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>
142 #include <linux/crash_dump.h>
144 #include "net-sysfs.h"
146 /* Instead of increasing this, you should create a hash table. */
147 #define MAX_GRO_SKBS 8
149 /* This should be increased if a protocol with a bigger head is added. */
150 #define GRO_MAX_HEAD (MAX_HEADER + 128)
152 static DEFINE_SPINLOCK(ptype_lock);
153 static DEFINE_SPINLOCK(offload_lock);
154 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
155 struct list_head ptype_all __read_mostly; /* Taps */
156 static struct list_head offload_base __read_mostly;
158 static int netif_rx_internal(struct sk_buff *skb);
159 static int call_netdevice_notifiers_info(unsigned long val,
160 struct net_device *dev,
161 struct netdev_notifier_info *info);
164 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
167 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
169 * Writers must hold the rtnl semaphore while they loop through the
170 * dev_base_head list, and hold dev_base_lock for writing when they do the
171 * actual updates. This allows pure readers to access the list even
172 * while a writer is preparing to update it.
174 * To put it another way, dev_base_lock is held for writing only to
175 * protect against pure readers; the rtnl semaphore provides the
176 * protection against other writers.
178 * See, for example usages, register_netdevice() and
179 * unregister_netdevice(), which must be called with the rtnl
182 DEFINE_RWLOCK(dev_base_lock);
183 EXPORT_SYMBOL(dev_base_lock);
185 /* protects napi_hash addition/deletion and napi_gen_id */
186 static DEFINE_SPINLOCK(napi_hash_lock);
188 static unsigned int napi_gen_id = NR_CPUS;
189 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
191 static seqcount_t devnet_rename_seq;
193 static inline void dev_base_seq_inc(struct net *net)
195 while (++net->dev_base_seq == 0);
198 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
200 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
202 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
205 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
207 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
210 static inline void rps_lock(struct softnet_data *sd)
213 spin_lock(&sd->input_pkt_queue.lock);
217 static inline void rps_unlock(struct softnet_data *sd)
220 spin_unlock(&sd->input_pkt_queue.lock);
224 /* Device list insertion */
225 static void list_netdevice(struct net_device *dev)
227 struct net *net = dev_net(dev);
231 write_lock_bh(&dev_base_lock);
232 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
233 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
234 hlist_add_head_rcu(&dev->index_hlist,
235 dev_index_hash(net, dev->ifindex));
236 write_unlock_bh(&dev_base_lock);
238 dev_base_seq_inc(net);
241 /* Device list removal
242 * caller must respect a RCU grace period before freeing/reusing dev
244 static void unlist_netdevice(struct net_device *dev)
248 /* Unlink dev from the device chain */
249 write_lock_bh(&dev_base_lock);
250 list_del_rcu(&dev->dev_list);
251 hlist_del_rcu(&dev->name_hlist);
252 hlist_del_rcu(&dev->index_hlist);
253 write_unlock_bh(&dev_base_lock);
255 dev_base_seq_inc(dev_net(dev));
262 static RAW_NOTIFIER_HEAD(netdev_chain);
265 * Device drivers call our routines to queue packets here. We empty the
266 * queue in the local softnet handler.
269 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
270 EXPORT_PER_CPU_SYMBOL(softnet_data);
272 #ifdef CONFIG_LOCKDEP
274 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
275 * according to dev->type
277 static const unsigned short netdev_lock_type[] =
278 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
279 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
280 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
281 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
282 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
283 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
284 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
285 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
286 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
287 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
288 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
289 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
290 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
291 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
292 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
294 static const char *const netdev_lock_name[] =
295 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
296 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
297 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
298 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
299 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
300 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
301 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
302 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
303 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
304 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
305 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
306 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
307 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
308 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
309 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
311 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
312 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
318 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
319 if (netdev_lock_type[i] == dev_type)
321 /* the last key is used by default */
322 return ARRAY_SIZE(netdev_lock_type) - 1;
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
326 unsigned short dev_type)
330 i = netdev_lock_pos(dev_type);
331 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
332 netdev_lock_name[i]);
335 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
339 i = netdev_lock_pos(dev->type);
340 lockdep_set_class_and_name(&dev->addr_list_lock,
341 &netdev_addr_lock_key[i],
342 netdev_lock_name[i]);
345 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
346 unsigned short dev_type)
349 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
354 /*******************************************************************************
356 Protocol management and registration routines
358 *******************************************************************************/
361 * Add a protocol ID to the list. Now that the input handler is
362 * smarter we can dispense with all the messy stuff that used to be
365 * BEWARE!!! Protocol handlers, mangling input packets,
366 * MUST BE last in hash buckets and checking protocol handlers
367 * MUST start from promiscuous ptype_all chain in net_bh.
368 * It is true now, do not change it.
369 * Explanation follows: if protocol handler, mangling packet, will
370 * be the first on list, it is not able to sense, that packet
371 * is cloned and should be copied-on-write, so that it will
372 * change it and subsequent readers will get broken packet.
376 static inline struct list_head *ptype_head(const struct packet_type *pt)
378 if (pt->type == htons(ETH_P_ALL))
379 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
381 return pt->dev ? &pt->dev->ptype_specific :
382 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
386 * dev_add_pack - add packet handler
387 * @pt: packet type declaration
389 * Add a protocol handler to the networking stack. The passed &packet_type
390 * is linked into kernel lists and may not be freed until it has been
391 * removed from the kernel lists.
393 * This call does not sleep therefore it can not
394 * guarantee all CPU's that are in middle of receiving packets
395 * will see the new packet type (until the next received packet).
398 void dev_add_pack(struct packet_type *pt)
400 struct list_head *head = ptype_head(pt);
402 spin_lock(&ptype_lock);
403 list_add_rcu(&pt->list, head);
404 spin_unlock(&ptype_lock);
406 EXPORT_SYMBOL(dev_add_pack);
409 * __dev_remove_pack - remove packet handler
410 * @pt: packet type declaration
412 * Remove a protocol handler that was previously added to the kernel
413 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
414 * from the kernel lists and can be freed or reused once this function
417 * The packet type might still be in use by receivers
418 * and must not be freed until after all the CPU's have gone
419 * through a quiescent state.
421 void __dev_remove_pack(struct packet_type *pt)
423 struct list_head *head = ptype_head(pt);
424 struct packet_type *pt1;
426 spin_lock(&ptype_lock);
428 list_for_each_entry(pt1, head, list) {
430 list_del_rcu(&pt->list);
435 pr_warn("dev_remove_pack: %p not found\n", pt);
437 spin_unlock(&ptype_lock);
439 EXPORT_SYMBOL(__dev_remove_pack);
442 * dev_remove_pack - remove packet handler
443 * @pt: packet type declaration
445 * Remove a protocol handler that was previously added to the kernel
446 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
447 * from the kernel lists and can be freed or reused once this function
450 * This call sleeps to guarantee that no CPU is looking at the packet
453 void dev_remove_pack(struct packet_type *pt)
455 __dev_remove_pack(pt);
459 EXPORT_SYMBOL(dev_remove_pack);
463 * dev_add_offload - register offload handlers
464 * @po: protocol offload declaration
466 * Add protocol offload handlers to the networking stack. The passed
467 * &proto_offload is linked into kernel lists and may not be freed until
468 * it has been removed from the kernel lists.
470 * This call does not sleep therefore it can not
471 * guarantee all CPU's that are in middle of receiving packets
472 * will see the new offload handlers (until the next received packet).
474 void dev_add_offload(struct packet_offload *po)
476 struct packet_offload *elem;
478 spin_lock(&offload_lock);
479 list_for_each_entry(elem, &offload_base, list) {
480 if (po->priority < elem->priority)
483 list_add_rcu(&po->list, elem->list.prev);
484 spin_unlock(&offload_lock);
486 EXPORT_SYMBOL(dev_add_offload);
489 * __dev_remove_offload - remove offload handler
490 * @po: packet offload declaration
492 * Remove a protocol offload handler that was previously added to the
493 * kernel offload handlers by dev_add_offload(). The passed &offload_type
494 * is removed from the kernel lists and can be freed or reused once this
497 * The packet type might still be in use by receivers
498 * and must not be freed until after all the CPU's have gone
499 * through a quiescent state.
501 static void __dev_remove_offload(struct packet_offload *po)
503 struct list_head *head = &offload_base;
504 struct packet_offload *po1;
506 spin_lock(&offload_lock);
508 list_for_each_entry(po1, head, list) {
510 list_del_rcu(&po->list);
515 pr_warn("dev_remove_offload: %p not found\n", po);
517 spin_unlock(&offload_lock);
521 * dev_remove_offload - remove packet offload handler
522 * @po: packet offload declaration
524 * Remove a packet offload handler that was previously added to the kernel
525 * offload handlers by dev_add_offload(). The passed &offload_type is
526 * removed from the kernel lists and can be freed or reused once this
529 * This call sleeps to guarantee that no CPU is looking at the packet
532 void dev_remove_offload(struct packet_offload *po)
534 __dev_remove_offload(po);
538 EXPORT_SYMBOL(dev_remove_offload);
540 /******************************************************************************
542 Device Boot-time Settings Routines
544 *******************************************************************************/
546 /* Boot time configuration table */
547 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
550 * netdev_boot_setup_add - add new setup entry
551 * @name: name of the device
552 * @map: configured settings for the device
554 * Adds new setup entry to the dev_boot_setup list. The function
555 * returns 0 on error and 1 on success. This is a generic routine to
558 static int netdev_boot_setup_add(char *name, struct ifmap *map)
560 struct netdev_boot_setup *s;
564 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
565 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
566 memset(s[i].name, 0, sizeof(s[i].name));
567 strlcpy(s[i].name, name, IFNAMSIZ);
568 memcpy(&s[i].map, map, sizeof(s[i].map));
573 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
577 * netdev_boot_setup_check - check boot time settings
578 * @dev: the netdevice
580 * Check boot time settings for the device.
581 * The found settings are set for the device to be used
582 * later in the device probing.
583 * Returns 0 if no settings found, 1 if they are.
585 int netdev_boot_setup_check(struct net_device *dev)
587 struct netdev_boot_setup *s = dev_boot_setup;
590 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
591 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
592 !strcmp(dev->name, s[i].name)) {
593 dev->irq = s[i].map.irq;
594 dev->base_addr = s[i].map.base_addr;
595 dev->mem_start = s[i].map.mem_start;
596 dev->mem_end = s[i].map.mem_end;
602 EXPORT_SYMBOL(netdev_boot_setup_check);
606 * netdev_boot_base - get address from boot time settings
607 * @prefix: prefix for network device
608 * @unit: id for network device
610 * Check boot time settings for the base address of device.
611 * The found settings are set for the device to be used
612 * later in the device probing.
613 * Returns 0 if no settings found.
615 unsigned long netdev_boot_base(const char *prefix, int unit)
617 const struct netdev_boot_setup *s = dev_boot_setup;
621 sprintf(name, "%s%d", prefix, unit);
624 * If device already registered then return base of 1
625 * to indicate not to probe for this interface
627 if (__dev_get_by_name(&init_net, name))
630 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
631 if (!strcmp(name, s[i].name))
632 return s[i].map.base_addr;
637 * Saves at boot time configured settings for any netdevice.
639 int __init netdev_boot_setup(char *str)
644 str = get_options(str, ARRAY_SIZE(ints), ints);
649 memset(&map, 0, sizeof(map));
653 map.base_addr = ints[2];
655 map.mem_start = ints[3];
657 map.mem_end = ints[4];
659 /* Add new entry to the list */
660 return netdev_boot_setup_add(str, &map);
663 __setup("netdev=", netdev_boot_setup);
665 /*******************************************************************************
667 Device Interface Subroutines
669 *******************************************************************************/
672 * dev_get_iflink - get 'iflink' value of a interface
673 * @dev: targeted interface
675 * Indicates the ifindex the interface is linked to.
676 * Physical interfaces have the same 'ifindex' and 'iflink' values.
679 int dev_get_iflink(const struct net_device *dev)
681 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
682 return dev->netdev_ops->ndo_get_iflink(dev);
686 EXPORT_SYMBOL(dev_get_iflink);
689 * dev_fill_metadata_dst - Retrieve tunnel egress information.
690 * @dev: targeted interface
693 * For better visibility of tunnel traffic OVS needs to retrieve
694 * egress tunnel information for a packet. Following API allows
695 * user to get this info.
697 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
699 struct ip_tunnel_info *info;
701 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
704 info = skb_tunnel_info_unclone(skb);
707 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
710 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
712 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
715 * __dev_get_by_name - find a device by its name
716 * @net: the applicable net namespace
717 * @name: name to find
719 * Find an interface by name. Must be called under RTNL semaphore
720 * or @dev_base_lock. If the name is found a pointer to the device
721 * is returned. If the name is not found then %NULL is returned. The
722 * reference counters are not incremented so the caller must be
723 * careful with locks.
726 struct net_device *__dev_get_by_name(struct net *net, const char *name)
728 struct net_device *dev;
729 struct hlist_head *head = dev_name_hash(net, name);
731 hlist_for_each_entry(dev, head, name_hlist)
732 if (!strncmp(dev->name, name, IFNAMSIZ))
737 EXPORT_SYMBOL(__dev_get_by_name);
740 * dev_get_by_name_rcu - find a device by its name
741 * @net: the applicable net namespace
742 * @name: name to find
744 * Find an interface by name.
745 * If the name is found a pointer to the device is returned.
746 * If the name is not found then %NULL is returned.
747 * The reference counters are not incremented so the caller must be
748 * careful with locks. The caller must hold RCU lock.
751 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
753 struct net_device *dev;
754 struct hlist_head *head = dev_name_hash(net, name);
756 hlist_for_each_entry_rcu(dev, head, name_hlist)
757 if (!strncmp(dev->name, name, IFNAMSIZ))
762 EXPORT_SYMBOL(dev_get_by_name_rcu);
765 * dev_get_by_name - find a device by its name
766 * @net: the applicable net namespace
767 * @name: name to find
769 * Find an interface by name. This can be called from any
770 * context and does its own locking. The returned handle has
771 * the usage count incremented and the caller must use dev_put() to
772 * release it when it is no longer needed. %NULL is returned if no
773 * matching device is found.
776 struct net_device *dev_get_by_name(struct net *net, const char *name)
778 struct net_device *dev;
781 dev = dev_get_by_name_rcu(net, name);
787 EXPORT_SYMBOL(dev_get_by_name);
790 * __dev_get_by_index - find a device by its ifindex
791 * @net: the applicable net namespace
792 * @ifindex: index of device
794 * Search for an interface by index. Returns %NULL if the device
795 * is not found or a pointer to the device. The device has not
796 * had its reference counter increased so the caller must be careful
797 * about locking. The caller must hold either the RTNL semaphore
801 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
803 struct net_device *dev;
804 struct hlist_head *head = dev_index_hash(net, ifindex);
806 hlist_for_each_entry(dev, head, index_hlist)
807 if (dev->ifindex == ifindex)
812 EXPORT_SYMBOL(__dev_get_by_index);
815 * dev_get_by_index_rcu - find a device by its ifindex
816 * @net: the applicable net namespace
817 * @ifindex: index of device
819 * Search for an interface by index. Returns %NULL if the device
820 * is not found or a pointer to the device. The device has not
821 * had its reference counter increased so the caller must be careful
822 * about locking. The caller must hold RCU lock.
825 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
827 struct net_device *dev;
828 struct hlist_head *head = dev_index_hash(net, ifindex);
830 hlist_for_each_entry_rcu(dev, head, index_hlist)
831 if (dev->ifindex == ifindex)
836 EXPORT_SYMBOL(dev_get_by_index_rcu);
840 * dev_get_by_index - find a device by its ifindex
841 * @net: the applicable net namespace
842 * @ifindex: index of device
844 * Search for an interface by index. Returns NULL if the device
845 * is not found or a pointer to the device. The device returned has
846 * had a reference added and the pointer is safe until the user calls
847 * dev_put to indicate they have finished with it.
850 struct net_device *dev_get_by_index(struct net *net, int ifindex)
852 struct net_device *dev;
855 dev = dev_get_by_index_rcu(net, ifindex);
861 EXPORT_SYMBOL(dev_get_by_index);
864 * netdev_get_name - get a netdevice name, knowing its ifindex.
865 * @net: network namespace
866 * @name: a pointer to the buffer where the name will be stored.
867 * @ifindex: the ifindex of the interface to get the name from.
869 * The use of raw_seqcount_begin() and cond_resched() before
870 * retrying is required as we want to give the writers a chance
871 * to complete when CONFIG_PREEMPT is not set.
873 int netdev_get_name(struct net *net, char *name, int ifindex)
875 struct net_device *dev;
879 seq = raw_seqcount_begin(&devnet_rename_seq);
881 dev = dev_get_by_index_rcu(net, ifindex);
887 strcpy(name, dev->name);
889 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
898 * dev_getbyhwaddr_rcu - find a device by its hardware address
899 * @net: the applicable net namespace
900 * @type: media type of device
901 * @ha: hardware address
903 * Search for an interface by MAC address. Returns NULL if the device
904 * is not found or a pointer to the device.
905 * The caller must hold RCU or RTNL.
906 * The returned device has not had its ref count increased
907 * and the caller must therefore be careful about locking
911 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
914 struct net_device *dev;
916 for_each_netdev_rcu(net, dev)
917 if (dev->type == type &&
918 !memcmp(dev->dev_addr, ha, dev->addr_len))
923 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
925 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
927 struct net_device *dev;
930 for_each_netdev(net, dev)
931 if (dev->type == type)
936 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
938 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
940 struct net_device *dev, *ret = NULL;
943 for_each_netdev_rcu(net, dev)
944 if (dev->type == type) {
952 EXPORT_SYMBOL(dev_getfirstbyhwtype);
955 * __dev_get_by_flags - find any device with given flags
956 * @net: the applicable net namespace
957 * @if_flags: IFF_* values
958 * @mask: bitmask of bits in if_flags to check
960 * Search for any interface with the given flags. Returns NULL if a device
961 * is not found or a pointer to the device. Must be called inside
962 * rtnl_lock(), and result refcount is unchanged.
965 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
968 struct net_device *dev, *ret;
973 for_each_netdev(net, dev) {
974 if (((dev->flags ^ if_flags) & mask) == 0) {
981 EXPORT_SYMBOL(__dev_get_by_flags);
984 * dev_valid_name - check if name is okay for network device
987 * Network device names need to be valid file names to
988 * to allow sysfs to work. We also disallow any kind of
991 bool dev_valid_name(const char *name)
995 if (strlen(name) >= IFNAMSIZ)
997 if (!strcmp(name, ".") || !strcmp(name, ".."))
1001 if (*name == '/' || *name == ':' || isspace(*name))
1007 EXPORT_SYMBOL(dev_valid_name);
1010 * __dev_alloc_name - allocate a name for a device
1011 * @net: network namespace to allocate the device name in
1012 * @name: name format string
1013 * @buf: scratch buffer and result name string
1015 * Passed a format string - eg "lt%d" it will try and find a suitable
1016 * id. It scans list of devices to build up a free map, then chooses
1017 * the first empty slot. The caller must hold the dev_base or rtnl lock
1018 * while allocating the name and adding the device in order to avoid
1020 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1021 * Returns the number of the unit assigned or a negative errno code.
1024 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1028 const int max_netdevices = 8*PAGE_SIZE;
1029 unsigned long *inuse;
1030 struct net_device *d;
1032 p = strnchr(name, IFNAMSIZ-1, '%');
1035 * Verify the string as this thing may have come from
1036 * the user. There must be either one "%d" and no other "%"
1039 if (p[1] != 'd' || strchr(p + 2, '%'))
1042 /* Use one page as a bit array of possible slots */
1043 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1047 for_each_netdev(net, d) {
1048 if (!sscanf(d->name, name, &i))
1050 if (i < 0 || i >= max_netdevices)
1053 /* avoid cases where sscanf is not exact inverse of printf */
1054 snprintf(buf, IFNAMSIZ, name, i);
1055 if (!strncmp(buf, d->name, IFNAMSIZ))
1059 i = find_first_zero_bit(inuse, max_netdevices);
1060 free_page((unsigned long) inuse);
1064 snprintf(buf, IFNAMSIZ, name, i);
1065 if (!__dev_get_by_name(net, buf))
1068 /* It is possible to run out of possible slots
1069 * when the name is long and there isn't enough space left
1070 * for the digits, or if all bits are used.
1076 * dev_alloc_name - allocate a name for a device
1078 * @name: name format string
1080 * Passed a format string - eg "lt%d" it will try and find a suitable
1081 * id. It scans list of devices to build up a free map, then chooses
1082 * the first empty slot. The caller must hold the dev_base or rtnl lock
1083 * while allocating the name and adding the device in order to avoid
1085 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1086 * Returns the number of the unit assigned or a negative errno code.
1089 int dev_alloc_name(struct net_device *dev, const char *name)
1095 BUG_ON(!dev_net(dev));
1097 ret = __dev_alloc_name(net, name, buf);
1099 strlcpy(dev->name, buf, IFNAMSIZ);
1102 EXPORT_SYMBOL(dev_alloc_name);
1104 static int dev_alloc_name_ns(struct net *net,
1105 struct net_device *dev,
1111 ret = __dev_alloc_name(net, name, buf);
1113 strlcpy(dev->name, buf, IFNAMSIZ);
1117 static int dev_get_valid_name(struct net *net,
1118 struct net_device *dev,
1123 if (!dev_valid_name(name))
1126 if (strchr(name, '%'))
1127 return dev_alloc_name_ns(net, dev, name);
1128 else if (__dev_get_by_name(net, name))
1130 else if (dev->name != name)
1131 strlcpy(dev->name, name, IFNAMSIZ);
1137 * dev_change_name - change name of a device
1139 * @newname: name (or format string) must be at least IFNAMSIZ
1141 * Change name of a device, can pass format strings "eth%d".
