2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <linux/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <linux/bpf.h>
98 #include <net/net_namespace.h>
100 #include <net/busy_poll.h>
101 #include <linux/rtnetlink.h>
102 #include <linux/stat.h>
104 #include <net/dst_metadata.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/iw_handler.h>
115 #include <asm/current.h>
116 #include <linux/audit.h>
117 #include <linux/dmaengine.h>
118 #include <linux/err.h>
119 #include <linux/ctype.h>
120 #include <linux/if_arp.h>
121 #include <linux/if_vlan.h>
122 #include <linux/ip.h>
124 #include <net/mpls.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/static_key.h>
136 #include <linux/hashtable.h>
137 #include <linux/vmalloc.h>
138 #include <linux/if_macvlan.h>
139 #include <linux/errqueue.h>
140 #include <linux/hrtimer.h>
141 #include <linux/netfilter_ingress.h>
142 #include <linux/crash_dump.h>
144 #include "net-sysfs.h"
146 /* Instead of increasing this, you should create a hash table. */
147 #define MAX_GRO_SKBS 8
149 /* This should be increased if a protocol with a bigger head is added. */
150 #define GRO_MAX_HEAD (MAX_HEADER + 128)
152 static DEFINE_SPINLOCK(ptype_lock);
153 static DEFINE_SPINLOCK(offload_lock);
154 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
155 struct list_head ptype_all __read_mostly; /* Taps */
156 static struct list_head offload_base __read_mostly;
158 static int netif_rx_internal(struct sk_buff *skb);
159 static int call_netdevice_notifiers_info(unsigned long val,
160 struct net_device *dev,
161 struct netdev_notifier_info *info);
164 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
167 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
169 * Writers must hold the rtnl semaphore while they loop through the
170 * dev_base_head list, and hold dev_base_lock for writing when they do the
171 * actual updates. This allows pure readers to access the list even
172 * while a writer is preparing to update it.
174 * To put it another way, dev_base_lock is held for writing only to
175 * protect against pure readers; the rtnl semaphore provides the
176 * protection against other writers.
178 * See, for example usages, register_netdevice() and
179 * unregister_netdevice(), which must be called with the rtnl
182 DEFINE_RWLOCK(dev_base_lock);
183 EXPORT_SYMBOL(dev_base_lock);
185 /* protects napi_hash addition/deletion and napi_gen_id */
186 static DEFINE_SPINLOCK(napi_hash_lock);
188 static unsigned int napi_gen_id = NR_CPUS;
189 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
191 static seqcount_t devnet_rename_seq;
193 static inline void dev_base_seq_inc(struct net *net)
195 while (++net->dev_base_seq == 0);
198 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
200 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
202 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
205 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
207 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
210 static inline void rps_lock(struct softnet_data *sd)
213 spin_lock(&sd->input_pkt_queue.lock);
217 static inline void rps_unlock(struct softnet_data *sd)
220 spin_unlock(&sd->input_pkt_queue.lock);
224 /* Device list insertion */
225 static void list_netdevice(struct net_device *dev)
227 struct net *net = dev_net(dev);
231 write_lock_bh(&dev_base_lock);
232 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
233 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
234 hlist_add_head_rcu(&dev->index_hlist,
235 dev_index_hash(net, dev->ifindex));
236 write_unlock_bh(&dev_base_lock);
238 dev_base_seq_inc(net);
241 /* Device list removal
242 * caller must respect a RCU grace period before freeing/reusing dev
244 static void unlist_netdevice(struct net_device *dev)
248 /* Unlink dev from the device chain */
249 write_lock_bh(&dev_base_lock);
250 list_del_rcu(&dev->dev_list);
251 hlist_del_rcu(&dev->name_hlist);
252 hlist_del_rcu(&dev->index_hlist);
253 write_unlock_bh(&dev_base_lock);
255 dev_base_seq_inc(dev_net(dev));
262 static RAW_NOTIFIER_HEAD(netdev_chain);
265 * Device drivers call our routines to queue packets here. We empty the
266 * queue in the local softnet handler.
269 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
270 EXPORT_PER_CPU_SYMBOL(softnet_data);
272 #ifdef CONFIG_LOCKDEP
274 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
275 * according to dev->type
277 static const unsigned short netdev_lock_type[] =
278 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
279 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
280 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
281 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
282 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
283 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
284 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
285 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
286 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
287 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
288 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
289 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
290 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
291 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
292 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
294 static const char *const netdev_lock_name[] =
295 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
296 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
297 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
298 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
299 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
300 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
301 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
302 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
303 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
304 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
305 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
306 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
307 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
308 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
309 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
311 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
312 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
318 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
319 if (netdev_lock_type[i] == dev_type)
321 /* the last key is used by default */
322 return ARRAY_SIZE(netdev_lock_type) - 1;
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
326 unsigned short dev_type)
330 i = netdev_lock_pos(dev_type);
331 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
332 netdev_lock_name[i]);
335 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
339 i = netdev_lock_pos(dev->type);
340 lockdep_set_class_and_name(&dev->addr_list_lock,
341 &netdev_addr_lock_key[i],
342 netdev_lock_name[i]);
345 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
346 unsigned short dev_type)
349 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
354 /*******************************************************************************
356 Protocol management and registration routines
358 *******************************************************************************/
361 * Add a protocol ID to the list. Now that the input handler is
362 * smarter we can dispense with all the messy stuff that used to be
365 * BEWARE!!! Protocol handlers, mangling input packets,
366 * MUST BE last in hash buckets and checking protocol handlers
367 * MUST start from promiscuous ptype_all chain in net_bh.
368 * It is true now, do not change it.
369 * Explanation follows: if protocol handler, mangling packet, will
370 * be the first on list, it is not able to sense, that packet
371 * is cloned and should be copied-on-write, so that it will
372 * change it and subsequent readers will get broken packet.
376 static inline struct list_head *ptype_head(const struct packet_type *pt)
378 if (pt->type == htons(ETH_P_ALL))
379 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
381 return pt->dev ? &pt->dev->ptype_specific :
382 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
386 * dev_add_pack - add packet handler
387 * @pt: packet type declaration
389 * Add a protocol handler to the networking stack. The passed &packet_type
390 * is linked into kernel lists and may not be freed until it has been
391 * removed from the kernel lists.
393 * This call does not sleep therefore it can not
394 * guarantee all CPU's that are in middle of receiving packets
395 * will see the new packet type (until the next received packet).
398 void dev_add_pack(struct packet_type *pt)
400 struct list_head *head = ptype_head(pt);
402 spin_lock(&ptype_lock);
403 list_add_rcu(&pt->list, head);
404 spin_unlock(&ptype_lock);
406 EXPORT_SYMBOL(dev_add_pack);
409 * __dev_remove_pack - remove packet handler
410 * @pt: packet type declaration
412 * Remove a protocol handler that was previously added to the kernel
413 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
414 * from the kernel lists and can be freed or reused once this function
417 * The packet type might still be in use by receivers
418 * and must not be freed until after all the CPU's have gone
419 * through a quiescent state.
421 void __dev_remove_pack(struct packet_type *pt)
423 struct list_head *head = ptype_head(pt);
424 struct packet_type *pt1;
426 spin_lock(&ptype_lock);
428 list_for_each_entry(pt1, head, list) {
430 list_del_rcu(&pt->list);
435 pr_warn("dev_remove_pack: %p not found\n", pt);
437 spin_unlock(&ptype_lock);
439 EXPORT_SYMBOL(__dev_remove_pack);
442 * dev_remove_pack - remove packet handler
443 * @pt: packet type declaration
445 * Remove a protocol handler that was previously added to the kernel
446 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
447 * from the kernel lists and can be freed or reused once this function
450 * This call sleeps to guarantee that no CPU is looking at the packet
453 void dev_remove_pack(struct packet_type *pt)
455 __dev_remove_pack(pt);
459 EXPORT_SYMBOL(dev_remove_pack);
463 * dev_add_offload - register offload handlers
464 * @po: protocol offload declaration
466 * Add protocol offload handlers to the networking stack. The passed
467 * &proto_offload is linked into kernel lists and may not be freed until
468 * it has been removed from the kernel lists.
470 * This call does not sleep therefore it can not
471 * guarantee all CPU's that are in middle of receiving packets
472 * will see the new offload handlers (until the next received packet).
474 void dev_add_offload(struct packet_offload *po)
476 struct packet_offload *elem;
478 spin_lock(&offload_lock);
479 list_for_each_entry(elem, &offload_base, list) {
480 if (po->priority < elem->priority)
483 list_add_rcu(&po->list, elem->list.prev);
484 spin_unlock(&offload_lock);
486 EXPORT_SYMBOL(dev_add_offload);
489 * __dev_remove_offload - remove offload handler
490 * @po: packet offload declaration
492 * Remove a protocol offload handler that was previously added to the
493 * kernel offload handlers by dev_add_offload(). The passed &offload_type
494 * is removed from the kernel lists and can be freed or reused once this
497 * The packet type might still be in use by receivers
498 * and must not be freed until after all the CPU's have gone
499 * through a quiescent state.
