONFIG_NET_INGRESS
1664 static struct static_key ingress_needed __read_mostly;
1666 void net_inc_ingress_queue(void)
1668 static_key_slow_inc(&ingress_needed);
1670 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1672 void net_dec_ingress_queue(void)
1674 static_key_slow_dec(&ingress_needed);
1676 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1679 #ifdef CONFIG_NET_EGRESS
1680 static struct static_key egress_needed __read_mostly;
1682 void net_inc_egress_queue(void)
1684 static_key_slow_inc(&egress_needed);
1686 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1688 void net_dec_egress_queue(void)
1690 static_key_slow_dec(&egress_needed);
1692 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1695 static struct static_key netstamp_needed __read_mostly;
1696 #ifdef HAVE_JUMP_LABEL
1697 /* We are not allowed to call static_key_slow_dec() from irq context
1698 * If net_disable_timestamp() is called from irq context, defer the
1699 * static_key_slow_dec() calls.
1701 static atomic_t netstamp_needed_deferred;
1704 void net_enable_timestamp(void)
1706 #ifdef HAVE_JUMP_LABEL
1707 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1711 static_key_slow_dec(&netstamp_needed);
1715 static_key_slow_inc(&netstamp_needed);
1717 EXPORT_SYMBOL(net_enable_timestamp);
1719 void net_disable_timestamp(void)
1721 #ifdef HAVE_JUMP_LABEL
1722 if (in_interrupt()) {
1723 atomic_inc(&netstamp_needed_deferred);
1727 static_key_slow_dec(&netstamp_needed);
1729 EXPORT_SYMBOL(net_disable_timestamp);
1731 static inline void net_timestamp_set(struct sk_buff *skb)
1733 skb->tstamp.tv64 = 0;
1734 if (static_key_false(&netstamp_needed))
1735 __net_timestamp(skb);
1738 #define net_timestamp_check(COND, SKB) \
1739 if (static_key_false(&netstamp_needed)) { \
1740 if ((COND) && !(SKB)->tstamp.tv64) \
1741 __net_timestamp(SKB); \
1744 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1748 if (!(dev->flags & IFF_UP))
1751 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1752 if (skb->len <= len)
1755 /* if TSO is enabled, we don't care about the length as the packet
1756 * could be forwarded without being segmented before
1758 if (skb_is_gso(skb))
1763 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1765 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1767 if (skb_orphan_frags(skb, GFP_ATOMIC) ||
1768 unlikely(!is_skb_forwardable(dev, skb))) {
1769 atomic_long_inc(&dev->rx_dropped);
1774 skb_scrub_packet(skb, true);
1776 skb->protocol = eth_type_trans(skb, dev);
1777 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1781 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1784 * dev_forward_skb - loopback an skb to another netif
1786 * @dev: destination network device
1787 * @skb: buffer to forward
1790 * NET_RX_SUCCESS (no congestion)
1791 * NET_RX_DROP (packet was dropped, but freed)
1793 * dev_forward_skb can be used for injecting an skb from the
1794 * start_xmit function of one device into the receive queue
1795 * of another device.
1797 * The receiving device may be in another namespace, so
1798 * we have to clear all information in the skb that could
1799 * impact namespace isolation.
1801 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1803 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1805 EXPORT_SYMBOL_GPL(dev_forward_skb);
1807 static inline int deliver_skb(struct sk_buff *skb,
1808 struct packet_type *pt_prev,
1809 struct net_device *orig_dev)
1811 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1813 atomic_inc(&skb->users);
1814 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1817 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1818 struct packet_type **pt,
1819 struct net_device *orig_dev,
1821 struct list_head *ptype_list)
1823 struct packet_type *ptype, *pt_prev = *pt;
1825 list_for_each_entry_rcu(ptype, ptype_list, list) {
1826 if (ptype->type != type)
1829 deliver_skb(skb, pt_prev, orig_dev);
1835 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1837 if (!ptype->af_packet_priv || !skb->sk)
1840 if (ptype->id_match)
1841 return ptype->id_match(ptype, skb->sk);
1842 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1849 * Support routine. Sends outgoing frames to any network
1850 * taps currently in use.
1853 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1855 struct packet_type *ptype;
1856 struct sk_buff *skb2 = NULL;
1857 struct packet_type *pt_prev = NULL;
1858 struct list_head *ptype_list = &ptype_all;
1862 list_for_each_entry_rcu(ptype, ptype_list, list) {
1863 /* Never send packets back to the socket
1864 * they originated from - MvS (miquels@drinkel.ow.org)
1866 if (skb_loop_sk(ptype, skb))
1870 deliver_skb(skb2, pt_prev, skb->dev);
1875 /* need to clone skb, done only once */
1876 skb2 = skb_clone(skb, GFP_ATOMIC);
1880 net_timestamp_set(skb2);
1882 /* skb->nh should be correctly
1883 * set by sender, so that the second statement is
1884 * just protection against buggy protocols.
1886 skb_reset_mac_header(skb2);
1888 if (skb_network_header(skb2) < skb2->data ||
1889 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1890 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1891 ntohs(skb2->protocol),
1893 skb_reset_network_header(skb2);
1896 skb2->transport_header = skb2->network_header;
1897 skb2->pkt_type = PACKET_OUTGOING;
1901 if (ptype_list == &ptype_all) {
1902 ptype_list = &dev->ptype_all;
1907 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1910 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
1913 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1914 * @dev: Network device
1915 * @txq: number of queues available
1917 * If real_num_tx_queues is changed the tc mappings may no longer be
1918 * valid. To resolve this verify the tc mapping remains valid and if
1919 * not NULL the mapping. With no priorities mapping to this
1920 * offset/count pair it will no longer be used. In the worst case TC0
1921 * is invalid nothing can be done so disable priority mappings. If is
1922 * expected that drivers will fix this mapping if they can before
1923 * calling netif_set_real_num_tx_queues.
1925 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1928 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1930 /* If TC0 is invalidated disable TC mapping */
1931 if (tc->offset + tc->count > txq) {
1932 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1937 /* Invalidated prio to tc mappings set to TC0 */
1938 for (i = 1; i < TC_BITMASK + 1; i++) {
1939 int q = netdev_get_prio_tc_map(dev, i);
1941 tc = &dev->tc_to_txq[q];
1942 if (tc->offset + tc->count > txq) {
1943 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1945 netdev_set_prio_tc_map(dev, i, 0);
1951 static DEFINE_MUTEX(xps_map_mutex);
1952 #define xmap_dereference(P) \
1953 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1955 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1958 struct xps_map *map = NULL;
1962 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1964 for (pos = 0; map && pos < map->len; pos++) {
1965 if (map->queues[pos] == index) {
1967 map->queues[pos] = map->queues[--map->len];
1969 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1970 kfree_rcu(map, rcu);
1980 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1982 struct xps_dev_maps *dev_maps;
1984 bool active = false;
1986 mutex_lock(&xps_map_mutex);
1987 dev_maps = xmap_dereference(dev->xps_maps);
1992 for_each_possible_cpu(cpu) {
1993 for (i = index; i < dev->num_tx_queues; i++) {
1994 if (!remove_xps_queue(dev_maps, cpu, i))
1997 if (i == dev->num_tx_queues)
2002 RCU_INIT_POINTER(dev->xps_maps, NULL);
2003 kfree_rcu(dev_maps, rcu);
2006 for (i = index; i < dev->num_tx_queues; i++)
2007 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2011 mutex_unlock(&xps_map_mutex);
2014 static struct xps_map *expand_xps_map(struct xps_map *map,
2017 struct xps_map *new_map;
2018 int alloc_len = XPS_MIN_MAP_ALLOC;
2021 for (pos = 0; map && pos < map->len; pos++) {
2022 if (map->queues[pos] != index)
2027 /* Need to add queue to this CPU's existing map */
2029 if (pos < map->alloc_len)
2032 alloc_len = map->alloc_len * 2;
2035 /* Need to allocate new map to store queue on this CPU's map */
2036 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2041 for (i = 0; i < pos; i++)
2042 new_map->queues[i] = map->queues[i];
2043 new_map->alloc_len = alloc_len;
2049 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2052 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2053 struct xps_map *map, *new_map;
2054 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2055 int cpu, numa_node_id = -2;
2056 bool active = false;
2058 mutex_lock(&xps_map_mutex);
2060 dev_maps = xmap_dereference(dev->xps_maps);
2062 /* allocate memory for queue storage */
2063 for_each_online_cpu(cpu) {
2064 if (!cpumask_test_cpu(cpu, mask))
2068 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2069 if (!new_dev_maps) {
2070 mutex_unlock(&xps_map_mutex);
2074 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2077 map = expand_xps_map(map, cpu, index);
2081 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2085 goto out_no_new_maps;
2087 for_each_possible_cpu(cpu) {
2088 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2089 /* add queue to CPU maps */
2092 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2093 while ((pos < map->len) && (map->queues[pos] != index))
2096 if (pos == map->len)
2097 map->queues[map->len++] = index;
2099 if (numa_node_id == -2)
2100 numa_node_id = cpu_to_node(cpu);
2101 else if (numa_node_id != cpu_to_node(cpu))
2104 } else if (dev_maps) {
2105 /* fill in the new device map from the old device map */
2106 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2107 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2112 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2114 /* Cleanup old maps */
2116 for_each_possible_cpu(cpu) {
2117 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2118 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2119 if (map && map != new_map)
2120 kfree_rcu(map, rcu);
2123 kfree_rcu(dev_maps, rcu);
2126 dev_maps = new_dev_maps;
2130 /* update Tx queue numa node */
2131 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2132 (numa_node_id >= 0) ? numa_node_id :
2138 /* removes queue from unused CPUs */
2139 for_each_possible_cpu(cpu) {
2140 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2143 if (remove_xps_queue(dev_maps, cpu, index))
2147 /* free map if not active */
2149 RCU_INIT_POINTER(dev->xps_maps, NULL);
2150 kfree_rcu(dev_maps, rcu);
2154 mutex_unlock(&xps_map_mutex);
2158 /* remove any maps that we added */
2159 for_each_possible_cpu(cpu) {
2160 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2161 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2163 if (new_map && new_map != map)
2167 mutex_unlock(&xps_map_mutex);
2169 kfree(new_dev_maps);
2172 EXPORT_SYMBOL(netif_set_xps_queue);
2176 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2177 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2179 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2183 if (txq < 1 || txq > dev->num_tx_queues)
2186 if (dev->reg_state == NETREG_REGISTERED ||
2187 dev->reg_state == NETREG_UNREGISTERING) {
2190 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2196 netif_setup_tc(dev, txq);
2198 if (txq < dev->real_num_tx_queues) {
2199 qdisc_reset_all_tx_gt(dev, txq);
2201 netif_reset_xps_queues_gt(dev, txq);
2206 dev->real_num_tx_queues = txq;
2209 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2213 * netif_set_real_num_rx_queues - set actual number of RX queues used
2214 * @dev: Network device
2215 * @rxq: Actual number of RX queues
2217 * This must be called either with the rtnl_lock held or before
2218 * registration of the net device. Returns 0 on success, or a
2219 * negative error code. If called before registration, it always
2222 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2226 if (rxq < 1 || rxq > dev->num_rx_queues)
2229 if (dev->reg_state == NETREG_REGISTERED) {
2232 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2238 dev->real_num_rx_queues = rxq;
2241 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2245 * netif_get_num_default_rss_queues - default number of RSS queues
2247 * This routine should set an upper limit on the number of RSS queues
2248 * used by default by multiqueue devices.
2250 int netif_get_num_default_rss_queues(void)
2252 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2254 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2256 static void __netif_reschedule(struct Qdisc *q)
2258 struct softnet_data *sd;
2259 unsigned long flags;
2261 local_irq_save(flags);
2262 sd = this_cpu_ptr(&softnet_data);
2263 q->next_sched = NULL;
2264 *sd->output_queue_tailp = q;
2265 sd->output_queue_tailp = &q->next_sched;
2266 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2267 local_irq_restore(flags);
2270 void __netif_schedule(struct Qdisc *q)
2272 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2273 __netif_reschedule(q);
2275 EXPORT_SYMBOL(__netif_schedule);
2277 struct dev_kfree_skb_cb {
2278 enum skb_free_reason reason;
2281 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2283 return (struct dev_kfree_skb_cb *)skb->cb;
2286 void netif_schedule_queue(struct netdev_queue *txq)
2289 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2290 struct Qdisc *q = rcu_dereference(txq->qdisc);
2292 __netif_schedule(q);
2296 EXPORT_SYMBOL(netif_schedule_queue);
2299 * netif_wake_subqueue - allow sending packets on subqueue
2300 * @dev: network device
2301 * @queue_index: sub queue index
2303 * Resume individual transmit queue of a device with multiple transmit queues.
2305 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2307 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2309 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2313 q = rcu_dereference(txq->qdisc);
2314 __netif_schedule(q);
2318 EXPORT_SYMBOL(netif_wake_subqueue);
2320 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2322 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2326 q = rcu_dereference(dev_queue->qdisc);
2327 __netif_schedule(q);
2331 EXPORT_SYMBOL(netif_tx_wake_queue);
2333 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2335 unsigned long flags;
2337 if (likely(atomic_read(&skb->users) == 1)) {
2339 atomic_set(&skb->users, 0);
2340 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2343 get_kfree_skb_cb(skb)->reason = reason;
2344 local_irq_save(flags);
2345 skb->next = __this_cpu_read(softnet_data.completion_queue);
2346 __this_cpu_write(softnet_data.completion_queue, skb);
2347 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2348 local_irq_restore(flags);
2350 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2352 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2354 if (in_irq() || irqs_disabled())
2355 __dev_kfree_skb_irq(skb, reason);
2359 EXPORT_SYMBOL(__dev_kfree_skb_any);
2363 * netif_device_detach - mark device as removed
2364 * @dev: network device
2366 * Mark device as removed from system and therefore no longer available.
2368 void netif_device_detach(struct net_device *dev)
2370 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2371 netif_running(dev)) {
2372 netif_tx_stop_all_queues(dev);
2375 EXPORT_SYMBOL(netif_device_detach);
2378 * netif_device_attach - mark device as attached
2379 * @dev: network device
2381 * Mark device as attached from system and restart if needed.
2383 void netif_device_attach(struct net_device *dev)
2385 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2386 netif_running(dev)) {
2387 netif_tx_wake_all_queues(dev);
2388 __netdev_watchdog_up(dev);
2391 EXPORT_SYMBOL(netif_device_attach);
2394 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2395 * to be used as a distribution range.
2397 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2398 unsigned int num_tx_queues)
2402 u16 qcount = num_tx_queues;
2404 if (skb_rx_queue_recorded(skb)) {
2405 hash = skb_get_rx_queue(skb);
2406 while (unlikely(hash >= num_tx_queues))
2407 hash -= num_tx_queues;
2412 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2413 qoffset = dev->tc_to_txq[tc].offset;
2414 qcount = dev->tc_to_txq[tc].count;
2417 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2419 EXPORT_SYMBOL(__skb_tx_hash);
2421 static void skb_warn_bad_offload(const struct sk_buff *skb)
2423 static const netdev_features_t null_features = 0;
2424 struct net_device *dev = skb->dev;
2425 const char *name = "";
2427 if (!net_ratelimit())
2431 if (dev->dev.parent)
2432 name = dev_driver_string(dev->dev.parent);
2434 name = netdev_name(dev);
2436 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2437 "gso_type=%d ip_summed=%d\n",
2438 name, dev ? &dev->features : &null_features,
2439 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2440 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2441 skb_shinfo(skb)->gso_type, skb->ip_summed);
2445 * Invalidate hardware checksum when packet is to be mangled, and
2446 * complete checksum manually on outgoing path.
2448 int skb_checksum_help(struct sk_buff *skb)
2451 int ret = 0, offset;
2453 if (skb->ip_summed == CHECKSUM_COMPLETE)
2454 goto out_set_summed;
2456 if (unlikely(skb_shinfo(skb)->gso_size)) {
2457 skb_warn_bad_offload(skb);
2461 /* Before computing a checksum, we should make sure no frag could
2462 * be modified by an external entity : checksum could be wrong.
2464 if (skb_has_shared_frag(skb)) {
2465 ret = __skb_linearize(skb);
2470 offset = skb_checksum_start_offset(skb);
2471 BUG_ON(offset >= skb_headlen(skb));
2472 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2474 offset += skb->csum_offset;
2475 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2477 if (skb_cloned(skb) &&
2478 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2479 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2484 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2486 skb->ip_summed = CHECKSUM_NONE;
2490 EXPORT_SYMBOL(skb_checksum_help);
2492 /* skb_csum_offload_check - Driver helper function to determine if a device
2493 * with limited checksum offload capabilities is able to offload the checksum
2494 * for a given packet.
2497 * skb - sk_buff for the packet in question
2498 * spec - contains the description of what device can offload
2499 * csum_encapped - returns true if the checksum being offloaded is
2500 * encpasulated. That is it is checksum for the transport header
2501 * in the inner headers.
2502 * checksum_help - when set indicates that helper function should
2503 * call skb_checksum_help if offload checks fail
2506 * true: Packet has passed the checksum checks and should be offloadable to
2507 * the device (a driver may still need to check for additional
2508 * restrictions of its device)
2509 * false: Checksum is not offloadable. If checksum_help was set then
2510 * skb_checksum_help was called to resolve checksum for non-GSO
2511 * packets and when IP protocol is not SCTP
2513 bool __skb_csum_offload_chk(struct sk_buff *skb,
2514 const struct skb_csum_offl_spec *spec,
2515 bool *csum_encapped,
2519 struct ipv6hdr *ipv6;
2524 if (skb->protocol == htons(ETH_P_8021Q) ||
2525 skb->protocol == htons(ETH_P_8021AD)) {
2526 if (!spec->vlan_okay)
2530 /* We check whether the checksum refers to a transport layer checksum in
2531 * the outermost header or an encapsulated transport layer checksum that
2532 * corresponds to the inner headers of the skb. If the checksum is for
2533 * something else in the packet we need help.
