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 <linux/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 <linux/bpf.h>
98 #include <net/net_namespace.h>
100 #include <net/busy_poll.h>
101 #include <linux/rtnetlink.h>
102 #include <linux/stat.h>
104 #include <net/dst_metadata.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/iw_handler.h>
115 #include <asm/current.h>
116 #include <linux/audit.h>
117 #include <linux/dmaengine.h>
118 #include <linux/err.h>
119 #include <linux/ctype.h>
120 #include <linux/if_arp.h>
121 #include <linux/if_vlan.h>
122 #include <linux/ip.h>
124 #include <net/mpls.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/static_key.h>
136 #include <linux/hashtable.h>
137 #include <linux/vmalloc.h>
138 #include <linux/if_macvlan.h>
139 #include <linux/errqueue.h>
140 #include <linux/hrtimer.h>
141 #include <linux/netfilter_ingress.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(net, 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)
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) \
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 int ret = ____dev_forward_skb(dev, skb);
1771 skb->protocol = eth_type_trans(skb, dev);
1772 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1777 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1780 * dev_forward_skb - loopback an skb to another netif
1782 * @dev: destination network device
1783 * @skb: buffer to forward
1786 * NET_RX_SUCCESS (no congestion)
1787 * NET_RX_DROP (packet was dropped, but freed)
1789 * dev_forward_skb can be used for injecting an skb from the
1790 * start_xmit function of one device into the receive queue
1791 * of another device.
1793 * The receiving device may be in another namespace, so
1794 * we have to clear all information in the skb that could
1795 * impact namespace isolation.
1797 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1799 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1801 EXPORT_SYMBOL_GPL(dev_forward_skb);
1803 static inline int deliver_skb(struct sk_buff *skb,
1804 struct packet_type *pt_prev,
1805 struct net_device *orig_dev)
1807 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1809 atomic_inc(&skb->users);
1810 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1813 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1814 struct packet_type **pt,
1815 struct net_device *orig_dev,
1817 struct list_head *ptype_list)
1819 struct packet_type *ptype, *pt_prev = *pt;
1821 list_for_each_entry_rcu(ptype, ptype_list, list) {
1822 if (ptype->type != type)
1825 deliver_skb(skb, pt_prev, orig_dev);
1831 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1833 if (!ptype->af_packet_priv || !skb->sk)
1836 if (ptype->id_match)
1837 return ptype->id_match(ptype, skb->sk);
1838 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1845 * Support routine. Sends outgoing frames to any network
1846 * taps currently in use.
1849 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1851 struct packet_type *ptype;
1852 struct sk_buff *skb2 = NULL;
1853 struct packet_type *pt_prev = NULL;
1854 struct list_head *ptype_list = &ptype_all;
1858 list_for_each_entry_rcu(ptype, ptype_list, list) {
1859 /* Never send packets back to the socket
1860 * they originated from - MvS (miquels@drinkel.ow.org)
1862 if (skb_loop_sk(ptype, skb))
1866 deliver_skb(skb2, pt_prev, skb->dev);
1871 /* need to clone skb, done only once */
1872 skb2 = skb_clone(skb, GFP_ATOMIC);
1876 net_timestamp_set(skb2);
1878 /* skb->nh should be correctly
1879 * set by sender, so that the second statement is
1880 * just protection against buggy protocols.
1882 skb_reset_mac_header(skb2);
1884 if (skb_network_header(skb2) < skb2->data ||
1885 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1886 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1887 ntohs(skb2->protocol),
1889 skb_reset_network_header(skb2);
1892 skb2->transport_header = skb2->network_header;
1893 skb2->pkt_type = PACKET_OUTGOING;
1897 if (ptype_list == &ptype_all) {
1898 ptype_list = &dev->ptype_all;
1903 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1906 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
1909 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1910 * @dev: Network device
1911 * @txq: number of queues available
1913 * If real_num_tx_queues is changed the tc mappings may no longer be
1914 * valid. To resolve this verify the tc mapping remains valid and if
1915 * not NULL the mapping. With no priorities mapping to this
1916 * offset/count pair it will no longer be used. In the worst case TC0
1917 * is invalid nothing can be done so disable priority mappings. If is
1918 * expected that drivers will fix this mapping if they can before
1919 * calling netif_set_real_num_tx_queues.
1921 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1924 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1926 /* If TC0 is invalidated disable TC mapping */
1927 if (tc->offset + tc->count > txq) {
1928 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1933 /* Invalidated prio to tc mappings set to TC0 */
1934 for (i = 1; i < TC_BITMASK + 1; i++) {
1935 int q = netdev_get_prio_tc_map(dev, i);
1937 tc = &dev->tc_to_txq[q];
1938 if (tc->offset + tc->count > txq) {
1939 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1941 netdev_set_prio_tc_map(dev, i, 0);
1946 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
1949 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1952 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
1953 if ((txq - tc->offset) < tc->count)
1964 static DEFINE_MUTEX(xps_map_mutex);
1965 #define xmap_dereference(P) \
1966 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1968 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
1971 struct xps_map *map = NULL;
1975 map = xmap_dereference(dev_maps->cpu_map[tci]);
1979 for (pos = map->len; pos--;) {
1980 if (map->queues[pos] != index)
1984 map->queues[pos] = map->queues[--map->len];
1988 RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL);
1989 kfree_rcu(map, rcu);
1996 static bool remove_xps_queue_cpu(struct net_device *dev,
1997 struct xps_dev_maps *dev_maps,
1998 int cpu, u16 offset, u16 count)
2000 int num_tc = dev->num_tc ? : 1;
2001 bool active = false;
2004 for (tci = cpu * num_tc; num_tc--; tci++) {
2007 for (i = count, j = offset; i--; j++) {
2008 if (!remove_xps_queue(dev_maps, cpu, j))
2018 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2021 struct xps_dev_maps *dev_maps;
2023 bool active = false;
2025 mutex_lock(&xps_map_mutex);
2026 dev_maps = xmap_dereference(dev->xps_maps);
2031 for_each_possible_cpu(cpu)
2032 active |= remove_xps_queue_cpu(dev, dev_maps, cpu,
2036 RCU_INIT_POINTER(dev->xps_maps, NULL);
2037 kfree_rcu(dev_maps, rcu);
2040 for (i = offset + (count - 1); count--; i--)
2041 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2045 mutex_unlock(&xps_map_mutex);
2048 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2050 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2053 static struct xps_map *expand_xps_map(struct xps_map *map,
2056 struct xps_map *new_map;
2057 int alloc_len = XPS_MIN_MAP_ALLOC;
2060 for (pos = 0; map && pos < map->len; pos++) {
2061 if (map->queues[pos] != index)
2066 /* Need to add queue to this CPU's existing map */
2068 if (pos < map->alloc_len)
2071 alloc_len = map->alloc_len * 2;
2074 /* Need to allocate new map to store queue on this CPU's map */
2075 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2080 for (i = 0; i < pos; i++)
2081 new_map->queues[i] = map->queues[i];
2082 new_map->alloc_len = alloc_len;
2088 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2091 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2092 int i, cpu, tci, numa_node_id = -2;
2093 int maps_sz, num_tc = 1, tc = 0;
2094 struct xps_map *map, *new_map;
2095 bool active = false;
2098 num_tc = dev->num_tc;
2099 tc = netdev_txq_to_tc(dev, index);
2104 maps_sz = XPS_DEV_MAPS_SIZE(num_tc);
2105 if (maps_sz < L1_CACHE_BYTES)
2106 maps_sz = L1_CACHE_BYTES;
2108 mutex_lock(&xps_map_mutex);
2110 dev_maps = xmap_dereference(dev->xps_maps);
2112 /* allocate memory for queue storage */
2113 for_each_cpu_and(cpu, cpu_online_mask, mask) {
2115 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2116 if (!new_dev_maps) {
2117 mutex_unlock(&xps_map_mutex);
2121 tci = cpu * num_tc + tc;
2122 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) :
2125 map = expand_xps_map(map, cpu, index);
2129 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2133 goto out_no_new_maps;
2135 for_each_possible_cpu(cpu) {
2136 /* copy maps belonging to foreign traffic classes */
2137 for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) {
2138 /* fill in the new device map from the old device map */
2139 map = xmap_dereference(dev_maps->cpu_map[tci]);
2140 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2143 /* We need to explicitly update tci as prevous loop
2144 * could break out early if dev_maps is NULL.
2146 tci = cpu * num_tc + tc;
2148 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2149 /* add queue to CPU maps */
2152 map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2153 while ((pos < map->len) && (map->queues[pos] != index))
2156 if (pos == map->len)
2157 map->queues[map->len++] = index;
2159 if (numa_node_id == -2)
2160 numa_node_id = cpu_to_node(cpu);
2161 else if (numa_node_id != cpu_to_node(cpu))
2164 } else if (dev_maps) {
2165 /* fill in the new device map from the old device map */
2166 map = xmap_dereference(dev_maps->cpu_map[tci]);
2167 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2170 /* copy maps belonging to foreign traffic classes */
2171 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2172 /* fill in the new device map from the old device map */
2173 map = xmap_dereference(dev_maps->cpu_map[tci]);
2174 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2178 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2180 /* Cleanup old maps */
2182 goto out_no_old_maps;
2184 for_each_possible_cpu(cpu) {
2185 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2186 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2187 map = xmap_dereference(dev_maps->cpu_map[tci]);
2188 if (map && map != new_map)
2189 kfree_rcu(map, rcu);
2193 kfree_rcu(dev_maps, rcu);
2196 dev_maps = new_dev_maps;
2200 /* update Tx queue numa node */
2201 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2202 (numa_node_id >= 0) ? numa_node_id :
2208 /* removes queue from unused CPUs */
2209 for_each_possible_cpu(cpu) {
2210 for (i = tc, tci = cpu * num_tc; i--; tci++)
2211 active |= remove_xps_queue(dev_maps, tci, index);
2212 if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu))
2213 active |= remove_xps_queue(dev_maps, tci, index);
2214 for (i = num_tc - tc, tci++; --i; tci++)
2215 active |= remove_xps_queue(dev_maps, tci, index);
2218 /* free map if not active */
2220 RCU_INIT_POINTER(dev->xps_maps, NULL);
2221 kfree_rcu(dev_maps, rcu);
2225 mutex_unlock(&xps_map_mutex);
2229 /* remove any maps that we added */
2230 for_each_possible_cpu(cpu) {
2231 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2232 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2234 xmap_dereference(dev_maps->cpu_map[tci]) :
2236 if (new_map && new_map != map)
2241 mutex_unlock(&xps_map_mutex);
2243 kfree(new_dev_maps);
2246 EXPORT_SYMBOL(netif_set_xps_queue);
2249 void netdev_reset_tc(struct net_device *dev)
2252 netif_reset_xps_queues_gt(dev, 0);
2255 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2256 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2258 EXPORT_SYMBOL(netdev_reset_tc);
2260 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2262 if (tc >= dev->num_tc)
2266 netif_reset_xps_queues(dev, offset, count);
2268 dev->tc_to_txq[tc].count = count;
2269 dev->tc_to_txq[tc].offset = offset;
2272 EXPORT_SYMBOL(netdev_set_tc_queue);
2274 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2276 if (num_tc > TC_MAX_QUEUE)
2280 netif_reset_xps_queues_gt(dev, 0);
2282 dev->num_tc = num_tc;
2285 EXPORT_SYMBOL(netdev_set_num_tc);
2288 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2289 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2291 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2295 if (txq < 1 || txq > dev->num_tx_queues)
2298 if (dev->reg_state == NETREG_REGISTERED ||
2299 dev->reg_state == NETREG_UNREGISTERING) {
2302 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2308 netif_setup_tc(dev, txq);
2310 if (txq < dev->real_num_tx_queues) {
2311 qdisc_reset_all_tx_gt(dev, txq);
2313 netif_reset_xps_queues_gt(dev, txq);
2318 dev->real_num_tx_queues = txq;
2321 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2325 * netif_set_real_num_rx_queues - set actual number of RX queues used
2326 * @dev: Network device
2327 * @rxq: Actual number of RX queues
2329 * This must be called either with the rtnl_lock held or before
2330 * registration of the net device. Returns 0 on success, or a
2331 * negative error code. If called before registration, it always
2334 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2338 if (rxq < 1 || rxq > dev->num_rx_queues)
2341 if (dev->reg_state == NETREG_REGISTERED) {
2344 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2350 dev->real_num_rx_queues = rxq;
2353 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2357 * netif_get_num_default_rss_queues - default number of RSS queues
2359 * This routine should set an upper limit on the number of RSS queues
2360 * used by default by multiqueue devices.
2362 int netif_get_num_default_rss_queues(void)
2364 return is_kdump_kernel() ?
2365 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2367 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2369 static void __netif_reschedule(struct Qdisc *q)
2371 struct softnet_data *sd;
2372 unsigned long flags;
2374 local_irq_save(flags);
2375 sd = this_cpu_ptr(&softnet_data);
2376 q->next_sched = NULL;
2377 *sd->output_queue_tailp = q;
2378 sd->output_queue_tailp = &q->next_sched;
2379 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2380 local_irq_restore(flags);
2383 void __netif_schedule(struct Qdisc *q)
2385 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2386 __netif_reschedule(q);
2388 EXPORT_SYMBOL(__netif_schedule);
2390 struct dev_kfree_skb_cb {
2391 enum skb_free_reason reason;
2394 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2396 return (struct dev_kfree_skb_cb *)skb->cb;
2399 void netif_schedule_queue(struct netdev_queue *txq)
2402 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2403 struct Qdisc *q = rcu_dereference(txq->qdisc);
2405 __netif_schedule(q);
2409 EXPORT_SYMBOL(netif_schedule_queue);
2411 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2413 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2417 q = rcu_dereference(dev_queue->qdisc);
2418 __netif_schedule(q);
2422 EXPORT_SYMBOL(netif_tx_wake_queue);
2424 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2426 unsigned long flags;
2428 if (likely(atomic_read(&skb->users) == 1)) {
2430 atomic_set(&skb->users, 0);
2431 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2434 get_kfree_skb_cb(skb)->reason = reason;
2435 local_irq_save(flags);
2436 skb->next = __this_cpu_read(softnet_data.completion_queue);
2437 __this_cpu_write(softnet_data.completion_queue, skb);
2438 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2439 local_irq_restore(flags);
2441 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2443 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2445 if (in_irq() || irqs_disabled())
2446 __dev_kfree_skb_irq(skb, reason);
2450 EXPORT_SYMBOL(__dev_kfree_skb_any);
2454 * netif_device_detach - mark device as removed
2455 * @dev: network device
2457 * Mark device as removed from system and therefore no longer available.
2459 void netif_device_detach(struct net_device *dev)
2461 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2462 netif_running(dev)) {
2463 netif_tx_stop_all_queues(dev);
2466 EXPORT_SYMBOL(netif_device_detach);
2469 * netif_device_attach - mark device as attached
2470 * @dev: network device
2472 * Mark device as attached from system and restart if needed.
2474 void netif_device_attach(struct net_device *dev)
2476 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2477 netif_running(dev)) {
2478 netif_tx_wake_all_queues(dev);
2479 __netdev_watchdog_up(dev);
2482 EXPORT_SYMBOL(netif_device_attach);
2485 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2486 * to be used as a distribution range.
2488 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2489 unsigned int num_tx_queues)
2493 u16 qcount = num_tx_queues;
2495 if (skb_rx_queue_recorded(skb)) {
2496 hash = skb_get_rx_queue(skb);
2497 while (unlikely(hash >= num_tx_queues))
2498 hash -= num_tx_queues;
2503 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2504 qoffset = dev->tc_to_txq[tc].offset;
2505 qcount = dev->tc_to_txq[tc].count;
2508 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2510 EXPORT_SYMBOL(__skb_tx_hash);
2512 static void skb_warn_bad_offload(const struct sk_buff *skb)
2514 static const netdev_features_t null_features;
2515 struct net_device *dev = skb->dev;
2516 const char *name = "";
2518 if (!net_ratelimit())
2522 if (dev->dev.parent)
2523 name = dev_driver_string(dev->dev.parent);
2525 name = netdev_name(dev);
2527 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2528 "gso_type=%d ip_summed=%d\n",
2529 name, dev ? &dev->features : &null_features,
2530 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2531 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2532 skb_shinfo(skb)->gso_type, skb->ip_summed);
2536 * Invalidate hardware checksum when packet is to be mangled, and
2537 * complete checksum manually on outgoing path.
2539 int skb_checksum_help(struct sk_buff *skb)
2542 int ret = 0, offset;
2544 if (skb->ip_summed == CHECKSUM_COMPLETE)
2545 goto out_set_summed;
2547 if (unlikely(skb_shinfo(skb)->gso_size)) {
2548 skb_warn_bad_offload(skb);
2552 /* Before computing a checksum, we should make sure no frag could
2553 * be modified by an external entity : checksum could be wrong.
2555 if (skb_has_shared_frag(skb)) {
2556 ret = __skb_linearize(skb);
2561 offset = skb_checksum_start_offset(skb);
2562 BUG_ON(offset >= skb_headlen(skb));
2563 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2565 offset += skb->csum_offset;
2566 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2568 if (skb_cloned(skb) &&
2569 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2570 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2575 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2577 skb->ip_summed = CHECKSUM_NONE;
2581 EXPORT_SYMBOL(skb_checksum_help);
2583 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2585 __be16 type = skb->protocol;
2587 /* Tunnel gso handlers can set protocol to ethernet. */
2588 if (type == htons(ETH_P_TEB)) {
2591 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2594 eth = (struct ethhdr *)skb_mac_header(skb);
2595 type = eth->h_proto;
2598 return __vlan_get_protocol(skb, type, depth);
2602 * skb_mac_gso_segment - mac layer segmentation handler.