501 static void __dev_remove_offload(struct packet_offload *po)
503 struct list_head *head = &offload_base;
504 struct packet_offload *po1;
506 spin_lock(&offload_lock);
508 list_for_each_entry(po1, head, list) {
510 list_del_rcu(&po->list);
515 pr_warn("dev_remove_offload: %p not found\n", po);
517 spin_unlock(&offload_lock);
521 * dev_remove_offload - remove packet offload handler
522 * @po: packet offload declaration
524 * Remove a packet offload handler that was previously added to the kernel
525 * offload handlers by dev_add_offload(). The passed &offload_type is
526 * removed from the kernel lists and can be freed or reused once this
529 * This call sleeps to guarantee that no CPU is looking at the packet
532 void dev_remove_offload(struct packet_offload *po)
534 __dev_remove_offload(po);
538 EXPORT_SYMBOL(dev_remove_offload);
540 /******************************************************************************
542 Device Boot-time Settings Routines
544 *******************************************************************************/
546 /* Boot time configuration table */
547 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
550 * netdev_boot_setup_add - add new setup entry
551 * @name: name of the device
552 * @map: configured settings for the device
554 * Adds new setup entry to the dev_boot_setup list. The function
555 * returns 0 on error and 1 on success. This is a generic routine to
558 static int netdev_boot_setup_add(char *name, struct ifmap *map)
560 struct netdev_boot_setup *s;
564 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
565 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
566 memset(s[i].name, 0, sizeof(s[i].name));
567 strlcpy(s[i].name, name, IFNAMSIZ);
568 memcpy(&s[i].map, map, sizeof(s[i].map));
573 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
577 * netdev_boot_setup_check - check boot time settings
578 * @dev: the netdevice
580 * Check boot time settings for the device.
581 * The found settings are set for the device to be used
582 * later in the device probing.
583 * Returns 0 if no settings found, 1 if they are.
585 int netdev_boot_setup_check(struct net_device *dev)
587 struct netdev_boot_setup *s = dev_boot_setup;
590 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
591 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
592 !strcmp(dev->name, s[i].name)) {
593 dev->irq = s[i].map.irq;
594 dev->base_addr = s[i].map.base_addr;
595 dev->mem_start = s[i].map.mem_start;
596 dev->mem_end = s[i].map.mem_end;
602 EXPORT_SYMBOL(netdev_boot_setup_check);
606 * netdev_boot_base - get address from boot time settings
607 * @prefix: prefix for network device
608 * @unit: id for network device
610 * Check boot time settings for the base address of device.
611 * The found settings are set for the device to be used
612 * later in the device probing.
613 * Returns 0 if no settings found.
615 unsigned long netdev_boot_base(const char *prefix, int unit)
617 const struct netdev_boot_setup *s = dev_boot_setup;
621 sprintf(name, "%s%d", prefix, unit);
624 * If device already registered then return base of 1
625 * to indicate not to probe for this interface
627 if (__dev_get_by_name(&init_net, name))
630 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
631 if (!strcmp(name, s[i].name))
632 return s[i].map.base_addr;
637 * Saves at boot time configured settings for any netdevice.
639 int __init netdev_boot_setup(char *str)
644 str = get_options(str, ARRAY_SIZE(ints), ints);
649 memset(&map, 0, sizeof(map));
653 map.base_addr = ints[2];
655 map.mem_start = ints[3];
657 map.mem_end = ints[4];
659 /* Add new entry to the list */
660 return netdev_boot_setup_add(str, &map);
663 __setup("netdev=", netdev_boot_setup);
665 /*******************************************************************************
667 Device Interface Subroutines
669 *******************************************************************************/
672 * dev_get_iflink - get 'iflink' value of a interface
673 * @dev: targeted interface
675 * Indicates the ifindex the interface is linked to.
676 * Physical interfaces have the same 'ifindex' and 'iflink' values.
679 int dev_get_iflink(const struct net_device *dev)
681 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
682 return dev->netdev_ops->ndo_get_iflink(dev);
686 EXPORT_SYMBOL(dev_get_iflink);
689 * dev_fill_metadata_dst - Retrieve tunnel egress information.
690 * @dev: targeted interface
693 * For better visibility of tunnel traffic OVS needs to retrieve
694 * egress tunnel information for a packet. Following API allows
695 * user to get this info.
697 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
699 struct ip_tunnel_info *info;
701 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
704 info = skb_tunnel_info_unclone(skb);
707 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
710 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
712 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
715 * __dev_get_by_name - find a device by its name
716 * @net: the applicable net namespace
717 * @name: name to find
719 * Find an interface by name. Must be called under RTNL semaphore
720 * or @dev_base_lock. If the name is found a pointer to the device
721 * is returned. If the name is not found then %NULL is returned. The
722 * reference counters are not incremented so the caller must be
723 * careful with locks.
726 struct net_device *__dev_get_by_name(struct net *net, const char *name)
728 struct net_device *dev;
729 struct hlist_head *head = dev_name_hash(net, name);
731 hlist_for_each_entry(dev, head, name_hlist)
732 if (!strncmp(dev->name, name, IFNAMSIZ))
737 EXPORT_SYMBOL(__dev_get_by_name);
740 * dev_get_by_name_rcu - find a device by its name
741 * @net: the applicable net namespace
742 * @name: name to find
744 * Find an interface by name.
745 * If the name is found a pointer to the device is returned.
746 * If the name is not found then %NULL is returned.
747 * The reference counters are not incremented so the caller must be
748 * careful with locks. The caller must hold RCU lock.
751 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
753 struct net_device *dev;
754 struct hlist_head *head = dev_name_hash(net, name);
756 hlist_for_each_entry_rcu(dev, head, name_hlist)
757 if (!strncmp(dev->name, name, IFNAMSIZ))
762 EXPORT_SYMBOL(dev_get_by_name_rcu);
765 * dev_get_by_name - find a device by its name
766 * @net: the applicable net namespace
767 * @name: name to find
769 * Find an interface by name. This can be called from any
770 * context and does its own locking. The returned handle has
771 * the usage count incremented and the caller must use dev_put() to
772 * release it when it is no longer needed. %NULL is returned if no
773 * matching device is found.
776 struct net_device *dev_get_by_name(struct net *net, const char *name)
778 struct net_device *dev;
781 dev = dev_get_by_name_rcu(net, name);
787 EXPORT_SYMBOL(dev_get_by_name);
790 * __dev_get_by_index - find a device by its ifindex
791 * @net: the applicable net namespace
792 * @ifindex: index of device
794 * Search for an interface by index. Returns %NULL if the device
795 * is not found or a pointer to the device. The device has not
796 * had its reference counter increased so the caller must be careful
797 * about locking. The caller must hold either the RTNL semaphore
801 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
803 struct net_device *dev;
804 struct hlist_head *head = dev_index_hash(net, ifindex);
806 hlist_for_each_entry(dev, head, index_hlist)
807 if (dev->ifindex == ifindex)
812 EXPORT_SYMBOL(__dev_get_by_index);
815 * dev_get_by_index_rcu - find a device by its ifindex
816 * @net: the applicable net namespace
817 * @ifindex: index of device
819 * Search for an interface by index. Returns %NULL if the device
820 * is not found or a pointer to the device. The device has not
821 * had its reference counter increased so the caller must be careful
822 * about locking. The caller must hold RCU lock.
825 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
827 struct net_device *dev;
828 struct hlist_head *head = dev_index_hash(net, ifindex);
830 hlist_for_each_entry_rcu(dev, head, index_hlist)
831 if (dev->ifindex == ifindex)
836 EXPORT_SYMBOL(dev_get_by_index_rcu);
840 * dev_get_by_index - find a device by its ifindex
841 * @net: the applicable net namespace
842 * @ifindex: index of device
844 * Search for an interface by index. Returns NULL if the device
845 * is not found or a pointer to the device. The device returned has
846 * had a reference added and the pointer is safe until the user calls
847 * dev_put to indicate they have finished with it.
850 struct net_device *dev_get_by_index(struct net *net, int ifindex)
852 struct net_device *dev;
855 dev = dev_get_by_index_rcu(net, ifindex);
861 EXPORT_SYMBOL(dev_get_by_index);
864 * netdev_get_name - get a netdevice name, knowing its ifindex.
865 * @net: network namespace
866 * @name: a pointer to the buffer where the name will be stored.
867 * @ifindex: the ifindex of the interface to get the name from.
869 * The use of raw_seqcount_begin() and cond_resched() before
870 * retrying is required as we want to give the writers a chance
871 * to complete when CONFIG_PREEMPT is not set.