2535 if (skb_checksum_start_offset(skb) == skb_transport_offset(skb)) {
2536 /* Non-encapsulated checksum */
2537 protocol = eproto_to_ipproto(vlan_get_protocol(skb));
2538 nhdr = skb_network_header(skb);
2539 *csum_encapped = false;
2540 if (spec->no_not_encapped)
2542 } else if (skb->encapsulation && spec->encap_okay &&
2543 skb_checksum_start_offset(skb) ==
2544 skb_inner_transport_offset(skb)) {
2545 /* Encapsulated checksum */
2546 *csum_encapped = true;
2547 switch (skb->inner_protocol_type) {
2548 case ENCAP_TYPE_ETHER:
2549 protocol = eproto_to_ipproto(skb->inner_protocol);
2551 case ENCAP_TYPE_IPPROTO:
2552 protocol = skb->inner_protocol;
2555 nhdr = skb_inner_network_header(skb);
2562 if (!spec->ipv4_okay)
2565 ip_proto = iph->protocol;
2566 if (iph->ihl != 5 && !spec->ip_options_okay)
2570 if (!spec->ipv6_okay)
2572 if (spec->no_encapped_ipv6 && *csum_encapped)
2575 nhdr += sizeof(*ipv6);
2576 ip_proto = ipv6->nexthdr;
2585 if (!spec->tcp_okay ||
2586 skb->csum_offset != offsetof(struct tcphdr, check))
2590 if (!spec->udp_okay ||
2591 skb->csum_offset != offsetof(struct udphdr, check))
2595 if (!spec->sctp_okay ||
2596 skb->csum_offset != offsetof(struct sctphdr, checksum))
2600 case NEXTHDR_ROUTING:
2601 case NEXTHDR_DEST: {
2604 if (protocol != IPPROTO_IPV6 || !spec->ext_hdrs_okay)
2607 ip_proto = opthdr[0];
2608 nhdr += (opthdr[1] + 1) << 3;
2610 goto ip_proto_again;
2616 /* Passed the tests for offloading checksum */
2620 if (csum_help && !skb_shinfo(skb)->gso_size)
2621 skb_checksum_help(skb);
2625 EXPORT_SYMBOL(__skb_csum_offload_chk);
2627 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2629 __be16 type = skb->protocol;
2631 /* Tunnel gso handlers can set protocol to ethernet. */
2632 if (type == htons(ETH_P_TEB)) {
2635 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2638 eth = (struct ethhdr *)skb_mac_header(skb);
2639 type = eth->h_proto;
2642 return __vlan_get_protocol(skb, type, depth);
2646 * skb_mac_gso_segment - mac layer segmentation handler.
2647 * @skb: buffer to segment
2648 * @features: features for the output path (see dev->features)
2650 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2651 netdev_features_t features)
2653 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2654 struct packet_offload *ptype;
2655 int vlan_depth = skb->mac_len;
2656 __be16 type = skb_network_protocol(skb, &vlan_depth);
2658 if (unlikely(!type))
2659 return ERR_PTR(-EINVAL);
2661 __skb_pull(skb, vlan_depth);
2664 list_for_each_entry_rcu(ptype, &offload_base, list) {
2665 if (ptype->type == type && ptype->callbacks.gso_segment) {
2666 segs = ptype->callbacks.gso_segment(skb, features);
2672 __skb_push(skb, skb->data - skb_mac_header(skb));
2676 EXPORT_SYMBOL(skb_mac_gso_segment);
2679 /* openvswitch calls this on rx path, so we need a different check.
2681 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2684 return skb->ip_summed != CHECKSUM_PARTIAL;
2686 return skb->ip_summed == CHECKSUM_NONE;
2690 * __skb_gso_segment - Perform segmentation on skb.
2691 * @skb: buffer to segment
2692 * @features: features for the output path (see dev->features)
2693 * @tx_path: whether it is called in TX path
2695 * This function segments the given skb and returns a list of segments.
2697 * It may return NULL if the skb requires no segmentation. This is
2698 * only possible when GSO is used for verifying header integrity.
2700 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2702 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2703 netdev_features_t features, bool tx_path)
2705 if (unlikely(skb_needs_check(skb, tx_path))) {
2708 skb_warn_bad_offload(skb);
2710 err = skb_cow_head(skb, 0);
2712 return ERR_PTR(err);
2715 /* Only report GSO partial support if it will enable us to
2716 * support segmentation on this frame without needing additional
2719 if (features & NETIF_F_GSO_PARTIAL) {
2720 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
2721 struct net_device *dev = skb->dev;
2723 partial_features |= dev->features & dev->gso_partial_features;
2724 if (!skb_gso_ok(skb, features | partial_features))
2725 features &= ~NETIF_F_GSO_PARTIAL;
2728 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2729 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2731 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2732 SKB_GSO_CB(skb)->encap_level = 0;
2734 skb_reset_mac_header(skb);
2735 skb_reset_mac_len(skb);
2737 return skb_mac_gso_segment(skb, features);
2739 EXPORT_SYMBOL(__skb_gso_segment);
2741 /* Take action when hardware reception checksum errors are detected. */
2743 void netdev_rx_csum_fault(struct net_device *dev)
2745 if (net_ratelimit()) {
2746 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2750 EXPORT_SYMBOL(netdev_rx_csum_fault);
2753 /* Actually, we should eliminate this check as soon as we know, that:
2754 * 1. IOMMU is present and allows to map all the memory.
2755 * 2. No high memory really exists on this machine.
2758 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2760 #ifdef CONFIG_HIGHMEM
2762 if (!(dev->features & NETIF_F_HIGHDMA)) {
2763 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2764 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2765 if (PageHighMem(skb_frag_page(frag)))
2770 if (PCI_DMA_BUS_IS_PHYS) {
2771 struct device *pdev = dev->dev.parent;
2775 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2776 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2777 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2778 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2786 /* If MPLS offload request, verify we are testing hardware MPLS features
2787 * instead of standard features for the netdev.
2789 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2790 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2791 netdev_features_t features,
2794 if (eth_p_mpls(type))
2795 features &= skb->dev->mpls_features;
2800 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2801 netdev_features_t features,
2808 static netdev_features_t harmonize_features(struct sk_buff *skb,
2809 netdev_features_t features)
2814 type = skb_network_protocol(skb, &tmp);
2815 features = net_mpls_features(skb, features, type);
2817 if (skb->ip_summed != CHECKSUM_NONE &&
2818 !can_checksum_protocol(features, type)) {
2819 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2820 } else if (illegal_highdma(skb->dev, skb)) {
2821 features &= ~NETIF_F_SG;
2827 netdev_features_t passthru_features_check(struct sk_buff *skb,
2828 struct net_device *dev,
2829 netdev_features_t features)
2833 EXPORT_SYMBOL(passthru_features_check);
2835 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2836 struct net_device *dev,
2837 netdev_features_t features)
2839 return vlan_features_check(skb, features);
2842 static netdev_features_t gso_features_check(const struct sk_buff *skb,
2843 struct net_device *dev,
2844 netdev_features_t features)
2846 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2848 if (gso_segs > dev->gso_max_segs)
2849 return features & ~NETIF_F_GSO_MASK;
2851 /* Support for GSO partial features requires software
2852 * intervention before we can actually process the packets
2853 * so we need to strip support for any partial features now
2854 * and we can pull them back in after we have partially
2855 * segmented the frame.
2857 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
2858 features &= ~dev->gso_partial_features;
2860 /* Make sure to clear the IPv4 ID mangling feature if the
2861 * IPv4 header has the potential to be fragmented.
2863 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2864 struct iphdr *iph = skb->encapsulation ?
2865 inner_ip_hdr(skb) : ip_hdr(skb);
2867 if (!(iph->frag_off & htons(IP_DF)))
2868 features &= ~NETIF_F_TSO_MANGLEID;
2874 netdev_features_t netif_skb_features(struct sk_buff *skb)
2876 struct net_device *dev = skb->dev;
2877 netdev_features_t features = dev->features;
2879 if (skb_is_gso(skb))
2880 features = gso_features_check(skb, dev, features);
2882 /* If encapsulation offload request, verify we are testing
2883 * hardware encapsulation features instead of standard
2884 * features for the netdev
2886 if (skb->encapsulation)
2887 features &= dev->hw_enc_features;
2889 if (skb_vlan_tagged(skb))
2890 features = netdev_intersect_features(features,
2891 dev->vlan_features |
2892 NETIF_F_HW_VLAN_CTAG_TX |
2893 NETIF_F_HW_VLAN_STAG_TX);
2895 if (dev->netdev_ops->ndo_features_check)
2896 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2899 features &= dflt_features_check(skb, dev, features);
2901 return harmonize_features(skb, features);
2903 EXPORT_SYMBOL(netif_skb_features);
2905 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2906 struct netdev_queue *txq, bool more)
2911 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2912 dev_queue_xmit_nit(skb, dev);
2915 trace_net_dev_start_xmit(skb, dev);
2916 rc = netdev_start_xmit(skb, dev, txq, more);
2917 trace_net_dev_xmit(skb, rc, dev, len);
2922 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2923 struct netdev_queue *txq, int *ret)
2925 struct sk_buff *skb = first;
2926 int rc = NETDEV_TX_OK;
2929 struct sk_buff *next = skb->next;
2932 rc = xmit_one(skb, dev, txq, next != NULL);
2933 if (unlikely(!dev_xmit_complete(rc))) {
2939 if (netif_xmit_stopped(txq) && skb) {
2940 rc = NETDEV_TX_BUSY;
2950 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2951 netdev_features_t features)
2953 if (skb_vlan_tag_present(skb) &&
2954 !vlan_hw_offload_capable(features, skb->vlan_proto))
2955 skb = __vlan_hwaccel_push_inside(skb);
2959 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2961 netdev_features_t features;
2963 features = netif_skb_features(skb);
2964 skb = validate_xmit_vlan(skb, features);
2968 if (netif_needs_gso(skb, features)) {
2969 struct sk_buff *segs;
2971 segs = skb_gso_segment(skb, features);
2979 if (skb_needs_linearize(skb, features) &&
2980 __skb_linearize(skb))
2983 /* If packet is not checksummed and device does not
2984 * support checksumming for this protocol, complete
2985 * checksumming here.
2987 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2988 if (skb->encapsulation)
2989 skb_set_inner_transport_header(skb,
2990 skb_checksum_start_offset(skb));
2992 skb_set_transport_header(skb,
2993 skb_checksum_start_offset(skb));
2994 if (!(features & NETIF_F_CSUM_MASK) &&
2995 skb_checksum_help(skb))
3005 atomic_long_inc(&dev->tx_dropped);
3009 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
3011 struct sk_buff *next, *head = NULL, *tail;
3013 for (; skb != NULL; skb = next) {
3017 /* in case skb wont be segmented, point to itself */
3020 skb = validate_xmit_skb(skb, dev);
3028 /* If skb was segmented, skb->prev points to
3029 * the last segment. If not, it still contains skb.
3036 static void qdisc_pkt_len_init(struct sk_buff *skb)
3038 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3040 qdisc_skb_cb(skb)->pkt_len = skb->len;
3042 /* To get more precise estimation of bytes sent on wire,
3043 * we add to pkt_len the headers size of all segments
3045 if (shinfo->gso_size) {
3046 unsigned int hdr_len;
3047 u16 gso_segs = shinfo->gso_segs;
3049 /* mac layer + network layer */
3050 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3052 /* + transport layer */
3053 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
3054 hdr_len += tcp_hdrlen(skb);
3056 hdr_len += sizeof(struct udphdr);
3058 if (shinfo->gso_type & SKB_GSO_DODGY)
3059 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3062 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3066 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3067 struct net_device *dev,
3068 struct netdev_queue *txq)
3070 spinlock_t *root_lock = qdisc_lock(q);
3074 qdisc_calculate_pkt_len(skb, q);
3076 * Heuristic to force contended enqueues to serialize on a
3077 * separate lock before trying to get qdisc main lock.
3078 * This permits __QDISC___STATE_RUNNING owner to get the lock more
3079 * often and dequeue packets faster.
3081 contended = qdisc_is_running(q);
3082 if (unlikely(contended))
3083 spin_lock(&q->busylock);
3085 spin_lock(root_lock);
3086 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3089 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3090 qdisc_run_begin(q)) {
3092 * This is a work-conserving queue; there are no old skbs
3093 * waiting to be sent out; and the qdisc is not running -
3094 * xmit the skb directly.
3097 qdisc_bstats_update(q, skb);
3099 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3100 if (unlikely(contended)) {
3101 spin_unlock(&q->busylock);
3108 rc = NET_XMIT_SUCCESS;
3110 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
3111 if (qdisc_run_begin(q)) {
3112 if (unlikely(contended)) {
3113 spin_unlock(&q->busylock);
3119 spin_unlock(root_lock);
3120 if (unlikely(contended))
3121 spin_unlock(&q->busylock);
3125 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3126 static void skb_update_prio(struct sk_buff *skb)
3128 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
3130 if (!skb->priority && skb->sk && map) {
3131 unsigned int prioidx =
3132 sock_cgroup_prioidx(&skb->sk->sk_cgrp_data);
3134 if (prioidx < map->priomap_len)
3135 skb->priority = map->priomap[prioidx];
3139 #define skb_update_prio(skb)
3142 DEFINE_PER_CPU(int, xmit_recursion);
3143 EXPORT_SYMBOL(xmit_recursion);
3145 #define RECURSION_LIMIT 10
3148 * dev_loopback_xmit - loop back @skb
3149 * @net: network namespace this loopback is happening in
3150 * @sk: sk needed to be a netfilter okfn
3151 * @skb: buffer to transmit
3153 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3155 skb_reset_mac_header(skb);
3156 __skb_pull(skb, skb_network_offset(skb));
3157 skb->pkt_type = PACKET_LOOPBACK;
3158 skb->ip_summed = CHECKSUM_UNNECESSARY;
3159 WARN_ON(!skb_dst(skb));
3164 EXPORT_SYMBOL(dev_loopback_xmit);
3166 #ifdef CONFIG_NET_EGRESS
3167 static struct sk_buff *
3168 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3170 struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list);
3171 struct tcf_result cl_res;
3176 /* skb->tc_verd and qdisc_skb_cb(skb)->pkt_len were already set
3177 * earlier by the caller.
3179 qdisc_bstats_cpu_update(cl->q, skb);
3181 switch (tc_classify(skb, cl, &cl_res, false)) {
3183 case TC_ACT_RECLASSIFY:
3184 skb->tc_index = TC_H_MIN(cl_res.classid);
3187 qdisc_qstats_cpu_drop(cl->q);
3188 *ret = NET_XMIT_DROP;
3193 *ret = NET_XMIT_SUCCESS;
3196 case TC_ACT_REDIRECT:
3197 /* No need to push/pop skb's mac_header here on egress! */
3198 skb_do_redirect(skb);
3199 *ret = NET_XMIT_SUCCESS;
3207 #endif /* CONFIG_NET_EGRESS */
3209 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3212 struct xps_dev_maps *dev_maps;
3213 struct xps_map *map;
3214 int queue_index = -1;
3217 dev_maps = rcu_dereference(dev->xps_maps);
3219 map = rcu_dereference(
3220 dev_maps->cpu_map[skb->sender_cpu - 1]);
3223 queue_index = map->queues[0];
3225 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3227 if (unlikely(queue_index >= dev->real_num_tx_queues))
3239 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3241 struct sock *sk = skb->sk;
3242 int queue_index = sk_tx_queue_get(sk);
3244 if (queue_index < 0 || skb->ooo_okay ||
3245 queue_index >= dev->real_num_tx_queues) {
3246 int new_index = get_xps_queue(dev, skb);
3248 new_index = skb_tx_hash(dev, skb);
3250 if (queue_index != new_index && sk &&
3252 rcu_access_pointer(sk->sk_dst_cache))
3253 sk_tx_queue_set(sk, new_index);
3255 queue_index = new_index;
3261 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3262 struct sk_buff *skb,
3265 int queue_index = 0;
3268 u32 sender_cpu = skb->sender_cpu - 1;
3270 if (sender_cpu >= (u32)NR_CPUS)
3271 skb->sender_cpu = raw_smp_processor_id() + 1;
3274 if (dev->real_num_tx_queues != 1) {
3275 const struct net_device_ops *ops = dev->netdev_ops;
3276 if (ops->ndo_select_queue)
3277 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3280 queue_index = __netdev_pick_tx(dev, skb);
3283 queue_index = netdev_cap_txqueue(dev, queue_index);
3286 skb_set_queue_mapping(skb, queue_index);
3287 return netdev_get_tx_queue(dev, queue_index);
3291 * __dev_queue_xmit - transmit a buffer
3292 * @skb: buffer to transmit
3293 * @accel_priv: private data used for L2 forwarding offload
3295 * Queue a buffer for transmission to a network device. The caller must
3296 * have set the device and priority and built the buffer before calling
3297 * this function. The function can be called from an interrupt.
3299 * A negative errno code is returned on a failure. A success does not
3300 * guarantee the frame will be transmitted as it may be dropped due
3301 * to congestion or traffic shaping.
3303 * -----------------------------------------------------------------------------------
3304 * I notice this method can also return errors from the queue disciplines,
3305 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3308 * Regardless of the return value, the skb is consumed, so it is currently
3309 * difficult to retry a send to this method. (You can bump the ref count
3310 * before sending to hold a reference for retry if you are careful.)
3312 * When calling this method, interrupts MUST be enabled. This is because
3313 * the BH enable code must have IRQs enabled so that it will not deadlock.
