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(struct net_device *dev, 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 static 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);
1912 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1913 * @dev: Network device
1914 * @txq: number of queues available
1916 * If real_num_tx_queues is changed the tc mappings may no longer be
1917 * valid. To resolve this verify the tc mapping remains valid and if
1918 * not NULL the mapping. With no priorities mapping to this
1919 * offset/count pair it will no longer be used. In the worst case TC0
1920 * is invalid nothing can be done so disable priority mappings. If is
1921 * expected that drivers will fix this mapping if they can before
1922 * calling netif_set_real_num_tx_queues.
1924 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1927 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1929 /* If TC0 is invalidated disable TC mapping */
1930 if (tc->offset + tc->count > txq) {
1931 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1936 /* Invalidated prio to tc mappings set to TC0 */
1937 for (i = 1; i < TC_BITMASK + 1; i++) {
1938 int q = netdev_get_prio_tc_map(dev, i);
1940 tc = &dev->tc_to_txq[q];
1941 if (tc->offset + tc->count > txq) {
1942 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1944 netdev_set_prio_tc_map(dev, i, 0);
1950 static DEFINE_MUTEX(xps_map_mutex);
1951 #define xmap_dereference(P) \
1952 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1954 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1957 struct xps_map *map = NULL;
1961 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1963 for (pos = 0; map && pos < map->len; pos++) {
1964 if (map->queues[pos] == index) {
1966 map->queues[pos] = map->queues[--map->len];
1968 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1969 kfree_rcu(map, rcu);
1979 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1981 struct xps_dev_maps *dev_maps;
1983 bool active = false;
1985 mutex_lock(&xps_map_mutex);
1986 dev_maps = xmap_dereference(dev->xps_maps);
1991 for_each_possible_cpu(cpu) {
1992 for (i = index; i < dev->num_tx_queues; i++) {
1993 if (!remove_xps_queue(dev_maps, cpu, i))
1996 if (i == dev->num_tx_queues)
2001 RCU_INIT_POINTER(dev->xps_maps, NULL);
2002 kfree_rcu(dev_maps, rcu);
2005 for (i = index; i < dev->num_tx_queues; i++)
2006 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2010 mutex_unlock(&xps_map_mutex);
2013 static struct xps_map *expand_xps_map(struct xps_map *map,
2016 struct xps_map *new_map;
2017 int alloc_len = XPS_MIN_MAP_ALLOC;
2020 for (pos = 0; map && pos < map->len; pos++) {
2021 if (map->queues[pos] != index)
2026 /* Need to add queue to this CPU's existing map */
2028 if (pos < map->alloc_len)
2031 alloc_len = map->alloc_len * 2;
2034 /* Need to allocate new map to store queue on this CPU's map */
2035 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2040 for (i = 0; i < pos; i++)
2041 new_map->queues[i] = map->queues[i];
2042 new_map->alloc_len = alloc_len;
2048 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2051 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2052 struct xps_map *map, *new_map;
2053 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2054 int cpu, numa_node_id = -2;
2055 bool active = false;
2057 mutex_lock(&xps_map_mutex);
2059 dev_maps = xmap_dereference(dev->xps_maps);
2061 /* allocate memory for queue storage */
2062 for_each_online_cpu(cpu) {
2063 if (!cpumask_test_cpu(cpu, mask))
2067 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2068 if (!new_dev_maps) {
2069 mutex_unlock(&xps_map_mutex);
2073 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2076 map = expand_xps_map(map, cpu, index);
2080 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2084 goto out_no_new_maps;
2086 for_each_possible_cpu(cpu) {
2087 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2088 /* add queue to CPU maps */
2091 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2092 while ((pos < map->len) && (map->queues[pos] != index))
2095 if (pos == map->len)
2096 map->queues[map->len++] = index;
2098 if (numa_node_id == -2)
2099 numa_node_id = cpu_to_node(cpu);
2100 else if (numa_node_id != cpu_to_node(cpu))
2103 } else if (dev_maps) {
2104 /* fill in the new device map from the old device map */
2105 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2106 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2111 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2113 /* Cleanup old maps */
2115 for_each_possible_cpu(cpu) {
2116 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2117 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2118 if (map && map != new_map)
2119 kfree_rcu(map, rcu);
2122 kfree_rcu(dev_maps, rcu);
2125 dev_maps = new_dev_maps;
2129 /* update Tx queue numa node */
2130 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2131 (numa_node_id >= 0) ? numa_node_id :
2137 /* removes queue from unused CPUs */
2138 for_each_possible_cpu(cpu) {
2139 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2142 if (remove_xps_queue(dev_maps, cpu, index))
2146 /* free map if not active */
2148 RCU_INIT_POINTER(dev->xps_maps, NULL);
2149 kfree_rcu(dev_maps, rcu);
2153 mutex_unlock(&xps_map_mutex);
2157 /* remove any maps that we added */
2158 for_each_possible_cpu(cpu) {
2159 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2160 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2162 if (new_map && new_map != map)
2166 mutex_unlock(&xps_map_mutex);
2168 kfree(new_dev_maps);
2171 EXPORT_SYMBOL(netif_set_xps_queue);
2175 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2176 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2178 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2182 if (txq < 1 || txq > dev->num_tx_queues)
2185 if (dev->reg_state == NETREG_REGISTERED ||
2186 dev->reg_state == NETREG_UNREGISTERING) {
2189 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2195 netif_setup_tc(dev, txq);
2197 if (txq < dev->real_num_tx_queues) {
2198 qdisc_reset_all_tx_gt(dev, txq);
2200 netif_reset_xps_queues_gt(dev, txq);
2205 dev->real_num_tx_queues = txq;
2208 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2212 * netif_set_real_num_rx_queues - set actual number of RX queues used
2213 * @dev: Network device
2214 * @rxq: Actual number of RX queues
2216 * This must be called either with the rtnl_lock held or before
2217 * registration of the net device. Returns 0 on success, or a
2218 * negative error code. If called before registration, it always
2221 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2225 if (rxq < 1 || rxq > dev->num_rx_queues)
2228 if (dev->reg_state == NETREG_REGISTERED) {
2231 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2237 dev->real_num_rx_queues = rxq;
2240 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2244 * netif_get_num_default_rss_queues - default number of RSS queues
2246 * This routine should set an upper limit on the number of RSS queues
2247 * used by default by multiqueue devices.
2249 int netif_get_num_default_rss_queues(void)
2251 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2253 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2255 static inline void __netif_reschedule(struct Qdisc *q)
2257 struct softnet_data *sd;
2258 unsigned long flags;
2260 local_irq_save(flags);
2261 sd = this_cpu_ptr(&softnet_data);
2262 q->next_sched = NULL;
2263 *sd->output_queue_tailp = q;
2264 sd->output_queue_tailp = &q->next_sched;
2265 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2266 local_irq_restore(flags);
2269 void __netif_schedule(struct Qdisc *q)
2271 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2272 __netif_reschedule(q);
2274 EXPORT_SYMBOL(__netif_schedule);
2276 struct dev_kfree_skb_cb {
2277 enum skb_free_reason reason;
2280 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2282 return (struct dev_kfree_skb_cb *)skb->cb;
2285 void netif_schedule_queue(struct netdev_queue *txq)
2288 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2289 struct Qdisc *q = rcu_dereference(txq->qdisc);
2291 __netif_schedule(q);
2295 EXPORT_SYMBOL(netif_schedule_queue);
2298 * netif_wake_subqueue - allow sending packets on subqueue
2299 * @dev: network device
2300 * @queue_index: sub queue index
2302 * Resume individual transmit queue of a device with multiple transmit queues.
2304 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2306 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2308 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2312 q = rcu_dereference(txq->qdisc);
2313 __netif_schedule(q);
2317 EXPORT_SYMBOL(netif_wake_subqueue);
2319 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2321 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2325 q = rcu_dereference(dev_queue->qdisc);
2326 __netif_schedule(q);
2330 EXPORT_SYMBOL(netif_tx_wake_queue);
2332 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2334 unsigned long flags;
2336 if (likely(atomic_read(&skb->users) == 1)) {
2338 atomic_set(&skb->users, 0);
2339 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2342 get_kfree_skb_cb(skb)->reason = reason;
2343 local_irq_save(flags);
2344 skb->next = __this_cpu_read(softnet_data.completion_queue);
2345 __this_cpu_write(softnet_data.completion_queue, skb);
2346 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2347 local_irq_restore(flags);
2349 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2351 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2353 if (in_irq() || irqs_disabled())
2354 __dev_kfree_skb_irq(skb, reason);
2358 EXPORT_SYMBOL(__dev_kfree_skb_any);
2362 * netif_device_detach - mark device as removed
2363 * @dev: network device
2365 * Mark device as removed from system and therefore no longer available.
2367 void netif_device_detach(struct net_device *dev)
2369 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2370 netif_running(dev)) {
2371 netif_tx_stop_all_queues(dev);
2374 EXPORT_SYMBOL(netif_device_detach);
2377 * netif_device_attach - mark device as attached
2378 * @dev: network device
2380 * Mark device as attached from system and restart if needed.
2382 void netif_device_attach(struct net_device *dev)
2384 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2385 netif_running(dev)) {
2386 netif_tx_wake_all_queues(dev);
2387 __netdev_watchdog_up(dev);
2390 EXPORT_SYMBOL(netif_device_attach);
2393 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2394 * to be used as a distribution range.
2396 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2397 unsigned int num_tx_queues)
2401 u16 qcount = num_tx_queues;
2403 if (skb_rx_queue_recorded(skb)) {
2404 hash = skb_get_rx_queue(skb);
2405 while (unlikely(hash >= num_tx_queues))
2406 hash -= num_tx_queues;
2411 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2412 qoffset = dev->tc_to_txq[tc].offset;
2413 qcount = dev->tc_to_txq[tc].count;
2416 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2418 EXPORT_SYMBOL(__skb_tx_hash);
2420 static void skb_warn_bad_offload(const struct sk_buff *skb)
2422 static const netdev_features_t null_features = 0;
2423 struct net_device *dev = skb->dev;
2424 const char *name = "";
2426 if (!net_ratelimit())
2430 if (dev->dev.parent)
2431 name = dev_driver_string(dev->dev.parent);
2433 name = netdev_name(dev);
2435 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2436 "gso_type=%d ip_summed=%d\n",
2437 name, dev ? &dev->features : &null_features,
2438 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2439 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2440 skb_shinfo(skb)->gso_type, skb->ip_summed);
2444 * Invalidate hardware checksum when packet is to be mangled, and
2445 * complete checksum manually on outgoing path.
2447 int skb_checksum_help(struct sk_buff *skb)
2450 int ret = 0, offset;
2452 if (skb->ip_summed == CHECKSUM_COMPLETE)
2453 goto out_set_summed;
2455 if (unlikely(skb_shinfo(skb)->gso_size)) {
2456 skb_warn_bad_offload(skb);
2460 /* Before computing a checksum, we should make sure no frag could
2461 * be modified by an external entity : checksum could be wrong.
2463 if (skb_has_shared_frag(skb)) {
2464 ret = __skb_linearize(skb);
2469 offset = skb_checksum_start_offset(skb);
2470 BUG_ON(offset >= skb_headlen(skb));
2471 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2473 offset += skb->csum_offset;
2474 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2476 if (skb_cloned(skb) &&
2477 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2478 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2483 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2485 skb->ip_summed = CHECKSUM_NONE;
2489 EXPORT_SYMBOL(skb_checksum_help);
2491 /* skb_csum_offload_check - Driver helper function to determine if a device
2492 * with limited checksum offload capabilities is able to offload the checksum
2493 * for a given packet.
2496 * skb - sk_buff for the packet in question
2497 * spec - contains the description of what device can offload
2498 * csum_encapped - returns true if the checksum being offloaded is
2499 * encpasulated. That is it is checksum for the transport header
2500 * in the inner headers.
2501 * checksum_help - when set indicates that helper function should
2502 * call skb_checksum_help if offload checks fail
2505 * true: Packet has passed the checksum checks and should be offloadable to
2506 * the device (a driver may still need to check for additional
2507 * restrictions of its device)
2508 * false: Checksum is not offloadable. If checksum_help was set then
2509 * skb_checksum_help was called to resolve checksum for non-GSO
2510 * packets and when IP protocol is not SCTP
2512 bool __skb_csum_offload_chk(struct sk_buff *skb,
2513 const struct skb_csum_offl_spec *spec,
2514 bool *csum_encapped,
2518 struct ipv6hdr *ipv6;
2523 if (skb->protocol == htons(ETH_P_8021Q) ||
2524 skb->protocol == htons(ETH_P_8021AD)) {
2525 if (!spec->vlan_okay)
2529 /* We check whether the checksum refers to a transport layer checksum in
2530 * the outermost header or an encapsulated transport layer checksum that
2531 * corresponds to the inner headers of the skb. If the checksum is for
2532 * something else in the packet we need help.