873 int netdev_get_name(struct net *net, char *name, int ifindex)
875 struct net_device *dev;
879 seq = raw_seqcount_begin(&devnet_rename_seq);
881 dev = dev_get_by_index_rcu(net, ifindex);
887 strcpy(name, dev->name);
889 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
898 * dev_getbyhwaddr_rcu - find a device by its hardware address
899 * @net: the applicable net namespace
900 * @type: media type of device
901 * @ha: hardware address
903 * Search for an interface by MAC address. Returns NULL if the device
904 * is not found or a pointer to the device.
905 * The caller must hold RCU or RTNL.
906 * The returned device has not had its ref count increased
907 * and the caller must therefore be careful about locking
911 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
914 struct net_device *dev;
916 for_each_netdev_rcu(net, dev)
917 if (dev->type == type &&
918 !memcmp(dev->dev_addr, ha, dev->addr_len))
923 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
925 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
927 struct net_device *dev;
930 for_each_netdev(net, dev)
931 if (dev->type == type)
936 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
938 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
940 struct net_device *dev, *ret = NULL;
943 for_each_netdev_rcu(net, dev)
944 if (dev->type == type) {
952 EXPORT_SYMBOL(dev_getfirstbyhwtype);
955 * __dev_get_by_flags - find any device with given flags
956 * @net: the applicable net namespace
957 * @if_flags: IFF_* values
958 * @mask: bitmask of bits in if_flags to check
960 * Search for any interface with the given flags. Returns NULL if a device
961 * is not found or a pointer to the device. Must be called inside
962 * rtnl_lock(), and result refcount is unchanged.
965 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
968 struct net_device *dev, *ret;
973 for_each_netdev(net, dev) {
974 if (((dev->flags ^ if_flags) & mask) == 0) {
981 EXPORT_SYMBOL(__dev_get_by_flags);
984 * dev_valid_name - check if name is okay for network device
987 * Network device names need to be valid file names to
988 * to allow sysfs to work. We also disallow any kind of
991 bool dev_valid_name(const char *name)
995 if (strlen(name) >= IFNAMSIZ)
997 if (!strcmp(name, ".") || !strcmp(name, ".."))
1001 if (*name == '/' || *name == ':' || isspace(*name))
1007 EXPORT_SYMBOL(dev_valid_name);
1010 * __dev_alloc_name - allocate a name for a device
1011 * @net: network namespace to allocate the device name in
1012 * @name: name format string
1013 * @buf: scratch buffer and result name string
1015 * Passed a format string - eg "lt%d" it will try and find a suitable
1016 * id. It scans list of devices to build up a free map, then chooses
1017 * the first empty slot. The caller must hold the dev_base or rtnl lock
1018 * while allocating the name and adding the device in order to avoid
1020 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1021 * Returns the number of the unit assigned or a negative errno code.
1024 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1028 const int max_netdevices = 8*PAGE_SIZE;
1029 unsigned long *inuse;
1030 struct net_device *d;
1032 p = strnchr(name, IFNAMSIZ-1, '%');
1035 * Verify the string as this thing may have come from
1036 * the user. There must be either one "%d" and no other "%"
1039 if (p[1] != 'd' || strchr(p + 2, '%'))
1042 /* Use one page as a bit array of possible slots */
1043 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1047 for_each_netdev(net, d) {
1048 if (!sscanf(d->name, name, &i))
1050 if (i < 0 || i >= max_netdevices)
1053 /* avoid cases where sscanf is not exact inverse of printf */
1054 snprintf(buf, IFNAMSIZ, name, i);
1055 if (!strncmp(buf, d->name, IFNAMSIZ))
1059 i = find_first_zero_bit(inuse, max_netdevices);
1060 free_page((unsigned long) inuse);
1064 snprintf(buf, IFNAMSIZ, name, i);
1065 if (!__dev_get_by_name(net, buf))
1068 /* It is possible to run out of possible slots
1069 * when the name is long and there isn't enough space left
1070 * for the digits, or if all bits are used.
1076 * dev_alloc_name - allocate a name for a device
1078 * @name: name format string
1080 * Passed a format string - eg "lt%d" it will try and find a suitable
1081 * id. It scans list of devices to build up a free map, then chooses
1082 * the first empty slot. The caller must hold the dev_base or rtnl lock
1083 * while allocating the name and adding the device in order to avoid
1085 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1086 * Returns the number of the unit assigned or a negative errno code.
1089 int dev_alloc_name(struct net_device *dev, const char *name)
1095 BUG_ON(!dev_net(dev));
1097 ret = __dev_alloc_name(net, name, buf);
1099 strlcpy(dev->name, buf, IFNAMSIZ);
1102 EXPORT_SYMBOL(dev_alloc_name);
1104 static int dev_alloc_name_ns(struct net *net,
1105 struct net_device *dev,
1111 ret = __dev_alloc_name(net, name, buf);
1113 strlcpy(dev->name, buf, IFNAMSIZ);
1117 static int dev_get_valid_name(struct net *net,
1118 struct net_device *dev,
1123 if (!dev_valid_name(name))
1126 if (strchr(name, '%'))
1127 return dev_alloc_name_ns(net, dev, name);
1128 else if (__dev_get_by_name(net, name))
1130 else if (dev->name != name)
1131 strlcpy(dev->name, name, IFNAMSIZ);
1137 * dev_change_name - change name of a device
1139 * @newname: name (or format string) must be at least IFNAMSIZ
1141 * Change name of a device, can pass format strings "eth%d".
1144 int dev_change_name(struct net_device *dev, const char *newname)
1146 unsigned char old_assign_type;
1147 char oldname[IFNAMSIZ];
1153 BUG_ON(!dev_net(dev));
1156 if (dev->flags & IFF_UP)
1159 write_seqcount_begin(&devnet_rename_seq);
1161 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1162 write_seqcount_end(&devnet_rename_seq);
1166 memcpy(oldname, dev->name, IFNAMSIZ);
1168 err = dev_get_valid_name(net, dev, newname);
1170 write_seqcount_end(&devnet_rename_seq);
1174 if (oldname[0] && !strchr(oldname, '%'))
1175 netdev_info(dev, "renamed from %s\n", oldname);
1177 old_assign_type = dev->name_assign_type;
1178 dev->name_assign_type = NET_NAME_RENAMED;
1181 ret = device_rename(&dev->dev, dev->name);
1183 memcpy(dev->name, oldname, IFNAMSIZ);
1184 dev->name_assign_type = old_assign_type;
1185 write_seqcount_end(&devnet_rename_seq);
1189 write_seqcount_end(&devnet_rename_seq);
1191 netdev_adjacent_rename_links(dev, oldname);
1193 write_lock_bh(&dev_base_lock);
1194 hlist_del_rcu(&dev->name_hlist);
1195 write_unlock_bh(&dev_base_lock);
1199 write_lock_bh(&dev_base_lock);
1200 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1201 write_unlock_bh(&dev_base_lock);
1203 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1204 ret = notifier_to_errno(ret);
1207 /* err >= 0 after dev_alloc_name() or stores the first errno */
1210 write_seqcount_begin(&devnet_rename_seq);
1211 memcpy(dev->name, oldname, IFNAMSIZ);
1212 memcpy(oldname, newname, IFNAMSIZ);
1213 dev->name_assign_type = old_assign_type;
1214 old_assign_type = NET_NAME_RENAMED;
1217 pr_err("%s: name change rollback failed: %d\n",
1226 * dev_set_alias - change ifalias of a device
1228 * @alias: name up to IFALIASZ
1229 * @len: limit of bytes to copy from info
1231 * Set ifalias for a device,
1233 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1239 if (len >= IFALIASZ)
1243 kfree(dev->ifalias);
1244 dev->ifalias = NULL;
1248 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1251 dev->ifalias = new_ifalias;
1253 strlcpy(dev->ifalias, alias, len+1);
1259 * netdev_features_change - device changes features
1260 * @dev: device to cause notification
1262 * Called to indicate a device has changed features.
1264 void netdev_features_change(struct net_device *dev)
1266 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1268 EXPORT_SYMBOL(netdev_features_change);
1271 * netdev_state_change - device changes state
1272 * @dev: device to cause notification
1274 * Called to indicate a device has changed state. This function calls
1275 * the notifier chains for netdev_chain and sends a NEWLINK message
1276 * to the routing socket.
1278 void netdev_state_change(struct net_device *dev)
1280 if (dev->flags & IFF_UP) {
1281 struct netdev_notifier_change_info change_info;
1283 change_info.flags_changed = 0;
1284 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1286 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1289 EXPORT_SYMBOL(netdev_state_change);
1292 * netdev_notify_peers - notify network peers about existence of @dev
1293 * @dev: network device
1295 * Generate traffic such that interested network peers are aware of
1296 * @dev, such as by generating a gratuitous ARP. This may be used when
1297 * a device wants to inform the rest of the network about some sort of
1298 * reconfiguration such as a failover event or virtual machine
1301 void netdev_notify_peers(struct net_device *dev)
1304 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1307 EXPORT_SYMBOL(netdev_notify_peers);
1309 static int __dev_open(struct net_device *dev)
1311 const struct net_device_ops *ops = dev->netdev_ops;
1316 if (!netif_device_present(dev))
1319 /* Block netpoll from trying to do any rx path servicing.
