2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <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)
1734 skb->tstamp.tv64 = 0;
1735 if (static_key_false(&netstamp_needed))
1736 __net_timestamp(skb);
1739 #define net_timestamp_check(COND, SKB) \
1740 if (static_key_false(&netstamp_needed)) { \
1741 if ((COND) && !(SKB)->tstamp.tv64) \
1742 __net_timestamp(SKB); \
1745 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1749 if (!(dev->flags & IFF_UP))
1752 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1753 if (skb->len <= len)
1756 /* if TSO is enabled, we don't care about the length as the packet
1757 * could be forwarded without being segmented before
1759 if (skb_is_gso(skb))
1764 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1766 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1768 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);
2412 * netif_wake_subqueue - allow sending packets on subqueue
2413 * @dev: network device
2414 * @queue_index: sub queue index
2416 * Resume individual transmit queue of a device with multiple transmit queues.
2418 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2420 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2422 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2426 q = rcu_dereference(txq->qdisc);
2427 __netif_schedule(q);
2431 EXPORT_SYMBOL(netif_wake_subqueue);
2433 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2435 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2439 q = rcu_dereference(dev_queue->qdisc);
2440 __netif_schedule(q);
2444 EXPORT_SYMBOL(netif_tx_wake_queue);
2446 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2448 unsigned long flags;
2450 if (likely(atomic_read(&skb->users) == 1)) {
2452 atomic_set(&skb->users, 0);
2453 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2456 get_kfree_skb_cb(skb)->reason = reason;
2457 local_irq_save(flags);
2458 skb->next = __this_cpu_read(softnet_data.completion_queue);
2459 __this_cpu_write(softnet_data.completion_queue, skb);
2460 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2461 local_irq_restore(flags);
2463 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2465 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2467 if (in_irq() || irqs_disabled())
2468 __dev_kfree_skb_irq(skb, reason);
2472 EXPORT_SYMBOL(__dev_kfree_skb_any);
2476 * netif_device_detach - mark device as removed
2477 * @dev: network device
2479 * Mark device as removed from system and therefore no longer available.
2481 void netif_device_detach(struct net_device *dev)
2483 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2484 netif_running(dev)) {
2485 netif_tx_stop_all_queues(dev);
2488 EXPORT_SYMBOL(netif_device_detach);
2491 * netif_device_attach - mark device as attached
2492 * @dev: network device
2494 * Mark device as attached from system and restart if needed.
2496 void netif_device_attach(struct net_device *dev)
2498 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2499 netif_running(dev)) {
2500 netif_tx_wake_all_queues(dev);
2501 __netdev_watchdog_up(dev);
2504 EXPORT_SYMBOL(netif_device_attach);
2507 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2508 * to be used as a distribution range.
2510 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2511 unsigned int num_tx_queues)
2515 u16 qcount = num_tx_queues;
2517 if (skb_rx_queue_recorded(skb)) {
2518 hash = skb_get_rx_queue(skb);
2519 while (unlikely(hash >= num_tx_queues))
2520 hash -= num_tx_queues;
2525 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2526 qoffset = dev->tc_to_txq[tc].offset;
2527 qcount = dev->tc_to_txq[tc].count;
2530 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2532 EXPORT_SYMBOL(__skb_tx_hash);
2534 static void skb_warn_bad_offload(const struct sk_buff *skb)
2536 static const netdev_features_t null_features;
2537 struct net_device *dev = skb->dev;
2538 const char *name = "";
2540 if (!net_ratelimit())
2544 if (dev->dev.parent)
2545 name = dev_driver_string(dev->dev.parent);
2547 name = netdev_name(dev);
2549 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2550 "gso_type=%d ip_summed=%d\n",
2551 name, dev ? &dev->features : &null_features,
2552 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2553 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2554 skb_shinfo(skb)->gso_type, skb->ip_summed);
2558 * Invalidate hardware checksum when packet is to be mangled, and
2559 * complete checksum manually on outgoing path.
2561 int skb_checksum_help(struct sk_buff *skb)
2564 int ret = 0, offset;
2566 if (skb->ip_summed == CHECKSUM_COMPLETE)
2567 goto out_set_summed;
2569 if (unlikely(skb_shinfo(skb)->gso_size)) {
2570 skb_warn_bad_offload(skb);
2574 /* Before computing a checksum, we should make sure no frag could
2575 * be modified by an external entity : checksum could be wrong.
2577 if (skb_has_shared_frag(skb)) {
2578 ret = __skb_linearize(skb);
2583 offset = skb_checksum_start_offset(skb);
2584 BUG_ON(offset >= skb_headlen(skb));
2585 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2587 offset += skb->csum_offset;
2588 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2590 if (skb_cloned(skb) &&
2591 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2592 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2597 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2599 skb->ip_summed = CHECKSUM_NONE;
2603 EXPORT_SYMBOL(skb_checksum_help);
2605 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2607 __be16 type = skb->protocol;
2609 /* Tunnel gso handlers can set protocol to ethernet. */
2610 if (type == htons(ETH_P_TEB)) {
2613 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2616 eth = (struct ethhdr *)skb_mac_header(skb);
2617 type = eth->h_proto;
2620 return __vlan_get_protocol(skb, type, depth);
2624 * skb_mac_gso_segment - mac layer segmentation handler.
2625 * @skb: buffer to segment
2626 * @features: features for the output path (see dev->features)
2628 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2629 netdev_features_t features)
2631 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2632 struct packet_offload *ptype;
2633 int vlan_depth = skb->mac_len;
2634 __be16 type = skb_network_protocol(skb, &vlan_depth);
2636 if (unlikely(!type))
2637 return ERR_PTR(-EINVAL);
2639 __skb_pull(skb, vlan_depth);
2642 list_for_each_entry_rcu(ptype, &offload_base, list) {
2643 if (ptype->type == type && ptype->callbacks.gso_segment) {
2644 segs = ptype->callbacks.gso_segment(skb, features);
2650 __skb_push(skb, skb->data - skb_mac_header(skb));
2654 EXPORT_SYMBOL(skb_mac_gso_segment);
2657 /* openvswitch calls this on rx path, so we need a different check.
2659 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2662 return skb->ip_summed != CHECKSUM_PARTIAL;
2664 return skb->ip_summed == CHECKSUM_NONE;
2668 * __skb_gso_segment - Perform segmentation on skb.
2669 * @skb: buffer to segment
2670 * @features: features for the output path (see dev->features)
2671 * @tx_path: whether it is called in TX path
2673 * This function segments the given skb and returns a list of segments.
2675 * It may return NULL if the skb requires no segmentation. This is
2676 * only possible when GSO is used for verifying header integrity.
2678 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2680 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2681 netdev_features_t features, bool tx_path)
2683 if (unlikely(skb_needs_check(skb, tx_path))) {
2686 skb_warn_bad_offload(skb);
2688 err = skb_cow_head(skb, 0);
2690 return ERR_PTR(err);
2693 /* Only report GSO partial support if it will enable us to
2694 * support segmentation on this frame without needing additional
2697 if (features & NETIF_F_GSO_PARTIAL) {
2698 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
2699 struct net_device *dev = skb->dev;
2701 partial_features |= dev->features & dev->gso_partial_features;
2702 if (!skb_gso_ok(skb, features | partial_features))
2703 features &= ~NETIF_F_GSO_PARTIAL;
2706 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2707 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2709 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2710 SKB_GSO_CB(skb)->encap_level = 0;
2712 skb_reset_mac_header(skb);
2713 skb_reset_mac_len(skb);
2715 return skb_mac_gso_segment(skb, features);
2717 EXPORT_SYMBOL(__skb_gso_segment);
2719 /* Take action when hardware reception checksum errors are detected. */
2721 void netdev_rx_csum_fault(struct net_device *dev)
2723 if (net_ratelimit()) {
2724 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2728 EXPORT_SYMBOL(netdev_rx_csum_fault);
2731 /* Actually, we should eliminate this check as soon as we know, that:
2732 * 1. IOMMU is present and allows to map all the memory.
2733 * 2. No high memory really exists on this machine.
2736 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2738 #ifdef CONFIG_HIGHMEM
2740 if (!(dev->features & NETIF_F_HIGHDMA)) {
2741 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2742 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2743 if (PageHighMem(skb_frag_page(frag)))
2748 if (PCI_DMA_BUS_IS_PHYS) {
2749 struct device *pdev = dev->dev.parent;
2753 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2754 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2755 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2756 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2764 /* If MPLS offload request, verify we are testing hardware MPLS features
2765 * instead of standard features for the netdev.
2767 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2768 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2769 netdev_features_t features,
2772 if (eth_p_mpls(type))
2773 features &= skb->dev->mpls_features;
2778 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2779 netdev_features_t features,
2786 static netdev_features_t harmonize_features(struct sk_buff *skb,
2787 netdev_features_t features)
2792 type = skb_network_protocol(skb, &tmp);
2793 features = net_mpls_features(skb, features, type);
2795 if (skb->ip_summed != CHECKSUM_NONE &&
2796 !can_checksum_protocol(features, type)) {
2797 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2798 } else if (illegal_highdma(skb->dev, skb)) {
2799 features &= ~NETIF_F_SG;
2805 netdev_features_t passthru_features_check(struct sk_buff *skb,
2806 struct net_device *dev,
2807 netdev_features_t features)
2811 EXPORT_SYMBOL(passthru_features_check);
2813 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2814 struct net_device *dev,
2815 netdev_features_t features)
2817 return vlan_features_check(skb, features);
2820 static netdev_features_t gso_features_check(const struct sk_buff *skb,
2821 struct net_device *dev,
2822 netdev_features_t features)
2824 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2826 if (gso_segs > dev->gso_max_segs)
2827 return features & ~NETIF_F_GSO_MASK;
2829 /* Support for GSO partial features requires software
2830 * intervention before we can actually process the packets
2831 * so we need to strip support for any partial features now
2832 * and we can pull them back in after we have partially
2833 * segmented the frame.
2835 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
2836 features &= ~dev->gso_partial_features;
2838 /* Make sure to clear the IPv4 ID mangling feature if the
2839 * IPv4 header has the potential to be fragmented.
2841 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2842 struct iphdr *iph = skb->encapsulation ?
2843 inner_ip_hdr(skb) : ip_hdr(skb);
2845 if (!(iph->frag_off & htons(IP_DF)))
2846 features &= ~NETIF_F_TSO_MANGLEID;
2852 netdev_features_t netif_skb_features(struct sk_buff *skb)
2854 struct net_device *dev = skb->dev;
2855 netdev_features_t features = dev->features;
2857 if (skb_is_gso(skb))
2858 features = gso_features_check(skb, dev, features);
2860 /* If encapsulation offload request, verify we are testing
2861 * hardware encapsulation features instead of standard
2862 * features for the netdev
2864 if (skb->encapsulation)
2865 features &= dev->hw_enc_features;
2867 if (skb_vlan_tagged(skb))
2868 features = netdev_intersect_features(features,
2869 dev->vlan_features |
2870 NETIF_F_HW_VLAN_CTAG_TX |
2871 NETIF_F_HW_VLAN_STAG_TX);
2873 if (dev->netdev_ops->ndo_features_check)
2874 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2877 features &= dflt_features_check(skb, dev, features);
2879 return harmonize_features(skb, features);
2881 EXPORT_SYMBOL(netif_skb_features);
2883 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2884 struct netdev_queue *txq, bool more)
2889 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2890 dev_queue_xmit_nit(skb, dev);
2893 trace_net_dev_start_xmit(skb, dev);
2894 rc = netdev_start_xmit(skb, dev, txq, more);
2895 trace_net_dev_xmit(skb, rc, dev, len);
2900 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2901 struct netdev_queue *txq, int *ret)
2903 struct sk_buff *skb = first;
2904 int rc = NETDEV_TX_OK;
2907 struct sk_buff *next = skb->next;
2910 rc = xmit_one(skb, dev, txq, next != NULL);
2911 if (unlikely(!dev_xmit_complete(rc))) {
2917 if (netif_xmit_stopped(txq) && skb) {
2918 rc = NETDEV_TX_BUSY;
2928 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2929 netdev_features_t features)
2931 if (skb_vlan_tag_present(skb) &&
2932 !vlan_hw_offload_capable(features, skb->vlan_proto))
2933 skb = __vlan_hwaccel_push_inside(skb);
2937 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2939 netdev_features_t features;
2941 features = netif_skb_features(skb);
2942 skb = validate_xmit_vlan(skb, features);
2946 if (netif_needs_gso(skb, features)) {
2947 struct sk_buff *segs;
2949 segs = skb_gso_segment(skb, features);
2957 if (skb_needs_linearize(skb, features) &&
2958 __skb_linearize(skb))
2961 /* If packet is not checksummed and device does not
2962 * support checksumming for this protocol, complete
2963 * checksumming here.
2965 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2966 if (skb->encapsulation)
2967 skb_set_inner_transport_header(skb,
2968 skb_checksum_start_offset(skb));
2970 skb_set_transport_header(skb,
2971 skb_checksum_start_offset(skb));
2972 if (!(features & NETIF_F_CSUM_MASK) &&
2973 skb_checksum_help(skb))
2983 atomic_long_inc(&dev->tx_dropped);
2987 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2989 struct sk_buff *next, *head = NULL, *tail;
2991 for (; skb != NULL; skb = next) {
2995 /* in case skb wont be segmented, point to itself */
2998 skb = validate_xmit_skb(skb, dev);
3006 /* If skb was segmented, skb->prev points to
3007 * the last segment. If not, it still contains skb.
3013 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3015 static void qdisc_pkt_len_init(struct sk_buff *skb)
3017 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3019 qdisc_skb_cb(skb)->pkt_len = skb->len;
3021 /* To get more precise estimation of bytes sent on wire,
3022 * we add to pkt_len the headers size of all segments
3024 if (shinfo->gso_size) {
3025 unsigned int hdr_len;
3026 u16 gso_segs = shinfo->gso_segs;
3028 /* mac layer + network layer */
3029 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3031 /* + transport layer */
3032 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
3033 hdr_len += tcp_hdrlen(skb);
3035 hdr_len += sizeof(struct udphdr);
3037 if (shinfo->gso_type & SKB_GSO_DODGY)
3038 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3041 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3045 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3046 struct net_device *dev,
3047 struct netdev_queue *txq)
3049 spinlock_t *root_lock = qdisc_lock(q);
3050 struct sk_buff *to_free = NULL;
3054 qdisc_calculate_pkt_len(skb, q);
3056 * Heuristic to force contended enqueues to serialize on a
3057 * separate lock before trying to get qdisc main lock.
3058 * This permits qdisc->running owner to get the lock more
3059 * often and dequeue packets faster.
3061 contended = qdisc_is_running(q);
3062 if (unlikely(contended))
3063 spin_lock(&q->busylock);
3065 spin_lock(root_lock);
3066 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3067 __qdisc_drop(skb, &to_free);
3069 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3070 qdisc_run_begin(q)) {
3072 * This is a work-conserving queue; there are no old skbs
3073 * waiting to be sent out; and the qdisc is not running -
3074 * xmit the skb directly.
3077 qdisc_bstats_update(q, skb);
3079 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3080 if (unlikely(contended)) {
3081 spin_unlock(&q->busylock);
3088 rc = NET_XMIT_SUCCESS;
3090 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3091 if (qdisc_run_begin(q)) {
3092 if (unlikely(contended)) {
3093 spin_unlock(&q->busylock);
3099 spin_unlock(root_lock);
3100 if (unlikely(to_free))
3101 kfree_skb_list(to_free);
3102 if (unlikely(contended))
3103 spin_unlock(&q->busylock);
3107 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3108 static void skb_update_prio(struct sk_buff *skb)
3110 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
3112 if (!skb->priority && skb->sk && map) {
3113 unsigned int prioidx =
3114 sock_cgroup_prioidx(&skb->sk->sk_cgrp_data);
3116 if (prioidx < map->priomap_len)
3117 skb->priority = map->priomap[prioidx];
3121 #define skb_update_prio(skb)
3124 DEFINE_PER_CPU(int, xmit_recursion);
3125 EXPORT_SYMBOL(xmit_recursion);
3128 * dev_loopback_xmit - loop back @skb
3129 * @net: network namespace this loopback is happening in
3130 * @sk: sk needed to be a netfilter okfn
3131 * @skb: buffer to transmit
3133 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3135 skb_reset_mac_header(skb);
3136 __skb_pull(skb, skb_network_offset(skb));
3137 skb->pkt_type = PACKET_LOOPBACK;
3138 skb->ip_summed = CHECKSUM_UNNECESSARY;
3139 WARN_ON(!skb_dst(skb));
3144 EXPORT_SYMBOL(dev_loopback_xmit);
3146 #ifdef CONFIG_NET_EGRESS
3147 static struct sk_buff *
3148 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3150 struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list);
3151 struct tcf_result cl_res;
3156 /* skb->tc_verd and qdisc_skb_cb(skb)->pkt_len were already set
3157 * earlier by the caller.