1144 int dev_change_name(struct net_device *dev, const char *newname)
1146 unsigned char old_assign_type;
1147 char oldname[IFNAMSIZ];
1153 BUG_ON(!dev_net(dev));
1156 if (dev->flags & IFF_UP)
1159 write_seqcount_begin(&devnet_rename_seq);
1161 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1162 write_seqcount_end(&devnet_rename_seq);
1166 memcpy(oldname, dev->name, IFNAMSIZ);
1168 err = dev_get_valid_name(net, dev, newname);
1170 write_seqcount_end(&devnet_rename_seq);
1174 if (oldname[0] && !strchr(oldname, '%'))
1175 netdev_info(dev, "renamed from %s\n", oldname);
1177 old_assign_type = dev->name_assign_type;
1178 dev->name_assign_type = NET_NAME_RENAMED;
1181 ret = device_rename(&dev->dev, dev->name);
1183 memcpy(dev->name, oldname, IFNAMSIZ);
1184 dev->name_assign_type = old_assign_type;
1185 write_seqcount_end(&devnet_rename_seq);
1189 write_seqcount_end(&devnet_rename_seq);
1191 netdev_adjacent_rename_links(dev, oldname);
1193 write_lock_bh(&dev_base_lock);
1194 hlist_del_rcu(&dev->name_hlist);
1195 write_unlock_bh(&dev_base_lock);
1199 write_lock_bh(&dev_base_lock);
1200 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1201 write_unlock_bh(&dev_base_lock);
1203 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1204 ret = notifier_to_errno(ret);
1207 /* err >= 0 after dev_alloc_name() or stores the first errno */
1210 write_seqcount_begin(&devnet_rename_seq);
1211 memcpy(dev->name, oldname, IFNAMSIZ);
1212 memcpy(oldname, newname, IFNAMSIZ);
1213 dev->name_assign_type = old_assign_type;
1214 old_assign_type = NET_NAME_RENAMED;
1217 pr_err("%s: name change rollback failed: %d\n",
1226 * dev_set_alias - change ifalias of a device
1228 * @alias: name up to IFALIASZ
1229 * @len: limit of bytes to copy from info
1231 * Set ifalias for a device,
1233 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1239 if (len >= IFALIASZ)
1243 kfree(dev->ifalias);
1244 dev->ifalias = NULL;
1248 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1251 dev->ifalias = new_ifalias;
1253 strlcpy(dev->ifalias, alias, len+1);
1259 * netdev_features_change - device changes features
1260 * @dev: device to cause notification
1262 * Called to indicate a device has changed features.
1264 void netdev_features_change(struct net_device *dev)
1266 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1268 EXPORT_SYMBOL(netdev_features_change);
1271 * netdev_state_change - device changes state
1272 * @dev: device to cause notification
1274 * Called to indicate a device has changed state. This function calls
1275 * the notifier chains for netdev_chain and sends a NEWLINK message
1276 * to the routing socket.
1278 void netdev_state_change(struct net_device *dev)
1280 if (dev->flags & IFF_UP) {
1281 struct netdev_notifier_change_info change_info;
1283 change_info.flags_changed = 0;
1284 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1286 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1289 EXPORT_SYMBOL(netdev_state_change);
1292 * netdev_notify_peers - notify network peers about existence of @dev
1293 * @dev: network device
1295 * Generate traffic such that interested network peers are aware of
1296 * @dev, such as by generating a gratuitous ARP. This may be used when
1297 * a device wants to inform the rest of the network about some sort of
1298 * reconfiguration such as a failover event or virtual machine
1301 void netdev_notify_peers(struct net_device *dev)
1304 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1307 EXPORT_SYMBOL(netdev_notify_peers);
1309 static int __dev_open(struct net_device *dev)
1311 const struct net_device_ops *ops = dev->netdev_ops;
1316 if (!netif_device_present(dev))
1319 /* Block netpoll from trying to do any rx path servicing.
1320 * If we don't do this there is a chance ndo_poll_controller
1321 * or ndo_poll may be running while we open the device
1323 netpoll_poll_disable(dev);
1325 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1326 ret = notifier_to_errno(ret);
1330 set_bit(__LINK_STATE_START, &dev->state);
1332 if (ops->ndo_validate_addr)
1333 ret = ops->ndo_validate_addr(dev);
1335 if (!ret && ops->ndo_open)
1336 ret = ops->ndo_open(dev);
1338 netpoll_poll_enable(dev);
1341 clear_bit(__LINK_STATE_START, &dev->state);
1343 dev->flags |= IFF_UP;
1344 dev_set_rx_mode(dev);
1346 add_device_randomness(dev->dev_addr, dev->addr_len);
1353 * dev_open - prepare an interface for use.
1354 * @dev: device to open
1356 * Takes a device from down to up state. The device's private open
1357 * function is invoked and then the multicast lists are loaded. Finally
1358 * the device is moved into the up state and a %NETDEV_UP message is
1359 * sent to the netdev notifier chain.
1361 * Calling this function on an active interface is a nop. On a failure
1362 * a negative errno code is returned.
1364 int dev_open(struct net_device *dev)
1368 if (dev->flags & IFF_UP)
1371 ret = __dev_open(dev);
1375 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1376 call_netdevice_notifiers(NETDEV_UP, dev);
1380 EXPORT_SYMBOL(dev_open);
1382 static int __dev_close_many(struct list_head *head)
1384 struct net_device *dev;
1389 list_for_each_entry(dev, head, close_list) {
1390 /* Temporarily disable netpoll until the interface is down */
1391 netpoll_poll_disable(dev);
1393 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1395 clear_bit(__LINK_STATE_START, &dev->state);
1397 /* Synchronize to scheduled poll. We cannot touch poll list, it
1398 * can be even on different cpu. So just clear netif_running().
1400 * dev->stop() will invoke napi_disable() on all of it's
1401 * napi_struct instances on this device.
1403 smp_mb__after_atomic(); /* Commit netif_running(). */
1406 dev_deactivate_many(head);
1408 list_for_each_entry(dev, head, close_list) {
1409 const struct net_device_ops *ops = dev->netdev_ops;
1412 * Call the device specific close. This cannot fail.
1413 * Only if device is UP
1415 * We allow it to be called even after a DETACH hot-plug
1421 dev->flags &= ~IFF_UP;
1422 netpoll_poll_enable(dev);
1428 static int __dev_close(struct net_device *dev)
1433 list_add(&dev->close_list, &single);
1434 retval = __dev_close_many(&single);
1440 int dev_close_many(struct list_head *head, bool unlink)
1442 struct net_device *dev, *tmp;
1444 /* Remove the devices that don't need to be closed */
1445 list_for_each_entry_safe(dev, tmp, head, close_list)
1446 if (!(dev->flags & IFF_UP))
1447 list_del_init(&dev->close_list);
1449 __dev_close_many(head);
1451 list_for_each_entry_safe(dev, tmp, head, close_list) {
1452 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1453 call_netdevice_notifiers(NETDEV_DOWN, dev);
1455 list_del_init(&dev->close_list);
1460 EXPORT_SYMBOL(dev_close_many);
1463 * dev_close - shutdown an interface.
1464 * @dev: device to shutdown
1466 * This function moves an active device into down state. A
1467 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1468 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1471 int dev_close(struct net_device *dev)
1473 if (dev->flags & IFF_UP) {
1476 list_add(&dev->close_list, &single);
1477 dev_close_many(&single, true);
1482 EXPORT_SYMBOL(dev_close);
1486 * dev_disable_lro - disable Large Receive Offload on a device
1489 * Disable Large Receive Offload (LRO) on a net device. Must be
1490 * called under RTNL. This is needed if received packets may be
1491 * forwarded to another interface.
1493 void dev_disable_lro(struct net_device *dev)
1495 struct net_device *lower_dev;
1496 struct list_head *iter;
1498 dev->wanted_features &= ~NETIF_F_LRO;
1499 netdev_update_features(dev);
1501 if (unlikely(dev->features & NETIF_F_LRO))
1502 netdev_WARN(dev, "failed to disable LRO!\n");
1504 netdev_for_each_lower_dev(dev, lower_dev, iter)
1505 dev_disable_lro(lower_dev);
1507 EXPORT_SYMBOL(dev_disable_lro);
1509 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1510 struct net_device *dev)
1512 struct netdev_notifier_info info;
1514 netdev_notifier_info_init(&info, dev);
1515 return nb->notifier_call(nb, val, &info);
1518 static int dev_boot_phase = 1;
1521 * register_netdevice_notifier - register a network notifier block
1524 * Register a notifier to be called when network device events occur.
1525 * The notifier passed is linked into the kernel structures and must
1526 * not be reused until it has been unregistered. A negative errno code
1527 * is returned on a failure.
1529 * When registered all registration and up events are replayed
1530 * to the new notifier to allow device to have a race free
1531 * view of the network device list.
1534 int register_netdevice_notifier(struct notifier_block *nb)
1536 struct net_device *dev;
1537 struct net_device *last;
1542 err = raw_notifier_chain_register(&netdev_chain, nb);
1548 for_each_netdev(net, dev) {
1549 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1550 err = notifier_to_errno(err);
1554 if (!(dev->flags & IFF_UP))
1557 call_netdevice_notifier(nb, NETDEV_UP, dev);
1568 for_each_netdev(net, dev) {
1572 if (dev->flags & IFF_UP) {
1573 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1575 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1577 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1582 raw_notifier_chain_unregister(&netdev_chain, nb);
1585 EXPORT_SYMBOL(register_netdevice_notifier);
1588 * unregister_netdevice_notifier - unregister a network notifier block
1591 * Unregister a notifier previously registered by
1592 * register_netdevice_notifier(). The notifier is unlinked into the
1593 * kernel structures and may then be reused. A negative errno code
1594 * is returned on a failure.
1596 * After unregistering unregister and down device events are synthesized
1597 * for all devices on the device list to the removed notifier to remove
1598 * the need for special case cleanup code.
1601 int unregister_netdevice_notifier(struct notifier_block *nb)
1603 struct net_device *dev;
1608 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1613 for_each_netdev(net, dev) {
1614 if (dev->flags & IFF_UP) {
1615 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1617 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1619 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1626 EXPORT_SYMBOL(unregister_netdevice_notifier);
1629 * call_netdevice_notifiers_info - call all network notifier blocks
1630 * @val: value passed unmodified to notifier function
1631 * @dev: net_device pointer passed unmodified to notifier function
1632 * @info: notifier information data
1634 * Call all network notifier blocks. Parameters and return value
1635 * are as for raw_notifier_call_chain().
1638 static int call_netdevice_notifiers_info(unsigned long val,
1639 struct net_device *dev,
1640 struct netdev_notifier_info *info)
1643 netdev_notifier_info_init(info, dev);
1644 return raw_notifier_call_chain(&netdev_chain, val, info);
1648 * call_netdevice_notifiers - call all network notifier blocks
1649 * @val: value passed unmodified to notifier function
1650 * @dev: net_device pointer passed unmodified to notifier function
1652 * Call all network notifier blocks. Parameters and return value
1653 * are as for raw_notifier_call_chain().
1656 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1658 struct netdev_notifier_info info;
1660 return call_netdevice_notifiers_info(val, dev, &info);
1662 EXPORT_SYMBOL(call_netdevice_notifiers);
1664 #ifdef CONFIG_NET_INGRESS
1665 static struct static_key ingress_needed __read_mostly;
1667 void net_inc_ingress_queue(void)
1669 static_key_slow_inc(&ingress_needed);
1671 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1673 void net_dec_ingress_queue(void)
1675 static_key_slow_dec(&ingress_needed);
1677 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1680 #ifdef CONFIG_NET_EGRESS
1681 static struct static_key egress_needed __read_mostly;
1683 void net_inc_egress_queue(void)
1685 static_key_slow_inc(&egress_needed);
1687 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1689 void net_dec_egress_queue(void)
1691 static_key_slow_dec(&egress_needed);
1693 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1696 static struct static_key netstamp_needed __read_mostly;
1697 #ifdef HAVE_JUMP_LABEL
1698 /* We are not allowed to call static_key_slow_dec() from irq context
1699 * If net_disable_timestamp() is called from irq context, defer the
1700 * static_key_slow_dec() calls.
1702 static atomic_t netstamp_needed_deferred;
1705 void net_enable_timestamp(void)
1707 #ifdef HAVE_JUMP_LABEL
1708 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1712 static_key_slow_dec(&netstamp_needed);
1716 static_key_slow_inc(&netstamp_needed);
1718 EXPORT_SYMBOL(net_enable_timestamp);
1720 void net_disable_timestamp(void)
1722 #ifdef HAVE_JUMP_LABEL
1723 if (in_interrupt()) {
1724 atomic_inc(&netstamp_needed_deferred);
1728 static_key_slow_dec(&netstamp_needed);
1730 EXPORT_SYMBOL(net_disable_timestamp);
1732 static inline void net_timestamp_set(struct sk_buff *skb)
1734 skb->tstamp.tv64 = 0;
1735 if (static_key_false(&netstamp_needed))
1736 __net_timestamp(skb);
1739 #define net_timestamp_check(COND, SKB) \
1740 if (static_key_false(&netstamp_needed)) { \
1741 if ((COND) && !(SKB)->tstamp.tv64) \
1742 __net_timestamp(SKB); \
1745 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1749 if (!(dev->flags & IFF_UP))
1752 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1753 if (skb->len <= len)
1756 /* if TSO is enabled, we don't care about the length as the packet
1757 * could be forwarded without being segmented before
1759 if (skb_is_gso(skb))
1764 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1766 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1768 if (skb_orphan_frags(skb, GFP_ATOMIC) ||
1769 unlikely(!is_skb_forwardable(dev, skb))) {
1770 atomic_long_inc(&dev->rx_dropped);
1775 skb_scrub_packet(skb, true);
1777 skb->protocol = eth_type_trans(skb, dev);
1778 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1782 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1785 * dev_forward_skb - loopback an skb to another netif
1787 * @dev: destination network device
1788 * @skb: buffer to forward
1791 * NET_RX_SUCCESS (no congestion)
1792 * NET_RX_DROP (packet was dropped, but freed)
1794 * dev_forward_skb can be used for injecting an skb from the
1795 * start_xmit function of one device into the receive queue
1796 * of another device.
1798 * The receiving device may be in another namespace, so
1799 * we have to clear all information in the skb that could
1800 * impact namespace isolation.
1802 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1804 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1806 EXPORT_SYMBOL_GPL(dev_forward_skb);
1808 static inline int deliver_skb(struct sk_buff *skb,
1809 struct packet_type *pt_prev,
1810 struct net_device *orig_dev)
1812 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1814 atomic_inc(&skb->users);
1815 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1818 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1819 struct packet_type **pt,
1820 struct net_device *orig_dev,
1822 struct list_head *ptype_list)
1824 struct packet_type *ptype, *pt_prev = *pt;
1826 list_for_each_entry_rcu(ptype, ptype_list, list) {
1827 if (ptype->type != type)
1830 deliver_skb(skb, pt_prev, orig_dev);
1836 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1838 if (!ptype->af_packet_priv || !skb->sk)
1841 if (ptype->id_match)
1842 return ptype->id_match(ptype, skb->sk);
1843 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1850 * Support routine. Sends outgoing frames to any network
1851 * taps currently in use.
1854 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1856 struct packet_type *ptype;
1857 struct sk_buff *skb2 = NULL;
1858 struct packet_type *pt_prev = NULL;
1859 struct list_head *ptype_list = &ptype_all;
1863 list_for_each_entry_rcu(ptype, ptype_list, list) {
1864 /* Never send packets back to the socket
1865 * they originated from - MvS (miquels@drinkel.ow.org)
1867 if (skb_loop_sk(ptype, skb))
1871 deliver_skb(skb2, pt_prev, skb->dev);
1876 /* need to clone skb, done only once */
1877 skb2 = skb_clone(skb, GFP_ATOMIC);
1881 net_timestamp_set(skb2);
1883 /* skb->nh should be correctly
1884 * set by sender, so that the second statement is
1885 * just protection against buggy protocols.
1887 skb_reset_mac_header(skb2);
1889 if (skb_network_header(skb2) < skb2->data ||
1890 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1891 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1892 ntohs(skb2->protocol),
1894 skb_reset_network_header(skb2);
1897 skb2->transport_header = skb2->network_header;
1898 skb2->pkt_type = PACKET_OUTGOING;
1902 if (ptype_list == &ptype_all) {
1903 ptype_list = &dev->ptype_all;
1908 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1911 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
1914 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1915 * @dev: Network device
1916 * @txq: number of queues available
1918 * If real_num_tx_queues is changed the tc mappings may no longer be
1919 * valid. To resolve this verify the tc mapping remains valid and if
1920 * not NULL the mapping. With no priorities mapping to this
1921 * offset/count pair it will no longer be used. In the worst case TC0
1922 * is invalid nothing can be done so disable priority mappings. If is
1923 * expected that drivers will fix this mapping if they can before
1924 * calling netif_set_real_num_tx_queues.
1926 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1929 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1931 /* If TC0 is invalidated disable TC mapping */
1932 if (tc->offset + tc->count > txq) {
1933 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1938 /* Invalidated prio to tc mappings set to TC0 */
1939 for (i = 1; i < TC_BITMASK + 1; i++) {
1940 int q = netdev_get_prio_tc_map(dev, i);
1942 tc = &dev->tc_to_txq[q];
1943 if (tc->offset + tc->count > txq) {
1944 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1946 netdev_set_prio_tc_map(dev, i, 0);
1952 static DEFINE_MUTEX(xps_map_mutex);
1953 #define xmap_dereference(P) \
1954 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1956 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1959 struct xps_map *map = NULL;
1963 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1965 for (pos = 0; map && pos < map->len; pos++) {
1966 if (map->queues[pos] == index) {
1968 map->queues[pos] = map->queues[--map->len];
1970 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1971 kfree_rcu(map, rcu);
1981 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1983 struct xps_dev_maps *dev_maps;
1985 bool active = false;
1987 mutex_lock(&xps_map_mutex);
1988 dev_maps = xmap_dereference(dev->xps_maps);
1993 for_each_possible_cpu(cpu) {
1994 for (i = index; i < dev->num_tx_queues; i++) {
1995 if (!remove_xps_queue(dev_maps, cpu, i))
1998 if (i == dev->num_tx_queues)
2003 RCU_INIT_POINTER(dev->xps_maps, NULL);
2004 kfree_rcu(dev_maps, rcu);
2007 for (i = index; i < dev->num_tx_queues; i++)
2008 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2012 mutex_unlock(&xps_map_mutex);
2015 static struct xps_map *expand_xps_map(struct xps_map *map,
2018 struct xps_map *new_map;
2019 int alloc_len = XPS_MIN_MAP_ALLOC;
2022 for (pos = 0; map && pos < map->len; pos++) {
2023 if (map->queues[pos] != index)
2028 /* Need to add queue to this CPU's existing map */
2030 if (pos < map->alloc_len)
2033 alloc_len = map->alloc_len * 2;
2036 /* Need to allocate new map to store queue on this CPU's map */
2037 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2042 for (i = 0; i < pos; i++)
2043 new_map->queues[i] = map->queues[i];
2044 new_map->alloc_len = alloc_len;
2050 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2053 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2054 struct xps_map *map, *new_map;
2055 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2056 int cpu, numa_node_id = -2;
2057 bool active = false;
2059 mutex_lock(&xps_map_mutex);
2061 dev_maps = xmap_dereference(dev->xps_maps);
2063 /* allocate memory for queue storage */
2064 for_each_online_cpu(cpu) {
2065 if (!cpumask_test_cpu(cpu, mask))
2069 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2070 if (!new_dev_maps) {
2071 mutex_unlock(&xps_map_mutex);
2075 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2078 map = expand_xps_map(map, cpu, index);
2082 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2086 goto out_no_new_maps;
2088 for_each_possible_cpu(cpu) {
2089 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2090 /* add queue to CPU maps */
2093 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2094 while ((pos < map->len) && (map->queues[pos] != index))
2097 if (pos == map->len)
2098 map->queues[map->len++] = index;
2100 if (numa_node_id == -2)
2101 numa_node_id = cpu_to_node(cpu);
2102 else if (numa_node_id != cpu_to_node(cpu))
2105 } else if (dev_maps) {
2106 /* fill in the new device map from the old device map */
2107 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2108 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2113 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2115 /* Cleanup old maps */
2117 for_each_possible_cpu(cpu) {
2118 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2119 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2120 if (map && map != new_map)
2121 kfree_rcu(map, rcu);
2124 kfree_rcu(dev_maps, rcu);
2127 dev_maps = new_dev_maps;
2131 /* update Tx queue numa node */
2132 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2133 (numa_node_id >= 0) ? numa_node_id :
2139 /* removes queue from unused CPUs */
2140 for_each_possible_cpu(cpu) {
2141 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2144 if (remove_xps_queue(dev_maps, cpu, index))
2148 /* free map if not active */
2150 RCU_INIT_POINTER(dev->xps_maps, NULL);
2151 kfree_rcu(dev_maps, rcu);
2155 mutex_unlock(&xps_map_mutex);
2159 /* remove any maps that we added */
2160 for_each_possible_cpu(cpu) {
2161 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2162 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2164 if (new_map && new_map != map)
2168 mutex_unlock(&xps_map_mutex);
2170 kfree(new_dev_maps);
2173 EXPORT_SYMBOL(netif_set_xps_queue);
2177 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2178 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2180 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2184 if (txq < 1 || txq > dev->num_tx_queues)
2187 if (dev->reg_state == NETREG_REGISTERED ||
2188 dev->reg_state == NETREG_UNREGISTERING) {
2191 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2197 netif_setup_tc(dev, txq);
2199 if (txq < dev->real_num_tx_queues) {
2200 qdisc_reset_all_tx_gt(dev, txq);
2202 netif_reset_xps_queues_gt(dev, txq);
2207 dev->real_num_tx_queues = txq;
2210 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2214 * netif_set_real_num_rx_queues - set actual number of RX queues used
2215 * @dev: Network device
2216 * @rxq: Actual number of RX queues
2218 * This must be called either with the rtnl_lock held or before
2219 * registration of the net device. Returns 0 on success, or a
2220 * negative error code. If called before registration, it always
2223 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2227 if (rxq < 1 || rxq > dev->num_rx_queues)
2230 if (dev->reg_state == NETREG_REGISTERED) {
2233 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2239 dev->real_num_rx_queues = rxq;
2242 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2246 * netif_get_num_default_rss_queues - default number of RSS queues
2248 * This routine should set an upper limit on the number of RSS queues
2249 * used by default by multiqueue devices.