1320 * If we don't do this there is a chance ndo_poll_controller
1321 * or ndo_poll may be running while we open the device
1323 netpoll_poll_disable(dev);
1325 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1326 ret = notifier_to_errno(ret);
1330 set_bit(__LINK_STATE_START, &dev->state);
1332 if (ops->ndo_validate_addr)
1333 ret = ops->ndo_validate_addr(dev);
1335 if (!ret && ops->ndo_open)
1336 ret = ops->ndo_open(dev);
1338 netpoll_poll_enable(dev);
1341 clear_bit(__LINK_STATE_START, &dev->state);
1343 dev->flags |= IFF_UP;
1344 dev_set_rx_mode(dev);
1346 add_device_randomness(dev->dev_addr, dev->addr_len);
1353 * dev_open - prepare an interface for use.
1354 * @dev: device to open
1356 * Takes a device from down to up state. The device's private open
1357 * function is invoked and then the multicast lists are loaded. Finally
1358 * the device is moved into the up state and a %NETDEV_UP message is
1359 * sent to the netdev notifier chain.
1361 * Calling this function on an active interface is a nop. On a failure
1362 * a negative errno code is returned.
1364 int dev_open(struct net_device *dev)
1368 if (dev->flags & IFF_UP)
1371 ret = __dev_open(dev);
1375 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1376 call_netdevice_notifiers(NETDEV_UP, dev);
1380 EXPORT_SYMBOL(dev_open);
1382 static int __dev_close_many(struct list_head *head)
1384 struct net_device *dev;
1389 list_for_each_entry(dev, head, close_list) {
1390 /* Temporarily disable netpoll until the interface is down */
1391 netpoll_poll_disable(dev);
1393 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1395 clear_bit(__LINK_STATE_START, &dev->state);
1397 /* Synchronize to scheduled poll. We cannot touch poll list, it
1398 * can be even on different cpu. So just clear netif_running().
1400 * dev->stop() will invoke napi_disable() on all of it's
1401 * napi_struct instances on this device.
1403 smp_mb__after_atomic(); /* Commit netif_running(). */
1406 dev_deactivate_many(head);
1408 list_for_each_entry(dev, head, close_list) {
1409 const struct net_device_ops *ops = dev->netdev_ops;
1412 * Call the device specific close. This cannot fail.
1413 * Only if device is UP
1415 * We allow it to be called even after a DETACH hot-plug
1421 dev->flags &= ~IFF_UP;
1422 netpoll_poll_enable(dev);
1428 static int __dev_close(struct net_device *dev)
1433 list_add(&dev->close_list, &single);
1434 retval = __dev_close_many(&single);
1440 int dev_close_many(struct list_head *head, bool unlink)
1442 struct net_device *dev, *tmp;
1444 /* Remove the devices that don't need to be closed */
1445 list_for_each_entry_safe(dev, tmp, head, close_list)
1446 if (!(dev->flags & IFF_UP))
1447 list_del_init(&dev->close_list);
1449 __dev_close_many(head);
1451 list_for_each_entry_safe(dev, tmp, head, close_list) {
1452 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1453 call_netdevice_notifiers(NETDEV_DOWN, dev);
1455 list_del_init(&dev->close_list);
1460 EXPORT_SYMBOL(dev_close_many);
1463 * dev_close - shutdown an interface.
1464 * @dev: device to shutdown
1466 * This function moves an active device into down state. A
1467 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1468 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1471 int dev_close(struct net_device *dev)
1473 if (dev->flags & IFF_UP) {
1476 list_add(&dev->close_list, &single);
1477 dev_close_many(&single, true);
1482 EXPORT_SYMBOL(dev_close);
1486 * dev_disable_lro - disable Large Receive Offload on a device
1489 * Disable Large Receive Offload (LRO) on a net device. Must be
1490 * called under RTNL. This is needed if received packets may be
1491 * forwarded to another interface.
1493 void dev_disable_lro(struct net_device *dev)
1495 struct net_device *lower_dev;
1496 struct list_head *iter;
1498 dev->wanted_features &= ~NETIF_F_LRO;
1499 netdev_update_features(dev);
1501 if (unlikely(dev->features & NETIF_F_LRO))
1502 netdev_WARN(dev, "failed to disable LRO!\n");
1504 netdev_for_each_lower_dev(dev, lower_dev, iter)
1505 dev_disable_lro(lower_dev);
1507 EXPORT_SYMBOL(dev_disable_lro);
1509 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1510 struct net_device *dev)
1512 struct netdev_notifier_info info;
1514 netdev_notifier_info_init(&info, dev);
1515 return nb->notifier_call(nb, val, &info);
1518 static int dev_boot_phase = 1;
1521 * register_netdevice_notifier - register a network notifier block
1524 * Register a notifier to be called when network device events occur.
1525 * The notifier passed is linked into the kernel structures and must
1526 * not be reused until it has been unregistered. A negative errno code
1527 * is returned on a failure.
1529 * When registered all registration and up events are replayed
1530 * to the new notifier to allow device to have a race free
1531 * view of the network device list.
1534 int register_netdevice_notifier(struct notifier_block *nb)
1536 struct net_device *dev;
1537 struct net_device *last;
1542 err = raw_notifier_chain_register(&netdev_chain, nb);
1548 for_each_netdev(net, dev) {
1549 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1550 err = notifier_to_errno(err);
1554 if (!(dev->flags & IFF_UP))
1557 call_netdevice_notifier(nb, NETDEV_UP, dev);
1568 for_each_netdev(net, dev) {
1572 if (dev->flags & IFF_UP) {
1573 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1575 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1577 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1582 raw_notifier_chain_unregister(&netdev_chain, nb);
1585 EXPORT_SYMBOL(register_netdevice_notifier);
1588 * unregister_netdevice_notifier - unregister a network notifier block
1591 * Unregister a notifier previously registered by
1592 * register_netdevice_notifier(). The notifier is unlinked into the
1593 * kernel structures and may then be reused. A negative errno code
1594 * is returned on a failure.
1596 * After unregistering unregister and down device events are synthesized
1597 * for all devices on the device list to the removed notifier to remove
1598 * the need for special case cleanup code.
1601 int unregister_netdevice_notifier(struct notifier_block *nb)
1603 struct net_device *dev;
1608 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1613 for_each_netdev(net, dev) {
1614 if (dev->flags & IFF_UP) {
1615 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1617 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1619 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1626 EXPORT_SYMBOL(unregister_netdevice_notifier);
1629 * call_netdevice_notifiers_info - call all network notifier blocks
1630 * @val: value passed unmodified to notifier function
1631 * @dev: net_device pointer passed unmodified to notifier function
1632 * @info: notifier information data
1634 * Call all network notifier blocks. Parameters and return value
1635 * are as for raw_notifier_call_chain().
1638 static int call_netdevice_notifiers_info(unsigned long val,
1639 struct net_device *dev,
1640 struct netdev_notifier_info *info)
1643 netdev_notifier_info_init(info, dev);
1644 return raw_notifier_call_chain(&netdev_chain, val, info);
1648 * call_netdevice_notifiers - call all network notifier blocks
1649 * @val: value passed unmodified to notifier function
1650 * @dev: net_device pointer passed unmodified to notifier function
1652 * Call all network notifier blocks. Parameters and return value
1653 * are as for raw_notifier_call_chain().
1656 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1658 struct netdev_notifier_info info;
1660 return call_netdevice_notifiers_info(val, dev, &info);
1662 EXPORT_SYMBOL(call_netdevice_notifiers);
1664 #ifdef CONFIG_NET_INGRESS
1665 static struct static_key ingress_needed __read_mostly;
1667 void net_inc_ingress_queue(void)
1669 static_key_slow_inc(&ingress_needed);
1671 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1673 void net_dec_ingress_queue(void)
1675 static_key_slow_dec(&ingress_needed);
1677 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1680 #ifdef CONFIG_NET_EGRESS
1681 static struct static_key egress_needed __read_mostly;
1683 void net_inc_egress_queue(void)
1685 static_key_slow_inc(&egress_needed);
1687 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1689 void net_dec_egress_queue(void)
1691 static_key_slow_dec(&egress_needed);
1693 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1696 static struct static_key netstamp_needed __read_mostly;
1697 #ifdef HAVE_JUMP_LABEL
1698 /* We are not allowed to call static_key_slow_dec() from irq context
1699 * If net_disable_timestamp() is called from irq context, defer the
1700 * static_key_slow_dec() calls.