3316 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3318 struct net_device *dev = skb->dev;
3319 struct netdev_queue *txq;
3323 skb_reset_mac_header(skb);
3325 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3326 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3328 /* Disable soft irqs for various locks below. Also
3329 * stops preemption for RCU.
3333 skb_update_prio(skb);
3335 qdisc_pkt_len_init(skb);
3336 #ifdef CONFIG_NET_CLS_ACT
3337 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3338 # ifdef CONFIG_NET_EGRESS
3339 if (static_key_false(&egress_needed)) {
3340 skb = sch_handle_egress(skb, &rc, dev);
3346 /* If device/qdisc don't need skb->dst, release it right now while
3347 * its hot in this cpu cache.
3349 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3354 #ifdef CONFIG_NET_SWITCHDEV
3355 /* Don't forward if offload device already forwarded */
3356 if (skb->offload_fwd_mark &&
3357 skb->offload_fwd_mark == dev->offload_fwd_mark) {
3359 rc = NET_XMIT_SUCCESS;
3364 txq = netdev_pick_tx(dev, skb, accel_priv);
3365 q = rcu_dereference_bh(txq->qdisc);
3367 trace_net_dev_queue(skb);
3369 rc = __dev_xmit_skb(skb, q, dev, txq);
3373 /* The device has no queue. Common case for software devices:
3374 loopback, all the sorts of tunnels...
3376 Really, it is unlikely that netif_tx_lock protection is necessary
3377 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3379 However, it is possible, that they rely on protection
3382 Check this and shot the lock. It is not prone from deadlocks.
3383 Either shot noqueue qdisc, it is even simpler 8)
3385 if (dev->flags & IFF_UP) {
3386 int cpu = smp_processor_id(); /* ok because BHs are off */
3388 if (txq->xmit_lock_owner != cpu) {
3390 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
3391 goto recursion_alert;
3393 skb = validate_xmit_skb(skb, dev);
3397 HARD_TX_LOCK(dev, txq, cpu);
3399 if (!netif_xmit_stopped(txq)) {
3400 __this_cpu_inc(xmit_recursion);
3401 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3402 __this_cpu_dec(xmit_recursion);
3403 if (dev_xmit_complete(rc)) {
3404 HARD_TX_UNLOCK(dev, txq);
3408 HARD_TX_UNLOCK(dev, txq);
3409 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3412 /* Recursion is detected! It is possible,
3416 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3422 rcu_read_unlock_bh();
3424 atomic_long_inc(&dev->tx_dropped);
3425 kfree_skb_list(skb);
3428 rcu_read_unlock_bh();
3432 int dev_queue_xmit(struct sk_buff *skb)
3434 return __dev_queue_xmit(skb, NULL);
3436 EXPORT_SYMBOL(dev_queue_xmit);
3438 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3440 return __dev_queue_xmit(skb, accel_priv);
3442 EXPORT_SYMBOL(dev_queue_xmit_accel);
3445 /*=======================================================================
3447 =======================================================================*/
3449 int netdev_max_backlog __read_mostly = 1000;
3450 EXPORT_SYMBOL(netdev_max_backlog);
3452 int netdev_tstamp_prequeue __read_mostly = 1;
3453 int netdev_budget __read_mostly = 300;
3454 int weight_p __read_mostly = 64; /* old backlog weight */
3456 /* Called with irq disabled */
3457 static inline void ____napi_schedule(struct softnet_data *sd,
3458 struct napi_struct *napi)
3460 list_add_tail(&napi->poll_list, &sd->poll_list);
3461 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3466 /* One global table that all flow-based protocols share. */
3467 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3468 EXPORT_SYMBOL(rps_sock_flow_table);
3469 u32 rps_cpu_mask __read_mostly;
3470 EXPORT_SYMBOL(rps_cpu_mask);
3472 struct static_key rps_needed __read_mostly;
3473 EXPORT_SYMBOL(rps_needed);
3475 static struct rps_dev_flow *
3476 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3477 struct rps_dev_flow *rflow, u16 next_cpu)
3479 if (next_cpu < nr_cpu_ids) {
3480 #ifdef CONFIG_RFS_ACCEL
3481 struct netdev_rx_queue *rxqueue;
3482 struct rps_dev_flow_table *flow_table;
3483 struct rps_dev_flow *old_rflow;
3488 /* Should we steer this flow to a different hardware queue? */
3489 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3490 !(dev->features & NETIF_F_NTUPLE))
3492 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3493 if (rxq_index == skb_get_rx_queue(skb))
3496 rxqueue = dev->_rx + rxq_index;
3497 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3500 flow_id = skb_get_hash(skb) & flow_table->mask;
3501 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3502 rxq_index, flow_id);
3506 rflow = &flow_table->flows[flow_id];
3508 if (old_rflow->filter == rflow->filter)
3509 old_rflow->filter = RPS_NO_FILTER;
3513 per_cpu(softnet_data, next_cpu).input_queue_head;
3516 rflow->cpu = next_cpu;
3521 * get_rps_cpu is called from netif_receive_skb and returns the target
3522 * CPU from the RPS map of the receiving queue for a given skb.
3523 * rcu_read_lock must be held on entry.
3525 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3526 struct rps_dev_flow **rflowp)
3528 const struct rps_sock_flow_table *sock_flow_table;
3529 struct netdev_rx_queue *rxqueue = dev->_rx;
3530 struct rps_dev_flow_table *flow_table;
3531 struct rps_map *map;
3536 if (skb_rx_queue_recorded(skb)) {
3537 u16 index = skb_get_rx_queue(skb);
3539 if (unlikely(index >= dev->real_num_rx_queues)) {
3540 WARN_ONCE(dev->real_num_rx_queues > 1,
3541 "%s received packet on queue %u, but number "
3542 "of RX queues is %u\n",
3543 dev->name, index, dev->real_num_rx_queues);
3549 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3551 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3552 map = rcu_dereference(rxqueue->rps_map);
3553 if (!flow_table && !map)
3556 skb_reset_network_header(skb);
3557 hash = skb_get_hash(skb);
3561 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3562 if (flow_table && sock_flow_table) {
3563 struct rps_dev_flow *rflow;
3567 /* First check into global flow table if there is a match */
3568 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3569 if ((ident ^ hash) & ~rps_cpu_mask)
3572 next_cpu = ident & rps_cpu_mask;
3574 /* OK, now we know there is a match,
3575 * we can look at the local (per receive queue) flow table
3577 rflow = &flow_table->flows[hash & flow_table->mask];
3581 * If the desired CPU (where last recvmsg was done) is
3582 * different from current CPU (one in the rx-queue flow
3583 * table entry), switch if one of the following holds:
3584 * - Current CPU is unset (>= nr_cpu_ids).
3585 * - Current CPU is offline.
3586 * - The current CPU's queue tail has advanced beyond the
3587 * last packet that was enqueued using this table entry.
3588 * This guarantees that all previous packets for the flow
3589 * have been dequeued, thus preserving in order delivery.
3591 if (unlikely(tcpu != next_cpu) &&
3592 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3593 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3594 rflow->last_qtail)) >= 0)) {
3596 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3599 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3609 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3610 if (cpu_online(tcpu)) {
3620 #ifdef CONFIG_RFS_ACCEL
3623 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3624 * @dev: Device on which the filter was set
3625 * @rxq_index: RX queue index
3626 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3627 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3629 * Drivers that implement ndo_rx_flow_steer() should periodically call
3630 * this function for each installed filter and remove the filters for
3631 * which it returns %true.
3633 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3634 u32 flow_id, u16 filter_id)
3636 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3637 struct rps_dev_flow_table *flow_table;
3638 struct rps_dev_flow *rflow;
3643 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3644 if (flow_table && flow_id <= flow_table->mask) {
3645 rflow = &flow_table->flows[flow_id];
3646 cpu = ACCESS_ONCE(rflow->cpu);
3647 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3648 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3649 rflow->last_qtail) <
3650 (int)(10 * flow_table->mask)))
3656 EXPORT_SYMBOL(rps_may_expire_flow);
3658 #endif /* CONFIG_RFS_ACCEL */
3660 /* Called from hardirq (IPI) context */
3661 static void rps_trigger_softirq(void *data)
3663 struct softnet_data *sd = data;
3665 ____napi_schedule(sd, &sd->backlog);
3669 #endif /* CONFIG_RPS */
3672 * Check if this softnet_data structure is another cpu one
3673 * If yes, queue it to our IPI list and return 1
3676 static int rps_ipi_queued(struct softnet_data *sd)
3679 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3682 sd->rps_ipi_next = mysd->rps_ipi_list;
3683 mysd->rps_ipi_list = sd;
3685 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3688 #endif /* CONFIG_RPS */
3692 #ifdef CONFIG_NET_FLOW_LIMIT
3693 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3696 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3698 #ifdef CONFIG_NET_FLOW_LIMIT
3699 struct sd_flow_limit *fl;
3700 struct softnet_data *sd;
3701 unsigned int old_flow, new_flow;
3703 if (qlen < (netdev_max_backlog >> 1))
3706 sd = this_cpu_ptr(&softnet_data);
3709 fl = rcu_dereference(sd->flow_limit);
3711 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3712 old_flow = fl->history[fl->history_head];
3713 fl->history[fl->history_head] = new_flow;
3716 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3718 if (likely(fl->buckets[old_flow]))
3719 fl->buckets[old_flow]--;
3721 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3733 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3734 * queue (may be a remote CPU queue).
3736 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3737 unsigned int *qtail)
3739 struct softnet_data *sd;
3740 unsigned long flags;
3743 sd = &per_cpu(softnet_data, cpu);
3745 local_irq_save(flags);
3748 if (!netif_running(skb->dev))
3750 qlen = skb_queue_len(&sd->input_pkt_queue);
3751 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3754 __skb_queue_tail(&sd->input_pkt_queue, skb);
3755 input_queue_tail_incr_save(sd, qtail);
3757 local_irq_restore(flags);
3758 return NET_RX_SUCCESS;
3761 /* Schedule NAPI for backlog device
3762 * We can use non atomic operation since we own the queue lock
3764 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3765 if (!rps_ipi_queued(sd))
3766 ____napi_schedule(sd, &sd->backlog);
3775 local_irq_restore(flags);
3777 atomic_long_inc(&skb->dev->rx_dropped);
3782 static int netif_rx_internal(struct sk_buff *skb)
3786 net_timestamp_check(netdev_tstamp_prequeue, skb);
3788 trace_netif_rx(skb);
3790 if (static_key_false(&rps_needed)) {
3791 struct rps_dev_flow voidflow, *rflow = &voidflow;
3797 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3799 cpu = smp_processor_id();
3801 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3809 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3816 * netif_rx - post buffer to the network code
3817 * @skb: buffer to post
3819 * This function receives a packet from a device driver and queues it for
3820 * the upper (protocol) levels to process. It always succeeds. The buffer
3821 * may be dropped during processing for congestion control or by the
3825 * NET_RX_SUCCESS (no congestion)
3826 * NET_RX_DROP (packet was dropped)
3830 int netif_rx(struct sk_buff *skb)
3832 trace_netif_rx_entry(skb);
3834 return netif_rx_internal(skb);
3836 EXPORT_SYMBOL(netif_rx);
3838 int netif_rx_ni(struct sk_buff *skb)
3842 trace_netif_rx_ni_entry(skb);
3845 err = netif_rx_internal(skb);
3846 if (local_softirq_pending())
3852 EXPORT_SYMBOL(netif_rx_ni);
3854 static void net_tx_action(struct softirq_action *h)
3856 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3858 if (sd->completion_queue) {
3859 struct sk_buff *clist;
3861 local_irq_disable();
3862 clist = sd->completion_queue;
3863 sd->completion_queue = NULL;
3867 struct sk_buff *skb = clist;
3868 clist = clist->next;
3870 WARN_ON(atomic_read(&skb->users));
3871 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3872 trace_consume_skb(skb);
3874 trace_kfree_skb(skb, net_tx_action);
3876 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
3879 __kfree_skb_defer(skb);
3882 __kfree_skb_flush();
3885 if (sd->output_queue) {
3888 local_irq_disable();
3889 head = sd->output_queue;
3890 sd->output_queue = NULL;
3891 sd->output_queue_tailp = &sd->output_queue;
3895 struct Qdisc *q = head;
3896 spinlock_t *root_lock;
3898 head = head->next_sched;
3900 root_lock = qdisc_lock(q);
3901 spin_lock(root_lock);
3902 /* We need to make sure head->next_sched is read
3903 * before clearing __QDISC_STATE_SCHED
3905 smp_mb__before_atomic();
3906 clear_bit(__QDISC_STATE_SCHED, &q->state);
3908 spin_unlock(root_lock);
3913 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3914 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3915 /* This hook is defined here for ATM LANE */
3916 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3917 unsigned char *addr) __read_mostly;
3918 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3921 static inline struct sk_buff *
3922 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
3923 struct net_device *orig_dev)
3925 #ifdef CONFIG_NET_CLS_ACT
3926 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3927 struct tcf_result cl_res;
3929 /* If there's at least one ingress present somewhere (so
3930 * we get here via enabled static key), remaining devices
3931 * that are not configured with an ingress qdisc will bail
3937 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3941 qdisc_skb_cb(skb)->pkt_len = skb->len;
3942 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3943 qdisc_bstats_cpu_update(cl->q, skb);
3945 switch (tc_classify(skb, cl, &cl_res, false)) {
3947 case TC_ACT_RECLASSIFY:
3948 skb->tc_index = TC_H_MIN(cl_res.classid);
3951 qdisc_qstats_cpu_drop(cl->q);
3958 case TC_ACT_REDIRECT:
3959 /* skb_mac_header check was done by cls/act_bpf, so
3960 * we can safely push the L2 header back before
3961 * redirecting to another netdev
3963 __skb_push(skb, skb->mac_len);
3964 skb_do_redirect(skb);
3969 #endif /* CONFIG_NET_CLS_ACT */
3974 * netdev_rx_handler_register - register receive handler
3975 * @dev: device to register a handler for
3976 * @rx_handler: receive handler to register
3977 * @rx_handler_data: data pointer that is used by rx handler
3979 * Register a receive handler for a device. This handler will then be
3980 * called from __netif_receive_skb. A negative errno code is returned
3983 * The caller must hold the rtnl_mutex.
3985 * For a general description of rx_handler, see enum rx_handler_result.
3987 int netdev_rx_handler_register(struct net_device *dev,
3988 rx_handler_func_t *rx_handler,
3989 void *rx_handler_data)
3993 if (dev->rx_handler)
3996 /* Note: rx_handler_data must be set before rx_handler */
3997 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3998 rcu_assign_pointer(dev->rx_handler, rx_handler);
4002 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4005 * netdev_rx_handler_unregister - unregister receive handler
4006 * @dev: device to unregister a handler from
4008 * Unregister a receive handler from a device.
4010 * The caller must hold the rtnl_mutex.
4012 void netdev_rx_handler_unregister(struct net_device *dev)
4016 RCU_INIT_POINTER(dev->rx_handler, NULL);
4017 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4018 * section has a guarantee to see a non NULL rx_handler_data
4022 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4024 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4027 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4028 * the special handling of PFMEMALLOC skbs.
4030 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4032 switch (skb->protocol) {
4033 case htons(ETH_P_ARP):
4034 case htons(ETH_P_IP):
4035 case htons(ETH_P_IPV6):
4036 case htons(ETH_P_8021Q):
4037 case htons(ETH_P_8021AD):
4044 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4045 int *ret, struct net_device *orig_dev)
4047 #ifdef CONFIG_NETFILTER_INGRESS
4048 if (nf_hook_ingress_active(skb)) {
4050 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4054 return nf_hook_ingress(skb);
4056 #endif /* CONFIG_NETFILTER_INGRESS */
4060 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
4062 struct packet_type *ptype, *pt_prev;
4063 rx_handler_func_t *rx_handler;
4064 struct net_device *orig_dev;
4065 bool deliver_exact = false;
4066 int ret = NET_RX_DROP;
4069 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4071 trace_netif_receive_skb(skb);
4073 orig_dev = skb->dev;
4075 skb_reset_network_header(skb);
4076 if (!skb_transport_header_was_set(skb))
4077 skb_reset_transport_header(skb);
4078 skb_reset_mac_len(skb);
4083 skb->skb_iif = skb->dev->ifindex;
4085 __this_cpu_inc(softnet_data.processed);
4087 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4088 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4089 skb = skb_vlan_untag(skb);
4094 #ifdef CONFIG_NET_CLS_ACT
4095 if (skb->tc_verd & TC_NCLS) {
4096 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
4104 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4106 ret = deliver_skb(skb, pt_prev, orig_dev);
4110 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4112 ret = deliver_skb(skb, pt_prev, orig_dev);
4117 #ifdef CONFIG_NET_INGRESS
4118 if (static_key_false(&ingress_needed)) {
4119 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4123 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4127 #ifdef CONFIG_NET_CLS_ACT
4131 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4134 if (skb_vlan_tag_present(skb)) {
4136 ret = deliver_skb(skb, pt_prev, orig_dev);
4139 if (vlan_do_receive(&skb))
4141 else if (unlikely(!skb))
4145 rx_handler = rcu_dereference(skb->dev->rx_handler);
4148 ret = deliver_skb(skb, pt_prev, orig_dev);
4151 switch (rx_handler(&skb)) {
4152 case RX_HANDLER_CONSUMED:
4153 ret = NET_RX_SUCCESS;
4155 case RX_HANDLER_ANOTHER:
4157 case RX_HANDLER_EXACT:
4158 deliver_exact = true;
4159 case RX_HANDLER_PASS:
4166 if (unlikely(skb_vlan_tag_present(skb))) {
4167 if (skb_vlan_tag_get_id(skb))
4168 skb->pkt_type = PACKET_OTHERHOST;
4169 /* Note: we might in the future use prio bits
4170 * and set skb->priority like in vlan_do_receive()
4171 * For the time being, just ignore Priority Code Point
4176 type = skb->protocol;
4178 /* deliver only exact match when indicated */
4179 if (likely(!deliver_exact)) {
4180 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4181 &ptype_base[ntohs(type) &
4185 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4186 &orig_dev->ptype_specific);
4188 if (unlikely(skb->dev != orig_dev)) {
4189 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4190 &skb->dev->ptype_specific);
4194 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
4197 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4201 atomic_long_inc(&skb->dev->rx_dropped);
4203 atomic_long_inc(&skb->dev->rx_nohandler);
4205 /* Jamal, now you will not able to escape explaining
4206 * me how you were going to use this. :-)
4215 static int __netif_receive_skb(struct sk_buff *skb)
4219 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4220 unsigned long pflags = current->flags;
4223 * PFMEMALLOC skbs are special, they should
4224 * - be delivered to SOCK_MEMALLOC sockets only
4225 * - stay away from userspace
4226 * - have bounded memory usage
4228 * Use PF_MEMALLOC as this saves us from propagating the allocation
4229 * context down to all allocation sites.