3159 qdisc_bstats_cpu_update(cl->q, skb);
3161 switch (tc_classify(skb, cl, &cl_res, false)) {
3163 case TC_ACT_RECLASSIFY:
3164 skb->tc_index = TC_H_MIN(cl_res.classid);
3167 qdisc_qstats_cpu_drop(cl->q);
3168 *ret = NET_XMIT_DROP;
3173 *ret = NET_XMIT_SUCCESS;
3176 case TC_ACT_REDIRECT:
3177 /* No need to push/pop skb's mac_header here on egress! */
3178 skb_do_redirect(skb);
3179 *ret = NET_XMIT_SUCCESS;
3187 #endif /* CONFIG_NET_EGRESS */
3189 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3192 struct xps_dev_maps *dev_maps;
3193 struct xps_map *map;
3194 int queue_index = -1;
3197 dev_maps = rcu_dereference(dev->xps_maps);
3199 unsigned int tci = skb->sender_cpu - 1;
3203 tci += netdev_get_prio_tc_map(dev, skb->priority);
3206 map = rcu_dereference(dev_maps->cpu_map[tci]);
3209 queue_index = map->queues[0];
3211 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3213 if (unlikely(queue_index >= dev->real_num_tx_queues))
3225 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3227 struct sock *sk = skb->sk;
3228 int queue_index = sk_tx_queue_get(sk);
3230 if (queue_index < 0 || skb->ooo_okay ||
3231 queue_index >= dev->real_num_tx_queues) {
3232 int new_index = get_xps_queue(dev, skb);
3234 new_index = skb_tx_hash(dev, skb);
3236 if (queue_index != new_index && sk &&
3238 rcu_access_pointer(sk->sk_dst_cache))
3239 sk_tx_queue_set(sk, new_index);
3241 queue_index = new_index;
3247 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3248 struct sk_buff *skb,
3251 int queue_index = 0;
3254 u32 sender_cpu = skb->sender_cpu - 1;
3256 if (sender_cpu >= (u32)NR_CPUS)
3257 skb->sender_cpu = raw_smp_processor_id() + 1;
3260 if (dev->real_num_tx_queues != 1) {
3261 const struct net_device_ops *ops = dev->netdev_ops;
3262 if (ops->ndo_select_queue)
3263 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3266 queue_index = __netdev_pick_tx(dev, skb);
3269 queue_index = netdev_cap_txqueue(dev, queue_index);
3272 skb_set_queue_mapping(skb, queue_index);
3273 return netdev_get_tx_queue(dev, queue_index);
3277 * __dev_queue_xmit - transmit a buffer
3278 * @skb: buffer to transmit
3279 * @accel_priv: private data used for L2 forwarding offload
3281 * Queue a buffer for transmission to a network device. The caller must
3282 * have set the device and priority and built the buffer before calling
3283 * this function. The function can be called from an interrupt.
3285 * A negative errno code is returned on a failure. A success does not
3286 * guarantee the frame will be transmitted as it may be dropped due
3287 * to congestion or traffic shaping.
3289 * -----------------------------------------------------------------------------------
3290 * I notice this method can also return errors from the queue disciplines,
3291 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3294 * Regardless of the return value, the skb is consumed, so it is currently
3295 * difficult to retry a send to this method. (You can bump the ref count
3296 * before sending to hold a reference for retry if you are careful.)
3298 * When calling this method, interrupts MUST be enabled. This is because
3299 * the BH enable code must have IRQs enabled so that it will not deadlock.
3302 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3304 struct net_device *dev = skb->dev;
3305 struct netdev_queue *txq;
3309 skb_reset_mac_header(skb);
3311 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3312 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3314 /* Disable soft irqs for various locks below. Also
3315 * stops preemption for RCU.
3319 skb_update_prio(skb);
3321 qdisc_pkt_len_init(skb);
3322 #ifdef CONFIG_NET_CLS_ACT
3323 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3324 # ifdef CONFIG_NET_EGRESS
3325 if (static_key_false(&egress_needed)) {
3326 skb = sch_handle_egress(skb, &rc, dev);
3332 /* If device/qdisc don't need skb->dst, release it right now while
3333 * its hot in this cpu cache.
3335 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3340 txq = netdev_pick_tx(dev, skb, accel_priv);
3341 q = rcu_dereference_bh(txq->qdisc);
3343 trace_net_dev_queue(skb);
3345 rc = __dev_xmit_skb(skb, q, dev, txq);
3349 /* The device has no queue. Common case for software devices:
3350 loopback, all the sorts of tunnels...
3352 Really, it is unlikely that netif_tx_lock protection is necessary
3353 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3355 However, it is possible, that they rely on protection
3358 Check this and shot the lock. It is not prone from deadlocks.
3359 Either shot noqueue qdisc, it is even simpler 8)
3361 if (dev->flags & IFF_UP) {
3362 int cpu = smp_processor_id(); /* ok because BHs are off */
3364 if (txq->xmit_lock_owner != cpu) {
3365 if (unlikely(__this_cpu_read(xmit_recursion) >
3366 XMIT_RECURSION_LIMIT))
3367 goto recursion_alert;
3369 skb = validate_xmit_skb(skb, dev);
3373 HARD_TX_LOCK(dev, txq, cpu);
3375 if (!netif_xmit_stopped(txq)) {
3376 __this_cpu_inc(xmit_recursion);
3377 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3378 __this_cpu_dec(xmit_recursion);
3379 if (dev_xmit_complete(rc)) {
3380 HARD_TX_UNLOCK(dev, txq);
3384 HARD_TX_UNLOCK(dev, txq);
3385 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3388 /* Recursion is detected! It is possible,
3392 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3398 rcu_read_unlock_bh();
3400 atomic_long_inc(&dev->tx_dropped);
3401 kfree_skb_list(skb);
3404 rcu_read_unlock_bh();
3408 int dev_queue_xmit(struct sk_buff *skb)
3410 return __dev_queue_xmit(skb, NULL);
3412 EXPORT_SYMBOL(dev_queue_xmit);
3414 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3416 return __dev_queue_xmit(skb, accel_priv);
3418 EXPORT_SYMBOL(dev_queue_xmit_accel);
3421 /*=======================================================================
3423 =======================================================================*/
3425 int netdev_max_backlog __read_mostly = 1000;
3426 EXPORT_SYMBOL(netdev_max_backlog);
3428 int netdev_tstamp_prequeue __read_mostly = 1;
3429 int netdev_budget __read_mostly = 300;
3430 int weight_p __read_mostly = 64; /* old backlog weight */
3432 /* Called with irq disabled */
3433 static inline void ____napi_schedule(struct softnet_data *sd,
3434 struct napi_struct *napi)
3436 list_add_tail(&napi->poll_list, &sd->poll_list);
3437 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3442 /* One global table that all flow-based protocols share. */
3443 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3444 EXPORT_SYMBOL(rps_sock_flow_table);
3445 u32 rps_cpu_mask __read_mostly;
3446 EXPORT_SYMBOL(rps_cpu_mask);
3448 struct static_key rps_needed __read_mostly;
3449 EXPORT_SYMBOL(rps_needed);
3451 static struct rps_dev_flow *
3452 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3453 struct rps_dev_flow *rflow, u16 next_cpu)
3455 if (next_cpu < nr_cpu_ids) {
3456 #ifdef CONFIG_RFS_ACCEL
3457 struct netdev_rx_queue *rxqueue;
3458 struct rps_dev_flow_table *flow_table;
3459 struct rps_dev_flow *old_rflow;
3464 /* Should we steer this flow to a different hardware queue? */
3465 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3466 !(dev->features & NETIF_F_NTUPLE))
3468 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3469 if (rxq_index == skb_get_rx_queue(skb))
3472 rxqueue = dev->_rx + rxq_index;
3473 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3476 flow_id = skb_get_hash(skb) & flow_table->mask;
3477 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3478 rxq_index, flow_id);
3482 rflow = &flow_table->flows[flow_id];
3484 if (old_rflow->filter == rflow->filter)
3485 old_rflow->filter = RPS_NO_FILTER;
3489 per_cpu(softnet_data, next_cpu).input_queue_head;
3492 rflow->cpu = next_cpu;
3497 * get_rps_cpu is called from netif_receive_skb and returns the target
3498 * CPU from the RPS map of the receiving queue for a given skb.
3499 * rcu_read_lock must be held on entry.
3501 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3502 struct rps_dev_flow **rflowp)
3504 const struct rps_sock_flow_table *sock_flow_table;
3505 struct netdev_rx_queue *rxqueue = dev->_rx;
3506 struct rps_dev_flow_table *flow_table;
3507 struct rps_map *map;
3512 if (skb_rx_queue_recorded(skb)) {
3513 u16 index = skb_get_rx_queue(skb);
3515 if (unlikely(index >= dev->real_num_rx_queues)) {
3516 WARN_ONCE(dev->real_num_rx_queues > 1,
3517 "%s received packet on queue %u, but number "
3518 "of RX queues is %u\n",
3519 dev->name, index, dev->real_num_rx_queues);
3525 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3527 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3528 map = rcu_dereference(rxqueue->rps_map);
3529 if (!flow_table && !map)
3532 skb_reset_network_header(skb);
3533 hash = skb_get_hash(skb);
3537 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3538 if (flow_table && sock_flow_table) {
3539 struct rps_dev_flow *rflow;
3543 /* First check into global flow table if there is a match */
3544 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3545 if ((ident ^ hash) & ~rps_cpu_mask)
3548 next_cpu = ident & rps_cpu_mask;
3550 /* OK, now we know there is a match,
3551 * we can look at the local (per receive queue) flow table
3553 rflow = &flow_table->flows[hash & flow_table->mask];
3557 * If the desired CPU (where last recvmsg was done) is
3558 * different from current CPU (one in the rx-queue flow
3559 * table entry), switch if one of the following holds:
3560 * - Current CPU is unset (>= nr_cpu_ids).
3561 * - Current CPU is offline.
3562 * - The current CPU's queue tail has advanced beyond the
3563 * last packet that was enqueued using this table entry.
3564 * This guarantees that all previous packets for the flow
3565 * have been dequeued, thus preserving in order delivery.
3567 if (unlikely(tcpu != next_cpu) &&
3568 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3569 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3570 rflow->last_qtail)) >= 0)) {
3572 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3575 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3585 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3586 if (cpu_online(tcpu)) {
3596 #ifdef CONFIG_RFS_ACCEL
3599 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3600 * @dev: Device on which the filter was set
3601 * @rxq_index: RX queue index
3602 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3603 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3605 * Drivers that implement ndo_rx_flow_steer() should periodically call
3606 * this function for each installed filter and remove the filters for
3607 * which it returns %true.
3609 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3610 u32 flow_id, u16 filter_id)
3612 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3613 struct rps_dev_flow_table *flow_table;
3614 struct rps_dev_flow *rflow;
3619 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3620 if (flow_table && flow_id <= flow_table->mask) {
3621 rflow = &flow_table->flows[flow_id];
3622 cpu = ACCESS_ONCE(rflow->cpu);
3623 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3624 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3625 rflow->last_qtail) <
3626 (int)(10 * flow_table->mask)))
3632 EXPORT_SYMBOL(rps_may_expire_flow);
3634 #endif /* CONFIG_RFS_ACCEL */
3636 /* Called from hardirq (IPI) context */
3637 static void rps_trigger_softirq(void *data)
3639 struct softnet_data *sd = data;
3641 ____napi_schedule(sd, &sd->backlog);
3645 #endif /* CONFIG_RPS */
3648 * Check if this softnet_data structure is another cpu one
3649 * If yes, queue it to our IPI list and return 1
3652 static int rps_ipi_queued(struct softnet_data *sd)
3655 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3658 sd->rps_ipi_next = mysd->rps_ipi_list;
3659 mysd->rps_ipi_list = sd;
3661 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3664 #endif /* CONFIG_RPS */
3668 #ifdef CONFIG_NET_FLOW_LIMIT
3669 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3672 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3674 #ifdef CONFIG_NET_FLOW_LIMIT
3675 struct sd_flow_limit *fl;
3676 struct softnet_data *sd;
3677 unsigned int old_flow, new_flow;
3679 if (qlen < (netdev_max_backlog >> 1))
3682 sd = this_cpu_ptr(&softnet_data);
3685 fl = rcu_dereference(sd->flow_limit);
3687 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3688 old_flow = fl->history[fl->history_head];
3689 fl->history[fl->history_head] = new_flow;
3692 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3694 if (likely(fl->buckets[old_flow]))
3695 fl->buckets[old_flow]--;
3697 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3709 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3710 * queue (may be a remote CPU queue).
3712 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3713 unsigned int *qtail)
3715 struct softnet_data *sd;
3716 unsigned long flags;
3719 sd = &per_cpu(softnet_data, cpu);
3721 local_irq_save(flags);
3724 if (!netif_running(skb->dev))
3726 qlen = skb_queue_len(&sd->input_pkt_queue);
3727 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3730 __skb_queue_tail(&sd->input_pkt_queue, skb);
3731 input_queue_tail_incr_save(sd, qtail);
3733 local_irq_restore(flags);
3734 return NET_RX_SUCCESS;
3737 /* Schedule NAPI for backlog device
3738 * We can use non atomic operation since we own the queue lock
3740 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3741 if (!rps_ipi_queued(sd))
3742 ____napi_schedule(sd, &sd->backlog);
3751 local_irq_restore(flags);
3753 atomic_long_inc(&skb->dev->rx_dropped);
3758 static int netif_rx_internal(struct sk_buff *skb)
3762 net_timestamp_check(netdev_tstamp_prequeue, skb);
3764 trace_netif_rx(skb);
3766 if (static_key_false(&rps_needed)) {
3767 struct rps_dev_flow voidflow, *rflow = &voidflow;
3773 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3775 cpu = smp_processor_id();
3777 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3785 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3792 * netif_rx - post buffer to the network code
3793 * @skb: buffer to post
3795 * This function receives a packet from a device driver and queues it for
3796 * the upper (protocol) levels to process. It always succeeds. The buffer
3797 * may be dropped during processing for congestion control or by the
3801 * NET_RX_SUCCESS (no congestion)
3802 * NET_RX_DROP (packet was dropped)
3806 int netif_rx(struct sk_buff *skb)
3808 trace_netif_rx_entry(skb);
3810 return netif_rx_internal(skb);
3812 EXPORT_SYMBOL(netif_rx);
3814 int netif_rx_ni(struct sk_buff *skb)
3818 trace_netif_rx_ni_entry(skb);
3821 err = netif_rx_internal(skb);
3822 if (local_softirq_pending())
3828 EXPORT_SYMBOL(netif_rx_ni);
3830 static __latent_entropy void net_tx_action(struct softirq_action *h)
3832 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3834 if (sd->completion_queue) {
3835 struct sk_buff *clist;
3837 local_irq_disable();
3838 clist = sd->completion_queue;
3839 sd->completion_queue = NULL;
3843 struct sk_buff *skb = clist;
3844 clist = clist->next;
3846 WARN_ON(atomic_read(&skb->users));
3847 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3848 trace_consume_skb(skb);
3850 trace_kfree_skb(skb, net_tx_action);
3852 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
3855 __kfree_skb_defer(skb);
3858 __kfree_skb_flush();
3861 if (sd->output_queue) {
3864 local_irq_disable();
3865 head = sd->output_queue;
3866 sd->output_queue = NULL;
3867 sd->output_queue_tailp = &sd->output_queue;
3871 struct Qdisc *q = head;
3872 spinlock_t *root_lock;
3874 head = head->next_sched;
3876 root_lock = qdisc_lock(q);
3877 spin_lock(root_lock);
3878 /* We need to make sure head->next_sched is read
3879 * before clearing __QDISC_STATE_SCHED
3881 smp_mb__before_atomic();
3882 clear_bit(__QDISC_STATE_SCHED, &q->state);
3884 spin_unlock(root_lock);
3889 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
3890 /* This hook is defined here for ATM LANE */
3891 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3892 unsigned char *addr) __read_mostly;
3893 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3896 static inline struct sk_buff *
3897 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
3898 struct net_device *orig_dev)
3900 #ifdef CONFIG_NET_CLS_ACT
3901 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3902 struct tcf_result cl_res;
3904 /* If there's at least one ingress present somewhere (so
3905 * we get here via enabled static key), remaining devices
3906 * that are not configured with an ingress qdisc will bail
3912 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3916 qdisc_skb_cb(skb)->pkt_len = skb->len;
3917 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3918 qdisc_bstats_cpu_update(cl->q, skb);
3920 switch (tc_classify(skb, cl, &cl_res, false)) {
3922 case TC_ACT_RECLASSIFY:
3923 skb->tc_index = TC_H_MIN(cl_res.classid);
3926 qdisc_qstats_cpu_drop(cl->q);
3933 case TC_ACT_REDIRECT:
3934 /* skb_mac_header check was done by cls/act_bpf, so
3935 * we can safely push the L2 header back before
3936 * redirecting to another netdev
3938 __skb_push(skb, skb->mac_len);
3939 skb_do_redirect(skb);
3944 #endif /* CONFIG_NET_CLS_ACT */
3949 * netdev_is_rx_handler_busy - check if receive handler is registered
3950 * @dev: device to check
3952 * Check if a receive handler is already registered for a given device.