1702 static atomic_t netstamp_needed_deferred;
1705 void net_enable_timestamp(void)
1707 #ifdef HAVE_JUMP_LABEL
1708 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1712 static_key_slow_dec(&netstamp_needed);
1716 static_key_slow_inc(&netstamp_needed);
1718 EXPORT_SYMBOL(net_enable_timestamp);
1720 void net_disable_timestamp(void)
1722 #ifdef HAVE_JUMP_LABEL
1723 if (in_interrupt()) {
1724 atomic_inc(&netstamp_needed_deferred);
1728 static_key_slow_dec(&netstamp_needed);
1730 EXPORT_SYMBOL(net_disable_timestamp);
1732 static inline void net_timestamp_set(struct sk_buff *skb)
1735 if (static_key_false(&netstamp_needed))
1736 __net_timestamp(skb);
1739 #define net_timestamp_check(COND, SKB) \
1740 if (static_key_false(&netstamp_needed)) { \
1741 if ((COND) && !(SKB)->tstamp) \
1742 __net_timestamp(SKB); \
1745 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1749 if (!(dev->flags & IFF_UP))
1752 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1753 if (skb->len <= len)
1756 /* if TSO is enabled, we don't care about the length as the packet
1757 * could be forwarded without being segmented before
1759 if (skb_is_gso(skb))
1764 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1766 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1768 int ret = ____dev_forward_skb(dev, skb);
1771 skb->protocol = eth_type_trans(skb, dev);
1772 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1777 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1780 * dev_forward_skb - loopback an skb to another netif
1782 * @dev: destination network device
1783 * @skb: buffer to forward
1786 * NET_RX_SUCCESS (no congestion)
1787 * NET_RX_DROP (packet was dropped, but freed)
1789 * dev_forward_skb can be used for injecting an skb from the
1790 * start_xmit function of one device into the receive queue
1791 * of another device.
1793 * The receiving device may be in another namespace, so
1794 * we have to clear all information in the skb that could
1795 * impact namespace isolation.
1797 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1799 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1801 EXPORT_SYMBOL_GPL(dev_forward_skb);
1803 static inline int deliver_skb(struct sk_buff *skb,
1804 struct packet_type *pt_prev,
1805 struct net_device *orig_dev)
1807 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1809 atomic_inc(&skb->users);
1810 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1813 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1814 struct packet_type **pt,
1815 struct net_device *orig_dev,
1817 struct list_head *ptype_list)
1819 struct packet_type *ptype, *pt_prev = *pt;
1821 list_for_each_entry_rcu(ptype, ptype_list, list) {
1822 if (ptype->type != type)
1825 deliver_skb(skb, pt_prev, orig_dev);
1831 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1833 if (!ptype->af_packet_priv || !skb->sk)
1836 if (ptype->id_match)
1837 return ptype->id_match(ptype, skb->sk);
1838 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1845 * Support routine. Sends outgoing frames to any network
1846 * taps currently in use.
1849 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1851 struct packet_type *ptype;
1852 struct sk_buff *skb2 = NULL;
1853 struct packet_type *pt_prev = NULL;
1854 struct list_head *ptype_list = &ptype_all;
1858 list_for_each_entry_rcu(ptype, ptype_list, list) {
1859 /* Never send packets back to the socket
1860 * they originated from - MvS (miquels@drinkel.ow.org)
1862 if (skb_loop_sk(ptype, skb))
1866 deliver_skb(skb2, pt_prev, skb->dev);
1871 /* need to clone skb, done only once */
1872 skb2 = skb_clone(skb, GFP_ATOMIC);
1876 net_timestamp_set(skb2);
1878 /* skb->nh should be correctly
1879 * set by sender, so that the second statement is
1880 * just protection against buggy protocols.
1882 skb_reset_mac_header(skb2);
1884 if (skb_network_header(skb2) < skb2->data ||
1885 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1886 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1887 ntohs(skb2->protocol),
1889 skb_reset_network_header(skb2);
1892 skb2->transport_header = skb2->network_header;
1893 skb2->pkt_type = PACKET_OUTGOING;
1897 if (ptype_list == &ptype_all) {
1898 ptype_list = &dev->ptype_all;
1903 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1906 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
1909 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1910 * @dev: Network device
1911 * @txq: number of queues available
1913 * If real_num_tx_queues is changed the tc mappings may no longer be
1914 * valid. To resolve this verify the tc mapping remains valid and if
1915 * not NULL the mapping. With no priorities mapping to this
1916 * offset/count pair it will no longer be used. In the worst case TC0
1917 * is invalid nothing can be done so disable priority mappings. If is
1918 * expected that drivers will fix this mapping if they can before
1919 * calling netif_set_real_num_tx_queues.
1921 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1924 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1926 /* If TC0 is invalidated disable TC mapping */
1927 if (tc->offset + tc->count > txq) {
1928 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1933 /* Invalidated prio to tc mappings set to TC0 */
1934 for (i = 1; i < TC_BITMASK + 1; i++) {
1935 int q = netdev_get_prio_tc_map(dev, i);
1937 tc = &dev->tc_to_txq[q];
1938 if (tc->offset + tc->count > txq) {
1939 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1941 netdev_set_prio_tc_map(dev, i, 0);
1946 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
1949 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1952 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
1953 if ((txq - tc->offset) < tc->count)
1964 static DEFINE_MUTEX(xps_map_mutex);
1965 #define xmap_dereference(P) \
1966 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1968 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
1971 struct xps_map *map = NULL;
1975 map = xmap_dereference(dev_maps->cpu_map[tci]);
1979 for (pos = map->len; pos--;) {
1980 if (map->queues[pos] != index)
1984 map->queues[pos] = map->queues[--map->len];
1988 RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL);
1989 kfree_rcu(map, rcu);
1996 static bool remove_xps_queue_cpu(struct net_device *dev,
1997 struct xps_dev_maps *dev_maps,
1998 int cpu, u16 offset, u16 count)
2000 int num_tc = dev->num_tc ? : 1;
2001 bool active = false;
2004 for (tci = cpu * num_tc; num_tc--; tci++) {
2007 for (i = count, j = offset; i--; j++) {
2008 if (!remove_xps_queue(dev_maps, cpu, j))
2018 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2021 struct xps_dev_maps *dev_maps;
2023 bool active = false;
2025 mutex_lock(&xps_map_mutex);
2026 dev_maps = xmap_dereference(dev->xps_maps);
2031 for_each_possible_cpu(cpu)
2032 active |= remove_xps_queue_cpu(dev, dev_maps, cpu,
2036 RCU_INIT_POINTER(dev->xps_maps, NULL);
2037 kfree_rcu(dev_maps, rcu);
2040 for (i = offset + (count - 1); count--; i--)
2041 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2045 mutex_unlock(&xps_map_mutex);
2048 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2050 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2053 static struct xps_map *expand_xps_map(struct xps_map *map,
2056 struct xps_map *new_map;
2057 int alloc_len = XPS_MIN_MAP_ALLOC;
2060 for (pos = 0; map && pos < map->len; pos++) {
2061 if (map->queues[pos] != index)
2066 /* Need to add queue to this CPU's existing map */
2068 if (pos < map->alloc_len)
2071 alloc_len = map->alloc_len * 2;
2074 /* Need to allocate new map to store queue on this CPU's map */
2075 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2080 for (i = 0; i < pos; i++)
2081 new_map->queues[i] = map->queues[i];
2082 new_map->alloc_len = alloc_len;
2088 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2091 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2092 int i, cpu, tci, numa_node_id = -2;
2093 int maps_sz, num_tc = 1, tc = 0;
2094 struct xps_map *map, *new_map;
2095 bool active = false;
2098 num_tc = dev->num_tc;
2099 tc = netdev_txq_to_tc(dev, index);
2104 maps_sz = XPS_DEV_MAPS_SIZE(num_tc);
2105 if (maps_sz < L1_CACHE_BYTES)
2106 maps_sz = L1_CACHE_BYTES;
2108 mutex_lock(&xps_map_mutex);
2110 dev_maps = xmap_dereference(dev->xps_maps);
2112 /* allocate memory for queue storage */
2113 for_each_cpu_and(cpu, cpu_online_mask, mask) {
2115 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2116 if (!new_dev_maps) {
2117 mutex_unlock(&xps_map_mutex);
2121 tci = cpu * num_tc + tc;
2122 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) :
2125 map = expand_xps_map(map, cpu, index);
2129 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2133 goto out_no_new_maps;
2135 for_each_possible_cpu(cpu) {
2136 /* copy maps belonging to foreign traffic classes */
2137 for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) {
2138 /* fill in the new device map from the old device map */
2139 map = xmap_dereference(dev_maps->cpu_map[tci]);
2140 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2143 /* We need to explicitly update tci as prevous loop
2144 * could break out early if dev_maps is NULL.