4231 current->flags |= PF_MEMALLOC;
4232 ret = __netif_receive_skb_core(skb, true);
4233 tsk_restore_flags(current, pflags, PF_MEMALLOC);
4235 ret = __netif_receive_skb_core(skb, false);
4240 static int netif_receive_skb_internal(struct sk_buff *skb)
4244 net_timestamp_check(netdev_tstamp_prequeue, skb);
4246 if (skb_defer_rx_timestamp(skb))
4247 return NET_RX_SUCCESS;
4252 if (static_key_false(&rps_needed)) {
4253 struct rps_dev_flow voidflow, *rflow = &voidflow;
4254 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4257 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4263 ret = __netif_receive_skb(skb);
4269 * netif_receive_skb - process receive buffer from network
4270 * @skb: buffer to process
4272 * netif_receive_skb() is the main receive data processing function.
4273 * It always succeeds. The buffer may be dropped during processing
4274 * for congestion control or by the protocol layers.
4276 * This function may only be called from softirq context and interrupts
4277 * should be enabled.
4279 * Return values (usually ignored):
4280 * NET_RX_SUCCESS: no congestion
4281 * NET_RX_DROP: packet was dropped
4283 int netif_receive_skb(struct sk_buff *skb)
4285 trace_netif_receive_skb_entry(skb);
4287 return netif_receive_skb_internal(skb);
4289 EXPORT_SYMBOL(netif_receive_skb);
4291 /* Network device is going away, flush any packets still pending
4292 * Called with irqs disabled.
4294 static void flush_backlog(void *arg)
4296 struct net_device *dev = arg;
4297 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4298 struct sk_buff *skb, *tmp;
4301 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4302 if (skb->dev == dev) {
4303 __skb_unlink(skb, &sd->input_pkt_queue);
4305 input_queue_head_incr(sd);
4310 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4311 if (skb->dev == dev) {
4312 __skb_unlink(skb, &sd->process_queue);
4314 input_queue_head_incr(sd);
4319 static int napi_gro_complete(struct sk_buff *skb)
4321 struct packet_offload *ptype;
4322 __be16 type = skb->protocol;
4323 struct list_head *head = &offload_base;
4326 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4328 if (NAPI_GRO_CB(skb)->count == 1) {
4329 skb_shinfo(skb)->gso_size = 0;
4334 list_for_each_entry_rcu(ptype, head, list) {
4335 if (ptype->type != type || !ptype->callbacks.gro_complete)
4338 err = ptype->callbacks.gro_complete(skb, 0);
4344 WARN_ON(&ptype->list == head);
4346 return NET_RX_SUCCESS;
4350 return netif_receive_skb_internal(skb);
4353 /* napi->gro_list contains packets ordered by age.
4354 * youngest packets at the head of it.
4355 * Complete skbs in reverse order to reduce latencies.
4357 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4359 struct sk_buff *skb, *prev = NULL;
4361 /* scan list and build reverse chain */
4362 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4367 for (skb = prev; skb; skb = prev) {
4370 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4374 napi_gro_complete(skb);
4378 napi->gro_list = NULL;
4380 EXPORT_SYMBOL(napi_gro_flush);
4382 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4385 unsigned int maclen = skb->dev->hard_header_len;
4386 u32 hash = skb_get_hash_raw(skb);
4388 for (p = napi->gro_list; p; p = p->next) {
4389 unsigned long diffs;
4391 NAPI_GRO_CB(p)->flush = 0;
4393 if (hash != skb_get_hash_raw(p)) {
4394 NAPI_GRO_CB(p)->same_flow = 0;
4398 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4399 diffs |= p->vlan_tci ^ skb->vlan_tci;
4400 diffs |= skb_metadata_dst_cmp(p, skb);
4401 if (maclen == ETH_HLEN)
4402 diffs |= compare_ether_header(skb_mac_header(p),
4403 skb_mac_header(skb));
4405 diffs = memcmp(skb_mac_header(p),
4406 skb_mac_header(skb),
4408 NAPI_GRO_CB(p)->same_flow = !diffs;
4412 static void skb_gro_reset_offset(struct sk_buff *skb)
4414 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4415 const skb_frag_t *frag0 = &pinfo->frags[0];
4417 NAPI_GRO_CB(skb)->data_offset = 0;
4418 NAPI_GRO_CB(skb)->frag0 = NULL;
4419 NAPI_GRO_CB(skb)->frag0_len = 0;
4421 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4423 !PageHighMem(skb_frag_page(frag0))) {
4424 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4425 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4429 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4431 struct skb_shared_info *pinfo = skb_shinfo(skb);
4433 BUG_ON(skb->end - skb->tail < grow);
4435 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4437 skb->data_len -= grow;
4440 pinfo->frags[0].page_offset += grow;
4441 skb_frag_size_sub(&pinfo->frags[0], grow);
4443 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4444 skb_frag_unref(skb, 0);
4445 memmove(pinfo->frags, pinfo->frags + 1,
4446 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4450 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4452 struct sk_buff **pp = NULL;
4453 struct packet_offload *ptype;
4454 __be16 type = skb->protocol;
4455 struct list_head *head = &offload_base;
4457 enum gro_result ret;
4460 if (!(skb->dev->features & NETIF_F_GRO))
4463 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4466 gro_list_prepare(napi, skb);
4469 list_for_each_entry_rcu(ptype, head, list) {
4470 if (ptype->type != type || !ptype->callbacks.gro_receive)
4473 skb_set_network_header(skb, skb_gro_offset(skb));
4474 skb_reset_mac_len(skb);
4475 NAPI_GRO_CB(skb)->same_flow = 0;
4476 NAPI_GRO_CB(skb)->flush = 0;
4477 NAPI_GRO_CB(skb)->free = 0;
4478 NAPI_GRO_CB(skb)->encap_mark = 0;
4479 NAPI_GRO_CB(skb)->is_fou = 0;
4480 NAPI_GRO_CB(skb)->is_atomic = 1;
4481 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4483 /* Setup for GRO checksum validation */
4484 switch (skb->ip_summed) {
4485 case CHECKSUM_COMPLETE:
4486 NAPI_GRO_CB(skb)->csum = skb->csum;
4487 NAPI_GRO_CB(skb)->csum_valid = 1;
4488 NAPI_GRO_CB(skb)->csum_cnt = 0;
4490 case CHECKSUM_UNNECESSARY:
4491 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4492 NAPI_GRO_CB(skb)->csum_valid = 0;
4495 NAPI_GRO_CB(skb)->csum_cnt = 0;
4496 NAPI_GRO_CB(skb)->csum_valid = 0;
4499 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4504 if (&ptype->list == head)
4507 same_flow = NAPI_GRO_CB(skb)->same_flow;
4508 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4511 struct sk_buff *nskb = *pp;
4515 napi_gro_complete(nskb);
4522 if (NAPI_GRO_CB(skb)->flush)
4525 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4526 struct sk_buff *nskb = napi->gro_list;
4528 /* locate the end of the list to select the 'oldest' flow */
4529 while (nskb->next) {
4535 napi_gro_complete(nskb);
4539 NAPI_GRO_CB(skb)->count = 1;
4540 NAPI_GRO_CB(skb)->age = jiffies;
4541 NAPI_GRO_CB(skb)->last = skb;
4542 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4543 skb->next = napi->gro_list;
4544 napi->gro_list = skb;
4548 grow = skb_gro_offset(skb) - skb_headlen(skb);
4550 gro_pull_from_frag0(skb, grow);
4559 struct packet_offload *gro_find_receive_by_type(__be16 type)
4561 struct list_head *offload_head = &offload_base;
4562 struct packet_offload *ptype;
4564 list_for_each_entry_rcu(ptype, offload_head, list) {
4565 if (ptype->type != type || !ptype->callbacks.gro_receive)
4571 EXPORT_SYMBOL(gro_find_receive_by_type);
4573 struct packet_offload *gro_find_complete_by_type(__be16 type)
4575 struct list_head *offload_head = &offload_base;
4576 struct packet_offload *ptype;
4578 list_for_each_entry_rcu(ptype, offload_head, list) {
4579 if (ptype->type != type || !ptype->callbacks.gro_complete)
4585 EXPORT_SYMBOL(gro_find_complete_by_type);
4587 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4591 if (netif_receive_skb_internal(skb))
4599 case GRO_MERGED_FREE:
4600 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) {
4602 kmem_cache_free(skbuff_head_cache, skb);
4616 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4618 skb_mark_napi_id(skb, napi);
4619 trace_napi_gro_receive_entry(skb);
4621 skb_gro_reset_offset(skb);
4623 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4625 EXPORT_SYMBOL(napi_gro_receive);
4627 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4629 if (unlikely(skb->pfmemalloc)) {
4633 __skb_pull(skb, skb_headlen(skb));
4634 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4635 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4637 skb->dev = napi->dev;
4639 skb->encapsulation = 0;
4640 skb_shinfo(skb)->gso_type = 0;
4641 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4646 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4648 struct sk_buff *skb = napi->skb;
4651 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4654 skb_mark_napi_id(skb, napi);
4659 EXPORT_SYMBOL(napi_get_frags);
4661 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4662 struct sk_buff *skb,
4668 __skb_push(skb, ETH_HLEN);
4669 skb->protocol = eth_type_trans(skb, skb->dev);
4670 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4675 case GRO_MERGED_FREE:
4676 napi_reuse_skb(napi, skb);
4686 /* Upper GRO stack assumes network header starts at gro_offset=0
4687 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4688 * We copy ethernet header into skb->data to have a common layout.
4690 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4692 struct sk_buff *skb = napi->skb;
4693 const struct ethhdr *eth;
4694 unsigned int hlen = sizeof(*eth);
4698 skb_reset_mac_header(skb);
4699 skb_gro_reset_offset(skb);
4701 eth = skb_gro_header_fast(skb, 0);
4702 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4703 eth = skb_gro_header_slow(skb, hlen, 0);
4704 if (unlikely(!eth)) {
4705 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
4706 __func__, napi->dev->name);
4707 napi_reuse_skb(napi, skb);
4711 gro_pull_from_frag0(skb, hlen);
4712 NAPI_GRO_CB(skb)->frag0 += hlen;
4713 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4715 __skb_pull(skb, hlen);
4718 * This works because the only protocols we care about don't require
4720 * We'll fix it up properly in napi_frags_finish()
4722 skb->protocol = eth->h_proto;
4727 gro_result_t napi_gro_frags(struct napi_struct *napi)
4729 struct sk_buff *skb = napi_frags_skb(napi);
4734 trace_napi_gro_frags_entry(skb);
4736 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4738 EXPORT_SYMBOL(napi_gro_frags);
4740 /* Compute the checksum from gro_offset and return the folded value
4741 * after adding in any pseudo checksum.
4743 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4748 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4750 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4751 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4753 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4754 !skb->csum_complete_sw)
4755 netdev_rx_csum_fault(skb->dev);
4758 NAPI_GRO_CB(skb)->csum = wsum;
4759 NAPI_GRO_CB(skb)->csum_valid = 1;
4763 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4766 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4767 * Note: called with local irq disabled, but exits with local irq enabled.
4769 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4772 struct softnet_data *remsd = sd->rps_ipi_list;
4775 sd->rps_ipi_list = NULL;
4779 /* Send pending IPI's to kick RPS processing on remote cpus. */
4781 struct softnet_data *next = remsd->rps_ipi_next;
4783 if (cpu_online(remsd->cpu))
4784 smp_call_function_single_async(remsd->cpu,
4793 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4796 return sd->rps_ipi_list != NULL;
4802 static int process_backlog(struct napi_struct *napi, int quota)
4805 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4807 /* Check if we have pending ipi, its better to send them now,
4808 * not waiting net_rx_action() end.
4810 if (sd_has_rps_ipi_waiting(sd)) {
4811 local_irq_disable();
4812 net_rps_action_and_irq_enable(sd);
4815 napi->weight = weight_p;
4816 local_irq_disable();
4818 struct sk_buff *skb;
4820 while ((skb = __skb_dequeue(&sd->process_queue))) {
4823 __netif_receive_skb(skb);
4825 local_irq_disable();
4826 input_queue_head_incr(sd);
4827 if (++work >= quota) {
4834 if (skb_queue_empty(&sd->input_pkt_queue)) {
4836 * Inline a custom version of __napi_complete().
4837 * only current cpu owns and manipulates this napi,
4838 * and NAPI_STATE_SCHED is the only possible flag set
4840 * We can use a plain write instead of clear_bit(),
4841 * and we dont need an smp_mb() memory barrier.
4849 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4850 &sd->process_queue);
4859 * __napi_schedule - schedule for receive
4860 * @n: entry to schedule
4862 * The entry's receive function will be scheduled to run.
4863 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4865 void __napi_schedule(struct napi_struct *n)
4867 unsigned long flags;
4869 local_irq_save(flags);
4870 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4871 local_irq_restore(flags);
4873 EXPORT_SYMBOL(__napi_schedule);
4876 * __napi_schedule_irqoff - schedule for receive
4877 * @n: entry to schedule
4879 * Variant of __napi_schedule() assuming hard irqs are masked
4881 void __napi_schedule_irqoff(struct napi_struct *n)
4883 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4885 EXPORT_SYMBOL(__napi_schedule_irqoff);
4887 void __napi_complete(struct napi_struct *n)
4889 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4891 list_del_init(&n->poll_list);
4892 smp_mb__before_atomic();
4893 clear_bit(NAPI_STATE_SCHED, &n->state);
4895 EXPORT_SYMBOL(__napi_complete);
4897 void napi_complete_done(struct napi_struct *n, int work_done)
4899 unsigned long flags;
4902 * don't let napi dequeue from the cpu poll list
4903 * just in case its running on a different cpu
4905 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4909 unsigned long timeout = 0;
4912 timeout = n->dev->gro_flush_timeout;
4915 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4916 HRTIMER_MODE_REL_PINNED);
4918 napi_gro_flush(n, false);
4920 if (likely(list_empty(&n->poll_list))) {
4921 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4923 /* If n->poll_list is not empty, we need to mask irqs */
4924 local_irq_save(flags);
4926 local_irq_restore(flags);
4929 EXPORT_SYMBOL(napi_complete_done);
4931 /* must be called under rcu_read_lock(), as we dont take a reference */
4932 static struct napi_struct *napi_by_id(unsigned int napi_id)
4934 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4935 struct napi_struct *napi;
4937 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4938 if (napi->napi_id == napi_id)
4944 #if defined(CONFIG_NET_RX_BUSY_POLL)
4945 #define BUSY_POLL_BUDGET 8
4946 bool sk_busy_loop(struct sock *sk, int nonblock)
4948 unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0;
4949 int (*busy_poll)(struct napi_struct *dev);
4950 struct napi_struct *napi;
4955 napi = napi_by_id(sk->sk_napi_id);
4959 /* Note: ndo_busy_poll method is optional in linux-4.5 */
4960 busy_poll = napi->dev->netdev_ops->ndo_busy_poll;
4966 rc = busy_poll(napi);
4967 } else if (napi_schedule_prep(napi)) {
4968 void *have = netpoll_poll_lock(napi);
4970 if (test_bit(NAPI_STATE_SCHED, &napi->state)) {
4971 rc = napi->poll(napi, BUSY_POLL_BUDGET);
4972 trace_napi_poll(napi);
4973 if (rc == BUSY_POLL_BUDGET) {
4974 napi_complete_done(napi, rc);
4975 napi_schedule(napi);
4978 netpoll_poll_unlock(have);
4981 __NET_ADD_STATS(sock_net(sk),
4982 LINUX_MIB_BUSYPOLLRXPACKETS, rc);
4985 if (rc == LL_FLUSH_FAILED)
4986 break; /* permanent failure */
4989 } while (!nonblock && skb_queue_empty(&sk->sk_receive_queue) &&
4990 !need_resched() && !busy_loop_timeout(end_time));
4992 rc = !skb_queue_empty(&sk->sk_receive_queue);
4997 EXPORT_SYMBOL(sk_busy_loop);
4999 #endif /* CONFIG_NET_RX_BUSY_POLL */
5001 void napi_hash_add(struct napi_struct *napi)
5003 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
5004 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
5007 spin_lock(&napi_hash_lock);
5009 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */
5011 if (unlikely(++napi_gen_id < NR_CPUS + 1))
5012 napi_gen_id = NR_CPUS + 1;
5013 } while (napi_by_id(napi_gen_id));
5014 napi->napi_id = napi_gen_id;
5016 hlist_add_head_rcu(&napi->napi_hash_node,
5017 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
5019 spin_unlock(&napi_hash_lock);
5021 EXPORT_SYMBOL_GPL(napi_hash_add);
5023 /* Warning : caller is responsible to make sure rcu grace period
5024 * is respected before freeing memory containing @napi
5026 bool napi_hash_del(struct napi_struct *napi)
5028 bool rcu_sync_needed = false;
5030 spin_lock(&napi_hash_lock);
5032 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
5033 rcu_sync_needed = true;
5034 hlist_del_rcu(&napi->napi_hash_node);
5036 spin_unlock(&napi_hash_lock);
5037 return rcu_sync_needed;
5039 EXPORT_SYMBOL_GPL(napi_hash_del);
5041 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
5043 struct napi_struct *napi;
5045 napi = container_of(timer, struct napi_struct, timer);
5047 napi_schedule(napi);
5049 return HRTIMER_NORESTART;
5052 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
5053 int (*poll)(struct napi_struct *, int), int weight)
5055 INIT_LIST_HEAD(&napi->poll_list);
5056 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
5057 napi->timer.function = napi_watchdog;
5058 napi->gro_count = 0;
5059 napi->gro_list = NULL;
5062 if (weight > NAPI_POLL_WEIGHT)
5063 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5065 napi->weight = weight;
5066 list_add(&napi->dev_list, &dev->napi_list);
5068 #ifdef CONFIG_NETPOLL
5069 spin_lock_init(&napi->poll_lock);
5070 napi->poll_owner = -1;
5072 set_bit(NAPI_STATE_SCHED, &napi->state);
5073 napi_hash_add(napi);
5075 EXPORT_SYMBOL(netif_napi_add);
5077 void napi_disable(struct napi_struct *n)
5080 set_bit(NAPI_STATE_DISABLE, &n->state);
5082 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
5084 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
5087 hrtimer_cancel(&n->timer);
5089 clear_bit(NAPI_STATE_DISABLE, &n->state);
5091 EXPORT_SYMBOL(napi_disable);
5093 /* Must be called in process context */
5094 void netif_napi_del(struct napi_struct *napi)
5097 if (napi_hash_del(napi))
5099 list_del_init(&napi->dev_list);
5100 napi_free_frags(napi);
5102 kfree_skb_list(napi->gro_list);
5103 napi->gro_list = NULL;
5104 napi->gro_count = 0;
5106 EXPORT_SYMBOL(netif_napi_del);
5108 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
5113 list_del_init(&n->poll_list);
5115 have = netpoll_poll_lock(n);
5119 /* This NAPI_STATE_SCHED test is for avoiding a race
5120 * with netpoll's poll_napi(). Only the entity which
5121 * obtains the lock and sees NAPI_STATE_SCHED set will
5122 * actually make the ->poll() call. Therefore we avoid
5123 * accidentally calling ->poll() when NAPI is not scheduled.