2534 if (skb_checksum_start_offset(skb) == skb_transport_offset(skb)) {
2535 /* Non-encapsulated checksum */
2536 protocol = eproto_to_ipproto(vlan_get_protocol(skb));
2537 nhdr = skb_network_header(skb);
2538 *csum_encapped = false;
2539 if (spec->no_not_encapped)
2541 } else if (skb->encapsulation && spec->encap_okay &&
2542 skb_checksum_start_offset(skb) ==
2543 skb_inner_transport_offset(skb)) {
2544 /* Encapsulated checksum */
2545 *csum_encapped = true;
2546 switch (skb->inner_protocol_type) {
2547 case ENCAP_TYPE_ETHER:
2548 protocol = eproto_to_ipproto(skb->inner_protocol);
2550 case ENCAP_TYPE_IPPROTO:
2551 protocol = skb->inner_protocol;
2554 nhdr = skb_inner_network_header(skb);
2561 if (!spec->ipv4_okay)
2564 ip_proto = iph->protocol;
2565 if (iph->ihl != 5 && !spec->ip_options_okay)
2569 if (!spec->ipv6_okay)
2571 if (spec->no_encapped_ipv6 && *csum_encapped)
2574 nhdr += sizeof(*ipv6);
2575 ip_proto = ipv6->nexthdr;
2584 if (!spec->tcp_okay ||
2585 skb->csum_offset != offsetof(struct tcphdr, check))
2589 if (!spec->udp_okay ||
2590 skb->csum_offset != offsetof(struct udphdr, check))
2594 if (!spec->sctp_okay ||
2595 skb->csum_offset != offsetof(struct sctphdr, checksum))
2599 case NEXTHDR_ROUTING:
2600 case NEXTHDR_DEST: {
2603 if (protocol != IPPROTO_IPV6 || !spec->ext_hdrs_okay)
2606 ip_proto = opthdr[0];
2607 nhdr += (opthdr[1] + 1) << 3;
2609 goto ip_proto_again;
2615 /* Passed the tests for offloading checksum */
2619 if (csum_help && !skb_shinfo(skb)->gso_size)
2620 skb_checksum_help(skb);
2624 EXPORT_SYMBOL(__skb_csum_offload_chk);
2626 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2628 __be16 type = skb->protocol;
2630 /* Tunnel gso handlers can set protocol to ethernet. */
2631 if (type == htons(ETH_P_TEB)) {
2634 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2637 eth = (struct ethhdr *)skb_mac_header(skb);
2638 type = eth->h_proto;
2641 return __vlan_get_protocol(skb, type, depth);
2645 * skb_mac_gso_segment - mac layer segmentation handler.
2646 * @skb: buffer to segment
2647 * @features: features for the output path (see dev->features)
2649 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2650 netdev_features_t features)
2652 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2653 struct packet_offload *ptype;
2654 int vlan_depth = skb->mac_len;
2655 __be16 type = skb_network_protocol(skb, &vlan_depth);
2657 if (unlikely(!type))
2658 return ERR_PTR(-EINVAL);
2660 __skb_pull(skb, vlan_depth);
2663 list_for_each_entry_rcu(ptype, &offload_base, list) {
2664 if (ptype->type == type && ptype->callbacks.gso_segment) {
2665 segs = ptype->callbacks.gso_segment(skb, features);
2671 __skb_push(skb, skb->data - skb_mac_header(skb));
2675 EXPORT_SYMBOL(skb_mac_gso_segment);
2678 /* openvswitch calls this on rx path, so we need a different check.
2680 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2683 return skb->ip_summed != CHECKSUM_PARTIAL;
2685 return skb->ip_summed == CHECKSUM_NONE;
2689 * __skb_gso_segment - Perform segmentation on skb.
2690 * @skb: buffer to segment
2691 * @features: features for the output path (see dev->features)
2692 * @tx_path: whether it is called in TX path
2694 * This function segments the given skb and returns a list of segments.
2696 * It may return NULL if the skb requires no segmentation. This is
2697 * only possible when GSO is used for verifying header integrity.
2699 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2701 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2702 netdev_features_t features, bool tx_path)
2704 if (unlikely(skb_needs_check(skb, tx_path))) {
2707 skb_warn_bad_offload(skb);
2709 err = skb_cow_head(skb, 0);
2711 return ERR_PTR(err);
2714 /* Only report GSO partial support if it will enable us to
2715 * support segmentation on this frame without needing additional
2718 if (features & NETIF_F_GSO_PARTIAL) {
2719 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
2720 struct net_device *dev = skb->dev;
2722 partial_features |= dev->features & dev->gso_partial_features;
2723 if (!skb_gso_ok(skb, features | partial_features))
2724 features &= ~NETIF_F_GSO_PARTIAL;
2727 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2728 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2730 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2731 SKB_GSO_CB(skb)->encap_level = 0;
2733 skb_reset_mac_header(skb);
2734 skb_reset_mac_len(skb);
2736 return skb_mac_gso_segment(skb, features);
2738 EXPORT_SYMBOL(__skb_gso_segment);
2740 /* Take action when hardware reception checksum errors are detected. */
2742 void netdev_rx_csum_fault(struct net_device *dev)
2744 if (net_ratelimit()) {
2745 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2749 EXPORT_SYMBOL(netdev_rx_csum_fault);
2752 /* Actually, we should eliminate this check as soon as we know, that:
2753 * 1. IOMMU is present and allows to map all the memory.
2754 * 2. No high memory really exists on this machine.
2757 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2759 #ifdef CONFIG_HIGHMEM
2761 if (!(dev->features & NETIF_F_HIGHDMA)) {
2762 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2763 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2764 if (PageHighMem(skb_frag_page(frag)))
2769 if (PCI_DMA_BUS_IS_PHYS) {
2770 struct device *pdev = dev->dev.parent;
2774 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2775 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2776 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2777 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2785 /* If MPLS offload request, verify we are testing hardware MPLS features
2786 * instead of standard features for the netdev.
2788 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2789 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2790 netdev_features_t features,
2793 if (eth_p_mpls(type))
2794 features &= skb->dev->mpls_features;
2799 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2800 netdev_features_t features,
2807 static netdev_features_t harmonize_features(struct sk_buff *skb,
2808 netdev_features_t features)
2813 type = skb_network_protocol(skb, &tmp);
2814 features = net_mpls_features(skb, features, type);
2816 if (skb->ip_summed != CHECKSUM_NONE &&
2817 !can_checksum_protocol(features, type)) {
2818 features &= ~NETIF_F_CSUM_MASK;
2819 } else if (illegal_highdma(skb->dev, skb)) {
2820 features &= ~NETIF_F_SG;
2826 netdev_features_t passthru_features_check(struct sk_buff *skb,
2827 struct net_device *dev,
2828 netdev_features_t features)
2832 EXPORT_SYMBOL(passthru_features_check);
2834 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2835 struct net_device *dev,
2836 netdev_features_t features)
2838 return vlan_features_check(skb, features);
2841 static netdev_features_t gso_features_check(const struct sk_buff *skb,
2842 struct net_device *dev,
2843 netdev_features_t features)
2845 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2847 if (gso_segs > dev->gso_max_segs)
2848 return features & ~NETIF_F_GSO_MASK;
2850 /* Support for GSO partial features requires software
2851 * intervention before we can actually process the packets
2852 * so we need to strip support for any partial features now
2853 * and we can pull them back in after we have partially
2854 * segmented the frame.
2856 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
2857 features &= ~dev->gso_partial_features;
2859 /* Make sure to clear the IPv4 ID mangling feature if the
2860 * IPv4 header has the potential to be fragmented.
2862 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2863 struct iphdr *iph = skb->encapsulation ?
2864 inner_ip_hdr(skb) : ip_hdr(skb);
2866 if (!(iph->frag_off & htons(IP_DF)))
2867 features &= ~NETIF_F_TSO_MANGLEID;
2873 netdev_features_t netif_skb_features(struct sk_buff *skb)
2875 struct net_device *dev = skb->dev;
2876 netdev_features_t features = dev->features;
2878 if (skb_is_gso(skb))
2879 features = gso_features_check(skb, dev, features);
2881 /* If encapsulation offload request, verify we are testing
2882 * hardware encapsulation features instead of standard
2883 * features for the netdev
2885 if (skb->encapsulation)
2886 features &= dev->hw_enc_features;
2888 if (skb_vlan_tagged(skb))
2889 features = netdev_intersect_features(features,
2890 dev->vlan_features |
2891 NETIF_F_HW_VLAN_CTAG_TX |
2892 NETIF_F_HW_VLAN_STAG_TX);
2894 if (dev->netdev_ops->ndo_features_check)
2895 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2898 features &= dflt_features_check(skb, dev, features);
2900 return harmonize_features(skb, features);
2902 EXPORT_SYMBOL(netif_skb_features);
2904 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2905 struct netdev_queue *txq, bool more)
2910 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2911 dev_queue_xmit_nit(skb, dev);
2914 trace_net_dev_start_xmit(skb, dev);
2915 rc = netdev_start_xmit(skb, dev, txq, more);
2916 trace_net_dev_xmit(skb, rc, dev, len);
2921 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2922 struct netdev_queue *txq, int *ret)
2924 struct sk_buff *skb = first;
2925 int rc = NETDEV_TX_OK;
2928 struct sk_buff *next = skb->next;
2931 rc = xmit_one(skb, dev, txq, next != NULL);
2932 if (unlikely(!dev_xmit_complete(rc))) {
2938 if (netif_xmit_stopped(txq) && skb) {
2939 rc = NETDEV_TX_BUSY;
2949 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2950 netdev_features_t features)
2952 if (skb_vlan_tag_present(skb) &&
2953 !vlan_hw_offload_capable(features, skb->vlan_proto))
2954 skb = __vlan_hwaccel_push_inside(skb);
2958 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2960 netdev_features_t features;
2962 features = netif_skb_features(skb);
2963 skb = validate_xmit_vlan(skb, features);
2967 if (netif_needs_gso(skb, features)) {
2968 struct sk_buff *segs;
2970 segs = skb_gso_segment(skb, features);
2978 if (skb_needs_linearize(skb, features) &&
2979 __skb_linearize(skb))
2982 /* If packet is not checksummed and device does not
2983 * support checksumming for this protocol, complete
2984 * checksumming here.
2986 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2987 if (skb->encapsulation)
2988 skb_set_inner_transport_header(skb,
2989 skb_checksum_start_offset(skb));
2991 skb_set_transport_header(skb,
2992 skb_checksum_start_offset(skb));
2993 if (!(features & NETIF_F_CSUM_MASK) &&
2994 skb_checksum_help(skb))
3004 atomic_long_inc(&dev->tx_dropped);
3008 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
3010 struct sk_buff *next, *head = NULL, *tail;
3012 for (; skb != NULL; skb = next) {
3016 /* in case skb wont be segmented, point to itself */
3019 skb = validate_xmit_skb(skb, dev);
3027 /* If skb was segmented, skb->prev points to
3028 * the last segment. If not, it still contains skb.
3035 static void qdisc_pkt_len_init(struct sk_buff *skb)
3037 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3039 qdisc_skb_cb(skb)->pkt_len = skb->len;
3041 /* To get more precise estimation of bytes sent on wire,
3042 * we add to pkt_len the headers size of all segments
3044 if (shinfo->gso_size) {
3045 unsigned int hdr_len;
3046 u16 gso_segs = shinfo->gso_segs;
3048 /* mac layer + network layer */
3049 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3051 /* + transport layer */
3052 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
3053 hdr_len += tcp_hdrlen(skb);
3055 hdr_len += sizeof(struct udphdr);
3057 if (shinfo->gso_type & SKB_GSO_DODGY)
3058 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3061 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3065 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3066 struct net_device *dev,
3067 struct netdev_queue *txq)
3069 spinlock_t *root_lock = qdisc_lock(q);
3073 qdisc_calculate_pkt_len(skb, q);
3075 * Heuristic to force contended enqueues to serialize on a
3076 * separate lock before trying to get qdisc main lock.
3077 * This permits __QDISC___STATE_RUNNING owner to get the lock more
3078 * often and dequeue packets faster.
3080 contended = qdisc_is_running(q);
3081 if (unlikely(contended))
3082 spin_lock(&q->busylock);
3084 spin_lock(root_lock);
3085 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3088 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3089 qdisc_run_begin(q)) {
3091 * This is a work-conserving queue; there are no old skbs
3092 * waiting to be sent out; and the qdisc is not running -
3093 * xmit the skb directly.