2603 * @skb: buffer to segment
2604 * @features: features for the output path (see dev->features)
2606 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2607 netdev_features_t features)
2609 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2610 struct packet_offload *ptype;
2611 int vlan_depth = skb->mac_len;
2612 __be16 type = skb_network_protocol(skb, &vlan_depth);
2614 if (unlikely(!type))
2615 return ERR_PTR(-EINVAL);
2617 __skb_pull(skb, vlan_depth);
2620 list_for_each_entry_rcu(ptype, &offload_base, list) {
2621 if (ptype->type == type && ptype->callbacks.gso_segment) {
2622 segs = ptype->callbacks.gso_segment(skb, features);
2628 __skb_push(skb, skb->data - skb_mac_header(skb));
2632 EXPORT_SYMBOL(skb_mac_gso_segment);
2635 /* openvswitch calls this on rx path, so we need a different check.
2637 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2640 return skb->ip_summed != CHECKSUM_PARTIAL;
2642 return skb->ip_summed == CHECKSUM_NONE;
2646 * __skb_gso_segment - Perform segmentation on skb.
2647 * @skb: buffer to segment
2648 * @features: features for the output path (see dev->features)
2649 * @tx_path: whether it is called in TX path
2651 * This function segments the given skb and returns a list of segments.
2653 * It may return NULL if the skb requires no segmentation. This is
2654 * only possible when GSO is used for verifying header integrity.
2656 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2658 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2659 netdev_features_t features, bool tx_path)
2661 if (unlikely(skb_needs_check(skb, tx_path))) {
2664 skb_warn_bad_offload(skb);
2666 err = skb_cow_head(skb, 0);
2668 return ERR_PTR(err);
2671 /* Only report GSO partial support if it will enable us to
2672 * support segmentation on this frame without needing additional
2675 if (features & NETIF_F_GSO_PARTIAL) {
2676 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
2677 struct net_device *dev = skb->dev;
2679 partial_features |= dev->features & dev->gso_partial_features;
2680 if (!skb_gso_ok(skb, features | partial_features))
2681 features &= ~NETIF_F_GSO_PARTIAL;
2684 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2685 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2687 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2688 SKB_GSO_CB(skb)->encap_level = 0;
2690 skb_reset_mac_header(skb);
2691 skb_reset_mac_len(skb);
2693 return skb_mac_gso_segment(skb, features);
2695 EXPORT_SYMBOL(__skb_gso_segment);
2697 /* Take action when hardware reception checksum errors are detected. */
2699 void netdev_rx_csum_fault(struct net_device *dev)
2701 if (net_ratelimit()) {
2702 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2706 EXPORT_SYMBOL(netdev_rx_csum_fault);
2709 /* Actually, we should eliminate this check as soon as we know, that:
2710 * 1. IOMMU is present and allows to map all the memory.
2711 * 2. No high memory really exists on this machine.
2714 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2716 #ifdef CONFIG_HIGHMEM
2718 if (!(dev->features & NETIF_F_HIGHDMA)) {
2719 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2720 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2721 if (PageHighMem(skb_frag_page(frag)))
2726 if (PCI_DMA_BUS_IS_PHYS) {
2727 struct device *pdev = dev->dev.parent;
2731 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2732 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2733 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2734 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2742 /* If MPLS offload request, verify we are testing hardware MPLS features
2743 * instead of standard features for the netdev.
2745 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2746 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2747 netdev_features_t features,
2750 if (eth_p_mpls(type))
2751 features &= skb->dev->mpls_features;
2756 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2757 netdev_features_t features,
2764 static netdev_features_t harmonize_features(struct sk_buff *skb,
2765 netdev_features_t features)
2770 type = skb_network_protocol(skb, &tmp);
2771 features = net_mpls_features(skb, features, type);
2773 if (skb->ip_summed != CHECKSUM_NONE &&
2774 !can_checksum_protocol(features, type)) {
2775 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2777 if (illegal_highdma(skb->dev, skb))
2778 features &= ~NETIF_F_SG;
2783 netdev_features_t passthru_features_check(struct sk_buff *skb,
2784 struct net_device *dev,
2785 netdev_features_t features)
2789 EXPORT_SYMBOL(passthru_features_check);
2791 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2792 struct net_device *dev,
2793 netdev_features_t features)
2795 return vlan_features_check(skb, features);
2798 static netdev_features_t gso_features_check(const struct sk_buff *skb,
2799 struct net_device *dev,
2800 netdev_features_t features)
2802 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2804 if (gso_segs > dev->gso_max_segs)
2805 return features & ~NETIF_F_GSO_MASK;
2807 /* Support for GSO partial features requires software
2808 * intervention before we can actually process the packets
2809 * so we need to strip support for any partial features now
2810 * and we can pull them back in after we have partially
2811 * segmented the frame.
2813 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
2814 features &= ~dev->gso_partial_features;
2816 /* Make sure to clear the IPv4 ID mangling feature if the
2817 * IPv4 header has the potential to be fragmented.
2819 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2820 struct iphdr *iph = skb->encapsulation ?
2821 inner_ip_hdr(skb) : ip_hdr(skb);
2823 if (!(iph->frag_off & htons(IP_DF)))
2824 features &= ~NETIF_F_TSO_MANGLEID;
2830 netdev_features_t netif_skb_features(struct sk_buff *skb)
2832 struct net_device *dev = skb->dev;
2833 netdev_features_t features = dev->features;
2835 if (skb_is_gso(skb))
2836 features = gso_features_check(skb, dev, features);
2838 /* If encapsulation offload request, verify we are testing
2839 * hardware encapsulation features instead of standard
2840 * features for the netdev
2842 if (skb->encapsulation)
2843 features &= dev->hw_enc_features;
2845 if (skb_vlan_tagged(skb))
2846 features = netdev_intersect_features(features,
2847 dev->vlan_features |
2848 NETIF_F_HW_VLAN_CTAG_TX |
2849 NETIF_F_HW_VLAN_STAG_TX);
2851 if (dev->netdev_ops->ndo_features_check)
2852 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2855 features &= dflt_features_check(skb, dev, features);
2857 return harmonize_features(skb, features);
2859 EXPORT_SYMBOL(netif_skb_features);
2861 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2862 struct netdev_queue *txq, bool more)
2867 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2868 dev_queue_xmit_nit(skb, dev);
2871 trace_net_dev_start_xmit(skb, dev);
2872 rc = netdev_start_xmit(skb, dev, txq, more);
2873 trace_net_dev_xmit(skb, rc, dev, len);
2878 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2879 struct netdev_queue *txq, int *ret)
2881 struct sk_buff *skb = first;
2882 int rc = NETDEV_TX_OK;
2885 struct sk_buff *next = skb->next;
2888 rc = xmit_one(skb, dev, txq, next != NULL);
2889 if (unlikely(!dev_xmit_complete(rc))) {
2895 if (netif_xmit_stopped(txq) && skb) {
2896 rc = NETDEV_TX_BUSY;
2906 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2907 netdev_features_t features)
2909 if (skb_vlan_tag_present(skb) &&
2910 !vlan_hw_offload_capable(features, skb->vlan_proto))
2911 skb = __vlan_hwaccel_push_inside(skb);
2915 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2917 netdev_features_t features;
2919 features = netif_skb_features(skb);
2920 skb = validate_xmit_vlan(skb, features);
2924 if (netif_needs_gso(skb, features)) {
2925 struct sk_buff *segs;
2927 segs = skb_gso_segment(skb, features);
2935 if (skb_needs_linearize(skb, features) &&
2936 __skb_linearize(skb))
2939 /* If packet is not checksummed and device does not
2940 * support checksumming for this protocol, complete
2941 * checksumming here.
2943 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2944 if (skb->encapsulation)
2945 skb_set_inner_transport_header(skb,
2946 skb_checksum_start_offset(skb));
2948 skb_set_transport_header(skb,
2949 skb_checksum_start_offset(skb));
2950 if (!(features & NETIF_F_CSUM_MASK) &&
2951 skb_checksum_help(skb))
2961 atomic_long_inc(&dev->tx_dropped);
2965 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2967 struct sk_buff *next, *head = NULL, *tail;
2969 for (; skb != NULL; skb = next) {
2973 /* in case skb wont be segmented, point to itself */
2976 skb = validate_xmit_skb(skb, dev);
2984 /* If skb was segmented, skb->prev points to
2985 * the last segment. If not, it still contains skb.
2991 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
2993 static void qdisc_pkt_len_init(struct sk_buff *skb)
2995 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2997 qdisc_skb_cb(skb)->pkt_len = skb->len;
2999 /* To get more precise estimation of bytes sent on wire,
3000 * we add to pkt_len the headers size of all segments
3002 if (shinfo->gso_size) {
3003 unsigned int hdr_len;
3004 u16 gso_segs = shinfo->gso_segs;
3006 /* mac layer + network layer */
3007 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3009 /* + transport layer */
3010 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
3011 hdr_len += tcp_hdrlen(skb);
3013 hdr_len += sizeof(struct udphdr);
3015 if (shinfo->gso_type & SKB_GSO_DODGY)
3016 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3019 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3023 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3024 struct net_device *dev,
3025 struct netdev_queue *txq)
3027 spinlock_t *root_lock = qdisc_lock(q);
3028 struct sk_buff *to_free = NULL;
3032 qdisc_calculate_pkt_len(skb, q);
3034 * Heuristic to force contended enqueues to serialize on a
3035 * separate lock before trying to get qdisc main lock.
3036 * This permits qdisc->running owner to get the lock more
3037 * often and dequeue packets faster.
3039 contended = qdisc_is_running(q);
3040 if (unlikely(contended))
3041 spin_lock(&q->busylock);
3043 spin_lock(root_lock);
3044 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3045 __qdisc_drop(skb, &to_free);
3047 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3048 qdisc_run_begin(q)) {
3050 * This is a work-conserving queue; there are no old skbs
3051 * waiting to be sent out; and the qdisc is not running -
3052 * xmit the skb directly.
3055 qdisc_bstats_update(q, skb);
3057 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3058 if (unlikely(contended)) {
3059 spin_unlock(&q->busylock);
3066 rc = NET_XMIT_SUCCESS;
3068 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3069 if (qdisc_run_begin(q)) {
3070 if (unlikely(contended)) {
3071 spin_unlock(&q->busylock);
3077 spin_unlock(root_lock);
3078 if (unlikely(to_free))
3079 kfree_skb_list(to_free);
3080 if (unlikely(contended))
3081 spin_unlock(&q->busylock);
3085 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3086 static void skb_update_prio(struct sk_buff *skb)
3088 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
3090 if (!skb->priority && skb->sk && map) {
3091 unsigned int prioidx =
3092 sock_cgroup_prioidx(&skb->sk->sk_cgrp_data);
3094 if (prioidx < map->priomap_len)
3095 skb->priority = map->priomap[prioidx];
3099 #define skb_update_prio(skb)
3102 DEFINE_PER_CPU(int, xmit_recursion);
3103 EXPORT_SYMBOL(xmit_recursion);
3106 * dev_loopback_xmit - loop back @skb
3107 * @net: network namespace this loopback is happening in
3108 * @sk: sk needed to be a netfilter okfn
3109 * @skb: buffer to transmit
3111 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3113 skb_reset_mac_header(skb);
3114 __skb_pull(skb, skb_network_offset(skb));
3115 skb->pkt_type = PACKET_LOOPBACK;
3116 skb->ip_summed = CHECKSUM_UNNECESSARY;
3117 WARN_ON(!skb_dst(skb));
3122 EXPORT_SYMBOL(dev_loopback_xmit);
3124 #ifdef CONFIG_NET_EGRESS
3125 static struct sk_buff *
3126 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3128 struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list);
3129 struct tcf_result cl_res;
3134 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3135 qdisc_bstats_cpu_update(cl->q, skb);
3137 switch (tc_classify(skb, cl, &cl_res, false)) {
3139 case TC_ACT_RECLASSIFY:
3140 skb->tc_index = TC_H_MIN(cl_res.classid);
3143 qdisc_qstats_cpu_drop(cl->q);
3144 *ret = NET_XMIT_DROP;
3149 *ret = NET_XMIT_SUCCESS;
3152 case TC_ACT_REDIRECT:
3153 /* No need to push/pop skb's mac_header here on egress! */
3154 skb_do_redirect(skb);
3155 *ret = NET_XMIT_SUCCESS;
3163 #endif /* CONFIG_NET_EGRESS */
3165 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3168 struct xps_dev_maps *dev_maps;
3169 struct xps_map *map;
3170 int queue_index = -1;
3173 dev_maps = rcu_dereference(dev->xps_maps);
3175 unsigned int tci = skb->sender_cpu - 1;
3179 tci += netdev_get_prio_tc_map(dev, skb->priority);
3182 map = rcu_dereference(dev_maps->cpu_map[tci]);
3185 queue_index = map->queues[0];
3187 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3189 if (unlikely(queue_index >= dev->real_num_tx_queues))
3201 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3203 struct sock *sk = skb->sk;
3204 int queue_index = sk_tx_queue_get(sk);
3206 if (queue_index < 0 || skb->ooo_okay ||
3207 queue_index >= dev->real_num_tx_queues) {
3208 int new_index = get_xps_queue(dev, skb);
3210 new_index = skb_tx_hash(dev, skb);
3212 if (queue_index != new_index && sk &&
3214 rcu_access_pointer(sk->sk_dst_cache))
3215 sk_tx_queue_set(sk, new_index);
3217 queue_index = new_index;
3223 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3224 struct sk_buff *skb,
3227 int queue_index = 0;
3230 u32 sender_cpu = skb->sender_cpu - 1;
3232 if (sender_cpu >= (u32)NR_CPUS)
3233 skb->sender_cpu = raw_smp_processor_id() + 1;
3236 if (dev->real_num_tx_queues != 1) {
3237 const struct net_device_ops *ops = dev->netdev_ops;
3238 if (ops->ndo_select_queue)
3239 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3242 queue_index = __netdev_pick_tx(dev, skb);
3245 queue_index = netdev_cap_txqueue(dev, queue_index);
3248 skb_set_queue_mapping(skb, queue_index);
3249 return netdev_get_tx_queue(dev, queue_index);
3253 * __dev_queue_xmit - transmit a buffer
3254 * @skb: buffer to transmit
3255 * @accel_priv: private data used for L2 forwarding offload
3257 * Queue a buffer for transmission to a network device. The caller must
3258 * have set the device and priority and built the buffer before calling
3259 * this function. The function can be called from an interrupt.
3261 * A negative errno code is returned on a failure. A success does not
3262 * guarantee the frame will be transmitted as it may be dropped due
3263 * to congestion or traffic shaping.
3265 * -----------------------------------------------------------------------------------
3266 * I notice this method can also return errors from the queue disciplines,
3267 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3270 * Regardless of the return value, the skb is consumed, so it is currently
3271 * difficult to retry a send to this method. (You can bump the ref count
3272 * before sending to hold a reference for retry if you are careful.)
3274 * When calling this method, interrupts MUST be enabled. This is because
3275 * the BH enable code must have IRQs enabled so that it will not deadlock.
3278 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3280 struct net_device *dev = skb->dev;
3281 struct netdev_queue *txq;
3285 skb_reset_mac_header(skb);
3287 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3288 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3290 /* Disable soft irqs for various locks below. Also
3291 * stops preemption for RCU.
3295 skb_update_prio(skb);
3297 qdisc_pkt_len_init(skb);
3298 #ifdef CONFIG_NET_CLS_ACT
3299 skb->tc_at_ingress = 0;
3300 # ifdef CONFIG_NET_EGRESS
3301 if (static_key_false(&egress_needed)) {
3302 skb = sch_handle_egress(skb, &rc, dev);
3308 /* If device/qdisc don't need skb->dst, release it right now while
3309 * its hot in this cpu cache.
3311 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3316 txq = netdev_pick_tx(dev, skb, accel_priv);
3317 q = rcu_dereference_bh(txq->qdisc);
3319 trace_net_dev_queue(skb);
3321 rc = __dev_xmit_skb(skb, q, dev, txq);
3325 /* The device has no queue. Common case for software devices:
3326 loopback, all the sorts of tunnels...
3328 Really, it is unlikely that netif_tx_lock protection is necessary
3329 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3331 However, it is possible, that they rely on protection
3334 Check this and shot the lock. It is not prone from deadlocks.
3335 Either shot noqueue qdisc, it is even simpler 8)
3337 if (dev->flags & IFF_UP) {
3338 int cpu = smp_processor_id(); /* ok because BHs are off */
3340 if (txq->xmit_lock_owner != cpu) {
3341 if (unlikely(__this_cpu_read(xmit_recursion) >
3342 XMIT_RECURSION_LIMIT))
3343 goto recursion_alert;
3345 skb = validate_xmit_skb(skb, dev);
3349 HARD_TX_LOCK(dev, txq, cpu);
3351 if (!netif_xmit_stopped(txq)) {
3352 __this_cpu_inc(xmit_recursion);
3353 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3354 __this_cpu_dec(xmit_recursion);
3355 if (dev_xmit_complete(rc)) {
3356 HARD_TX_UNLOCK(dev, txq);
3360 HARD_TX_UNLOCK(dev, txq);
3361 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3364 /* Recursion is detected! It is possible,
3368 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3374 rcu_read_unlock_bh();
3376 atomic_long_inc(&dev->tx_dropped);
3377 kfree_skb_list(skb);
3380 rcu_read_unlock_bh();
3384 int dev_queue_xmit(struct sk_buff *skb)
3386 return __dev_queue_xmit(skb, NULL);
3388 EXPORT_SYMBOL(dev_queue_xmit);
3390 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3392 return __dev_queue_xmit(skb, accel_priv);
3394 EXPORT_SYMBOL(dev_queue_xmit_accel);
3397 /*=======================================================================
3399 =======================================================================*/
3401 int netdev_max_backlog __read_mostly = 1000;
3402 EXPORT_SYMBOL(netdev_max_backlog);
3404 int netdev_tstamp_prequeue __read_mostly = 1;
3405 int netdev_budget __read_mostly = 300;
3406 int weight_p __read_mostly = 64; /* old backlog weight */
3407 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
3408 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
3409 int dev_rx_weight __read_mostly = 64;
3410 int dev_tx_weight __read_mostly = 64;
3412 /* Called with irq disabled */
3413 static inline void ____napi_schedule(struct softnet_data *sd,
3414 struct napi_struct *napi)
3416 list_add_tail(&napi->poll_list, &sd->poll_list);
3417 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3422 /* One global table that all flow-based protocols share. */
3423 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3424 EXPORT_SYMBOL(rps_sock_flow_table);
3425 u32 rps_cpu_mask __read_mostly;
3426 EXPORT_SYMBOL(rps_cpu_mask);
3428 struct static_key rps_needed __read_mostly;
3429 EXPORT_SYMBOL(rps_needed);
3430 struct static_key rfs_needed __read_mostly;
3431 EXPORT_SYMBOL(rfs_needed);
3433 static struct rps_dev_flow *
3434 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3435 struct rps_dev_flow *rflow, u16 next_cpu)
3437 if (next_cpu < nr_cpu_ids) {
3438 #ifdef CONFIG_RFS_ACCEL
3439 struct netdev_rx_queue *rxqueue;
3440 struct rps_dev_flow_table *flow_table;
3441 struct rps_dev_flow *old_rflow;
3446 /* Should we steer this flow to a different hardware queue? */
3447 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3448 !(dev->features & NETIF_F_NTUPLE))
3450 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3451 if (rxq_index == skb_get_rx_queue(skb))
3454 rxqueue = dev->_rx + rxq_index;
3455 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3458 flow_id = skb_get_hash(skb) & flow_table->mask;
3459 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3460 rxq_index, flow_id);
3464 rflow = &flow_table->flows[flow_id];
3466 if (old_rflow->filter == rflow->filter)
3467 old_rflow->filter = RPS_NO_FILTER;
3471 per_cpu(softnet_data, next_cpu).input_queue_head;
3474 rflow->cpu = next_cpu;
3479 * get_rps_cpu is called from netif_receive_skb and returns the target
3480 * CPU from the RPS map of the receiving queue for a given skb.