5126 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
5127 work = n->poll(n, weight);
5131 WARN_ON_ONCE(work > weight);
5133 if (likely(work < weight))
5136 /* Drivers must not modify the NAPI state if they
5137 * consume the entire weight. In such cases this code
5138 * still "owns" the NAPI instance and therefore can
5139 * move the instance around on the list at-will.
5141 if (unlikely(napi_disable_pending(n))) {
5147 /* flush too old packets
5148 * If HZ < 1000, flush all packets.
5150 napi_gro_flush(n, HZ >= 1000);
5153 /* Some drivers may have called napi_schedule
5154 * prior to exhausting their budget.
5156 if (unlikely(!list_empty(&n->poll_list))) {
5157 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5158 n->dev ? n->dev->name : "backlog");
5162 list_add_tail(&n->poll_list, repoll);
5165 netpoll_poll_unlock(have);
5170 static void net_rx_action(struct softirq_action *h)
5172 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5173 unsigned long time_limit = jiffies + 2;
5174 int budget = netdev_budget;
5178 local_irq_disable();
5179 list_splice_init(&sd->poll_list, &list);
5183 struct napi_struct *n;
5185 if (list_empty(&list)) {
5186 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
5191 n = list_first_entry(&list, struct napi_struct, poll_list);
5192 budget -= napi_poll(n, &repoll);
5194 /* If softirq window is exhausted then punt.
5195 * Allow this to run for 2 jiffies since which will allow
5196 * an average latency of 1.5/HZ.
5198 if (unlikely(budget <= 0 ||
5199 time_after_eq(jiffies, time_limit))) {
5205 __kfree_skb_flush();
5206 local_irq_disable();
5208 list_splice_tail_init(&sd->poll_list, &list);
5209 list_splice_tail(&repoll, &list);
5210 list_splice(&list, &sd->poll_list);
5211 if (!list_empty(&sd->poll_list))
5212 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5214 net_rps_action_and_irq_enable(sd);
5217 struct netdev_adjacent {
5218 struct net_device *dev;
5220 /* upper master flag, there can only be one master device per list */
5223 /* counter for the number of times this device was added to us */
5226 /* private field for the users */
5229 struct list_head list;
5230 struct rcu_head rcu;
5233 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
5234 struct list_head *adj_list)
5236 struct netdev_adjacent *adj;
5238 list_for_each_entry(adj, adj_list, list) {
5239 if (adj->dev == adj_dev)
5246 * netdev_has_upper_dev - Check if device is linked to an upper device
5248 * @upper_dev: upper device to check
5250 * Find out if a device is linked to specified upper device and return true
5251 * in case it is. Note that this checks only immediate upper device,
5252 * not through a complete stack of devices. The caller must hold the RTNL lock.
5254 bool netdev_has_upper_dev(struct net_device *dev,
5255 struct net_device *upper_dev)
5259 return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper);
5261 EXPORT_SYMBOL(netdev_has_upper_dev);
5264 * netdev_has_any_upper_dev - Check if device is linked to some device
5267 * Find out if a device is linked to an upper device and return true in case
5268 * it is. The caller must hold the RTNL lock.
5270 static bool netdev_has_any_upper_dev(struct net_device *dev)
5274 return !list_empty(&dev->all_adj_list.upper);
5278 * netdev_master_upper_dev_get - Get master upper device
5281 * Find a master upper device and return pointer to it or NULL in case
5282 * it's not there. The caller must hold the RTNL lock.
5284 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5286 struct netdev_adjacent *upper;
5290 if (list_empty(&dev->adj_list.upper))
5293 upper = list_first_entry(&dev->adj_list.upper,
5294 struct netdev_adjacent, list);
5295 if (likely(upper->master))
5299 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5301 void *netdev_adjacent_get_private(struct list_head *adj_list)
5303 struct netdev_adjacent *adj;
5305 adj = list_entry(adj_list, struct netdev_adjacent, list);
5307 return adj->private;
5309 EXPORT_SYMBOL(netdev_adjacent_get_private);
5312 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5314 * @iter: list_head ** of the current position
5316 * Gets the next device from the dev's upper list, starting from iter
5317 * position. The caller must hold RCU read lock.
5319 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5320 struct list_head **iter)
5322 struct netdev_adjacent *upper;
5324 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5326 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5328 if (&upper->list == &dev->adj_list.upper)
5331 *iter = &upper->list;
5335 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5338 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
5340 * @iter: list_head ** of the current position
5342 * Gets the next device from the dev's upper list, starting from iter
5343 * position. The caller must hold RCU read lock.
5345 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
5346 struct list_head **iter)
5348 struct netdev_adjacent *upper;
5350 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5352 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5354 if (&upper->list == &dev->all_adj_list.upper)
5357 *iter = &upper->list;
5361 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
5364 * netdev_lower_get_next_private - Get the next ->private from the
5365 * lower neighbour list
5367 * @iter: list_head ** of the current position
5369 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5370 * list, starting from iter position. The caller must hold either hold the
5371 * RTNL lock or its own locking that guarantees that the neighbour lower
5372 * list will remain unchanged.
5374 void *netdev_lower_get_next_private(struct net_device *dev,
5375 struct list_head **iter)
5377 struct netdev_adjacent *lower;
5379 lower = list_entry(*iter, struct netdev_adjacent, list);
5381 if (&lower->list == &dev->adj_list.lower)
5384 *iter = lower->list.next;
5386 return lower->private;
5388 EXPORT_SYMBOL(netdev_lower_get_next_private);
5391 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5392 * lower neighbour list, RCU
5395 * @iter: list_head ** of the current position
5397 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5398 * list, starting from iter position. The caller must hold RCU read lock.
5400 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5401 struct list_head **iter)
5403 struct netdev_adjacent *lower;
5405 WARN_ON_ONCE(!rcu_read_lock_held());
5407 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5409 if (&lower->list == &dev->adj_list.lower)
5412 *iter = &lower->list;
5414 return lower->private;
5416 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5419 * netdev_lower_get_next - Get the next device from the lower neighbour
5422 * @iter: list_head ** of the current position
5424 * Gets the next netdev_adjacent from the dev's lower neighbour
5425 * list, starting from iter position. The caller must hold RTNL lock or
5426 * its own locking that guarantees that the neighbour lower
5427 * list will remain unchanged.
5429 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5431 struct netdev_adjacent *lower;
5433 lower = list_entry(*iter, struct netdev_adjacent, list);
5435 if (&lower->list == &dev->adj_list.lower)
5438 *iter = lower->list.next;
5442 EXPORT_SYMBOL(netdev_lower_get_next);
5445 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5446 * lower neighbour list, RCU
5450 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5451 * list. The caller must hold RCU read lock.
5453 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5455 struct netdev_adjacent *lower;
5457 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5458 struct netdev_adjacent, list);
5460 return lower->private;
5463 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5466 * netdev_master_upper_dev_get_rcu - Get master upper device
5469 * Find a master upper device and return pointer to it or NULL in case
5470 * it's not there. The caller must hold the RCU read lock.
5472 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5474 struct netdev_adjacent *upper;
5476 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5477 struct netdev_adjacent, list);
5478 if (upper && likely(upper->master))
5482 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5484 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5485 struct net_device *adj_dev,
5486 struct list_head *dev_list)
5488 char linkname[IFNAMSIZ+7];
5489 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5490 "upper_%s" : "lower_%s", adj_dev->name);
5491 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5494 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5496 struct list_head *dev_list)
5498 char linkname[IFNAMSIZ+7];
5499 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5500 "upper_%s" : "lower_%s", name);
5501 sysfs_remove_link(&(dev->dev.kobj), linkname);
5504 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5505 struct net_device *adj_dev,
5506 struct list_head *dev_list)
5508 return (dev_list == &dev->adj_list.upper ||
5509 dev_list == &dev->adj_list.lower) &&
5510 net_eq(dev_net(dev), dev_net(adj_dev));
5513 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5514 struct net_device *adj_dev,
5515 struct list_head *dev_list,
5516 void *private, bool master)
5518 struct netdev_adjacent *adj;
5521 adj = __netdev_find_adj(adj_dev, dev_list);
5528 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5533 adj->master = master;
5535 adj->private = private;
5538 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5539 adj_dev->name, dev->name, adj_dev->name);
5541 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5542 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5547 /* Ensure that master link is always the first item in list. */
5549 ret = sysfs_create_link(&(dev->dev.kobj),
5550 &(adj_dev->dev.kobj), "master");
5552 goto remove_symlinks;
5554 list_add_rcu(&adj->list, dev_list);
5556 list_add_tail_rcu(&adj->list, dev_list);
5562 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5563 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5571 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5572 struct net_device *adj_dev,
5573 struct list_head *dev_list)
5575 struct netdev_adjacent *adj;
5577 adj = __netdev_find_adj(adj_dev, dev_list);
5580 pr_err("tried to remove device %s from %s\n",
5581 dev->name, adj_dev->name);
5585 if (adj->ref_nr > 1) {
5586 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5593 sysfs_remove_link(&(dev->dev.kobj), "master");
5595 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5596 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5598 list_del_rcu(&adj->list);
5599 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5600 adj_dev->name, dev->name, adj_dev->name);
5602 kfree_rcu(adj, rcu);
5605 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5606 struct net_device *upper_dev,
5607 struct list_head *up_list,
5608 struct list_head *down_list,
5609 void *private, bool master)
5613 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5618 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5621 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5628 static int __netdev_adjacent_dev_link(struct net_device *dev,
5629 struct net_device *upper_dev)
5631 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5632 &dev->all_adj_list.upper,
5633 &upper_dev->all_adj_list.lower,
5637 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5638 struct net_device *upper_dev,
5639 struct list_head *up_list,
5640 struct list_head *down_list)
5642 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5643 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5646 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5647 struct net_device *upper_dev)
5649 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5650 &dev->all_adj_list.upper,
5651 &upper_dev->all_adj_list.lower);
5654 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5655 struct net_device *upper_dev,
5656 void *private, bool master)
5658 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5663 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5664 &dev->adj_list.upper,
5665 &upper_dev->adj_list.lower,
5668 __netdev_adjacent_dev_unlink(dev, upper_dev);
5675 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5676 struct net_device *upper_dev)
5678 __netdev_adjacent_dev_unlink(dev, upper_dev);
5679 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5680 &dev->adj_list.upper,
5681 &upper_dev->adj_list.lower);
5684 static int __netdev_upper_dev_link(struct net_device *dev,
5685 struct net_device *upper_dev, bool master,
5686 void *upper_priv, void *upper_info)
5688 struct netdev_notifier_changeupper_info changeupper_info;
5689 struct netdev_adjacent *i, *j, *to_i, *to_j;
5694 if (dev == upper_dev)
5697 /* To prevent loops, check if dev is not upper device to upper_dev. */
5698 if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper))
5701 if (__netdev_find_adj(upper_dev, &dev->adj_list.upper))
5704 if (master && netdev_master_upper_dev_get(dev))
5707 changeupper_info.upper_dev = upper_dev;
5708 changeupper_info.master = master;
5709 changeupper_info.linking = true;
5710 changeupper_info.upper_info = upper_info;
5712 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5713 &changeupper_info.info);
5714 ret = notifier_to_errno(ret);
5718 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
5723 /* Now that we linked these devs, make all the upper_dev's
5724 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5725 * versa, and don't forget the devices itself. All of these
5726 * links are non-neighbours.
5728 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5729 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5730 pr_debug("Interlinking %s with %s, non-neighbour\n",
5731 i->dev->name, j->dev->name);
5732 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5738 /* add dev to every upper_dev's upper device */
5739 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5740 pr_debug("linking %s's upper device %s with %s\n",
5741 upper_dev->name, i->dev->name, dev->name);
5742 ret = __netdev_adjacent_dev_link(dev, i->dev);
5744 goto rollback_upper_mesh;
5747 /* add upper_dev to every dev's lower device */
5748 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5749 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5750 i->dev->name, upper_dev->name);
5751 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5753 goto rollback_lower_mesh;
5756 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5757 &changeupper_info.info);
5758 ret = notifier_to_errno(ret);
5760 goto rollback_lower_mesh;
5764 rollback_lower_mesh:
5766 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5769 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5774 rollback_upper_mesh:
5776 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5779 __netdev_adjacent_dev_unlink(dev, i->dev);
5787 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5788 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5789 if (i == to_i && j == to_j)
5791 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5797 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5803 * netdev_upper_dev_link - Add a link to the upper device
5805 * @upper_dev: new upper device
5807 * Adds a link to device which is upper to this one. The caller must hold
5808 * the RTNL lock. On a failure a negative errno code is returned.
5809 * On success the reference counts are adjusted and the function
5812 int netdev_upper_dev_link(struct net_device *dev,
5813 struct net_device *upper_dev)
5815 return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL);
5817 EXPORT_SYMBOL(netdev_upper_dev_link);
5820 * netdev_master_upper_dev_link - Add a master link to the upper device
5822 * @upper_dev: new upper device
5823 * @upper_priv: upper device private
5824 * @upper_info: upper info to be passed down via notifier
5826 * Adds a link to device which is upper to this one. In this case, only
5827 * one master upper device can be linked, although other non-master devices
5828 * might be linked as well. The caller must hold the RTNL lock.
5829 * On a failure a negative errno code is returned. On success the reference
5830 * counts are adjusted and the function returns zero.
5832 int netdev_master_upper_dev_link(struct net_device *dev,
5833 struct net_device *upper_dev,
5834 void *upper_priv, void *upper_info)
5836 return __netdev_upper_dev_link(dev, upper_dev, true,
5837 upper_priv, upper_info);
5839 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5842 * netdev_upper_dev_unlink - Removes a link to upper device
5844 * @upper_dev: new upper device
5846 * Removes a link to device which is upper to this one. The caller must hold
5849 void netdev_upper_dev_unlink(struct net_device *dev,
5850 struct net_device *upper_dev)
5852 struct netdev_notifier_changeupper_info changeupper_info;
5853 struct netdev_adjacent *i, *j;
5856 changeupper_info.upper_dev = upper_dev;
5857 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
5858 changeupper_info.linking = false;
5860 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5861 &changeupper_info.info);
5863 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5865 /* Here is the tricky part. We must remove all dev's lower
5866 * devices from all upper_dev's upper devices and vice
5867 * versa, to maintain the graph relationship.
5869 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5870 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5871 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5873 /* remove also the devices itself from lower/upper device
5876 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5877 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5879 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5880 __netdev_adjacent_dev_unlink(dev, i->dev);
5882 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5883 &changeupper_info.info);
5885 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5888 * netdev_bonding_info_change - Dispatch event about slave change
5890 * @bonding_info: info to dispatch
5892 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5893 * The caller must hold the RTNL lock.