3953 * Return true if there one.
3955 * The caller must hold the rtnl_mutex.
3957 bool netdev_is_rx_handler_busy(struct net_device *dev)
3960 return dev && rtnl_dereference(dev->rx_handler);
3962 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
3965 * netdev_rx_handler_register - register receive handler
3966 * @dev: device to register a handler for
3967 * @rx_handler: receive handler to register
3968 * @rx_handler_data: data pointer that is used by rx handler
3970 * Register a receive handler for a device. This handler will then be
3971 * called from __netif_receive_skb. A negative errno code is returned
3974 * The caller must hold the rtnl_mutex.
3976 * For a general description of rx_handler, see enum rx_handler_result.
3978 int netdev_rx_handler_register(struct net_device *dev,
3979 rx_handler_func_t *rx_handler,
3980 void *rx_handler_data)
3984 if (dev->rx_handler)
3987 /* Note: rx_handler_data must be set before rx_handler */
3988 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3989 rcu_assign_pointer(dev->rx_handler, rx_handler);
3993 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3996 * netdev_rx_handler_unregister - unregister receive handler
3997 * @dev: device to unregister a handler from
3999 * Unregister a receive handler from a device.
4001 * The caller must hold the rtnl_mutex.
4003 void netdev_rx_handler_unregister(struct net_device *dev)
4007 RCU_INIT_POINTER(dev->rx_handler, NULL);
4008 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4009 * section has a guarantee to see a non NULL rx_handler_data
4013 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4015 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4018 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4019 * the special handling of PFMEMALLOC skbs.
4021 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4023 switch (skb->protocol) {
4024 case htons(ETH_P_ARP):
4025 case htons(ETH_P_IP):
4026 case htons(ETH_P_IPV6):
4027 case htons(ETH_P_8021Q):
4028 case htons(ETH_P_8021AD):
4035 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4036 int *ret, struct net_device *orig_dev)
4038 #ifdef CONFIG_NETFILTER_INGRESS
4039 if (nf_hook_ingress_active(skb)) {
4043 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4048 ingress_retval = nf_hook_ingress(skb);
4050 return ingress_retval;
4052 #endif /* CONFIG_NETFILTER_INGRESS */
4056 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
4058 struct packet_type *ptype, *pt_prev;
4059 rx_handler_func_t *rx_handler;
4060 struct net_device *orig_dev;
4061 bool deliver_exact = false;
4062 int ret = NET_RX_DROP;
4065 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4067 trace_netif_receive_skb(skb);
4069 orig_dev = skb->dev;
4071 skb_reset_network_header(skb);
4072 if (!skb_transport_header_was_set(skb))
4073 skb_reset_transport_header(skb);
4074 skb_reset_mac_len(skb);
4079 skb->skb_iif = skb->dev->ifindex;
4081 __this_cpu_inc(softnet_data.processed);
4083 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4084 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4085 skb = skb_vlan_untag(skb);
4090 #ifdef CONFIG_NET_CLS_ACT
4091 if (skb->tc_verd & TC_NCLS) {
4092 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
4100 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4102 ret = deliver_skb(skb, pt_prev, orig_dev);
4106 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4108 ret = deliver_skb(skb, pt_prev, orig_dev);
4113 #ifdef CONFIG_NET_INGRESS
4114 if (static_key_false(&ingress_needed)) {
4115 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4119 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4123 #ifdef CONFIG_NET_CLS_ACT
4127 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4130 if (skb_vlan_tag_present(skb)) {
4132 ret = deliver_skb(skb, pt_prev, orig_dev);
4135 if (vlan_do_receive(&skb))
4137 else if (unlikely(!skb))
4141 rx_handler = rcu_dereference(skb->dev->rx_handler);
4144 ret = deliver_skb(skb, pt_prev, orig_dev);
4147 switch (rx_handler(&skb)) {
4148 case RX_HANDLER_CONSUMED:
4149 ret = NET_RX_SUCCESS;
4151 case RX_HANDLER_ANOTHER:
4153 case RX_HANDLER_EXACT:
4154 deliver_exact = true;
4155 case RX_HANDLER_PASS:
4162 if (unlikely(skb_vlan_tag_present(skb))) {
4163 if (skb_vlan_tag_get_id(skb))
4164 skb->pkt_type = PACKET_OTHERHOST;
4165 /* Note: we might in the future use prio bits
4166 * and set skb->priority like in vlan_do_receive()
4167 * For the time being, just ignore Priority Code Point
4172 type = skb->protocol;
4174 /* deliver only exact match when indicated */
4175 if (likely(!deliver_exact)) {
4176 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4177 &ptype_base[ntohs(type) &
4181 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4182 &orig_dev->ptype_specific);
4184 if (unlikely(skb->dev != orig_dev)) {
4185 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4186 &skb->dev->ptype_specific);
4190 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
4193 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4197 atomic_long_inc(&skb->dev->rx_dropped);
4199 atomic_long_inc(&skb->dev->rx_nohandler);
4201 /* Jamal, now you will not able to escape explaining
4202 * me how you were going to use this. :-)
4211 static int __netif_receive_skb(struct sk_buff *skb)
4215 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4216 unsigned long pflags = current->flags;
4219 * PFMEMALLOC skbs are special, they should
4220 * - be delivered to SOCK_MEMALLOC sockets only
4221 * - stay away from userspace
4222 * - have bounded memory usage
4224 * Use PF_MEMALLOC as this saves us from propagating the allocation
4225 * context down to all allocation sites.
4227 current->flags |= PF_MEMALLOC;
4228 ret = __netif_receive_skb_core(skb, true);
4229 tsk_restore_flags(current, pflags, PF_MEMALLOC);
4231 ret = __netif_receive_skb_core(skb, false);
4236 static int netif_receive_skb_internal(struct sk_buff *skb)
4240 net_timestamp_check(netdev_tstamp_prequeue, skb);
4242 if (skb_defer_rx_timestamp(skb))
4243 return NET_RX_SUCCESS;
4248 if (static_key_false(&rps_needed)) {
4249 struct rps_dev_flow voidflow, *rflow = &voidflow;
4250 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4253 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4259 ret = __netif_receive_skb(skb);
4265 * netif_receive_skb - process receive buffer from network
4266 * @skb: buffer to process
4268 * netif_receive_skb() is the main receive data processing function.
4269 * It always succeeds. The buffer may be dropped during processing
4270 * for congestion control or by the protocol layers.
4272 * This function may only be called from softirq context and interrupts
4273 * should be enabled.
4275 * Return values (usually ignored):
4276 * NET_RX_SUCCESS: no congestion
4277 * NET_RX_DROP: packet was dropped
4279 int netif_receive_skb(struct sk_buff *skb)
4281 trace_netif_receive_skb_entry(skb);
4283 return netif_receive_skb_internal(skb);
4285 EXPORT_SYMBOL(netif_receive_skb);
4287 DEFINE_PER_CPU(struct work_struct, flush_works);
4289 /* Network device is going away, flush any packets still pending */
4290 static void flush_backlog(struct work_struct *work)
4292 struct sk_buff *skb, *tmp;
4293 struct softnet_data *sd;
4296 sd = this_cpu_ptr(&softnet_data);
4298 local_irq_disable();
4300 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4301 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4302 __skb_unlink(skb, &sd->input_pkt_queue);
4304 input_queue_head_incr(sd);
4310 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4311 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4312 __skb_unlink(skb, &sd->process_queue);
4314 input_queue_head_incr(sd);
4320 static void flush_all_backlogs(void)
4326 for_each_online_cpu(cpu)
4327 queue_work_on(cpu, system_highpri_wq,
4328 per_cpu_ptr(&flush_works, cpu));
4330 for_each_online_cpu(cpu)
4331 flush_work(per_cpu_ptr(&flush_works, cpu));
4336 static int napi_gro_complete(struct sk_buff *skb)
4338 struct packet_offload *ptype;
4339 __be16 type = skb->protocol;
4340 struct list_head *head = &offload_base;
4343 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4345 if (NAPI_GRO_CB(skb)->count == 1) {
4346 skb_shinfo(skb)->gso_size = 0;
4351 list_for_each_entry_rcu(ptype, head, list) {
4352 if (ptype->type != type || !ptype->callbacks.gro_complete)
4355 err = ptype->callbacks.gro_complete(skb, 0);
4361 WARN_ON(&ptype->list == head);
4363 return NET_RX_SUCCESS;
4367 return netif_receive_skb_internal(skb);
4370 /* napi->gro_list contains packets ordered by age.
4371 * youngest packets at the head of it.
4372 * Complete skbs in reverse order to reduce latencies.
4374 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4376 struct sk_buff *skb, *prev = NULL;
4378 /* scan list and build reverse chain */
4379 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4384 for (skb = prev; skb; skb = prev) {
4387 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4391 napi_gro_complete(skb);
4395 napi->gro_list = NULL;
4397 EXPORT_SYMBOL(napi_gro_flush);
4399 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4402 unsigned int maclen = skb->dev->hard_header_len;
4403 u32 hash = skb_get_hash_raw(skb);
4405 for (p = napi->gro_list; p; p = p->next) {
4406 unsigned long diffs;
4408 NAPI_GRO_CB(p)->flush = 0;
4410 if (hash != skb_get_hash_raw(p)) {
4411 NAPI_GRO_CB(p)->same_flow = 0;
4415 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4416 diffs |= p->vlan_tci ^ skb->vlan_tci;
4417 diffs |= skb_metadata_dst_cmp(p, skb);
4418 if (maclen == ETH_HLEN)
4419 diffs |= compare_ether_header(skb_mac_header(p),
4420 skb_mac_header(skb));
4422 diffs = memcmp(skb_mac_header(p),
4423 skb_mac_header(skb),
4425 NAPI_GRO_CB(p)->same_flow = !diffs;
4429 static void skb_gro_reset_offset(struct sk_buff *skb)
4431 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4432 const skb_frag_t *frag0 = &pinfo->frags[0];
4434 NAPI_GRO_CB(skb)->data_offset = 0;
4435 NAPI_GRO_CB(skb)->frag0 = NULL;
4436 NAPI_GRO_CB(skb)->frag0_len = 0;
4438 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4440 !PageHighMem(skb_frag_page(frag0))) {
4441 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4442 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4446 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4448 struct skb_shared_info *pinfo = skb_shinfo(skb);
4450 BUG_ON(skb->end - skb->tail < grow);
4452 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4454 skb->data_len -= grow;
4457 pinfo->frags[0].page_offset += grow;
4458 skb_frag_size_sub(&pinfo->frags[0], grow);
4460 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4461 skb_frag_unref(skb, 0);
4462 memmove(pinfo->frags, pinfo->frags + 1,
4463 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4467 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4469 struct sk_buff **pp = NULL;
4470 struct packet_offload *ptype;
4471 __be16 type = skb->protocol;
4472 struct list_head *head = &offload_base;
4474 enum gro_result ret;
4477 if (!(skb->dev->features & NETIF_F_GRO))
4483 gro_list_prepare(napi, skb);
4486 list_for_each_entry_rcu(ptype, head, list) {
4487 if (ptype->type != type || !ptype->callbacks.gro_receive)
4490 skb_set_network_header(skb, skb_gro_offset(skb));
4491 skb_reset_mac_len(skb);
4492 NAPI_GRO_CB(skb)->same_flow = 0;
4493 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
4494 NAPI_GRO_CB(skb)->free = 0;
4495 NAPI_GRO_CB(skb)->encap_mark = 0;
4496 NAPI_GRO_CB(skb)->recursion_counter = 0;
4497 NAPI_GRO_CB(skb)->is_fou = 0;
4498 NAPI_GRO_CB(skb)->is_atomic = 1;
4499 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4501 /* Setup for GRO checksum validation */
4502 switch (skb->ip_summed) {
4503 case CHECKSUM_COMPLETE:
4504 NAPI_GRO_CB(skb)->csum = skb->csum;
4505 NAPI_GRO_CB(skb)->csum_valid = 1;
4506 NAPI_GRO_CB(skb)->csum_cnt = 0;
4508 case CHECKSUM_UNNECESSARY:
4509 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4510 NAPI_GRO_CB(skb)->csum_valid = 0;
4513 NAPI_GRO_CB(skb)->csum_cnt = 0;
4514 NAPI_GRO_CB(skb)->csum_valid = 0;
4517 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4522 if (&ptype->list == head)
4525 same_flow = NAPI_GRO_CB(skb)->same_flow;
4526 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4529 struct sk_buff *nskb = *pp;
4533 napi_gro_complete(nskb);
4540 if (NAPI_GRO_CB(skb)->flush)
4543 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4544 struct sk_buff *nskb = napi->gro_list;
4546 /* locate the end of the list to select the 'oldest' flow */
4547 while (nskb->next) {
4553 napi_gro_complete(nskb);
4557 NAPI_GRO_CB(skb)->count = 1;
4558 NAPI_GRO_CB(skb)->age = jiffies;
4559 NAPI_GRO_CB(skb)->last = skb;
4560 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4561 skb->next = napi->gro_list;
4562 napi->gro_list = skb;
4566 grow = skb_gro_offset(skb) - skb_headlen(skb);
4568 gro_pull_from_frag0(skb, grow);
4577 struct packet_offload *gro_find_receive_by_type(__be16 type)
4579 struct list_head *offload_head = &offload_base;
4580 struct packet_offload *ptype;
4582 list_for_each_entry_rcu(ptype, offload_head, list) {
4583 if (ptype->type != type || !ptype->callbacks.gro_receive)
4589 EXPORT_SYMBOL(gro_find_receive_by_type);
4591 struct packet_offload *gro_find_complete_by_type(__be16 type)
4593 struct list_head *offload_head = &offload_base;
4594 struct packet_offload *ptype;
4596 list_for_each_entry_rcu(ptype, offload_head, list) {
4597 if (ptype->type != type || !ptype->callbacks.gro_complete)
4603 EXPORT_SYMBOL(gro_find_complete_by_type);
4605 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4609 if (netif_receive_skb_internal(skb))
4617 case GRO_MERGED_FREE:
4618 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) {
4620 kmem_cache_free(skbuff_head_cache, skb);
4634 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4636 skb_mark_napi_id(skb, napi);
4637 trace_napi_gro_receive_entry(skb);
4639 skb_gro_reset_offset(skb);
4641 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4643 EXPORT_SYMBOL(napi_gro_receive);
4645 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4647 if (unlikely(skb->pfmemalloc)) {
4651 __skb_pull(skb, skb_headlen(skb));
4652 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4653 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4655 skb->dev = napi->dev;
4657 skb->encapsulation = 0;
4658 skb_shinfo(skb)->gso_type = 0;
4659 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4664 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4666 struct sk_buff *skb = napi->skb;
4669 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4672 skb_mark_napi_id(skb, napi);
4677 EXPORT_SYMBOL(napi_get_frags);
4679 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4680 struct sk_buff *skb,
4686 __skb_push(skb, ETH_HLEN);
4687 skb->protocol = eth_type_trans(skb, skb->dev);
4688 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4693 case GRO_MERGED_FREE:
4694 napi_reuse_skb(napi, skb);
4704 /* Upper GRO stack assumes network header starts at gro_offset=0
4705 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4706 * We copy ethernet header into skb->data to have a common layout.