2251 int netif_get_num_default_rss_queues(void)
2253 return is_kdump_kernel() ?
2254 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2256 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2258 static void __netif_reschedule(struct Qdisc *q)
2260 struct softnet_data *sd;
2261 unsigned long flags;
2263 local_irq_save(flags);
2264 sd = this_cpu_ptr(&softnet_data);
2265 q->next_sched = NULL;
2266 *sd->output_queue_tailp = q;
2267 sd->output_queue_tailp = &q->next_sched;
2268 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2269 local_irq_restore(flags);
2272 void __netif_schedule(struct Qdisc *q)
2274 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2275 __netif_reschedule(q);
2277 EXPORT_SYMBOL(__netif_schedule);
2279 struct dev_kfree_skb_cb {
2280 enum skb_free_reason reason;
2283 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2285 return (struct dev_kfree_skb_cb *)skb->cb;
2288 void netif_schedule_queue(struct netdev_queue *txq)
2291 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2292 struct Qdisc *q = rcu_dereference(txq->qdisc);
2294 __netif_schedule(q);
2298 EXPORT_SYMBOL(netif_schedule_queue);
2301 * netif_wake_subqueue - allow sending packets on subqueue
2302 * @dev: network device
2303 * @queue_index: sub queue index
2305 * Resume individual transmit queue of a device with multiple transmit queues.
2307 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2309 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2311 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2315 q = rcu_dereference(txq->qdisc);
2316 __netif_schedule(q);
2320 EXPORT_SYMBOL(netif_wake_subqueue);
2322 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2324 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2328 q = rcu_dereference(dev_queue->qdisc);
2329 __netif_schedule(q);
2333 EXPORT_SYMBOL(netif_tx_wake_queue);
2335 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2337 unsigned long flags;
2339 if (likely(atomic_read(&skb->users) == 1)) {
2341 atomic_set(&skb->users, 0);
2342 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2345 get_kfree_skb_cb(skb)->reason = reason;
2346 local_irq_save(flags);
2347 skb->next = __this_cpu_read(softnet_data.completion_queue);
2348 __this_cpu_write(softnet_data.completion_queue, skb);
2349 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2350 local_irq_restore(flags);
2352 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2354 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2356 if (in_irq() || irqs_disabled())
2357 __dev_kfree_skb_irq(skb, reason);
2361 EXPORT_SYMBOL(__dev_kfree_skb_any);
2365 * netif_device_detach - mark device as removed
2366 * @dev: network device
2368 * Mark device as removed from system and therefore no longer available.
2370 void netif_device_detach(struct net_device *dev)
2372 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2373 netif_running(dev)) {
2374 netif_tx_stop_all_queues(dev);
2377 EXPORT_SYMBOL(netif_device_detach);
2380 * netif_device_attach - mark device as attached
2381 * @dev: network device
2383 * Mark device as attached from system and restart if needed.
2385 void netif_device_attach(struct net_device *dev)
2387 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2388 netif_running(dev)) {
2389 netif_tx_wake_all_queues(dev);
2390 __netdev_watchdog_up(dev);
2393 EXPORT_SYMBOL(netif_device_attach);
2396 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2397 * to be used as a distribution range.
2399 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2400 unsigned int num_tx_queues)
2404 u16 qcount = num_tx_queues;
2406 if (skb_rx_queue_recorded(skb)) {
2407 hash = skb_get_rx_queue(skb);
2408 while (unlikely(hash >= num_tx_queues))
2409 hash -= num_tx_queues;
2414 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2415 qoffset = dev->tc_to_txq[tc].offset;
2416 qcount = dev->tc_to_txq[tc].count;
2419 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2421 EXPORT_SYMBOL(__skb_tx_hash);
2423 static void skb_warn_bad_offload(const struct sk_buff *skb)
2425 static const netdev_features_t null_features;
2426 struct net_device *dev = skb->dev;
2427 const char *name = "";
2429 if (!net_ratelimit())
2433 if (dev->dev.parent)
2434 name = dev_driver_string(dev->dev.parent);
2436 name = netdev_name(dev);
2438 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2439 "gso_type=%d ip_summed=%d\n",
2440 name, dev ? &dev->features : &null_features,
2441 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2442 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2443 skb_shinfo(skb)->gso_type, skb->ip_summed);
2447 * Invalidate hardware checksum when packet is to be mangled, and
2448 * complete checksum manually on outgoing path.
2450 int skb_checksum_help(struct sk_buff *skb)
2453 int ret = 0, offset;
2455 if (skb->ip_summed == CHECKSUM_COMPLETE)
2456 goto out_set_summed;
2458 if (unlikely(skb_shinfo(skb)->gso_size)) {
2459 skb_warn_bad_offload(skb);
2463 /* Before computing a checksum, we should make sure no frag could
2464 * be modified by an external entity : checksum could be wrong.
2466 if (skb_has_shared_frag(skb)) {
2467 ret = __skb_linearize(skb);
2472 offset = skb_checksum_start_offset(skb);
2473 BUG_ON(offset >= skb_headlen(skb));
2474 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2476 offset += skb->csum_offset;
2477 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2479 if (skb_cloned(skb) &&
2480 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2481 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2486 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2488 skb->ip_summed = CHECKSUM_NONE;
2492 EXPORT_SYMBOL(skb_checksum_help);
2494 /* skb_csum_offload_check - Driver helper function to determine if a device
2495 * with limited checksum offload capabilities is able to offload the checksum
2496 * for a given packet.
2499 * skb - sk_buff for the packet in question
2500 * spec - contains the description of what device can offload
2501 * csum_encapped - returns true if the checksum being offloaded is
2502 * encpasulated. That is it is checksum for the transport header
2503 * in the inner headers.
2504 * checksum_help - when set indicates that helper function should
2505 * call skb_checksum_help if offload checks fail
2508 * true: Packet has passed the checksum checks and should be offloadable to
2509 * the device (a driver may still need to check for additional
2510 * restrictions of its device)
2511 * false: Checksum is not offloadable. If checksum_help was set then
2512 * skb_checksum_help was called to resolve checksum for non-GSO
2513 * packets and when IP protocol is not SCTP
2515 bool __skb_csum_offload_chk(struct sk_buff *skb,
2516 const struct skb_csum_offl_spec *spec,
2517 bool *csum_encapped,
2521 struct ipv6hdr *ipv6;
2526 if (skb->protocol == htons(ETH_P_8021Q) ||
2527 skb->protocol == htons(ETH_P_8021AD)) {
2528 if (!spec->vlan_okay)
2532 /* We check whether the checksum refers to a transport layer checksum in
2533 * the outermost header or an encapsulated transport layer checksum that
2534 * corresponds to the inner headers of the skb. If the checksum is for
2535 * something else in the packet we need help.
2537 if (skb_checksum_start_offset(skb) == skb_transport_offset(skb)) {
2538 /* Non-encapsulated checksum */
2539 protocol = eproto_to_ipproto(vlan_get_protocol(skb));
2540 nhdr = skb_network_header(skb);
2541 *csum_encapped = false;
2542 if (spec->no_not_encapped)
2544 } else if (skb->encapsulation && spec->encap_okay &&
2545 skb_checksum_start_offset(skb) ==
2546 skb_inner_transport_offset(skb)) {
2547 /* Encapsulated checksum */
2548 *csum_encapped = true;
2549 switch (skb->inner_protocol_type) {
2550 case ENCAP_TYPE_ETHER:
2551 protocol = eproto_to_ipproto(skb->inner_protocol);
2553 case ENCAP_TYPE_IPPROTO:
2554 protocol = skb->inner_protocol;
2557 nhdr = skb_inner_network_header(skb);
2564 if (!spec->ipv4_okay)
2567 ip_proto = iph->protocol;
2568 if (iph->ihl != 5 && !spec->ip_options_okay)
2572 if (!spec->ipv6_okay)
2574 if (spec->no_encapped_ipv6 && *csum_encapped)
2577 nhdr += sizeof(*ipv6);
2578 ip_proto = ipv6->nexthdr;
2587 if (!spec->tcp_okay ||
2588 skb->csum_offset != offsetof(struct tcphdr, check))
2592 if (!spec->udp_okay ||
2593 skb->csum_offset != offsetof(struct udphdr, check))
2597 if (!spec->sctp_okay ||
2598 skb->csum_offset != offsetof(struct sctphdr, checksum))
2602 case NEXTHDR_ROUTING:
2603 case NEXTHDR_DEST: {
2606 if (protocol != IPPROTO_IPV6 || !spec->ext_hdrs_okay)
2609 ip_proto = opthdr[0];
2610 nhdr += (opthdr[1] + 1) << 3;
2612 goto ip_proto_again;
2618 /* Passed the tests for offloading checksum */
2622 if (csum_help && !skb_shinfo(skb)->gso_size)
2623 skb_checksum_help(skb);
2627 EXPORT_SYMBOL(__skb_csum_offload_chk);
2629 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2631 __be16 type = skb->protocol;
2633 /* Tunnel gso handlers can set protocol to ethernet. */
2634 if (type == htons(ETH_P_TEB)) {
2637 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2640 eth = (struct ethhdr *)skb_mac_header(skb);
2641 type = eth->h_proto;
2644 return __vlan_get_protocol(skb, type, depth);
2648 * skb_mac_gso_segment - mac layer segmentation handler.
2649 * @skb: buffer to segment
2650 * @features: features for the output path (see dev->features)
2652 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2653 netdev_features_t features)
2655 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2656 struct packet_offload *ptype;
2657 int vlan_depth = skb->mac_len;
2658 __be16 type = skb_network_protocol(skb, &vlan_depth);
2660 if (unlikely(!type))
2661 return ERR_PTR(-EINVAL);
2663 __skb_pull(skb, vlan_depth);
2666 list_for_each_entry_rcu(ptype, &offload_base, list) {
2667 if (ptype->type == type && ptype->callbacks.gso_segment) {
2668 segs = ptype->callbacks.gso_segment(skb, features);
2674 __skb_push(skb, skb->data - skb_mac_header(skb));
2678 EXPORT_SYMBOL(skb_mac_gso_segment);
2681 /* openvswitch calls this on rx path, so we need a different check.
2683 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2686 return skb->ip_summed != CHECKSUM_PARTIAL;
2688 return skb->ip_summed == CHECKSUM_NONE;
2692 * __skb_gso_segment - Perform segmentation on skb.
2693 * @skb: buffer to segment
2694 * @features: features for the output path (see dev->features)
2695 * @tx_path: whether it is called in TX path
2697 * This function segments the given skb and returns a list of segments.
2699 * It may return NULL if the skb requires no segmentation. This is
2700 * only possible when GSO is used for verifying header integrity.
2702 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2704 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2705 netdev_features_t features, bool tx_path)
2707 if (unlikely(skb_needs_check(skb, tx_path))) {
2710 skb_warn_bad_offload(skb);
2712 err = skb_cow_head(skb, 0);
2714 return ERR_PTR(err);
2717 /* Only report GSO partial support if it will enable us to
2718 * support segmentation on this frame without needing additional
2721 if (features & NETIF_F_GSO_PARTIAL) {
2722 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
2723 struct net_device *dev = skb->dev;
2725 partial_features |= dev->features & dev->gso_partial_features;
2726 if (!skb_gso_ok(skb, features | partial_features))
2727 features &= ~NETIF_F_GSO_PARTIAL;
2730 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2731 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2733 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2734 SKB_GSO_CB(skb)->encap_level = 0;
2736 skb_reset_mac_header(skb);
2737 skb_reset_mac_len(skb);
2739 return skb_mac_gso_segment(skb, features);
2741 EXPORT_SYMBOL(__skb_gso_segment);
2743 /* Take action when hardware reception checksum errors are detected. */
2745 void netdev_rx_csum_fault(struct net_device *dev)
2747 if (net_ratelimit()) {
2748 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2752 EXPORT_SYMBOL(netdev_rx_csum_fault);
2755 /* Actually, we should eliminate this check as soon as we know, that:
2756 * 1. IOMMU is present and allows to map all the memory.
2757 * 2. No high memory really exists on this machine.
2760 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2762 #ifdef CONFIG_HIGHMEM
2764 if (!(dev->features & NETIF_F_HIGHDMA)) {
2765 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2766 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2767 if (PageHighMem(skb_frag_page(frag)))
2772 if (PCI_DMA_BUS_IS_PHYS) {
2773 struct device *pdev = dev->dev.parent;
2777 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2778 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2779 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2780 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2788 /* If MPLS offload request, verify we are testing hardware MPLS features
2789 * instead of standard features for the netdev.
2791 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2792 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2793 netdev_features_t features,
2796 if (eth_p_mpls(type))
2797 features &= skb->dev->mpls_features;
2802 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2803 netdev_features_t features,
2810 static netdev_features_t harmonize_features(struct sk_buff *skb,
2811 netdev_features_t features)
2816 type = skb_network_protocol(skb, &tmp);
2817 features = net_mpls_features(skb, features, type);
2819 if (skb->ip_summed != CHECKSUM_NONE &&
2820 !can_checksum_protocol(features, type)) {
2821 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2822 } else if (illegal_highdma(skb->dev, skb)) {
2823 features &= ~NETIF_F_SG;
2829 netdev_features_t passthru_features_check(struct sk_buff *skb,
2830 struct net_device *dev,
2831 netdev_features_t features)
2835 EXPORT_SYMBOL(passthru_features_check);
2837 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2838 struct net_device *dev,
2839 netdev_features_t features)
2841 return vlan_features_check(skb, features);
2844 static netdev_features_t gso_features_check(const struct sk_buff *skb,
2845 struct net_device *dev,
2846 netdev_features_t features)
2848 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2850 if (gso_segs > dev->gso_max_segs)
2851 return features & ~NETIF_F_GSO_MASK;
2853 /* Support for GSO partial features requires software
2854 * intervention before we can actually process the packets
2855 * so we need to strip support for any partial features now
2856 * and we can pull them back in after we have partially
2857 * segmented the frame.
2859 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
2860 features &= ~dev->gso_partial_features;
2862 /* Make sure to clear the IPv4 ID mangling feature if the
2863 * IPv4 header has the potential to be fragmented.
2865 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2866 struct iphdr *iph = skb->encapsulation ?
2867 inner_ip_hdr(skb) : ip_hdr(skb);
2869 if (!(iph->frag_off & htons(IP_DF)))
2870 features &= ~NETIF_F_TSO_MANGLEID;
2876 netdev_features_t netif_skb_features(struct sk_buff *skb)
2878 struct net_device *dev = skb->dev;
2879 netdev_features_t features = dev->features;
2881 if (skb_is_gso(skb))
2882 features = gso_features_check(skb, dev, features);
2884 /* If encapsulation offload request, verify we are testing
2885 * hardware encapsulation features instead of standard
2886 * features for the netdev
2888 if (skb->encapsulation)
2889 features &= dev->hw_enc_features;
2891 if (skb_vlan_tagged(skb))
2892 features = netdev_intersect_features(features,
2893 dev->vlan_features |
2894 NETIF_F_HW_VLAN_CTAG_TX |
2895 NETIF_F_HW_VLAN_STAG_TX);
2897 if (dev->netdev_ops->ndo_features_check)
2898 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2901 features &= dflt_features_check(skb, dev, features);
2903 return harmonize_features(skb, features);
2905 EXPORT_SYMBOL(netif_skb_features);
2907 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2908 struct netdev_queue *txq, bool more)
2913 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2914 dev_queue_xmit_nit(skb, dev);
2917 trace_net_dev_start_xmit(skb, dev);
2918 rc = netdev_start_xmit(skb, dev, txq, more);
2919 trace_net_dev_xmit(skb, rc, dev, len);
2924 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2925 struct netdev_queue *txq, int *ret)
2927 struct sk_buff *skb = first;
2928 int rc = NETDEV_TX_OK;
2931 struct sk_buff *next = skb->next;
2934 rc = xmit_one(skb, dev, txq, next != NULL);
2935 if (unlikely(!dev_xmit_complete(rc))) {
2941 if (netif_xmit_stopped(txq) && skb) {
2942 rc = NETDEV_TX_BUSY;
2952 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2953 netdev_features_t features)
2955 if (skb_vlan_tag_present(skb) &&
2956 !vlan_hw_offload_capable(features, skb->vlan_proto))
2957 skb = __vlan_hwaccel_push_inside(skb);
2961 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2963 netdev_features_t features;
2965 features = netif_skb_features(skb);
2966 skb = validate_xmit_vlan(skb, features);
2970 if (netif_needs_gso(skb, features)) {
2971 struct sk_buff *segs;
2973 segs = skb_gso_segment(skb, features);
2981 if (skb_needs_linearize(skb, features) &&
2982 __skb_linearize(skb))
2985 /* If packet is not checksummed and device does not
2986 * support checksumming for this protocol, complete
2987 * checksumming here.
2989 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2990 if (skb->encapsulation)
2991 skb_set_inner_transport_header(skb,
2992 skb_checksum_start_offset(skb));
2994 skb_set_transport_header(skb,
2995 skb_checksum_start_offset(skb));
2996 if (!(features & NETIF_F_CSUM_MASK) &&
2997 skb_checksum_help(skb))
3007 atomic_long_inc(&dev->tx_dropped);
3011 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
3013 struct sk_buff *next, *head = NULL, *tail;
3015 for (; skb != NULL; skb = next) {
3019 /* in case skb wont be segmented, point to itself */
3022 skb = validate_xmit_skb(skb, dev);
3030 /* If skb was segmented, skb->prev points to
3031 * the last segment. If not, it still contains skb.
3038 static void qdisc_pkt_len_init(struct sk_buff *skb)
3040 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3042 qdisc_skb_cb(skb)->pkt_len = skb->len;
3044 /* To get more precise estimation of bytes sent on wire,
3045 * we add to pkt_len the headers size of all segments
3047 if (shinfo->gso_size) {
3048 unsigned int hdr_len;
3049 u16 gso_segs = shinfo->gso_segs;
3051 /* mac layer + network layer */
3052 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3054 /* + transport layer */
3055 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
3056 hdr_len += tcp_hdrlen(skb);
3058 hdr_len += sizeof(struct udphdr);
3060 if (shinfo->gso_type & SKB_GSO_DODGY)
3061 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3064 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3068 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3069 struct net_device *dev,
3070 struct netdev_queue *txq)
3072 spinlock_t *root_lock = qdisc_lock(q);
3073 struct sk_buff *to_free = NULL;
3077 qdisc_calculate_pkt_len(skb, q);
3079 * Heuristic to force contended enqueues to serialize on a
3080 * separate lock before trying to get qdisc main lock.
3081 * This permits qdisc->running owner to get the lock more
3082 * often and dequeue packets faster.
3084 contended = qdisc_is_running(q);
3085 if (unlikely(contended))
3086 spin_lock(&q->busylock);
3088 spin_lock(root_lock);
3089 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3090 __qdisc_drop(skb, &to_free);
3092 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3093 qdisc_run_begin(q)) {
3095 * This is a work-conserving queue; there are no old skbs
3096 * waiting to be sent out; and the qdisc is not running -
3097 * xmit the skb directly.
3100 qdisc_bstats_update(q, skb);
3102 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3103 if (unlikely(contended)) {
3104 spin_unlock(&q->busylock);
3111 rc = NET_XMIT_SUCCESS;
3113 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3114 if (qdisc_run_begin(q)) {
3115 if (unlikely(contended)) {
3116 spin_unlock(&q->busylock);
3122 spin_unlock(root_lock);
3123 if (unlikely(to_free))
3124 kfree_skb_list(to_free);
3125 if (unlikely(contended))
3126 spin_unlock(&q->busylock);
3130 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3131 static void skb_update_prio(struct sk_buff *skb)
3133 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
3135 if (!skb->priority && skb->sk && map) {
3136 unsigned int prioidx =
3137 sock_cgroup_prioidx(&skb->sk->sk_cgrp_data);
3139 if (prioidx < map->priomap_len)
3140 skb->priority = map->priomap[prioidx];
3144 #define skb_update_prio(skb)
3147 DEFINE_PER_CPU(int, xmit_recursion);
3148 EXPORT_SYMBOL(xmit_recursion);
3151 * dev_loopback_xmit - loop back @skb
3152 * @net: network namespace this loopback is happening in
3153 * @sk: sk needed to be a netfilter okfn
3154 * @skb: buffer to transmit
3156 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3158 skb_reset_mac_header(skb);
3159 __skb_pull(skb, skb_network_offset(skb));
3160 skb->pkt_type = PACKET_LOOPBACK;
3161 skb->ip_summed = CHECKSUM_UNNECESSARY;
3162 WARN_ON(!skb_dst(skb));
3167 EXPORT_SYMBOL(dev_loopback_xmit);
3169 #ifdef CONFIG_NET_EGRESS
3170 static struct sk_buff *
3171 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3173 struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list);
3174 struct tcf_result cl_res;
3179 /* skb->tc_verd and qdisc_skb_cb(skb)->pkt_len were already set
3180 * earlier by the caller.