3096 qdisc_bstats_update(q, skb);
3098 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3099 if (unlikely(contended)) {
3100 spin_unlock(&q->busylock);
3107 rc = NET_XMIT_SUCCESS;
3109 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
3110 if (qdisc_run_begin(q)) {
3111 if (unlikely(contended)) {
3112 spin_unlock(&q->busylock);
3118 spin_unlock(root_lock);
3119 if (unlikely(contended))
3120 spin_unlock(&q->busylock);
3124 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3125 static void skb_update_prio(struct sk_buff *skb)
3127 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
3129 if (!skb->priority && skb->sk && map) {
3130 unsigned int prioidx =
3131 sock_cgroup_prioidx(&skb->sk->sk_cgrp_data);
3133 if (prioidx < map->priomap_len)
3134 skb->priority = map->priomap[prioidx];
3138 #define skb_update_prio(skb)
3141 DEFINE_PER_CPU(int, xmit_recursion);
3142 EXPORT_SYMBOL(xmit_recursion);
3144 #define RECURSION_LIMIT 10
3147 * dev_loopback_xmit - loop back @skb
3148 * @net: network namespace this loopback is happening in
3149 * @sk: sk needed to be a netfilter okfn
3150 * @skb: buffer to transmit
3152 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3154 skb_reset_mac_header(skb);
3155 __skb_pull(skb, skb_network_offset(skb));
3156 skb->pkt_type = PACKET_LOOPBACK;
3157 skb->ip_summed = CHECKSUM_UNNECESSARY;
3158 WARN_ON(!skb_dst(skb));
3163 EXPORT_SYMBOL(dev_loopback_xmit);
3165 #ifdef CONFIG_NET_EGRESS
3166 static struct sk_buff *
3167 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3169 struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list);
3170 struct tcf_result cl_res;
3175 /* skb->tc_verd and qdisc_skb_cb(skb)->pkt_len were already set
3176 * earlier by the caller.
3178 qdisc_bstats_cpu_update(cl->q, skb);
3180 switch (tc_classify(skb, cl, &cl_res, false)) {
3182 case TC_ACT_RECLASSIFY:
3183 skb->tc_index = TC_H_MIN(cl_res.classid);
3186 qdisc_qstats_cpu_drop(cl->q);
3187 *ret = NET_XMIT_DROP;
3191 *ret = NET_XMIT_SUCCESS;
3195 case TC_ACT_REDIRECT:
3196 /* No need to push/pop skb's mac_header here on egress! */
3197 skb_do_redirect(skb);
3198 *ret = NET_XMIT_SUCCESS;
3206 #endif /* CONFIG_NET_EGRESS */
3208 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3211 struct xps_dev_maps *dev_maps;
3212 struct xps_map *map;
3213 int queue_index = -1;
3216 dev_maps = rcu_dereference(dev->xps_maps);
3218 map = rcu_dereference(
3219 dev_maps->cpu_map[skb->sender_cpu - 1]);
3222 queue_index = map->queues[0];
3224 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3226 if (unlikely(queue_index >= dev->real_num_tx_queues))
3238 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3240 struct sock *sk = skb->sk;
3241 int queue_index = sk_tx_queue_get(sk);
3243 if (queue_index < 0 || skb->ooo_okay ||
3244 queue_index >= dev->real_num_tx_queues) {
3245 int new_index = get_xps_queue(dev, skb);
3247 new_index = skb_tx_hash(dev, skb);
3249 if (queue_index != new_index && sk &&
3251 rcu_access_pointer(sk->sk_dst_cache))
3252 sk_tx_queue_set(sk, new_index);
3254 queue_index = new_index;
3260 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3261 struct sk_buff *skb,
3264 int queue_index = 0;
3267 u32 sender_cpu = skb->sender_cpu - 1;
3269 if (sender_cpu >= (u32)NR_CPUS)
3270 skb->sender_cpu = raw_smp_processor_id() + 1;
3273 if (dev->real_num_tx_queues != 1) {
3274 const struct net_device_ops *ops = dev->netdev_ops;
3275 if (ops->ndo_select_queue)
3276 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3279 queue_index = __netdev_pick_tx(dev, skb);
3282 queue_index = netdev_cap_txqueue(dev, queue_index);
3285 skb_set_queue_mapping(skb, queue_index);
3286 return netdev_get_tx_queue(dev, queue_index);
3290 * __dev_queue_xmit - transmit a buffer
3291 * @skb: buffer to transmit
3292 * @accel_priv: private data used for L2 forwarding offload
3294 * Queue a buffer for transmission to a network device. The caller must
3295 * have set the device and priority and built the buffer before calling
3296 * this function. The function can be called from an interrupt.
3298 * A negative errno code is returned on a failure. A success does not
3299 * guarantee the frame will be transmitted as it may be dropped due
3300 * to congestion or traffic shaping.
3302 * -----------------------------------------------------------------------------------
3303 * I notice this method can also return errors from the queue disciplines,
3304 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3307 * Regardless of the return value, the skb is consumed, so it is currently
3308 * difficult to retry a send to this method. (You can bump the ref count
3309 * before sending to hold a reference for retry if you are careful.)
3311 * When calling this method, interrupts MUST be enabled. This is because
3312 * the BH enable code must have IRQs enabled so that it will not deadlock.
3315 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3317 struct net_device *dev = skb->dev;
3318 struct netdev_queue *txq;
3322 skb_reset_mac_header(skb);
3324 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3325 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3327 /* Disable soft irqs for various locks below. Also
3328 * stops preemption for RCU.
3332 skb_update_prio(skb);
3334 qdisc_pkt_len_init(skb);
3335 #ifdef CONFIG_NET_CLS_ACT
3336 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3337 # ifdef CONFIG_NET_EGRESS
3338 if (static_key_false(&egress_needed)) {
3339 skb = sch_handle_egress(skb, &rc, dev);
3345 /* If device/qdisc don't need skb->dst, release it right now while
3346 * its hot in this cpu cache.
3348 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3353 #ifdef CONFIG_NET_SWITCHDEV
3354 /* Don't forward if offload device already forwarded */
3355 if (skb->offload_fwd_mark &&
3356 skb->offload_fwd_mark == dev->offload_fwd_mark) {
3358 rc = NET_XMIT_SUCCESS;
3363 txq = netdev_pick_tx(dev, skb, accel_priv);
3364 q = rcu_dereference_bh(txq->qdisc);
3366 trace_net_dev_queue(skb);
3368 rc = __dev_xmit_skb(skb, q, dev, txq);
3372 /* The device has no queue. Common case for software devices:
3373 loopback, all the sorts of tunnels...
3375 Really, it is unlikely that netif_tx_lock protection is necessary
3376 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3378 However, it is possible, that they rely on protection
3381 Check this and shot the lock. It is not prone from deadlocks.
3382 Either shot noqueue qdisc, it is even simpler 8)
3384 if (dev->flags & IFF_UP) {
3385 int cpu = smp_processor_id(); /* ok because BHs are off */
3387 if (txq->xmit_lock_owner != cpu) {
3389 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
3390 goto recursion_alert;
3392 skb = validate_xmit_skb(skb, dev);
3396 HARD_TX_LOCK(dev, txq, cpu);
3398 if (!netif_xmit_stopped(txq)) {
3399 __this_cpu_inc(xmit_recursion);
3400 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3401 __this_cpu_dec(xmit_recursion);
3402 if (dev_xmit_complete(rc)) {
3403 HARD_TX_UNLOCK(dev, txq);
3407 HARD_TX_UNLOCK(dev, txq);
3408 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3411 /* Recursion is detected! It is possible,
3415 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3421 rcu_read_unlock_bh();
3423 atomic_long_inc(&dev->tx_dropped);
3424 kfree_skb_list(skb);
3427 rcu_read_unlock_bh();
3431 int dev_queue_xmit(struct sk_buff *skb)
3433 return __dev_queue_xmit(skb, NULL);
3435 EXPORT_SYMBOL(dev_queue_xmit);
3437 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3439 return __dev_queue_xmit(skb, accel_priv);
3441 EXPORT_SYMBOL(dev_queue_xmit_accel);
3444 /*=======================================================================
3446 =======================================================================*/
3448 int netdev_max_backlog __read_mostly = 1000;
3449 EXPORT_SYMBOL(netdev_max_backlog);
3451 int netdev_tstamp_prequeue __read_mostly = 1;
3452 int netdev_budget __read_mostly = 300;
3453 int weight_p __read_mostly = 64; /* old backlog weight */
3455 /* Called with irq disabled */
3456 static inline void ____napi_schedule(struct softnet_data *sd,
3457 struct napi_struct *napi)
3459 list_add_tail(&napi->poll_list, &sd->poll_list);
3460 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3465 /* One global table that all flow-based protocols share. */
3466 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3467 EXPORT_SYMBOL(rps_sock_flow_table);
3468 u32 rps_cpu_mask __read_mostly;
3469 EXPORT_SYMBOL(rps_cpu_mask);
3471 struct static_key rps_needed __read_mostly;
3473 static struct rps_dev_flow *
3474 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3475 struct rps_dev_flow *rflow, u16 next_cpu)
3477 if (next_cpu < nr_cpu_ids) {
3478 #ifdef CONFIG_RFS_ACCEL
3479 struct netdev_rx_queue *rxqueue;
3480 struct rps_dev_flow_table *flow_table;
3481 struct rps_dev_flow *old_rflow;
3486 /* Should we steer this flow to a different hardware queue? */
3487 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3488 !(dev->features & NETIF_F_NTUPLE))
3490 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3491 if (rxq_index == skb_get_rx_queue(skb))
3494 rxqueue = dev->_rx + rxq_index;
3495 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3498 flow_id = skb_get_hash(skb) & flow_table->mask;
3499 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3500 rxq_index, flow_id);
3504 rflow = &flow_table->flows[flow_id];
3506 if (old_rflow->filter == rflow->filter)
3507 old_rflow->filter = RPS_NO_FILTER;
3511 per_cpu(softnet_data, next_cpu).input_queue_head;
3514 rflow->cpu = next_cpu;
3519 * get_rps_cpu is called from netif_receive_skb and returns the target
3520 * CPU from the RPS map of the receiving queue for a given skb.
3521 * rcu_read_lock must be held on entry.
3523 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3524 struct rps_dev_flow **rflowp)
3526 const struct rps_sock_flow_table *sock_flow_table;
3527 struct netdev_rx_queue *rxqueue = dev->_rx;
3528 struct rps_dev_flow_table *flow_table;
3529 struct rps_map *map;
3534 if (skb_rx_queue_recorded(skb)) {
3535 u16 index = skb_get_rx_queue(skb);
3537 if (unlikely(index >= dev->real_num_rx_queues)) {
3538 WARN_ONCE(dev->real_num_rx_queues > 1,
3539 "%s received packet on queue %u, but number "
3540 "of RX queues is %u\n",
3541 dev->name, index, dev->real_num_rx_queues);
3547 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3549 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3550 map = rcu_dereference(rxqueue->rps_map);
3551 if (!flow_table && !map)
3554 skb_reset_network_header(skb);
3555 hash = skb_get_hash(skb);
3559 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3560 if (flow_table && sock_flow_table) {
3561 struct rps_dev_flow *rflow;
3565 /* First check into global flow table if there is a match */
3566 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3567 if ((ident ^ hash) & ~rps_cpu_mask)
3570 next_cpu = ident & rps_cpu_mask;
3572 /* OK, now we know there is a match,
3573 * we can look at the local (per receive queue) flow table
3575 rflow = &flow_table->flows[hash & flow_table->mask];
3579 * If the desired CPU (where last recvmsg was done) is
3580 * different from current CPU (one in the rx-queue flow
3581 * table entry), switch if one of the following holds:
3582 * - Current CPU is unset (>= nr_cpu_ids).
3583 * - Current CPU is offline.
3584 * - The current CPU's queue tail has advanced beyond the
3585 * last packet that was enqueued using this table entry.
3586 * This guarantees that all previous packets for the flow
3587 * have been dequeued, thus preserving in order delivery.
3589 if (unlikely(tcpu != next_cpu) &&
3590 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3591 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3592 rflow->last_qtail)) >= 0)) {
3594 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3597 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3607 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3608 if (cpu_online(tcpu)) {
3618 #ifdef CONFIG_RFS_ACCEL
3621 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3622 * @dev: Device on which the filter was set
3623 * @rxq_index: RX queue index
3624 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3625 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3627 * Drivers that implement ndo_rx_flow_steer() should periodically call
3628 * this function for each installed filter and remove the filters for
3629 * which it returns %true.