4708 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4710 struct sk_buff *skb = napi->skb;
4711 const struct ethhdr *eth;
4712 unsigned int hlen = sizeof(*eth);
4716 skb_reset_mac_header(skb);
4717 skb_gro_reset_offset(skb);
4719 eth = skb_gro_header_fast(skb, 0);
4720 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4721 eth = skb_gro_header_slow(skb, hlen, 0);
4722 if (unlikely(!eth)) {
4723 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
4724 __func__, napi->dev->name);
4725 napi_reuse_skb(napi, skb);
4729 gro_pull_from_frag0(skb, hlen);
4730 NAPI_GRO_CB(skb)->frag0 += hlen;
4731 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4733 __skb_pull(skb, hlen);
4736 * This works because the only protocols we care about don't require
4738 * We'll fix it up properly in napi_frags_finish()
4740 skb->protocol = eth->h_proto;
4745 gro_result_t napi_gro_frags(struct napi_struct *napi)
4747 struct sk_buff *skb = napi_frags_skb(napi);
4752 trace_napi_gro_frags_entry(skb);
4754 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4756 EXPORT_SYMBOL(napi_gro_frags);
4758 /* Compute the checksum from gro_offset and return the folded value
4759 * after adding in any pseudo checksum.
4761 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4766 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4768 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4769 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4771 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4772 !skb->csum_complete_sw)
4773 netdev_rx_csum_fault(skb->dev);
4776 NAPI_GRO_CB(skb)->csum = wsum;
4777 NAPI_GRO_CB(skb)->csum_valid = 1;
4781 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4784 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4785 * Note: called with local irq disabled, but exits with local irq enabled.
4787 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4790 struct softnet_data *remsd = sd->rps_ipi_list;
4793 sd->rps_ipi_list = NULL;
4797 /* Send pending IPI's to kick RPS processing on remote cpus. */
4799 struct softnet_data *next = remsd->rps_ipi_next;
4801 if (cpu_online(remsd->cpu))
4802 smp_call_function_single_async(remsd->cpu,
4811 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4814 return sd->rps_ipi_list != NULL;
4820 static int process_backlog(struct napi_struct *napi, int quota)
4822 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4826 /* Check if we have pending ipi, its better to send them now,
4827 * not waiting net_rx_action() end.
4829 if (sd_has_rps_ipi_waiting(sd)) {
4830 local_irq_disable();
4831 net_rps_action_and_irq_enable(sd);
4834 napi->weight = weight_p;
4836 struct sk_buff *skb;
4838 while ((skb = __skb_dequeue(&sd->process_queue))) {
4840 __netif_receive_skb(skb);
4842 input_queue_head_incr(sd);
4843 if (++work >= quota)
4848 local_irq_disable();
4850 if (skb_queue_empty(&sd->input_pkt_queue)) {
4852 * Inline a custom version of __napi_complete().
4853 * only current cpu owns and manipulates this napi,
4854 * and NAPI_STATE_SCHED is the only possible flag set
4856 * We can use a plain write instead of clear_bit(),
4857 * and we dont need an smp_mb() memory barrier.
4862 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4863 &sd->process_queue);
4873 * __napi_schedule - schedule for receive
4874 * @n: entry to schedule
4876 * The entry's receive function will be scheduled to run.
4877 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4879 void __napi_schedule(struct napi_struct *n)
4881 unsigned long flags;
4883 local_irq_save(flags);
4884 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4885 local_irq_restore(flags);
4887 EXPORT_SYMBOL(__napi_schedule);
4890 * __napi_schedule_irqoff - schedule for receive
4891 * @n: entry to schedule
4893 * Variant of __napi_schedule() assuming hard irqs are masked
4895 void __napi_schedule_irqoff(struct napi_struct *n)
4897 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4899 EXPORT_SYMBOL(__napi_schedule_irqoff);
4901 void __napi_complete(struct napi_struct *n)
4903 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4905 /* Some drivers call us directly, instead of calling
4906 * napi_complete_done().
4908 if (unlikely(test_bit(NAPI_STATE_IN_BUSY_POLL, &n->state)))
4911 list_del_init(&n->poll_list);
4912 smp_mb__before_atomic();
4913 clear_bit(NAPI_STATE_SCHED, &n->state);
4915 EXPORT_SYMBOL(__napi_complete);
4917 void napi_complete_done(struct napi_struct *n, int work_done)
4919 unsigned long flags;
4922 * 1) Don't let napi dequeue from the cpu poll list
4923 * just in case its running on a different cpu.
4924 * 2) If we are busy polling, do nothing here, we have
4925 * the guarantee we will be called later.
4927 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
4928 NAPIF_STATE_IN_BUSY_POLL)))
4932 unsigned long timeout = 0;
4935 timeout = n->dev->gro_flush_timeout;
4938 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4939 HRTIMER_MODE_REL_PINNED);
4941 napi_gro_flush(n, false);
4943 if (likely(list_empty(&n->poll_list))) {
4944 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4946 /* If n->poll_list is not empty, we need to mask irqs */
4947 local_irq_save(flags);
4949 local_irq_restore(flags);
4952 EXPORT_SYMBOL(napi_complete_done);
4954 /* must be called under rcu_read_lock(), as we dont take a reference */
4955 static struct napi_struct *napi_by_id(unsigned int napi_id)
4957 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4958 struct napi_struct *napi;
4960 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4961 if (napi->napi_id == napi_id)
4967 #if defined(CONFIG_NET_RX_BUSY_POLL)
4969 #define BUSY_POLL_BUDGET 8
4971 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
4975 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
4979 /* All we really want here is to re-enable device interrupts.
4980 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
4982 rc = napi->poll(napi, BUSY_POLL_BUDGET);
4983 netpoll_poll_unlock(have_poll_lock);
4984 if (rc == BUSY_POLL_BUDGET)
4985 __napi_schedule(napi);
4987 if (local_softirq_pending())
4991 bool sk_busy_loop(struct sock *sk, int nonblock)
4993 unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0;
4994 int (*napi_poll)(struct napi_struct *napi, int budget);
4995 int (*busy_poll)(struct napi_struct *dev);
4996 void *have_poll_lock = NULL;
4997 struct napi_struct *napi;
5006 napi = napi_by_id(sk->sk_napi_id);
5010 /* Note: ndo_busy_poll method is optional in linux-4.5 */
5011 busy_poll = napi->dev->netdev_ops->ndo_busy_poll;
5018 rc = busy_poll(napi);
5022 unsigned long val = READ_ONCE(napi->state);
5024 /* If multiple threads are competing for this napi,
5025 * we avoid dirtying napi->state as much as we can.
5027 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
5028 NAPIF_STATE_IN_BUSY_POLL))
5030 if (cmpxchg(&napi->state, val,
5031 val | NAPIF_STATE_IN_BUSY_POLL |
5032 NAPIF_STATE_SCHED) != val)
5034 have_poll_lock = netpoll_poll_lock(napi);
5035 napi_poll = napi->poll;
5037 rc = napi_poll(napi, BUSY_POLL_BUDGET);
5038 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
5041 __NET_ADD_STATS(sock_net(sk),
5042 LINUX_MIB_BUSYPOLLRXPACKETS, rc);
5045 if (rc == LL_FLUSH_FAILED)
5046 break; /* permanent failure */
5048 if (nonblock || !skb_queue_empty(&sk->sk_receive_queue) ||
5049 busy_loop_timeout(end_time))
5052 if (unlikely(need_resched())) {
5054 busy_poll_stop(napi, have_poll_lock);
5058 rc = !skb_queue_empty(&sk->sk_receive_queue);
5059 if (rc || busy_loop_timeout(end_time))
5063 cpu_relax_lowlatency();
5066 busy_poll_stop(napi, have_poll_lock);
5068 rc = !skb_queue_empty(&sk->sk_receive_queue);
5073 EXPORT_SYMBOL(sk_busy_loop);
5075 #endif /* CONFIG_NET_RX_BUSY_POLL */
5077 static void napi_hash_add(struct napi_struct *napi)
5079 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
5080 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
5083 spin_lock(&napi_hash_lock);
5085 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */
5087 if (unlikely(++napi_gen_id < NR_CPUS + 1))
5088 napi_gen_id = NR_CPUS + 1;
5089 } while (napi_by_id(napi_gen_id));
5090 napi->napi_id = napi_gen_id;
5092 hlist_add_head_rcu(&napi->napi_hash_node,
5093 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
5095 spin_unlock(&napi_hash_lock);
5098 /* Warning : caller is responsible to make sure rcu grace period
5099 * is respected before freeing memory containing @napi
5101 bool napi_hash_del(struct napi_struct *napi)
5103 bool rcu_sync_needed = false;
5105 spin_lock(&napi_hash_lock);
5107 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
5108 rcu_sync_needed = true;
5109 hlist_del_rcu(&napi->napi_hash_node);
5111 spin_unlock(&napi_hash_lock);
5112 return rcu_sync_needed;
5114 EXPORT_SYMBOL_GPL(napi_hash_del);
5116 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
5118 struct napi_struct *napi;
5120 napi = container_of(timer, struct napi_struct, timer);
5122 napi_schedule(napi);
5124 return HRTIMER_NORESTART;
5127 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
5128 int (*poll)(struct napi_struct *, int), int weight)
5130 INIT_LIST_HEAD(&napi->poll_list);
5131 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
5132 napi->timer.function = napi_watchdog;
5133 napi->gro_count = 0;
5134 napi->gro_list = NULL;
5137 if (weight > NAPI_POLL_WEIGHT)
5138 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5140 napi->weight = weight;
5141 list_add(&napi->dev_list, &dev->napi_list);
5143 #ifdef CONFIG_NETPOLL
5144 spin_lock_init(&napi->poll_lock);
5145 napi->poll_owner = -1;
5147 set_bit(NAPI_STATE_SCHED, &napi->state);
5148 napi_hash_add(napi);
5150 EXPORT_SYMBOL(netif_napi_add);
5152 void napi_disable(struct napi_struct *n)
5155 set_bit(NAPI_STATE_DISABLE, &n->state);
5157 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
5159 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
5162 hrtimer_cancel(&n->timer);
5164 clear_bit(NAPI_STATE_DISABLE, &n->state);
5166 EXPORT_SYMBOL(napi_disable);
5168 /* Must be called in process context */
5169 void netif_napi_del(struct napi_struct *napi)
5172 if (napi_hash_del(napi))
5174 list_del_init(&napi->dev_list);
5175 napi_free_frags(napi);
5177 kfree_skb_list(napi->gro_list);
5178 napi->gro_list = NULL;
5179 napi->gro_count = 0;
5181 EXPORT_SYMBOL(netif_napi_del);
5183 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
5188 list_del_init(&n->poll_list);
5190 have = netpoll_poll_lock(n);
5194 /* This NAPI_STATE_SCHED test is for avoiding a race
5195 * with netpoll's poll_napi(). Only the entity which
5196 * obtains the lock and sees NAPI_STATE_SCHED set will
5197 * actually make the ->poll() call. Therefore we avoid
5198 * accidentally calling ->poll() when NAPI is not scheduled.
5201 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
5202 work = n->poll(n, weight);
5203 trace_napi_poll(n, work, weight);
5206 WARN_ON_ONCE(work > weight);
5208 if (likely(work < weight))
5211 /* Drivers must not modify the NAPI state if they
5212 * consume the entire weight. In such cases this code
5213 * still "owns" the NAPI instance and therefore can
5214 * move the instance around on the list at-will.
5216 if (unlikely(napi_disable_pending(n))) {
5222 /* flush too old packets
5223 * If HZ < 1000, flush all packets.
5225 napi_gro_flush(n, HZ >= 1000);
5228 /* Some drivers may have called napi_schedule
5229 * prior to exhausting their budget.
5231 if (unlikely(!list_empty(&n->poll_list))) {
5232 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5233 n->dev ? n->dev->name : "backlog");
5237 list_add_tail(&n->poll_list, repoll);
5240 netpoll_poll_unlock(have);
5245 static __latent_entropy void net_rx_action(struct softirq_action *h)
5247 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5248 unsigned long time_limit = jiffies + 2;
5249 int budget = netdev_budget;
5253 local_irq_disable();
5254 list_splice_init(&sd->poll_list, &list);
5258 struct napi_struct *n;
5260 if (list_empty(&list)) {
5261 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
5266 n = list_first_entry(&list, struct napi_struct, poll_list);
5267 budget -= napi_poll(n, &repoll);
5269 /* If softirq window is exhausted then punt.
5270 * Allow this to run for 2 jiffies since which will allow
5271 * an average latency of 1.5/HZ.
5273 if (unlikely(budget <= 0 ||
5274 time_after_eq(jiffies, time_limit))) {
5280 __kfree_skb_flush();
5281 local_irq_disable();
5283 list_splice_tail_init(&sd->poll_list, &list);
5284 list_splice_tail(&repoll, &list);
5285 list_splice(&list, &sd->poll_list);
5286 if (!list_empty(&sd->poll_list))
5287 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5289 net_rps_action_and_irq_enable(sd);
5292 struct netdev_adjacent {
5293 struct net_device *dev;
5295 /* upper master flag, there can only be one master device per list */
5298 /* counter for the number of times this device was added to us */
5301 /* private field for the users */
5304 struct list_head list;
5305 struct rcu_head rcu;
5308 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
5309 struct list_head *adj_list)
5311 struct netdev_adjacent *adj;
5313 list_for_each_entry(adj, adj_list, list) {
5314 if (adj->dev == adj_dev)
5320 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
5322 struct net_device *dev = data;
5324 return upper_dev == dev;
5328 * netdev_has_upper_dev - Check if device is linked to an upper device
5330 * @upper_dev: upper device to check
5332 * Find out if a device is linked to specified upper device and return true
5333 * in case it is. Note that this checks only immediate upper device,
5334 * not through a complete stack of devices. The caller must hold the RTNL lock.
5336 bool netdev_has_upper_dev(struct net_device *dev,
5337 struct net_device *upper_dev)
5341 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5344 EXPORT_SYMBOL(netdev_has_upper_dev);
5347 * netdev_has_upper_dev_all - Check if device is linked to an upper device
5349 * @upper_dev: upper device to check
5351 * Find out if a device is linked to specified upper device and return true
5352 * in case it is. Note that this checks the entire upper device chain.
5353 * The caller must hold rcu lock.
5356 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
5357 struct net_device *upper_dev)
5359 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5362 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
5365 * netdev_has_any_upper_dev - Check if device is linked to some device
5368 * Find out if a device is linked to an upper device and return true in case
5369 * it is. The caller must hold the RTNL lock.
5371 static bool netdev_has_any_upper_dev(struct net_device *dev)
5375 return !list_empty(&dev->adj_list.upper);
5379 * netdev_master_upper_dev_get - Get master upper device
5382 * Find a master upper device and return pointer to it or NULL in case
5383 * it's not there. The caller must hold the RTNL lock.
5385 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5387 struct netdev_adjacent *upper;
5391 if (list_empty(&dev->adj_list.upper))
5394 upper = list_first_entry(&dev->adj_list.upper,
5395 struct netdev_adjacent, list);
5396 if (likely(upper->master))
5400 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5403 * netdev_has_any_lower_dev - Check if device is linked to some device
5406 * Find out if a device is linked to a lower device and return true in case
5407 * it is. The caller must hold the RTNL lock.
5409 static bool netdev_has_any_lower_dev(struct net_device *dev)
5413 return !list_empty(&dev->adj_list.lower);
5416 void *netdev_adjacent_get_private(struct list_head *adj_list)
5418 struct netdev_adjacent *adj;
5420 adj = list_entry(adj_list, struct netdev_adjacent, list);
5422 return adj->private;
5424 EXPORT_SYMBOL(netdev_adjacent_get_private);
5427 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5429 * @iter: list_head ** of the current position
5431 * Gets the next device from the dev's upper list, starting from iter
5432 * position. The caller must hold RCU read lock.
5434 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5435 struct list_head **iter)
5437 struct netdev_adjacent *upper;
5439 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5441 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5443 if (&upper->list == &dev->adj_list.upper)
5446 *iter = &upper->list;
5450 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5452 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
5453 struct list_head **iter)
5455 struct netdev_adjacent *upper;
5457 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5459 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5461 if (&upper->list == &dev->adj_list.upper)
5464 *iter = &upper->list;
5469 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
5470 int (*fn)(struct net_device *dev,
5474 struct net_device *udev;
5475 struct list_head *iter;
5478 for (iter = &dev->adj_list.upper,
5479 udev = netdev_next_upper_dev_rcu(dev, &iter);
5481 udev = netdev_next_upper_dev_rcu(dev, &iter)) {
5482 /* first is the upper device itself */
5483 ret = fn(udev, data);
5487 /* then look at all of its upper devices */
5488 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data);
5495 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
5498 * netdev_lower_get_next_private - Get the next ->private from the
5499 * lower neighbour list
5501 * @iter: list_head ** of the current position
5503 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5504 * list, starting from iter position. The caller must hold either hold the
5505 * RTNL lock or its own locking that guarantees that the neighbour lower
5506 * list will remain unchanged.