3182 qdisc_bstats_cpu_update(cl->q, skb);
3184 switch (tc_classify(skb, cl, &cl_res, false)) {
3186 case TC_ACT_RECLASSIFY:
3187 skb->tc_index = TC_H_MIN(cl_res.classid);
3190 qdisc_qstats_cpu_drop(cl->q);
3191 *ret = NET_XMIT_DROP;
3196 *ret = NET_XMIT_SUCCESS;
3199 case TC_ACT_REDIRECT:
3200 /* No need to push/pop skb's mac_header here on egress! */
3201 skb_do_redirect(skb);
3202 *ret = NET_XMIT_SUCCESS;
3210 #endif /* CONFIG_NET_EGRESS */
3212 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3215 struct xps_dev_maps *dev_maps;
3216 struct xps_map *map;
3217 int queue_index = -1;
3220 dev_maps = rcu_dereference(dev->xps_maps);
3222 map = rcu_dereference(
3223 dev_maps->cpu_map[skb->sender_cpu - 1]);
3226 queue_index = map->queues[0];
3228 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3230 if (unlikely(queue_index >= dev->real_num_tx_queues))
3242 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3244 struct sock *sk = skb->sk;
3245 int queue_index = sk_tx_queue_get(sk);
3247 if (queue_index < 0 || skb->ooo_okay ||
3248 queue_index >= dev->real_num_tx_queues) {
3249 int new_index = get_xps_queue(dev, skb);
3251 new_index = skb_tx_hash(dev, skb);
3253 if (queue_index != new_index && sk &&
3255 rcu_access_pointer(sk->sk_dst_cache))
3256 sk_tx_queue_set(sk, new_index);
3258 queue_index = new_index;
3264 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3265 struct sk_buff *skb,
3268 int queue_index = 0;
3271 u32 sender_cpu = skb->sender_cpu - 1;
3273 if (sender_cpu >= (u32)NR_CPUS)
3274 skb->sender_cpu = raw_smp_processor_id() + 1;
3277 if (dev->real_num_tx_queues != 1) {
3278 const struct net_device_ops *ops = dev->netdev_ops;
3279 if (ops->ndo_select_queue)
3280 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3283 queue_index = __netdev_pick_tx(dev, skb);
3286 queue_index = netdev_cap_txqueue(dev, queue_index);
3289 skb_set_queue_mapping(skb, queue_index);
3290 return netdev_get_tx_queue(dev, queue_index);
3294 * __dev_queue_xmit - transmit a buffer
3295 * @skb: buffer to transmit
3296 * @accel_priv: private data used for L2 forwarding offload
3298 * Queue a buffer for transmission to a network device. The caller must
3299 * have set the device and priority and built the buffer before calling
3300 * this function. The function can be called from an interrupt.
3302 * A negative errno code is returned on a failure. A success does not
3303 * guarantee the frame will be transmitted as it may be dropped due
3304 * to congestion or traffic shaping.
3306 * -----------------------------------------------------------------------------------
3307 * I notice this method can also return errors from the queue disciplines,
3308 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3311 * Regardless of the return value, the skb is consumed, so it is currently
3312 * difficult to retry a send to this method. (You can bump the ref count
3313 * before sending to hold a reference for retry if you are careful.)
3315 * When calling this method, interrupts MUST be enabled. This is because
3316 * the BH enable code must have IRQs enabled so that it will not deadlock.
3319 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3321 struct net_device *dev = skb->dev;
3322 struct netdev_queue *txq;
3326 skb_reset_mac_header(skb);
3328 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3329 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3331 /* Disable soft irqs for various locks below. Also
3332 * stops preemption for RCU.
3336 skb_update_prio(skb);
3338 qdisc_pkt_len_init(skb);
3339 #ifdef CONFIG_NET_CLS_ACT
3340 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3341 # ifdef CONFIG_NET_EGRESS
3342 if (static_key_false(&egress_needed)) {
3343 skb = sch_handle_egress(skb, &rc, dev);
3349 /* If device/qdisc don't need skb->dst, release it right now while
3350 * its hot in this cpu cache.
3352 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3357 #ifdef CONFIG_NET_SWITCHDEV
3358 /* Don't forward if offload device already forwarded */
3359 if (skb->offload_fwd_mark &&
3360 skb->offload_fwd_mark == dev->offload_fwd_mark) {
3362 rc = NET_XMIT_SUCCESS;
3367 txq = netdev_pick_tx(dev, skb, accel_priv);
3368 q = rcu_dereference_bh(txq->qdisc);
3370 trace_net_dev_queue(skb);
3372 rc = __dev_xmit_skb(skb, q, dev, txq);
3376 /* The device has no queue. Common case for software devices:
3377 loopback, all the sorts of tunnels...
3379 Really, it is unlikely that netif_tx_lock protection is necessary
3380 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3382 However, it is possible, that they rely on protection
3385 Check this and shot the lock. It is not prone from deadlocks.
3386 Either shot noqueue qdisc, it is even simpler 8)
3388 if (dev->flags & IFF_UP) {
3389 int cpu = smp_processor_id(); /* ok because BHs are off */
3391 if (txq->xmit_lock_owner != cpu) {
3392 if (unlikely(__this_cpu_read(xmit_recursion) >
3393 XMIT_RECURSION_LIMIT))
3394 goto recursion_alert;
3396 skb = validate_xmit_skb(skb, dev);
3400 HARD_TX_LOCK(dev, txq, cpu);
3402 if (!netif_xmit_stopped(txq)) {
3403 __this_cpu_inc(xmit_recursion);
3404 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3405 __this_cpu_dec(xmit_recursion);
3406 if (dev_xmit_complete(rc)) {
3407 HARD_TX_UNLOCK(dev, txq);
3411 HARD_TX_UNLOCK(dev, txq);
3412 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3415 /* Recursion is detected! It is possible,
3419 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3425 rcu_read_unlock_bh();
3427 atomic_long_inc(&dev->tx_dropped);
3428 kfree_skb_list(skb);
3431 rcu_read_unlock_bh();
3435 int dev_queue_xmit(struct sk_buff *skb)
3437 return __dev_queue_xmit(skb, NULL);
3439 EXPORT_SYMBOL(dev_queue_xmit);
3441 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3443 return __dev_queue_xmit(skb, accel_priv);
3445 EXPORT_SYMBOL(dev_queue_xmit_accel);
3448 /*=======================================================================
3450 =======================================================================*/
3452 int netdev_max_backlog __read_mostly = 1000;
3453 EXPORT_SYMBOL(netdev_max_backlog);
3455 int netdev_tstamp_prequeue __read_mostly = 1;
3456 int netdev_budget __read_mostly = 300;
3457 int weight_p __read_mostly = 64; /* old backlog weight */
3459 /* Called with irq disabled */
3460 static inline void ____napi_schedule(struct softnet_data *sd,
3461 struct napi_struct *napi)
3463 list_add_tail(&napi->poll_list, &sd->poll_list);
3464 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3469 /* One global table that all flow-based protocols share. */
3470 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3471 EXPORT_SYMBOL(rps_sock_flow_table);
3472 u32 rps_cpu_mask __read_mostly;
3473 EXPORT_SYMBOL(rps_cpu_mask);
3475 struct static_key rps_needed __read_mostly;
3476 EXPORT_SYMBOL(rps_needed);
3478 static struct rps_dev_flow *
3479 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3480 struct rps_dev_flow *rflow, u16 next_cpu)
3482 if (next_cpu < nr_cpu_ids) {
3483 #ifdef CONFIG_RFS_ACCEL
3484 struct netdev_rx_queue *rxqueue;
3485 struct rps_dev_flow_table *flow_table;
3486 struct rps_dev_flow *old_rflow;
3491 /* Should we steer this flow to a different hardware queue? */
3492 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3493 !(dev->features & NETIF_F_NTUPLE))
3495 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3496 if (rxq_index == skb_get_rx_queue(skb))
3499 rxqueue = dev->_rx + rxq_index;
3500 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3503 flow_id = skb_get_hash(skb) & flow_table->mask;
3504 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3505 rxq_index, flow_id);
3509 rflow = &flow_table->flows[flow_id];
3511 if (old_rflow->filter == rflow->filter)
3512 old_rflow->filter = RPS_NO_FILTER;
3516 per_cpu(softnet_data, next_cpu).input_queue_head;
3519 rflow->cpu = next_cpu;
3524 * get_rps_cpu is called from netif_receive_skb and returns the target
3525 * CPU from the RPS map of the receiving queue for a given skb.
3526 * rcu_read_lock must be held on entry.
3528 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3529 struct rps_dev_flow **rflowp)
3531 const struct rps_sock_flow_table *sock_flow_table;
3532 struct netdev_rx_queue *rxqueue = dev->_rx;
3533 struct rps_dev_flow_table *flow_table;
3534 struct rps_map *map;
3539 if (skb_rx_queue_recorded(skb)) {
3540 u16 index = skb_get_rx_queue(skb);
3542 if (unlikely(index >= dev->real_num_rx_queues)) {
3543 WARN_ONCE(dev->real_num_rx_queues > 1,
3544 "%s received packet on queue %u, but number "
3545 "of RX queues is %u\n",
3546 dev->name, index, dev->real_num_rx_queues);
3552 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3554 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3555 map = rcu_dereference(rxqueue->rps_map);
3556 if (!flow_table && !map)
3559 skb_reset_network_header(skb);
3560 hash = skb_get_hash(skb);
3564 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3565 if (flow_table && sock_flow_table) {
3566 struct rps_dev_flow *rflow;
3570 /* First check into global flow table if there is a match */
3571 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3572 if ((ident ^ hash) & ~rps_cpu_mask)
3575 next_cpu = ident & rps_cpu_mask;
3577 /* OK, now we know there is a match,
3578 * we can look at the local (per receive queue) flow table
3580 rflow = &flow_table->flows[hash & flow_table->mask];
3584 * If the desired CPU (where last recvmsg was done) is
3585 * different from current CPU (one in the rx-queue flow
3586 * table entry), switch if one of the following holds:
3587 * - Current CPU is unset (>= nr_cpu_ids).
3588 * - Current CPU is offline.
3589 * - The current CPU's queue tail has advanced beyond the
3590 * last packet that was enqueued using this table entry.
3591 * This guarantees that all previous packets for the flow
3592 * have been dequeued, thus preserving in order delivery.
3594 if (unlikely(tcpu != next_cpu) &&
3595 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3596 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3597 rflow->last_qtail)) >= 0)) {
3599 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3602 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3612 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3613 if (cpu_online(tcpu)) {
3623 #ifdef CONFIG_RFS_ACCEL
3626 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3627 * @dev: Device on which the filter was set
3628 * @rxq_index: RX queue index
3629 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3630 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3632 * Drivers that implement ndo_rx_flow_steer() should periodically call
3633 * this function for each installed filter and remove the filters for
3634 * which it returns %true.
3636 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3637 u32 flow_id, u16 filter_id)
3639 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3640 struct rps_dev_flow_table *flow_table;
3641 struct rps_dev_flow *rflow;
3646 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3647 if (flow_table && flow_id <= flow_table->mask) {
3648 rflow = &flow_table->flows[flow_id];
3649 cpu = ACCESS_ONCE(rflow->cpu);
3650 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3651 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3652 rflow->last_qtail) <
3653 (int)(10 * flow_table->mask)))
3659 EXPORT_SYMBOL(rps_may_expire_flow);
3661 #endif /* CONFIG_RFS_ACCEL */
3663 /* Called from hardirq (IPI) context */
3664 static void rps_trigger_softirq(void *data)
3666 struct softnet_data *sd = data;
3668 ____napi_schedule(sd, &sd->backlog);
3672 #endif /* CONFIG_RPS */
3675 * Check if this softnet_data structure is another cpu one
3676 * If yes, queue it to our IPI list and return 1
3679 static int rps_ipi_queued(struct softnet_data *sd)
3682 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3685 sd->rps_ipi_next = mysd->rps_ipi_list;
3686 mysd->rps_ipi_list = sd;
3688 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3691 #endif /* CONFIG_RPS */
3695 #ifdef CONFIG_NET_FLOW_LIMIT
3696 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3699 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3701 #ifdef CONFIG_NET_FLOW_LIMIT
3702 struct sd_flow_limit *fl;
3703 struct softnet_data *sd;
3704 unsigned int old_flow, new_flow;
3706 if (qlen < (netdev_max_backlog >> 1))
3709 sd = this_cpu_ptr(&softnet_data);
3712 fl = rcu_dereference(sd->flow_limit);
3714 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3715 old_flow = fl->history[fl->history_head];
3716 fl->history[fl->history_head] = new_flow;
3719 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3721 if (likely(fl->buckets[old_flow]))
3722 fl->buckets[old_flow]--;
3724 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3736 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3737 * queue (may be a remote CPU queue).
3739 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3740 unsigned int *qtail)
3742 struct softnet_data *sd;
3743 unsigned long flags;
3746 sd = &per_cpu(softnet_data, cpu);
3748 local_irq_save(flags);
3751 if (!netif_running(skb->dev))
3753 qlen = skb_queue_len(&sd->input_pkt_queue);
3754 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3757 __skb_queue_tail(&sd->input_pkt_queue, skb);
3758 input_queue_tail_incr_save(sd, qtail);
3760 local_irq_restore(flags);
3761 return NET_RX_SUCCESS;
3764 /* Schedule NAPI for backlog device
3765 * We can use non atomic operation since we own the queue lock
3767 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3768 if (!rps_ipi_queued(sd))
3769 ____napi_schedule(sd, &sd->backlog);
3778 local_irq_restore(flags);
3780 atomic_long_inc(&skb->dev->rx_dropped);
3785 static int netif_rx_internal(struct sk_buff *skb)
3789 net_timestamp_check(netdev_tstamp_prequeue, skb);
3791 trace_netif_rx(skb);
3793 if (static_key_false(&rps_needed)) {
3794 struct rps_dev_flow voidflow, *rflow = &voidflow;
3800 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3802 cpu = smp_processor_id();
3804 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3812 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3819 * netif_rx - post buffer to the network code
3820 * @skb: buffer to post
3822 * This function receives a packet from a device driver and queues it for
3823 * the upper (protocol) levels to process. It always succeeds. The buffer
3824 * may be dropped during processing for congestion control or by the
3828 * NET_RX_SUCCESS (no congestion)
3829 * NET_RX_DROP (packet was dropped)
3833 int netif_rx(struct sk_buff *skb)
3835 trace_netif_rx_entry(skb);
3837 return netif_rx_internal(skb);
3839 EXPORT_SYMBOL(netif_rx);
3841 int netif_rx_ni(struct sk_buff *skb)
3845 trace_netif_rx_ni_entry(skb);
3848 err = netif_rx_internal(skb);
3849 if (local_softirq_pending())
3855 EXPORT_SYMBOL(netif_rx_ni);
3857 static void net_tx_action(struct softirq_action *h)
3859 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3861 if (sd->completion_queue) {
3862 struct sk_buff *clist;
3864 local_irq_disable();
3865 clist = sd->completion_queue;
3866 sd->completion_queue = NULL;
3870 struct sk_buff *skb = clist;
3871 clist = clist->next;
3873 WARN_ON(atomic_read(&skb->users));
3874 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3875 trace_consume_skb(skb);
3877 trace_kfree_skb(skb, net_tx_action);
3879 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
3882 __kfree_skb_defer(skb);
3885 __kfree_skb_flush();
3888 if (sd->output_queue) {
3891 local_irq_disable();
3892 head = sd->output_queue;
3893 sd->output_queue = NULL;
3894 sd->output_queue_tailp = &sd->output_queue;
3898 struct Qdisc *q = head;
3899 spinlock_t *root_lock;
3901 head = head->next_sched;
3903 root_lock = qdisc_lock(q);
3904 spin_lock(root_lock);
3905 /* We need to make sure head->next_sched is read
3906 * before clearing __QDISC_STATE_SCHED
3908 smp_mb__before_atomic();
3909 clear_bit(__QDISC_STATE_SCHED, &q->state);
3911 spin_unlock(root_lock);
3916 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3917 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3918 /* This hook is defined here for ATM LANE */
3919 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3920 unsigned char *addr) __read_mostly;
3921 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3924 static inline struct sk_buff *
3925 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
3926 struct net_device *orig_dev)
3928 #ifdef CONFIG_NET_CLS_ACT
3929 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3930 struct tcf_result cl_res;
3932 /* If there's at least one ingress present somewhere (so
3933 * we get here via enabled static key), remaining devices
3934 * that are not configured with an ingress qdisc will bail
3940 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3944 qdisc_skb_cb(skb)->pkt_len = skb->len;
3945 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3946 qdisc_bstats_cpu_update(cl->q, skb);
3948 switch (tc_classify(skb, cl, &cl_res, false)) {
3950 case TC_ACT_RECLASSIFY:
3951 skb->tc_index = TC_H_MIN(cl_res.classid);
3954 qdisc_qstats_cpu_drop(cl->q);
3961 case TC_ACT_REDIRECT:
3962 /* skb_mac_header check was done by cls/act_bpf, so
3963 * we can safely push the L2 header back before
3964 * redirecting to another netdev
3966 __skb_push(skb, skb->mac_len);
3967 skb_do_redirect(skb);
3972 #endif /* CONFIG_NET_CLS_ACT */
3977 * netdev_rx_handler_register - register receive handler
3978 * @dev: device to register a handler for
3979 * @rx_handler: receive handler to register
3980 * @rx_handler_data: data pointer that is used by rx handler
3982 * Register a receive handler for a device. This handler will then be
3983 * called from __netif_receive_skb. A negative errno code is returned
3986 * The caller must hold the rtnl_mutex.
3988 * For a general description of rx_handler, see enum rx_handler_result.
3990 int netdev_rx_handler_register(struct net_device *dev,
3991 rx_handler_func_t *rx_handler,
3992 void *rx_handler_data)
3996 if (dev->rx_handler)
3999 /* Note: rx_handler_data must be set before rx_handler */
4000 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4001 rcu_assign_pointer(dev->rx_handler, rx_handler);
4005 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4008 * netdev_rx_handler_unregister - unregister receive handler
4009 * @dev: device to unregister a handler from
4011 * Unregister a receive handler from a device.
4013 * The caller must hold the rtnl_mutex.
4015 void netdev_rx_handler_unregister(struct net_device *dev)
4019 RCU_INIT_POINTER(dev->rx_handler, NULL);
4020 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4021 * section has a guarantee to see a non NULL rx_handler_data
4025 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4027 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4030 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4031 * the special handling of PFMEMALLOC skbs.
4033 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4035 switch (skb->protocol) {
4036 case htons(ETH_P_ARP):
4037 case htons(ETH_P_IP):
4038 case htons(ETH_P_IPV6):
4039 case htons(ETH_P_8021Q):
4040 case htons(ETH_P_8021AD):
4047 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4048 int *ret, struct net_device *orig_dev)
4050 #ifdef CONFIG_NETFILTER_INGRESS
4051 if (nf_hook_ingress_active(skb)) {
4053 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4057 return nf_hook_ingress(skb);
4059 #endif /* CONFIG_NETFILTER_INGRESS */
4063 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
4065 struct packet_type *ptype, *pt_prev;
4066 rx_handler_func_t *rx_handler;
4067 struct net_device *orig_dev;
4068 bool deliver_exact = false;
4069 int ret = NET_RX_DROP;
4072 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4074 trace_netif_receive_skb(skb);
4076 orig_dev = skb->dev;
4078 skb_reset_network_header(skb);
4079 if (!skb_transport_header_was_set(skb))
4080 skb_reset_transport_header(skb);
4081 skb_reset_mac_len(skb);
4086 skb->skb_iif = skb->dev->ifindex;
4088 __this_cpu_inc(softnet_data.processed);
4090 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4091 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4092 skb = skb_vlan_untag(skb);
4097 #ifdef CONFIG_NET_CLS_ACT
4098 if (skb->tc_verd & TC_NCLS) {
4099 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
4107 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4109 ret = deliver_skb(skb, pt_prev, orig_dev);
4113 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4115 ret = deliver_skb(skb, pt_prev, orig_dev);
4120 #ifdef CONFIG_NET_INGRESS
4121 if (static_key_false(&ingress_needed)) {
4122 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4126 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4130 #ifdef CONFIG_NET_CLS_ACT
4134 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4137 if (skb_vlan_tag_present(skb)) {
4139 ret = deliver_skb(skb, pt_prev, orig_dev);
4142 if (vlan_do_receive(&skb))
4144 else if (unlikely(!skb))
4148 rx_handler = rcu_dereference(skb->dev->rx_handler);
4151 ret = deliver_skb(skb, pt_prev, orig_dev);
4154 switch (rx_handler(&skb)) {
4155 case RX_HANDLER_CONSUMED:
4156 ret = NET_RX_SUCCESS;
4158 case RX_HANDLER_ANOTHER:
4160 case RX_HANDLER_EXACT:
4161 deliver_exact = true;
4162 case RX_HANDLER_PASS:
4169 if (unlikely(skb_vlan_tag_present(skb))) {
4170 if (skb_vlan_tag_get_id(skb))
4171 skb->pkt_type = PACKET_OTHERHOST;
4172 /* Note: we might in the future use prio bits
4173 * and set skb->priority like in vlan_do_receive()
4174 * For the time being, just ignore Priority Code Point
4179 type = skb->protocol;
4181 /* deliver only exact match when indicated */
4182 if (likely(!deliver_exact)) {
4183 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4184 &ptype_base[ntohs(type) &
4188 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4189 &orig_dev->ptype_specific);
4191 if (unlikely(skb->dev != orig_dev)) {
4192 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4193 &skb->dev->ptype_specific);
4197 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
4200 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4204 atomic_long_inc(&skb->dev->rx_dropped);
4206 atomic_long_inc(&skb->dev->rx_nohandler);
4208 /* Jamal, now you will not able to escape explaining
4209 * me how you were going to use this. :-)
4218 static int __netif_receive_skb(struct sk_buff *skb)
4222 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4223 unsigned long pflags = current->flags;
4226 * PFMEMALLOC skbs are special, they should
4227 * - be delivered to SOCK_MEMALLOC sockets only
4228 * - stay away from userspace
4229 * - have bounded memory usage
4231 * Use PF_MEMALLOC as this saves us from propagating the allocation
4232 * context down to all allocation sites.
4234 current->flags |= PF_MEMALLOC;
4235 ret = __netif_receive_skb_core(skb, true);
4236 tsk_restore_flags(current, pflags, PF_MEMALLOC);
4238 ret = __netif_receive_skb_core(skb, false);
4243 static int netif_receive_skb_internal(struct sk_buff *skb)
4247 net_timestamp_check(netdev_tstamp_prequeue, skb);
4249 if (skb_defer_rx_timestamp(skb))
4250 return NET_RX_SUCCESS;
4255 if (static_key_false(&rps_needed)) {
4256 struct rps_dev_flow voidflow, *rflow = &voidflow;
4257 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4260 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4266 ret = __netif_receive_skb(skb);
4272 * netif_receive_skb - process receive buffer from network
4273 * @skb: buffer to process
4275 * netif_receive_skb() is the main receive data processing function.
4276 * It always succeeds. The buffer may be dropped during processing
4277 * for congestion control or by the protocol layers.
4279 * This function may only be called from softirq context and interrupts
4280 * should be enabled.