5895 void netdev_bonding_info_change(struct net_device *dev,
5896 struct netdev_bonding_info *bonding_info)
5898 struct netdev_notifier_bonding_info info;
5900 memcpy(&info.bonding_info, bonding_info,
5901 sizeof(struct netdev_bonding_info));
5902 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5905 EXPORT_SYMBOL(netdev_bonding_info_change);
5907 static void netdev_adjacent_add_links(struct net_device *dev)
5909 struct netdev_adjacent *iter;
5911 struct net *net = dev_net(dev);
5913 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5914 if (!net_eq(net,dev_net(iter->dev)))
5916 netdev_adjacent_sysfs_add(iter->dev, dev,
5917 &iter->dev->adj_list.lower);
5918 netdev_adjacent_sysfs_add(dev, iter->dev,
5919 &dev->adj_list.upper);
5922 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5923 if (!net_eq(net,dev_net(iter->dev)))
5925 netdev_adjacent_sysfs_add(iter->dev, dev,
5926 &iter->dev->adj_list.upper);
5927 netdev_adjacent_sysfs_add(dev, iter->dev,
5928 &dev->adj_list.lower);
5932 static void netdev_adjacent_del_links(struct net_device *dev)
5934 struct netdev_adjacent *iter;
5936 struct net *net = dev_net(dev);
5938 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5939 if (!net_eq(net,dev_net(iter->dev)))
5941 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5942 &iter->dev->adj_list.lower);
5943 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5944 &dev->adj_list.upper);
5947 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5948 if (!net_eq(net,dev_net(iter->dev)))
5950 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5951 &iter->dev->adj_list.upper);
5952 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5953 &dev->adj_list.lower);
5957 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5959 struct netdev_adjacent *iter;
5961 struct net *net = dev_net(dev);
5963 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5964 if (!net_eq(net,dev_net(iter->dev)))
5966 netdev_adjacent_sysfs_del(iter->dev, oldname,
5967 &iter->dev->adj_list.lower);
5968 netdev_adjacent_sysfs_add(iter->dev, dev,
5969 &iter->dev->adj_list.lower);
5972 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5973 if (!net_eq(net,dev_net(iter->dev)))
5975 netdev_adjacent_sysfs_del(iter->dev, oldname,
5976 &iter->dev->adj_list.upper);
5977 netdev_adjacent_sysfs_add(iter->dev, dev,
5978 &iter->dev->adj_list.upper);
5982 void *netdev_lower_dev_get_private(struct net_device *dev,
5983 struct net_device *lower_dev)
5985 struct netdev_adjacent *lower;
5989 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
5993 return lower->private;
5995 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5998 int dev_get_nest_level(struct net_device *dev,
5999 bool (*type_check)(const struct net_device *dev))
6001 struct net_device *lower = NULL;
6002 struct list_head *iter;
6008 netdev_for_each_lower_dev(dev, lower, iter) {
6009 nest = dev_get_nest_level(lower, type_check);
6010 if (max_nest < nest)
6014 if (type_check(dev))
6019 EXPORT_SYMBOL(dev_get_nest_level);
6022 * netdev_lower_change - Dispatch event about lower device state change
6023 * @lower_dev: device
6024 * @lower_state_info: state to dispatch
6026 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
6027 * The caller must hold the RTNL lock.
6029 void netdev_lower_state_changed(struct net_device *lower_dev,
6030 void *lower_state_info)
6032 struct netdev_notifier_changelowerstate_info changelowerstate_info;
6035 changelowerstate_info.lower_state_info = lower_state_info;
6036 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev,
6037 &changelowerstate_info.info);
6039 EXPORT_SYMBOL(netdev_lower_state_changed);
6041 static void dev_change_rx_flags(struct net_device *dev, int flags)
6043 const struct net_device_ops *ops = dev->netdev_ops;
6045 if (ops->ndo_change_rx_flags)
6046 ops->ndo_change_rx_flags(dev, flags);
6049 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
6051 unsigned int old_flags = dev->flags;
6057 dev->flags |= IFF_PROMISC;
6058 dev->promiscuity += inc;
6059 if (dev->promiscuity == 0) {
6062 * If inc causes overflow, untouch promisc and return error.
6065 dev->flags &= ~IFF_PROMISC;
6067 dev->promiscuity -= inc;
6068 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6073 if (dev->flags != old_flags) {
6074 pr_info("device %s %s promiscuous mode\n",
6076 dev->flags & IFF_PROMISC ? "entered" : "left");
6077 if (audit_enabled) {
6078 current_uid_gid(&uid, &gid);
6079 audit_log(current->audit_context, GFP_ATOMIC,
6080 AUDIT_ANOM_PROMISCUOUS,
6081 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6082 dev->name, (dev->flags & IFF_PROMISC),
6083 (old_flags & IFF_PROMISC),
6084 from_kuid(&init_user_ns, audit_get_loginuid(current)),
6085 from_kuid(&init_user_ns, uid),
6086 from_kgid(&init_user_ns, gid),
6087 audit_get_sessionid(current));
6090 dev_change_rx_flags(dev, IFF_PROMISC);
6093 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
6098 * dev_set_promiscuity - update promiscuity count on a device
6102 * Add or remove promiscuity from a device. While the count in the device
6103 * remains above zero the interface remains promiscuous. Once it hits zero
6104 * the device reverts back to normal filtering operation. A negative inc
6105 * value is used to drop promiscuity on the device.
6106 * Return 0 if successful or a negative errno code on error.
6108 int dev_set_promiscuity(struct net_device *dev, int inc)
6110 unsigned int old_flags = dev->flags;
6113 err = __dev_set_promiscuity(dev, inc, true);
6116 if (dev->flags != old_flags)
6117 dev_set_rx_mode(dev);
6120 EXPORT_SYMBOL(dev_set_promiscuity);
6122 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
6124 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
6128 dev->flags |= IFF_ALLMULTI;
6129 dev->allmulti += inc;
6130 if (dev->allmulti == 0) {
6133 * If inc causes overflow, untouch allmulti and return error.
6136 dev->flags &= ~IFF_ALLMULTI;
6138 dev->allmulti -= inc;
6139 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6144 if (dev->flags ^ old_flags) {
6145 dev_change_rx_flags(dev, IFF_ALLMULTI);
6146 dev_set_rx_mode(dev);
6148 __dev_notify_flags(dev, old_flags,
6149 dev->gflags ^ old_gflags);
6155 * dev_set_allmulti - update allmulti count on a device
6159 * Add or remove reception of all multicast frames to a device. While the
6160 * count in the device remains above zero the interface remains listening
6161 * to all interfaces. Once it hits zero the device reverts back to normal
6162 * filtering operation. A negative @inc value is used to drop the counter
6163 * when releasing a resource needing all multicasts.
6164 * Return 0 if successful or a negative errno code on error.
6167 int dev_set_allmulti(struct net_device *dev, int inc)
6169 return __dev_set_allmulti(dev, inc, true);
6171 EXPORT_SYMBOL(dev_set_allmulti);
6174 * Upload unicast and multicast address lists to device and
6175 * configure RX filtering. When the device doesn't support unicast
6176 * filtering it is put in promiscuous mode while unicast addresses
6179 void __dev_set_rx_mode(struct net_device *dev)
6181 const struct net_device_ops *ops = dev->netdev_ops;
6183 /* dev_open will call this function so the list will stay sane. */
6184 if (!(dev->flags&IFF_UP))
6187 if (!netif_device_present(dev))
6190 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
6191 /* Unicast addresses changes may only happen under the rtnl,
6192 * therefore calling __dev_set_promiscuity here is safe.
6194 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
6195 __dev_set_promiscuity(dev, 1, false);
6196 dev->uc_promisc = true;
6197 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
6198 __dev_set_promiscuity(dev, -1, false);
6199 dev->uc_promisc = false;
6203 if (ops->ndo_set_rx_mode)
6204 ops->ndo_set_rx_mode(dev);
6207 void dev_set_rx_mode(struct net_device *dev)
6209 netif_addr_lock_bh(dev);
6210 __dev_set_rx_mode(dev);
6211 netif_addr_unlock_bh(dev);
6215 * dev_get_flags - get flags reported to userspace
6218 * Get the combination of flag bits exported through APIs to userspace.
6220 unsigned int dev_get_flags(const struct net_device *dev)
6224 flags = (dev->flags & ~(IFF_PROMISC |
6229 (dev->gflags & (IFF_PROMISC |
6232 if (netif_running(dev)) {
6233 if (netif_oper_up(dev))
6234 flags |= IFF_RUNNING;
6235 if (netif_carrier_ok(dev))
6236 flags |= IFF_LOWER_UP;
6237 if (netif_dormant(dev))
6238 flags |= IFF_DORMANT;
6243 EXPORT_SYMBOL(dev_get_flags);
6245 int __dev_change_flags(struct net_device *dev, unsigned int flags)
6247 unsigned int old_flags = dev->flags;
6253 * Set the flags on our device.
6256 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
6257 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
6259 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6263 * Load in the correct multicast list now the flags have changed.
6266 if ((old_flags ^ flags) & IFF_MULTICAST)
6267 dev_change_rx_flags(dev, IFF_MULTICAST);
6269 dev_set_rx_mode(dev);
6272 * Have we downed the interface. We handle IFF_UP ourselves
6273 * according to user attempts to set it, rather than blindly
6278 if ((old_flags ^ flags) & IFF_UP)
6279 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
6281 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6282 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6283 unsigned int old_flags = dev->flags;
6285 dev->gflags ^= IFF_PROMISC;
6287 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6288 if (dev->flags != old_flags)
6289 dev_set_rx_mode(dev);
6292 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6293 is important. Some (broken) drivers set IFF_PROMISC, when
6294 IFF_ALLMULTI is requested not asking us and not reporting.
6296 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6297 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6299 dev->gflags ^= IFF_ALLMULTI;
6300 __dev_set_allmulti(dev, inc, false);
6306 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6307 unsigned int gchanges)
6309 unsigned int changes = dev->flags ^ old_flags;
6312 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6314 if (changes & IFF_UP) {
6315 if (dev->flags & IFF_UP)
6316 call_netdevice_notifiers(NETDEV_UP, dev);
6318 call_netdevice_notifiers(NETDEV_DOWN, dev);
6321 if (dev->flags & IFF_UP &&
6322 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6323 struct netdev_notifier_change_info change_info;
6325 change_info.flags_changed = changes;
6326 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6332 * dev_change_flags - change device settings
6334 * @flags: device state flags
6336 * Change settings on device based state flags. The flags are
6337 * in the userspace exported format.
6339 int dev_change_flags(struct net_device *dev, unsigned int flags)
6342 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6344 ret = __dev_change_flags(dev, flags);
6348 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6349 __dev_notify_flags(dev, old_flags, changes);
6352 EXPORT_SYMBOL(dev_change_flags);
6354 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6356 const struct net_device_ops *ops = dev->netdev_ops;
6358 if (ops->ndo_change_mtu)
6359 return ops->ndo_change_mtu(dev, new_mtu);
6366 * dev_set_mtu - Change maximum transfer unit
6368 * @new_mtu: new transfer unit
6370 * Change the maximum transfer size of the network device.
6372 int dev_set_mtu(struct net_device *dev, int new_mtu)
6376 if (new_mtu == dev->mtu)
6379 /* MTU must be positive. */
6383 if (!netif_device_present(dev))
6386 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6387 err = notifier_to_errno(err);
6391 orig_mtu = dev->mtu;
6392 err = __dev_set_mtu(dev, new_mtu);
6395 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6396 err = notifier_to_errno(err);
6398 /* setting mtu back and notifying everyone again,
6399 * so that they have a chance to revert changes.
6401 __dev_set_mtu(dev, orig_mtu);
6402 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6407 EXPORT_SYMBOL(dev_set_mtu);
6410 * dev_set_group - Change group this device belongs to
6412 * @new_group: group this device should belong to
6414 void dev_set_group(struct net_device *dev, int new_group)
6416 dev->group = new_group;
6418 EXPORT_SYMBOL(dev_set_group);
6421 * dev_set_mac_address - Change Media Access Control Address
6425 * Change the hardware (MAC) address of the device
6427 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6429 const struct net_device_ops *ops = dev->netdev_ops;
6432 if (!ops->ndo_set_mac_address)
6434 if (sa->sa_family != dev->type)
6436 if (!netif_device_present(dev))
6438 err = ops->ndo_set_mac_address(dev, sa);
6441 dev->addr_assign_type = NET_ADDR_SET;
6442 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6443 add_device_randomness(dev->dev_addr, dev->addr_len);
6446 EXPORT_SYMBOL(dev_set_mac_address);
6449 * dev_change_carrier - Change device carrier
6451 * @new_carrier: new value
6453 * Change device carrier
6455 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6457 const struct net_device_ops *ops = dev->netdev_ops;
6459 if (!ops->ndo_change_carrier)
6461 if (!netif_device_present(dev))
6463 return ops->ndo_change_carrier(dev, new_carrier);
6465 EXPORT_SYMBOL(dev_change_carrier);
6468 * dev_get_phys_port_id - Get device physical port ID
6472 * Get device physical port ID
6474 int dev_get_phys_port_id(struct net_device *dev,
6475 struct netdev_phys_item_id *ppid)
6477 const struct net_device_ops *ops = dev->netdev_ops;
6479 if (!ops->ndo_get_phys_port_id)
6481 return ops->ndo_get_phys_port_id(dev, ppid);
6483 EXPORT_SYMBOL(dev_get_phys_port_id);
6486 * dev_get_phys_port_name - Get device physical port name
6489 * @len: limit of bytes to copy to name
6491 * Get device physical port name
6493 int dev_get_phys_port_name(struct net_device *dev,
6494 char *name, size_t len)
6496 const struct net_device_ops *ops = dev->netdev_ops;
6498 if (!ops->ndo_get_phys_port_name)
6500 return ops->ndo_get_phys_port_name(dev, name, len);
6502 EXPORT_SYMBOL(dev_get_phys_port_name);
6505 * dev_change_proto_down - update protocol port state information
6507 * @proto_down: new value
6509 * This info can be used by switch drivers to set the phys state of the
6512 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6514 const struct net_device_ops *ops = dev->netdev_ops;
6516 if (!ops->ndo_change_proto_down)
6518 if (!netif_device_present(dev))
6520 return ops->ndo_change_proto_down(dev, proto_down);
6522 EXPORT_SYMBOL(dev_change_proto_down);
6525 * dev_new_index - allocate an ifindex
6526 * @net: the applicable net namespace
6528 * Returns a suitable unique value for a new device interface
6529 * number. The caller must hold the rtnl semaphore or the
6530 * dev_base_lock to be sure it remains unique.
6532 static int dev_new_index(struct net *net)
6534 int ifindex = net->ifindex;
6538 if (!__dev_get_by_index(net, ifindex))
6539 return net->ifindex = ifindex;
6543 /* Delayed registration/unregisteration */
6544 static LIST_HEAD(net_todo_list);
6545 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6547 static void net_set_todo(struct net_device *dev)
6549 list_add_tail(&dev->todo_list, &net_todo_list);
6550 dev_net(dev)->dev_unreg_count++;
6553 static void rollback_registered_many(struct list_head *head)
6555 struct net_device *dev, *tmp;
6556 LIST_HEAD(close_head);
6558 BUG_ON(dev_boot_phase);
6561 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6562 /* Some devices call without registering
6563 * for initialization unwind. Remove those
6564 * devices and proceed with the remaining.
6566 if (dev->reg_state == NETREG_UNINITIALIZED) {
6567 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6571 list_del(&dev->unreg_list);
6574 dev->dismantle = true;
6575 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6578 /* If device is running, close it first. */
6579 list_for_each_entry(dev, head, unreg_list)
6580 list_add_tail(&dev->close_list, &close_head);
6581 dev_close_many(&close_head, true);
6583 list_for_each_entry(dev, head, unreg_list) {
6584 /* And unlink it from device chain. */
6585 unlist_netdevice(dev);
6587 dev->reg_state = NETREG_UNREGISTERING;
6588 on_each_cpu(flush_backlog, dev, 1);
6593 list_for_each_entry(dev, head, unreg_list) {
6594 struct sk_buff *skb = NULL;
6596 /* Shutdown queueing discipline. */
6600 /* Notify protocols, that we are about to destroy
6601 this device. They should clean all the things.