3481 * rcu_read_lock must be held on entry.
3483 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3484 struct rps_dev_flow **rflowp)
3486 const struct rps_sock_flow_table *sock_flow_table;
3487 struct netdev_rx_queue *rxqueue = dev->_rx;
3488 struct rps_dev_flow_table *flow_table;
3489 struct rps_map *map;
3494 if (skb_rx_queue_recorded(skb)) {
3495 u16 index = skb_get_rx_queue(skb);
3497 if (unlikely(index >= dev->real_num_rx_queues)) {
3498 WARN_ONCE(dev->real_num_rx_queues > 1,
3499 "%s received packet on queue %u, but number "
3500 "of RX queues is %u\n",
3501 dev->name, index, dev->real_num_rx_queues);
3507 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3509 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3510 map = rcu_dereference(rxqueue->rps_map);
3511 if (!flow_table && !map)
3514 skb_reset_network_header(skb);
3515 hash = skb_get_hash(skb);
3519 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3520 if (flow_table && sock_flow_table) {
3521 struct rps_dev_flow *rflow;
3525 /* First check into global flow table if there is a match */
3526 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3527 if ((ident ^ hash) & ~rps_cpu_mask)
3530 next_cpu = ident & rps_cpu_mask;
3532 /* OK, now we know there is a match,
3533 * we can look at the local (per receive queue) flow table
3535 rflow = &flow_table->flows[hash & flow_table->mask];
3539 * If the desired CPU (where last recvmsg was done) is
3540 * different from current CPU (one in the rx-queue flow
3541 * table entry), switch if one of the following holds:
3542 * - Current CPU is unset (>= nr_cpu_ids).
3543 * - Current CPU is offline.
3544 * - The current CPU's queue tail has advanced beyond the
3545 * last packet that was enqueued using this table entry.
3546 * This guarantees that all previous packets for the flow
3547 * have been dequeued, thus preserving in order delivery.
3549 if (unlikely(tcpu != next_cpu) &&
3550 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3551 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3552 rflow->last_qtail)) >= 0)) {
3554 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3557 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3567 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3568 if (cpu_online(tcpu)) {
3578 #ifdef CONFIG_RFS_ACCEL
3581 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3582 * @dev: Device on which the filter was set
3583 * @rxq_index: RX queue index
3584 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3585 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3587 * Drivers that implement ndo_rx_flow_steer() should periodically call
3588 * this function for each installed filter and remove the filters for
3589 * which it returns %true.
3591 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3592 u32 flow_id, u16 filter_id)
3594 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3595 struct rps_dev_flow_table *flow_table;
3596 struct rps_dev_flow *rflow;
3601 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3602 if (flow_table && flow_id <= flow_table->mask) {
3603 rflow = &flow_table->flows[flow_id];
3604 cpu = ACCESS_ONCE(rflow->cpu);
3605 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3606 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3607 rflow->last_qtail) <
3608 (int)(10 * flow_table->mask)))
3614 EXPORT_SYMBOL(rps_may_expire_flow);
3616 #endif /* CONFIG_RFS_ACCEL */
3618 /* Called from hardirq (IPI) context */
3619 static void rps_trigger_softirq(void *data)
3621 struct softnet_data *sd = data;
3623 ____napi_schedule(sd, &sd->backlog);
3627 #endif /* CONFIG_RPS */
3630 * Check if this softnet_data structure is another cpu one
3631 * If yes, queue it to our IPI list and return 1
3634 static int rps_ipi_queued(struct softnet_data *sd)
3637 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3640 sd->rps_ipi_next = mysd->rps_ipi_list;
3641 mysd->rps_ipi_list = sd;
3643 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3646 #endif /* CONFIG_RPS */
3650 #ifdef CONFIG_NET_FLOW_LIMIT
3651 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3654 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3656 #ifdef CONFIG_NET_FLOW_LIMIT
3657 struct sd_flow_limit *fl;
3658 struct softnet_data *sd;
3659 unsigned int old_flow, new_flow;
3661 if (qlen < (netdev_max_backlog >> 1))
3664 sd = this_cpu_ptr(&softnet_data);
3667 fl = rcu_dereference(sd->flow_limit);
3669 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3670 old_flow = fl->history[fl->history_head];
3671 fl->history[fl->history_head] = new_flow;
3674 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3676 if (likely(fl->buckets[old_flow]))
3677 fl->buckets[old_flow]--;
3679 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3691 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3692 * queue (may be a remote CPU queue).
3694 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3695 unsigned int *qtail)
3697 struct softnet_data *sd;
3698 unsigned long flags;
3701 sd = &per_cpu(softnet_data, cpu);
3703 local_irq_save(flags);
3706 if (!netif_running(skb->dev))
3708 qlen = skb_queue_len(&sd->input_pkt_queue);
3709 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3712 __skb_queue_tail(&sd->input_pkt_queue, skb);
3713 input_queue_tail_incr_save(sd, qtail);
3715 local_irq_restore(flags);
3716 return NET_RX_SUCCESS;
3719 /* Schedule NAPI for backlog device
3720 * We can use non atomic operation since we own the queue lock
3722 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3723 if (!rps_ipi_queued(sd))
3724 ____napi_schedule(sd, &sd->backlog);
3733 local_irq_restore(flags);
3735 atomic_long_inc(&skb->dev->rx_dropped);
3740 static int netif_rx_internal(struct sk_buff *skb)
3744 net_timestamp_check(netdev_tstamp_prequeue, skb);
3746 trace_netif_rx(skb);
3748 if (static_key_false(&rps_needed)) {
3749 struct rps_dev_flow voidflow, *rflow = &voidflow;
3755 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3757 cpu = smp_processor_id();
3759 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3767 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3774 * netif_rx - post buffer to the network code
3775 * @skb: buffer to post
3777 * This function receives a packet from a device driver and queues it for
3778 * the upper (protocol) levels to process. It always succeeds. The buffer
3779 * may be dropped during processing for congestion control or by the
3783 * NET_RX_SUCCESS (no congestion)
3784 * NET_RX_DROP (packet was dropped)
3788 int netif_rx(struct sk_buff *skb)
3790 trace_netif_rx_entry(skb);
3792 return netif_rx_internal(skb);
3794 EXPORT_SYMBOL(netif_rx);
3796 int netif_rx_ni(struct sk_buff *skb)
3800 trace_netif_rx_ni_entry(skb);
3803 err = netif_rx_internal(skb);
3804 if (local_softirq_pending())
3810 EXPORT_SYMBOL(netif_rx_ni);
3812 static __latent_entropy void net_tx_action(struct softirq_action *h)
3814 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3816 if (sd->completion_queue) {
3817 struct sk_buff *clist;
3819 local_irq_disable();
3820 clist = sd->completion_queue;
3821 sd->completion_queue = NULL;
3825 struct sk_buff *skb = clist;
3826 clist = clist->next;
3828 WARN_ON(atomic_read(&skb->users));
3829 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3830 trace_consume_skb(skb);
3832 trace_kfree_skb(skb, net_tx_action);
3834 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
3837 __kfree_skb_defer(skb);
3840 __kfree_skb_flush();
3843 if (sd->output_queue) {
3846 local_irq_disable();
3847 head = sd->output_queue;
3848 sd->output_queue = NULL;
3849 sd->output_queue_tailp = &sd->output_queue;
3853 struct Qdisc *q = head;
3854 spinlock_t *root_lock;
3856 head = head->next_sched;
3858 root_lock = qdisc_lock(q);
3859 spin_lock(root_lock);
3860 /* We need to make sure head->next_sched is read
3861 * before clearing __QDISC_STATE_SCHED
3863 smp_mb__before_atomic();
3864 clear_bit(__QDISC_STATE_SCHED, &q->state);
3866 spin_unlock(root_lock);
3871 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
3872 /* This hook is defined here for ATM LANE */
3873 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3874 unsigned char *addr) __read_mostly;
3875 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3878 static inline struct sk_buff *
3879 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
3880 struct net_device *orig_dev)
3882 #ifdef CONFIG_NET_CLS_ACT
3883 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3884 struct tcf_result cl_res;
3886 /* If there's at least one ingress present somewhere (so
3887 * we get here via enabled static key), remaining devices
3888 * that are not configured with an ingress qdisc will bail
3894 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3898 qdisc_skb_cb(skb)->pkt_len = skb->len;
3899 skb->tc_at_ingress = 1;
3900 qdisc_bstats_cpu_update(cl->q, skb);
3902 switch (tc_classify(skb, cl, &cl_res, false)) {
3904 case TC_ACT_RECLASSIFY:
3905 skb->tc_index = TC_H_MIN(cl_res.classid);
3908 qdisc_qstats_cpu_drop(cl->q);
3915 case TC_ACT_REDIRECT:
3916 /* skb_mac_header check was done by cls/act_bpf, so
3917 * we can safely push the L2 header back before
3918 * redirecting to another netdev
3920 __skb_push(skb, skb->mac_len);
3921 skb_do_redirect(skb);
3926 #endif /* CONFIG_NET_CLS_ACT */
3931 * netdev_is_rx_handler_busy - check if receive handler is registered
3932 * @dev: device to check
3934 * Check if a receive handler is already registered for a given device.
3935 * Return true if there one.
3937 * The caller must hold the rtnl_mutex.
3939 bool netdev_is_rx_handler_busy(struct net_device *dev)
3942 return dev && rtnl_dereference(dev->rx_handler);
3944 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
3947 * netdev_rx_handler_register - register receive handler
3948 * @dev: device to register a handler for
3949 * @rx_handler: receive handler to register
3950 * @rx_handler_data: data pointer that is used by rx handler
3952 * Register a receive handler for a device. This handler will then be
3953 * called from __netif_receive_skb. A negative errno code is returned
3956 * The caller must hold the rtnl_mutex.
3958 * For a general description of rx_handler, see enum rx_handler_result.
3960 int netdev_rx_handler_register(struct net_device *dev,
3961 rx_handler_func_t *rx_handler,
3962 void *rx_handler_data)
3964 if (netdev_is_rx_handler_busy(dev))
3967 /* Note: rx_handler_data must be set before rx_handler */
3968 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3969 rcu_assign_pointer(dev->rx_handler, rx_handler);
3973 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3976 * netdev_rx_handler_unregister - unregister receive handler
3977 * @dev: device to unregister a handler from
3979 * Unregister a receive handler from a device.
3981 * The caller must hold the rtnl_mutex.
3983 void netdev_rx_handler_unregister(struct net_device *dev)
3987 RCU_INIT_POINTER(dev->rx_handler, NULL);
3988 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3989 * section has a guarantee to see a non NULL rx_handler_data
3993 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3995 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3998 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3999 * the special handling of PFMEMALLOC skbs.
4001 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4003 switch (skb->protocol) {
4004 case htons(ETH_P_ARP):
4005 case htons(ETH_P_IP):
4006 case htons(ETH_P_IPV6):
4007 case htons(ETH_P_8021Q):
4008 case htons(ETH_P_8021AD):
4015 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4016 int *ret, struct net_device *orig_dev)
4018 #ifdef CONFIG_NETFILTER_INGRESS
4019 if (nf_hook_ingress_active(skb)) {
4023 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4028 ingress_retval = nf_hook_ingress(skb);
4030 return ingress_retval;
4032 #endif /* CONFIG_NETFILTER_INGRESS */
4036 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
4038 struct packet_type *ptype, *pt_prev;
4039 rx_handler_func_t *rx_handler;
4040 struct net_device *orig_dev;
4041 bool deliver_exact = false;
4042 int ret = NET_RX_DROP;
4045 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4047 trace_netif_receive_skb(skb);
4049 orig_dev = skb->dev;
4051 skb_reset_network_header(skb);
4052 if (!skb_transport_header_was_set(skb))
4053 skb_reset_transport_header(skb);
4054 skb_reset_mac_len(skb);
4059 skb->skb_iif = skb->dev->ifindex;
4061 __this_cpu_inc(softnet_data.processed);
4063 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4064 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4065 skb = skb_vlan_untag(skb);
4070 if (skb_skip_tc_classify(skb))
4076 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4078 ret = deliver_skb(skb, pt_prev, orig_dev);
4082 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4084 ret = deliver_skb(skb, pt_prev, orig_dev);
4089 #ifdef CONFIG_NET_INGRESS
4090 if (static_key_false(&ingress_needed)) {
4091 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4095 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4101 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4104 if (skb_vlan_tag_present(skb)) {
4106 ret = deliver_skb(skb, pt_prev, orig_dev);
4109 if (vlan_do_receive(&skb))
4111 else if (unlikely(!skb))
4115 rx_handler = rcu_dereference(skb->dev->rx_handler);
4118 ret = deliver_skb(skb, pt_prev, orig_dev);
4121 switch (rx_handler(&skb)) {
4122 case RX_HANDLER_CONSUMED:
4123 ret = NET_RX_SUCCESS;
4125 case RX_HANDLER_ANOTHER:
4127 case RX_HANDLER_EXACT:
4128 deliver_exact = true;
4129 case RX_HANDLER_PASS:
4136 if (unlikely(skb_vlan_tag_present(skb))) {
4137 if (skb_vlan_tag_get_id(skb))
4138 skb->pkt_type = PACKET_OTHERHOST;
4139 /* Note: we might in the future use prio bits
4140 * and set skb->priority like in vlan_do_receive()
4141 * For the time being, just ignore Priority Code Point
4146 type = skb->protocol;
4148 /* deliver only exact match when indicated */
4149 if (likely(!deliver_exact)) {
4150 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4151 &ptype_base[ntohs(type) &
4155 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4156 &orig_dev->ptype_specific);
4158 if (unlikely(skb->dev != orig_dev)) {
4159 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4160 &skb->dev->ptype_specific);
4164 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
4167 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4171 atomic_long_inc(&skb->dev->rx_dropped);
4173 atomic_long_inc(&skb->dev->rx_nohandler);
4175 /* Jamal, now you will not able to escape explaining
4176 * me how you were going to use this. :-)
4185 static int __netif_receive_skb(struct sk_buff *skb)
4189 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4190 unsigned long pflags = current->flags;
4193 * PFMEMALLOC skbs are special, they should
4194 * - be delivered to SOCK_MEMALLOC sockets only
4195 * - stay away from userspace
4196 * - have bounded memory usage
4198 * Use PF_MEMALLOC as this saves us from propagating the allocation
4199 * context down to all allocation sites.
4201 current->flags |= PF_MEMALLOC;
4202 ret = __netif_receive_skb_core(skb, true);
4203 tsk_restore_flags(current, pflags, PF_MEMALLOC);
4205 ret = __netif_receive_skb_core(skb, false);
4210 static int netif_receive_skb_internal(struct sk_buff *skb)
4214 net_timestamp_check(netdev_tstamp_prequeue, skb);
4216 if (skb_defer_rx_timestamp(skb))
4217 return NET_RX_SUCCESS;
4222 if (static_key_false(&rps_needed)) {
4223 struct rps_dev_flow voidflow, *rflow = &voidflow;
4224 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4227 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4233 ret = __netif_receive_skb(skb);
4239 * netif_receive_skb - process receive buffer from network
4240 * @skb: buffer to process
4242 * netif_receive_skb() is the main receive data processing function.
4243 * It always succeeds. The buffer may be dropped during processing
4244 * for congestion control or by the protocol layers.
4246 * This function may only be called from softirq context and interrupts
4247 * should be enabled.