4282 * Return values (usually ignored):
4283 * NET_RX_SUCCESS: no congestion
4284 * NET_RX_DROP: packet was dropped
4286 int netif_receive_skb(struct sk_buff *skb)
4288 trace_netif_receive_skb_entry(skb);
4290 return netif_receive_skb_internal(skb);
4292 EXPORT_SYMBOL(netif_receive_skb);
4294 /* Network device is going away, flush any packets still pending
4295 * Called with irqs disabled.
4297 static void flush_backlog(void *arg)
4299 struct net_device *dev = arg;
4300 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4301 struct sk_buff *skb, *tmp;
4304 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4305 if (skb->dev == dev) {
4306 __skb_unlink(skb, &sd->input_pkt_queue);
4308 input_queue_head_incr(sd);
4313 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4314 if (skb->dev == dev) {
4315 __skb_unlink(skb, &sd->process_queue);
4317 input_queue_head_incr(sd);
4322 static int napi_gro_complete(struct sk_buff *skb)
4324 struct packet_offload *ptype;
4325 __be16 type = skb->protocol;
4326 struct list_head *head = &offload_base;
4329 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4331 if (NAPI_GRO_CB(skb)->count == 1) {
4332 skb_shinfo(skb)->gso_size = 0;
4337 list_for_each_entry_rcu(ptype, head, list) {
4338 if (ptype->type != type || !ptype->callbacks.gro_complete)
4341 err = ptype->callbacks.gro_complete(skb, 0);
4347 WARN_ON(&ptype->list == head);
4349 return NET_RX_SUCCESS;
4353 return netif_receive_skb_internal(skb);
4356 /* napi->gro_list contains packets ordered by age.
4357 * youngest packets at the head of it.
4358 * Complete skbs in reverse order to reduce latencies.
4360 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4362 struct sk_buff *skb, *prev = NULL;
4364 /* scan list and build reverse chain */
4365 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4370 for (skb = prev; skb; skb = prev) {
4373 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4377 napi_gro_complete(skb);
4381 napi->gro_list = NULL;
4383 EXPORT_SYMBOL(napi_gro_flush);
4385 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4388 unsigned int maclen = skb->dev->hard_header_len;
4389 u32 hash = skb_get_hash_raw(skb);
4391 for (p = napi->gro_list; p; p = p->next) {
4392 unsigned long diffs;
4394 NAPI_GRO_CB(p)->flush = 0;
4396 if (hash != skb_get_hash_raw(p)) {
4397 NAPI_GRO_CB(p)->same_flow = 0;
4401 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4402 diffs |= p->vlan_tci ^ skb->vlan_tci;
4403 diffs |= skb_metadata_dst_cmp(p, skb);
4404 if (maclen == ETH_HLEN)
4405 diffs |= compare_ether_header(skb_mac_header(p),
4406 skb_mac_header(skb));
4408 diffs = memcmp(skb_mac_header(p),
4409 skb_mac_header(skb),
4411 NAPI_GRO_CB(p)->same_flow = !diffs;
4415 static void skb_gro_reset_offset(struct sk_buff *skb)
4417 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4418 const skb_frag_t *frag0 = &pinfo->frags[0];
4420 NAPI_GRO_CB(skb)->data_offset = 0;
4421 NAPI_GRO_CB(skb)->frag0 = NULL;
4422 NAPI_GRO_CB(skb)->frag0_len = 0;
4424 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4426 !PageHighMem(skb_frag_page(frag0))) {
4427 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4428 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4432 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4434 struct skb_shared_info *pinfo = skb_shinfo(skb);
4436 BUG_ON(skb->end - skb->tail < grow);
4438 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4440 skb->data_len -= grow;
4443 pinfo->frags[0].page_offset += grow;
4444 skb_frag_size_sub(&pinfo->frags[0], grow);
4446 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4447 skb_frag_unref(skb, 0);
4448 memmove(pinfo->frags, pinfo->frags + 1,
4449 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4453 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4455 struct sk_buff **pp = NULL;
4456 struct packet_offload *ptype;
4457 __be16 type = skb->protocol;
4458 struct list_head *head = &offload_base;
4460 enum gro_result ret;
4463 if (!(skb->dev->features & NETIF_F_GRO))
4466 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4469 gro_list_prepare(napi, skb);
4472 list_for_each_entry_rcu(ptype, head, list) {
4473 if (ptype->type != type || !ptype->callbacks.gro_receive)
4476 skb_set_network_header(skb, skb_gro_offset(skb));
4477 skb_reset_mac_len(skb);
4478 NAPI_GRO_CB(skb)->same_flow = 0;
4479 NAPI_GRO_CB(skb)->flush = 0;
4480 NAPI_GRO_CB(skb)->free = 0;
4481 NAPI_GRO_CB(skb)->encap_mark = 0;
4482 NAPI_GRO_CB(skb)->is_fou = 0;
4483 NAPI_GRO_CB(skb)->is_atomic = 1;
4484 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4486 /* Setup for GRO checksum validation */
4487 switch (skb->ip_summed) {
4488 case CHECKSUM_COMPLETE:
4489 NAPI_GRO_CB(skb)->csum = skb->csum;
4490 NAPI_GRO_CB(skb)->csum_valid = 1;
4491 NAPI_GRO_CB(skb)->csum_cnt = 0;
4493 case CHECKSUM_UNNECESSARY:
4494 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4495 NAPI_GRO_CB(skb)->csum_valid = 0;
4498 NAPI_GRO_CB(skb)->csum_cnt = 0;
4499 NAPI_GRO_CB(skb)->csum_valid = 0;
4502 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4507 if (&ptype->list == head)
4510 same_flow = NAPI_GRO_CB(skb)->same_flow;
4511 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4514 struct sk_buff *nskb = *pp;
4518 napi_gro_complete(nskb);
4525 if (NAPI_GRO_CB(skb)->flush)
4528 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4529 struct sk_buff *nskb = napi->gro_list;
4531 /* locate the end of the list to select the 'oldest' flow */
4532 while (nskb->next) {
4538 napi_gro_complete(nskb);
4542 NAPI_GRO_CB(skb)->count = 1;
4543 NAPI_GRO_CB(skb)->age = jiffies;
4544 NAPI_GRO_CB(skb)->last = skb;
4545 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4546 skb->next = napi->gro_list;
4547 napi->gro_list = skb;
4551 grow = skb_gro_offset(skb) - skb_headlen(skb);
4553 gro_pull_from_frag0(skb, grow);
4562 struct packet_offload *gro_find_receive_by_type(__be16 type)
4564 struct list_head *offload_head = &offload_base;
4565 struct packet_offload *ptype;
4567 list_for_each_entry_rcu(ptype, offload_head, list) {
4568 if (ptype->type != type || !ptype->callbacks.gro_receive)
4574 EXPORT_SYMBOL(gro_find_receive_by_type);
4576 struct packet_offload *gro_find_complete_by_type(__be16 type)
4578 struct list_head *offload_head = &offload_base;
4579 struct packet_offload *ptype;
4581 list_for_each_entry_rcu(ptype, offload_head, list) {
4582 if (ptype->type != type || !ptype->callbacks.gro_complete)
4588 EXPORT_SYMBOL(gro_find_complete_by_type);
4590 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4594 if (netif_receive_skb_internal(skb))
4602 case GRO_MERGED_FREE:
4603 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) {
4605 kmem_cache_free(skbuff_head_cache, skb);
4619 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4621 skb_mark_napi_id(skb, napi);
4622 trace_napi_gro_receive_entry(skb);
4624 skb_gro_reset_offset(skb);
4626 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4628 EXPORT_SYMBOL(napi_gro_receive);
4630 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4632 if (unlikely(skb->pfmemalloc)) {
4636 __skb_pull(skb, skb_headlen(skb));
4637 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4638 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4640 skb->dev = napi->dev;
4642 skb->encapsulation = 0;
4643 skb_shinfo(skb)->gso_type = 0;
4644 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4649 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4651 struct sk_buff *skb = napi->skb;
4654 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4657 skb_mark_napi_id(skb, napi);
4662 EXPORT_SYMBOL(napi_get_frags);
4664 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4665 struct sk_buff *skb,
4671 __skb_push(skb, ETH_HLEN);
4672 skb->protocol = eth_type_trans(skb, skb->dev);
4673 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4678 case GRO_MERGED_FREE:
4679 napi_reuse_skb(napi, skb);
4689 /* Upper GRO stack assumes network header starts at gro_offset=0
4690 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4691 * We copy ethernet header into skb->data to have a common layout.
4693 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4695 struct sk_buff *skb = napi->skb;
4696 const struct ethhdr *eth;
4697 unsigned int hlen = sizeof(*eth);
4701 skb_reset_mac_header(skb);
4702 skb_gro_reset_offset(skb);
4704 eth = skb_gro_header_fast(skb, 0);
4705 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4706 eth = skb_gro_header_slow(skb, hlen, 0);
4707 if (unlikely(!eth)) {
4708 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
4709 __func__, napi->dev->name);
4710 napi_reuse_skb(napi, skb);
4714 gro_pull_from_frag0(skb, hlen);
4715 NAPI_GRO_CB(skb)->frag0 += hlen;
4716 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4718 __skb_pull(skb, hlen);
4721 * This works because the only protocols we care about don't require
4723 * We'll fix it up properly in napi_frags_finish()
4725 skb->protocol = eth->h_proto;
4730 gro_result_t napi_gro_frags(struct napi_struct *napi)
4732 struct sk_buff *skb = napi_frags_skb(napi);
4737 trace_napi_gro_frags_entry(skb);
4739 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4741 EXPORT_SYMBOL(napi_gro_frags);
4743 /* Compute the checksum from gro_offset and return the folded value
4744 * after adding in any pseudo checksum.
4746 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4751 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4753 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4754 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4756 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4757 !skb->csum_complete_sw)
4758 netdev_rx_csum_fault(skb->dev);
4761 NAPI_GRO_CB(skb)->csum = wsum;
4762 NAPI_GRO_CB(skb)->csum_valid = 1;
4766 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4769 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4770 * Note: called with local irq disabled, but exits with local irq enabled.
4772 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4775 struct softnet_data *remsd = sd->rps_ipi_list;
4778 sd->rps_ipi_list = NULL;
4782 /* Send pending IPI's to kick RPS processing on remote cpus. */
4784 struct softnet_data *next = remsd->rps_ipi_next;
4786 if (cpu_online(remsd->cpu))
4787 smp_call_function_single_async(remsd->cpu,
4796 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4799 return sd->rps_ipi_list != NULL;
4805 static int process_backlog(struct napi_struct *napi, int quota)
4808 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4810 /* Check if we have pending ipi, its better to send them now,
4811 * not waiting net_rx_action() end.
4813 if (sd_has_rps_ipi_waiting(sd)) {
4814 local_irq_disable();
4815 net_rps_action_and_irq_enable(sd);
4818 napi->weight = weight_p;
4819 local_irq_disable();
4821 struct sk_buff *skb;
4823 while ((skb = __skb_dequeue(&sd->process_queue))) {
4826 __netif_receive_skb(skb);
4828 local_irq_disable();
4829 input_queue_head_incr(sd);
4830 if (++work >= quota) {
4837 if (skb_queue_empty(&sd->input_pkt_queue)) {
4839 * Inline a custom version of __napi_complete().
4840 * only current cpu owns and manipulates this napi,
4841 * and NAPI_STATE_SCHED is the only possible flag set
4843 * We can use a plain write instead of clear_bit(),
4844 * and we dont need an smp_mb() memory barrier.
4852 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4853 &sd->process_queue);
4862 * __napi_schedule - schedule for receive
4863 * @n: entry to schedule
4865 * The entry's receive function will be scheduled to run.
4866 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4868 void __napi_schedule(struct napi_struct *n)
4870 unsigned long flags;
4872 local_irq_save(flags);
4873 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4874 local_irq_restore(flags);
4876 EXPORT_SYMBOL(__napi_schedule);
4879 * __napi_schedule_irqoff - schedule for receive
4880 * @n: entry to schedule
4882 * Variant of __napi_schedule() assuming hard irqs are masked
4884 void __napi_schedule_irqoff(struct napi_struct *n)
4886 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4888 EXPORT_SYMBOL(__napi_schedule_irqoff);
4890 void __napi_complete(struct napi_struct *n)
4892 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4894 list_del_init(&n->poll_list);
4895 smp_mb__before_atomic();
4896 clear_bit(NAPI_STATE_SCHED, &n->state);
4898 EXPORT_SYMBOL(__napi_complete);
4900 void napi_complete_done(struct napi_struct *n, int work_done)
4902 unsigned long flags;
4905 * don't let napi dequeue from the cpu poll list
4906 * just in case its running on a different cpu
4908 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4912 unsigned long timeout = 0;
4915 timeout = n->dev->gro_flush_timeout;
4918 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4919 HRTIMER_MODE_REL_PINNED);
4921 napi_gro_flush(n, false);
4923 if (likely(list_empty(&n->poll_list))) {
4924 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4926 /* If n->poll_list is not empty, we need to mask irqs */
4927 local_irq_save(flags);
4929 local_irq_restore(flags);
4932 EXPORT_SYMBOL(napi_complete_done);
4934 /* must be called under rcu_read_lock(), as we dont take a reference */
4935 static struct napi_struct *napi_by_id(unsigned int napi_id)
4937 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4938 struct napi_struct *napi;
4940 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4941 if (napi->napi_id == napi_id)
4947 #if defined(CONFIG_NET_RX_BUSY_POLL)
4948 #define BUSY_POLL_BUDGET 8
4949 bool sk_busy_loop(struct sock *sk, int nonblock)
4951 unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0;
4952 int (*busy_poll)(struct napi_struct *dev);
4953 struct napi_struct *napi;
4958 napi = napi_by_id(sk->sk_napi_id);
4962 /* Note: ndo_busy_poll method is optional in linux-4.5 */
4963 busy_poll = napi->dev->netdev_ops->ndo_busy_poll;
4969 rc = busy_poll(napi);
4970 } else if (napi_schedule_prep(napi)) {
4971 void *have = netpoll_poll_lock(napi);
4973 if (test_bit(NAPI_STATE_SCHED, &napi->state)) {
4974 rc = napi->poll(napi, BUSY_POLL_BUDGET);
4975 trace_napi_poll(napi);
4976 if (rc == BUSY_POLL_BUDGET) {
4977 napi_complete_done(napi, rc);
4978 napi_schedule(napi);
4981 netpoll_poll_unlock(have);
4984 __NET_ADD_STATS(sock_net(sk),
4985 LINUX_MIB_BUSYPOLLRXPACKETS, rc);
4988 if (rc == LL_FLUSH_FAILED)
4989 break; /* permanent failure */
4992 } while (!nonblock && skb_queue_empty(&sk->sk_receive_queue) &&
4993 !need_resched() && !busy_loop_timeout(end_time));
4995 rc = !skb_queue_empty(&sk->sk_receive_queue);
5000 EXPORT_SYMBOL(sk_busy_loop);
5002 #endif /* CONFIG_NET_RX_BUSY_POLL */
5004 void napi_hash_add(struct napi_struct *napi)
5006 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
5007 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
5010 spin_lock(&napi_hash_lock);
5012 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */
5014 if (unlikely(++napi_gen_id < NR_CPUS + 1))
5015 napi_gen_id = NR_CPUS + 1;
5016 } while (napi_by_id(napi_gen_id));
5017 napi->napi_id = napi_gen_id;
5019 hlist_add_head_rcu(&napi->napi_hash_node,
5020 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
5022 spin_unlock(&napi_hash_lock);
5024 EXPORT_SYMBOL_GPL(napi_hash_add);
5026 /* Warning : caller is responsible to make sure rcu grace period
5027 * is respected before freeing memory containing @napi
5029 bool napi_hash_del(struct napi_struct *napi)
5031 bool rcu_sync_needed = false;
5033 spin_lock(&napi_hash_lock);
5035 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
5036 rcu_sync_needed = true;
5037 hlist_del_rcu(&napi->napi_hash_node);
5039 spin_unlock(&napi_hash_lock);
5040 return rcu_sync_needed;
5042 EXPORT_SYMBOL_GPL(napi_hash_del);
5044 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
5046 struct napi_struct *napi;
5048 napi = container_of(timer, struct napi_struct, timer);
5050 napi_schedule(napi);
5052 return HRTIMER_NORESTART;
5055 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
5056 int (*poll)(struct napi_struct *, int), int weight)
5058 INIT_LIST_HEAD(&napi->poll_list);
5059 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
5060 napi->timer.function = napi_watchdog;
5061 napi->gro_count = 0;
5062 napi->gro_list = NULL;
5065 if (weight > NAPI_POLL_WEIGHT)
5066 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5068 napi->weight = weight;
5069 list_add(&napi->dev_list, &dev->napi_list);
5071 #ifdef CONFIG_NETPOLL
5072 spin_lock_init(&napi->poll_lock);
5073 napi->poll_owner = -1;
5075 set_bit(NAPI_STATE_SCHED, &napi->state);
5076 napi_hash_add(napi);
5078 EXPORT_SYMBOL(netif_napi_add);
5080 void napi_disable(struct napi_struct *n)
5083 set_bit(NAPI_STATE_DISABLE, &n->state);
5085 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
5087 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
5090 hrtimer_cancel(&n->timer);
5092 clear_bit(NAPI_STATE_DISABLE, &n->state);
5094 EXPORT_SYMBOL(napi_disable);
5096 /* Must be called in process context */
5097 void netif_napi_del(struct napi_struct *napi)
5100 if (napi_hash_del(napi))
5102 list_del_init(&napi->dev_list);
5103 napi_free_frags(napi);
5105 kfree_skb_list(napi->gro_list);
5106 napi->gro_list = NULL;
5107 napi->gro_count = 0;
5109 EXPORT_SYMBOL(netif_napi_del);
5111 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
5116 list_del_init(&n->poll_list);
5118 have = netpoll_poll_lock(n);
5122 /* This NAPI_STATE_SCHED test is for avoiding a race
5123 * with netpoll's poll_napi(). Only the entity which
5124 * obtains the lock and sees NAPI_STATE_SCHED set will
5125 * actually make the ->poll() call. Therefore we avoid
5126 * accidentally calling ->poll() when NAPI is not scheduled.
5129 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
5130 work = n->poll(n, weight);
5134 WARN_ON_ONCE(work > weight);
5136 if (likely(work < weight))
5139 /* Drivers must not modify the NAPI state if they
5140 * consume the entire weight. In such cases this code
5141 * still "owns" the NAPI instance and therefore can
5142 * move the instance around on the list at-will.
5144 if (unlikely(napi_disable_pending(n))) {
5150 /* flush too old packets
5151 * If HZ < 1000, flush all packets.
5153 napi_gro_flush(n, HZ >= 1000);
5156 /* Some drivers may have called napi_schedule
5157 * prior to exhausting their budget.
5159 if (unlikely(!list_empty(&n->poll_list))) {
5160 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5161 n->dev ? n->dev->name : "backlog");
5165 list_add_tail(&n->poll_list, repoll);
5168 netpoll_poll_unlock(have);
5173 static void net_rx_action(struct softirq_action *h)
5175 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5176 unsigned long time_limit = jiffies + 2;
5177 int budget = netdev_budget;
5181 local_irq_disable();
5182 list_splice_init(&sd->poll_list, &list);
5186 struct napi_struct *n;
5188 if (list_empty(&list)) {
5189 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
5194 n = list_first_entry(&list, struct napi_struct, poll_list);
5195 budget -= napi_poll(n, &repoll);
5197 /* If softirq window is exhausted then punt.
5198 * Allow this to run for 2 jiffies since which will allow
5199 * an average latency of 1.5/HZ.
5201 if (unlikely(budget <= 0 ||
5202 time_after_eq(jiffies, time_limit))) {
5208 __kfree_skb_flush();
5209 local_irq_disable();
5211 list_splice_tail_init(&sd->poll_list, &list);
5212 list_splice_tail(&repoll, &list);
5213 list_splice(&list, &sd->poll_list);
5214 if (!list_empty(&sd->poll_list))
5215 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5217 net_rps_action_and_irq_enable(sd);
5220 struct netdev_adjacent {
5221 struct net_device *dev;
5223 /* upper master flag, there can only be one master device per list */
5226 /* counter for the number of times this device was added to us */
5229 /* private field for the users */
5232 struct list_head list;
5233 struct rcu_head rcu;
5236 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
5237 struct list_head *adj_list)
5239 struct netdev_adjacent *adj;
5241 list_for_each_entry(adj, adj_list, list) {
5242 if (adj->dev == adj_dev)
5249 * netdev_has_upper_dev - Check if device is linked to an upper device
5251 * @upper_dev: upper device to check
5253 * Find out if a device is linked to specified upper device and return true
5254 * in case it is. Note that this checks only immediate upper device,
5255 * not through a complete stack of devices. The caller must hold the RTNL lock.
5257 bool netdev_has_upper_dev(struct net_device *dev,
5258 struct net_device *upper_dev)
5262 return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper);
5264 EXPORT_SYMBOL(netdev_has_upper_dev);
5267 * netdev_has_any_upper_dev - Check if device is linked to some device
5270 * Find out if a device is linked to an upper device and return true in case
5271 * it is. The caller must hold the RTNL lock.
5273 static bool netdev_has_any_upper_dev(struct net_device *dev)
5277 return !list_empty(&dev->all_adj_list.upper);
5281 * netdev_master_upper_dev_get - Get master upper device
5284 * Find a master upper device and return pointer to it or NULL in case
5285 * it's not there. The caller must hold the RTNL lock.
5287 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5289 struct netdev_adjacent *upper;
5293 if (list_empty(&dev->adj_list.upper))
5296 upper = list_first_entry(&dev->adj_list.upper,
5297 struct netdev_adjacent, list);
5298 if (likely(upper->master))
5302 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5304 void *netdev_adjacent_get_private(struct list_head *adj_list)
5306 struct netdev_adjacent *adj;
5308 adj = list_entry(adj_list, struct netdev_adjacent, list);
5310 return adj->private;
5312 EXPORT_SYMBOL(netdev_adjacent_get_private);
5315 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5317 * @iter: list_head ** of the current position
5319 * Gets the next device from the dev's upper list, starting from iter
5320 * position. The caller must hold RCU read lock.
5322 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5323 struct list_head **iter)
5325 struct netdev_adjacent *upper;
5327 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5329 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5331 if (&upper->list == &dev->adj_list.upper)
5334 *iter = &upper->list;
5338 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5341 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
5343 * @iter: list_head ** of the current position
5345 * Gets the next device from the dev's upper list, starting from iter
5346 * position. The caller must hold RCU read lock.