6603 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6605 if (!dev->rtnl_link_ops ||
6606 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6607 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6611 * Flush the unicast and multicast chains
6616 if (dev->netdev_ops->ndo_uninit)
6617 dev->netdev_ops->ndo_uninit(dev);
6620 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6622 /* Notifier chain MUST detach us all upper devices. */
6623 WARN_ON(netdev_has_any_upper_dev(dev));
6625 /* Remove entries from kobject tree */
6626 netdev_unregister_kobject(dev);
6628 /* Remove XPS queueing entries */
6629 netif_reset_xps_queues_gt(dev, 0);
6635 list_for_each_entry(dev, head, unreg_list)
6639 static void rollback_registered(struct net_device *dev)
6643 list_add(&dev->unreg_list, &single);
6644 rollback_registered_many(&single);
6648 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6649 struct net_device *upper, netdev_features_t features)
6651 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6652 netdev_features_t feature;
6655 for_each_netdev_feature(&upper_disables, feature_bit) {
6656 feature = __NETIF_F_BIT(feature_bit);
6657 if (!(upper->wanted_features & feature)
6658 && (features & feature)) {
6659 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6660 &feature, upper->name);
6661 features &= ~feature;
6668 static void netdev_sync_lower_features(struct net_device *upper,
6669 struct net_device *lower, netdev_features_t features)
6671 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6672 netdev_features_t feature;
6675 for_each_netdev_feature(&upper_disables, feature_bit) {
6676 feature = __NETIF_F_BIT(feature_bit);
6677 if (!(features & feature) && (lower->features & feature)) {
6678 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6679 &feature, lower->name);
6680 lower->wanted_features &= ~feature;
6681 netdev_update_features(lower);
6683 if (unlikely(lower->features & feature))
6684 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6685 &feature, lower->name);
6690 static netdev_features_t netdev_fix_features(struct net_device *dev,
6691 netdev_features_t features)
6693 /* Fix illegal checksum combinations */
6694 if ((features & NETIF_F_HW_CSUM) &&
6695 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6696 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6697 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6700 /* TSO requires that SG is present as well. */
6701 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6702 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6703 features &= ~NETIF_F_ALL_TSO;
6706 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6707 !(features & NETIF_F_IP_CSUM)) {
6708 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6709 features &= ~NETIF_F_TSO;
6710 features &= ~NETIF_F_TSO_ECN;
6713 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6714 !(features & NETIF_F_IPV6_CSUM)) {
6715 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6716 features &= ~NETIF_F_TSO6;
6719 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
6720 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
6721 features &= ~NETIF_F_TSO_MANGLEID;
6723 /* TSO ECN requires that TSO is present as well. */
6724 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6725 features &= ~NETIF_F_TSO_ECN;
6727 /* Software GSO depends on SG. */
6728 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6729 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6730 features &= ~NETIF_F_GSO;
6733 /* UFO needs SG and checksumming */
6734 if (features & NETIF_F_UFO) {
6735 /* maybe split UFO into V4 and V6? */
6736 if (!(features & NETIF_F_HW_CSUM) &&
6737 ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) !=
6738 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) {
6740 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6741 features &= ~NETIF_F_UFO;
6744 if (!(features & NETIF_F_SG)) {
6746 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6747 features &= ~NETIF_F_UFO;
6751 /* GSO partial features require GSO partial be set */
6752 if ((features & dev->gso_partial_features) &&
6753 !(features & NETIF_F_GSO_PARTIAL)) {
6755 "Dropping partially supported GSO features since no GSO partial.\n");
6756 features &= ~dev->gso_partial_features;
6759 #ifdef CONFIG_NET_RX_BUSY_POLL
6760 if (dev->netdev_ops->ndo_busy_poll)
6761 features |= NETIF_F_BUSY_POLL;
6764 features &= ~NETIF_F_BUSY_POLL;
6769 int __netdev_update_features(struct net_device *dev)
6771 struct net_device *upper, *lower;
6772 netdev_features_t features;
6773 struct list_head *iter;
6778 features = netdev_get_wanted_features(dev);
6780 if (dev->netdev_ops->ndo_fix_features)
6781 features = dev->netdev_ops->ndo_fix_features(dev, features);
6783 /* driver might be less strict about feature dependencies */
6784 features = netdev_fix_features(dev, features);
6786 /* some features can't be enabled if they're off an an upper device */
6787 netdev_for_each_upper_dev_rcu(dev, upper, iter)
6788 features = netdev_sync_upper_features(dev, upper, features);
6790 if (dev->features == features)
6793 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6794 &dev->features, &features);
6796 if (dev->netdev_ops->ndo_set_features)
6797 err = dev->netdev_ops->ndo_set_features(dev, features);
6801 if (unlikely(err < 0)) {
6803 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6804 err, &features, &dev->features);
6805 /* return non-0 since some features might have changed and
6806 * it's better to fire a spurious notification than miss it
6812 /* some features must be disabled on lower devices when disabled
6813 * on an upper device (think: bonding master or bridge)
6815 netdev_for_each_lower_dev(dev, lower, iter)
6816 netdev_sync_lower_features(dev, lower, features);
6819 dev->features = features;
6821 return err < 0 ? 0 : 1;
6825 * netdev_update_features - recalculate device features
6826 * @dev: the device to check
6828 * Recalculate dev->features set and send notifications if it
6829 * has changed. Should be called after driver or hardware dependent
6830 * conditions might have changed that influence the features.
6832 void netdev_update_features(struct net_device *dev)
6834 if (__netdev_update_features(dev))
6835 netdev_features_change(dev);
6837 EXPORT_SYMBOL(netdev_update_features);
6840 * netdev_change_features - recalculate device features
6841 * @dev: the device to check
6843 * Recalculate dev->features set and send notifications even
6844 * if they have not changed. Should be called instead of
6845 * netdev_update_features() if also dev->vlan_features might
6846 * have changed to allow the changes to be propagated to stacked
6849 void netdev_change_features(struct net_device *dev)
6851 __netdev_update_features(dev);
6852 netdev_features_change(dev);
6854 EXPORT_SYMBOL(netdev_change_features);
6857 * netif_stacked_transfer_operstate - transfer operstate
6858 * @rootdev: the root or lower level device to transfer state from
6859 * @dev: the device to transfer operstate to
6861 * Transfer operational state from root to device. This is normally
6862 * called when a stacking relationship exists between the root
6863 * device and the device(a leaf device).
6865 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6866 struct net_device *dev)
6868 if (rootdev->operstate == IF_OPER_DORMANT)
6869 netif_dormant_on(dev);
6871 netif_dormant_off(dev);
6873 if (netif_carrier_ok(rootdev)) {
6874 if (!netif_carrier_ok(dev))
6875 netif_carrier_on(dev);
6877 if (netif_carrier_ok(dev))
6878 netif_carrier_off(dev);
6881 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6884 static int netif_alloc_rx_queues(struct net_device *dev)
6886 unsigned int i, count = dev->num_rx_queues;
6887 struct netdev_rx_queue *rx;
6888 size_t sz = count * sizeof(*rx);
6892 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6900 for (i = 0; i < count; i++)
6906 static void netdev_init_one_queue(struct net_device *dev,
6907 struct netdev_queue *queue, void *_unused)
6909 /* Initialize queue lock */
6910 spin_lock_init(&queue->_xmit_lock);
6911 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6912 queue->xmit_lock_owner = -1;
6913 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6916 dql_init(&queue->dql, HZ);
6920 static void netif_free_tx_queues(struct net_device *dev)
6925 static int netif_alloc_netdev_queues(struct net_device *dev)
6927 unsigned int count = dev->num_tx_queues;
6928 struct netdev_queue *tx;
6929 size_t sz = count * sizeof(*tx);
6931 if (count < 1 || count > 0xffff)
6934 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6942 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6943 spin_lock_init(&dev->tx_global_lock);
6948 void netif_tx_stop_all_queues(struct net_device *dev)
6952 for (i = 0; i < dev->num_tx_queues; i++) {
6953 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
6954 netif_tx_stop_queue(txq);
6957 EXPORT_SYMBOL(netif_tx_stop_all_queues);
6960 * register_netdevice - register a network device
6961 * @dev: device to register
6963 * Take a completed network device structure and add it to the kernel
6964 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6965 * chain. 0 is returned on success. A negative errno code is returned
6966 * on a failure to set up the device, or if the name is a duplicate.
6968 * Callers must hold the rtnl semaphore. You may want
6969 * register_netdev() instead of this.
6972 * The locking appears insufficient to guarantee two parallel registers
6973 * will not get the same name.
6976 int register_netdevice(struct net_device *dev)
6979 struct net *net = dev_net(dev);
6981 BUG_ON(dev_boot_phase);
6986 /* When net_device's are persistent, this will be fatal. */
6987 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6990 spin_lock_init(&dev->addr_list_lock);
6991 netdev_set_addr_lockdep_class(dev);
6993 ret = dev_get_valid_name(net, dev, dev->name);
6997 /* Init, if this function is available */
6998 if (dev->netdev_ops->ndo_init) {
6999 ret = dev->netdev_ops->ndo_init(dev);
7007 if (((dev->hw_features | dev->features) &
7008 NETIF_F_HW_VLAN_CTAG_FILTER) &&
7009 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
7010 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
7011 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
7018 dev->ifindex = dev_new_index(net);
7019 else if (__dev_get_by_index(net, dev->ifindex))
7022 /* Transfer changeable features to wanted_features and enable
7023 * software offloads (GSO and GRO).
7025 dev->hw_features |= NETIF_F_SOFT_FEATURES;
7026 dev->features |= NETIF_F_SOFT_FEATURES;
7027 dev->wanted_features = dev->features & dev->hw_features;
7029 if (!(dev->flags & IFF_LOOPBACK))
7030 dev->hw_features |= NETIF_F_NOCACHE_COPY;
7032 /* If IPv4 TCP segmentation offload is supported we should also
7033 * allow the device to enable segmenting the frame with the option
7034 * of ignoring a static IP ID value. This doesn't enable the
7035 * feature itself but allows the user to enable it later.
7037 if (dev->hw_features & NETIF_F_TSO)
7038 dev->hw_features |= NETIF_F_TSO_MANGLEID;
7039 if (dev->vlan_features & NETIF_F_TSO)
7040 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
7041 if (dev->mpls_features & NETIF_F_TSO)
7042 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
7043 if (dev->hw_enc_features & NETIF_F_TSO)
7044 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
7046 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
7048 dev->vlan_features |= NETIF_F_HIGHDMA;
7050 /* Make NETIF_F_SG inheritable to tunnel devices.
7052 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
7054 /* Make NETIF_F_SG inheritable to MPLS.
7056 dev->mpls_features |= NETIF_F_SG;
7058 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
7059 ret = notifier_to_errno(ret);
7063 ret = netdev_register_kobject(dev);
7066 dev->reg_state = NETREG_REGISTERED;
7068 __netdev_update_features(dev);
7071 * Default initial state at registry is that the
7072 * device is present.
7075 set_bit(__LINK_STATE_PRESENT, &dev->state);
7077 linkwatch_init_dev(dev);
7079 dev_init_scheduler(dev);
7081 list_netdevice(dev);
7082 add_device_randomness(dev->dev_addr, dev->addr_len);
7084 /* If the device has permanent device address, driver should
7085 * set dev_addr and also addr_assign_type should be set to
7086 * NET_ADDR_PERM (default value).
7088 if (dev->addr_assign_type == NET_ADDR_PERM)
7089 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
7091 /* Notify protocols, that a new device appeared. */
7092 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
7093 ret = notifier_to_errno(ret);
7095 rollback_registered(dev);
7096 dev->reg_state = NETREG_UNREGISTERED;
7099 * Prevent userspace races by waiting until the network
7100 * device is fully setup before sending notifications.
7102 if (!dev->rtnl_link_ops ||
7103 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7104 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7110 if (dev->netdev_ops->ndo_uninit)
7111 dev->netdev_ops->ndo_uninit(dev);
7114 EXPORT_SYMBOL(register_netdevice);
7117 * init_dummy_netdev - init a dummy network device for NAPI
7118 * @dev: device to init
7120 * This takes a network device structure and initialize the minimum
7121 * amount of fields so it can be used to schedule NAPI polls without
7122 * registering a full blown interface. This is to be used by drivers
7123 * that need to tie several hardware interfaces to a single NAPI
7124 * poll scheduler due to HW limitations.
7126 int init_dummy_netdev(struct net_device *dev)
7128 /* Clear everything. Note we don't initialize spinlocks
7129 * are they aren't supposed to be taken by any of the
7130 * NAPI code and this dummy netdev is supposed to be
7131 * only ever used for NAPI polls
7133 memset(dev, 0, sizeof(struct net_device));
7135 /* make sure we BUG if trying to hit standard
7136 * register/unregister code path
7138 dev->reg_state = NETREG_DUMMY;
7140 /* NAPI wants this */
7141 INIT_LIST_HEAD(&dev->napi_list);
7143 /* a dummy interface is started by default */
7144 set_bit(__LINK_STATE_PRESENT, &dev->state);
7145 set_bit(__LINK_STATE_START, &dev->state);
7147 /* Note : We dont allocate pcpu_refcnt for dummy devices,
7148 * because users of this 'device' dont need to change
7154 EXPORT_SYMBOL_GPL(init_dummy_netdev);
7158 * register_netdev - register a network device
7159 * @dev: device to register
7161 * Take a completed network device structure and add it to the kernel
7162 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7163 * chain. 0 is returned on success. A negative errno code is returned
7164 * on a failure to set up the device, or if the name is a duplicate.
7166 * This is a wrapper around register_netdevice that takes the rtnl semaphore
7167 * and expands the device name if you passed a format string to
7170 int register_netdev(struct net_device *dev)
7175 err = register_netdevice(dev);
7179 EXPORT_SYMBOL(register_netdev);
7181 int netdev_refcnt_read(const struct net_device *dev)
7185 for_each_possible_cpu(i)
7186 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
7189 EXPORT_SYMBOL(netdev_refcnt_read);
7192 * netdev_wait_allrefs - wait until all references are gone.
7193 * @dev: target net_device
7195 * This is called when unregistering network devices.
7197 * Any protocol or device that holds a reference should register
7198 * for netdevice notification, and cleanup and put back the
7199 * reference if they receive an UNREGISTER event.
7200 * We can get stuck here if buggy protocols don't correctly
7203 static void netdev_wait_allrefs(struct net_device *dev)
7205 unsigned long rebroadcast_time, warning_time;
7208 linkwatch_forget_dev(dev);
7210 rebroadcast_time = warning_time = jiffies;
7211 refcnt = netdev_refcnt_read(dev);
7213 while (refcnt != 0) {
7214 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
7217 /* Rebroadcast unregister notification */
7218 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7224 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7225 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
7227 /* We must not have linkwatch events
7228 * pending on unregister. If this
7229 * happens, we simply run the queue
7230 * unscheduled, resulting in a noop
7233 linkwatch_run_queue();
7238 rebroadcast_time = jiffies;
7243 refcnt = netdev_refcnt_read(dev);
7245 if (time_after(jiffies, warning_time + 10 * HZ)) {
7246 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
7248 warning_time = jiffies;
7257 * register_netdevice(x1);
7258 * register_netdevice(x2);
7260 * unregister_netdevice(y1);
7261 * unregister_netdevice(y2);
7267 * We are invoked by rtnl_unlock().
7268 * This allows us to deal with problems:
7269 * 1) We can delete sysfs objects which invoke hotplug
7270 * without deadlocking with linkwatch via keventd.
7271 * 2) Since we run with the RTNL semaphore not held, we can sleep
7272 * safely in order to wait for the netdev refcnt to drop to zero.
7274 * We must not return until all unregister events added during
7275 * the interval the lock was held have been completed.
7277 void netdev_run_todo(void)
7279 struct list_head list;
7281 /* Snapshot list, allow later requests */
7282 list_replace_init(&net_todo_list, &list);
7287 /* Wait for rcu callbacks to finish before next phase */
7288 if (!list_empty(&list))
7291 while (!list_empty(&list)) {
7292 struct net_device *dev
7293 = list_first_entry(&list, struct net_device, todo_list);
7294 list_del(&dev->todo_list);
7297 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7300 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
7301 pr_err("network todo '%s' but state %d\n",
7302 dev->name, dev->reg_state);
7307 dev->reg_state = NETREG_UNREGISTERED;
7309 netdev_wait_allrefs(dev);
7312 BUG_ON(netdev_refcnt_read(dev));
7313 BUG_ON(!list_empty(&dev->ptype_all));
7314 BUG_ON(!list_empty(&dev->ptype_specific));
7315 WARN_ON(rcu_access_pointer(dev->ip_ptr));
7316 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
7317 WARN_ON(dev->dn_ptr);
7319 if (dev->destructor)
7320 dev->destructor(dev);
7322 /* Report a network device has been unregistered */
7324 dev_net(dev)->dev_unreg_count--;
7326 wake_up(&netdev_unregistering_wq);
7328 /* Free network device */
7329 kobject_put(&dev->dev.kobj);
7333 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
7334 * all the same fields in the same order as net_device_stats, with only
7335 * the type differing, but rtnl_link_stats64 may have additional fields
7336 * at the end for newer counters.
7338 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7339 const struct net_device_stats *netdev_stats)
7341 #if BITS_PER_LONG == 64
7342 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
7343 memcpy(stats64, netdev_stats, sizeof(*stats64));
7344 /* zero out counters that only exist in rtnl_link_stats64 */
7345 memset((char *)stats64 + sizeof(*netdev_stats), 0,
7346 sizeof(*stats64) - sizeof(*netdev_stats));
7348 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
7349 const unsigned long *src = (const unsigned long *)netdev_stats;
7350 u64 *dst = (u64 *)stats64;
7352 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
7353 for (i = 0; i < n; i++)
7355 /* zero out counters that only exist in rtnl_link_stats64 */
7356 memset((char *)stats64 + n * sizeof(u64), 0,
7357 sizeof(*stats64) - n * sizeof(u64));
7360 EXPORT_SYMBOL(netdev_stats_to_stats64);
7363 * dev_get_stats - get network device statistics
7364 * @dev: device to get statistics from
7365 * @storage: place to store stats
7367 * Get network statistics from device. Return @storage.
7368 * The device driver may provide its own method by setting
7369 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7370 * otherwise the internal statistics structure is used.
7372 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
7373 struct rtnl_link_stats64 *storage)
7375 const struct net_device_ops *ops = dev->netdev_ops;
7377 if (ops->ndo_get_stats64) {
7378 memset(storage, 0, sizeof(*storage));
7379 ops->ndo_get_stats64(dev, storage);
7380 } else if (ops->ndo_get_stats) {
7381 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7383 netdev_stats_to_stats64(storage, &dev->stats);
7385 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
7386 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
7387 storage->rx_nohandler += atomic_long_read(&dev->rx_nohandler);
7390 EXPORT_SYMBOL(dev_get_stats);
7392 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7394 struct netdev_queue *queue = dev_ingress_queue(dev);
7396 #ifdef CONFIG_NET_CLS_ACT
7399 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7402 netdev_init_one_queue(dev, queue, NULL);
7403 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7404 queue->qdisc_sleeping = &noop_qdisc;
7405 rcu_assign_pointer(dev->ingress_queue, queue);
7410 static const struct ethtool_ops default_ethtool_ops;
7412 void netdev_set_default_ethtool_ops(struct net_device *dev,
7413 const struct ethtool_ops *ops)
7415 if (dev->ethtool_ops == &default_ethtool_ops)
7416 dev->ethtool_ops = ops;
7418 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7420 void netdev_freemem(struct net_device *dev)
7422 char *addr = (char *)dev - dev->padded;
7428 * alloc_netdev_mqs - allocate network device
7429 * @sizeof_priv: size of private data to allocate space for
7430 * @name: device name format string
7431 * @name_assign_type: origin of device name
7432 * @setup: callback to initialize device
7433 * @txqs: the number of TX subqueues to allocate
7434 * @rxqs: the number of RX subqueues to allocate
7436 * Allocates a struct net_device with private data area for driver use
7437 * and performs basic initialization. Also allocates subqueue structs
7438 * for each queue on the device.