5508 void *netdev_lower_get_next_private(struct net_device *dev,
5509 struct list_head **iter)
5511 struct netdev_adjacent *lower;
5513 lower = list_entry(*iter, struct netdev_adjacent, list);
5515 if (&lower->list == &dev->adj_list.lower)
5518 *iter = lower->list.next;
5520 return lower->private;
5522 EXPORT_SYMBOL(netdev_lower_get_next_private);
5525 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5526 * lower neighbour list, RCU
5529 * @iter: list_head ** of the current position
5531 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5532 * list, starting from iter position. The caller must hold RCU read lock.
5534 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5535 struct list_head **iter)
5537 struct netdev_adjacent *lower;
5539 WARN_ON_ONCE(!rcu_read_lock_held());
5541 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5543 if (&lower->list == &dev->adj_list.lower)
5546 *iter = &lower->list;
5548 return lower->private;
5550 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5553 * netdev_lower_get_next - Get the next device from the lower neighbour
5556 * @iter: list_head ** of the current position
5558 * Gets the next netdev_adjacent from the dev's lower neighbour
5559 * list, starting from iter position. The caller must hold RTNL lock or
5560 * its own locking that guarantees that the neighbour lower
5561 * list will remain unchanged.
5563 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5565 struct netdev_adjacent *lower;
5567 lower = list_entry(*iter, struct netdev_adjacent, list);
5569 if (&lower->list == &dev->adj_list.lower)
5572 *iter = lower->list.next;
5576 EXPORT_SYMBOL(netdev_lower_get_next);
5578 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
5579 struct list_head **iter)
5581 struct netdev_adjacent *lower;
5583 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5585 if (&lower->list == &dev->adj_list.lower)
5588 *iter = &lower->list;
5593 int netdev_walk_all_lower_dev(struct net_device *dev,
5594 int (*fn)(struct net_device *dev,
5598 struct net_device *ldev;
5599 struct list_head *iter;
5602 for (iter = &dev->adj_list.lower,
5603 ldev = netdev_next_lower_dev(dev, &iter);
5605 ldev = netdev_next_lower_dev(dev, &iter)) {
5606 /* first is the lower device itself */
5607 ret = fn(ldev, data);
5611 /* then look at all of its lower devices */
5612 ret = netdev_walk_all_lower_dev(ldev, fn, data);
5619 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
5621 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
5622 struct list_head **iter)
5624 struct netdev_adjacent *lower;
5626 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5627 if (&lower->list == &dev->adj_list.lower)
5630 *iter = &lower->list;
5635 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
5636 int (*fn)(struct net_device *dev,
5640 struct net_device *ldev;
5641 struct list_head *iter;
5644 for (iter = &dev->adj_list.lower,
5645 ldev = netdev_next_lower_dev_rcu(dev, &iter);
5647 ldev = netdev_next_lower_dev_rcu(dev, &iter)) {
5648 /* first is the lower device itself */
5649 ret = fn(ldev, data);
5653 /* then look at all of its lower devices */
5654 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data);
5661 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
5664 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5665 * lower neighbour list, RCU
5669 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5670 * list. The caller must hold RCU read lock.
5672 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5674 struct netdev_adjacent *lower;
5676 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5677 struct netdev_adjacent, list);
5679 return lower->private;
5682 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5685 * netdev_master_upper_dev_get_rcu - Get master upper device
5688 * Find a master upper device and return pointer to it or NULL in case
5689 * it's not there. The caller must hold the RCU read lock.
5691 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5693 struct netdev_adjacent *upper;
5695 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5696 struct netdev_adjacent, list);
5697 if (upper && likely(upper->master))
5701 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5703 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5704 struct net_device *adj_dev,
5705 struct list_head *dev_list)
5707 char linkname[IFNAMSIZ+7];
5708 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5709 "upper_%s" : "lower_%s", adj_dev->name);
5710 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5713 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5715 struct list_head *dev_list)
5717 char linkname[IFNAMSIZ+7];
5718 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5719 "upper_%s" : "lower_%s", name);
5720 sysfs_remove_link(&(dev->dev.kobj), linkname);
5723 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5724 struct net_device *adj_dev,
5725 struct list_head *dev_list)
5727 return (dev_list == &dev->adj_list.upper ||
5728 dev_list == &dev->adj_list.lower) &&
5729 net_eq(dev_net(dev), dev_net(adj_dev));
5732 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5733 struct net_device *adj_dev,
5734 struct list_head *dev_list,
5735 void *private, bool master)
5737 struct netdev_adjacent *adj;
5740 adj = __netdev_find_adj(adj_dev, dev_list);
5744 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
5745 dev->name, adj_dev->name, adj->ref_nr);
5750 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5755 adj->master = master;
5757 adj->private = private;
5760 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
5761 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
5763 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5764 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5769 /* Ensure that master link is always the first item in list. */
5771 ret = sysfs_create_link(&(dev->dev.kobj),
5772 &(adj_dev->dev.kobj), "master");
5774 goto remove_symlinks;
5776 list_add_rcu(&adj->list, dev_list);
5778 list_add_tail_rcu(&adj->list, dev_list);
5784 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5785 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5793 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5794 struct net_device *adj_dev,
5796 struct list_head *dev_list)
5798 struct netdev_adjacent *adj;
5800 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
5801 dev->name, adj_dev->name, ref_nr);
5803 adj = __netdev_find_adj(adj_dev, dev_list);
5806 pr_err("Adjacency does not exist for device %s from %s\n",
5807 dev->name, adj_dev->name);
5812 if (adj->ref_nr > ref_nr) {
5813 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
5814 dev->name, adj_dev->name, ref_nr,
5815 adj->ref_nr - ref_nr);
5816 adj->ref_nr -= ref_nr;
5821 sysfs_remove_link(&(dev->dev.kobj), "master");
5823 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5824 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5826 list_del_rcu(&adj->list);
5827 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
5828 adj_dev->name, dev->name, adj_dev->name);
5830 kfree_rcu(adj, rcu);
5833 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5834 struct net_device *upper_dev,
5835 struct list_head *up_list,
5836 struct list_head *down_list,
5837 void *private, bool master)
5841 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
5846 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
5849 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
5856 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5857 struct net_device *upper_dev,
5859 struct list_head *up_list,
5860 struct list_head *down_list)
5862 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
5863 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
5866 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5867 struct net_device *upper_dev,
5868 void *private, bool master)
5870 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5871 &dev->adj_list.upper,
5872 &upper_dev->adj_list.lower,
5876 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5877 struct net_device *upper_dev)
5879 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
5880 &dev->adj_list.upper,
5881 &upper_dev->adj_list.lower);
5884 static int __netdev_upper_dev_link(struct net_device *dev,
5885 struct net_device *upper_dev, bool master,
5886 void *upper_priv, void *upper_info)
5888 struct netdev_notifier_changeupper_info changeupper_info;
5893 if (dev == upper_dev)
5896 /* To prevent loops, check if dev is not upper device to upper_dev. */
5897 if (netdev_has_upper_dev(upper_dev, dev))
5900 if (netdev_has_upper_dev(dev, upper_dev))
5903 if (master && netdev_master_upper_dev_get(dev))
5906 changeupper_info.upper_dev = upper_dev;
5907 changeupper_info.master = master;
5908 changeupper_info.linking = true;
5909 changeupper_info.upper_info = upper_info;
5911 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5912 &changeupper_info.info);
5913 ret = notifier_to_errno(ret);
5917 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
5922 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5923 &changeupper_info.info);
5924 ret = notifier_to_errno(ret);
5931 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5937 * netdev_upper_dev_link - Add a link to the upper device
5939 * @upper_dev: new upper device
5941 * Adds a link to device which is upper to this one. The caller must hold
5942 * the RTNL lock. On a failure a negative errno code is returned.
5943 * On success the reference counts are adjusted and the function
5946 int netdev_upper_dev_link(struct net_device *dev,
5947 struct net_device *upper_dev)
5949 return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL);
5951 EXPORT_SYMBOL(netdev_upper_dev_link);
5954 * netdev_master_upper_dev_link - Add a master link to the upper device
5956 * @upper_dev: new upper device
5957 * @upper_priv: upper device private
5958 * @upper_info: upper info to be passed down via notifier
5960 * Adds a link to device which is upper to this one. In this case, only
5961 * one master upper device can be linked, although other non-master devices
5962 * might be linked as well. The caller must hold the RTNL lock.
5963 * On a failure a negative errno code is returned. On success the reference
5964 * counts are adjusted and the function returns zero.
5966 int netdev_master_upper_dev_link(struct net_device *dev,
5967 struct net_device *upper_dev,
5968 void *upper_priv, void *upper_info)
5970 return __netdev_upper_dev_link(dev, upper_dev, true,
5971 upper_priv, upper_info);
5973 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5976 * netdev_upper_dev_unlink - Removes a link to upper device
5978 * @upper_dev: new upper device
5980 * Removes a link to device which is upper to this one. The caller must hold
5983 void netdev_upper_dev_unlink(struct net_device *dev,
5984 struct net_device *upper_dev)
5986 struct netdev_notifier_changeupper_info changeupper_info;
5989 changeupper_info.upper_dev = upper_dev;
5990 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
5991 changeupper_info.linking = false;
5993 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5994 &changeupper_info.info);
5996 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5998 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5999 &changeupper_info.info);
6001 EXPORT_SYMBOL(netdev_upper_dev_unlink);
6004 * netdev_bonding_info_change - Dispatch event about slave change
6006 * @bonding_info: info to dispatch
6008 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
6009 * The caller must hold the RTNL lock.
6011 void netdev_bonding_info_change(struct net_device *dev,
6012 struct netdev_bonding_info *bonding_info)
6014 struct netdev_notifier_bonding_info info;
6016 memcpy(&info.bonding_info, bonding_info,
6017 sizeof(struct netdev_bonding_info));
6018 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
6021 EXPORT_SYMBOL(netdev_bonding_info_change);
6023 static void netdev_adjacent_add_links(struct net_device *dev)
6025 struct netdev_adjacent *iter;
6027 struct net *net = dev_net(dev);
6029 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6030 if (!net_eq(net, dev_net(iter->dev)))
6032 netdev_adjacent_sysfs_add(iter->dev, dev,
6033 &iter->dev->adj_list.lower);
6034 netdev_adjacent_sysfs_add(dev, iter->dev,
6035 &dev->adj_list.upper);
6038 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6039 if (!net_eq(net, dev_net(iter->dev)))
6041 netdev_adjacent_sysfs_add(iter->dev, dev,
6042 &iter->dev->adj_list.upper);
6043 netdev_adjacent_sysfs_add(dev, iter->dev,
6044 &dev->adj_list.lower);
6048 static void netdev_adjacent_del_links(struct net_device *dev)
6050 struct netdev_adjacent *iter;
6052 struct net *net = dev_net(dev);
6054 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6055 if (!net_eq(net, dev_net(iter->dev)))
6057 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6058 &iter->dev->adj_list.lower);
6059 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6060 &dev->adj_list.upper);
6063 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6064 if (!net_eq(net, dev_net(iter->dev)))
6066 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6067 &iter->dev->adj_list.upper);
6068 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6069 &dev->adj_list.lower);
6073 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
6075 struct netdev_adjacent *iter;
6077 struct net *net = dev_net(dev);
6079 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6080 if (!net_eq(net, dev_net(iter->dev)))
6082 netdev_adjacent_sysfs_del(iter->dev, oldname,
6083 &iter->dev->adj_list.lower);
6084 netdev_adjacent_sysfs_add(iter->dev, dev,
6085 &iter->dev->adj_list.lower);
6088 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6089 if (!net_eq(net, dev_net(iter->dev)))
6091 netdev_adjacent_sysfs_del(iter->dev, oldname,
6092 &iter->dev->adj_list.upper);
6093 netdev_adjacent_sysfs_add(iter->dev, dev,
6094 &iter->dev->adj_list.upper);
6098 void *netdev_lower_dev_get_private(struct net_device *dev,
6099 struct net_device *lower_dev)
6101 struct netdev_adjacent *lower;
6105 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
6109 return lower->private;
6111 EXPORT_SYMBOL(netdev_lower_dev_get_private);
6114 int dev_get_nest_level(struct net_device *dev)
6116 struct net_device *lower = NULL;
6117 struct list_head *iter;
6123 netdev_for_each_lower_dev(dev, lower, iter) {
6124 nest = dev_get_nest_level(lower);
6125 if (max_nest < nest)
6129 return max_nest + 1;
6131 EXPORT_SYMBOL(dev_get_nest_level);
6134 * netdev_lower_change - Dispatch event about lower device state change
6135 * @lower_dev: device
6136 * @lower_state_info: state to dispatch
6138 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
6139 * The caller must hold the RTNL lock.
6141 void netdev_lower_state_changed(struct net_device *lower_dev,
6142 void *lower_state_info)
6144 struct netdev_notifier_changelowerstate_info changelowerstate_info;
6147 changelowerstate_info.lower_state_info = lower_state_info;
6148 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev,
6149 &changelowerstate_info.info);
6151 EXPORT_SYMBOL(netdev_lower_state_changed);
6153 int netdev_default_l2upper_neigh_construct(struct net_device *dev,
6154 struct neighbour *n)
6156 struct net_device *lower_dev, *stop_dev;
6157 struct list_head *iter;
6160 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6161 if (!lower_dev->netdev_ops->ndo_neigh_construct)
6163 err = lower_dev->netdev_ops->ndo_neigh_construct(lower_dev, n);
6165 stop_dev = lower_dev;
6172 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6173 if (lower_dev == stop_dev)
6175 if (!lower_dev->netdev_ops->ndo_neigh_destroy)
6177 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n);
6181 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_construct);
6183 void netdev_default_l2upper_neigh_destroy(struct net_device *dev,
6184 struct neighbour *n)
6186 struct net_device *lower_dev;
6187 struct list_head *iter;
6189 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6190 if (!lower_dev->netdev_ops->ndo_neigh_destroy)
6192 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n);
6195 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_destroy);
6197 static void dev_change_rx_flags(struct net_device *dev, int flags)
6199 const struct net_device_ops *ops = dev->netdev_ops;
6201 if (ops->ndo_change_rx_flags)
6202 ops->ndo_change_rx_flags(dev, flags);
6205 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
6207 unsigned int old_flags = dev->flags;
6213 dev->flags |= IFF_PROMISC;
6214 dev->promiscuity += inc;
6215 if (dev->promiscuity == 0) {
6218 * If inc causes overflow, untouch promisc and return error.
6221 dev->flags &= ~IFF_PROMISC;
6223 dev->promiscuity -= inc;
6224 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6229 if (dev->flags != old_flags) {
6230 pr_info("device %s %s promiscuous mode\n",
6232 dev->flags & IFF_PROMISC ? "entered" : "left");
6233 if (audit_enabled) {
6234 current_uid_gid(&uid, &gid);
6235 audit_log(current->audit_context, GFP_ATOMIC,
6236 AUDIT_ANOM_PROMISCUOUS,
6237 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6238 dev->name, (dev->flags & IFF_PROMISC),
6239 (old_flags & IFF_PROMISC),
6240 from_kuid(&init_user_ns, audit_get_loginuid(current)),
6241 from_kuid(&init_user_ns, uid),
6242 from_kgid(&init_user_ns, gid),
6243 audit_get_sessionid(current));
6246 dev_change_rx_flags(dev, IFF_PROMISC);
6249 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
6254 * dev_set_promiscuity - update promiscuity count on a device
6258 * Add or remove promiscuity from a device. While the count in the device
6259 * remains above zero the interface remains promiscuous. Once it hits zero
6260 * the device reverts back to normal filtering operation. A negative inc
6261 * value is used to drop promiscuity on the device.
6262 * Return 0 if successful or a negative errno code on error.
6264 int dev_set_promiscuity(struct net_device *dev, int inc)
6266 unsigned int old_flags = dev->flags;
6269 err = __dev_set_promiscuity(dev, inc, true);
6272 if (dev->flags != old_flags)
6273 dev_set_rx_mode(dev);
6276 EXPORT_SYMBOL(dev_set_promiscuity);
6278 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
6280 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
6284 dev->flags |= IFF_ALLMULTI;
6285 dev->allmulti += inc;
6286 if (dev->allmulti == 0) {
6289 * If inc causes overflow, untouch allmulti and return error.