5348 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
5349 struct list_head **iter)
5351 struct netdev_adjacent *upper;
5353 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5355 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5357 if (&upper->list == &dev->all_adj_list.upper)
5360 *iter = &upper->list;
5364 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
5367 * netdev_lower_get_next_private - Get the next ->private from the
5368 * lower neighbour list
5370 * @iter: list_head ** of the current position
5372 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5373 * list, starting from iter position. The caller must hold either hold the
5374 * RTNL lock or its own locking that guarantees that the neighbour lower
5375 * list will remain unchanged.
5377 void *netdev_lower_get_next_private(struct net_device *dev,
5378 struct list_head **iter)
5380 struct netdev_adjacent *lower;
5382 lower = list_entry(*iter, struct netdev_adjacent, list);
5384 if (&lower->list == &dev->adj_list.lower)
5387 *iter = lower->list.next;
5389 return lower->private;
5391 EXPORT_SYMBOL(netdev_lower_get_next_private);
5394 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5395 * lower neighbour list, RCU
5398 * @iter: list_head ** of the current position
5400 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5401 * list, starting from iter position. The caller must hold RCU read lock.
5403 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5404 struct list_head **iter)
5406 struct netdev_adjacent *lower;
5408 WARN_ON_ONCE(!rcu_read_lock_held());
5410 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5412 if (&lower->list == &dev->adj_list.lower)
5415 *iter = &lower->list;
5417 return lower->private;
5419 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5422 * netdev_lower_get_next - Get the next device from the lower neighbour
5425 * @iter: list_head ** of the current position
5427 * Gets the next netdev_adjacent from the dev's lower neighbour
5428 * list, starting from iter position. The caller must hold RTNL lock or
5429 * its own locking that guarantees that the neighbour lower
5430 * list will remain unchanged.
5432 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5434 struct netdev_adjacent *lower;
5436 lower = list_entry(*iter, struct netdev_adjacent, list);
5438 if (&lower->list == &dev->adj_list.lower)
5441 *iter = lower->list.next;
5445 EXPORT_SYMBOL(netdev_lower_get_next);
5448 * netdev_all_lower_get_next - Get the next device from all lower neighbour list
5450 * @iter: list_head ** of the current position
5452 * Gets the next netdev_adjacent from the dev's all lower neighbour
5453 * list, starting from iter position. The caller must hold RTNL lock or
5454 * its own locking that guarantees that the neighbour all lower
5455 * list will remain unchanged.
5457 struct net_device *netdev_all_lower_get_next(struct net_device *dev, struct list_head **iter)
5459 struct netdev_adjacent *lower;
5461 lower = list_entry(*iter, struct netdev_adjacent, list);
5463 if (&lower->list == &dev->all_adj_list.lower)
5466 *iter = lower->list.next;
5470 EXPORT_SYMBOL(netdev_all_lower_get_next);
5473 * netdev_all_lower_get_next_rcu - Get the next device from all
5474 * lower neighbour list, RCU variant
5476 * @iter: list_head ** of the current position
5478 * Gets the next netdev_adjacent from the dev's all lower neighbour
5479 * list, starting from iter position. The caller must hold RCU read lock.
5481 struct net_device *netdev_all_lower_get_next_rcu(struct net_device *dev,
5482 struct list_head **iter)
5484 struct netdev_adjacent *lower;
5486 lower = list_first_or_null_rcu(&dev->all_adj_list.lower,
5487 struct netdev_adjacent, list);
5489 return lower ? lower->dev : NULL;
5491 EXPORT_SYMBOL(netdev_all_lower_get_next_rcu);
5494 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5495 * lower neighbour list, RCU
5499 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5500 * list. The caller must hold RCU read lock.
5502 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5504 struct netdev_adjacent *lower;
5506 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5507 struct netdev_adjacent, list);
5509 return lower->private;
5512 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5515 * netdev_master_upper_dev_get_rcu - Get master upper device
5518 * Find a master upper device and return pointer to it or NULL in case
5519 * it's not there. The caller must hold the RCU read lock.
5521 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5523 struct netdev_adjacent *upper;
5525 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5526 struct netdev_adjacent, list);
5527 if (upper && likely(upper->master))
5531 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5533 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5534 struct net_device *adj_dev,
5535 struct list_head *dev_list)
5537 char linkname[IFNAMSIZ+7];
5538 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5539 "upper_%s" : "lower_%s", adj_dev->name);
5540 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5543 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5545 struct list_head *dev_list)
5547 char linkname[IFNAMSIZ+7];
5548 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5549 "upper_%s" : "lower_%s", name);
5550 sysfs_remove_link(&(dev->dev.kobj), linkname);
5553 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5554 struct net_device *adj_dev,
5555 struct list_head *dev_list)
5557 return (dev_list == &dev->adj_list.upper ||
5558 dev_list == &dev->adj_list.lower) &&
5559 net_eq(dev_net(dev), dev_net(adj_dev));
5562 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5563 struct net_device *adj_dev,
5564 struct list_head *dev_list,
5565 void *private, bool master)
5567 struct netdev_adjacent *adj;
5570 adj = __netdev_find_adj(adj_dev, dev_list);
5577 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5582 adj->master = master;
5584 adj->private = private;
5587 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5588 adj_dev->name, dev->name, adj_dev->name);
5590 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5591 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5596 /* Ensure that master link is always the first item in list. */
5598 ret = sysfs_create_link(&(dev->dev.kobj),
5599 &(adj_dev->dev.kobj), "master");
5601 goto remove_symlinks;
5603 list_add_rcu(&adj->list, dev_list);
5605 list_add_tail_rcu(&adj->list, dev_list);
5611 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5612 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5620 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5621 struct net_device *adj_dev,
5622 struct list_head *dev_list)
5624 struct netdev_adjacent *adj;
5626 adj = __netdev_find_adj(adj_dev, dev_list);
5629 pr_err("tried to remove device %s from %s\n",
5630 dev->name, adj_dev->name);
5634 if (adj->ref_nr > 1) {
5635 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5642 sysfs_remove_link(&(dev->dev.kobj), "master");
5644 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5645 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5647 list_del_rcu(&adj->list);
5648 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5649 adj_dev->name, dev->name, adj_dev->name);
5651 kfree_rcu(adj, rcu);
5654 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5655 struct net_device *upper_dev,
5656 struct list_head *up_list,
5657 struct list_head *down_list,
5658 void *private, bool master)
5662 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5667 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5670 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5677 static int __netdev_adjacent_dev_link(struct net_device *dev,
5678 struct net_device *upper_dev)
5680 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5681 &dev->all_adj_list.upper,
5682 &upper_dev->all_adj_list.lower,
5686 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5687 struct net_device *upper_dev,
5688 struct list_head *up_list,
5689 struct list_head *down_list)
5691 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5692 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5695 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5696 struct net_device *upper_dev)
5698 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5699 &dev->all_adj_list.upper,
5700 &upper_dev->all_adj_list.lower);
5703 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5704 struct net_device *upper_dev,
5705 void *private, bool master)
5707 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5712 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5713 &dev->adj_list.upper,
5714 &upper_dev->adj_list.lower,
5717 __netdev_adjacent_dev_unlink(dev, upper_dev);
5724 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5725 struct net_device *upper_dev)
5727 __netdev_adjacent_dev_unlink(dev, upper_dev);
5728 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5729 &dev->adj_list.upper,
5730 &upper_dev->adj_list.lower);
5733 static int __netdev_upper_dev_link(struct net_device *dev,
5734 struct net_device *upper_dev, bool master,
5735 void *upper_priv, void *upper_info)
5737 struct netdev_notifier_changeupper_info changeupper_info;
5738 struct netdev_adjacent *i, *j, *to_i, *to_j;
5743 if (dev == upper_dev)
5746 /* To prevent loops, check if dev is not upper device to upper_dev. */
5747 if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper))
5750 if (__netdev_find_adj(upper_dev, &dev->adj_list.upper))
5753 if (master && netdev_master_upper_dev_get(dev))
5756 changeupper_info.upper_dev = upper_dev;
5757 changeupper_info.master = master;
5758 changeupper_info.linking = true;
5759 changeupper_info.upper_info = upper_info;
5761 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5762 &changeupper_info.info);
5763 ret = notifier_to_errno(ret);
5767 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
5772 /* Now that we linked these devs, make all the upper_dev's
5773 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5774 * versa, and don't forget the devices itself. All of these
5775 * links are non-neighbours.
5777 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5778 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5779 pr_debug("Interlinking %s with %s, non-neighbour\n",
5780 i->dev->name, j->dev->name);
5781 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5787 /* add dev to every upper_dev's upper device */
5788 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5789 pr_debug("linking %s's upper device %s with %s\n",
5790 upper_dev->name, i->dev->name, dev->name);
5791 ret = __netdev_adjacent_dev_link(dev, i->dev);
5793 goto rollback_upper_mesh;
5796 /* add upper_dev to every dev's lower device */
5797 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5798 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5799 i->dev->name, upper_dev->name);
5800 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5802 goto rollback_lower_mesh;
5805 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5806 &changeupper_info.info);
5807 ret = notifier_to_errno(ret);
5809 goto rollback_lower_mesh;
5813 rollback_lower_mesh:
5815 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5818 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5823 rollback_upper_mesh:
5825 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5828 __netdev_adjacent_dev_unlink(dev, i->dev);
5836 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5837 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5838 if (i == to_i && j == to_j)
5840 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5846 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5852 * netdev_upper_dev_link - Add a link to the upper device
5854 * @upper_dev: new upper device
5856 * Adds a link to device which is upper to this one. The caller must hold
5857 * the RTNL lock. On a failure a negative errno code is returned.
5858 * On success the reference counts are adjusted and the function
5861 int netdev_upper_dev_link(struct net_device *dev,
5862 struct net_device *upper_dev)
5864 return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL);
5866 EXPORT_SYMBOL(netdev_upper_dev_link);
5869 * netdev_master_upper_dev_link - Add a master link to the upper device
5871 * @upper_dev: new upper device
5872 * @upper_priv: upper device private
5873 * @upper_info: upper info to be passed down via notifier
5875 * Adds a link to device which is upper to this one. In this case, only
5876 * one master upper device can be linked, although other non-master devices
5877 * might be linked as well. The caller must hold the RTNL lock.
5878 * On a failure a negative errno code is returned. On success the reference
5879 * counts are adjusted and the function returns zero.
5881 int netdev_master_upper_dev_link(struct net_device *dev,
5882 struct net_device *upper_dev,
5883 void *upper_priv, void *upper_info)
5885 return __netdev_upper_dev_link(dev, upper_dev, true,
5886 upper_priv, upper_info);
5888 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5891 * netdev_upper_dev_unlink - Removes a link to upper device
5893 * @upper_dev: new upper device
5895 * Removes a link to device which is upper to this one. The caller must hold
5898 void netdev_upper_dev_unlink(struct net_device *dev,
5899 struct net_device *upper_dev)
5901 struct netdev_notifier_changeupper_info changeupper_info;
5902 struct netdev_adjacent *i, *j;
5905 changeupper_info.upper_dev = upper_dev;
5906 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
5907 changeupper_info.linking = false;
5909 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5910 &changeupper_info.info);
5912 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5914 /* Here is the tricky part. We must remove all dev's lower
5915 * devices from all upper_dev's upper devices and vice
5916 * versa, to maintain the graph relationship.
5918 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5919 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5920 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5922 /* remove also the devices itself from lower/upper device
5925 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5926 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5928 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5929 __netdev_adjacent_dev_unlink(dev, i->dev);
5931 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5932 &changeupper_info.info);
5934 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5937 * netdev_bonding_info_change - Dispatch event about slave change
5939 * @bonding_info: info to dispatch
5941 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5942 * The caller must hold the RTNL lock.
5944 void netdev_bonding_info_change(struct net_device *dev,
5945 struct netdev_bonding_info *bonding_info)
5947 struct netdev_notifier_bonding_info info;
5949 memcpy(&info.bonding_info, bonding_info,
5950 sizeof(struct netdev_bonding_info));
5951 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5954 EXPORT_SYMBOL(netdev_bonding_info_change);
5956 static void netdev_adjacent_add_links(struct net_device *dev)
5958 struct netdev_adjacent *iter;
5960 struct net *net = dev_net(dev);
5962 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5963 if (!net_eq(net, dev_net(iter->dev)))
5965 netdev_adjacent_sysfs_add(iter->dev, dev,
5966 &iter->dev->adj_list.lower);
5967 netdev_adjacent_sysfs_add(dev, iter->dev,
5968 &dev->adj_list.upper);
5971 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5972 if (!net_eq(net, dev_net(iter->dev)))
5974 netdev_adjacent_sysfs_add(iter->dev, dev,
5975 &iter->dev->adj_list.upper);
5976 netdev_adjacent_sysfs_add(dev, iter->dev,
5977 &dev->adj_list.lower);
5981 static void netdev_adjacent_del_links(struct net_device *dev)
5983 struct netdev_adjacent *iter;
5985 struct net *net = dev_net(dev);
5987 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5988 if (!net_eq(net, dev_net(iter->dev)))
5990 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5991 &iter->dev->adj_list.lower);
5992 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5993 &dev->adj_list.upper);
5996 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5997 if (!net_eq(net, dev_net(iter->dev)))
5999 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6000 &iter->dev->adj_list.upper);
6001 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6002 &dev->adj_list.lower);
6006 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
6008 struct netdev_adjacent *iter;
6010 struct net *net = dev_net(dev);
6012 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6013 if (!net_eq(net, dev_net(iter->dev)))
6015 netdev_adjacent_sysfs_del(iter->dev, oldname,
6016 &iter->dev->adj_list.lower);
6017 netdev_adjacent_sysfs_add(iter->dev, dev,
6018 &iter->dev->adj_list.lower);
6021 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6022 if (!net_eq(net, dev_net(iter->dev)))
6024 netdev_adjacent_sysfs_del(iter->dev, oldname,
6025 &iter->dev->adj_list.upper);
6026 netdev_adjacent_sysfs_add(iter->dev, dev,
6027 &iter->dev->adj_list.upper);
6031 void *netdev_lower_dev_get_private(struct net_device *dev,
6032 struct net_device *lower_dev)
6034 struct netdev_adjacent *lower;
6038 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
6042 return lower->private;
6044 EXPORT_SYMBOL(netdev_lower_dev_get_private);
6047 int dev_get_nest_level(struct net_device *dev,
6048 bool (*type_check)(const struct net_device *dev))
6050 struct net_device *lower = NULL;
6051 struct list_head *iter;
6057 netdev_for_each_lower_dev(dev, lower, iter) {
6058 nest = dev_get_nest_level(lower, type_check);
6059 if (max_nest < nest)
6063 if (type_check(dev))
6068 EXPORT_SYMBOL(dev_get_nest_level);
6071 * netdev_lower_change - Dispatch event about lower device state change
6072 * @lower_dev: device
6073 * @lower_state_info: state to dispatch
6075 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
6076 * The caller must hold the RTNL lock.
6078 void netdev_lower_state_changed(struct net_device *lower_dev,
6079 void *lower_state_info)
6081 struct netdev_notifier_changelowerstate_info changelowerstate_info;
6084 changelowerstate_info.lower_state_info = lower_state_info;
6085 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev,
6086 &changelowerstate_info.info);
6088 EXPORT_SYMBOL(netdev_lower_state_changed);
6090 int netdev_default_l2upper_neigh_construct(struct net_device *dev,
6091 struct neighbour *n)
6093 struct net_device *lower_dev, *stop_dev;
6094 struct list_head *iter;
6097 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6098 if (!lower_dev->netdev_ops->ndo_neigh_construct)
6100 err = lower_dev->netdev_ops->ndo_neigh_construct(lower_dev, n);
6102 stop_dev = lower_dev;
6109 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6110 if (lower_dev == stop_dev)
6112 if (!lower_dev->netdev_ops->ndo_neigh_destroy)
6114 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n);
6118 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_construct);
6120 void netdev_default_l2upper_neigh_destroy(struct net_device *dev,
6121 struct neighbour *n)
6123 struct net_device *lower_dev;
6124 struct list_head *iter;
6126 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6127 if (!lower_dev->netdev_ops->ndo_neigh_destroy)
6129 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n);
6132 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_destroy);
6134 static void dev_change_rx_flags(struct net_device *dev, int flags)
6136 const struct net_device_ops *ops = dev->netdev_ops;
6138 if (ops->ndo_change_rx_flags)
6139 ops->ndo_change_rx_flags(dev, flags);
6142 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
6144 unsigned int old_flags = dev->flags;
6150 dev->flags |= IFF_PROMISC;
6151 dev->promiscuity += inc;
6152 if (dev->promiscuity == 0) {
6155 * If inc causes overflow, untouch promisc and return error.
6158 dev->flags &= ~IFF_PROMISC;
6160 dev->promiscuity -= inc;
6161 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6166 if (dev->flags != old_flags) {
6167 pr_info("device %s %s promiscuous mode\n",
6169 dev->flags & IFF_PROMISC ? "entered" : "left");
6170 if (audit_enabled) {
6171 current_uid_gid(&uid, &gid);
6172 audit_log(current->audit_context, GFP_ATOMIC,
6173 AUDIT_ANOM_PROMISCUOUS,
6174 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6175 dev->name, (dev->flags & IFF_PROMISC),
6176 (old_flags & IFF_PROMISC),
6177 from_kuid(&init_user_ns, audit_get_loginuid(current)),
6178 from_kuid(&init_user_ns, uid),
6179 from_kgid(&init_user_ns, gid),
6180 audit_get_sessionid(current));
6183 dev_change_rx_flags(dev, IFF_PROMISC);
6186 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
6191 * dev_set_promiscuity - update promiscuity count on a device
6195 * Add or remove promiscuity from a device. While the count in the device
6196 * remains above zero the interface remains promiscuous. Once it hits zero
6197 * the device reverts back to normal filtering operation. A negative inc
6198 * value is used to drop promiscuity on the device.
6199 * Return 0 if successful or a negative errno code on error.
6201 int dev_set_promiscuity(struct net_device *dev, int inc)
6203 unsigned int old_flags = dev->flags;
6206 err = __dev_set_promiscuity(dev, inc, true);
6209 if (dev->flags != old_flags)
6210 dev_set_rx_mode(dev);
6213 EXPORT_SYMBOL(dev_set_promiscuity);
6215 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
6217 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
6221 dev->flags |= IFF_ALLMULTI;
6222 dev->allmulti += inc;
6223 if (dev->allmulti == 0) {
6226 * If inc causes overflow, untouch allmulti and return error.
6229 dev->flags &= ~IFF_ALLMULTI;
6231 dev->allmulti -= inc;
6232 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6237 if (dev->flags ^ old_flags) {
6238 dev_change_rx_flags(dev, IFF_ALLMULTI);
6239 dev_set_rx_mode(dev);
6241 __dev_notify_flags(dev, old_flags,
6242 dev->gflags ^ old_gflags);
6248 * dev_set_allmulti - update allmulti count on a device
6252 * Add or remove reception of all multicast frames to a device. While the
6253 * count in the device remains above zero the interface remains listening
6254 * to all interfaces. Once it hits zero the device reverts back to normal
6255 * filtering operation. A negative @inc value is used to drop the counter
6256 * when releasing a resource needing all multicasts.
6257 * Return 0 if successful or a negative errno code on error.
6260 int dev_set_allmulti(struct net_device *dev, int inc)
6262 return __dev_set_allmulti(dev, inc, true);
6264 EXPORT_SYMBOL(dev_set_allmulti);
6267 * Upload unicast and multicast address lists to device and
6268 * configure RX filtering. When the device doesn't support unicast
6269 * filtering it is put in promiscuous mode while unicast addresses
6272 void __dev_set_rx_mode(struct net_device *dev)
6274 const struct net_device_ops *ops = dev->netdev_ops;
6276 /* dev_open will call this function so the list will stay sane. */
6277 if (!(dev->flags&IFF_UP))
6280 if (!netif_device_present(dev))
6283 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
6284 /* Unicast addresses changes may only happen under the rtnl,
6285 * therefore calling __dev_set_promiscuity here is safe.
6287 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
6288 __dev_set_promiscuity(dev, 1, false);
6289 dev->uc_promisc = true;
6290 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
6291 __dev_set_promiscuity(dev, -1, false);
6292 dev->uc_promisc = false;
6296 if (ops->ndo_set_rx_mode)
6297 ops->ndo_set_rx_mode(dev);
6300 void dev_set_rx_mode(struct net_device *dev)
6302 netif_addr_lock_bh(dev);
6303 __dev_set_rx_mode(dev);
6304 netif_addr_unlock_bh(dev);
6308 * dev_get_flags - get flags reported to userspace
6311 * Get the combination of flag bits exported through APIs to userspace.
6313 unsigned int dev_get_flags(const struct net_device *dev)
6317 flags = (dev->flags & ~(IFF_PROMISC |
6322 (dev->gflags & (IFF_PROMISC |
6325 if (netif_running(dev)) {
6326 if (netif_oper_up(dev))
6327 flags |= IFF_RUNNING;
6328 if (netif_carrier_ok(dev))
6329 flags |= IFF_LOWER_UP;
6330 if (netif_dormant(dev))
6331 flags |= IFF_DORMANT;
6336 EXPORT_SYMBOL(dev_get_flags);
6338 int __dev_change_flags(struct net_device *dev, unsigned int flags)
6340 unsigned int old_flags = dev->flags;
6346 * Set the flags on our device.
6349 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
6350 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
6352 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6356 * Load in the correct multicast list now the flags have changed.