3631 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3632 u32 flow_id, u16 filter_id)
3634 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3635 struct rps_dev_flow_table *flow_table;
3636 struct rps_dev_flow *rflow;
3641 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3642 if (flow_table && flow_id <= flow_table->mask) {
3643 rflow = &flow_table->flows[flow_id];
3644 cpu = ACCESS_ONCE(rflow->cpu);
3645 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3646 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3647 rflow->last_qtail) <
3648 (int)(10 * flow_table->mask)))
3654 EXPORT_SYMBOL(rps_may_expire_flow);
3656 #endif /* CONFIG_RFS_ACCEL */
3658 /* Called from hardirq (IPI) context */
3659 static void rps_trigger_softirq(void *data)
3661 struct softnet_data *sd = data;
3663 ____napi_schedule(sd, &sd->backlog);
3667 #endif /* CONFIG_RPS */
3670 * Check if this softnet_data structure is another cpu one
3671 * If yes, queue it to our IPI list and return 1
3674 static int rps_ipi_queued(struct softnet_data *sd)
3677 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3680 sd->rps_ipi_next = mysd->rps_ipi_list;
3681 mysd->rps_ipi_list = sd;
3683 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3686 #endif /* CONFIG_RPS */
3690 #ifdef CONFIG_NET_FLOW_LIMIT
3691 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3694 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3696 #ifdef CONFIG_NET_FLOW_LIMIT
3697 struct sd_flow_limit *fl;
3698 struct softnet_data *sd;
3699 unsigned int old_flow, new_flow;
3701 if (qlen < (netdev_max_backlog >> 1))
3704 sd = this_cpu_ptr(&softnet_data);
3707 fl = rcu_dereference(sd->flow_limit);
3709 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3710 old_flow = fl->history[fl->history_head];
3711 fl->history[fl->history_head] = new_flow;
3714 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3716 if (likely(fl->buckets[old_flow]))
3717 fl->buckets[old_flow]--;
3719 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3731 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3732 * queue (may be a remote CPU queue).
3734 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3735 unsigned int *qtail)
3737 struct softnet_data *sd;
3738 unsigned long flags;
3741 sd = &per_cpu(softnet_data, cpu);
3743 local_irq_save(flags);
3746 if (!netif_running(skb->dev))
3748 qlen = skb_queue_len(&sd->input_pkt_queue);
3749 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3752 __skb_queue_tail(&sd->input_pkt_queue, skb);
3753 input_queue_tail_incr_save(sd, qtail);
3755 local_irq_restore(flags);
3756 return NET_RX_SUCCESS;
3759 /* Schedule NAPI for backlog device
3760 * We can use non atomic operation since we own the queue lock
3762 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3763 if (!rps_ipi_queued(sd))
3764 ____napi_schedule(sd, &sd->backlog);
3773 local_irq_restore(flags);
3775 atomic_long_inc(&skb->dev->rx_dropped);
3780 static int netif_rx_internal(struct sk_buff *skb)
3784 net_timestamp_check(netdev_tstamp_prequeue, skb);
3786 trace_netif_rx(skb);
3788 if (static_key_false(&rps_needed)) {
3789 struct rps_dev_flow voidflow, *rflow = &voidflow;
3795 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3797 cpu = smp_processor_id();
3799 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3807 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3814 * netif_rx - post buffer to the network code
3815 * @skb: buffer to post
3817 * This function receives a packet from a device driver and queues it for
3818 * the upper (protocol) levels to process. It always succeeds. The buffer
3819 * may be dropped during processing for congestion control or by the
3823 * NET_RX_SUCCESS (no congestion)
3824 * NET_RX_DROP (packet was dropped)
3828 int netif_rx(struct sk_buff *skb)
3830 trace_netif_rx_entry(skb);
3832 return netif_rx_internal(skb);
3834 EXPORT_SYMBOL(netif_rx);
3836 int netif_rx_ni(struct sk_buff *skb)
3840 trace_netif_rx_ni_entry(skb);
3843 err = netif_rx_internal(skb);
3844 if (local_softirq_pending())
3850 EXPORT_SYMBOL(netif_rx_ni);
3852 static void net_tx_action(struct softirq_action *h)
3854 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3856 if (sd->completion_queue) {
3857 struct sk_buff *clist;
3859 local_irq_disable();
3860 clist = sd->completion_queue;
3861 sd->completion_queue = NULL;
3865 struct sk_buff *skb = clist;
3866 clist = clist->next;
3868 WARN_ON(atomic_read(&skb->users));
3869 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3870 trace_consume_skb(skb);
3872 trace_kfree_skb(skb, net_tx_action);
3874 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
3877 __kfree_skb_defer(skb);
3880 __kfree_skb_flush();
3883 if (sd->output_queue) {
3886 local_irq_disable();
3887 head = sd->output_queue;
3888 sd->output_queue = NULL;
3889 sd->output_queue_tailp = &sd->output_queue;
3893 struct Qdisc *q = head;
3894 spinlock_t *root_lock;
3896 head = head->next_sched;
3898 root_lock = qdisc_lock(q);
3899 if (spin_trylock(root_lock)) {
3900 smp_mb__before_atomic();
3901 clear_bit(__QDISC_STATE_SCHED,
3904 spin_unlock(root_lock);
3906 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3908 __netif_reschedule(q);
3910 smp_mb__before_atomic();
3911 clear_bit(__QDISC_STATE_SCHED,
3919 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3920 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3921 /* This hook is defined here for ATM LANE */
3922 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3923 unsigned char *addr) __read_mostly;
3924 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3927 static inline struct sk_buff *
3928 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
3929 struct net_device *orig_dev)
3931 #ifdef CONFIG_NET_CLS_ACT
3932 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3933 struct tcf_result cl_res;
3935 /* If there's at least one ingress present somewhere (so
3936 * we get here via enabled static key), remaining devices
3937 * that are not configured with an ingress qdisc will bail
3943 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3947 qdisc_skb_cb(skb)->pkt_len = skb->len;
3948 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3949 qdisc_bstats_cpu_update(cl->q, skb);
3951 switch (tc_classify(skb, cl, &cl_res, false)) {
3953 case TC_ACT_RECLASSIFY:
3954 skb->tc_index = TC_H_MIN(cl_res.classid);
3957 qdisc_qstats_cpu_drop(cl->q);
3962 case TC_ACT_REDIRECT:
3963 /* skb_mac_header check was done by cls/act_bpf, so
3964 * we can safely push the L2 header back before
3965 * redirecting to another netdev
3967 __skb_push(skb, skb->mac_len);
3968 skb_do_redirect(skb);
3973 #endif /* CONFIG_NET_CLS_ACT */
3978 * netdev_rx_handler_register - register receive handler
3979 * @dev: device to register a handler for
3980 * @rx_handler: receive handler to register
3981 * @rx_handler_data: data pointer that is used by rx handler
3983 * Register a receive handler for a device. This handler will then be
3984 * called from __netif_receive_skb. A negative errno code is returned
3987 * The caller must hold the rtnl_mutex.
3989 * For a general description of rx_handler, see enum rx_handler_result.
3991 int netdev_rx_handler_register(struct net_device *dev,
3992 rx_handler_func_t *rx_handler,
3993 void *rx_handler_data)
3997 if (dev->rx_handler)
4000 /* Note: rx_handler_data must be set before rx_handler */
4001 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4002 rcu_assign_pointer(dev->rx_handler, rx_handler);
4006 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4009 * netdev_rx_handler_unregister - unregister receive handler
4010 * @dev: device to unregister a handler from
4012 * Unregister a receive handler from a device.
4014 * The caller must hold the rtnl_mutex.
4016 void netdev_rx_handler_unregister(struct net_device *dev)
4020 RCU_INIT_POINTER(dev->rx_handler, NULL);
4021 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4022 * section has a guarantee to see a non NULL rx_handler_data
4026 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4028 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4031 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4032 * the special handling of PFMEMALLOC skbs.
4034 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4036 switch (skb->protocol) {
4037 case htons(ETH_P_ARP):
4038 case htons(ETH_P_IP):
4039 case htons(ETH_P_IPV6):
4040 case htons(ETH_P_8021Q):
4041 case htons(ETH_P_8021AD):
4048 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4049 int *ret, struct net_device *orig_dev)
4051 #ifdef CONFIG_NETFILTER_INGRESS
4052 if (nf_hook_ingress_active(skb)) {
4054 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4058 return nf_hook_ingress(skb);
4060 #endif /* CONFIG_NETFILTER_INGRESS */
4064 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
4066 struct packet_type *ptype, *pt_prev;
4067 rx_handler_func_t *rx_handler;
4068 struct net_device *orig_dev;
4069 bool deliver_exact = false;
4070 int ret = NET_RX_DROP;
4073 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4075 trace_netif_receive_skb(skb);
4077 orig_dev = skb->dev;
4079 skb_reset_network_header(skb);
4080 if (!skb_transport_header_was_set(skb))
4081 skb_reset_transport_header(skb);
4082 skb_reset_mac_len(skb);
4087 skb->skb_iif = skb->dev->ifindex;
4089 __this_cpu_inc(softnet_data.processed);
4091 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4092 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4093 skb = skb_vlan_untag(skb);
4098 #ifdef CONFIG_NET_CLS_ACT
4099 if (skb->tc_verd & TC_NCLS) {
4100 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
4108 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4110 ret = deliver_skb(skb, pt_prev, orig_dev);
4114 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4116 ret = deliver_skb(skb, pt_prev, orig_dev);
4121 #ifdef CONFIG_NET_INGRESS
4122 if (static_key_false(&ingress_needed)) {
4123 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4127 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4131 #ifdef CONFIG_NET_CLS_ACT
4135 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4138 if (skb_vlan_tag_present(skb)) {
4140 ret = deliver_skb(skb, pt_prev, orig_dev);
4143 if (vlan_do_receive(&skb))
4145 else if (unlikely(!skb))
4149 rx_handler = rcu_dereference(skb->dev->rx_handler);
4152 ret = deliver_skb(skb, pt_prev, orig_dev);
4155 switch (rx_handler(&skb)) {
4156 case RX_HANDLER_CONSUMED:
4157 ret = NET_RX_SUCCESS;
4159 case RX_HANDLER_ANOTHER:
4161 case RX_HANDLER_EXACT:
4162 deliver_exact = true;
4163 case RX_HANDLER_PASS:
4170 if (unlikely(skb_vlan_tag_present(skb))) {
4171 if (skb_vlan_tag_get_id(skb))
4172 skb->pkt_type = PACKET_OTHERHOST;
4173 /* Note: we might in the future use prio bits
4174 * and set skb->priority like in vlan_do_receive()
4175 * For the time being, just ignore Priority Code Point
4180 type = skb->protocol;
4182 /* deliver only exact match when indicated */
4183 if (likely(!deliver_exact)) {
4184 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4185 &ptype_base[ntohs(type) &
4189 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4190 &orig_dev->ptype_specific);
4192 if (unlikely(skb->dev != orig_dev)) {
4193 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4194 &skb->dev->ptype_specific);
4198 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
4201 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4205 atomic_long_inc(&skb->dev->rx_dropped);
4207 atomic_long_inc(&skb->dev->rx_nohandler);
4209 /* Jamal, now you will not able to escape explaining
4210 * me how you were going to use this. :-)
4219 static int __netif_receive_skb(struct sk_buff *skb)
4223 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4224 unsigned long pflags = current->flags;
4227 * PFMEMALLOC skbs are special, they should
4228 * - be delivered to SOCK_MEMALLOC sockets only
4229 * - stay away from userspace
4230 * - have bounded memory usage
4232 * Use PF_MEMALLOC as this saves us from propagating the allocation
4233 * context down to all allocation sites.
4235 current->flags |= PF_MEMALLOC;
4236 ret = __netif_receive_skb_core(skb, true);
4237 tsk_restore_flags(current, pflags, PF_MEMALLOC);
4239 ret = __netif_receive_skb_core(skb, false);
4244 static int netif_receive_skb_internal(struct sk_buff *skb)
4248 net_timestamp_check(netdev_tstamp_prequeue, skb);
4250 if (skb_defer_rx_timestamp(skb))
4251 return NET_RX_SUCCESS;
4256 if (static_key_false(&rps_needed)) {
4257 struct rps_dev_flow voidflow, *rflow = &voidflow;
4258 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4261 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4267 ret = __netif_receive_skb(skb);
4273 * netif_receive_skb - process receive buffer from network
4274 * @skb: buffer to process
4276 * netif_receive_skb() is the main receive data processing function.
4277 * It always succeeds. The buffer may be dropped during processing
4278 * for congestion control or by the protocol layers.
4280 * This function may only be called from softirq context and interrupts
4281 * should be enabled.
4283 * Return values (usually ignored):
4284 * NET_RX_SUCCESS: no congestion
4285 * NET_RX_DROP: packet was dropped
4287 int netif_receive_skb(struct sk_buff *skb)
4289 trace_netif_receive_skb_entry(skb);
4291 return netif_receive_skb_internal(skb);
4293 EXPORT_SYMBOL(netif_receive_skb);
4295 /* Network device is going away, flush any packets still pending
4296 * Called with irqs disabled.