6292 dev->flags &= ~IFF_ALLMULTI;
6294 dev->allmulti -= inc;
6295 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6300 if (dev->flags ^ old_flags) {
6301 dev_change_rx_flags(dev, IFF_ALLMULTI);
6302 dev_set_rx_mode(dev);
6304 __dev_notify_flags(dev, old_flags,
6305 dev->gflags ^ old_gflags);
6311 * dev_set_allmulti - update allmulti count on a device
6315 * Add or remove reception of all multicast frames to a device. While the
6316 * count in the device remains above zero the interface remains listening
6317 * to all interfaces. Once it hits zero the device reverts back to normal
6318 * filtering operation. A negative @inc value is used to drop the counter
6319 * when releasing a resource needing all multicasts.
6320 * Return 0 if successful or a negative errno code on error.
6323 int dev_set_allmulti(struct net_device *dev, int inc)
6325 return __dev_set_allmulti(dev, inc, true);
6327 EXPORT_SYMBOL(dev_set_allmulti);
6330 * Upload unicast and multicast address lists to device and
6331 * configure RX filtering. When the device doesn't support unicast
6332 * filtering it is put in promiscuous mode while unicast addresses
6335 void __dev_set_rx_mode(struct net_device *dev)
6337 const struct net_device_ops *ops = dev->netdev_ops;
6339 /* dev_open will call this function so the list will stay sane. */
6340 if (!(dev->flags&IFF_UP))
6343 if (!netif_device_present(dev))
6346 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
6347 /* Unicast addresses changes may only happen under the rtnl,
6348 * therefore calling __dev_set_promiscuity here is safe.
6350 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
6351 __dev_set_promiscuity(dev, 1, false);
6352 dev->uc_promisc = true;
6353 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
6354 __dev_set_promiscuity(dev, -1, false);
6355 dev->uc_promisc = false;
6359 if (ops->ndo_set_rx_mode)
6360 ops->ndo_set_rx_mode(dev);
6363 void dev_set_rx_mode(struct net_device *dev)
6365 netif_addr_lock_bh(dev);
6366 __dev_set_rx_mode(dev);
6367 netif_addr_unlock_bh(dev);
6371 * dev_get_flags - get flags reported to userspace
6374 * Get the combination of flag bits exported through APIs to userspace.
6376 unsigned int dev_get_flags(const struct net_device *dev)
6380 flags = (dev->flags & ~(IFF_PROMISC |
6385 (dev->gflags & (IFF_PROMISC |
6388 if (netif_running(dev)) {
6389 if (netif_oper_up(dev))
6390 flags |= IFF_RUNNING;
6391 if (netif_carrier_ok(dev))
6392 flags |= IFF_LOWER_UP;
6393 if (netif_dormant(dev))
6394 flags |= IFF_DORMANT;
6399 EXPORT_SYMBOL(dev_get_flags);
6401 int __dev_change_flags(struct net_device *dev, unsigned int flags)
6403 unsigned int old_flags = dev->flags;
6409 * Set the flags on our device.
6412 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
6413 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
6415 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6419 * Load in the correct multicast list now the flags have changed.
6422 if ((old_flags ^ flags) & IFF_MULTICAST)
6423 dev_change_rx_flags(dev, IFF_MULTICAST);
6425 dev_set_rx_mode(dev);
6428 * Have we downed the interface. We handle IFF_UP ourselves
6429 * according to user attempts to set it, rather than blindly
6434 if ((old_flags ^ flags) & IFF_UP)
6435 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
6437 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6438 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6439 unsigned int old_flags = dev->flags;
6441 dev->gflags ^= IFF_PROMISC;
6443 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6444 if (dev->flags != old_flags)
6445 dev_set_rx_mode(dev);
6448 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6449 is important. Some (broken) drivers set IFF_PROMISC, when
6450 IFF_ALLMULTI is requested not asking us and not reporting.
6452 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6453 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6455 dev->gflags ^= IFF_ALLMULTI;
6456 __dev_set_allmulti(dev, inc, false);
6462 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6463 unsigned int gchanges)
6465 unsigned int changes = dev->flags ^ old_flags;
6468 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6470 if (changes & IFF_UP) {
6471 if (dev->flags & IFF_UP)
6472 call_netdevice_notifiers(NETDEV_UP, dev);
6474 call_netdevice_notifiers(NETDEV_DOWN, dev);
6477 if (dev->flags & IFF_UP &&
6478 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6479 struct netdev_notifier_change_info change_info;
6481 change_info.flags_changed = changes;
6482 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6488 * dev_change_flags - change device settings
6490 * @flags: device state flags
6492 * Change settings on device based state flags. The flags are
6493 * in the userspace exported format.
6495 int dev_change_flags(struct net_device *dev, unsigned int flags)
6498 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6500 ret = __dev_change_flags(dev, flags);
6504 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6505 __dev_notify_flags(dev, old_flags, changes);
6508 EXPORT_SYMBOL(dev_change_flags);
6510 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6512 const struct net_device_ops *ops = dev->netdev_ops;
6514 if (ops->ndo_change_mtu)
6515 return ops->ndo_change_mtu(dev, new_mtu);
6522 * dev_set_mtu - Change maximum transfer unit
6524 * @new_mtu: new transfer unit
6526 * Change the maximum transfer size of the network device.
6528 int dev_set_mtu(struct net_device *dev, int new_mtu)
6532 if (new_mtu == dev->mtu)
6535 /* MTU must be positive, and in range */
6536 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
6537 net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n",
6538 dev->name, new_mtu, dev->min_mtu);
6542 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
6543 net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n",
6544 dev->name, new_mtu, dev->max_mtu);
6548 if (!netif_device_present(dev))
6551 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6552 err = notifier_to_errno(err);
6556 orig_mtu = dev->mtu;
6557 err = __dev_set_mtu(dev, new_mtu);
6560 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6561 err = notifier_to_errno(err);
6563 /* setting mtu back and notifying everyone again,
6564 * so that they have a chance to revert changes.
6566 __dev_set_mtu(dev, orig_mtu);
6567 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6572 EXPORT_SYMBOL(dev_set_mtu);
6575 * dev_set_group - Change group this device belongs to
6577 * @new_group: group this device should belong to
6579 void dev_set_group(struct net_device *dev, int new_group)
6581 dev->group = new_group;
6583 EXPORT_SYMBOL(dev_set_group);
6586 * dev_set_mac_address - Change Media Access Control Address
6590 * Change the hardware (MAC) address of the device
6592 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6594 const struct net_device_ops *ops = dev->netdev_ops;
6597 if (!ops->ndo_set_mac_address)
6599 if (sa->sa_family != dev->type)
6601 if (!netif_device_present(dev))
6603 err = ops->ndo_set_mac_address(dev, sa);
6606 dev->addr_assign_type = NET_ADDR_SET;
6607 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6608 add_device_randomness(dev->dev_addr, dev->addr_len);
6611 EXPORT_SYMBOL(dev_set_mac_address);
6614 * dev_change_carrier - Change device carrier
6616 * @new_carrier: new value
6618 * Change device carrier
6620 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6622 const struct net_device_ops *ops = dev->netdev_ops;
6624 if (!ops->ndo_change_carrier)
6626 if (!netif_device_present(dev))
6628 return ops->ndo_change_carrier(dev, new_carrier);
6630 EXPORT_SYMBOL(dev_change_carrier);
6633 * dev_get_phys_port_id - Get device physical port ID
6637 * Get device physical port ID
6639 int dev_get_phys_port_id(struct net_device *dev,
6640 struct netdev_phys_item_id *ppid)
6642 const struct net_device_ops *ops = dev->netdev_ops;
6644 if (!ops->ndo_get_phys_port_id)
6646 return ops->ndo_get_phys_port_id(dev, ppid);
6648 EXPORT_SYMBOL(dev_get_phys_port_id);
6651 * dev_get_phys_port_name - Get device physical port name
6654 * @len: limit of bytes to copy to name
6656 * Get device physical port name
6658 int dev_get_phys_port_name(struct net_device *dev,
6659 char *name, size_t len)
6661 const struct net_device_ops *ops = dev->netdev_ops;
6663 if (!ops->ndo_get_phys_port_name)
6665 return ops->ndo_get_phys_port_name(dev, name, len);
6667 EXPORT_SYMBOL(dev_get_phys_port_name);
6670 * dev_change_proto_down - update protocol port state information
6672 * @proto_down: new value
6674 * This info can be used by switch drivers to set the phys state of the
6677 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6679 const struct net_device_ops *ops = dev->netdev_ops;
6681 if (!ops->ndo_change_proto_down)
6683 if (!netif_device_present(dev))
6685 return ops->ndo_change_proto_down(dev, proto_down);
6687 EXPORT_SYMBOL(dev_change_proto_down);
6690 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
6692 * @fd: new program fd or negative value to clear
6694 * Set or clear a bpf program for a device
6696 int dev_change_xdp_fd(struct net_device *dev, int fd)
6698 const struct net_device_ops *ops = dev->netdev_ops;
6699 struct bpf_prog *prog = NULL;
6700 struct netdev_xdp xdp = {};
6706 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP);
6708 return PTR_ERR(prog);
6711 xdp.command = XDP_SETUP_PROG;
6713 err = ops->ndo_xdp(dev, &xdp);
6714 if (err < 0 && prog)
6719 EXPORT_SYMBOL(dev_change_xdp_fd);
6722 * dev_new_index - allocate an ifindex
6723 * @net: the applicable net namespace
6725 * Returns a suitable unique value for a new device interface
6726 * number. The caller must hold the rtnl semaphore or the
6727 * dev_base_lock to be sure it remains unique.
6729 static int dev_new_index(struct net *net)
6731 int ifindex = net->ifindex;
6735 if (!__dev_get_by_index(net, ifindex))
6736 return net->ifindex = ifindex;
6740 /* Delayed registration/unregisteration */
6741 static LIST_HEAD(net_todo_list);
6742 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6744 static void net_set_todo(struct net_device *dev)
6746 list_add_tail(&dev->todo_list, &net_todo_list);
6747 dev_net(dev)->dev_unreg_count++;
6750 static void rollback_registered_many(struct list_head *head)
6752 struct net_device *dev, *tmp;
6753 LIST_HEAD(close_head);
6755 BUG_ON(dev_boot_phase);
6758 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6759 /* Some devices call without registering
6760 * for initialization unwind. Remove those
6761 * devices and proceed with the remaining.
6763 if (dev->reg_state == NETREG_UNINITIALIZED) {
6764 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6768 list_del(&dev->unreg_list);
6771 dev->dismantle = true;
6772 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6775 /* If device is running, close it first. */
6776 list_for_each_entry(dev, head, unreg_list)
6777 list_add_tail(&dev->close_list, &close_head);
6778 dev_close_many(&close_head, true);
6780 list_for_each_entry(dev, head, unreg_list) {
6781 /* And unlink it from device chain. */
6782 unlist_netdevice(dev);
6784 dev->reg_state = NETREG_UNREGISTERING;
6786 flush_all_backlogs();
6790 list_for_each_entry(dev, head, unreg_list) {
6791 struct sk_buff *skb = NULL;
6793 /* Shutdown queueing discipline. */
6797 /* Notify protocols, that we are about to destroy
6798 this device. They should clean all the things.
6800 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6802 if (!dev->rtnl_link_ops ||
6803 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6804 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6808 * Flush the unicast and multicast chains
6813 if (dev->netdev_ops->ndo_uninit)
6814 dev->netdev_ops->ndo_uninit(dev);
6817 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6819 /* Notifier chain MUST detach us all upper devices. */
6820 WARN_ON(netdev_has_any_upper_dev(dev));
6821 WARN_ON(netdev_has_any_lower_dev(dev));
6823 /* Remove entries from kobject tree */
6824 netdev_unregister_kobject(dev);
6826 /* Remove XPS queueing entries */
6827 netif_reset_xps_queues_gt(dev, 0);
6833 list_for_each_entry(dev, head, unreg_list)
6837 static void rollback_registered(struct net_device *dev)
6841 list_add(&dev->unreg_list, &single);
6842 rollback_registered_many(&single);
6846 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6847 struct net_device *upper, netdev_features_t features)
6849 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6850 netdev_features_t feature;
6853 for_each_netdev_feature(&upper_disables, feature_bit) {
6854 feature = __NETIF_F_BIT(feature_bit);
6855 if (!(upper->wanted_features & feature)
6856 && (features & feature)) {
6857 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6858 &feature, upper->name);
6859 features &= ~feature;
6866 static void netdev_sync_lower_features(struct net_device *upper,
6867 struct net_device *lower, netdev_features_t features)
6869 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6870 netdev_features_t feature;
6873 for_each_netdev_feature(&upper_disables, feature_bit) {
6874 feature = __NETIF_F_BIT(feature_bit);
6875 if (!(features & feature) && (lower->features & feature)) {
6876 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6877 &feature, lower->name);
6878 lower->wanted_features &= ~feature;
6879 netdev_update_features(lower);
6881 if (unlikely(lower->features & feature))
6882 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6883 &feature, lower->name);
6888 static netdev_features_t netdev_fix_features(struct net_device *dev,
6889 netdev_features_t features)
6891 /* Fix illegal checksum combinations */
6892 if ((features & NETIF_F_HW_CSUM) &&
6893 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6894 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6895 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6898 /* TSO requires that SG is present as well. */
6899 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6900 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6901 features &= ~NETIF_F_ALL_TSO;
6904 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6905 !(features & NETIF_F_IP_CSUM)) {
6906 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6907 features &= ~NETIF_F_TSO;
6908 features &= ~NETIF_F_TSO_ECN;
6911 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6912 !(features & NETIF_F_IPV6_CSUM)) {
6913 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6914 features &= ~NETIF_F_TSO6;
6917 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
6918 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
6919 features &= ~NETIF_F_TSO_MANGLEID;
6921 /* TSO ECN requires that TSO is present as well. */
6922 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6923 features &= ~NETIF_F_TSO_ECN;
6925 /* Software GSO depends on SG. */
6926 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6927 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6928 features &= ~NETIF_F_GSO;
6931 /* UFO needs SG and checksumming */
6932 if (features & NETIF_F_UFO) {
6933 /* maybe split UFO into V4 and V6? */
6934 if (!(features & NETIF_F_HW_CSUM) &&
6935 ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) !=
6936 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) {
6938 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6939 features &= ~NETIF_F_UFO;
6942 if (!(features & NETIF_F_SG)) {
6944 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6945 features &= ~NETIF_F_UFO;
6949 /* GSO partial features require GSO partial be set */
6950 if ((features & dev->gso_partial_features) &&
6951 !(features & NETIF_F_GSO_PARTIAL)) {
6953 "Dropping partially supported GSO features since no GSO partial.\n");
6954 features &= ~dev->gso_partial_features;
6957 #ifdef CONFIG_NET_RX_BUSY_POLL
6958 if (dev->netdev_ops->ndo_busy_poll)
6959 features |= NETIF_F_BUSY_POLL;
6962 features &= ~NETIF_F_BUSY_POLL;
6967 int __netdev_update_features(struct net_device *dev)
6969 struct net_device *upper, *lower;
6970 netdev_features_t features;
6971 struct list_head *iter;
6976 features = netdev_get_wanted_features(dev);
6978 if (dev->netdev_ops->ndo_fix_features)
6979 features = dev->netdev_ops->ndo_fix_features(dev, features);
6981 /* driver might be less strict about feature dependencies */
6982 features = netdev_fix_features(dev, features);
6984 /* some features can't be enabled if they're off an an upper device */
6985 netdev_for_each_upper_dev_rcu(dev, upper, iter)
6986 features = netdev_sync_upper_features(dev, upper, features);
6988 if (dev->features == features)
6991 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6992 &dev->features, &features);
6994 if (dev->netdev_ops->ndo_set_features)
6995 err = dev->netdev_ops->ndo_set_features(dev, features);
6999 if (unlikely(err < 0)) {
7001 "set_features() failed (%d); wanted %pNF, left %pNF\n",
7002 err, &features, &dev->features);
7003 /* return non-0 since some features might have changed and
7004 * it's better to fire a spurious notification than miss it
7010 /* some features must be disabled on lower devices when disabled
7011 * on an upper device (think: bonding master or bridge)
7013 netdev_for_each_lower_dev(dev, lower, iter)
7014 netdev_sync_lower_features(dev, lower, features);
7017 dev->features = features;
7019 return err < 0 ? 0 : 1;
7023 * netdev_update_features - recalculate device features
7024 * @dev: the device to check
7026 * Recalculate dev->features set and send notifications if it
7027 * has changed. Should be called after driver or hardware dependent
7028 * conditions might have changed that influence the features.