2146 tci = cpu * num_tc + tc;
2148 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2149 /* add queue to CPU maps */
2152 map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2153 while ((pos < map->len) && (map->queues[pos] != index))
2156 if (pos == map->len)
2157 map->queues[map->len++] = index;
2159 if (numa_node_id == -2)
2160 numa_node_id = cpu_to_node(cpu);
2161 else if (numa_node_id != cpu_to_node(cpu))
2164 } else if (dev_maps) {
2165 /* fill in the new device map from the old device map */
2166 map = xmap_dereference(dev_maps->cpu_map[tci]);
2167 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2170 /* copy maps belonging to foreign traffic classes */
2171 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2172 /* fill in the new device map from the old device map */
2173 map = xmap_dereference(dev_maps->cpu_map[tci]);
2174 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2178 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2180 /* Cleanup old maps */
2182 goto out_no_old_maps;
2184 for_each_possible_cpu(cpu) {
2185 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2186 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2187 map = xmap_dereference(dev_maps->cpu_map[tci]);
2188 if (map && map != new_map)
2189 kfree_rcu(map, rcu);
2193 kfree_rcu(dev_maps, rcu);
2196 dev_maps = new_dev_maps;
2200 /* update Tx queue numa node */
2201 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2202 (numa_node_id >= 0) ? numa_node_id :
2208 /* removes queue from unused CPUs */
2209 for_each_possible_cpu(cpu) {
2210 for (i = tc, tci = cpu * num_tc; i--; tci++)
2211 active |= remove_xps_queue(dev_maps, tci, index);
2212 if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu))
2213 active |= remove_xps_queue(dev_maps, tci, index);
2214 for (i = num_tc - tc, tci++; --i; tci++)
2215 active |= remove_xps_queue(dev_maps, tci, index);
2218 /* free map if not active */
2220 RCU_INIT_POINTER(dev->xps_maps, NULL);
2221 kfree_rcu(dev_maps, rcu);
2225 mutex_unlock(&xps_map_mutex);
2229 /* remove any maps that we added */
2230 for_each_possible_cpu(cpu) {
2231 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2232 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2234 xmap_dereference(dev_maps->cpu_map[tci]) :
2236 if (new_map && new_map != map)
2241 mutex_unlock(&xps_map_mutex);
2243 kfree(new_dev_maps);
2246 EXPORT_SYMBOL(netif_set_xps_queue);
2249 void netdev_reset_tc(struct net_device *dev)
2252 netif_reset_xps_queues_gt(dev, 0);
2255 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2256 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2258 EXPORT_SYMBOL(netdev_reset_tc);
2260 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2262 if (tc >= dev->num_tc)
2266 netif_reset_xps_queues(dev, offset, count);
2268 dev->tc_to_txq[tc].count = count;
2269 dev->tc_to_txq[tc].offset = offset;
2272 EXPORT_SYMBOL(netdev_set_tc_queue);
2274 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2276 if (num_tc > TC_MAX_QUEUE)
2280 netif_reset_xps_queues_gt(dev, 0);
2282 dev->num_tc = num_tc;
2285 EXPORT_SYMBOL(netdev_set_num_tc);
2288 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2289 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2291 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2295 if (txq < 1 || txq > dev->num_tx_queues)
2298 if (dev->reg_state == NETREG_REGISTERED ||
2299 dev->reg_state == NETREG_UNREGISTERING) {
2302 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2308 netif_setup_tc(dev, txq);
2310 if (txq < dev->real_num_tx_queues) {
2311 qdisc_reset_all_tx_gt(dev, txq);
2313 netif_reset_xps_queues_gt(dev, txq);
2318 dev->real_num_tx_queues = txq;
2321 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2325 * netif_set_real_num_rx_queues - set actual number of RX queues used
2326 * @dev: Network device
2327 * @rxq: Actual number of RX queues
2329 * This must be called either with the rtnl_lock held or before
2330 * registration of the net device. Returns 0 on success, or a
2331 * negative error code. If called before registration, it always
2334 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2338 if (rxq < 1 || rxq > dev->num_rx_queues)
2341 if (dev->reg_state == NETREG_REGISTERED) {
2344 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2350 dev->real_num_rx_queues = rxq;
2353 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2357 * netif_get_num_default_rss_queues - default number of RSS queues
2359 * This routine should set an upper limit on the number of RSS queues
2360 * used by default by multiqueue devices.
2362 int netif_get_num_default_rss_queues(void)
2364 return is_kdump_kernel() ?
2365 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2367 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2369 static void __netif_reschedule(struct Qdisc *q)
2371 struct softnet_data *sd;
2372 unsigned long flags;
2374 local_irq_save(flags);
2375 sd = this_cpu_ptr(&softnet_data);
2376 q->next_sched = NULL;
2377 *sd->output_queue_tailp = q;
2378 sd->output_queue_tailp = &q->next_sched;
2379 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2380 local_irq_restore(flags);
2383 void __netif_schedule(struct Qdisc *q)
2385 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2386 __netif_reschedule(q);
2388 EXPORT_SYMBOL(__netif_schedule);
2390 struct dev_kfree_skb_cb {
2391 enum skb_free_reason reason;
2394 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2396 return (struct dev_kfree_skb_cb *)skb->cb;
2399 void netif_schedule_queue(struct netdev_queue *txq)
2402 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2403 struct Qdisc *q = rcu_dereference(txq->qdisc);
2405 __netif_schedule(q);
2409 EXPORT_SYMBOL(netif_schedule_queue);
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);
3450 struct static_key rfs_needed __read_mostly;
3451 EXPORT_SYMBOL(rfs_needed);
3453 static struct rps_dev_flow *
3454 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3455 struct rps_dev_flow *rflow, u16 next_cpu)
3457 if (next_cpu < nr_cpu_ids) {
3458 #ifdef CONFIG_RFS_ACCEL
3459 struct netdev_rx_queue *rxqueue;
3460 struct rps_dev_flow_table *flow_table;
3461 struct rps_dev_flow *old_rflow;
3466 /* Should we steer this flow to a different hardware queue? */
3467 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3468 !(dev->features & NETIF_F_NTUPLE))
3470 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3471 if (rxq_index == skb_get_rx_queue(skb))
3474 rxqueue = dev->_rx + rxq_index;
3475 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3478 flow_id = skb_get_hash(skb) & flow_table->mask;
3479 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3480 rxq_index, flow_id);
3484 rflow = &flow_table->flows[flow_id];
3486 if (old_rflow->filter == rflow->filter)
3487 old_rflow->filter = RPS_NO_FILTER;
3491 per_cpu(softnet_data, next_cpu).input_queue_head;
3494 rflow->cpu = next_cpu;
3499 * get_rps_cpu is called from netif_receive_skb and returns the target
3500 * CPU from the RPS map of the receiving queue for a given skb.
3501 * rcu_read_lock must be held on entry.
3503 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3504 struct rps_dev_flow **rflowp)
3506 const struct rps_sock_flow_table *sock_flow_table;
3507 struct netdev_rx_queue *rxqueue = dev->_rx;
3508 struct rps_dev_flow_table *flow_table;
3509 struct rps_map *map;
3514 if (skb_rx_queue_recorded(skb)) {
3515 u16 index = skb_get_rx_queue(skb);
3517 if (unlikely(index >= dev->real_num_rx_queues)) {
3518 WARN_ONCE(dev->real_num_rx_queues > 1,
3519 "%s received packet on queue %u, but number "
3520 "of RX queues is %u\n",
3521 dev->name, index, dev->real_num_rx_queues);
3527 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3529 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3530 map = rcu_dereference(rxqueue->rps_map);
3531 if (!flow_table && !map)
3534 skb_reset_network_header(skb);
3535 hash = skb_get_hash(skb);
3539 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3540 if (flow_table && sock_flow_table) {
3541 struct rps_dev_flow *rflow;
3545 /* First check into global flow table if there is a match */
3546 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3547 if ((ident ^ hash) & ~rps_cpu_mask)
3550 next_cpu = ident & rps_cpu_mask;
3552 /* OK, now we know there is a match,
3553 * we can look at the local (per receive queue) flow table
3555 rflow = &flow_table->flows[hash & flow_table->mask];
3559 * If the desired CPU (where last recvmsg was done) is
3560 * different from current CPU (one in the rx-queue flow
3561 * table entry), switch if one of the following holds:
3562 * - Current CPU is unset (>= nr_cpu_ids).
3563 * - Current CPU is offline.
3564 * - The current CPU's queue tail has advanced beyond the
3565 * last packet that was enqueued using this table entry.
3566 * This guarantees that all previous packets for the flow
3567 * have been dequeued, thus preserving in order delivery.
3569 if (unlikely(tcpu != next_cpu) &&
3570 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3571 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3572 rflow->last_qtail)) >= 0)) {
3574 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3577 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3587 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3588 if (cpu_online(tcpu)) {
3598 #ifdef CONFIG_RFS_ACCEL
3601 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3602 * @dev: Device on which the filter was set
3603 * @rxq_index: RX queue index
3604 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3605 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3607 * Drivers that implement ndo_rx_flow_steer() should periodically call
3608 * this function for each installed filter and remove the filters for
3609 * which it returns %true.