4298 static void flush_backlog(void *arg)
4300 struct net_device *dev = arg;
4301 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4302 struct sk_buff *skb, *tmp;
4305 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4306 if (skb->dev == dev) {
4307 __skb_unlink(skb, &sd->input_pkt_queue);
4309 input_queue_head_incr(sd);
4314 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4315 if (skb->dev == dev) {
4316 __skb_unlink(skb, &sd->process_queue);
4318 input_queue_head_incr(sd);
4323 static int napi_gro_complete(struct sk_buff *skb)
4325 struct packet_offload *ptype;
4326 __be16 type = skb->protocol;
4327 struct list_head *head = &offload_base;
4330 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4332 if (NAPI_GRO_CB(skb)->count == 1) {
4333 skb_shinfo(skb)->gso_size = 0;
4338 list_for_each_entry_rcu(ptype, head, list) {
4339 if (ptype->type != type || !ptype->callbacks.gro_complete)
4342 err = ptype->callbacks.gro_complete(skb, 0);
4348 WARN_ON(&ptype->list == head);
4350 return NET_RX_SUCCESS;
4354 return netif_receive_skb_internal(skb);
4357 /* napi->gro_list contains packets ordered by age.
4358 * youngest packets at the head of it.
4359 * Complete skbs in reverse order to reduce latencies.
4361 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4363 struct sk_buff *skb, *prev = NULL;
4365 /* scan list and build reverse chain */
4366 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4371 for (skb = prev; skb; skb = prev) {
4374 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4378 napi_gro_complete(skb);
4382 napi->gro_list = NULL;
4384 EXPORT_SYMBOL(napi_gro_flush);
4386 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4389 unsigned int maclen = skb->dev->hard_header_len;
4390 u32 hash = skb_get_hash_raw(skb);
4392 for (p = napi->gro_list; p; p = p->next) {
4393 unsigned long diffs;
4395 NAPI_GRO_CB(p)->flush = 0;
4397 if (hash != skb_get_hash_raw(p)) {
4398 NAPI_GRO_CB(p)->same_flow = 0;
4402 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4403 diffs |= p->vlan_tci ^ skb->vlan_tci;
4404 diffs |= skb_metadata_dst_cmp(p, skb);
4405 if (maclen == ETH_HLEN)
4406 diffs |= compare_ether_header(skb_mac_header(p),
4407 skb_mac_header(skb));
4409 diffs = memcmp(skb_mac_header(p),
4410 skb_mac_header(skb),
4412 NAPI_GRO_CB(p)->same_flow = !diffs;
4416 static void skb_gro_reset_offset(struct sk_buff *skb)
4418 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4419 const skb_frag_t *frag0 = &pinfo->frags[0];
4421 NAPI_GRO_CB(skb)->data_offset = 0;
4422 NAPI_GRO_CB(skb)->frag0 = NULL;
4423 NAPI_GRO_CB(skb)->frag0_len = 0;
4425 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4427 !PageHighMem(skb_frag_page(frag0))) {
4428 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4429 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4433 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4435 struct skb_shared_info *pinfo = skb_shinfo(skb);
4437 BUG_ON(skb->end - skb->tail < grow);
4439 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4441 skb->data_len -= grow;
4444 pinfo->frags[0].page_offset += grow;
4445 skb_frag_size_sub(&pinfo->frags[0], grow);
4447 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4448 skb_frag_unref(skb, 0);
4449 memmove(pinfo->frags, pinfo->frags + 1,
4450 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4454 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4456 struct sk_buff **pp = NULL;
4457 struct packet_offload *ptype;
4458 __be16 type = skb->protocol;
4459 struct list_head *head = &offload_base;
4461 enum gro_result ret;
4464 if (!(skb->dev->features & NETIF_F_GRO))
4467 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4470 gro_list_prepare(napi, skb);
4473 list_for_each_entry_rcu(ptype, head, list) {
4474 if (ptype->type != type || !ptype->callbacks.gro_receive)
4477 skb_set_network_header(skb, skb_gro_offset(skb));
4478 skb_reset_mac_len(skb);
4479 NAPI_GRO_CB(skb)->same_flow = 0;
4480 NAPI_GRO_CB(skb)->flush = 0;
4481 NAPI_GRO_CB(skb)->free = 0;
4482 NAPI_GRO_CB(skb)->encap_mark = 0;
4483 NAPI_GRO_CB(skb)->is_fou = 0;
4484 NAPI_GRO_CB(skb)->is_atomic = 1;
4485 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4487 /* Setup for GRO checksum validation */
4488 switch (skb->ip_summed) {
4489 case CHECKSUM_COMPLETE:
4490 NAPI_GRO_CB(skb)->csum = skb->csum;
4491 NAPI_GRO_CB(skb)->csum_valid = 1;
4492 NAPI_GRO_CB(skb)->csum_cnt = 0;
4494 case CHECKSUM_UNNECESSARY:
4495 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4496 NAPI_GRO_CB(skb)->csum_valid = 0;
4499 NAPI_GRO_CB(skb)->csum_cnt = 0;
4500 NAPI_GRO_CB(skb)->csum_valid = 0;
4503 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4508 if (&ptype->list == head)
4511 same_flow = NAPI_GRO_CB(skb)->same_flow;
4512 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4515 struct sk_buff *nskb = *pp;
4519 napi_gro_complete(nskb);
4526 if (NAPI_GRO_CB(skb)->flush)
4529 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4530 struct sk_buff *nskb = napi->gro_list;
4532 /* locate the end of the list to select the 'oldest' flow */
4533 while (nskb->next) {
4539 napi_gro_complete(nskb);
4543 NAPI_GRO_CB(skb)->count = 1;
4544 NAPI_GRO_CB(skb)->age = jiffies;
4545 NAPI_GRO_CB(skb)->last = skb;
4546 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4547 skb->next = napi->gro_list;
4548 napi->gro_list = skb;
4552 grow = skb_gro_offset(skb) - skb_headlen(skb);
4554 gro_pull_from_frag0(skb, grow);
4563 struct packet_offload *gro_find_receive_by_type(__be16 type)
4565 struct list_head *offload_head = &offload_base;
4566 struct packet_offload *ptype;
4568 list_for_each_entry_rcu(ptype, offload_head, list) {
4569 if (ptype->type != type || !ptype->callbacks.gro_receive)
4575 EXPORT_SYMBOL(gro_find_receive_by_type);
4577 struct packet_offload *gro_find_complete_by_type(__be16 type)
4579 struct list_head *offload_head = &offload_base;
4580 struct packet_offload *ptype;
4582 list_for_each_entry_rcu(ptype, offload_head, list) {
4583 if (ptype->type != type || !ptype->callbacks.gro_complete)
4589 EXPORT_SYMBOL(gro_find_complete_by_type);
4591 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4595 if (netif_receive_skb_internal(skb))
4603 case GRO_MERGED_FREE:
4604 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) {
4606 kmem_cache_free(skbuff_head_cache, skb);
4620 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4622 skb_mark_napi_id(skb, napi);
4623 trace_napi_gro_receive_entry(skb);
4625 skb_gro_reset_offset(skb);
4627 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4629 EXPORT_SYMBOL(napi_gro_receive);
4631 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4633 if (unlikely(skb->pfmemalloc)) {
4637 __skb_pull(skb, skb_headlen(skb));
4638 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4639 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4641 skb->dev = napi->dev;
4643 skb->encapsulation = 0;
4644 skb_shinfo(skb)->gso_type = 0;
4645 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4650 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4652 struct sk_buff *skb = napi->skb;
4655 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4658 skb_mark_napi_id(skb, napi);
4663 EXPORT_SYMBOL(napi_get_frags);
4665 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4666 struct sk_buff *skb,
4672 __skb_push(skb, ETH_HLEN);
4673 skb->protocol = eth_type_trans(skb, skb->dev);
4674 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4679 case GRO_MERGED_FREE:
4680 napi_reuse_skb(napi, skb);
4690 /* Upper GRO stack assumes network header starts at gro_offset=0
4691 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4692 * We copy ethernet header into skb->data to have a common layout.
4694 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4696 struct sk_buff *skb = napi->skb;
4697 const struct ethhdr *eth;
4698 unsigned int hlen = sizeof(*eth);
4702 skb_reset_mac_header(skb);
4703 skb_gro_reset_offset(skb);
4705 eth = skb_gro_header_fast(skb, 0);
4706 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4707 eth = skb_gro_header_slow(skb, hlen, 0);
4708 if (unlikely(!eth)) {
4709 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
4710 __func__, napi->dev->name);
4711 napi_reuse_skb(napi, skb);
4715 gro_pull_from_frag0(skb, hlen);
4716 NAPI_GRO_CB(skb)->frag0 += hlen;
4717 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4719 __skb_pull(skb, hlen);
4722 * This works because the only protocols we care about don't require
4724 * We'll fix it up properly in napi_frags_finish()
4726 skb->protocol = eth->h_proto;
4731 gro_result_t napi_gro_frags(struct napi_struct *napi)
4733 struct sk_buff *skb = napi_frags_skb(napi);
4738 trace_napi_gro_frags_entry(skb);
4740 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4742 EXPORT_SYMBOL(napi_gro_frags);
4744 /* Compute the checksum from gro_offset and return the folded value
4745 * after adding in any pseudo checksum.
4747 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4752 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4754 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4755 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4757 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4758 !skb->csum_complete_sw)
4759 netdev_rx_csum_fault(skb->dev);
4762 NAPI_GRO_CB(skb)->csum = wsum;
4763 NAPI_GRO_CB(skb)->csum_valid = 1;
4767 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4770 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4771 * Note: called with local irq disabled, but exits with local irq enabled.
4773 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4776 struct softnet_data *remsd = sd->rps_ipi_list;
4779 sd->rps_ipi_list = NULL;
4783 /* Send pending IPI's to kick RPS processing on remote cpus. */
4785 struct softnet_data *next = remsd->rps_ipi_next;
4787 if (cpu_online(remsd->cpu))
4788 smp_call_function_single_async(remsd->cpu,
4797 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4800 return sd->rps_ipi_list != NULL;
4806 static int process_backlog(struct napi_struct *napi, int quota)
4809 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4811 /* Check if we have pending ipi, its better to send them now,
4812 * not waiting net_rx_action() end.
4814 if (sd_has_rps_ipi_waiting(sd)) {
4815 local_irq_disable();
4816 net_rps_action_and_irq_enable(sd);
4819 napi->weight = weight_p;
4820 local_irq_disable();
4822 struct sk_buff *skb;
4824 while ((skb = __skb_dequeue(&sd->process_queue))) {
4827 __netif_receive_skb(skb);
4829 local_irq_disable();
4830 input_queue_head_incr(sd);
4831 if (++work >= quota) {
4838 if (skb_queue_empty(&sd->input_pkt_queue)) {
4840 * Inline a custom version of __napi_complete().
4841 * only current cpu owns and manipulates this napi,
4842 * and NAPI_STATE_SCHED is the only possible flag set
4844 * We can use a plain write instead of clear_bit(),
4845 * and we dont need an smp_mb() memory barrier.
4853 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4854 &sd->process_queue);
4863 * __napi_schedule - schedule for receive
4864 * @n: entry to schedule
4866 * The entry's receive function will be scheduled to run.