7440 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7441 unsigned char name_assign_type,
7442 void (*setup)(struct net_device *),
7443 unsigned int txqs, unsigned int rxqs)
7445 struct net_device *dev;
7447 struct net_device *p;
7449 BUG_ON(strlen(name) >= sizeof(dev->name));
7452 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7458 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7463 alloc_size = sizeof(struct net_device);
7465 /* ensure 32-byte alignment of private area */
7466 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7467 alloc_size += sizeof_priv;
7469 /* ensure 32-byte alignment of whole construct */
7470 alloc_size += NETDEV_ALIGN - 1;
7472 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7474 p = vzalloc(alloc_size);
7478 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7479 dev->padded = (char *)dev - (char *)p;
7481 dev->pcpu_refcnt = alloc_percpu(int);
7482 if (!dev->pcpu_refcnt)
7485 if (dev_addr_init(dev))
7491 dev_net_set(dev, &init_net);
7493 dev->gso_max_size = GSO_MAX_SIZE;
7494 dev->gso_max_segs = GSO_MAX_SEGS;
7496 INIT_LIST_HEAD(&dev->napi_list);
7497 INIT_LIST_HEAD(&dev->unreg_list);
7498 INIT_LIST_HEAD(&dev->close_list);
7499 INIT_LIST_HEAD(&dev->link_watch_list);
7500 INIT_LIST_HEAD(&dev->adj_list.upper);
7501 INIT_LIST_HEAD(&dev->adj_list.lower);
7502 INIT_LIST_HEAD(&dev->all_adj_list.upper);
7503 INIT_LIST_HEAD(&dev->all_adj_list.lower);
7504 INIT_LIST_HEAD(&dev->ptype_all);
7505 INIT_LIST_HEAD(&dev->ptype_specific);
7506 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7509 if (!dev->tx_queue_len) {
7510 dev->priv_flags |= IFF_NO_QUEUE;
7511 dev->tx_queue_len = 1;
7514 dev->num_tx_queues = txqs;
7515 dev->real_num_tx_queues = txqs;
7516 if (netif_alloc_netdev_queues(dev))
7520 dev->num_rx_queues = rxqs;
7521 dev->real_num_rx_queues = rxqs;
7522 if (netif_alloc_rx_queues(dev))
7526 strcpy(dev->name, name);
7527 dev->name_assign_type = name_assign_type;
7528 dev->group = INIT_NETDEV_GROUP;
7529 if (!dev->ethtool_ops)
7530 dev->ethtool_ops = &default_ethtool_ops;
7532 nf_hook_ingress_init(dev);
7541 free_percpu(dev->pcpu_refcnt);
7543 netdev_freemem(dev);
7546 EXPORT_SYMBOL(alloc_netdev_mqs);
7549 * free_netdev - free network device
7552 * This function does the last stage of destroying an allocated device
7553 * interface. The reference to the device object is released.
7554 * If this is the last reference then it will be freed.
7555 * Must be called in process context.
7557 void free_netdev(struct net_device *dev)
7559 struct napi_struct *p, *n;
7562 netif_free_tx_queues(dev);
7567 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7569 /* Flush device addresses */
7570 dev_addr_flush(dev);
7572 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7575 free_percpu(dev->pcpu_refcnt);
7576 dev->pcpu_refcnt = NULL;
7578 /* Compatibility with error handling in drivers */
7579 if (dev->reg_state == NETREG_UNINITIALIZED) {
7580 netdev_freemem(dev);
7584 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7585 dev->reg_state = NETREG_RELEASED;
7587 /* will free via device release */
7588 put_device(&dev->dev);
7590 EXPORT_SYMBOL(free_netdev);
7593 * synchronize_net - Synchronize with packet receive processing
7595 * Wait for packets currently being received to be done.
7596 * Does not block later packets from starting.
7598 void synchronize_net(void)
7601 if (rtnl_is_locked())
7602 synchronize_rcu_expedited();
7606 EXPORT_SYMBOL(synchronize_net);
7609 * unregister_netdevice_queue - remove device from the kernel
7613 * This function shuts down a device interface and removes it
7614 * from the kernel tables.
7615 * If head not NULL, device is queued to be unregistered later.
7617 * Callers must hold the rtnl semaphore. You may want
7618 * unregister_netdev() instead of this.
7621 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7626 list_move_tail(&dev->unreg_list, head);
7628 rollback_registered(dev);
7629 /* Finish processing unregister after unlock */
7633 EXPORT_SYMBOL(unregister_netdevice_queue);
7636 * unregister_netdevice_many - unregister many devices
7637 * @head: list of devices
7639 * Note: As most callers use a stack allocated list_head,
7640 * we force a list_del() to make sure stack wont be corrupted later.
7642 void unregister_netdevice_many(struct list_head *head)
7644 struct net_device *dev;
7646 if (!list_empty(head)) {
7647 rollback_registered_many(head);
7648 list_for_each_entry(dev, head, unreg_list)
7653 EXPORT_SYMBOL(unregister_netdevice_many);
7656 * unregister_netdev - remove device from the kernel
7659 * This function shuts down a device interface and removes it
7660 * from the kernel tables.
7662 * This is just a wrapper for unregister_netdevice that takes
7663 * the rtnl semaphore. In general you want to use this and not
7664 * unregister_netdevice.
7666 void unregister_netdev(struct net_device *dev)
7669 unregister_netdevice(dev);
7672 EXPORT_SYMBOL(unregister_netdev);
7675 * dev_change_net_namespace - move device to different nethost namespace
7677 * @net: network namespace
7678 * @pat: If not NULL name pattern to try if the current device name
7679 * is already taken in the destination network namespace.
7681 * This function shuts down a device interface and moves it
7682 * to a new network namespace. On success 0 is returned, on
7683 * a failure a netagive errno code is returned.
7685 * Callers must hold the rtnl semaphore.
7688 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7694 /* Don't allow namespace local devices to be moved. */
7696 if (dev->features & NETIF_F_NETNS_LOCAL)
7699 /* Ensure the device has been registrered */
7700 if (dev->reg_state != NETREG_REGISTERED)
7703 /* Get out if there is nothing todo */
7705 if (net_eq(dev_net(dev), net))
7708 /* Pick the destination device name, and ensure
7709 * we can use it in the destination network namespace.
7712 if (__dev_get_by_name(net, dev->name)) {
7713 /* We get here if we can't use the current device name */
7716 if (dev_get_valid_name(net, dev, pat) < 0)
7721 * And now a mini version of register_netdevice unregister_netdevice.
7724 /* If device is running close it first. */
7727 /* And unlink it from device chain */
7729 unlist_netdevice(dev);
7733 /* Shutdown queueing discipline. */
7736 /* Notify protocols, that we are about to destroy
7737 this device. They should clean all the things.
7739 Note that dev->reg_state stays at NETREG_REGISTERED.
7740 This is wanted because this way 8021q and macvlan know
7741 the device is just moving and can keep their slaves up.
7743 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7745 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7746 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7749 * Flush the unicast and multicast chains
7754 /* Send a netdev-removed uevent to the old namespace */
7755 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7756 netdev_adjacent_del_links(dev);
7758 /* Actually switch the network namespace */
7759 dev_net_set(dev, net);
7761 /* If there is an ifindex conflict assign a new one */
7762 if (__dev_get_by_index(net, dev->ifindex))
7763 dev->ifindex = dev_new_index(net);
7765 /* Send a netdev-add uevent to the new namespace */
7766 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7767 netdev_adjacent_add_links(dev);
7769 /* Fixup kobjects */
7770 err = device_rename(&dev->dev, dev->name);
7773 /* Add the device back in the hashes */
7774 list_netdevice(dev);
7776 /* Notify protocols, that a new device appeared. */
7777 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7780 * Prevent userspace races by waiting until the network
7781 * device is fully setup before sending notifications.
7783 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7790 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7792 static int dev_cpu_callback(struct notifier_block *nfb,
7793 unsigned long action,
7796 struct sk_buff **list_skb;
7797 struct sk_buff *skb;
7798 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7799 struct softnet_data *sd, *oldsd;
7801 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7804 local_irq_disable();
7805 cpu = smp_processor_id();
7806 sd = &per_cpu(softnet_data, cpu);
7807 oldsd = &per_cpu(softnet_data, oldcpu);
7809 /* Find end of our completion_queue. */
7810 list_skb = &sd->completion_queue;
7812 list_skb = &(*list_skb)->next;
7813 /* Append completion queue from offline CPU. */
7814 *list_skb = oldsd->completion_queue;
7815 oldsd->completion_queue = NULL;
7817 /* Append output queue from offline CPU. */
7818 if (oldsd->output_queue) {
7819 *sd->output_queue_tailp = oldsd->output_queue;
7820 sd->output_queue_tailp = oldsd->output_queue_tailp;
7821 oldsd->output_queue = NULL;
7822 oldsd->output_queue_tailp = &oldsd->output_queue;
7824 /* Append NAPI poll list from offline CPU, with one exception :
7825 * process_backlog() must be called by cpu owning percpu backlog.
7826 * We properly handle process_queue & input_pkt_queue later.
7828 while (!list_empty(&oldsd->poll_list)) {
7829 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7833 list_del_init(&napi->poll_list);
7834 if (napi->poll == process_backlog)
7837 ____napi_schedule(sd, napi);
7840 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7843 /* Process offline CPU's input_pkt_queue */
7844 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7846 input_queue_head_incr(oldsd);
7848 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7850 input_queue_head_incr(oldsd);
7858 * netdev_increment_features - increment feature set by one
7859 * @all: current feature set
7860 * @one: new feature set
7861 * @mask: mask feature set
7863 * Computes a new feature set after adding a device with feature set
7864 * @one to the master device with current feature set @all. Will not
7865 * enable anything that is off in @mask. Returns the new feature set.
7867 netdev_features_t netdev_increment_features(netdev_features_t all,
7868 netdev_features_t one, netdev_features_t mask)
7870 if (mask & NETIF_F_HW_CSUM)
7871 mask |= NETIF_F_CSUM_MASK;
7872 mask |= NETIF_F_VLAN_CHALLENGED;
7874 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
7875 all &= one | ~NETIF_F_ALL_FOR_ALL;
7877 /* If one device supports hw checksumming, set for all. */
7878 if (all & NETIF_F_HW_CSUM)
7879 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
7883 EXPORT_SYMBOL(netdev_increment_features);
7885 static struct hlist_head * __net_init netdev_create_hash(void)
7888 struct hlist_head *hash;
7890 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7892 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7893 INIT_HLIST_HEAD(&hash[i]);
7898 /* Initialize per network namespace state */
7899 static int __net_init netdev_init(struct net *net)
7901 if (net != &init_net)
7902 INIT_LIST_HEAD(&net->dev_base_head);
7904 net->dev_name_head = netdev_create_hash();
7905 if (net->dev_name_head == NULL)
7908 net->dev_index_head = netdev_create_hash();
7909 if (net->dev_index_head == NULL)
7915 kfree(net->dev_name_head);
7921 * netdev_drivername - network driver for the device
7922 * @dev: network device
7924 * Determine network driver for device.
7926 const char *netdev_drivername(const struct net_device *dev)
7928 const struct device_driver *driver;
7929 const struct device *parent;
7930 const char *empty = "";
7932 parent = dev->dev.parent;
7936 driver = parent->driver;
7937 if (driver && driver->name)
7938 return driver->name;
7942 static void __netdev_printk(const char *level, const struct net_device *dev,
7943 struct va_format *vaf)
7945 if (dev && dev->dev.parent) {
7946 dev_printk_emit(level[1] - '0',
7949 dev_driver_string(dev->dev.parent),
7950 dev_name(dev->dev.parent),
7951 netdev_name(dev), netdev_reg_state(dev),
7954 printk("%s%s%s: %pV",
7955 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7957 printk("%s(NULL net_device): %pV", level, vaf);
7961 void netdev_printk(const char *level, const struct net_device *dev,
7962 const char *format, ...)
7964 struct va_format vaf;
7967 va_start(args, format);
7972 __netdev_printk(level, dev, &vaf);
7976 EXPORT_SYMBOL(netdev_printk);
7978 #define define_netdev_printk_level(func, level) \
7979 void func(const struct net_device *dev, const char *fmt, ...) \
7981 struct va_format vaf; \
7984 va_start(args, fmt); \
7989 __netdev_printk(level, dev, &vaf); \
7993 EXPORT_SYMBOL(func);
7995 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7996 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7997 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7998 define_netdev_printk_level(netdev_err, KERN_ERR);
7999 define_netdev_printk_level(netdev_warn, KERN_WARNING);
8000 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
8001 define_netdev_printk_level(netdev_info, KERN_INFO);
8003 static void __net_exit netdev_exit(struct net *net)
8005 kfree(net->dev_name_head);
8006 kfree(net->dev_index_head);
8009 static struct pernet_operations __net_initdata netdev_net_ops = {
8010 .init = netdev_init,
8011 .exit = netdev_exit,
8014 static void __net_exit default_device_exit(struct net *net)
8016 struct net_device *dev, *aux;
8018 * Push all migratable network devices back to the
8019 * initial network namespace
8022 for_each_netdev_safe(net, dev, aux) {
8024 char fb_name[IFNAMSIZ];
8026 /* Ignore unmoveable devices (i.e. loopback) */
8027 if (dev->features & NETIF_F_NETNS_LOCAL)
8030 /* Leave virtual devices for the generic cleanup */
8031 if (dev->rtnl_link_ops)
8034 /* Push remaining network devices to init_net */
8035 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
8036 err = dev_change_net_namespace(dev, &init_net, fb_name);
8038 pr_emerg("%s: failed to move %s to init_net: %d\n",
8039 __func__, dev->name, err);
8046 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
8048 /* Return with the rtnl_lock held when there are no network
8049 * devices unregistering in any network namespace in net_list.
8053 DEFINE_WAIT_FUNC(wait, woken_wake_function);
8055 add_wait_queue(&netdev_unregistering_wq, &wait);
8057 unregistering = false;
8059 list_for_each_entry(net, net_list, exit_list) {
8060 if (net->dev_unreg_count > 0) {
8061 unregistering = true;
8069 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
8071 remove_wait_queue(&netdev_unregistering_wq, &wait);
8074 static void __net_exit default_device_exit_batch(struct list_head *net_list)
8076 /* At exit all network devices most be removed from a network
8077 * namespace. Do this in the reverse order of registration.
8078 * Do this across as many network namespaces as possible to
8079 * improve batching efficiency.
8081 struct net_device *dev;
8083 LIST_HEAD(dev_kill_list);
8085 /* To prevent network device cleanup code from dereferencing
8086 * loopback devices or network devices that have been freed
8087 * wait here for all pending unregistrations to complete,
8088 * before unregistring the loopback device and allowing the
8089 * network namespace be freed.
8091 * The netdev todo list containing all network devices
8092 * unregistrations that happen in default_device_exit_batch
8093 * will run in the rtnl_unlock() at the end of
8094 * default_device_exit_batch.
8096 rtnl_lock_unregistering(net_list);
8097 list_for_each_entry(net, net_list, exit_list) {
8098 for_each_netdev_reverse(net, dev) {
8099 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
8100 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
8102 unregister_netdevice_queue(dev, &dev_kill_list);
8105 unregister_netdevice_many(&dev_kill_list);
8109 static struct pernet_operations __net_initdata default_device_ops = {
8110 .exit = default_device_exit,
8111 .exit_batch = default_device_exit_batch,
8115 * Initialize the DEV module. At boot time this walks the device list and
8116 * unhooks any devices that fail to initialise (normally hardware not
8117 * present) and leaves us with a valid list of present and active devices.
8122 * This is called single threaded during boot, so no need
8123 * to take the rtnl semaphore.
8125 static int __init net_dev_init(void)
8127 int i, rc = -ENOMEM;
8129 BUG_ON(!dev_boot_phase);
8131 if (dev_proc_init())
8134 if (netdev_kobject_init())
8137 INIT_LIST_HEAD(&ptype_all);
8138 for (i = 0; i < PTYPE_HASH_SIZE; i++)
8139 INIT_LIST_HEAD(&ptype_base[i]);
8141 INIT_LIST_HEAD(&offload_base);
8143 if (register_pernet_subsys(&netdev_net_ops))
8147 * Initialise the packet receive queues.
8150 for_each_possible_cpu(i) {
8151 struct softnet_data *sd = &per_cpu(softnet_data, i);
8153 skb_queue_head_init(&sd->input_pkt_queue);
8154 skb_queue_head_init(&sd->process_queue);
8155 INIT_LIST_HEAD(&sd->poll_list);
8156 sd->output_queue_tailp = &sd->output_queue;
8158 sd->csd.func = rps_trigger_softirq;
8163 sd->backlog.poll = process_backlog;
8164 sd->backlog.weight = weight_p;
8169 /* The loopback device is special if any other network devices
8170 * is present in a network namespace the loopback device must
8171 * be present. Since we now dynamically allocate and free the
8172 * loopback device ensure this invariant is maintained by
8173 * keeping the loopback device as the first device on the
8174 * list of network devices. Ensuring the loopback devices
8175 * is the first device that appears and the last network device
8178 if (register_pernet_device(&loopback_net_ops))
8181 if (register_pernet_device(&default_device_ops))
8184 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
8185 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
8187 hotcpu_notifier(dev_cpu_callback, 0);
8194 subsys_initcall(net_dev_init);