4249 * Return values (usually ignored):
4250 * NET_RX_SUCCESS: no congestion
4251 * NET_RX_DROP: packet was dropped
4253 int netif_receive_skb(struct sk_buff *skb)
4255 trace_netif_receive_skb_entry(skb);
4257 return netif_receive_skb_internal(skb);
4259 EXPORT_SYMBOL(netif_receive_skb);
4261 DEFINE_PER_CPU(struct work_struct, flush_works);
4263 /* Network device is going away, flush any packets still pending */
4264 static void flush_backlog(struct work_struct *work)
4266 struct sk_buff *skb, *tmp;
4267 struct softnet_data *sd;
4270 sd = this_cpu_ptr(&softnet_data);
4272 local_irq_disable();
4274 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4275 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4276 __skb_unlink(skb, &sd->input_pkt_queue);
4278 input_queue_head_incr(sd);
4284 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4285 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4286 __skb_unlink(skb, &sd->process_queue);
4288 input_queue_head_incr(sd);
4294 static void flush_all_backlogs(void)
4300 for_each_online_cpu(cpu)
4301 queue_work_on(cpu, system_highpri_wq,
4302 per_cpu_ptr(&flush_works, cpu));
4304 for_each_online_cpu(cpu)
4305 flush_work(per_cpu_ptr(&flush_works, cpu));
4310 static int napi_gro_complete(struct sk_buff *skb)
4312 struct packet_offload *ptype;
4313 __be16 type = skb->protocol;
4314 struct list_head *head = &offload_base;
4317 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4319 if (NAPI_GRO_CB(skb)->count == 1) {
4320 skb_shinfo(skb)->gso_size = 0;
4325 list_for_each_entry_rcu(ptype, head, list) {
4326 if (ptype->type != type || !ptype->callbacks.gro_complete)
4329 err = ptype->callbacks.gro_complete(skb, 0);
4335 WARN_ON(&ptype->list == head);
4337 return NET_RX_SUCCESS;
4341 return netif_receive_skb_internal(skb);
4344 /* napi->gro_list contains packets ordered by age.
4345 * youngest packets at the head of it.
4346 * Complete skbs in reverse order to reduce latencies.
4348 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4350 struct sk_buff *skb, *prev = NULL;
4352 /* scan list and build reverse chain */
4353 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4358 for (skb = prev; skb; skb = prev) {
4361 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4365 napi_gro_complete(skb);
4369 napi->gro_list = NULL;
4371 EXPORT_SYMBOL(napi_gro_flush);
4373 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4376 unsigned int maclen = skb->dev->hard_header_len;
4377 u32 hash = skb_get_hash_raw(skb);
4379 for (p = napi->gro_list; p; p = p->next) {
4380 unsigned long diffs;
4382 NAPI_GRO_CB(p)->flush = 0;
4384 if (hash != skb_get_hash_raw(p)) {
4385 NAPI_GRO_CB(p)->same_flow = 0;
4389 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4390 diffs |= p->vlan_tci ^ skb->vlan_tci;
4391 diffs |= skb_metadata_dst_cmp(p, skb);
4392 if (maclen == ETH_HLEN)
4393 diffs |= compare_ether_header(skb_mac_header(p),
4394 skb_mac_header(skb));
4396 diffs = memcmp(skb_mac_header(p),
4397 skb_mac_header(skb),
4399 NAPI_GRO_CB(p)->same_flow = !diffs;
4403 static void skb_gro_reset_offset(struct sk_buff *skb)
4405 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4406 const skb_frag_t *frag0 = &pinfo->frags[0];
4408 NAPI_GRO_CB(skb)->data_offset = 0;
4409 NAPI_GRO_CB(skb)->frag0 = NULL;
4410 NAPI_GRO_CB(skb)->frag0_len = 0;
4412 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4414 !PageHighMem(skb_frag_page(frag0))) {
4415 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4416 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
4417 skb_frag_size(frag0),
4418 skb->end - skb->tail);
4422 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4424 struct skb_shared_info *pinfo = skb_shinfo(skb);
4426 BUG_ON(skb->end - skb->tail < grow);
4428 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4430 skb->data_len -= grow;
4433 pinfo->frags[0].page_offset += grow;
4434 skb_frag_size_sub(&pinfo->frags[0], grow);
4436 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4437 skb_frag_unref(skb, 0);
4438 memmove(pinfo->frags, pinfo->frags + 1,
4439 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4443 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4445 struct sk_buff **pp = NULL;
4446 struct packet_offload *ptype;
4447 __be16 type = skb->protocol;
4448 struct list_head *head = &offload_base;
4450 enum gro_result ret;
4453 if (!(skb->dev->features & NETIF_F_GRO))
4459 gro_list_prepare(napi, skb);
4462 list_for_each_entry_rcu(ptype, head, list) {
4463 if (ptype->type != type || !ptype->callbacks.gro_receive)
4466 skb_set_network_header(skb, skb_gro_offset(skb));
4467 skb_reset_mac_len(skb);
4468 NAPI_GRO_CB(skb)->same_flow = 0;
4469 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
4470 NAPI_GRO_CB(skb)->free = 0;
4471 NAPI_GRO_CB(skb)->encap_mark = 0;
4472 NAPI_GRO_CB(skb)->recursion_counter = 0;
4473 NAPI_GRO_CB(skb)->is_fou = 0;
4474 NAPI_GRO_CB(skb)->is_atomic = 1;
4475 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4477 /* Setup for GRO checksum validation */
4478 switch (skb->ip_summed) {
4479 case CHECKSUM_COMPLETE:
4480 NAPI_GRO_CB(skb)->csum = skb->csum;
4481 NAPI_GRO_CB(skb)->csum_valid = 1;
4482 NAPI_GRO_CB(skb)->csum_cnt = 0;
4484 case CHECKSUM_UNNECESSARY:
4485 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4486 NAPI_GRO_CB(skb)->csum_valid = 0;
4489 NAPI_GRO_CB(skb)->csum_cnt = 0;
4490 NAPI_GRO_CB(skb)->csum_valid = 0;
4493 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4498 if (&ptype->list == head)
4501 same_flow = NAPI_GRO_CB(skb)->same_flow;
4502 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4505 struct sk_buff *nskb = *pp;
4509 napi_gro_complete(nskb);
4516 if (NAPI_GRO_CB(skb)->flush)
4519 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4520 struct sk_buff *nskb = napi->gro_list;
4522 /* locate the end of the list to select the 'oldest' flow */
4523 while (nskb->next) {
4529 napi_gro_complete(nskb);
4533 NAPI_GRO_CB(skb)->count = 1;
4534 NAPI_GRO_CB(skb)->age = jiffies;
4535 NAPI_GRO_CB(skb)->last = skb;
4536 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4537 skb->next = napi->gro_list;
4538 napi->gro_list = skb;
4542 grow = skb_gro_offset(skb) - skb_headlen(skb);
4544 gro_pull_from_frag0(skb, grow);
4553 struct packet_offload *gro_find_receive_by_type(__be16 type)
4555 struct list_head *offload_head = &offload_base;
4556 struct packet_offload *ptype;
4558 list_for_each_entry_rcu(ptype, offload_head, list) {
4559 if (ptype->type != type || !ptype->callbacks.gro_receive)
4565 EXPORT_SYMBOL(gro_find_receive_by_type);
4567 struct packet_offload *gro_find_complete_by_type(__be16 type)
4569 struct list_head *offload_head = &offload_base;
4570 struct packet_offload *ptype;
4572 list_for_each_entry_rcu(ptype, offload_head, list) {
4573 if (ptype->type != type || !ptype->callbacks.gro_complete)
4579 EXPORT_SYMBOL(gro_find_complete_by_type);
4581 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4585 if (netif_receive_skb_internal(skb))
4593 case GRO_MERGED_FREE:
4594 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) {
4597 kmem_cache_free(skbuff_head_cache, skb);
4611 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4613 skb_mark_napi_id(skb, napi);
4614 trace_napi_gro_receive_entry(skb);
4616 skb_gro_reset_offset(skb);
4618 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4620 EXPORT_SYMBOL(napi_gro_receive);
4622 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4624 if (unlikely(skb->pfmemalloc)) {
4628 __skb_pull(skb, skb_headlen(skb));
4629 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4630 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4632 skb->dev = napi->dev;
4634 skb->encapsulation = 0;
4635 skb_shinfo(skb)->gso_type = 0;
4636 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4642 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4644 struct sk_buff *skb = napi->skb;
4647 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4650 skb_mark_napi_id(skb, napi);
4655 EXPORT_SYMBOL(napi_get_frags);
4657 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4658 struct sk_buff *skb,
4664 __skb_push(skb, ETH_HLEN);
4665 skb->protocol = eth_type_trans(skb, skb->dev);
4666 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4671 case GRO_MERGED_FREE:
4672 napi_reuse_skb(napi, skb);
4682 /* Upper GRO stack assumes network header starts at gro_offset=0
4683 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4684 * We copy ethernet header into skb->data to have a common layout.
4686 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4688 struct sk_buff *skb = napi->skb;
4689 const struct ethhdr *eth;
4690 unsigned int hlen = sizeof(*eth);
4694 skb_reset_mac_header(skb);
4695 skb_gro_reset_offset(skb);
4697 eth = skb_gro_header_fast(skb, 0);
4698 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4699 eth = skb_gro_header_slow(skb, hlen, 0);
4700 if (unlikely(!eth)) {
4701 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
4702 __func__, napi->dev->name);
4703 napi_reuse_skb(napi, skb);
4707 gro_pull_from_frag0(skb, hlen);
4708 NAPI_GRO_CB(skb)->frag0 += hlen;
4709 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4711 __skb_pull(skb, hlen);
4714 * This works because the only protocols we care about don't require
4716 * We'll fix it up properly in napi_frags_finish()
4718 skb->protocol = eth->h_proto;
4723 gro_result_t napi_gro_frags(struct napi_struct *napi)
4725 struct sk_buff *skb = napi_frags_skb(napi);
4730 trace_napi_gro_frags_entry(skb);
4732 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4734 EXPORT_SYMBOL(napi_gro_frags);
4736 /* Compute the checksum from gro_offset and return the folded value
4737 * after adding in any pseudo checksum.
4739 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4744 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4746 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4747 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4749 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4750 !skb->csum_complete_sw)
4751 netdev_rx_csum_fault(skb->dev);
4754 NAPI_GRO_CB(skb)->csum = wsum;
4755 NAPI_GRO_CB(skb)->csum_valid = 1;
4759 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4762 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4763 * Note: called with local irq disabled, but exits with local irq enabled.
4765 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4768 struct softnet_data *remsd = sd->rps_ipi_list;
4771 sd->rps_ipi_list = NULL;
4775 /* Send pending IPI's to kick RPS processing on remote cpus. */
4777 struct softnet_data *next = remsd->rps_ipi_next;
4779 if (cpu_online(remsd->cpu))
4780 smp_call_function_single_async(remsd->cpu,
4789 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4792 return sd->rps_ipi_list != NULL;
4798 static int process_backlog(struct napi_struct *napi, int quota)
4800 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4804 /* Check if we have pending ipi, its better to send them now,
4805 * not waiting net_rx_action() end.
4807 if (sd_has_rps_ipi_waiting(sd)) {
4808 local_irq_disable();
4809 net_rps_action_and_irq_enable(sd);
4812 napi->weight = dev_rx_weight;
4814 struct sk_buff *skb;
4816 while ((skb = __skb_dequeue(&sd->process_queue))) {
4818 __netif_receive_skb(skb);
4820 input_queue_head_incr(sd);
4821 if (++work >= quota)
4826 local_irq_disable();
4828 if (skb_queue_empty(&sd->input_pkt_queue)) {
4830 * Inline a custom version of __napi_complete().
4831 * only current cpu owns and manipulates this napi,
4832 * and NAPI_STATE_SCHED is the only possible flag set
4834 * We can use a plain write instead of clear_bit(),
4835 * and we dont need an smp_mb() memory barrier.
4840 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4841 &sd->process_queue);
4851 * __napi_schedule - schedule for receive
4852 * @n: entry to schedule
4854 * The entry's receive function will be scheduled to run.
4855 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4857 void __napi_schedule(struct napi_struct *n)
4859 unsigned long flags;
4861 local_irq_save(flags);
4862 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4863 local_irq_restore(flags);
4865 EXPORT_SYMBOL(__napi_schedule);
4868 * __napi_schedule_irqoff - schedule for receive
4869 * @n: entry to schedule
4871 * Variant of __napi_schedule() assuming hard irqs are masked
4873 void __napi_schedule_irqoff(struct napi_struct *n)
4875 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4877 EXPORT_SYMBOL(__napi_schedule_irqoff);
4879 bool __napi_complete(struct napi_struct *n)
4881 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4883 /* Some drivers call us directly, instead of calling
4884 * napi_complete_done().
4886 if (unlikely(test_bit(NAPI_STATE_IN_BUSY_POLL, &n->state)))
4889 list_del_init(&n->poll_list);
4890 smp_mb__before_atomic();
4891 clear_bit(NAPI_STATE_SCHED, &n->state);
4894 EXPORT_SYMBOL(__napi_complete);
4896 bool napi_complete_done(struct napi_struct *n, int work_done)
4898 unsigned long flags;
4901 * 1) Don't let napi dequeue from the cpu poll list
4902 * just in case its running on a different cpu.
4903 * 2) If we are busy polling, do nothing here, we have
4904 * the guarantee we will be called later.
4906 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
4907 NAPIF_STATE_IN_BUSY_POLL)))
4911 unsigned long timeout = 0;
4914 timeout = n->dev->gro_flush_timeout;
4917 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4918 HRTIMER_MODE_REL_PINNED);
4920 napi_gro_flush(n, false);
4922 if (likely(list_empty(&n->poll_list))) {
4923 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4925 /* If n->poll_list is not empty, we need to mask irqs */
4926 local_irq_save(flags);
4928 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)
4949 #define BUSY_POLL_BUDGET 8
4951 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
4955 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
4959 /* All we really want here is to re-enable device interrupts.
4960 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
4962 rc = napi->poll(napi, BUSY_POLL_BUDGET);
4963 netpoll_poll_unlock(have_poll_lock);
4964 if (rc == BUSY_POLL_BUDGET)
4965 __napi_schedule(napi);
4967 if (local_softirq_pending())
4971 bool sk_busy_loop(struct sock *sk, int nonblock)
4973 unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0;
4974 int (*napi_poll)(struct napi_struct *napi, int budget);
4975 int (*busy_poll)(struct napi_struct *dev);
4976 void *have_poll_lock = NULL;
4977 struct napi_struct *napi;
4986 napi = napi_by_id(sk->sk_napi_id);
4990 /* Note: ndo_busy_poll method is optional in linux-4.5 */
4991 busy_poll = napi->dev->netdev_ops->ndo_busy_poll;
4998 rc = busy_poll(napi);
5002 unsigned long val = READ_ONCE(napi->state);
5004 /* If multiple threads are competing for this napi,
5005 * we avoid dirtying napi->state as much as we can.
5007 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
5008 NAPIF_STATE_IN_BUSY_POLL))
5010 if (cmpxchg(&napi->state, val,
5011 val | NAPIF_STATE_IN_BUSY_POLL |
5012 NAPIF_STATE_SCHED) != val)
5014 have_poll_lock = netpoll_poll_lock(napi);
5015 napi_poll = napi->poll;
5017 rc = napi_poll(napi, BUSY_POLL_BUDGET);
5018 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
5021 __NET_ADD_STATS(sock_net(sk),
5022 LINUX_MIB_BUSYPOLLRXPACKETS, rc);
5025 if (rc == LL_FLUSH_FAILED)
5026 break; /* permanent failure */
5028 if (nonblock || !skb_queue_empty(&sk->sk_receive_queue) ||
5029 busy_loop_timeout(end_time))
5032 if (unlikely(need_resched())) {
5034 busy_poll_stop(napi, have_poll_lock);
5038 rc = !skb_queue_empty(&sk->sk_receive_queue);
5039 if (rc || busy_loop_timeout(end_time))
5046 busy_poll_stop(napi, have_poll_lock);
5048 rc = !skb_queue_empty(&sk->sk_receive_queue);
5053 EXPORT_SYMBOL(sk_busy_loop);
5055 #endif /* CONFIG_NET_RX_BUSY_POLL */
5057 static void napi_hash_add(struct napi_struct *napi)
5059 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
5060 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
5063 spin_lock(&napi_hash_lock);
5065 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */
5067 if (unlikely(++napi_gen_id < NR_CPUS + 1))
5068 napi_gen_id = NR_CPUS + 1;
5069 } while (napi_by_id(napi_gen_id));
5070 napi->napi_id = napi_gen_id;
5072 hlist_add_head_rcu(&napi->napi_hash_node,
5073 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
5075 spin_unlock(&napi_hash_lock);
5078 /* Warning : caller is responsible to make sure rcu grace period
5079 * is respected before freeing memory containing @napi
5081 bool napi_hash_del(struct napi_struct *napi)
5083 bool rcu_sync_needed = false;
5085 spin_lock(&napi_hash_lock);
5087 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
5088 rcu_sync_needed = true;
5089 hlist_del_rcu(&napi->napi_hash_node);
5091 spin_unlock(&napi_hash_lock);
5092 return rcu_sync_needed;
5094 EXPORT_SYMBOL_GPL(napi_hash_del);
5096 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
5098 struct napi_struct *napi;
5100 napi = container_of(timer, struct napi_struct, timer);
5102 napi_schedule(napi);
5104 return HRTIMER_NORESTART;
5107 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
5108 int (*poll)(struct napi_struct *, int), int weight)
5110 INIT_LIST_HEAD(&napi->poll_list);
5111 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
5112 napi->timer.function = napi_watchdog;
5113 napi->gro_count = 0;
5114 napi->gro_list = NULL;
5117 if (weight > NAPI_POLL_WEIGHT)
5118 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5120 napi->weight = weight;
5121 list_add(&napi->dev_list, &dev->napi_list);
5123 #ifdef CONFIG_NETPOLL
5124 napi->poll_owner = -1;
5126 set_bit(NAPI_STATE_SCHED, &napi->state);
5127 napi_hash_add(napi);
5129 EXPORT_SYMBOL(netif_napi_add);
5131 void napi_disable(struct napi_struct *n)
5134 set_bit(NAPI_STATE_DISABLE, &n->state);
5136 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
5138 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
5141 hrtimer_cancel(&n->timer);
5143 clear_bit(NAPI_STATE_DISABLE, &n->state);
5145 EXPORT_SYMBOL(napi_disable);
5147 /* Must be called in process context */
5148 void netif_napi_del(struct napi_struct *napi)
5151 if (napi_hash_del(napi))
5153 list_del_init(&napi->dev_list);
5154 napi_free_frags(napi);
5156 kfree_skb_list(napi->gro_list);
5157 napi->gro_list = NULL;
5158 napi->gro_count = 0;
5160 EXPORT_SYMBOL(netif_napi_del);
5162 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
5167 list_del_init(&n->poll_list);
5169 have = netpoll_poll_lock(n);
5173 /* This NAPI_STATE_SCHED test is for avoiding a race
5174 * with netpoll's poll_napi(). Only the entity which
5175 * obtains the lock and sees NAPI_STATE_SCHED set will
5176 * actually make the ->poll() call. Therefore we avoid
5177 * accidentally calling ->poll() when NAPI is not scheduled.