3611 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3612 u32 flow_id, u16 filter_id)
3614 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3615 struct rps_dev_flow_table *flow_table;
3616 struct rps_dev_flow *rflow;
3621 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3622 if (flow_table && flow_id <= flow_table->mask) {
3623 rflow = &flow_table->flows[flow_id];
3624 cpu = ACCESS_ONCE(rflow->cpu);
3625 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3626 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3627 rflow->last_qtail) <
3628 (int)(10 * flow_table->mask)))
3634 EXPORT_SYMBOL(rps_may_expire_flow);
3636 #endif /* CONFIG_RFS_ACCEL */
3638 /* Called from hardirq (IPI) context */
3639 static void rps_trigger_softirq(void *data)
3641 struct softnet_data *sd = data;
3643 ____napi_schedule(sd, &sd->backlog);
3647 #endif /* CONFIG_RPS */
3650 * Check if this softnet_data structure is another cpu one
3651 * If yes, queue it to our IPI list and return 1
3654 static int rps_ipi_queued(struct softnet_data *sd)
3657 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3660 sd->rps_ipi_next = mysd->rps_ipi_list;
3661 mysd->rps_ipi_list = sd;
3663 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3666 #endif /* CONFIG_RPS */
3670 #ifdef CONFIG_NET_FLOW_LIMIT
3671 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3674 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3676 #ifdef CONFIG_NET_FLOW_LIMIT
3677 struct sd_flow_limit *fl;
3678 struct softnet_data *sd;
3679 unsigned int old_flow, new_flow;
3681 if (qlen < (netdev_max_backlog >> 1))
3684 sd = this_cpu_ptr(&softnet_data);
3687 fl = rcu_dereference(sd->flow_limit);
3689 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3690 old_flow = fl->history[fl->history_head];
3691 fl->history[fl->history_head] = new_flow;
3694 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3696 if (likely(fl->buckets[old_flow]))
3697 fl->buckets[old_flow]--;
3699 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3711 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3712 * queue (may be a remote CPU queue).
3714 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3715 unsigned int *qtail)
3717 struct softnet_data *sd;
3718 unsigned long flags;
3721 sd = &per_cpu(softnet_data, cpu);
3723 local_irq_save(flags);
3726 if (!netif_running(skb->dev))
3728 qlen = skb_queue_len(&sd->input_pkt_queue);
3729 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3732 __skb_queue_tail(&sd->input_pkt_queue, skb);
3733 input_queue_tail_incr_save(sd, qtail);
3735 local_irq_restore(flags);
3736 return NET_RX_SUCCESS;
3739 /* Schedule NAPI for backlog device
3740 * We can use non atomic operation since we own the queue lock
3742 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3743 if (!rps_ipi_queued(sd))
3744 ____napi_schedule(sd, &sd->backlog);
3753 local_irq_restore(flags);
3755 atomic_long_inc(&skb->dev->rx_dropped);
3760 static int netif_rx_internal(struct sk_buff *skb)
3764 net_timestamp_check(netdev_tstamp_prequeue, skb);
3766 trace_netif_rx(skb);
3768 if (static_key_false(&rps_needed)) {
3769 struct rps_dev_flow voidflow, *rflow = &voidflow;
3775 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3777 cpu = smp_processor_id();
3779 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3787 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3794 * netif_rx - post buffer to the network code
3795 * @skb: buffer to post
3797 * This function receives a packet from a device driver and queues it for
3798 * the upper (protocol) levels to process. It always succeeds. The buffer
3799 * may be dropped during processing for congestion control or by the
3803 * NET_RX_SUCCESS (no congestion)
3804 * NET_RX_DROP (packet was dropped)
3808 int netif_rx(struct sk_buff *skb)
3810 trace_netif_rx_entry(skb);
3812 return netif_rx_internal(skb);
3814 EXPORT_SYMBOL(netif_rx);
3816 int netif_rx_ni(struct sk_buff *skb)
3820 trace_netif_rx_ni_entry(skb);
3823 err = netif_rx_internal(skb);
3824 if (local_softirq_pending())
3830 EXPORT_SYMBOL(netif_rx_ni);
3832 static __latent_entropy void net_tx_action(struct softirq_action *h)
3834 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3836 if (sd->completion_queue) {
3837 struct sk_buff *clist;
3839 local_irq_disable();
3840 clist = sd->completion_queue;
3841 sd->completion_queue = NULL;
3845 struct sk_buff *skb = clist;
3846 clist = clist->next;
3848 WARN_ON(atomic_read(&skb->users));
3849 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3850 trace_consume_skb(skb);
3852 trace_kfree_skb(skb, net_tx_action);
3854 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
3857 __kfree_skb_defer(skb);
3860 __kfree_skb_flush();
3863 if (sd->output_queue) {
3866 local_irq_disable();
3867 head = sd->output_queue;
3868 sd->output_queue = NULL;
3869 sd->output_queue_tailp = &sd->output_queue;
3873 struct Qdisc *q = head;
3874 spinlock_t *root_lock;
3876 head = head->next_sched;
3878 root_lock = qdisc_lock(q);
3879 spin_lock(root_lock);
3880 /* We need to make sure head->next_sched is read
3881 * before clearing __QDISC_STATE_SCHED
3883 smp_mb__before_atomic();
3884 clear_bit(__QDISC_STATE_SCHED, &q->state);
3886 spin_unlock(root_lock);
3891 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
3892 /* This hook is defined here for ATM LANE */
3893 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3894 unsigned char *addr) __read_mostly;
3895 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3898 static inline struct sk_buff *
3899 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
3900 struct net_device *orig_dev)
3902 #ifdef CONFIG_NET_CLS_ACT
3903 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3904 struct tcf_result cl_res;
3906 /* If there's at least one ingress present somewhere (so
3907 * we get here via enabled static key), remaining devices
3908 * that are not configured with an ingress qdisc will bail
3914 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3918 qdisc_skb_cb(skb)->pkt_len = skb->len;
3919 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3920 qdisc_bstats_cpu_update(cl->q, skb);
3922 switch (tc_classify(skb, cl, &cl_res, false)) {
3924 case TC_ACT_RECLASSIFY:
3925 skb->tc_index = TC_H_MIN(cl_res.classid);
3928 qdisc_qstats_cpu_drop(cl->q);
3935 case TC_ACT_REDIRECT:
3936 /* skb_mac_header check was done by cls/act_bpf, so
3937 * we can safely push the L2 header back before
3938 * redirecting to another netdev
3940 __skb_push(skb, skb->mac_len);
3941 skb_do_redirect(skb);
3946 #endif /* CONFIG_NET_CLS_ACT */
3951 * netdev_is_rx_handler_busy - check if receive handler is registered
3952 * @dev: device to check
3954 * Check if a receive handler is already registered for a given device.
3955 * Return true if there one.
3957 * The caller must hold the rtnl_mutex.
3959 bool netdev_is_rx_handler_busy(struct net_device *dev)
3962 return dev && rtnl_dereference(dev->rx_handler);
3964 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
3967 * netdev_rx_handler_register - register receive handler
3968 * @dev: device to register a handler for
3969 * @rx_handler: receive handler to register
3970 * @rx_handler_data: data pointer that is used by rx handler
3972 * Register a receive handler for a device. This handler will then be
3973 * called from __netif_receive_skb. A negative errno code is returned
3976 * The caller must hold the rtnl_mutex.
3978 * For a general description of rx_handler, see enum rx_handler_result.
3980 int netdev_rx_handler_register(struct net_device *dev,
3981 rx_handler_func_t *rx_handler,
3982 void *rx_handler_data)
3986 if (dev->rx_handler)
3989 /* Note: rx_handler_data must be set before rx_handler */
3990 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3991 rcu_assign_pointer(dev->rx_handler, rx_handler);
3995 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3998 * netdev_rx_handler_unregister - unregister receive handler
3999 * @dev: device to unregister a handler from
4001 * Unregister a receive handler from a device.
4003 * The caller must hold the rtnl_mutex.
4005 void netdev_rx_handler_unregister(struct net_device *dev)
4009 RCU_INIT_POINTER(dev->rx_handler, NULL);
4010 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4011 * section has a guarantee to see a non NULL rx_handler_data
4015 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4017 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4020 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4021 * the special handling of PFMEMALLOC skbs.