4867 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4869 void __napi_schedule(struct napi_struct *n)
4871 unsigned long flags;
4873 local_irq_save(flags);
4874 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4875 local_irq_restore(flags);
4877 EXPORT_SYMBOL(__napi_schedule);
4880 * __napi_schedule_irqoff - schedule for receive
4881 * @n: entry to schedule
4883 * Variant of __napi_schedule() assuming hard irqs are masked
4885 void __napi_schedule_irqoff(struct napi_struct *n)
4887 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4889 EXPORT_SYMBOL(__napi_schedule_irqoff);
4891 void __napi_complete(struct napi_struct *n)
4893 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4895 list_del_init(&n->poll_list);
4896 smp_mb__before_atomic();
4897 clear_bit(NAPI_STATE_SCHED, &n->state);
4899 EXPORT_SYMBOL(__napi_complete);
4901 void napi_complete_done(struct napi_struct *n, int work_done)
4903 unsigned long flags;
4906 * don't let napi dequeue from the cpu poll list
4907 * just in case its running on a different cpu
4909 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4913 unsigned long timeout = 0;
4916 timeout = n->dev->gro_flush_timeout;
4919 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4920 HRTIMER_MODE_REL_PINNED);
4922 napi_gro_flush(n, false);
4924 if (likely(list_empty(&n->poll_list))) {
4925 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4927 /* If n->poll_list is not empty, we need to mask irqs */
4928 local_irq_save(flags);
4930 local_irq_restore(flags);
4933 EXPORT_SYMBOL(napi_complete_done);
4935 /* must be called under rcu_read_lock(), as we dont take a reference */
4936 static struct napi_struct *napi_by_id(unsigned int napi_id)
4938 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4939 struct napi_struct *napi;
4941 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4942 if (napi->napi_id == napi_id)
4948 #if defined(CONFIG_NET_RX_BUSY_POLL)
4949 #define BUSY_POLL_BUDGET 8
4950 bool sk_busy_loop(struct sock *sk, int nonblock)
4952 unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0;
4953 int (*busy_poll)(struct napi_struct *dev);
4954 struct napi_struct *napi;
4959 napi = napi_by_id(sk->sk_napi_id);
4963 /* Note: ndo_busy_poll method is optional in linux-4.5 */
4964 busy_poll = napi->dev->netdev_ops->ndo_busy_poll;
4970 rc = busy_poll(napi);
4971 } else if (napi_schedule_prep(napi)) {
4972 void *have = netpoll_poll_lock(napi);
4974 if (test_bit(NAPI_STATE_SCHED, &napi->state)) {
4975 rc = napi->poll(napi, BUSY_POLL_BUDGET);
4976 trace_napi_poll(napi);
4977 if (rc == BUSY_POLL_BUDGET) {
4978 napi_complete_done(napi, rc);
4979 napi_schedule(napi);
4982 netpoll_poll_unlock(have);
4985 __NET_ADD_STATS(sock_net(sk),
4986 LINUX_MIB_BUSYPOLLRXPACKETS, rc);
4989 if (rc == LL_FLUSH_FAILED)
4990 break; /* permanent failure */
4993 } while (!nonblock && skb_queue_empty(&sk->sk_receive_queue) &&
4994 !need_resched() && !busy_loop_timeout(end_time));
4996 rc = !skb_queue_empty(&sk->sk_receive_queue);
5001 EXPORT_SYMBOL(sk_busy_loop);
5003 #endif /* CONFIG_NET_RX_BUSY_POLL */
5005 void napi_hash_add(struct napi_struct *napi)
5007 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
5008 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
5011 spin_lock(&napi_hash_lock);
5013 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */
5015 if (unlikely(++napi_gen_id < NR_CPUS + 1))
5016 napi_gen_id = NR_CPUS + 1;
5017 } while (napi_by_id(napi_gen_id));
5018 napi->napi_id = napi_gen_id;
5020 hlist_add_head_rcu(&napi->napi_hash_node,
5021 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
5023 spin_unlock(&napi_hash_lock);
5025 EXPORT_SYMBOL_GPL(napi_hash_add);
5027 /* Warning : caller is responsible to make sure rcu grace period
5028 * is respected before freeing memory containing @napi
5030 bool napi_hash_del(struct napi_struct *napi)
5032 bool rcu_sync_needed = false;
5034 spin_lock(&napi_hash_lock);
5036 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
5037 rcu_sync_needed = true;
5038 hlist_del_rcu(&napi->napi_hash_node);
5040 spin_unlock(&napi_hash_lock);
5041 return rcu_sync_needed;
5043 EXPORT_SYMBOL_GPL(napi_hash_del);
5045 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
5047 struct napi_struct *napi;
5049 napi = container_of(timer, struct napi_struct, timer);
5051 napi_schedule(napi);
5053 return HRTIMER_NORESTART;
5056 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
5057 int (*poll)(struct napi_struct *, int), int weight)
5059 INIT_LIST_HEAD(&napi->poll_list);
5060 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
5061 napi->timer.function = napi_watchdog;
5062 napi->gro_count = 0;
5063 napi->gro_list = NULL;
5066 if (weight > NAPI_POLL_WEIGHT)
5067 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5069 napi->weight = weight;
5070 list_add(&napi->dev_list, &dev->napi_list);
5072 #ifdef CONFIG_NETPOLL
5073 spin_lock_init(&napi->poll_lock);
5074 napi->poll_owner = -1;
5076 set_bit(NAPI_STATE_SCHED, &napi->state);
5077 napi_hash_add(napi);
5079 EXPORT_SYMBOL(netif_napi_add);
5081 void napi_disable(struct napi_struct *n)
5084 set_bit(NAPI_STATE_DISABLE, &n->state);
5086 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
5088 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
5091 hrtimer_cancel(&n->timer);
5093 clear_bit(NAPI_STATE_DISABLE, &n->state);
5095 EXPORT_SYMBOL(napi_disable);
5097 /* Must be called in process context */
5098 void netif_napi_del(struct napi_struct *napi)
5101 if (napi_hash_del(napi))
5103 list_del_init(&napi->dev_list);
5104 napi_free_frags(napi);
5106 kfree_skb_list(napi->gro_list);
5107 napi->gro_list = NULL;
5108 napi->gro_count = 0;
5110 EXPORT_SYMBOL(netif_napi_del);
5112 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
5117 list_del_init(&n->poll_list);
5119 have = netpoll_poll_lock(n);
5123 /* This NAPI_STATE_SCHED test is for avoiding a race
5124 * with netpoll's poll_napi(). Only the entity which
5125 * obtains the lock and sees NAPI_STATE_SCHED set will
5126 * actually make the ->poll() call. Therefore we avoid
5127 * accidentally calling ->poll() when NAPI is not scheduled.
5130 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
5131 work = n->poll(n, weight);
5135 WARN_ON_ONCE(work > weight);
5137 if (likely(work < weight))
5140 /* Drivers must not modify the NAPI state if they
5141 * consume the entire weight. In such cases this code
5142 * still "owns" the NAPI instance and therefore can
5143 * move the instance around on the list at-will.
5145 if (unlikely(napi_disable_pending(n))) {
5151 /* flush too old packets
5152 * If HZ < 1000, flush all packets.
5154 napi_gro_flush(n, HZ >= 1000);
5157 /* Some drivers may have called napi_schedule
5158 * prior to exhausting their budget.
5160 if (unlikely(!list_empty(&n->poll_list))) {
5161 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5162 n->dev ? n->dev->name : "backlog");
5166 list_add_tail(&n->poll_list, repoll);
5169 netpoll_poll_unlock(have);
5174 static void net_rx_action(struct softirq_action *h)
5176 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5177 unsigned long time_limit = jiffies + 2;
5178 int budget = netdev_budget;
5182 local_irq_disable();
5183 list_splice_init(&sd->poll_list, &list);
5187 struct napi_struct *n;
5189 if (list_empty(&list)) {
5190 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
5195 n = list_first_entry(&list, struct napi_struct, poll_list);
5196 budget -= napi_poll(n, &repoll);
5198 /* If softirq window is exhausted then punt.
5199 * Allow this to run for 2 jiffies since which will allow
5200 * an average latency of 1.5/HZ.
5202 if (unlikely(budget <= 0 ||
5203 time_after_eq(jiffies, time_limit))) {
5209 __kfree_skb_flush();
5210 local_irq_disable();
5212 list_splice_tail_init(&sd->poll_list, &list);
5213 list_splice_tail(&repoll, &list);
5214 list_splice(&list, &sd->poll_list);
5215 if (!list_empty(&sd->poll_list))
5216 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5218 net_rps_action_and_irq_enable(sd);
5221 struct netdev_adjacent {
5222 struct net_device *dev;
5224 /* upper master flag, there can only be one master device per list */
5227 /* counter for the number of times this device was added to us */
5230 /* private field for the users */
5233 struct list_head list;
5234 struct rcu_head rcu;
5237 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
5238 struct list_head *adj_list)
5240 struct netdev_adjacent *adj;
5242 list_for_each_entry(adj, adj_list, list) {
5243 if (adj->dev == adj_dev)
5250 * netdev_has_upper_dev - Check if device is linked to an upper device
5252 * @upper_dev: upper device to check
5254 * Find out if a device is linked to specified upper device and return true
5255 * in case it is. Note that this checks only immediate upper device,
5256 * not through a complete stack of devices. The caller must hold the RTNL lock.
5258 bool netdev_has_upper_dev(struct net_device *dev,
5259 struct net_device *upper_dev)
5263 return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper);
5265 EXPORT_SYMBOL(netdev_has_upper_dev);
5268 * netdev_has_any_upper_dev - Check if device is linked to some device
5271 * Find out if a device is linked to an upper device and return true in case
5272 * it is. The caller must hold the RTNL lock.
5274 static bool netdev_has_any_upper_dev(struct net_device *dev)
5278 return !list_empty(&dev->all_adj_list.upper);
5282 * netdev_master_upper_dev_get - Get master upper device
5285 * Find a master upper device and return pointer to it or NULL in case
5286 * it's not there. The caller must hold the RTNL lock.
5288 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5290 struct netdev_adjacent *upper;
5294 if (list_empty(&dev->adj_list.upper))
5297 upper = list_first_entry(&dev->adj_list.upper,
5298 struct netdev_adjacent, list);
5299 if (likely(upper->master))
5303 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5305 void *netdev_adjacent_get_private(struct list_head *adj_list)
5307 struct netdev_adjacent *adj;
5309 adj = list_entry(adj_list, struct netdev_adjacent, list);
5311 return adj->private;
5313 EXPORT_SYMBOL(netdev_adjacent_get_private);
5316 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5318 * @iter: list_head ** of the current position
5320 * Gets the next device from the dev's upper list, starting from iter
5321 * position. The caller must hold RCU read lock.
5323 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5324 struct list_head **iter)
5326 struct netdev_adjacent *upper;
5328 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5330 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5332 if (&upper->list == &dev->adj_list.upper)
5335 *iter = &upper->list;
5339 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5342 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
5344 * @iter: list_head ** of the current position
5346 * Gets the next device from the dev's upper list, starting from iter
5347 * position. The caller must hold RCU read lock.
5349 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
5350 struct list_head **iter)
5352 struct netdev_adjacent *upper;
5354 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5356 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5358 if (&upper->list == &dev->all_adj_list.upper)
5361 *iter = &upper->list;
5365 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
5368 * netdev_lower_get_next_private - Get the next ->private from the
5369 * lower neighbour list
5371 * @iter: list_head ** of the current position
5373 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5374 * list, starting from iter position. The caller must hold either hold the
5375 * RTNL lock or its own locking that guarantees that the neighbour lower
5376 * list will remain unchanged.
5378 void *netdev_lower_get_next_private(struct net_device *dev,
5379 struct list_head **iter)
5381 struct netdev_adjacent *lower;
5383 lower = list_entry(*iter, struct netdev_adjacent, list);
5385 if (&lower->list == &dev->adj_list.lower)
5388 *iter = lower->list.next;
5390 return lower->private;
5392 EXPORT_SYMBOL(netdev_lower_get_next_private);
5395 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5396 * lower neighbour list, RCU
5399 * @iter: list_head ** of the current position
5401 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5402 * list, starting from iter position. The caller must hold RCU read lock.
5404 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5405 struct list_head **iter)
5407 struct netdev_adjacent *lower;
5409 WARN_ON_ONCE(!rcu_read_lock_held());
5411 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5413 if (&lower->list == &dev->adj_list.lower)
5416 *iter = &lower->list;
5418 return lower->private;
5420 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5423 * netdev_lower_get_next - Get the next device from the lower neighbour
5426 * @iter: list_head ** of the current position
5428 * Gets the next netdev_adjacent from the dev's lower neighbour
5429 * list, starting from iter position. The caller must hold RTNL lock or
5430 * its own locking that guarantees that the neighbour lower
5431 * list will remain unchanged.
5433 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5435 struct netdev_adjacent *lower;
5437 lower = list_entry(*iter, struct netdev_adjacent, list);
5439 if (&lower->list == &dev->adj_list.lower)
5442 *iter = lower->list.next;
5446 EXPORT_SYMBOL(netdev_lower_get_next);
5449 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5450 * lower neighbour list, RCU
5454 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5455 * list. The caller must hold RCU read lock.
5457 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5459 struct netdev_adjacent *lower;
5461 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5462 struct netdev_adjacent, list);
5464 return lower->private;
5467 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5470 * netdev_master_upper_dev_get_rcu - Get master upper device
5473 * Find a master upper device and return pointer to it or NULL in case
5474 * it's not there. The caller must hold the RCU read lock.