5180 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
5181 work = n->poll(n, weight);
5182 trace_napi_poll(n, work, weight);
5185 WARN_ON_ONCE(work > weight);
5187 if (likely(work < weight))
5190 /* Drivers must not modify the NAPI state if they
5191 * consume the entire weight. In such cases this code
5192 * still "owns" the NAPI instance and therefore can
5193 * move the instance around on the list at-will.
5195 if (unlikely(napi_disable_pending(n))) {
5201 /* flush too old packets
5202 * If HZ < 1000, flush all packets.
5204 napi_gro_flush(n, HZ >= 1000);
5207 /* Some drivers may have called napi_schedule
5208 * prior to exhausting their budget.
5210 if (unlikely(!list_empty(&n->poll_list))) {
5211 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5212 n->dev ? n->dev->name : "backlog");
5216 list_add_tail(&n->poll_list, repoll);
5219 netpoll_poll_unlock(have);
5224 static __latent_entropy void net_rx_action(struct softirq_action *h)
5226 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5227 unsigned long time_limit = jiffies + 2;
5228 int budget = netdev_budget;
5232 local_irq_disable();
5233 list_splice_init(&sd->poll_list, &list);
5237 struct napi_struct *n;
5239 if (list_empty(&list)) {
5240 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
5245 n = list_first_entry(&list, struct napi_struct, poll_list);
5246 budget -= napi_poll(n, &repoll);
5248 /* If softirq window is exhausted then punt.
5249 * Allow this to run for 2 jiffies since which will allow
5250 * an average latency of 1.5/HZ.
5252 if (unlikely(budget <= 0 ||
5253 time_after_eq(jiffies, time_limit))) {
5259 local_irq_disable();
5261 list_splice_tail_init(&sd->poll_list, &list);
5262 list_splice_tail(&repoll, &list);
5263 list_splice(&list, &sd->poll_list);
5264 if (!list_empty(&sd->poll_list))
5265 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5267 net_rps_action_and_irq_enable(sd);
5269 __kfree_skb_flush();
5272 struct netdev_adjacent {
5273 struct net_device *dev;
5275 /* upper master flag, there can only be one master device per list */
5278 /* counter for the number of times this device was added to us */
5281 /* private field for the users */
5284 struct list_head list;
5285 struct rcu_head rcu;
5288 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
5289 struct list_head *adj_list)
5291 struct netdev_adjacent *adj;
5293 list_for_each_entry(adj, adj_list, list) {
5294 if (adj->dev == adj_dev)
5300 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
5302 struct net_device *dev = data;
5304 return upper_dev == dev;
5308 * netdev_has_upper_dev - Check if device is linked to an upper device
5310 * @upper_dev: upper device to check
5312 * Find out if a device is linked to specified upper device and return true
5313 * in case it is. Note that this checks only immediate upper device,
5314 * not through a complete stack of devices. The caller must hold the RTNL lock.
5316 bool netdev_has_upper_dev(struct net_device *dev,
5317 struct net_device *upper_dev)
5321 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5324 EXPORT_SYMBOL(netdev_has_upper_dev);
5327 * netdev_has_upper_dev_all - Check if device is linked to an upper device
5329 * @upper_dev: upper device to check
5331 * Find out if a device is linked to specified upper device and return true
5332 * in case it is. Note that this checks the entire upper device chain.
5333 * The caller must hold rcu lock.
5336 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
5337 struct net_device *upper_dev)
5339 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5342 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
5345 * netdev_has_any_upper_dev - Check if device is linked to some device
5348 * Find out if a device is linked to an upper device and return true in case
5349 * it is. The caller must hold the RTNL lock.
5351 static bool netdev_has_any_upper_dev(struct net_device *dev)
5355 return !list_empty(&dev->adj_list.upper);
5359 * netdev_master_upper_dev_get - Get master upper device
5362 * Find a master upper device and return pointer to it or NULL in case
5363 * it's not there. The caller must hold the RTNL lock.
5365 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5367 struct netdev_adjacent *upper;
5371 if (list_empty(&dev->adj_list.upper))
5374 upper = list_first_entry(&dev->adj_list.upper,
5375 struct netdev_adjacent, list);
5376 if (likely(upper->master))
5380 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5383 * netdev_has_any_lower_dev - Check if device is linked to some device
5386 * Find out if a device is linked to a lower device and return true in case
5387 * it is. The caller must hold the RTNL lock.
5389 static bool netdev_has_any_lower_dev(struct net_device *dev)
5393 return !list_empty(&dev->adj_list.lower);
5396 void *netdev_adjacent_get_private(struct list_head *adj_list)
5398 struct netdev_adjacent *adj;
5400 adj = list_entry(adj_list, struct netdev_adjacent, list);
5402 return adj->private;
5404 EXPORT_SYMBOL(netdev_adjacent_get_private);
5407 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5409 * @iter: list_head ** of the current position
5411 * Gets the next device from the dev's upper list, starting from iter
5412 * position. The caller must hold RCU read lock.
5414 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5415 struct list_head **iter)
5417 struct netdev_adjacent *upper;
5419 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5421 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5423 if (&upper->list == &dev->adj_list.upper)
5426 *iter = &upper->list;
5430 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5432 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
5433 struct list_head **iter)
5435 struct netdev_adjacent *upper;
5437 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5439 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5441 if (&upper->list == &dev->adj_list.upper)
5444 *iter = &upper->list;
5449 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
5450 int (*fn)(struct net_device *dev,
5454 struct net_device *udev;
5455 struct list_head *iter;
5458 for (iter = &dev->adj_list.upper,
5459 udev = netdev_next_upper_dev_rcu(dev, &iter);
5461 udev = netdev_next_upper_dev_rcu(dev, &iter)) {
5462 /* first is the upper device itself */
5463 ret = fn(udev, data);
5467 /* then look at all of its upper devices */
5468 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data);
5475 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
5478 * netdev_lower_get_next_private - Get the next ->private from the
5479 * lower neighbour list
5481 * @iter: list_head ** of the current position
5483 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5484 * list, starting from iter position. The caller must hold either hold the
5485 * RTNL lock or its own locking that guarantees that the neighbour lower
5486 * list will remain unchanged.
5488 void *netdev_lower_get_next_private(struct net_device *dev,
5489 struct list_head **iter)
5491 struct netdev_adjacent *lower;
5493 lower = list_entry(*iter, struct netdev_adjacent, list);
5495 if (&lower->list == &dev->adj_list.lower)
5498 *iter = lower->list.next;
5500 return lower->private;
5502 EXPORT_SYMBOL(netdev_lower_get_next_private);
5505 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5506 * lower neighbour list, RCU
5509 * @iter: list_head ** of the current position
5511 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5512 * list, starting from iter position. The caller must hold RCU read lock.
5514 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5515 struct list_head **iter)
5517 struct netdev_adjacent *lower;
5519 WARN_ON_ONCE(!rcu_read_lock_held());
5521 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5523 if (&lower->list == &dev->adj_list.lower)
5526 *iter = &lower->list;
5528 return lower->private;
5530 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5533 * netdev_lower_get_next - Get the next device from the lower neighbour
5536 * @iter: list_head ** of the current position
5538 * Gets the next netdev_adjacent from the dev's lower neighbour
5539 * list, starting from iter position. The caller must hold RTNL lock or
5540 * its own locking that guarantees that the neighbour lower
5541 * list will remain unchanged.
5543 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5545 struct netdev_adjacent *lower;
5547 lower = list_entry(*iter, struct netdev_adjacent, list);
5549 if (&lower->list == &dev->adj_list.lower)
5552 *iter = lower->list.next;
5556 EXPORT_SYMBOL(netdev_lower_get_next);
5558 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
5559 struct list_head **iter)
5561 struct netdev_adjacent *lower;
5563 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5565 if (&lower->list == &dev->adj_list.lower)
5568 *iter = &lower->list;
5573 int netdev_walk_all_lower_dev(struct net_device *dev,
5574 int (*fn)(struct net_device *dev,
5578 struct net_device *ldev;
5579 struct list_head *iter;
5582 for (iter = &dev->adj_list.lower,
5583 ldev = netdev_next_lower_dev(dev, &iter);
5585 ldev = netdev_next_lower_dev(dev, &iter)) {
5586 /* first is the lower device itself */
5587 ret = fn(ldev, data);
5591 /* then look at all of its lower devices */
5592 ret = netdev_walk_all_lower_dev(ldev, fn, data);
5599 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
5601 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
5602 struct list_head **iter)
5604 struct netdev_adjacent *lower;
5606 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5607 if (&lower->list == &dev->adj_list.lower)
5610 *iter = &lower->list;
5615 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
5616 int (*fn)(struct net_device *dev,
5620 struct net_device *ldev;
5621 struct list_head *iter;
5624 for (iter = &dev->adj_list.lower,
5625 ldev = netdev_next_lower_dev_rcu(dev, &iter);
5627 ldev = netdev_next_lower_dev_rcu(dev, &iter)) {
5628 /* first is the lower device itself */
5629 ret = fn(ldev, data);
5633 /* then look at all of its lower devices */
5634 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data);
5641 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
5644 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5645 * lower neighbour list, RCU
5649 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5650 * list. The caller must hold RCU read lock.
5652 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5654 struct netdev_adjacent *lower;
5656 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5657 struct netdev_adjacent, list);
5659 return lower->private;
5662 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5665 * netdev_master_upper_dev_get_rcu - Get master upper device
5668 * Find a master upper device and return pointer to it or NULL in case
5669 * it's not there. The caller must hold the RCU read lock.
5671 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5673 struct netdev_adjacent *upper;
5675 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5676 struct netdev_adjacent, list);
5677 if (upper && likely(upper->master))
5681 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5683 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5684 struct net_device *adj_dev,
5685 struct list_head *dev_list)
5687 char linkname[IFNAMSIZ+7];
5688 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5689 "upper_%s" : "lower_%s", adj_dev->name);
5690 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5693 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5695 struct list_head *dev_list)
5697 char linkname[IFNAMSIZ+7];
5698 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5699 "upper_%s" : "lower_%s", name);
5700 sysfs_remove_link(&(dev->dev.kobj), linkname);
5703 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5704 struct net_device *adj_dev,
5705 struct list_head *dev_list)
5707 return (dev_list == &dev->adj_list.upper ||
5708 dev_list == &dev->adj_list.lower) &&
5709 net_eq(dev_net(dev), dev_net(adj_dev));
5712 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5713 struct net_device *adj_dev,
5714 struct list_head *dev_list,
5715 void *private, bool master)
5717 struct netdev_adjacent *adj;
5720 adj = __netdev_find_adj(adj_dev, dev_list);
5724 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
5725 dev->name, adj_dev->name, adj->ref_nr);
5730 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5735 adj->master = master;
5737 adj->private = private;
5740 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
5741 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
5743 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5744 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5749 /* Ensure that master link is always the first item in list. */
5751 ret = sysfs_create_link(&(dev->dev.kobj),
5752 &(adj_dev->dev.kobj), "master");
5754 goto remove_symlinks;
5756 list_add_rcu(&adj->list, dev_list);
5758 list_add_tail_rcu(&adj->list, dev_list);
5764 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5765 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5773 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5774 struct net_device *adj_dev,
5776 struct list_head *dev_list)
5778 struct netdev_adjacent *adj;
5780 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
5781 dev->name, adj_dev->name, ref_nr);
5783 adj = __netdev_find_adj(adj_dev, dev_list);
5786 pr_err("Adjacency does not exist for device %s from %s\n",
5787 dev->name, adj_dev->name);
5792 if (adj->ref_nr > ref_nr) {
5793 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
5794 dev->name, adj_dev->name, ref_nr,
5795 adj->ref_nr - ref_nr);
5796 adj->ref_nr -= ref_nr;
5801 sysfs_remove_link(&(dev->dev.kobj), "master");
5803 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5804 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5806 list_del_rcu(&adj->list);
5807 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
5808 adj_dev->name, dev->name, adj_dev->name);
5810 kfree_rcu(adj, rcu);
5813 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5814 struct net_device *upper_dev,
5815 struct list_head *up_list,
5816 struct list_head *down_list,
5817 void *private, bool master)
5821 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
5826 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
5829 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
5836 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5837 struct net_device *upper_dev,
5839 struct list_head *up_list,
5840 struct list_head *down_list)
5842 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
5843 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
5846 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5847 struct net_device *upper_dev,
5848 void *private, bool master)
5850 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5851 &dev->adj_list.upper,
5852 &upper_dev->adj_list.lower,
5856 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5857 struct net_device *upper_dev)
5859 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
5860 &dev->adj_list.upper,
5861 &upper_dev->adj_list.lower);
5864 static int __netdev_upper_dev_link(struct net_device *dev,
5865 struct net_device *upper_dev, bool master,
5866 void *upper_priv, void *upper_info)
5868 struct netdev_notifier_changeupper_info changeupper_info;
5873 if (dev == upper_dev)
5876 /* To prevent loops, check if dev is not upper device to upper_dev. */
5877 if (netdev_has_upper_dev(upper_dev, dev))
5880 if (netdev_has_upper_dev(dev, upper_dev))
5883 if (master && netdev_master_upper_dev_get(dev))
5886 changeupper_info.upper_dev = upper_dev;
5887 changeupper_info.master = master;
5888 changeupper_info.linking = true;
5889 changeupper_info.upper_info = upper_info;
5891 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5892 &changeupper_info.info);
5893 ret = notifier_to_errno(ret);
5897 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
5902 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5903 &changeupper_info.info);
5904 ret = notifier_to_errno(ret);
5911 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5917 * netdev_upper_dev_link - Add a link to the upper device
5919 * @upper_dev: new upper device
5921 * Adds a link to device which is upper to this one. The caller must hold
5922 * the RTNL lock. On a failure a negative errno code is returned.
5923 * On success the reference counts are adjusted and the function
5926 int netdev_upper_dev_link(struct net_device *dev,
5927 struct net_device *upper_dev)
5929 return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL);
5931 EXPORT_SYMBOL(netdev_upper_dev_link);
5934 * netdev_master_upper_dev_link - Add a master link to the upper device
5936 * @upper_dev: new upper device
5937 * @upper_priv: upper device private
5938 * @upper_info: upper info to be passed down via notifier
5940 * Adds a link to device which is upper to this one. In this case, only
5941 * one master upper device can be linked, although other non-master devices
5942 * might be linked as well. The caller must hold the RTNL lock.
5943 * On a failure a negative errno code is returned. On success the reference
5944 * counts are adjusted and the function returns zero.
5946 int netdev_master_upper_dev_link(struct net_device *dev,
5947 struct net_device *upper_dev,
5948 void *upper_priv, void *upper_info)
5950 return __netdev_upper_dev_link(dev, upper_dev, true,
5951 upper_priv, upper_info);
5953 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5956 * netdev_upper_dev_unlink - Removes a link to upper device
5958 * @upper_dev: new upper device
5960 * Removes a link to device which is upper to this one. The caller must hold
5963 void netdev_upper_dev_unlink(struct net_device *dev,
5964 struct net_device *upper_dev)
5966 struct netdev_notifier_changeupper_info changeupper_info;
5969 changeupper_info.upper_dev = upper_dev;
5970 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
5971 changeupper_info.linking = false;
5973 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5974 &changeupper_info.info);
5976 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5978 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5979 &changeupper_info.info);
5981 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5984 * netdev_bonding_info_change - Dispatch event about slave change
5986 * @bonding_info: info to dispatch
5988 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5989 * The caller must hold the RTNL lock.
5991 void netdev_bonding_info_change(struct net_device *dev,
5992 struct netdev_bonding_info *bonding_info)
5994 struct netdev_notifier_bonding_info info;
5996 memcpy(&info.bonding_info, bonding_info,
5997 sizeof(struct netdev_bonding_info));
5998 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
6001 EXPORT_SYMBOL(netdev_bonding_info_change);
6003 static void netdev_adjacent_add_links(struct net_device *dev)
6005 struct netdev_adjacent *iter;
6007 struct net *net = dev_net(dev);
6009 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6010 if (!net_eq(net, dev_net(iter->dev)))
6012 netdev_adjacent_sysfs_add(iter->dev, dev,
6013 &iter->dev->adj_list.lower);
6014 netdev_adjacent_sysfs_add(dev, iter->dev,
6015 &dev->adj_list.upper);
6018 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6019 if (!net_eq(net, dev_net(iter->dev)))
6021 netdev_adjacent_sysfs_add(iter->dev, dev,
6022 &iter->dev->adj_list.upper);
6023 netdev_adjacent_sysfs_add(dev, iter->dev,
6024 &dev->adj_list.lower);
6028 static void netdev_adjacent_del_links(struct net_device *dev)
6030 struct netdev_adjacent *iter;
6032 struct net *net = dev_net(dev);
6034 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6035 if (!net_eq(net, dev_net(iter->dev)))
6037 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6038 &iter->dev->adj_list.lower);
6039 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6040 &dev->adj_list.upper);
6043 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6044 if (!net_eq(net, dev_net(iter->dev)))
6046 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6047 &iter->dev->adj_list.upper);
6048 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6049 &dev->adj_list.lower);
6053 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
6055 struct netdev_adjacent *iter;
6057 struct net *net = dev_net(dev);
6059 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6060 if (!net_eq(net, dev_net(iter->dev)))
6062 netdev_adjacent_sysfs_del(iter->dev, oldname,
6063 &iter->dev->adj_list.lower);
6064 netdev_adjacent_sysfs_add(iter->dev, dev,
6065 &iter->dev->adj_list.lower);
6068 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6069 if (!net_eq(net, dev_net(iter->dev)))
6071 netdev_adjacent_sysfs_del(iter->dev, oldname,
6072 &iter->dev->adj_list.upper);
6073 netdev_adjacent_sysfs_add(iter->dev, dev,
6074 &iter->dev->adj_list.upper);
6078 void *netdev_lower_dev_get_private(struct net_device *dev,
6079 struct net_device *lower_dev)
6081 struct netdev_adjacent *lower;
6085 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
6089 return lower->private;
6091 EXPORT_SYMBOL(netdev_lower_dev_get_private);
6094 int dev_get_nest_level(struct net_device *dev)
6096 struct net_device *lower = NULL;
6097 struct list_head *iter;
6103 netdev_for_each_lower_dev(dev, lower, iter) {
6104 nest = dev_get_nest_level(lower);
6105 if (max_nest < nest)
6109 return max_nest + 1;
6111 EXPORT_SYMBOL(dev_get_nest_level);
6114 * netdev_lower_change - Dispatch event about lower device state change
6115 * @lower_dev: device
6116 * @lower_state_info: state to dispatch
6118 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
6119 * The caller must hold the RTNL lock.