7030 void netdev_update_features(struct net_device *dev)
7032 if (__netdev_update_features(dev))
7033 netdev_features_change(dev);
7035 EXPORT_SYMBOL(netdev_update_features);
7038 * netdev_change_features - recalculate device features
7039 * @dev: the device to check
7041 * Recalculate dev->features set and send notifications even
7042 * if they have not changed. Should be called instead of
7043 * netdev_update_features() if also dev->vlan_features might
7044 * have changed to allow the changes to be propagated to stacked
7047 void netdev_change_features(struct net_device *dev)
7049 __netdev_update_features(dev);
7050 netdev_features_change(dev);
7052 EXPORT_SYMBOL(netdev_change_features);
7055 * netif_stacked_transfer_operstate - transfer operstate
7056 * @rootdev: the root or lower level device to transfer state from
7057 * @dev: the device to transfer operstate to
7059 * Transfer operational state from root to device. This is normally
7060 * called when a stacking relationship exists between the root
7061 * device and the device(a leaf device).
7063 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
7064 struct net_device *dev)
7066 if (rootdev->operstate == IF_OPER_DORMANT)
7067 netif_dormant_on(dev);
7069 netif_dormant_off(dev);
7071 if (netif_carrier_ok(rootdev)) {
7072 if (!netif_carrier_ok(dev))
7073 netif_carrier_on(dev);
7075 if (netif_carrier_ok(dev))
7076 netif_carrier_off(dev);
7079 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
7082 static int netif_alloc_rx_queues(struct net_device *dev)
7084 unsigned int i, count = dev->num_rx_queues;
7085 struct netdev_rx_queue *rx;
7086 size_t sz = count * sizeof(*rx);
7090 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7098 for (i = 0; i < count; i++)
7104 static void netdev_init_one_queue(struct net_device *dev,
7105 struct netdev_queue *queue, void *_unused)
7107 /* Initialize queue lock */
7108 spin_lock_init(&queue->_xmit_lock);
7109 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
7110 queue->xmit_lock_owner = -1;
7111 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
7114 dql_init(&queue->dql, HZ);
7118 static void netif_free_tx_queues(struct net_device *dev)
7123 static int netif_alloc_netdev_queues(struct net_device *dev)
7125 unsigned int count = dev->num_tx_queues;
7126 struct netdev_queue *tx;
7127 size_t sz = count * sizeof(*tx);
7129 if (count < 1 || count > 0xffff)
7132 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7140 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
7141 spin_lock_init(&dev->tx_global_lock);
7146 void netif_tx_stop_all_queues(struct net_device *dev)
7150 for (i = 0; i < dev->num_tx_queues; i++) {
7151 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
7152 netif_tx_stop_queue(txq);
7155 EXPORT_SYMBOL(netif_tx_stop_all_queues);
7158 * register_netdevice - register a network device
7159 * @dev: device to register
7161 * Take a completed network device structure and add it to the kernel
7162 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7163 * chain. 0 is returned on success. A negative errno code is returned
7164 * on a failure to set up the device, or if the name is a duplicate.
7166 * Callers must hold the rtnl semaphore. You may want
7167 * register_netdev() instead of this.
7170 * The locking appears insufficient to guarantee two parallel registers
7171 * will not get the same name.
7174 int register_netdevice(struct net_device *dev)
7177 struct net *net = dev_net(dev);
7179 BUG_ON(dev_boot_phase);
7184 /* When net_device's are persistent, this will be fatal. */
7185 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
7188 spin_lock_init(&dev->addr_list_lock);
7189 netdev_set_addr_lockdep_class(dev);
7191 ret = dev_get_valid_name(net, dev, dev->name);
7195 /* Init, if this function is available */
7196 if (dev->netdev_ops->ndo_init) {
7197 ret = dev->netdev_ops->ndo_init(dev);
7205 if (((dev->hw_features | dev->features) &
7206 NETIF_F_HW_VLAN_CTAG_FILTER) &&
7207 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
7208 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
7209 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
7216 dev->ifindex = dev_new_index(net);
7217 else if (__dev_get_by_index(net, dev->ifindex))
7220 /* Transfer changeable features to wanted_features and enable
7221 * software offloads (GSO and GRO).
7223 dev->hw_features |= NETIF_F_SOFT_FEATURES;
7224 dev->features |= NETIF_F_SOFT_FEATURES;
7225 dev->wanted_features = dev->features & dev->hw_features;
7227 if (!(dev->flags & IFF_LOOPBACK))
7228 dev->hw_features |= NETIF_F_NOCACHE_COPY;
7230 /* If IPv4 TCP segmentation offload is supported we should also
7231 * allow the device to enable segmenting the frame with the option
7232 * of ignoring a static IP ID value. This doesn't enable the
7233 * feature itself but allows the user to enable it later.
7235 if (dev->hw_features & NETIF_F_TSO)
7236 dev->hw_features |= NETIF_F_TSO_MANGLEID;
7237 if (dev->vlan_features & NETIF_F_TSO)
7238 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
7239 if (dev->mpls_features & NETIF_F_TSO)
7240 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
7241 if (dev->hw_enc_features & NETIF_F_TSO)
7242 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
7244 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
7246 dev->vlan_features |= NETIF_F_HIGHDMA;
7248 /* Make NETIF_F_SG inheritable to tunnel devices.
7250 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
7252 /* Make NETIF_F_SG inheritable to MPLS.
7254 dev->mpls_features |= NETIF_F_SG;
7256 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
7257 ret = notifier_to_errno(ret);
7261 ret = netdev_register_kobject(dev);
7264 dev->reg_state = NETREG_REGISTERED;
7266 __netdev_update_features(dev);
7269 * Default initial state at registry is that the
7270 * device is present.
7273 set_bit(__LINK_STATE_PRESENT, &dev->state);
7275 linkwatch_init_dev(dev);
7277 dev_init_scheduler(dev);
7279 list_netdevice(dev);
7280 add_device_randomness(dev->dev_addr, dev->addr_len);
7282 /* If the device has permanent device address, driver should
7283 * set dev_addr and also addr_assign_type should be set to
7284 * NET_ADDR_PERM (default value).
7286 if (dev->addr_assign_type == NET_ADDR_PERM)
7287 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
7289 /* Notify protocols, that a new device appeared. */
7290 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
7291 ret = notifier_to_errno(ret);
7293 rollback_registered(dev);
7294 dev->reg_state = NETREG_UNREGISTERED;
7297 * Prevent userspace races by waiting until the network
7298 * device is fully setup before sending notifications.
7300 if (!dev->rtnl_link_ops ||
7301 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7302 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7308 if (dev->netdev_ops->ndo_uninit)
7309 dev->netdev_ops->ndo_uninit(dev);
7312 EXPORT_SYMBOL(register_netdevice);
7315 * init_dummy_netdev - init a dummy network device for NAPI
7316 * @dev: device to init
7318 * This takes a network device structure and initialize the minimum
7319 * amount of fields so it can be used to schedule NAPI polls without
7320 * registering a full blown interface. This is to be used by drivers
7321 * that need to tie several hardware interfaces to a single NAPI
7322 * poll scheduler due to HW limitations.
7324 int init_dummy_netdev(struct net_device *dev)
7326 /* Clear everything. Note we don't initialize spinlocks
7327 * are they aren't supposed to be taken by any of the
7328 * NAPI code and this dummy netdev is supposed to be
7329 * only ever used for NAPI polls
7331 memset(dev, 0, sizeof(struct net_device));
7333 /* make sure we BUG if trying to hit standard
7334 * register/unregister code path
7336 dev->reg_state = NETREG_DUMMY;
7338 /* NAPI wants this */
7339 INIT_LIST_HEAD(&dev->napi_list);
7341 /* a dummy interface is started by default */
7342 set_bit(__LINK_STATE_PRESENT, &dev->state);
7343 set_bit(__LINK_STATE_START, &dev->state);
7345 /* Note : We dont allocate pcpu_refcnt for dummy devices,
7346 * because users of this 'device' dont need to change
7352 EXPORT_SYMBOL_GPL(init_dummy_netdev);
7356 * register_netdev - register a network device
7357 * @dev: device to register
7359 * Take a completed network device structure and add it to the kernel
7360 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7361 * chain. 0 is returned on success. A negative errno code is returned
7362 * on a failure to set up the device, or if the name is a duplicate.
7364 * This is a wrapper around register_netdevice that takes the rtnl semaphore
7365 * and expands the device name if you passed a format string to
7368 int register_netdev(struct net_device *dev)
7373 err = register_netdevice(dev);
7377 EXPORT_SYMBOL(register_netdev);
7379 int netdev_refcnt_read(const struct net_device *dev)
7383 for_each_possible_cpu(i)
7384 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
7387 EXPORT_SYMBOL(netdev_refcnt_read);
7390 * netdev_wait_allrefs - wait until all references are gone.
7391 * @dev: target net_device
7393 * This is called when unregistering network devices.
7395 * Any protocol or device that holds a reference should register
7396 * for netdevice notification, and cleanup and put back the
7397 * reference if they receive an UNREGISTER event.
7398 * We can get stuck here if buggy protocols don't correctly
7401 static void netdev_wait_allrefs(struct net_device *dev)
7403 unsigned long rebroadcast_time, warning_time;
7406 linkwatch_forget_dev(dev);
7408 rebroadcast_time = warning_time = jiffies;
7409 refcnt = netdev_refcnt_read(dev);
7411 while (refcnt != 0) {
7412 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
7415 /* Rebroadcast unregister notification */
7416 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7422 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7423 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
7425 /* We must not have linkwatch events
7426 * pending on unregister. If this
7427 * happens, we simply run the queue
7428 * unscheduled, resulting in a noop
7431 linkwatch_run_queue();
7436 rebroadcast_time = jiffies;
7441 refcnt = netdev_refcnt_read(dev);
7443 if (time_after(jiffies, warning_time + 10 * HZ)) {
7444 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
7446 warning_time = jiffies;
7455 * register_netdevice(x1);
7456 * register_netdevice(x2);
7458 * unregister_netdevice(y1);
7459 * unregister_netdevice(y2);
7465 * We are invoked by rtnl_unlock().
7466 * This allows us to deal with problems:
7467 * 1) We can delete sysfs objects which invoke hotplug
7468 * without deadlocking with linkwatch via keventd.
7469 * 2) Since we run with the RTNL semaphore not held, we can sleep
7470 * safely in order to wait for the netdev refcnt to drop to zero.
7472 * We must not return until all unregister events added during
7473 * the interval the lock was held have been completed.
7475 void netdev_run_todo(void)
7477 struct list_head list;
7479 /* Snapshot list, allow later requests */
7480 list_replace_init(&net_todo_list, &list);
7485 /* Wait for rcu callbacks to finish before next phase */
7486 if (!list_empty(&list))
7489 while (!list_empty(&list)) {
7490 struct net_device *dev
7491 = list_first_entry(&list, struct net_device, todo_list);
7492 list_del(&dev->todo_list);
7495 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7498 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
7499 pr_err("network todo '%s' but state %d\n",
7500 dev->name, dev->reg_state);
7505 dev->reg_state = NETREG_UNREGISTERED;
7507 netdev_wait_allrefs(dev);
7510 BUG_ON(netdev_refcnt_read(dev));
7511 BUG_ON(!list_empty(&dev->ptype_all));
7512 BUG_ON(!list_empty(&dev->ptype_specific));
7513 WARN_ON(rcu_access_pointer(dev->ip_ptr));
7514 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
7515 WARN_ON(dev->dn_ptr);
7517 if (dev->destructor)
7518 dev->destructor(dev);
7520 /* Report a network device has been unregistered */
7522 dev_net(dev)->dev_unreg_count--;
7524 wake_up(&netdev_unregistering_wq);
7526 /* Free network device */
7527 kobject_put(&dev->dev.kobj);
7531 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
7532 * all the same fields in the same order as net_device_stats, with only
7533 * the type differing, but rtnl_link_stats64 may have additional fields
7534 * at the end for newer counters.
7536 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7537 const struct net_device_stats *netdev_stats)
7539 #if BITS_PER_LONG == 64
7540 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
7541 memcpy(stats64, netdev_stats, sizeof(*stats64));
7542 /* zero out counters that only exist in rtnl_link_stats64 */
7543 memset((char *)stats64 + sizeof(*netdev_stats), 0,
7544 sizeof(*stats64) - sizeof(*netdev_stats));
7546 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
7547 const unsigned long *src = (const unsigned long *)netdev_stats;
7548 u64 *dst = (u64 *)stats64;
7550 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
7551 for (i = 0; i < n; i++)
7553 /* zero out counters that only exist in rtnl_link_stats64 */
7554 memset((char *)stats64 + n * sizeof(u64), 0,
7555 sizeof(*stats64) - n * sizeof(u64));
7558 EXPORT_SYMBOL(netdev_stats_to_stats64);
7561 * dev_get_stats - get network device statistics
7562 * @dev: device to get statistics from
7563 * @storage: place to store stats
7565 * Get network statistics from device. Return @storage.
7566 * The device driver may provide its own method by setting
7567 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7568 * otherwise the internal statistics structure is used.
7570 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
7571 struct rtnl_link_stats64 *storage)
7573 const struct net_device_ops *ops = dev->netdev_ops;
7575 if (ops->ndo_get_stats64) {
7576 memset(storage, 0, sizeof(*storage));
7577 ops->ndo_get_stats64(dev, storage);
7578 } else if (ops->ndo_get_stats) {
7579 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7581 netdev_stats_to_stats64(storage, &dev->stats);
7583 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
7584 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
7585 storage->rx_nohandler += atomic_long_read(&dev->rx_nohandler);
7588 EXPORT_SYMBOL(dev_get_stats);
7590 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7592 struct netdev_queue *queue = dev_ingress_queue(dev);
7594 #ifdef CONFIG_NET_CLS_ACT
7597 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7600 netdev_init_one_queue(dev, queue, NULL);
7601 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7602 queue->qdisc_sleeping = &noop_qdisc;
7603 rcu_assign_pointer(dev->ingress_queue, queue);
7608 static const struct ethtool_ops default_ethtool_ops;
7610 void netdev_set_default_ethtool_ops(struct net_device *dev,
7611 const struct ethtool_ops *ops)
7613 if (dev->ethtool_ops == &default_ethtool_ops)
7614 dev->ethtool_ops = ops;
7616 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7618 void netdev_freemem(struct net_device *dev)
7620 char *addr = (char *)dev - dev->padded;
7626 * alloc_netdev_mqs - allocate network device
7627 * @sizeof_priv: size of private data to allocate space for
7628 * @name: device name format string
7629 * @name_assign_type: origin of device name
7630 * @setup: callback to initialize device
7631 * @txqs: the number of TX subqueues to allocate
7632 * @rxqs: the number of RX subqueues to allocate
7634 * Allocates a struct net_device with private data area for driver use
7635 * and performs basic initialization. Also allocates subqueue structs
7636 * for each queue on the device.