5476 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5478 struct netdev_adjacent *upper;
5480 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5481 struct netdev_adjacent, list);
5482 if (upper && likely(upper->master))
5486 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5488 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5489 struct net_device *adj_dev,
5490 struct list_head *dev_list)
5492 char linkname[IFNAMSIZ+7];
5493 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5494 "upper_%s" : "lower_%s", adj_dev->name);
5495 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5498 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5500 struct list_head *dev_list)
5502 char linkname[IFNAMSIZ+7];
5503 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5504 "upper_%s" : "lower_%s", name);
5505 sysfs_remove_link(&(dev->dev.kobj), linkname);
5508 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5509 struct net_device *adj_dev,
5510 struct list_head *dev_list)
5512 return (dev_list == &dev->adj_list.upper ||
5513 dev_list == &dev->adj_list.lower) &&
5514 net_eq(dev_net(dev), dev_net(adj_dev));
5517 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5518 struct net_device *adj_dev,
5519 struct list_head *dev_list,
5520 void *private, bool master)
5522 struct netdev_adjacent *adj;
5525 adj = __netdev_find_adj(adj_dev, dev_list);
5532 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5537 adj->master = master;
5539 adj->private = private;
5542 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5543 adj_dev->name, dev->name, adj_dev->name);
5545 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5546 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5551 /* Ensure that master link is always the first item in list. */
5553 ret = sysfs_create_link(&(dev->dev.kobj),
5554 &(adj_dev->dev.kobj), "master");
5556 goto remove_symlinks;
5558 list_add_rcu(&adj->list, dev_list);
5560 list_add_tail_rcu(&adj->list, dev_list);
5566 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5567 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5575 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5576 struct net_device *adj_dev,
5577 struct list_head *dev_list)
5579 struct netdev_adjacent *adj;
5581 adj = __netdev_find_adj(adj_dev, dev_list);
5584 pr_err("tried to remove device %s from %s\n",
5585 dev->name, adj_dev->name);
5589 if (adj->ref_nr > 1) {
5590 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5597 sysfs_remove_link(&(dev->dev.kobj), "master");
5599 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5600 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5602 list_del_rcu(&adj->list);
5603 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5604 adj_dev->name, dev->name, adj_dev->name);
5606 kfree_rcu(adj, rcu);
5609 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5610 struct net_device *upper_dev,
5611 struct list_head *up_list,
5612 struct list_head *down_list,
5613 void *private, bool master)
5617 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5622 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5625 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5632 static int __netdev_adjacent_dev_link(struct net_device *dev,
5633 struct net_device *upper_dev)
5635 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5636 &dev->all_adj_list.upper,
5637 &upper_dev->all_adj_list.lower,
5641 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5642 struct net_device *upper_dev,
5643 struct list_head *up_list,
5644 struct list_head *down_list)
5646 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5647 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5650 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5651 struct net_device *upper_dev)
5653 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5654 &dev->all_adj_list.upper,
5655 &upper_dev->all_adj_list.lower);
5658 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5659 struct net_device *upper_dev,
5660 void *private, bool master)
5662 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5667 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5668 &dev->adj_list.upper,
5669 &upper_dev->adj_list.lower,
5672 __netdev_adjacent_dev_unlink(dev, upper_dev);
5679 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5680 struct net_device *upper_dev)
5682 __netdev_adjacent_dev_unlink(dev, upper_dev);
5683 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5684 &dev->adj_list.upper,
5685 &upper_dev->adj_list.lower);
5688 static int __netdev_upper_dev_link(struct net_device *dev,
5689 struct net_device *upper_dev, bool master,
5690 void *upper_priv, void *upper_info)
5692 struct netdev_notifier_changeupper_info changeupper_info;
5693 struct netdev_adjacent *i, *j, *to_i, *to_j;
5698 if (dev == upper_dev)
5701 /* To prevent loops, check if dev is not upper device to upper_dev. */
5702 if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper))
5705 if (__netdev_find_adj(upper_dev, &dev->adj_list.upper))
5708 if (master && netdev_master_upper_dev_get(dev))
5711 changeupper_info.upper_dev = upper_dev;
5712 changeupper_info.master = master;
5713 changeupper_info.linking = true;
5714 changeupper_info.upper_info = upper_info;
5716 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5717 &changeupper_info.info);
5718 ret = notifier_to_errno(ret);
5722 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
5727 /* Now that we linked these devs, make all the upper_dev's
5728 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5729 * versa, and don't forget the devices itself. All of these
5730 * links are non-neighbours.
5732 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5733 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5734 pr_debug("Interlinking %s with %s, non-neighbour\n",
5735 i->dev->name, j->dev->name);
5736 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5742 /* add dev to every upper_dev's upper device */
5743 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5744 pr_debug("linking %s's upper device %s with %s\n",
5745 upper_dev->name, i->dev->name, dev->name);
5746 ret = __netdev_adjacent_dev_link(dev, i->dev);
5748 goto rollback_upper_mesh;
5751 /* add upper_dev to every dev's lower device */
5752 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5753 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5754 i->dev->name, upper_dev->name);
5755 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5757 goto rollback_lower_mesh;
5760 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5761 &changeupper_info.info);
5762 ret = notifier_to_errno(ret);
5764 goto rollback_lower_mesh;
5768 rollback_lower_mesh:
5770 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5773 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5778 rollback_upper_mesh:
5780 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5783 __netdev_adjacent_dev_unlink(dev, i->dev);
5791 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5792 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5793 if (i == to_i && j == to_j)
5795 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5801 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5807 * netdev_upper_dev_link - Add a link to the upper device
5809 * @upper_dev: new upper device
5811 * Adds a link to device which is upper to this one. The caller must hold
5812 * the RTNL lock. On a failure a negative errno code is returned.
5813 * On success the reference counts are adjusted and the function
5816 int netdev_upper_dev_link(struct net_device *dev,
5817 struct net_device *upper_dev)
5819 return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL);
5821 EXPORT_SYMBOL(netdev_upper_dev_link);
5824 * netdev_master_upper_dev_link - Add a master link to the upper device
5826 * @upper_dev: new upper device
5827 * @upper_priv: upper device private
5828 * @upper_info: upper info to be passed down via notifier
5830 * Adds a link to device which is upper to this one. In this case, only
5831 * one master upper device can be linked, although other non-master devices
5832 * might be linked as well. The caller must hold the RTNL lock.
5833 * On a failure a negative errno code is returned. On success the reference
5834 * counts are adjusted and the function returns zero.
5836 int netdev_master_upper_dev_link(struct net_device *dev,
5837 struct net_device *upper_dev,
5838 void *upper_priv, void *upper_info)
5840 return __netdev_upper_dev_link(dev, upper_dev, true,
5841 upper_priv, upper_info);
5843 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5846 * netdev_upper_dev_unlink - Removes a link to upper device
5848 * @upper_dev: new upper device
5850 * Removes a link to device which is upper to this one. The caller must hold
5853 void netdev_upper_dev_unlink(struct net_device *dev,
5854 struct net_device *upper_dev)
5856 struct netdev_notifier_changeupper_info changeupper_info;
5857 struct netdev_adjacent *i, *j;
5860 changeupper_info.upper_dev = upper_dev;
5861 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
5862 changeupper_info.linking = false;
5864 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5865 &changeupper_info.info);
5867 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5869 /* Here is the tricky part. We must remove all dev's lower
5870 * devices from all upper_dev's upper devices and vice
5871 * versa, to maintain the graph relationship.
5873 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5874 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5875 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5877 /* remove also the devices itself from lower/upper device
5880 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5881 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5883 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5884 __netdev_adjacent_dev_unlink(dev, i->dev);
5886 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5887 &changeupper_info.info);
5889 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5892 * netdev_bonding_info_change - Dispatch event about slave change
5894 * @bonding_info: info to dispatch
5896 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5897 * The caller must hold the RTNL lock.
5899 void netdev_bonding_info_change(struct net_device *dev,
5900 struct netdev_bonding_info *bonding_info)
5902 struct netdev_notifier_bonding_info info;
5904 memcpy(&info.bonding_info, bonding_info,
5905 sizeof(struct netdev_bonding_info));
5906 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5909 EXPORT_SYMBOL(netdev_bonding_info_change);
5911 static void netdev_adjacent_add_links(struct net_device *dev)
5913 struct netdev_adjacent *iter;
5915 struct net *net = dev_net(dev);
5917 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5918 if (!net_eq(net,dev_net(iter->dev)))
5920 netdev_adjacent_sysfs_add(iter->dev, dev,
5921 &iter->dev->adj_list.lower);
5922 netdev_adjacent_sysfs_add(dev, iter->dev,
5923 &dev->adj_list.upper);
5926 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5927 if (!net_eq(net,dev_net(iter->dev)))
5929 netdev_adjacent_sysfs_add(iter->dev, dev,
5930 &iter->dev->adj_list.upper);
5931 netdev_adjacent_sysfs_add(dev, iter->dev,
5932 &dev->adj_list.lower);
5936 static void netdev_adjacent_del_links(struct net_device *dev)
5938 struct netdev_adjacent *iter;
5940 struct net *net = dev_net(dev);
5942 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5943 if (!net_eq(net,dev_net(iter->dev)))
5945 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5946 &iter->dev->adj_list.lower);
5947 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5948 &dev->adj_list.upper);
5951 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5952 if (!net_eq(net,dev_net(iter->dev)))
5954 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5955 &iter->dev->adj_list.upper);
5956 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5957 &dev->adj_list.lower);
5961 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5963 struct netdev_adjacent *iter;
5965 struct net *net = dev_net(dev);
5967 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5968 if (!net_eq(net,dev_net(iter->dev)))
5970 netdev_adjacent_sysfs_del(iter->dev, oldname,
5971 &iter->dev->adj_list.lower);
5972 netdev_adjacent_sysfs_add(iter->dev, dev,
5973 &iter->dev->adj_list.lower);
5976 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5977 if (!net_eq(net,dev_net(iter->dev)))
5979 netdev_adjacent_sysfs_del(iter->dev, oldname,
5980 &iter->dev->adj_list.upper);
5981 netdev_adjacent_sysfs_add(iter->dev, dev,
5982 &iter->dev->adj_list.upper);
5986 void *netdev_lower_dev_get_private(struct net_device *dev,
5987 struct net_device *lower_dev)
5989 struct netdev_adjacent *lower;
5993 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
5997 return lower->private;
5999 EXPORT_SYMBOL(netdev_lower_dev_get_private);
6002 int dev_get_nest_level(struct net_device *dev,
6003 bool (*type_check)(const struct net_device *dev))
6005 struct net_device *lower = NULL;
6006 struct list_head *iter;
6012 netdev_for_each_lower_dev(dev, lower, iter) {
6013 nest = dev_get_nest_level(lower, type_check);
6014 if (max_nest < nest)
6018 if (type_check(dev))
6023 EXPORT_SYMBOL(dev_get_nest_level);
6026 * netdev_lower_change - Dispatch event about lower device state change
6027 * @lower_dev: device
6028 * @lower_state_info: state to dispatch
6030 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
6031 * The caller must hold the RTNL lock.
6033 void netdev_lower_state_changed(struct net_device *lower_dev,
6034 void *lower_state_info)
6036 struct netdev_notifier_changelowerstate_info changelowerstate_info;
6039 changelowerstate_info.lower_state_info = lower_state_info;
6040 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev,
6041 &changelowerstate_info.info);
6043 EXPORT_SYMBOL(netdev_lower_state_changed);
6045 static void dev_change_rx_flags(struct net_device *dev, int flags)
6047 const struct net_device_ops *ops = dev->netdev_ops;
6049 if (ops->ndo_change_rx_flags)
6050 ops->ndo_change_rx_flags(dev, flags);
6053 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
6055 unsigned int old_flags = dev->flags;
6061 dev->flags |= IFF_PROMISC;
6062 dev->promiscuity += inc;
6063 if (dev->promiscuity == 0) {
6066 * If inc causes overflow, untouch promisc and return error.
6069 dev->flags &= ~IFF_PROMISC;
6071 dev->promiscuity -= inc;
6072 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6077 if (dev->flags != old_flags) {
6078 pr_info("device %s %s promiscuous mode\n",
6080 dev->flags & IFF_PROMISC ? "entered" : "left");
6081 if (audit_enabled) {
6082 current_uid_gid(&uid, &gid);
6083 audit_log(current->audit_context, GFP_ATOMIC,
6084 AUDIT_ANOM_PROMISCUOUS,
6085 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6086 dev->name, (dev->flags & IFF_PROMISC),
6087 (old_flags & IFF_PROMISC),
6088 from_kuid(&init_user_ns, audit_get_loginuid(current)),
6089 from_kuid(&init_user_ns, uid),
6090 from_kgid(&init_user_ns, gid),
6091 audit_get_sessionid(current));
6094 dev_change_rx_flags(dev, IFF_PROMISC);
6097 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
6102 * dev_set_promiscuity - update promiscuity count on a device
6106 * Add or remove promiscuity from a device. While the count in the device
6107 * remains above zero the interface remains promiscuous. Once it hits zero
6108 * the device reverts back to normal filtering operation. A negative inc
6109 * value is used to drop promiscuity on the device.
6110 * Return 0 if successful or a negative errno code on error.
6112 int dev_set_promiscuity(struct net_device *dev, int inc)
6114 unsigned int old_flags = dev->flags;
6117 err = __dev_set_promiscuity(dev, inc, true);
6120 if (dev->flags != old_flags)
6121 dev_set_rx_mode(dev);
6124 EXPORT_SYMBOL(dev_set_promiscuity);
6126 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
6128 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
6132 dev->flags |= IFF_ALLMULTI;
6133 dev->allmulti += inc;
6134 if (dev->allmulti == 0) {
6137 * If inc causes overflow, untouch allmulti and return error.