6121 void netdev_lower_state_changed(struct net_device *lower_dev,
6122 void *lower_state_info)
6124 struct netdev_notifier_changelowerstate_info changelowerstate_info;
6127 changelowerstate_info.lower_state_info = lower_state_info;
6128 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev,
6129 &changelowerstate_info.info);
6131 EXPORT_SYMBOL(netdev_lower_state_changed);
6133 int netdev_default_l2upper_neigh_construct(struct net_device *dev,
6134 struct neighbour *n)
6136 struct net_device *lower_dev, *stop_dev;
6137 struct list_head *iter;
6140 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6141 if (!lower_dev->netdev_ops->ndo_neigh_construct)
6143 err = lower_dev->netdev_ops->ndo_neigh_construct(lower_dev, n);
6145 stop_dev = lower_dev;
6152 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6153 if (lower_dev == stop_dev)
6155 if (!lower_dev->netdev_ops->ndo_neigh_destroy)
6157 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n);
6161 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_construct);
6163 void netdev_default_l2upper_neigh_destroy(struct net_device *dev,
6164 struct neighbour *n)
6166 struct net_device *lower_dev;
6167 struct list_head *iter;
6169 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6170 if (!lower_dev->netdev_ops->ndo_neigh_destroy)
6172 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n);
6175 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_destroy);
6177 static void dev_change_rx_flags(struct net_device *dev, int flags)
6179 const struct net_device_ops *ops = dev->netdev_ops;
6181 if (ops->ndo_change_rx_flags)
6182 ops->ndo_change_rx_flags(dev, flags);
6185 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
6187 unsigned int old_flags = dev->flags;
6193 dev->flags |= IFF_PROMISC;
6194 dev->promiscuity += inc;
6195 if (dev->promiscuity == 0) {
6198 * If inc causes overflow, untouch promisc and return error.
6201 dev->flags &= ~IFF_PROMISC;
6203 dev->promiscuity -= inc;
6204 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6209 if (dev->flags != old_flags) {
6210 pr_info("device %s %s promiscuous mode\n",
6212 dev->flags & IFF_PROMISC ? "entered" : "left");
6213 if (audit_enabled) {
6214 current_uid_gid(&uid, &gid);
6215 audit_log(current->audit_context, GFP_ATOMIC,
6216 AUDIT_ANOM_PROMISCUOUS,
6217 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6218 dev->name, (dev->flags & IFF_PROMISC),
6219 (old_flags & IFF_PROMISC),
6220 from_kuid(&init_user_ns, audit_get_loginuid(current)),
6221 from_kuid(&init_user_ns, uid),
6222 from_kgid(&init_user_ns, gid),
6223 audit_get_sessionid(current));
6226 dev_change_rx_flags(dev, IFF_PROMISC);
6229 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
6234 * dev_set_promiscuity - update promiscuity count on a device
6238 * Add or remove promiscuity from a device. While the count in the device
6239 * remains above zero the interface remains promiscuous. Once it hits zero
6240 * the device reverts back to normal filtering operation. A negative inc
6241 * value is used to drop promiscuity on the device.
6242 * Return 0 if successful or a negative errno code on error.
6244 int dev_set_promiscuity(struct net_device *dev, int inc)
6246 unsigned int old_flags = dev->flags;
6249 err = __dev_set_promiscuity(dev, inc, true);
6252 if (dev->flags != old_flags)
6253 dev_set_rx_mode(dev);
6256 EXPORT_SYMBOL(dev_set_promiscuity);
6258 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
6260 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
6264 dev->flags |= IFF_ALLMULTI;
6265 dev->allmulti += inc;
6266 if (dev->allmulti == 0) {
6269 * If inc causes overflow, untouch allmulti and return error.
6272 dev->flags &= ~IFF_ALLMULTI;
6274 dev->allmulti -= inc;
6275 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6280 if (dev->flags ^ old_flags) {
6281 dev_change_rx_flags(dev, IFF_ALLMULTI);
6282 dev_set_rx_mode(dev);
6284 __dev_notify_flags(dev, old_flags,
6285 dev->gflags ^ old_gflags);
6291 * dev_set_allmulti - update allmulti count on a device
6295 * Add or remove reception of all multicast frames to a device. While the
6296 * count in the device remains above zero the interface remains listening
6297 * to all interfaces. Once it hits zero the device reverts back to normal
6298 * filtering operation. A negative @inc value is used to drop the counter
6299 * when releasing a resource needing all multicasts.
6300 * Return 0 if successful or a negative errno code on error.
6303 int dev_set_allmulti(struct net_device *dev, int inc)
6305 return __dev_set_allmulti(dev, inc, true);
6307 EXPORT_SYMBOL(dev_set_allmulti);
6310 * Upload unicast and multicast address lists to device and
6311 * configure RX filtering. When the device doesn't support unicast
6312 * filtering it is put in promiscuous mode while unicast addresses
6315 void __dev_set_rx_mode(struct net_device *dev)
6317 const struct net_device_ops *ops = dev->netdev_ops;
6319 /* dev_open will call this function so the list will stay sane. */
6320 if (!(dev->flags&IFF_UP))
6323 if (!netif_device_present(dev))
6326 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
6327 /* Unicast addresses changes may only happen under the rtnl,
6328 * therefore calling __dev_set_promiscuity here is safe.
6330 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
6331 __dev_set_promiscuity(dev, 1, false);
6332 dev->uc_promisc = true;
6333 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
6334 __dev_set_promiscuity(dev, -1, false);
6335 dev->uc_promisc = false;
6339 if (ops->ndo_set_rx_mode)
6340 ops->ndo_set_rx_mode(dev);
6343 void dev_set_rx_mode(struct net_device *dev)
6345 netif_addr_lock_bh(dev);
6346 __dev_set_rx_mode(dev);
6347 netif_addr_unlock_bh(dev);
6351 * dev_get_flags - get flags reported to userspace
6354 * Get the combination of flag bits exported through APIs to userspace.
6356 unsigned int dev_get_flags(const struct net_device *dev)
6360 flags = (dev->flags & ~(IFF_PROMISC |
6365 (dev->gflags & (IFF_PROMISC |
6368 if (netif_running(dev)) {
6369 if (netif_oper_up(dev))
6370 flags |= IFF_RUNNING;
6371 if (netif_carrier_ok(dev))
6372 flags |= IFF_LOWER_UP;
6373 if (netif_dormant(dev))
6374 flags |= IFF_DORMANT;
6379 EXPORT_SYMBOL(dev_get_flags);
6381 int __dev_change_flags(struct net_device *dev, unsigned int flags)
6383 unsigned int old_flags = dev->flags;
6389 * Set the flags on our device.
6392 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
6393 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
6395 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6399 * Load in the correct multicast list now the flags have changed.
6402 if ((old_flags ^ flags) & IFF_MULTICAST)
6403 dev_change_rx_flags(dev, IFF_MULTICAST);
6405 dev_set_rx_mode(dev);
6408 * Have we downed the interface. We handle IFF_UP ourselves
6409 * according to user attempts to set it, rather than blindly
6414 if ((old_flags ^ flags) & IFF_UP)
6415 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
6417 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6418 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6419 unsigned int old_flags = dev->flags;
6421 dev->gflags ^= IFF_PROMISC;
6423 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6424 if (dev->flags != old_flags)
6425 dev_set_rx_mode(dev);
6428 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6429 is important. Some (broken) drivers set IFF_PROMISC, when
6430 IFF_ALLMULTI is requested not asking us and not reporting.
6432 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6433 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6435 dev->gflags ^= IFF_ALLMULTI;
6436 __dev_set_allmulti(dev, inc, false);
6442 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6443 unsigned int gchanges)
6445 unsigned int changes = dev->flags ^ old_flags;
6448 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6450 if (changes & IFF_UP) {
6451 if (dev->flags & IFF_UP)
6452 call_netdevice_notifiers(NETDEV_UP, dev);
6454 call_netdevice_notifiers(NETDEV_DOWN, dev);
6457 if (dev->flags & IFF_UP &&
6458 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6459 struct netdev_notifier_change_info change_info;
6461 change_info.flags_changed = changes;
6462 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6468 * dev_change_flags - change device settings
6470 * @flags: device state flags
6472 * Change settings on device based state flags. The flags are
6473 * in the userspace exported format.
6475 int dev_change_flags(struct net_device *dev, unsigned int flags)
6478 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6480 ret = __dev_change_flags(dev, flags);
6484 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6485 __dev_notify_flags(dev, old_flags, changes);
6488 EXPORT_SYMBOL(dev_change_flags);
6490 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6492 const struct net_device_ops *ops = dev->netdev_ops;
6494 if (ops->ndo_change_mtu)
6495 return ops->ndo_change_mtu(dev, new_mtu);
6502 * dev_set_mtu - Change maximum transfer unit
6504 * @new_mtu: new transfer unit
6506 * Change the maximum transfer size of the network device.
6508 int dev_set_mtu(struct net_device *dev, int new_mtu)
6512 if (new_mtu == dev->mtu)
6515 /* MTU must be positive, and in range */
6516 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
6517 net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n",
6518 dev->name, new_mtu, dev->min_mtu);
6522 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
6523 net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n",
6524 dev->name, new_mtu, dev->max_mtu);
6528 if (!netif_device_present(dev))
6531 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6532 err = notifier_to_errno(err);
6536 orig_mtu = dev->mtu;
6537 err = __dev_set_mtu(dev, new_mtu);
6540 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6541 err = notifier_to_errno(err);
6543 /* setting mtu back and notifying everyone again,
6544 * so that they have a chance to revert changes.
6546 __dev_set_mtu(dev, orig_mtu);
6547 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6552 EXPORT_SYMBOL(dev_set_mtu);
6555 * dev_set_group - Change group this device belongs to
6557 * @new_group: group this device should belong to
6559 void dev_set_group(struct net_device *dev, int new_group)
6561 dev->group = new_group;
6563 EXPORT_SYMBOL(dev_set_group);
6566 * dev_set_mac_address - Change Media Access Control Address
6570 * Change the hardware (MAC) address of the device
6572 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6574 const struct net_device_ops *ops = dev->netdev_ops;
6577 if (!ops->ndo_set_mac_address)
6579 if (sa->sa_family != dev->type)
6581 if (!netif_device_present(dev))
6583 err = ops->ndo_set_mac_address(dev, sa);
6586 dev->addr_assign_type = NET_ADDR_SET;
6587 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6588 add_device_randomness(dev->dev_addr, dev->addr_len);
6591 EXPORT_SYMBOL(dev_set_mac_address);
6594 * dev_change_carrier - Change device carrier
6596 * @new_carrier: new value
6598 * Change device carrier
6600 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6602 const struct net_device_ops *ops = dev->netdev_ops;
6604 if (!ops->ndo_change_carrier)
6606 if (!netif_device_present(dev))
6608 return ops->ndo_change_carrier(dev, new_carrier);
6610 EXPORT_SYMBOL(dev_change_carrier);
6613 * dev_get_phys_port_id - Get device physical port ID
6617 * Get device physical port ID
6619 int dev_get_phys_port_id(struct net_device *dev,
6620 struct netdev_phys_item_id *ppid)
6622 const struct net_device_ops *ops = dev->netdev_ops;
6624 if (!ops->ndo_get_phys_port_id)
6626 return ops->ndo_get_phys_port_id(dev, ppid);
6628 EXPORT_SYMBOL(dev_get_phys_port_id);
6631 * dev_get_phys_port_name - Get device physical port name
6634 * @len: limit of bytes to copy to name
6636 * Get device physical port name
6638 int dev_get_phys_port_name(struct net_device *dev,
6639 char *name, size_t len)
6641 const struct net_device_ops *ops = dev->netdev_ops;
6643 if (!ops->ndo_get_phys_port_name)
6645 return ops->ndo_get_phys_port_name(dev, name, len);
6647 EXPORT_SYMBOL(dev_get_phys_port_name);
6650 * dev_change_proto_down - update protocol port state information
6652 * @proto_down: new value
6654 * This info can be used by switch drivers to set the phys state of the
6657 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6659 const struct net_device_ops *ops = dev->netdev_ops;
6661 if (!ops->ndo_change_proto_down)
6663 if (!netif_device_present(dev))
6665 return ops->ndo_change_proto_down(dev, proto_down);
6667 EXPORT_SYMBOL(dev_change_proto_down);
6670 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
6672 * @fd: new program fd or negative value to clear
6673 * @flags: xdp-related flags
6675 * Set or clear a bpf program for a device
6677 int dev_change_xdp_fd(struct net_device *dev, int fd, u32 flags)
6679 const struct net_device_ops *ops = dev->netdev_ops;
6680 struct bpf_prog *prog = NULL;
6681 struct netdev_xdp xdp;
6689 if (flags & XDP_FLAGS_UPDATE_IF_NOEXIST) {
6690 memset(&xdp, 0, sizeof(xdp));
6691 xdp.command = XDP_QUERY_PROG;
6693 err = ops->ndo_xdp(dev, &xdp);
6696 if (xdp.prog_attached)
6700 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP);
6702 return PTR_ERR(prog);
6705 memset(&xdp, 0, sizeof(xdp));
6706 xdp.command = XDP_SETUP_PROG;
6709 err = ops->ndo_xdp(dev, &xdp);
6710 if (err < 0 && prog)
6715 EXPORT_SYMBOL(dev_change_xdp_fd);
6718 * dev_new_index - allocate an ifindex
6719 * @net: the applicable net namespace
6721 * Returns a suitable unique value for a new device interface
6722 * number. The caller must hold the rtnl semaphore or the
6723 * dev_base_lock to be sure it remains unique.
6725 static int dev_new_index(struct net *net)
6727 int ifindex = net->ifindex;
6731 if (!__dev_get_by_index(net, ifindex))
6732 return net->ifindex = ifindex;
6736 /* Delayed registration/unregisteration */
6737 static LIST_HEAD(net_todo_list);
6738 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6740 static void net_set_todo(struct net_device *dev)
6742 list_add_tail(&dev->todo_list, &net_todo_list);
6743 dev_net(dev)->dev_unreg_count++;
6746 static void rollback_registered_many(struct list_head *head)
6748 struct net_device *dev, *tmp;
6749 LIST_HEAD(close_head);
6751 BUG_ON(dev_boot_phase);
6754 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6755 /* Some devices call without registering
6756 * for initialization unwind. Remove those
6757 * devices and proceed with the remaining.
6759 if (dev->reg_state == NETREG_UNINITIALIZED) {
6760 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6764 list_del(&dev->unreg_list);
6767 dev->dismantle = true;
6768 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6771 /* If device is running, close it first. */
6772 list_for_each_entry(dev, head, unreg_list)
6773 list_add_tail(&dev->close_list, &close_head);
6774 dev_close_many(&close_head, true);
6776 list_for_each_entry(dev, head, unreg_list) {
6777 /* And unlink it from device chain. */
6778 unlist_netdevice(dev);
6780 dev->reg_state = NETREG_UNREGISTERING;
6782 flush_all_backlogs();
6786 list_for_each_entry(dev, head, unreg_list) {
6787 struct sk_buff *skb = NULL;
6789 /* Shutdown queueing discipline. */
6793 /* Notify protocols, that we are about to destroy
6794 this device. They should clean all the things.