7638 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7639 unsigned char name_assign_type,
7640 void (*setup)(struct net_device *),
7641 unsigned int txqs, unsigned int rxqs)
7643 struct net_device *dev;
7645 struct net_device *p;
7647 BUG_ON(strlen(name) >= sizeof(dev->name));
7650 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7656 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7661 alloc_size = sizeof(struct net_device);
7663 /* ensure 32-byte alignment of private area */
7664 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7665 alloc_size += sizeof_priv;
7667 /* ensure 32-byte alignment of whole construct */
7668 alloc_size += NETDEV_ALIGN - 1;
7670 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7672 p = vzalloc(alloc_size);
7676 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7677 dev->padded = (char *)dev - (char *)p;
7679 dev->pcpu_refcnt = alloc_percpu(int);
7680 if (!dev->pcpu_refcnt)
7683 if (dev_addr_init(dev))
7689 dev_net_set(dev, &init_net);
7691 dev->gso_max_size = GSO_MAX_SIZE;
7692 dev->gso_max_segs = GSO_MAX_SEGS;
7694 INIT_LIST_HEAD(&dev->napi_list);
7695 INIT_LIST_HEAD(&dev->unreg_list);
7696 INIT_LIST_HEAD(&dev->close_list);
7697 INIT_LIST_HEAD(&dev->link_watch_list);
7698 INIT_LIST_HEAD(&dev->adj_list.upper);
7699 INIT_LIST_HEAD(&dev->adj_list.lower);
7700 INIT_LIST_HEAD(&dev->ptype_all);
7701 INIT_LIST_HEAD(&dev->ptype_specific);
7702 #ifdef CONFIG_NET_SCHED
7703 hash_init(dev->qdisc_hash);
7705 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7708 if (!dev->tx_queue_len) {
7709 dev->priv_flags |= IFF_NO_QUEUE;
7710 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
7713 dev->num_tx_queues = txqs;
7714 dev->real_num_tx_queues = txqs;
7715 if (netif_alloc_netdev_queues(dev))
7719 dev->num_rx_queues = rxqs;
7720 dev->real_num_rx_queues = rxqs;
7721 if (netif_alloc_rx_queues(dev))
7725 strcpy(dev->name, name);
7726 dev->name_assign_type = name_assign_type;
7727 dev->group = INIT_NETDEV_GROUP;
7728 if (!dev->ethtool_ops)
7729 dev->ethtool_ops = &default_ethtool_ops;
7731 nf_hook_ingress_init(dev);
7740 free_percpu(dev->pcpu_refcnt);
7742 netdev_freemem(dev);
7745 EXPORT_SYMBOL(alloc_netdev_mqs);
7748 * free_netdev - free network device
7751 * This function does the last stage of destroying an allocated device
7752 * interface. The reference to the device object is released.
7753 * If this is the last reference then it will be freed.
7754 * Must be called in process context.
7756 void free_netdev(struct net_device *dev)
7758 struct napi_struct *p, *n;
7761 netif_free_tx_queues(dev);
7766 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7768 /* Flush device addresses */
7769 dev_addr_flush(dev);
7771 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7774 free_percpu(dev->pcpu_refcnt);
7775 dev->pcpu_refcnt = NULL;
7777 /* Compatibility with error handling in drivers */
7778 if (dev->reg_state == NETREG_UNINITIALIZED) {
7779 netdev_freemem(dev);
7783 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7784 dev->reg_state = NETREG_RELEASED;
7786 /* will free via device release */
7787 put_device(&dev->dev);
7789 EXPORT_SYMBOL(free_netdev);
7792 * synchronize_net - Synchronize with packet receive processing
7794 * Wait for packets currently being received to be done.
7795 * Does not block later packets from starting.
7797 void synchronize_net(void)
7800 if (rtnl_is_locked())
7801 synchronize_rcu_expedited();
7805 EXPORT_SYMBOL(synchronize_net);
7808 * unregister_netdevice_queue - remove device from the kernel
7812 * This function shuts down a device interface and removes it
7813 * from the kernel tables.
7814 * If head not NULL, device is queued to be unregistered later.
7816 * Callers must hold the rtnl semaphore. You may want
7817 * unregister_netdev() instead of this.
7820 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7825 list_move_tail(&dev->unreg_list, head);
7827 rollback_registered(dev);
7828 /* Finish processing unregister after unlock */
7832 EXPORT_SYMBOL(unregister_netdevice_queue);
7835 * unregister_netdevice_many - unregister many devices
7836 * @head: list of devices
7838 * Note: As most callers use a stack allocated list_head,
7839 * we force a list_del() to make sure stack wont be corrupted later.
7841 void unregister_netdevice_many(struct list_head *head)
7843 struct net_device *dev;
7845 if (!list_empty(head)) {
7846 rollback_registered_many(head);
7847 list_for_each_entry(dev, head, unreg_list)
7852 EXPORT_SYMBOL(unregister_netdevice_many);
7855 * unregister_netdev - remove device from the kernel
7858 * This function shuts down a device interface and removes it
7859 * from the kernel tables.
7861 * This is just a wrapper for unregister_netdevice that takes
7862 * the rtnl semaphore. In general you want to use this and not
7863 * unregister_netdevice.
7865 void unregister_netdev(struct net_device *dev)
7868 unregister_netdevice(dev);
7871 EXPORT_SYMBOL(unregister_netdev);
7874 * dev_change_net_namespace - move device to different nethost namespace
7876 * @net: network namespace
7877 * @pat: If not NULL name pattern to try if the current device name
7878 * is already taken in the destination network namespace.
7880 * This function shuts down a device interface and moves it
7881 * to a new network namespace. On success 0 is returned, on
7882 * a failure a netagive errno code is returned.
7884 * Callers must hold the rtnl semaphore.
7887 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7893 /* Don't allow namespace local devices to be moved. */
7895 if (dev->features & NETIF_F_NETNS_LOCAL)
7898 /* Ensure the device has been registrered */
7899 if (dev->reg_state != NETREG_REGISTERED)
7902 /* Get out if there is nothing todo */
7904 if (net_eq(dev_net(dev), net))
7907 /* Pick the destination device name, and ensure
7908 * we can use it in the destination network namespace.
7911 if (__dev_get_by_name(net, dev->name)) {
7912 /* We get here if we can't use the current device name */
7915 if (dev_get_valid_name(net, dev, pat) < 0)
7920 * And now a mini version of register_netdevice unregister_netdevice.
7923 /* If device is running close it first. */
7926 /* And unlink it from device chain */
7928 unlist_netdevice(dev);
7932 /* Shutdown queueing discipline. */
7935 /* Notify protocols, that we are about to destroy
7936 this device. They should clean all the things.
7938 Note that dev->reg_state stays at NETREG_REGISTERED.
7939 This is wanted because this way 8021q and macvlan know
7940 the device is just moving and can keep their slaves up.
7942 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7944 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7945 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7948 * Flush the unicast and multicast chains
7953 /* Send a netdev-removed uevent to the old namespace */
7954 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7955 netdev_adjacent_del_links(dev);
7957 /* Actually switch the network namespace */
7958 dev_net_set(dev, net);
7960 /* If there is an ifindex conflict assign a new one */
7961 if (__dev_get_by_index(net, dev->ifindex))
7962 dev->ifindex = dev_new_index(net);
7964 /* Send a netdev-add uevent to the new namespace */
7965 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7966 netdev_adjacent_add_links(dev);
7968 /* Fixup kobjects */
7969 err = device_rename(&dev->dev, dev->name);
7972 /* Add the device back in the hashes */
7973 list_netdevice(dev);
7975 /* Notify protocols, that a new device appeared. */
7976 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7979 * Prevent userspace races by waiting until the network
7980 * device is fully setup before sending notifications.
7982 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7989 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7991 static int dev_cpu_callback(struct notifier_block *nfb,
7992 unsigned long action,
7995 struct sk_buff **list_skb;
7996 struct sk_buff *skb;
7997 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7998 struct softnet_data *sd, *oldsd;
8000 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
8003 local_irq_disable();
8004 cpu = smp_processor_id();
8005 sd = &per_cpu(softnet_data, cpu);
8006 oldsd = &per_cpu(softnet_data, oldcpu);
8008 /* Find end of our completion_queue. */
8009 list_skb = &sd->completion_queue;
8011 list_skb = &(*list_skb)->next;
8012 /* Append completion queue from offline CPU. */
8013 *list_skb = oldsd->completion_queue;
8014 oldsd->completion_queue = NULL;
8016 /* Append output queue from offline CPU. */
8017 if (oldsd->output_queue) {
8018 *sd->output_queue_tailp = oldsd->output_queue;
8019 sd->output_queue_tailp = oldsd->output_queue_tailp;
8020 oldsd->output_queue = NULL;
8021 oldsd->output_queue_tailp = &oldsd->output_queue;
8023 /* Append NAPI poll list from offline CPU, with one exception :
8024 * process_backlog() must be called by cpu owning percpu backlog.
8025 * We properly handle process_queue & input_pkt_queue later.
8027 while (!list_empty(&oldsd->poll_list)) {
8028 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
8032 list_del_init(&napi->poll_list);
8033 if (napi->poll == process_backlog)
8036 ____napi_schedule(sd, napi);
8039 raise_softirq_irqoff(NET_TX_SOFTIRQ);
8042 /* Process offline CPU's input_pkt_queue */
8043 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
8045 input_queue_head_incr(oldsd);
8047 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
8049 input_queue_head_incr(oldsd);
8057 * netdev_increment_features - increment feature set by one
8058 * @all: current feature set
8059 * @one: new feature set
8060 * @mask: mask feature set
8062 * Computes a new feature set after adding a device with feature set
8063 * @one to the master device with current feature set @all. Will not
8064 * enable anything that is off in @mask. Returns the new feature set.
8066 netdev_features_t netdev_increment_features(netdev_features_t all,
8067 netdev_features_t one, netdev_features_t mask)
8069 if (mask & NETIF_F_HW_CSUM)
8070 mask |= NETIF_F_CSUM_MASK;
8071 mask |= NETIF_F_VLAN_CHALLENGED;
8073 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
8074 all &= one | ~NETIF_F_ALL_FOR_ALL;
8076 /* If one device supports hw checksumming, set for all. */
8077 if (all & NETIF_F_HW_CSUM)
8078 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
8082 EXPORT_SYMBOL(netdev_increment_features);
8084 static struct hlist_head * __net_init netdev_create_hash(void)
8087 struct hlist_head *hash;
8089 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
8091 for (i = 0; i < NETDEV_HASHENTRIES; i++)
8092 INIT_HLIST_HEAD(&hash[i]);
8097 /* Initialize per network namespace state */
8098 static int __net_init netdev_init(struct net *net)
8100 if (net != &init_net)
8101 INIT_LIST_HEAD(&net->dev_base_head);
8103 net->dev_name_head = netdev_create_hash();
8104 if (net->dev_name_head == NULL)
8107 net->dev_index_head = netdev_create_hash();
8108 if (net->dev_index_head == NULL)
8114 kfree(net->dev_name_head);
8120 * netdev_drivername - network driver for the device
8121 * @dev: network device
8123 * Determine network driver for device.
8125 const char *netdev_drivername(const struct net_device *dev)
8127 const struct device_driver *driver;
8128 const struct device *parent;
8129 const char *empty = "";
8131 parent = dev->dev.parent;
8135 driver = parent->driver;
8136 if (driver && driver->name)
8137 return driver->name;
8141 static void __netdev_printk(const char *level, const struct net_device *dev,
8142 struct va_format *vaf)
8144 if (dev && dev->dev.parent) {
8145 dev_printk_emit(level[1] - '0',
8148 dev_driver_string(dev->dev.parent),
8149 dev_name(dev->dev.parent),
8150 netdev_name(dev), netdev_reg_state(dev),
8153 printk("%s%s%s: %pV",
8154 level, netdev_name(dev), netdev_reg_state(dev), vaf);
8156 printk("%s(NULL net_device): %pV", level, vaf);
8160 void netdev_printk(const char *level, const struct net_device *dev,
8161 const char *format, ...)
8163 struct va_format vaf;
8166 va_start(args, format);
8171 __netdev_printk(level, dev, &vaf);
8175 EXPORT_SYMBOL(netdev_printk);
8177 #define define_netdev_printk_level(func, level) \
8178 void func(const struct net_device *dev, const char *fmt, ...) \
8180 struct va_format vaf; \
8183 va_start(args, fmt); \
8188 __netdev_printk(level, dev, &vaf); \
8192 EXPORT_SYMBOL(func);
8194 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
8195 define_netdev_printk_level(netdev_alert, KERN_ALERT);
8196 define_netdev_printk_level(netdev_crit, KERN_CRIT);
8197 define_netdev_printk_level(netdev_err, KERN_ERR);
8198 define_netdev_printk_level(netdev_warn, KERN_WARNING);
8199 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
8200 define_netdev_printk_level(netdev_info, KERN_INFO);
8202 static void __net_exit netdev_exit(struct net *net)
8204 kfree(net->dev_name_head);
8205 kfree(net->dev_index_head);
8208 static struct pernet_operations __net_initdata netdev_net_ops = {
8209 .init = netdev_init,
8210 .exit = netdev_exit,
8213 static void __net_exit default_device_exit(struct net *net)
8215 struct net_device *dev, *aux;
8217 * Push all migratable network devices back to the
8218 * initial network namespace
8221 for_each_netdev_safe(net, dev, aux) {
8223 char fb_name[IFNAMSIZ];
8225 /* Ignore unmoveable devices (i.e. loopback) */
8226 if (dev->features & NETIF_F_NETNS_LOCAL)
8229 /* Leave virtual devices for the generic cleanup */
8230 if (dev->rtnl_link_ops)
8233 /* Push remaining network devices to init_net */
8234 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
8235 err = dev_change_net_namespace(dev, &init_net, fb_name);
8237 pr_emerg("%s: failed to move %s to init_net: %d\n",
8238 __func__, dev->name, err);
8245 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
8247 /* Return with the rtnl_lock held when there are no network
8248 * devices unregistering in any network namespace in net_list.
8252 DEFINE_WAIT_FUNC(wait, woken_wake_function);
8254 add_wait_queue(&netdev_unregistering_wq, &wait);
8256 unregistering = false;
8258 list_for_each_entry(net, net_list, exit_list) {
8259 if (net->dev_unreg_count > 0) {
8260 unregistering = true;
8268 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
8270 remove_wait_queue(&netdev_unregistering_wq, &wait);
8273 static void __net_exit default_device_exit_batch(struct list_head *net_list)
8275 /* At exit all network devices most be removed from a network
8276 * namespace. Do this in the reverse order of registration.
8277 * Do this across as many network namespaces as possible to
8278 * improve batching efficiency.
8280 struct net_device *dev;
8282 LIST_HEAD(dev_kill_list);
8284 /* To prevent network device cleanup code from dereferencing
8285 * loopback devices or network devices that have been freed
8286 * wait here for all pending unregistrations to complete,
8287 * before unregistring the loopback device and allowing the
8288 * network namespace be freed.
8290 * The netdev todo list containing all network devices
8291 * unregistrations that happen in default_device_exit_batch
8292 * will run in the rtnl_unlock() at the end of
8293 * default_device_exit_batch.
8295 rtnl_lock_unregistering(net_list);
8296 list_for_each_entry(net, net_list, exit_list) {
8297 for_each_netdev_reverse(net, dev) {
8298 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
8299 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
8301 unregister_netdevice_queue(dev, &dev_kill_list);
8304 unregister_netdevice_many(&dev_kill_list);
8308 static struct pernet_operations __net_initdata default_device_ops = {
8309 .exit = default_device_exit,
8310 .exit_batch = default_device_exit_batch,
8314 * Initialize the DEV module. At boot time this walks the device list and
8315 * unhooks any devices that fail to initialise (normally hardware not
8316 * present) and leaves us with a valid list of present and active devices.
8321 * This is called single threaded during boot, so no need
8322 * to take the rtnl semaphore.
8324 static int __init net_dev_init(void)
8326 int i, rc = -ENOMEM;
8328 BUG_ON(!dev_boot_phase);
8330 if (dev_proc_init())
8333 if (netdev_kobject_init())
8336 INIT_LIST_HEAD(&ptype_all);
8337 for (i = 0; i < PTYPE_HASH_SIZE; i++)
8338 INIT_LIST_HEAD(&ptype_base[i]);
8340 INIT_LIST_HEAD(&offload_base);
8342 if (register_pernet_subsys(&netdev_net_ops))
8346 * Initialise the packet receive queues.
8349 for_each_possible_cpu(i) {
8350 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
8351 struct softnet_data *sd = &per_cpu(softnet_data, i);
8353 INIT_WORK(flush, flush_backlog);
8355 skb_queue_head_init(&sd->input_pkt_queue);
8356 skb_queue_head_init(&sd->process_queue);
8357 INIT_LIST_HEAD(&sd->poll_list);
8358 sd->output_queue_tailp = &sd->output_queue;
8360 sd->csd.func = rps_trigger_softirq;
8365 sd->backlog.poll = process_backlog;
8366 sd->backlog.weight = weight_p;
8371 /* The loopback device is special if any other network devices
8372 * is present in a network namespace the loopback device must
8373 * be present. Since we now dynamically allocate and free the
8374 * loopback device ensure this invariant is maintained by
8375 * keeping the loopback device as the first device on the
8376 * list of network devices. Ensuring the loopback devices
8377 * is the first device that appears and the last network device
8380 if (register_pernet_device(&loopback_net_ops))
8383 if (register_pernet_device(&default_device_ops))
8386 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
8387 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
8389 hotcpu_notifier(dev_cpu_callback, 0);
8396 subsys_initcall(net_dev_init);