6140 dev->flags &= ~IFF_ALLMULTI;
6142 dev->allmulti -= inc;
6143 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6148 if (dev->flags ^ old_flags) {
6149 dev_change_rx_flags(dev, IFF_ALLMULTI);
6150 dev_set_rx_mode(dev);
6152 __dev_notify_flags(dev, old_flags,
6153 dev->gflags ^ old_gflags);
6159 * dev_set_allmulti - update allmulti count on a device
6163 * Add or remove reception of all multicast frames to a device. While the
6164 * count in the device remains above zero the interface remains listening
6165 * to all interfaces. Once it hits zero the device reverts back to normal
6166 * filtering operation. A negative @inc value is used to drop the counter
6167 * when releasing a resource needing all multicasts.
6168 * Return 0 if successful or a negative errno code on error.
6171 int dev_set_allmulti(struct net_device *dev, int inc)
6173 return __dev_set_allmulti(dev, inc, true);
6175 EXPORT_SYMBOL(dev_set_allmulti);
6178 * Upload unicast and multicast address lists to device and
6179 * configure RX filtering. When the device doesn't support unicast
6180 * filtering it is put in promiscuous mode while unicast addresses
6183 void __dev_set_rx_mode(struct net_device *dev)
6185 const struct net_device_ops *ops = dev->netdev_ops;
6187 /* dev_open will call this function so the list will stay sane. */
6188 if (!(dev->flags&IFF_UP))
6191 if (!netif_device_present(dev))
6194 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
6195 /* Unicast addresses changes may only happen under the rtnl,
6196 * therefore calling __dev_set_promiscuity here is safe.
6198 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
6199 __dev_set_promiscuity(dev, 1, false);
6200 dev->uc_promisc = true;
6201 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
6202 __dev_set_promiscuity(dev, -1, false);
6203 dev->uc_promisc = false;
6207 if (ops->ndo_set_rx_mode)
6208 ops->ndo_set_rx_mode(dev);
6211 void dev_set_rx_mode(struct net_device *dev)
6213 netif_addr_lock_bh(dev);
6214 __dev_set_rx_mode(dev);
6215 netif_addr_unlock_bh(dev);
6219 * dev_get_flags - get flags reported to userspace
6222 * Get the combination of flag bits exported through APIs to userspace.
6224 unsigned int dev_get_flags(const struct net_device *dev)
6228 flags = (dev->flags & ~(IFF_PROMISC |
6233 (dev->gflags & (IFF_PROMISC |
6236 if (netif_running(dev)) {
6237 if (netif_oper_up(dev))
6238 flags |= IFF_RUNNING;
6239 if (netif_carrier_ok(dev))
6240 flags |= IFF_LOWER_UP;
6241 if (netif_dormant(dev))
6242 flags |= IFF_DORMANT;
6247 EXPORT_SYMBOL(dev_get_flags);
6249 int __dev_change_flags(struct net_device *dev, unsigned int flags)
6251 unsigned int old_flags = dev->flags;
6257 * Set the flags on our device.
6260 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
6261 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
6263 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6267 * Load in the correct multicast list now the flags have changed.
6270 if ((old_flags ^ flags) & IFF_MULTICAST)
6271 dev_change_rx_flags(dev, IFF_MULTICAST);
6273 dev_set_rx_mode(dev);
6276 * Have we downed the interface. We handle IFF_UP ourselves
6277 * according to user attempts to set it, rather than blindly
6282 if ((old_flags ^ flags) & IFF_UP)
6283 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
6285 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6286 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6287 unsigned int old_flags = dev->flags;
6289 dev->gflags ^= IFF_PROMISC;
6291 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6292 if (dev->flags != old_flags)
6293 dev_set_rx_mode(dev);
6296 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6297 is important. Some (broken) drivers set IFF_PROMISC, when
6298 IFF_ALLMULTI is requested not asking us and not reporting.
6300 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6301 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6303 dev->gflags ^= IFF_ALLMULTI;
6304 __dev_set_allmulti(dev, inc, false);
6310 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6311 unsigned int gchanges)
6313 unsigned int changes = dev->flags ^ old_flags;
6316 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6318 if (changes & IFF_UP) {
6319 if (dev->flags & IFF_UP)
6320 call_netdevice_notifiers(NETDEV_UP, dev);
6322 call_netdevice_notifiers(NETDEV_DOWN, dev);
6325 if (dev->flags & IFF_UP &&
6326 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6327 struct netdev_notifier_change_info change_info;
6329 change_info.flags_changed = changes;
6330 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6336 * dev_change_flags - change device settings
6338 * @flags: device state flags
6340 * Change settings on device based state flags. The flags are
6341 * in the userspace exported format.
6343 int dev_change_flags(struct net_device *dev, unsigned int flags)
6346 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6348 ret = __dev_change_flags(dev, flags);
6352 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6353 __dev_notify_flags(dev, old_flags, changes);
6356 EXPORT_SYMBOL(dev_change_flags);
6358 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6360 const struct net_device_ops *ops = dev->netdev_ops;
6362 if (ops->ndo_change_mtu)
6363 return ops->ndo_change_mtu(dev, new_mtu);
6370 * dev_set_mtu - Change maximum transfer unit
6372 * @new_mtu: new transfer unit
6374 * Change the maximum transfer size of the network device.
6376 int dev_set_mtu(struct net_device *dev, int new_mtu)
6380 if (new_mtu == dev->mtu)
6383 /* MTU must be positive. */
6387 if (!netif_device_present(dev))
6390 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6391 err = notifier_to_errno(err);
6395 orig_mtu = dev->mtu;
6396 err = __dev_set_mtu(dev, new_mtu);
6399 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6400 err = notifier_to_errno(err);
6402 /* setting mtu back and notifying everyone again,
6403 * so that they have a chance to revert changes.
6405 __dev_set_mtu(dev, orig_mtu);
6406 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6411 EXPORT_SYMBOL(dev_set_mtu);
6414 * dev_set_group - Change group this device belongs to
6416 * @new_group: group this device should belong to
6418 void dev_set_group(struct net_device *dev, int new_group)
6420 dev->group = new_group;
6422 EXPORT_SYMBOL(dev_set_group);
6425 * dev_set_mac_address - Change Media Access Control Address
6429 * Change the hardware (MAC) address of the device
6431 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6433 const struct net_device_ops *ops = dev->netdev_ops;
6436 if (!ops->ndo_set_mac_address)
6438 if (sa->sa_family != dev->type)
6440 if (!netif_device_present(dev))
6442 err = ops->ndo_set_mac_address(dev, sa);
6445 dev->addr_assign_type = NET_ADDR_SET;
6446 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6447 add_device_randomness(dev->dev_addr, dev->addr_len);
6450 EXPORT_SYMBOL(dev_set_mac_address);
6453 * dev_change_carrier - Change device carrier
6455 * @new_carrier: new value
6457 * Change device carrier
6459 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6461 const struct net_device_ops *ops = dev->netdev_ops;
6463 if (!ops->ndo_change_carrier)
6465 if (!netif_device_present(dev))
6467 return ops->ndo_change_carrier(dev, new_carrier);
6469 EXPORT_SYMBOL(dev_change_carrier);
6472 * dev_get_phys_port_id - Get device physical port ID
6476 * Get device physical port ID
6478 int dev_get_phys_port_id(struct net_device *dev,
6479 struct netdev_phys_item_id *ppid)
6481 const struct net_device_ops *ops = dev->netdev_ops;
6483 if (!ops->ndo_get_phys_port_id)
6485 return ops->ndo_get_phys_port_id(dev, ppid);
6487 EXPORT_SYMBOL(dev_get_phys_port_id);
6490 * dev_get_phys_port_name - Get device physical port name
6493 * @len: limit of bytes to copy to name
6495 * Get device physical port name
6497 int dev_get_phys_port_name(struct net_device *dev,
6498 char *name, size_t len)
6500 const struct net_device_ops *ops = dev->netdev_ops;
6502 if (!ops->ndo_get_phys_port_name)
6504 return ops->ndo_get_phys_port_name(dev, name, len);
6506 EXPORT_SYMBOL(dev_get_phys_port_name);
6509 * dev_change_proto_down - update protocol port state information
6511 * @proto_down: new value
6513 * This info can be used by switch drivers to set the phys state of the
6516 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6518 const struct net_device_ops *ops = dev->netdev_ops;
6520 if (!ops->ndo_change_proto_down)
6522 if (!netif_device_present(dev))
6524 return ops->ndo_change_proto_down(dev, proto_down);
6526 EXPORT_SYMBOL(dev_change_proto_down);
6529 * dev_new_index - allocate an ifindex
6530 * @net: the applicable net namespace
6532 * Returns a suitable unique value for a new device interface
6533 * number. The caller must hold the rtnl semaphore or the
6534 * dev_base_lock to be sure it remains unique.
6536 static int dev_new_index(struct net *net)
6538 int ifindex = net->ifindex;
6542 if (!__dev_get_by_index(net, ifindex))
6543 return net->ifindex = ifindex;
6547 /* Delayed registration/unregisteration */
6548 static LIST_HEAD(net_todo_list);
6549 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6551 static void net_set_todo(struct net_device *dev)
6553 list_add_tail(&dev->todo_list, &net_todo_list);
6554 dev_net(dev)->dev_unreg_count++;
6557 static void rollback_registered_many(struct list_head *head)
6559 struct net_device *dev, *tmp;
6560 LIST_HEAD(close_head);
6562 BUG_ON(dev_boot_phase);
6565 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6566 /* Some devices call without registering
6567 * for initialization unwind. Remove those
6568 * devices and proceed with the remaining.
6570 if (dev->reg_state == NETREG_UNINITIALIZED) {
6571 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6575 list_del(&dev->unreg_list);
6578 dev->dismantle = true;
6579 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6582 /* If device is running, close it first. */
6583 list_for_each_entry(dev, head, unreg_list)
6584 list_add_tail(&dev->close_list, &close_head);
6585 dev_close_many(&close_head, true);
6587 list_for_each_entry(dev, head, unreg_list) {
6588 /* And unlink it from device chain. */
6589 unlist_netdevice(dev);
6591 dev->reg_state = NETREG_UNREGISTERING;
6592 on_each_cpu(flush_backlog, dev, 1);
6597 list_for_each_entry(dev, head, unreg_list) {
6598 struct sk_buff *skb = NULL;
6600 /* Shutdown queueing discipline. */
6604 /* Notify protocols, that we are about to destroy
6605 this device. They should clean all the things.
6607 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6609 if (!dev->rtnl_link_ops ||
6610 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6611 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6615 * Flush the unicast and multicast chains
6620 if (dev->netdev_ops->ndo_uninit)
6621 dev->netdev_ops->ndo_uninit(dev);
6624 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6626 /* Notifier chain MUST detach us all upper devices. */
6627 WARN_ON(netdev_has_any_upper_dev(dev));
6629 /* Remove entries from kobject tree */
6630 netdev_unregister_kobject(dev);
6632 /* Remove XPS queueing entries */
6633 netif_reset_xps_queues_gt(dev, 0);
6639 list_for_each_entry(dev, head, unreg_list)
6643 static void rollback_registered(struct net_device *dev)
6647 list_add(&dev->unreg_list, &single);
6648 rollback_registered_many(&single);
6652 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6653 struct net_device *upper, netdev_features_t features)
6655 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6656 netdev_features_t feature;
6659 for_each_netdev_feature(&upper_disables, feature_bit) {
6660 feature = __NETIF_F_BIT(feature_bit);
6661 if (!(upper->wanted_features & feature)
6662 && (features & feature)) {
6663 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6664 &feature, upper->name);
6665 features &= ~feature;
6672 static void netdev_sync_lower_features(struct net_device *upper,
6673 struct net_device *lower, netdev_features_t features)
6675 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6676 netdev_features_t feature;
6679 for_each_netdev_feature(&upper_disables, feature_bit) {
6680 feature = __NETIF_F_BIT(feature_bit);
6681 if (!(features & feature) && (lower->features & feature)) {
6682 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6683 &feature, lower->name);
6684 lower->wanted_features &= ~feature;
6685 netdev_update_features(lower);
6687 if (unlikely(lower->features & feature))
6688 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6689 &feature, lower->name);
6694 static netdev_features_t netdev_fix_features(struct net_device *dev,
6695 netdev_features_t features)
6697 /* Fix illegal checksum combinations */
6698 if ((features & NETIF_F_HW_CSUM) &&
6699 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6700 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6701 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6704 /* TSO requires that SG is present as well. */
6705 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6706 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6707 features &= ~NETIF_F_ALL_TSO;
6710 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6711 !(features & NETIF_F_IP_CSUM)) {
6712 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6713 features &= ~NETIF_F_TSO;
6714 features &= ~NETIF_F_TSO_ECN;
6717 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6718 !(features & NETIF_F_IPV6_CSUM)) {
6719 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6720 features &= ~NETIF_F_TSO6;
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);