6796 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6798 if (!dev->rtnl_link_ops ||
6799 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6800 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6804 * Flush the unicast and multicast chains
6809 if (dev->netdev_ops->ndo_uninit)
6810 dev->netdev_ops->ndo_uninit(dev);
6813 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6815 /* Notifier chain MUST detach us all upper devices. */
6816 WARN_ON(netdev_has_any_upper_dev(dev));
6817 WARN_ON(netdev_has_any_lower_dev(dev));
6819 /* Remove entries from kobject tree */
6820 netdev_unregister_kobject(dev);
6822 /* Remove XPS queueing entries */
6823 netif_reset_xps_queues_gt(dev, 0);
6829 list_for_each_entry(dev, head, unreg_list)
6833 static void rollback_registered(struct net_device *dev)
6837 list_add(&dev->unreg_list, &single);
6838 rollback_registered_many(&single);
6842 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6843 struct net_device *upper, netdev_features_t features)
6845 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6846 netdev_features_t feature;
6849 for_each_netdev_feature(&upper_disables, feature_bit) {
6850 feature = __NETIF_F_BIT(feature_bit);
6851 if (!(upper->wanted_features & feature)
6852 && (features & feature)) {
6853 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6854 &feature, upper->name);
6855 features &= ~feature;
6862 static void netdev_sync_lower_features(struct net_device *upper,
6863 struct net_device *lower, netdev_features_t features)
6865 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6866 netdev_features_t feature;
6869 for_each_netdev_feature(&upper_disables, feature_bit) {
6870 feature = __NETIF_F_BIT(feature_bit);
6871 if (!(features & feature) && (lower->features & feature)) {
6872 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6873 &feature, lower->name);
6874 lower->wanted_features &= ~feature;
6875 netdev_update_features(lower);
6877 if (unlikely(lower->features & feature))
6878 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6879 &feature, lower->name);
6884 static netdev_features_t netdev_fix_features(struct net_device *dev,
6885 netdev_features_t features)
6887 /* Fix illegal checksum combinations */
6888 if ((features & NETIF_F_HW_CSUM) &&
6889 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6890 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6891 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6894 /* TSO requires that SG is present as well. */
6895 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6896 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6897 features &= ~NETIF_F_ALL_TSO;
6900 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6901 !(features & NETIF_F_IP_CSUM)) {
6902 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6903 features &= ~NETIF_F_TSO;
6904 features &= ~NETIF_F_TSO_ECN;
6907 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6908 !(features & NETIF_F_IPV6_CSUM)) {
6909 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6910 features &= ~NETIF_F_TSO6;
6913 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
6914 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
6915 features &= ~NETIF_F_TSO_MANGLEID;
6917 /* TSO ECN requires that TSO is present as well. */
6918 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6919 features &= ~NETIF_F_TSO_ECN;
6921 /* Software GSO depends on SG. */
6922 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6923 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6924 features &= ~NETIF_F_GSO;
6927 /* UFO needs SG and checksumming */
6928 if (features & NETIF_F_UFO) {
6929 /* maybe split UFO into V4 and V6? */
6930 if (!(features & NETIF_F_HW_CSUM) &&
6931 ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) !=
6932 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) {
6934 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6935 features &= ~NETIF_F_UFO;
6938 if (!(features & NETIF_F_SG)) {
6940 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6941 features &= ~NETIF_F_UFO;
6945 /* GSO partial features require GSO partial be set */
6946 if ((features & dev->gso_partial_features) &&
6947 !(features & NETIF_F_GSO_PARTIAL)) {
6949 "Dropping partially supported GSO features since no GSO partial.\n");
6950 features &= ~dev->gso_partial_features;
6953 #ifdef CONFIG_NET_RX_BUSY_POLL
6954 if (dev->netdev_ops->ndo_busy_poll)
6955 features |= NETIF_F_BUSY_POLL;
6958 features &= ~NETIF_F_BUSY_POLL;
6963 int __netdev_update_features(struct net_device *dev)
6965 struct net_device *upper, *lower;
6966 netdev_features_t features;
6967 struct list_head *iter;
6972 features = netdev_get_wanted_features(dev);
6974 if (dev->netdev_ops->ndo_fix_features)
6975 features = dev->netdev_ops->ndo_fix_features(dev, features);
6977 /* driver might be less strict about feature dependencies */
6978 features = netdev_fix_features(dev, features);
6980 /* some features can't be enabled if they're off an an upper device */
6981 netdev_for_each_upper_dev_rcu(dev, upper, iter)
6982 features = netdev_sync_upper_features(dev, upper, features);
6984 if (dev->features == features)
6987 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6988 &dev->features, &features);
6990 if (dev->netdev_ops->ndo_set_features)
6991 err = dev->netdev_ops->ndo_set_features(dev, features);
6995 if (unlikely(err < 0)) {
6997 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6998 err, &features, &dev->features);
6999 /* return non-0 since some features might have changed and
7000 * it's better to fire a spurious notification than miss it
7006 /* some features must be disabled on lower devices when disabled
7007 * on an upper device (think: bonding master or bridge)
7009 netdev_for_each_lower_dev(dev, lower, iter)
7010 netdev_sync_lower_features(dev, lower, features);
7013 dev->features = features;
7015 return err < 0 ? 0 : 1;
7019 * netdev_update_features - recalculate device features
7020 * @dev: the device to check
7022 * Recalculate dev->features set and send notifications if it
7023 * has changed. Should be called after driver or hardware dependent
7024 * conditions might have changed that influence the features.
7026 void netdev_update_features(struct net_device *dev)
7028 if (__netdev_update_features(dev))
7029 netdev_features_change(dev);
7031 EXPORT_SYMBOL(netdev_update_features);
7034 * netdev_change_features - recalculate device features
7035 * @dev: the device to check
7037 * Recalculate dev->features set and send notifications even
7038 * if they have not changed. Should be called instead of
7039 * netdev_update_features() if also dev->vlan_features might
7040 * have changed to allow the changes to be propagated to stacked
7043 void netdev_change_features(struct net_device *dev)
7045 __netdev_update_features(dev);
7046 netdev_features_change(dev);
7048 EXPORT_SYMBOL(netdev_change_features);
7051 * netif_stacked_transfer_operstate - transfer operstate
7052 * @rootdev: the root or lower level device to transfer state from
7053 * @dev: the device to transfer operstate to
7055 * Transfer operational state from root to device. This is normally
7056 * called when a stacking relationship exists between the root
7057 * device and the device(a leaf device).
7059 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
7060 struct net_device *dev)
7062 if (rootdev->operstate == IF_OPER_DORMANT)
7063 netif_dormant_on(dev);
7065 netif_dormant_off(dev);
7067 if (netif_carrier_ok(rootdev)) {
7068 if (!netif_carrier_ok(dev))
7069 netif_carrier_on(dev);
7071 if (netif_carrier_ok(dev))
7072 netif_carrier_off(dev);
7075 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
7078 static int netif_alloc_rx_queues(struct net_device *dev)
7080 unsigned int i, count = dev->num_rx_queues;
7081 struct netdev_rx_queue *rx;
7082 size_t sz = count * sizeof(*rx);
7086 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7094 for (i = 0; i < count; i++)
7100 static void netdev_init_one_queue(struct net_device *dev,
7101 struct netdev_queue *queue, void *_unused)
7103 /* Initialize queue lock */
7104 spin_lock_init(&queue->_xmit_lock);
7105 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
7106 queue->xmit_lock_owner = -1;
7107 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
7110 dql_init(&queue->dql, HZ);
7114 static void netif_free_tx_queues(struct net_device *dev)
7119 static int netif_alloc_netdev_queues(struct net_device *dev)
7121 unsigned int count = dev->num_tx_queues;
7122 struct netdev_queue *tx;
7123 size_t sz = count * sizeof(*tx);
7125 if (count < 1 || count > 0xffff)
7128 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7136 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
7137 spin_lock_init(&dev->tx_global_lock);
7142 void netif_tx_stop_all_queues(struct net_device *dev)
7146 for (i = 0; i < dev->num_tx_queues; i++) {
7147 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
7148 netif_tx_stop_queue(txq);
7151 EXPORT_SYMBOL(netif_tx_stop_all_queues);
7154 * register_netdevice - register a network device
7155 * @dev: device to register
7157 * Take a completed network device structure and add it to the kernel
7158 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7159 * chain. 0 is returned on success. A negative errno code is returned
7160 * on a failure to set up the device, or if the name is a duplicate.
7162 * Callers must hold the rtnl semaphore. You may want
7163 * register_netdev() instead of this.
7166 * The locking appears insufficient to guarantee two parallel registers
7167 * will not get the same name.
7170 int register_netdevice(struct net_device *dev)
7173 struct net *net = dev_net(dev);
7175 BUG_ON(dev_boot_phase);
7180 /* When net_device's are persistent, this will be fatal. */
7181 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
7184 spin_lock_init(&dev->addr_list_lock);
7185 netdev_set_addr_lockdep_class(dev);
7187 ret = dev_get_valid_name(net, dev, dev->name);
7191 /* Init, if this function is available */
7192 if (dev->netdev_ops->ndo_init) {
7193 ret = dev->netdev_ops->ndo_init(dev);
7201 if (((dev->hw_features | dev->features) &
7202 NETIF_F_HW_VLAN_CTAG_FILTER) &&
7203 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
7204 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
7205 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
7212 dev->ifindex = dev_new_index(net);
7213 else if (__dev_get_by_index(net, dev->ifindex))
7216 /* Transfer changeable features to wanted_features and enable
7217 * software offloads (GSO and GRO).
7219 dev->hw_features |= NETIF_F_SOFT_FEATURES;
7220 dev->features |= NETIF_F_SOFT_FEATURES;
7221 dev->wanted_features = dev->features & dev->hw_features;
7223 if (!(dev->flags & IFF_LOOPBACK))
7224 dev->hw_features |= NETIF_F_NOCACHE_COPY;
7226 /* If IPv4 TCP segmentation offload is supported we should also
7227 * allow the device to enable segmenting the frame with the option
7228 * of ignoring a static IP ID value. This doesn't enable the
7229 * feature itself but allows the user to enable it later.
7231 if (dev->hw_features & NETIF_F_TSO)
7232 dev->hw_features |= NETIF_F_TSO_MANGLEID;
7233 if (dev->vlan_features & NETIF_F_TSO)
7234 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
7235 if (dev->mpls_features & NETIF_F_TSO)
7236 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
7237 if (dev->hw_enc_features & NETIF_F_TSO)
7238 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
7240 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
7242 dev->vlan_features |= NETIF_F_HIGHDMA;
7244 /* Make NETIF_F_SG inheritable to tunnel devices.
7246 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
7248 /* Make NETIF_F_SG inheritable to MPLS.
7250 dev->mpls_features |= NETIF_F_SG;
7252 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
7253 ret = notifier_to_errno(ret);
7257 ret = netdev_register_kobject(dev);
7260 dev->reg_state = NETREG_REGISTERED;
7262 __netdev_update_features(dev);
7265 * Default initial state at registry is that the
7266 * device is present.
7269 set_bit(__LINK_STATE_PRESENT, &dev->state);
7271 linkwatch_init_dev(dev);
7273 dev_init_scheduler(dev);
7275 list_netdevice(dev);
7276 add_device_randomness(dev->dev_addr, dev->addr_len);
7278 /* If the device has permanent device address, driver should
7279 * set dev_addr and also addr_assign_type should be set to
7280 * NET_ADDR_PERM (default value).
7282 if (dev->addr_assign_type == NET_ADDR_PERM)
7283 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
7285 /* Notify protocols, that a new device appeared. */
7286 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
7287 ret = notifier_to_errno(ret);
7289 rollback_registered(dev);
7290 dev->reg_state = NETREG_UNREGISTERED;
7293 * Prevent userspace races by waiting until the network
7294 * device is fully setup before sending notifications.
7296 if (!dev->rtnl_link_ops ||
7297 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7298 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7304 if (dev->netdev_ops->ndo_uninit)
7305 dev->netdev_ops->ndo_uninit(dev);
7308 EXPORT_SYMBOL(register_netdevice);
7311 * init_dummy_netdev - init a dummy network device for NAPI
7312 * @dev: device to init
7314 * This takes a network device structure and initialize the minimum
7315 * amount of fields so it can be used to schedule NAPI polls without
7316 * registering a full blown interface. This is to be used by drivers
7317 * that need to tie several hardware interfaces to a single NAPI
7318 * poll scheduler due to HW limitations.
7320 int init_dummy_netdev(struct net_device *dev)
7322 /* Clear everything. Note we don't initialize spinlocks
7323 * are they aren't supposed to be taken by any of the
7324 * NAPI code and this dummy netdev is supposed to be
7325 * only ever used for NAPI polls
7327 memset(dev, 0, sizeof(struct net_device));
7329 /* make sure we BUG if trying to hit standard
7330 * register/unregister code path
7332 dev->reg_state = NETREG_DUMMY;
7334 /* NAPI wants this */
7335 INIT_LIST_HEAD(&dev->napi_list);
7337 /* a dummy interface is started by default */
7338 set_bit(__LINK_STATE_PRESENT, &dev->state);
7339 set_bit(__LINK_STATE_START, &dev->state);
7341 /* Note : We dont allocate pcpu_refcnt for dummy devices,
7342 * because users of this 'device' dont need to change
7348 EXPORT_SYMBOL_GPL(init_dummy_netdev);
7352 * register_netdev - register a network device
7353 * @dev: device to register
7355 * Take a completed network device structure and add it to the kernel
7356 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7357 * chain. 0 is returned on success. A negative errno code is returned
7358 * on a failure to set up the device, or if the name is a duplicate.
7360 * This is a wrapper around register_netdevice that takes the rtnl semaphore
7361 * and expands the device name if you passed a format string to
7364 int register_netdev(struct net_device *dev)
7369 err = register_netdevice(dev);
7373 EXPORT_SYMBOL(register_netdev);
7375 int netdev_refcnt_read(const struct net_device *dev)
7379 for_each_possible_cpu(i)
7380 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
7383 EXPORT_SYMBOL(netdev_refcnt_read);
7386 * netdev_wait_allrefs - wait until all references are gone.
7387 * @dev: target net_device
7389 * This is called when unregistering network devices.
7391 * Any protocol or device that holds a reference should register
7392 * for netdevice notification, and cleanup and put back the
7393 * reference if they receive an UNREGISTER event.
7394 * We can get stuck here if buggy protocols don't correctly
7397 static void netdev_wait_allrefs(struct net_device *dev)
7399 unsigned long rebroadcast_time, warning_time;
7402 linkwatch_forget_dev(dev);
7404 rebroadcast_time = warning_time = jiffies;
7405 refcnt = netdev_refcnt_read(dev);
7407 while (refcnt != 0) {
7408 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
7411 /* Rebroadcast unregister notification */
7412 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7418 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7419 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
7421 /* We must not have linkwatch events
7422 * pending on unregister. If this
7423 * happens, we simply run the queue
7424 * unscheduled, resulting in a noop
7427 linkwatch_run_queue();
7432 rebroadcast_time = jiffies;
7437 refcnt = netdev_refcnt_read(dev);
7439 if (time_after(jiffies, warning_time + 10 * HZ)) {
7440 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
7442 warning_time = jiffies;
7451 * register_netdevice(x1);
7452 * register_netdevice(x2);
7454 * unregister_netdevice(y1);
7455 * unregister_netdevice(y2);
7461 * We are invoked by rtnl_unlock().
7462 * This allows us to deal with problems:
7463 * 1) We can delete sysfs objects which invoke hotplug
7464 * without deadlocking with linkwatch via keventd.
7465 * 2) Since we run with the RTNL semaphore not held, we can sleep
7466 * safely in order to wait for the netdev refcnt to drop to zero.
7468 * We must not return until all unregister events added during
7469 * the interval the lock was held have been completed.
7471 void netdev_run_todo(void)
7473 struct list_head list;
7475 /* Snapshot list, allow later requests */
7476 list_replace_init(&net_todo_list, &list);
7481 /* Wait for rcu callbacks to finish before next phase */
7482 if (!list_empty(&list))
7485 while (!list_empty(&list)) {
7486 struct net_device *dev
7487 = list_first_entry(&list, struct net_device, todo_list);
7488 list_del(&dev->todo_list);
7491 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7494 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
7495 pr_err("network todo '%s' but state %d\n",
7496 dev->name, dev->reg_state);
7501 dev->reg_state = NETREG_UNREGISTERED;
7503 netdev_wait_allrefs(dev);
7506 BUG_ON(netdev_refcnt_read(dev));
7507 BUG_ON(!list_empty(&dev->ptype_all));
7508 BUG_ON(!list_empty(&dev->ptype_specific));
7509 WARN_ON(rcu_access_pointer(dev->ip_ptr));
7510 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
7511 WARN_ON(dev->dn_ptr);
7513 if (dev->destructor)
7514 dev->destructor(dev);
7516 /* Report a network device has been unregistered */
7518 dev_net(dev)->dev_unreg_count--;
7520 wake_up(&netdev_unregistering_wq);
7522 /* Free network device */
7523 kobject_put(&dev->dev.kobj);
7527 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
7528 * all the same fields in the same order as net_device_stats, with only
7529 * the type differing, but rtnl_link_stats64 may have additional fields
7530 * at the end for newer counters.
7532 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7533 const struct net_device_stats *netdev_stats)
7535 #if BITS_PER_LONG == 64
7536 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
7537 memcpy(stats64, netdev_stats, sizeof(*stats64));
7538 /* zero out counters that only exist in rtnl_link_stats64 */
7539 memset((char *)stats64 + sizeof(*netdev_stats), 0,
7540 sizeof(*stats64) - sizeof(*netdev_stats));
7542 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
7543 const unsigned long *src = (const unsigned long *)netdev_stats;
7544 u64 *dst = (u64 *)stats64;
7546 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
7547 for (i = 0; i < n; i++)
7549 /* zero out counters that only exist in rtnl_link_stats64 */
7550 memset((char *)stats64 + n * sizeof(u64), 0,
7551 sizeof(*stats64) - n * sizeof(u64));
7554 EXPORT_SYMBOL(netdev_stats_to_stats64);
7557 * dev_get_stats - get network device statistics
7558 * @dev: device to get statistics from
7559 * @storage: place to store stats
7561 * Get network statistics from device. Return @storage.
7562 * The device driver may provide its own method by setting
7563 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7564 * otherwise the internal statistics structure is used.
7566 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
7567 struct rtnl_link_stats64 *storage)
7569 const struct net_device_ops *ops = dev->netdev_ops;
7571 if (ops->ndo_get_stats64) {
7572 memset(storage, 0, sizeof(*storage));
7573 ops->ndo_get_stats64(dev, storage);
7574 } else if (ops->ndo_get_stats) {
7575 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7577 netdev_stats_to_stats64(storage, &dev->stats);
7579 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
7580 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
7581 storage->rx_nohandler += atomic_long_read(&dev->rx_nohandler);
7584 EXPORT_SYMBOL(dev_get_stats);
7586 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7588 struct netdev_queue *queue = dev_ingress_queue(dev);
7590 #ifdef CONFIG_NET_CLS_ACT
7593 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7596 netdev_init_one_queue(dev, queue, NULL);
7597 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7598 queue->qdisc_sleeping = &noop_qdisc;
7599 rcu_assign_pointer(dev->ingress_queue, queue);
7604 static const struct ethtool_ops default_ethtool_ops;
7606 void netdev_set_default_ethtool_ops(struct net_device *dev,
7607 const struct ethtool_ops *ops)
7609 if (dev->ethtool_ops == &default_ethtool_ops)
7610 dev->ethtool_ops = ops;
7612 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7614 void netdev_freemem(struct net_device *dev)
7616 char *addr = (char *)dev - dev->padded;
7622 * alloc_netdev_mqs - allocate network device
7623 * @sizeof_priv: size of private data to allocate space for
7624 * @name: device name format string
7625 * @name_assign_type: origin of device name
7626 * @setup: callback to initialize device
7627 * @txqs: the number of TX subqueues to allocate
7628 * @rxqs: the number of RX subqueues to allocate
7630 * Allocates a struct net_device with private data area for driver use
7631 * and performs basic initialization. Also allocates subqueue structs
7632 * for each queue on the device.