6359 if ((old_flags ^ flags) & IFF_MULTICAST)
6360 dev_change_rx_flags(dev, IFF_MULTICAST);
6362 dev_set_rx_mode(dev);
6365 * Have we downed the interface. We handle IFF_UP ourselves
6366 * according to user attempts to set it, rather than blindly
6371 if ((old_flags ^ flags) & IFF_UP)
6372 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
6374 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6375 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6376 unsigned int old_flags = dev->flags;
6378 dev->gflags ^= IFF_PROMISC;
6380 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6381 if (dev->flags != old_flags)
6382 dev_set_rx_mode(dev);
6385 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6386 is important. Some (broken) drivers set IFF_PROMISC, when
6387 IFF_ALLMULTI is requested not asking us and not reporting.
6389 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6390 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6392 dev->gflags ^= IFF_ALLMULTI;
6393 __dev_set_allmulti(dev, inc, false);
6399 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6400 unsigned int gchanges)
6402 unsigned int changes = dev->flags ^ old_flags;
6405 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6407 if (changes & IFF_UP) {
6408 if (dev->flags & IFF_UP)
6409 call_netdevice_notifiers(NETDEV_UP, dev);
6411 call_netdevice_notifiers(NETDEV_DOWN, dev);
6414 if (dev->flags & IFF_UP &&
6415 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6416 struct netdev_notifier_change_info change_info;
6418 change_info.flags_changed = changes;
6419 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6425 * dev_change_flags - change device settings
6427 * @flags: device state flags
6429 * Change settings on device based state flags. The flags are
6430 * in the userspace exported format.
6432 int dev_change_flags(struct net_device *dev, unsigned int flags)
6435 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6437 ret = __dev_change_flags(dev, flags);
6441 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6442 __dev_notify_flags(dev, old_flags, changes);
6445 EXPORT_SYMBOL(dev_change_flags);
6447 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6449 const struct net_device_ops *ops = dev->netdev_ops;
6451 if (ops->ndo_change_mtu)
6452 return ops->ndo_change_mtu(dev, new_mtu);
6459 * dev_set_mtu - Change maximum transfer unit
6461 * @new_mtu: new transfer unit
6463 * Change the maximum transfer size of the network device.
6465 int dev_set_mtu(struct net_device *dev, int new_mtu)
6469 if (new_mtu == dev->mtu)
6472 /* MTU must be positive. */
6476 if (!netif_device_present(dev))
6479 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6480 err = notifier_to_errno(err);
6484 orig_mtu = dev->mtu;
6485 err = __dev_set_mtu(dev, new_mtu);
6488 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6489 err = notifier_to_errno(err);
6491 /* setting mtu back and notifying everyone again,
6492 * so that they have a chance to revert changes.
6494 __dev_set_mtu(dev, orig_mtu);
6495 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6500 EXPORT_SYMBOL(dev_set_mtu);
6503 * dev_set_group - Change group this device belongs to
6505 * @new_group: group this device should belong to
6507 void dev_set_group(struct net_device *dev, int new_group)
6509 dev->group = new_group;
6511 EXPORT_SYMBOL(dev_set_group);
6514 * dev_set_mac_address - Change Media Access Control Address
6518 * Change the hardware (MAC) address of the device
6520 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6522 const struct net_device_ops *ops = dev->netdev_ops;
6525 if (!ops->ndo_set_mac_address)
6527 if (sa->sa_family != dev->type)
6529 if (!netif_device_present(dev))
6531 err = ops->ndo_set_mac_address(dev, sa);
6534 dev->addr_assign_type = NET_ADDR_SET;
6535 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6536 add_device_randomness(dev->dev_addr, dev->addr_len);
6539 EXPORT_SYMBOL(dev_set_mac_address);
6542 * dev_change_carrier - Change device carrier
6544 * @new_carrier: new value
6546 * Change device carrier
6548 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6550 const struct net_device_ops *ops = dev->netdev_ops;
6552 if (!ops->ndo_change_carrier)
6554 if (!netif_device_present(dev))
6556 return ops->ndo_change_carrier(dev, new_carrier);
6558 EXPORT_SYMBOL(dev_change_carrier);
6561 * dev_get_phys_port_id - Get device physical port ID
6565 * Get device physical port ID
6567 int dev_get_phys_port_id(struct net_device *dev,
6568 struct netdev_phys_item_id *ppid)
6570 const struct net_device_ops *ops = dev->netdev_ops;
6572 if (!ops->ndo_get_phys_port_id)
6574 return ops->ndo_get_phys_port_id(dev, ppid);
6576 EXPORT_SYMBOL(dev_get_phys_port_id);
6579 * dev_get_phys_port_name - Get device physical port name
6582 * @len: limit of bytes to copy to name
6584 * Get device physical port name
6586 int dev_get_phys_port_name(struct net_device *dev,
6587 char *name, size_t len)
6589 const struct net_device_ops *ops = dev->netdev_ops;
6591 if (!ops->ndo_get_phys_port_name)
6593 return ops->ndo_get_phys_port_name(dev, name, len);
6595 EXPORT_SYMBOL(dev_get_phys_port_name);
6598 * dev_change_proto_down - update protocol port state information
6600 * @proto_down: new value
6602 * This info can be used by switch drivers to set the phys state of the
6605 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6607 const struct net_device_ops *ops = dev->netdev_ops;
6609 if (!ops->ndo_change_proto_down)
6611 if (!netif_device_present(dev))
6613 return ops->ndo_change_proto_down(dev, proto_down);
6615 EXPORT_SYMBOL(dev_change_proto_down);
6618 * dev_new_index - allocate an ifindex
6619 * @net: the applicable net namespace
6621 * Returns a suitable unique value for a new device interface
6622 * number. The caller must hold the rtnl semaphore or the
6623 * dev_base_lock to be sure it remains unique.
6625 static int dev_new_index(struct net *net)
6627 int ifindex = net->ifindex;
6631 if (!__dev_get_by_index(net, ifindex))
6632 return net->ifindex = ifindex;
6636 /* Delayed registration/unregisteration */
6637 static LIST_HEAD(net_todo_list);
6638 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6640 static void net_set_todo(struct net_device *dev)
6642 list_add_tail(&dev->todo_list, &net_todo_list);
6643 dev_net(dev)->dev_unreg_count++;
6646 static void rollback_registered_many(struct list_head *head)
6648 struct net_device *dev, *tmp;
6649 LIST_HEAD(close_head);
6651 BUG_ON(dev_boot_phase);
6654 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6655 /* Some devices call without registering
6656 * for initialization unwind. Remove those
6657 * devices and proceed with the remaining.
6659 if (dev->reg_state == NETREG_UNINITIALIZED) {
6660 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6664 list_del(&dev->unreg_list);
6667 dev->dismantle = true;
6668 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6671 /* If device is running, close it first. */
6672 list_for_each_entry(dev, head, unreg_list)
6673 list_add_tail(&dev->close_list, &close_head);
6674 dev_close_many(&close_head, true);
6676 list_for_each_entry(dev, head, unreg_list) {
6677 /* And unlink it from device chain. */
6678 unlist_netdevice(dev);
6680 dev->reg_state = NETREG_UNREGISTERING;
6681 on_each_cpu(flush_backlog, dev, 1);
6686 list_for_each_entry(dev, head, unreg_list) {
6687 struct sk_buff *skb = NULL;
6689 /* Shutdown queueing discipline. */
6693 /* Notify protocols, that we are about to destroy
6694 this device. They should clean all the things.
6696 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6698 if (!dev->rtnl_link_ops ||
6699 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6700 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6704 * Flush the unicast and multicast chains
6709 if (dev->netdev_ops->ndo_uninit)
6710 dev->netdev_ops->ndo_uninit(dev);
6713 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6715 /* Notifier chain MUST detach us all upper devices. */
6716 WARN_ON(netdev_has_any_upper_dev(dev));
6718 /* Remove entries from kobject tree */
6719 netdev_unregister_kobject(dev);
6721 /* Remove XPS queueing entries */
6722 netif_reset_xps_queues_gt(dev, 0);
6728 list_for_each_entry(dev, head, unreg_list)
6732 static void rollback_registered(struct net_device *dev)
6736 list_add(&dev->unreg_list, &single);
6737 rollback_registered_many(&single);
6741 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6742 struct net_device *upper, netdev_features_t features)
6744 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6745 netdev_features_t feature;
6748 for_each_netdev_feature(&upper_disables, feature_bit) {
6749 feature = __NETIF_F_BIT(feature_bit);
6750 if (!(upper->wanted_features & feature)
6751 && (features & feature)) {
6752 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6753 &feature, upper->name);
6754 features &= ~feature;
6761 static void netdev_sync_lower_features(struct net_device *upper,
6762 struct net_device *lower, netdev_features_t features)
6764 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6765 netdev_features_t feature;
6768 for_each_netdev_feature(&upper_disables, feature_bit) {
6769 feature = __NETIF_F_BIT(feature_bit);
6770 if (!(features & feature) && (lower->features & feature)) {
6771 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6772 &feature, lower->name);
6773 lower->wanted_features &= ~feature;
6774 netdev_update_features(lower);
6776 if (unlikely(lower->features & feature))
6777 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6778 &feature, lower->name);
6783 static netdev_features_t netdev_fix_features(struct net_device *dev,
6784 netdev_features_t features)
6786 /* Fix illegal checksum combinations */
6787 if ((features & NETIF_F_HW_CSUM) &&
6788 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6789 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6790 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6793 /* TSO requires that SG is present as well. */
6794 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6795 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6796 features &= ~NETIF_F_ALL_TSO;
6799 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6800 !(features & NETIF_F_IP_CSUM)) {
6801 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6802 features &= ~NETIF_F_TSO;
6803 features &= ~NETIF_F_TSO_ECN;
6806 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6807 !(features & NETIF_F_IPV6_CSUM)) {
6808 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6809 features &= ~NETIF_F_TSO6;
6812 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
6813 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
6814 features &= ~NETIF_F_TSO_MANGLEID;
6816 /* TSO ECN requires that TSO is present as well. */
6817 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6818 features &= ~NETIF_F_TSO_ECN;
6820 /* Software GSO depends on SG. */
6821 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6822 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6823 features &= ~NETIF_F_GSO;
6826 /* UFO needs SG and checksumming */
6827 if (features & NETIF_F_UFO) {
6828 /* maybe split UFO into V4 and V6? */
6829 if (!(features & NETIF_F_HW_CSUM) &&
6830 ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) !=
6831 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) {
6833 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6834 features &= ~NETIF_F_UFO;
6837 if (!(features & NETIF_F_SG)) {
6839 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6840 features &= ~NETIF_F_UFO;
6844 /* GSO partial features require GSO partial be set */
6845 if ((features & dev->gso_partial_features) &&
6846 !(features & NETIF_F_GSO_PARTIAL)) {
6848 "Dropping partially supported GSO features since no GSO partial.\n");
6849 features &= ~dev->gso_partial_features;
6852 #ifdef CONFIG_NET_RX_BUSY_POLL
6853 if (dev->netdev_ops->ndo_busy_poll)
6854 features |= NETIF_F_BUSY_POLL;
6857 features &= ~NETIF_F_BUSY_POLL;
6862 int __netdev_update_features(struct net_device *dev)
6864 struct net_device *upper, *lower;
6865 netdev_features_t features;
6866 struct list_head *iter;
6871 features = netdev_get_wanted_features(dev);
6873 if (dev->netdev_ops->ndo_fix_features)
6874 features = dev->netdev_ops->ndo_fix_features(dev, features);
6876 /* driver might be less strict about feature dependencies */
6877 features = netdev_fix_features(dev, features);
6879 /* some features can't be enabled if they're off an an upper device */
6880 netdev_for_each_upper_dev_rcu(dev, upper, iter)
6881 features = netdev_sync_upper_features(dev, upper, features);
6883 if (dev->features == features)
6886 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6887 &dev->features, &features);
6889 if (dev->netdev_ops->ndo_set_features)
6890 err = dev->netdev_ops->ndo_set_features(dev, features);
6894 if (unlikely(err < 0)) {
6896 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6897 err, &features, &dev->features);
6898 /* return non-0 since some features might have changed and
6899 * it's better to fire a spurious notification than miss it
6905 /* some features must be disabled on lower devices when disabled
6906 * on an upper device (think: bonding master or bridge)
6908 netdev_for_each_lower_dev(dev, lower, iter)
6909 netdev_sync_lower_features(dev, lower, features);
6912 dev->features = features;
6914 return err < 0 ? 0 : 1;
6918 * netdev_update_features - recalculate device features
6919 * @dev: the device to check
6921 * Recalculate dev->features set and send notifications if it
6922 * has changed. Should be called after driver or hardware dependent
6923 * conditions might have changed that influence the features.
6925 void netdev_update_features(struct net_device *dev)
6927 if (__netdev_update_features(dev))
6928 netdev_features_change(dev);
6930 EXPORT_SYMBOL(netdev_update_features);
6933 * netdev_change_features - recalculate device features
6934 * @dev: the device to check
6936 * Recalculate dev->features set and send notifications even
6937 * if they have not changed. Should be called instead of
6938 * netdev_update_features() if also dev->vlan_features might
6939 * have changed to allow the changes to be propagated to stacked
6942 void netdev_change_features(struct net_device *dev)
6944 __netdev_update_features(dev);
6945 netdev_features_change(dev);
6947 EXPORT_SYMBOL(netdev_change_features);
6950 * netif_stacked_transfer_operstate - transfer operstate
6951 * @rootdev: the root or lower level device to transfer state from
6952 * @dev: the device to transfer operstate to
6954 * Transfer operational state from root to device. This is normally
6955 * called when a stacking relationship exists between the root
6956 * device and the device(a leaf device).
6958 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6959 struct net_device *dev)
6961 if (rootdev->operstate == IF_OPER_DORMANT)
6962 netif_dormant_on(dev);
6964 netif_dormant_off(dev);
6966 if (netif_carrier_ok(rootdev)) {
6967 if (!netif_carrier_ok(dev))
6968 netif_carrier_on(dev);
6970 if (netif_carrier_ok(dev))
6971 netif_carrier_off(dev);
6974 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6977 static int netif_alloc_rx_queues(struct net_device *dev)
6979 unsigned int i, count = dev->num_rx_queues;
6980 struct netdev_rx_queue *rx;
6981 size_t sz = count * sizeof(*rx);
6985 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6993 for (i = 0; i < count; i++)
6999 static void netdev_init_one_queue(struct net_device *dev,
7000 struct netdev_queue *queue, void *_unused)
7002 /* Initialize queue lock */
7003 spin_lock_init(&queue->_xmit_lock);
7004 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
7005 queue->xmit_lock_owner = -1;
7006 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
7009 dql_init(&queue->dql, HZ);
7013 static void netif_free_tx_queues(struct net_device *dev)
7018 static int netif_alloc_netdev_queues(struct net_device *dev)
7020 unsigned int count = dev->num_tx_queues;
7021 struct netdev_queue *tx;
7022 size_t sz = count * sizeof(*tx);
7024 if (count < 1 || count > 0xffff)
7027 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7035 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
7036 spin_lock_init(&dev->tx_global_lock);
7041 void netif_tx_stop_all_queues(struct net_device *dev)
7045 for (i = 0; i < dev->num_tx_queues; i++) {
7046 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
7047 netif_tx_stop_queue(txq);
7050 EXPORT_SYMBOL(netif_tx_stop_all_queues);
7053 * register_netdevice - register a network device
7054 * @dev: device to register
7056 * Take a completed network device structure and add it to the kernel
7057 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7058 * chain. 0 is returned on success. A negative errno code is returned
7059 * on a failure to set up the device, or if the name is a duplicate.
7061 * Callers must hold the rtnl semaphore. You may want
7062 * register_netdev() instead of this.
7065 * The locking appears insufficient to guarantee two parallel registers
7066 * will not get the same name.
7069 int register_netdevice(struct net_device *dev)
7072 struct net *net = dev_net(dev);
7074 BUG_ON(dev_boot_phase);
7079 /* When net_device's are persistent, this will be fatal. */
7080 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
7083 spin_lock_init(&dev->addr_list_lock);
7084 netdev_set_addr_lockdep_class(dev);
7086 ret = dev_get_valid_name(net, dev, dev->name);
7090 /* Init, if this function is available */
7091 if (dev->netdev_ops->ndo_init) {
7092 ret = dev->netdev_ops->ndo_init(dev);
7100 if (((dev->hw_features | dev->features) &
7101 NETIF_F_HW_VLAN_CTAG_FILTER) &&
7102 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
7103 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
7104 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
7111 dev->ifindex = dev_new_index(net);
7112 else if (__dev_get_by_index(net, dev->ifindex))
7115 /* Transfer changeable features to wanted_features and enable
7116 * software offloads (GSO and GRO).
7118 dev->hw_features |= NETIF_F_SOFT_FEATURES;
7119 dev->features |= NETIF_F_SOFT_FEATURES;
7120 dev->wanted_features = dev->features & dev->hw_features;
7122 if (!(dev->flags & IFF_LOOPBACK))
7123 dev->hw_features |= NETIF_F_NOCACHE_COPY;
7125 /* If IPv4 TCP segmentation offload is supported we should also
7126 * allow the device to enable segmenting the frame with the option
7127 * of ignoring a static IP ID value. This doesn't enable the
7128 * feature itself but allows the user to enable it later.
7130 if (dev->hw_features & NETIF_F_TSO)
7131 dev->hw_features |= NETIF_F_TSO_MANGLEID;
7132 if (dev->vlan_features & NETIF_F_TSO)
7133 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
7134 if (dev->mpls_features & NETIF_F_TSO)
7135 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
7136 if (dev->hw_enc_features & NETIF_F_TSO)
7137 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
7139 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
7141 dev->vlan_features |= NETIF_F_HIGHDMA;
7143 /* Make NETIF_F_SG inheritable to tunnel devices.
7145 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
7147 /* Make NETIF_F_SG inheritable to MPLS.
7149 dev->mpls_features |= NETIF_F_SG;
7151 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
7152 ret = notifier_to_errno(ret);
7156 ret = netdev_register_kobject(dev);
7159 dev->reg_state = NETREG_REGISTERED;
7161 __netdev_update_features(dev);
7164 * Default initial state at registry is that the
7165 * device is present.
7168 set_bit(__LINK_STATE_PRESENT, &dev->state);
7170 linkwatch_init_dev(dev);
7172 dev_init_scheduler(dev);
7174 list_netdevice(dev);
7175 add_device_randomness(dev->dev_addr, dev->addr_len);
7177 /* If the device has permanent device address, driver should
7178 * set dev_addr and also addr_assign_type should be set to
7179 * NET_ADDR_PERM (default value).
7181 if (dev->addr_assign_type == NET_ADDR_PERM)
7182 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
7184 /* Notify protocols, that a new device appeared. */
7185 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
7186 ret = notifier_to_errno(ret);
7188 rollback_registered(dev);
7189 dev->reg_state = NETREG_UNREGISTERED;
7192 * Prevent userspace races by waiting until the network
7193 * device is fully setup before sending notifications.
7195 if (!dev->rtnl_link_ops ||
7196 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7197 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7203 if (dev->netdev_ops->ndo_uninit)
7204 dev->netdev_ops->ndo_uninit(dev);
7207 EXPORT_SYMBOL(register_netdevice);
7210 * init_dummy_netdev - init a dummy network device for NAPI
7211 * @dev: device to init
7213 * This takes a network device structure and initialize the minimum
7214 * amount of fields so it can be used to schedule NAPI polls without
7215 * registering a full blown interface. This is to be used by drivers
7216 * that need to tie several hardware interfaces to a single NAPI
7217 * poll scheduler due to HW limitations.
7219 int init_dummy_netdev(struct net_device *dev)
7221 /* Clear everything. Note we don't initialize spinlocks
7222 * are they aren't supposed to be taken by any of the
7223 * NAPI code and this dummy netdev is supposed to be
7224 * only ever used for NAPI polls
7226 memset(dev, 0, sizeof(struct net_device));
7228 /* make sure we BUG if trying to hit standard
7229 * register/unregister code path
7231 dev->reg_state = NETREG_DUMMY;
7233 /* NAPI wants this */
7234 INIT_LIST_HEAD(&dev->napi_list);
7236 /* a dummy interface is started by default */
7237 set_bit(__LINK_STATE_PRESENT, &dev->state);
7238 set_bit(__LINK_STATE_START, &dev->state);
7240 /* Note : We dont allocate pcpu_refcnt for dummy devices,
7241 * because users of this 'device' dont need to change
7247 EXPORT_SYMBOL_GPL(init_dummy_netdev);
7251 * register_netdev - register a network device
7252 * @dev: device to register
7254 * Take a completed network device structure and add it to the kernel
7255 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7256 * chain. 0 is returned on success. A negative errno code is returned
7257 * on a failure to set up the device, or if the name is a duplicate.
7259 * This is a wrapper around register_netdevice that takes the rtnl semaphore
7260 * and expands the device name if you passed a format string to
7263 int register_netdev(struct net_device *dev)
7268 err = register_netdevice(dev);
7272 EXPORT_SYMBOL(register_netdev);
7274 int netdev_refcnt_read(const struct net_device *dev)
7278 for_each_possible_cpu(i)
7279 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
7282 EXPORT_SYMBOL(netdev_refcnt_read);
7285 * netdev_wait_allrefs - wait until all references are gone.
7286 * @dev: target net_device
7288 * This is called when unregistering network devices.
7290 * Any protocol or device that holds a reference should register
7291 * for netdevice notification, and cleanup and put back the
7292 * reference if they receive an UNREGISTER event.
7293 * We can get stuck here if buggy protocols don't correctly
7296 static void netdev_wait_allrefs(struct net_device *dev)
7298 unsigned long rebroadcast_time, warning_time;
7301 linkwatch_forget_dev(dev);
7303 rebroadcast_time = warning_time = jiffies;
7304 refcnt = netdev_refcnt_read(dev);
7306 while (refcnt != 0) {
7307 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
7310 /* Rebroadcast unregister notification */
7311 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7317 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7318 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
7320 /* We must not have linkwatch events
7321 * pending on unregister. If this
7322 * happens, we simply run the queue
7323 * unscheduled, resulting in a noop
7326 linkwatch_run_queue();
7331 rebroadcast_time = jiffies;
7336 refcnt = netdev_refcnt_read(dev);
7338 if (time_after(jiffies, warning_time + 10 * HZ)) {
7339 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
7341 warning_time = jiffies;
7350 * register_netdevice(x1);
7351 * register_netdevice(x2);
7353 * unregister_netdevice(y1);
7354 * unregister_netdevice(y2);
7360 * We are invoked by rtnl_unlock().
7361 * This allows us to deal with problems:
7362 * 1) We can delete sysfs objects which invoke hotplug
7363 * without deadlocking with linkwatch via keventd.