7634 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7635 unsigned char name_assign_type,
7636 void (*setup)(struct net_device *),
7637 unsigned int txqs, unsigned int rxqs)
7639 struct net_device *dev;
7641 struct net_device *p;
7643 BUG_ON(strlen(name) >= sizeof(dev->name));
7646 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7652 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7657 alloc_size = sizeof(struct net_device);
7659 /* ensure 32-byte alignment of private area */
7660 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7661 alloc_size += sizeof_priv;
7663 /* ensure 32-byte alignment of whole construct */
7664 alloc_size += NETDEV_ALIGN - 1;
7666 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7668 p = vzalloc(alloc_size);
7672 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7673 dev->padded = (char *)dev - (char *)p;
7675 dev->pcpu_refcnt = alloc_percpu(int);
7676 if (!dev->pcpu_refcnt)
7679 if (dev_addr_init(dev))
7685 dev_net_set(dev, &init_net);
7687 dev->gso_max_size = GSO_MAX_SIZE;
7688 dev->gso_max_segs = GSO_MAX_SEGS;
7690 INIT_LIST_HEAD(&dev->napi_list);
7691 INIT_LIST_HEAD(&dev->unreg_list);
7692 INIT_LIST_HEAD(&dev->close_list);
7693 INIT_LIST_HEAD(&dev->link_watch_list);
7694 INIT_LIST_HEAD(&dev->adj_list.upper);
7695 INIT_LIST_HEAD(&dev->adj_list.lower);
7696 INIT_LIST_HEAD(&dev->ptype_all);
7697 INIT_LIST_HEAD(&dev->ptype_specific);
7698 #ifdef CONFIG_NET_SCHED
7699 hash_init(dev->qdisc_hash);
7701 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7704 if (!dev->tx_queue_len) {
7705 dev->priv_flags |= IFF_NO_QUEUE;
7706 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
7709 dev->num_tx_queues = txqs;
7710 dev->real_num_tx_queues = txqs;
7711 if (netif_alloc_netdev_queues(dev))
7715 dev->num_rx_queues = rxqs;
7716 dev->real_num_rx_queues = rxqs;
7717 if (netif_alloc_rx_queues(dev))
7721 strcpy(dev->name, name);
7722 dev->name_assign_type = name_assign_type;
7723 dev->group = INIT_NETDEV_GROUP;
7724 if (!dev->ethtool_ops)
7725 dev->ethtool_ops = &default_ethtool_ops;
7727 nf_hook_ingress_init(dev);
7736 free_percpu(dev->pcpu_refcnt);
7738 netdev_freemem(dev);
7741 EXPORT_SYMBOL(alloc_netdev_mqs);
7744 * free_netdev - free network device
7747 * This function does the last stage of destroying an allocated device
7748 * interface. The reference to the device object is released.
7749 * If this is the last reference then it will be freed.
7750 * Must be called in process context.
7752 void free_netdev(struct net_device *dev)
7754 struct napi_struct *p, *n;
7757 netif_free_tx_queues(dev);
7762 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7764 /* Flush device addresses */
7765 dev_addr_flush(dev);
7767 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7770 free_percpu(dev->pcpu_refcnt);
7771 dev->pcpu_refcnt = NULL;
7773 /* Compatibility with error handling in drivers */
7774 if (dev->reg_state == NETREG_UNINITIALIZED) {
7775 netdev_freemem(dev);
7779 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7780 dev->reg_state = NETREG_RELEASED;
7782 /* will free via device release */
7783 put_device(&dev->dev);
7785 EXPORT_SYMBOL(free_netdev);
7788 * synchronize_net - Synchronize with packet receive processing
7790 * Wait for packets currently being received to be done.
7791 * Does not block later packets from starting.
7793 void synchronize_net(void)
7796 if (rtnl_is_locked())
7797 synchronize_rcu_expedited();
7801 EXPORT_SYMBOL(synchronize_net);
7804 * unregister_netdevice_queue - remove device from the kernel
7808 * This function shuts down a device interface and removes it
7809 * from the kernel tables.
7810 * If head not NULL, device is queued to be unregistered later.
7812 * Callers must hold the rtnl semaphore. You may want
7813 * unregister_netdev() instead of this.
7816 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7821 list_move_tail(&dev->unreg_list, head);
7823 rollback_registered(dev);
7824 /* Finish processing unregister after unlock */
7828 EXPORT_SYMBOL(unregister_netdevice_queue);
7831 * unregister_netdevice_many - unregister many devices
7832 * @head: list of devices
7834 * Note: As most callers use a stack allocated list_head,
7835 * we force a list_del() to make sure stack wont be corrupted later.
7837 void unregister_netdevice_many(struct list_head *head)
7839 struct net_device *dev;
7841 if (!list_empty(head)) {
7842 rollback_registered_many(head);
7843 list_for_each_entry(dev, head, unreg_list)
7848 EXPORT_SYMBOL(unregister_netdevice_many);
7851 * unregister_netdev - remove device from the kernel
7854 * This function shuts down a device interface and removes it
7855 * from the kernel tables.
7857 * This is just a wrapper for unregister_netdevice that takes
7858 * the rtnl semaphore. In general you want to use this and not
7859 * unregister_netdevice.
7861 void unregister_netdev(struct net_device *dev)
7864 unregister_netdevice(dev);
7867 EXPORT_SYMBOL(unregister_netdev);
7870 * dev_change_net_namespace - move device to different nethost namespace
7872 * @net: network namespace
7873 * @pat: If not NULL name pattern to try if the current device name
7874 * is already taken in the destination network namespace.
7876 * This function shuts down a device interface and moves it
7877 * to a new network namespace. On success 0 is returned, on
7878 * a failure a netagive errno code is returned.
7880 * Callers must hold the rtnl semaphore.
7883 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7889 /* Don't allow namespace local devices to be moved. */
7891 if (dev->features & NETIF_F_NETNS_LOCAL)
7894 /* Ensure the device has been registrered */
7895 if (dev->reg_state != NETREG_REGISTERED)
7898 /* Get out if there is nothing todo */
7900 if (net_eq(dev_net(dev), net))
7903 /* Pick the destination device name, and ensure
7904 * we can use it in the destination network namespace.
7907 if (__dev_get_by_name(net, dev->name)) {
7908 /* We get here if we can't use the current device name */
7911 if (dev_get_valid_name(net, dev, pat) < 0)
7916 * And now a mini version of register_netdevice unregister_netdevice.
7919 /* If device is running close it first. */
7922 /* And unlink it from device chain */
7924 unlist_netdevice(dev);
7928 /* Shutdown queueing discipline. */
7931 /* Notify protocols, that we are about to destroy
7932 this device. They should clean all the things.
7934 Note that dev->reg_state stays at NETREG_REGISTERED.
7935 This is wanted because this way 8021q and macvlan know
7936 the device is just moving and can keep their slaves up.
7938 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7940 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7941 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7944 * Flush the unicast and multicast chains
7949 /* Send a netdev-removed uevent to the old namespace */
7950 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7951 netdev_adjacent_del_links(dev);
7953 /* Actually switch the network namespace */
7954 dev_net_set(dev, net);
7956 /* If there is an ifindex conflict assign a new one */
7957 if (__dev_get_by_index(net, dev->ifindex))
7958 dev->ifindex = dev_new_index(net);
7960 /* Send a netdev-add uevent to the new namespace */
7961 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7962 netdev_adjacent_add_links(dev);
7964 /* Fixup kobjects */
7965 err = device_rename(&dev->dev, dev->name);
7968 /* Add the device back in the hashes */
7969 list_netdevice(dev);
7971 /* Notify protocols, that a new device appeared. */
7972 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7975 * Prevent userspace races by waiting until the network
7976 * device is fully setup before sending notifications.
7978 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7985 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7987 static int dev_cpu_dead(unsigned int oldcpu)
7989 struct sk_buff **list_skb;
7990 struct sk_buff *skb;
7992 struct softnet_data *sd, *oldsd;
7994 local_irq_disable();
7995 cpu = smp_processor_id();
7996 sd = &per_cpu(softnet_data, cpu);
7997 oldsd = &per_cpu(softnet_data, oldcpu);
7999 /* Find end of our completion_queue. */
8000 list_skb = &sd->completion_queue;
8002 list_skb = &(*list_skb)->next;
8003 /* Append completion queue from offline CPU. */
8004 *list_skb = oldsd->completion_queue;
8005 oldsd->completion_queue = NULL;
8007 /* Append output queue from offline CPU. */
8008 if (oldsd->output_queue) {
8009 *sd->output_queue_tailp = oldsd->output_queue;
8010 sd->output_queue_tailp = oldsd->output_queue_tailp;
8011 oldsd->output_queue = NULL;
8012 oldsd->output_queue_tailp = &oldsd->output_queue;
8014 /* Append NAPI poll list from offline CPU, with one exception :
8015 * process_backlog() must be called by cpu owning percpu backlog.
8016 * We properly handle process_queue & input_pkt_queue later.
8018 while (!list_empty(&oldsd->poll_list)) {
8019 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
8023 list_del_init(&napi->poll_list);
8024 if (napi->poll == process_backlog)
8027 ____napi_schedule(sd, napi);
8030 raise_softirq_irqoff(NET_TX_SOFTIRQ);
8033 /* Process offline CPU's input_pkt_queue */
8034 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
8036 input_queue_head_incr(oldsd);
8038 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
8040 input_queue_head_incr(oldsd);
8047 * netdev_increment_features - increment feature set by one
8048 * @all: current feature set
8049 * @one: new feature set
8050 * @mask: mask feature set
8052 * Computes a new feature set after adding a device with feature set
8053 * @one to the master device with current feature set @all. Will not
8054 * enable anything that is off in @mask. Returns the new feature set.
8056 netdev_features_t netdev_increment_features(netdev_features_t all,
8057 netdev_features_t one, netdev_features_t mask)
8059 if (mask & NETIF_F_HW_CSUM)
8060 mask |= NETIF_F_CSUM_MASK;
8061 mask |= NETIF_F_VLAN_CHALLENGED;
8063 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
8064 all &= one | ~NETIF_F_ALL_FOR_ALL;
8066 /* If one device supports hw checksumming, set for all. */
8067 if (all & NETIF_F_HW_CSUM)
8068 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
8072 EXPORT_SYMBOL(netdev_increment_features);
8074 static struct hlist_head * __net_init netdev_create_hash(void)
8077 struct hlist_head *hash;
8079 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
8081 for (i = 0; i < NETDEV_HASHENTRIES; i++)
8082 INIT_HLIST_HEAD(&hash[i]);
8087 /* Initialize per network namespace state */
8088 static int __net_init netdev_init(struct net *net)
8090 if (net != &init_net)
8091 INIT_LIST_HEAD(&net->dev_base_head);
8093 net->dev_name_head = netdev_create_hash();
8094 if (net->dev_name_head == NULL)
8097 net->dev_index_head = netdev_create_hash();
8098 if (net->dev_index_head == NULL)
8104 kfree(net->dev_name_head);
8110 * netdev_drivername - network driver for the device
8111 * @dev: network device
8113 * Determine network driver for device.
8115 const char *netdev_drivername(const struct net_device *dev)
8117 const struct device_driver *driver;
8118 const struct device *parent;
8119 const char *empty = "";
8121 parent = dev->dev.parent;
8125 driver = parent->driver;
8126 if (driver && driver->name)
8127 return driver->name;
8131 static void __netdev_printk(const char *level, const struct net_device *dev,
8132 struct va_format *vaf)
8134 if (dev && dev->dev.parent) {
8135 dev_printk_emit(level[1] - '0',
8138 dev_driver_string(dev->dev.parent),
8139 dev_name(dev->dev.parent),
8140 netdev_name(dev), netdev_reg_state(dev),
8143 printk("%s%s%s: %pV",
8144 level, netdev_name(dev), netdev_reg_state(dev), vaf);
8146 printk("%s(NULL net_device): %pV", level, vaf);
8150 void netdev_printk(const char *level, const struct net_device *dev,
8151 const char *format, ...)
8153 struct va_format vaf;
8156 va_start(args, format);
8161 __netdev_printk(level, dev, &vaf);
8165 EXPORT_SYMBOL(netdev_printk);
8167 #define define_netdev_printk_level(func, level) \
8168 void func(const struct net_device *dev, const char *fmt, ...) \
8170 struct va_format vaf; \
8173 va_start(args, fmt); \
8178 __netdev_printk(level, dev, &vaf); \
8182 EXPORT_SYMBOL(func);
8184 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
8185 define_netdev_printk_level(netdev_alert, KERN_ALERT);
8186 define_netdev_printk_level(netdev_crit, KERN_CRIT);
8187 define_netdev_printk_level(netdev_err, KERN_ERR);
8188 define_netdev_printk_level(netdev_warn, KERN_WARNING);
8189 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
8190 define_netdev_printk_level(netdev_info, KERN_INFO);
8192 static void __net_exit netdev_exit(struct net *net)
8194 kfree(net->dev_name_head);
8195 kfree(net->dev_index_head);
8198 static struct pernet_operations __net_initdata netdev_net_ops = {
8199 .init = netdev_init,
8200 .exit = netdev_exit,
8203 static void __net_exit default_device_exit(struct net *net)
8205 struct net_device *dev, *aux;
8207 * Push all migratable network devices back to the
8208 * initial network namespace
8211 for_each_netdev_safe(net, dev, aux) {
8213 char fb_name[IFNAMSIZ];
8215 /* Ignore unmoveable devices (i.e. loopback) */
8216 if (dev->features & NETIF_F_NETNS_LOCAL)
8219 /* Leave virtual devices for the generic cleanup */
8220 if (dev->rtnl_link_ops)
8223 /* Push remaining network devices to init_net */
8224 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
8225 err = dev_change_net_namespace(dev, &init_net, fb_name);
8227 pr_emerg("%s: failed to move %s to init_net: %d\n",
8228 __func__, dev->name, err);
8235 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
8237 /* Return with the rtnl_lock held when there are no network
8238 * devices unregistering in any network namespace in net_list.
8242 DEFINE_WAIT_FUNC(wait, woken_wake_function);
8244 add_wait_queue(&netdev_unregistering_wq, &wait);
8246 unregistering = false;
8248 list_for_each_entry(net, net_list, exit_list) {
8249 if (net->dev_unreg_count > 0) {
8250 unregistering = true;
8258 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
8260 remove_wait_queue(&netdev_unregistering_wq, &wait);
8263 static void __net_exit default_device_exit_batch(struct list_head *net_list)
8265 /* At exit all network devices most be removed from a network
8266 * namespace. Do this in the reverse order of registration.
8267 * Do this across as many network namespaces as possible to
8268 * improve batching efficiency.
8270 struct net_device *dev;
8272 LIST_HEAD(dev_kill_list);
8274 /* To prevent network device cleanup code from dereferencing
8275 * loopback devices or network devices that have been freed
8276 * wait here for all pending unregistrations to complete,
8277 * before unregistring the loopback device and allowing the
8278 * network namespace be freed.
8280 * The netdev todo list containing all network devices
8281 * unregistrations that happen in default_device_exit_batch
8282 * will run in the rtnl_unlock() at the end of
8283 * default_device_exit_batch.
8285 rtnl_lock_unregistering(net_list);
8286 list_for_each_entry(net, net_list, exit_list) {
8287 for_each_netdev_reverse(net, dev) {
8288 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
8289 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
8291 unregister_netdevice_queue(dev, &dev_kill_list);
8294 unregister_netdevice_many(&dev_kill_list);
8298 static struct pernet_operations __net_initdata default_device_ops = {
8299 .exit = default_device_exit,
8300 .exit_batch = default_device_exit_batch,
8304 * Initialize the DEV module. At boot time this walks the device list and
8305 * unhooks any devices that fail to initialise (normally hardware not
8306 * present) and leaves us with a valid list of present and active devices.
8311 * This is called single threaded during boot, so no need
8312 * to take the rtnl semaphore.
8314 static int __init net_dev_init(void)
8316 int i, rc = -ENOMEM;
8318 BUG_ON(!dev_boot_phase);
8320 if (dev_proc_init())
8323 if (netdev_kobject_init())
8326 INIT_LIST_HEAD(&ptype_all);
8327 for (i = 0; i < PTYPE_HASH_SIZE; i++)
8328 INIT_LIST_HEAD(&ptype_base[i]);
8330 INIT_LIST_HEAD(&offload_base);
8332 if (register_pernet_subsys(&netdev_net_ops))
8336 * Initialise the packet receive queues.
8339 for_each_possible_cpu(i) {
8340 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
8341 struct softnet_data *sd = &per_cpu(softnet_data, i);
8343 INIT_WORK(flush, flush_backlog);
8345 skb_queue_head_init(&sd->input_pkt_queue);
8346 skb_queue_head_init(&sd->process_queue);
8347 INIT_LIST_HEAD(&sd->poll_list);
8348 sd->output_queue_tailp = &sd->output_queue;
8350 sd->csd.func = rps_trigger_softirq;
8355 sd->backlog.poll = process_backlog;
8356 sd->backlog.weight = weight_p;
8361 /* The loopback device is special if any other network devices
8362 * is present in a network namespace the loopback device must
8363 * be present. Since we now dynamically allocate and free the
8364 * loopback device ensure this invariant is maintained by
8365 * keeping the loopback device as the first device on the
8366 * list of network devices. Ensuring the loopback devices
8367 * is the first device that appears and the last network device
8370 if (register_pernet_device(&loopback_net_ops))
8373 if (register_pernet_device(&default_device_ops))
8376 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
8377 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
8379 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
8380 NULL, dev_cpu_dead);
8388 subsys_initcall(net_dev_init);