7364 * 2) Since we run with the RTNL semaphore not held, we can sleep
7365 * safely in order to wait for the netdev refcnt to drop to zero.
7367 * We must not return until all unregister events added during
7368 * the interval the lock was held have been completed.
7370 void netdev_run_todo(void)
7372 struct list_head list;
7374 /* Snapshot list, allow later requests */
7375 list_replace_init(&net_todo_list, &list);
7380 /* Wait for rcu callbacks to finish before next phase */
7381 if (!list_empty(&list))
7384 while (!list_empty(&list)) {
7385 struct net_device *dev
7386 = list_first_entry(&list, struct net_device, todo_list);
7387 list_del(&dev->todo_list);
7390 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7393 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
7394 pr_err("network todo '%s' but state %d\n",
7395 dev->name, dev->reg_state);
7400 dev->reg_state = NETREG_UNREGISTERED;
7402 netdev_wait_allrefs(dev);
7405 BUG_ON(netdev_refcnt_read(dev));
7406 BUG_ON(!list_empty(&dev->ptype_all));
7407 BUG_ON(!list_empty(&dev->ptype_specific));
7408 WARN_ON(rcu_access_pointer(dev->ip_ptr));
7409 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
7410 WARN_ON(dev->dn_ptr);
7412 if (dev->destructor)
7413 dev->destructor(dev);
7415 /* Report a network device has been unregistered */
7417 dev_net(dev)->dev_unreg_count--;
7419 wake_up(&netdev_unregistering_wq);
7421 /* Free network device */
7422 kobject_put(&dev->dev.kobj);
7426 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
7427 * all the same fields in the same order as net_device_stats, with only
7428 * the type differing, but rtnl_link_stats64 may have additional fields
7429 * at the end for newer counters.
7431 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7432 const struct net_device_stats *netdev_stats)
7434 #if BITS_PER_LONG == 64
7435 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
7436 memcpy(stats64, netdev_stats, sizeof(*stats64));
7437 /* zero out counters that only exist in rtnl_link_stats64 */
7438 memset((char *)stats64 + sizeof(*netdev_stats), 0,
7439 sizeof(*stats64) - sizeof(*netdev_stats));
7441 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
7442 const unsigned long *src = (const unsigned long *)netdev_stats;
7443 u64 *dst = (u64 *)stats64;
7445 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
7446 for (i = 0; i < n; i++)
7448 /* zero out counters that only exist in rtnl_link_stats64 */
7449 memset((char *)stats64 + n * sizeof(u64), 0,
7450 sizeof(*stats64) - n * sizeof(u64));
7453 EXPORT_SYMBOL(netdev_stats_to_stats64);
7456 * dev_get_stats - get network device statistics
7457 * @dev: device to get statistics from
7458 * @storage: place to store stats
7460 * Get network statistics from device. Return @storage.
7461 * The device driver may provide its own method by setting
7462 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7463 * otherwise the internal statistics structure is used.
7465 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
7466 struct rtnl_link_stats64 *storage)
7468 const struct net_device_ops *ops = dev->netdev_ops;
7470 if (ops->ndo_get_stats64) {
7471 memset(storage, 0, sizeof(*storage));
7472 ops->ndo_get_stats64(dev, storage);
7473 } else if (ops->ndo_get_stats) {
7474 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7476 netdev_stats_to_stats64(storage, &dev->stats);
7478 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
7479 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
7480 storage->rx_nohandler += atomic_long_read(&dev->rx_nohandler);
7483 EXPORT_SYMBOL(dev_get_stats);
7485 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7487 struct netdev_queue *queue = dev_ingress_queue(dev);
7489 #ifdef CONFIG_NET_CLS_ACT
7492 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7495 netdev_init_one_queue(dev, queue, NULL);
7496 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7497 queue->qdisc_sleeping = &noop_qdisc;
7498 rcu_assign_pointer(dev->ingress_queue, queue);
7503 static const struct ethtool_ops default_ethtool_ops;
7505 void netdev_set_default_ethtool_ops(struct net_device *dev,
7506 const struct ethtool_ops *ops)
7508 if (dev->ethtool_ops == &default_ethtool_ops)
7509 dev->ethtool_ops = ops;
7511 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7513 void netdev_freemem(struct net_device *dev)
7515 char *addr = (char *)dev - dev->padded;
7521 * alloc_netdev_mqs - allocate network device
7522 * @sizeof_priv: size of private data to allocate space for
7523 * @name: device name format string
7524 * @name_assign_type: origin of device name
7525 * @setup: callback to initialize device
7526 * @txqs: the number of TX subqueues to allocate
7527 * @rxqs: the number of RX subqueues to allocate
7529 * Allocates a struct net_device with private data area for driver use
7530 * and performs basic initialization. Also allocates subqueue structs
7531 * for each queue on the device.
7533 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7534 unsigned char name_assign_type,
7535 void (*setup)(struct net_device *),
7536 unsigned int txqs, unsigned int rxqs)
7538 struct net_device *dev;
7540 struct net_device *p;
7542 BUG_ON(strlen(name) >= sizeof(dev->name));
7545 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7551 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7556 alloc_size = sizeof(struct net_device);
7558 /* ensure 32-byte alignment of private area */
7559 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7560 alloc_size += sizeof_priv;
7562 /* ensure 32-byte alignment of whole construct */
7563 alloc_size += NETDEV_ALIGN - 1;
7565 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7567 p = vzalloc(alloc_size);
7571 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7572 dev->padded = (char *)dev - (char *)p;
7574 dev->pcpu_refcnt = alloc_percpu(int);
7575 if (!dev->pcpu_refcnt)
7578 if (dev_addr_init(dev))
7584 dev_net_set(dev, &init_net);
7586 dev->gso_max_size = GSO_MAX_SIZE;
7587 dev->gso_max_segs = GSO_MAX_SEGS;
7589 INIT_LIST_HEAD(&dev->napi_list);
7590 INIT_LIST_HEAD(&dev->unreg_list);
7591 INIT_LIST_HEAD(&dev->close_list);
7592 INIT_LIST_HEAD(&dev->link_watch_list);
7593 INIT_LIST_HEAD(&dev->adj_list.upper);
7594 INIT_LIST_HEAD(&dev->adj_list.lower);
7595 INIT_LIST_HEAD(&dev->all_adj_list.upper);
7596 INIT_LIST_HEAD(&dev->all_adj_list.lower);
7597 INIT_LIST_HEAD(&dev->ptype_all);
7598 INIT_LIST_HEAD(&dev->ptype_specific);
7599 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7602 if (!dev->tx_queue_len) {
7603 dev->priv_flags |= IFF_NO_QUEUE;
7604 dev->tx_queue_len = 1;
7607 dev->num_tx_queues = txqs;
7608 dev->real_num_tx_queues = txqs;
7609 if (netif_alloc_netdev_queues(dev))
7613 dev->num_rx_queues = rxqs;
7614 dev->real_num_rx_queues = rxqs;
7615 if (netif_alloc_rx_queues(dev))
7619 strcpy(dev->name, name);
7620 dev->name_assign_type = name_assign_type;
7621 dev->group = INIT_NETDEV_GROUP;
7622 if (!dev->ethtool_ops)
7623 dev->ethtool_ops = &default_ethtool_ops;
7625 nf_hook_ingress_init(dev);
7634 free_percpu(dev->pcpu_refcnt);
7636 netdev_freemem(dev);
7639 EXPORT_SYMBOL(alloc_netdev_mqs);
7642 * free_netdev - free network device
7645 * This function does the last stage of destroying an allocated device
7646 * interface. The reference to the device object is released.
7647 * If this is the last reference then it will be freed.
7648 * Must be called in process context.
7650 void free_netdev(struct net_device *dev)
7652 struct napi_struct *p, *n;
7655 netif_free_tx_queues(dev);
7660 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7662 /* Flush device addresses */
7663 dev_addr_flush(dev);
7665 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7668 free_percpu(dev->pcpu_refcnt);
7669 dev->pcpu_refcnt = NULL;
7671 /* Compatibility with error handling in drivers */
7672 if (dev->reg_state == NETREG_UNINITIALIZED) {
7673 netdev_freemem(dev);
7677 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7678 dev->reg_state = NETREG_RELEASED;
7680 /* will free via device release */
7681 put_device(&dev->dev);
7683 EXPORT_SYMBOL(free_netdev);
7686 * synchronize_net - Synchronize with packet receive processing
7688 * Wait for packets currently being received to be done.
7689 * Does not block later packets from starting.
7691 void synchronize_net(void)
7694 if (rtnl_is_locked())
7695 synchronize_rcu_expedited();
7699 EXPORT_SYMBOL(synchronize_net);
7702 * unregister_netdevice_queue - remove device from the kernel
7706 * This function shuts down a device interface and removes it
7707 * from the kernel tables.
7708 * If head not NULL, device is queued to be unregistered later.
7710 * Callers must hold the rtnl semaphore. You may want
7711 * unregister_netdev() instead of this.
7714 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7719 list_move_tail(&dev->unreg_list, head);
7721 rollback_registered(dev);
7722 /* Finish processing unregister after unlock */
7726 EXPORT_SYMBOL(unregister_netdevice_queue);
7729 * unregister_netdevice_many - unregister many devices
7730 * @head: list of devices
7732 * Note: As most callers use a stack allocated list_head,
7733 * we force a list_del() to make sure stack wont be corrupted later.
7735 void unregister_netdevice_many(struct list_head *head)
7737 struct net_device *dev;
7739 if (!list_empty(head)) {
7740 rollback_registered_many(head);
7741 list_for_each_entry(dev, head, unreg_list)
7746 EXPORT_SYMBOL(unregister_netdevice_many);
7749 * unregister_netdev - remove device from the kernel
7752 * This function shuts down a device interface and removes it
7753 * from the kernel tables.
7755 * This is just a wrapper for unregister_netdevice that takes
7756 * the rtnl semaphore. In general you want to use this and not
7757 * unregister_netdevice.
7759 void unregister_netdev(struct net_device *dev)
7762 unregister_netdevice(dev);
7765 EXPORT_SYMBOL(unregister_netdev);
7768 * dev_change_net_namespace - move device to different nethost namespace
7770 * @net: network namespace
7771 * @pat: If not NULL name pattern to try if the current device name
7772 * is already taken in the destination network namespace.
7774 * This function shuts down a device interface and moves it
7775 * to a new network namespace. On success 0 is returned, on
7776 * a failure a netagive errno code is returned.
7778 * Callers must hold the rtnl semaphore.
7781 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7787 /* Don't allow namespace local devices to be moved. */
7789 if (dev->features & NETIF_F_NETNS_LOCAL)
7792 /* Ensure the device has been registrered */
7793 if (dev->reg_state != NETREG_REGISTERED)
7796 /* Get out if there is nothing todo */
7798 if (net_eq(dev_net(dev), net))
7801 /* Pick the destination device name, and ensure
7802 * we can use it in the destination network namespace.
7805 if (__dev_get_by_name(net, dev->name)) {
7806 /* We get here if we can't use the current device name */
7809 if (dev_get_valid_name(net, dev, pat) < 0)
7814 * And now a mini version of register_netdevice unregister_netdevice.
7817 /* If device is running close it first. */
7820 /* And unlink it from device chain */
7822 unlist_netdevice(dev);
7826 /* Shutdown queueing discipline. */
7829 /* Notify protocols, that we are about to destroy
7830 this device. They should clean all the things.
7832 Note that dev->reg_state stays at NETREG_REGISTERED.
7833 This is wanted because this way 8021q and macvlan know
7834 the device is just moving and can keep their slaves up.
7836 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7838 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7839 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7842 * Flush the unicast and multicast chains
7847 /* Send a netdev-removed uevent to the old namespace */
7848 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7849 netdev_adjacent_del_links(dev);
7851 /* Actually switch the network namespace */
7852 dev_net_set(dev, net);
7854 /* If there is an ifindex conflict assign a new one */
7855 if (__dev_get_by_index(net, dev->ifindex))
7856 dev->ifindex = dev_new_index(net);
7858 /* Send a netdev-add uevent to the new namespace */
7859 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7860 netdev_adjacent_add_links(dev);
7862 /* Fixup kobjects */
7863 err = device_rename(&dev->dev, dev->name);
7866 /* Add the device back in the hashes */
7867 list_netdevice(dev);
7869 /* Notify protocols, that a new device appeared. */
7870 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7873 * Prevent userspace races by waiting until the network
7874 * device is fully setup before sending notifications.
7876 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7883 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7885 static int dev_cpu_callback(struct notifier_block *nfb,
7886 unsigned long action,
7889 struct sk_buff **list_skb;
7890 struct sk_buff *skb;
7891 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7892 struct softnet_data *sd, *oldsd;
7894 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7897 local_irq_disable();
7898 cpu = smp_processor_id();
7899 sd = &per_cpu(softnet_data, cpu);
7900 oldsd = &per_cpu(softnet_data, oldcpu);
7902 /* Find end of our completion_queue. */
7903 list_skb = &sd->completion_queue;
7905 list_skb = &(*list_skb)->next;
7906 /* Append completion queue from offline CPU. */
7907 *list_skb = oldsd->completion_queue;
7908 oldsd->completion_queue = NULL;
7910 /* Append output queue from offline CPU. */
7911 if (oldsd->output_queue) {
7912 *sd->output_queue_tailp = oldsd->output_queue;
7913 sd->output_queue_tailp = oldsd->output_queue_tailp;
7914 oldsd->output_queue = NULL;
7915 oldsd->output_queue_tailp = &oldsd->output_queue;
7917 /* Append NAPI poll list from offline CPU, with one exception :
7918 * process_backlog() must be called by cpu owning percpu backlog.
7919 * We properly handle process_queue & input_pkt_queue later.
7921 while (!list_empty(&oldsd->poll_list)) {
7922 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7926 list_del_init(&napi->poll_list);
7927 if (napi->poll == process_backlog)
7930 ____napi_schedule(sd, napi);
7933 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7936 /* Process offline CPU's input_pkt_queue */
7937 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7939 input_queue_head_incr(oldsd);
7941 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7943 input_queue_head_incr(oldsd);
7951 * netdev_increment_features - increment feature set by one
7952 * @all: current feature set
7953 * @one: new feature set
7954 * @mask: mask feature set
7956 * Computes a new feature set after adding a device with feature set
7957 * @one to the master device with current feature set @all. Will not
7958 * enable anything that is off in @mask. Returns the new feature set.
7960 netdev_features_t netdev_increment_features(netdev_features_t all,
7961 netdev_features_t one, netdev_features_t mask)
7963 if (mask & NETIF_F_HW_CSUM)
7964 mask |= NETIF_F_CSUM_MASK;
7965 mask |= NETIF_F_VLAN_CHALLENGED;
7967 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
7968 all &= one | ~NETIF_F_ALL_FOR_ALL;
7970 /* If one device supports hw checksumming, set for all. */
7971 if (all & NETIF_F_HW_CSUM)
7972 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
7976 EXPORT_SYMBOL(netdev_increment_features);
7978 static struct hlist_head * __net_init netdev_create_hash(void)
7981 struct hlist_head *hash;
7983 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7985 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7986 INIT_HLIST_HEAD(&hash[i]);
7991 /* Initialize per network namespace state */
7992 static int __net_init netdev_init(struct net *net)
7994 if (net != &init_net)
7995 INIT_LIST_HEAD(&net->dev_base_head);
7997 net->dev_name_head = netdev_create_hash();
7998 if (net->dev_name_head == NULL)
8001 net->dev_index_head = netdev_create_hash();
8002 if (net->dev_index_head == NULL)
8008 kfree(net->dev_name_head);
8014 * netdev_drivername - network driver for the device
8015 * @dev: network device
8017 * Determine network driver for device.
8019 const char *netdev_drivername(const struct net_device *dev)
8021 const struct device_driver *driver;
8022 const struct device *parent;
8023 const char *empty = "";
8025 parent = dev->dev.parent;
8029 driver = parent->driver;
8030 if (driver && driver->name)
8031 return driver->name;
8035 static void __netdev_printk(const char *level, const struct net_device *dev,
8036 struct va_format *vaf)
8038 if (dev && dev->dev.parent) {
8039 dev_printk_emit(level[1] - '0',
8042 dev_driver_string(dev->dev.parent),
8043 dev_name(dev->dev.parent),
8044 netdev_name(dev), netdev_reg_state(dev),
8047 printk("%s%s%s: %pV",
8048 level, netdev_name(dev), netdev_reg_state(dev), vaf);
8050 printk("%s(NULL net_device): %pV", level, vaf);
8054 void netdev_printk(const char *level, const struct net_device *dev,
8055 const char *format, ...)
8057 struct va_format vaf;
8060 va_start(args, format);
8065 __netdev_printk(level, dev, &vaf);
8069 EXPORT_SYMBOL(netdev_printk);
8071 #define define_netdev_printk_level(func, level) \
8072 void func(const struct net_device *dev, const char *fmt, ...) \
8074 struct va_format vaf; \
8077 va_start(args, fmt); \
8082 __netdev_printk(level, dev, &vaf); \
8086 EXPORT_SYMBOL(func);
8088 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
8089 define_netdev_printk_level(netdev_alert, KERN_ALERT);
8090 define_netdev_printk_level(netdev_crit, KERN_CRIT);
8091 define_netdev_printk_level(netdev_err, KERN_ERR);
8092 define_netdev_printk_level(netdev_warn, KERN_WARNING);
8093 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
8094 define_netdev_printk_level(netdev_info, KERN_INFO);
8096 static void __net_exit netdev_exit(struct net *net)
8098 kfree(net->dev_name_head);
8099 kfree(net->dev_index_head);
8102 static struct pernet_operations __net_initdata netdev_net_ops = {
8103 .init = netdev_init,
8104 .exit = netdev_exit,
8107 static void __net_exit default_device_exit(struct net *net)
8109 struct net_device *dev, *aux;
8111 * Push all migratable network devices back to the
8112 * initial network namespace
8115 for_each_netdev_safe(net, dev, aux) {
8117 char fb_name[IFNAMSIZ];
8119 /* Ignore unmoveable devices (i.e. loopback) */
8120 if (dev->features & NETIF_F_NETNS_LOCAL)
8123 /* Leave virtual devices for the generic cleanup */
8124 if (dev->rtnl_link_ops)
8127 /* Push remaining network devices to init_net */
8128 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
8129 err = dev_change_net_namespace(dev, &init_net, fb_name);
8131 pr_emerg("%s: failed to move %s to init_net: %d\n",
8132 __func__, dev->name, err);
8139 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
8141 /* Return with the rtnl_lock held when there are no network
8142 * devices unregistering in any network namespace in net_list.
8146 DEFINE_WAIT_FUNC(wait, woken_wake_function);
8148 add_wait_queue(&netdev_unregistering_wq, &wait);
8150 unregistering = false;
8152 list_for_each_entry(net, net_list, exit_list) {
8153 if (net->dev_unreg_count > 0) {
8154 unregistering = true;
8162 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
8164 remove_wait_queue(&netdev_unregistering_wq, &wait);
8167 static void __net_exit default_device_exit_batch(struct list_head *net_list)
8169 /* At exit all network devices most be removed from a network
8170 * namespace. Do this in the reverse order of registration.
8171 * Do this across as many network namespaces as possible to
8172 * improve batching efficiency.
8174 struct net_device *dev;
8176 LIST_HEAD(dev_kill_list);
8178 /* To prevent network device cleanup code from dereferencing
8179 * loopback devices or network devices that have been freed
8180 * wait here for all pending unregistrations to complete,
8181 * before unregistring the loopback device and allowing the
8182 * network namespace be freed.
8184 * The netdev todo list containing all network devices
8185 * unregistrations that happen in default_device_exit_batch
8186 * will run in the rtnl_unlock() at the end of
8187 * default_device_exit_batch.
8189 rtnl_lock_unregistering(net_list);
8190 list_for_each_entry(net, net_list, exit_list) {
8191 for_each_netdev_reverse(net, dev) {
8192 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
8193 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
8195 unregister_netdevice_queue(dev, &dev_kill_list);
8198 unregister_netdevice_many(&dev_kill_list);
8202 static struct pernet_operations __net_initdata default_device_ops = {
8203 .exit = default_device_exit,
8204 .exit_batch = default_device_exit_batch,
8208 * Initialize the DEV module. At boot time this walks the device list and
8209 * unhooks any devices that fail to initialise (normally hardware not
8210 * present) and leaves us with a valid list of present and active devices.
8215 * This is called single threaded during boot, so no need
8216 * to take the rtnl semaphore.
8218 static int __init net_dev_init(void)
8220 int i, rc = -ENOMEM;
8222 BUG_ON(!dev_boot_phase);
8224 if (dev_proc_init())
8227 if (netdev_kobject_init())
8230 INIT_LIST_HEAD(&ptype_all);
8231 for (i = 0; i < PTYPE_HASH_SIZE; i++)
8232 INIT_LIST_HEAD(&ptype_base[i]);
8234 INIT_LIST_HEAD(&offload_base);
8236 if (register_pernet_subsys(&netdev_net_ops))
8240 * Initialise the packet receive queues.
8243 for_each_possible_cpu(i) {
8244 struct softnet_data *sd = &per_cpu(softnet_data, i);
8246 skb_queue_head_init(&sd->input_pkt_queue);
8247 skb_queue_head_init(&sd->process_queue);
8248 INIT_LIST_HEAD(&sd->poll_list);
8249 sd->output_queue_tailp = &sd->output_queue;
8251 sd->csd.func = rps_trigger_softirq;
8256 sd->backlog.poll = process_backlog;
8257 sd->backlog.weight = weight_p;
8262 /* The loopback device is special if any other network devices
8263 * is present in a network namespace the loopback device must
8264 * be present. Since we now dynamically allocate and free the
8265 * loopback device ensure this invariant is maintained by
8266 * keeping the loopback device as the first device on the
8267 * list of network devices. Ensuring the loopback devices
8268 * is the first device that appears and the last network device
8271 if (register_pernet_device(&loopback_net_ops))
8274 if (register_pernet_device(&default_device_ops))
8277 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
8278 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
8280 hotcpu_notifier(dev_cpu_callback, 0);
8287 subsys_initcall(net_dev_init);