4023 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4025 switch (skb->protocol) {
4026 case htons(ETH_P_ARP):
4027 case htons(ETH_P_IP):
4028 case htons(ETH_P_IPV6):
4029 case htons(ETH_P_8021Q):
4030 case htons(ETH_P_8021AD):
4037 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4038 int *ret, struct net_device *orig_dev)
4040 #ifdef CONFIG_NETFILTER_INGRESS
4041 if (nf_hook_ingress_active(skb)) {
4045 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4050 ingress_retval = nf_hook_ingress(skb);
4052 return ingress_retval;
4054 #endif /* CONFIG_NETFILTER_INGRESS */
4058 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
4060 struct packet_type *ptype, *pt_prev;
4061 rx_handler_func_t *rx_handler;
4062 struct net_device *orig_dev;
4063 bool deliver_exact = false;
4064 int ret = NET_RX_DROP;
4067 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4069 trace_netif_receive_skb(skb);
4071 orig_dev = skb->dev;
4073 skb_reset_network_header(skb);
4074 if (!skb_transport_header_was_set(skb))
4075 skb_reset_transport_header(skb);
4076 skb_reset_mac_len(skb);
4081 skb->skb_iif = skb->dev->ifindex;
4083 __this_cpu_inc(softnet_data.processed);
4085 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4086 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4087 skb = skb_vlan_untag(skb);
4092 #ifdef CONFIG_NET_CLS_ACT
4093 if (skb->tc_verd & TC_NCLS) {
4094 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
4102 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4104 ret = deliver_skb(skb, pt_prev, orig_dev);
4108 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4110 ret = deliver_skb(skb, pt_prev, orig_dev);
4115 #ifdef CONFIG_NET_INGRESS
4116 if (static_key_false(&ingress_needed)) {
4117 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4121 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4125 #ifdef CONFIG_NET_CLS_ACT
4129 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4132 if (skb_vlan_tag_present(skb)) {
4134 ret = deliver_skb(skb, pt_prev, orig_dev);
4137 if (vlan_do_receive(&skb))
4139 else if (unlikely(!skb))
4143 rx_handler = rcu_dereference(skb->dev->rx_handler);
4146 ret = deliver_skb(skb, pt_prev, orig_dev);
4149 switch (rx_handler(&skb)) {
4150 case RX_HANDLER_CONSUMED:
4151 ret = NET_RX_SUCCESS;
4153 case RX_HANDLER_ANOTHER:
4155 case RX_HANDLER_EXACT:
4156 deliver_exact = true;
4157 case RX_HANDLER_PASS:
4164 if (unlikely(skb_vlan_tag_present(skb))) {
4165 if (skb_vlan_tag_get_id(skb))
4166 skb->pkt_type = PACKET_OTHERHOST;
4167 /* Note: we might in the future use prio bits
4168 * and set skb->priority like in vlan_do_receive()
4169 * For the time being, just ignore Priority Code Point
4174 type = skb->protocol;
4176 /* deliver only exact match when indicated */
4177 if (likely(!deliver_exact)) {
4178 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4179 &ptype_base[ntohs(type) &
4183 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4184 &orig_dev->ptype_specific);
4186 if (unlikely(skb->dev != orig_dev)) {
4187 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4188 &skb->dev->ptype_specific);
4192 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
4195 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4199 atomic_long_inc(&skb->dev->rx_dropped);
4201 atomic_long_inc(&skb->dev->rx_nohandler);
4203 /* Jamal, now you will not able to escape explaining
4204 * me how you were going to use this. :-)
4213 static int __netif_receive_skb(struct sk_buff *skb)
4217 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4218 unsigned long pflags = current->flags;
4221 * PFMEMALLOC skbs are special, they should
4222 * - be delivered to SOCK_MEMALLOC sockets only
4223 * - stay away from userspace
4224 * - have bounded memory usage
4226 * Use PF_MEMALLOC as this saves us from propagating the allocation
4227 * context down to all allocation sites.
4229 current->flags |= PF_MEMALLOC;
4230 ret = __netif_receive_skb_core(skb, true);
4231 tsk_restore_flags(current, pflags, PF_MEMALLOC);
4233 ret = __netif_receive_skb_core(skb, false);
4238 static int netif_receive_skb_internal(struct sk_buff *skb)
4242 net_timestamp_check(netdev_tstamp_prequeue, skb);
4244 if (skb_defer_rx_timestamp(skb))
4245 return NET_RX_SUCCESS;
4250 if (static_key_false(&rps_needed)) {
4251 struct rps_dev_flow voidflow, *rflow = &voidflow;
4252 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4255 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4261 ret = __netif_receive_skb(skb);
4267 * netif_receive_skb - process receive buffer from network
4268 * @skb: buffer to process
4270 * netif_receive_skb() is the main receive data processing function.
4271 * It always succeeds. The buffer may be dropped during processing
4272 * for congestion control or by the protocol layers.
4274 * This function may only be called from softirq context and interrupts
4275 * should be enabled.
4277 * Return values (usually ignored):
4278 * NET_RX_SUCCESS: no congestion
4279 * NET_RX_DROP: packet was dropped
4281 int netif_receive_skb(struct sk_buff *skb)
4283 trace_netif_receive_skb_entry(skb);
4285 return netif_receive_skb_internal(skb);
4287 EXPORT_SYMBOL(netif_receive_skb);
4289 DEFINE_PER_CPU(struct work_struct, flush_works);
4291 /* Network device is going away, flush any packets still pending */
4292 static void flush_backlog(struct work_struct *work)
4294 struct sk_buff *skb, *tmp;
4295 struct softnet_data *sd;
4298 sd = this_cpu_ptr(&softnet_data);
4300 local_irq_disable();
4302 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4303 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4304 __skb_unlink(skb, &sd->input_pkt_queue);
4306 input_queue_head_incr(sd);
4312 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4313 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4314 __skb_unlink(skb, &sd->process_queue);
4316 input_queue_head_incr(sd);
4322 static void flush_all_backlogs(void)
4328 for_each_online_cpu(cpu)
4329 queue_work_on(cpu, system_highpri_wq,
4330 per_cpu_ptr(&flush_works, cpu));
4332 for_each_online_cpu(cpu)
4333 flush_work(per_cpu_ptr(&flush_works, cpu));
4338 static int napi_gro_complete(struct sk_buff *skb)
4340 struct packet_offload *ptype;
4341 __be16 type = skb->protocol;
4342 struct list_head *head = &offload_base;
4345 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4347 if (NAPI_GRO_CB(skb)->count == 1) {
4348 skb_shinfo(skb)->gso_size = 0;
4353 list_for_each_entry_rcu(ptype, head, list) {
4354 if (ptype->type != type || !ptype->callbacks.gro_complete)
4357 err = ptype->callbacks.gro_complete(skb, 0);
4363 WARN_ON(&ptype->list == head);
4365 return NET_RX_SUCCESS;
4369 return netif_receive_skb_internal(skb);
4372 /* napi->gro_list contains packets ordered by age.
4373 * youngest packets at the head of it.
4374 * Complete skbs in reverse order to reduce latencies.
4376 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4378 struct sk_buff *skb, *prev = NULL;
4380 /* scan list and build reverse chain */
4381 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4386 for (skb = prev; skb; skb = prev) {
4389 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4393 napi_gro_complete(skb);
4397 napi->gro_list = NULL;
4399 EXPORT_SYMBOL(napi_gro_flush);
4401 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4404 unsigned int maclen = skb->dev->hard_header_len;
4405 u32 hash = skb_get_hash_raw(skb);
4407 for (p = napi->gro_list; p; p = p->next) {
4408 unsigned long diffs;
4410 NAPI_GRO_CB(p)->flush = 0;
4412 if (hash != skb_get_hash_raw(p)) {
4413 NAPI_GRO_CB(p)->same_flow = 0;
4417 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4418 diffs |= p->vlan_tci ^ skb->vlan_tci;
4419 diffs |= skb_metadata_dst_cmp(p, skb);
4420 if (maclen == ETH_HLEN)
4421 diffs |= compare_ether_header(skb_mac_header(p),
4422 skb_mac_header(skb));
4424 diffs = memcmp(skb_mac_header(p),
4425 skb_mac_header(skb),
4427 NAPI_GRO_CB(p)->same_flow = !diffs;
4431 static void skb_gro_reset_offset(struct sk_buff *skb)
4433 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4434 const skb_frag_t *frag0 = &pinfo->frags[0];
4436 NAPI_GRO_CB(skb)->data_offset = 0;
4437 NAPI_GRO_CB(skb)->frag0 = NULL;
4438 NAPI_GRO_CB(skb)->frag0_len = 0;
4440 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4442 !PageHighMem(skb_frag_page(frag0))) {
4443 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4444 NAPI_GRO_CB(skb)->frag0_len = min(skb_frag_size(frag0),
4445 skb->end - skb->tail);
4449 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4451 struct skb_shared_info *pinfo = skb_shinfo(skb);
4453 BUG_ON(skb->end - skb->tail < grow);
4455 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4457 skb->data_len -= grow;
4460 pinfo->frags[0].page_offset += grow;
4461 skb_frag_size_sub(&pinfo->frags[0], grow);
4463 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4464 skb_frag_unref(skb, 0);
4465 memmove(pinfo->frags, pinfo->frags + 1,
4466 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4470 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4472 struct sk_buff **pp = NULL;
4473 struct packet_offload *ptype;
4474 __be16 type = skb->protocol;
4475 struct list_head *head = &offload_base;
4477 enum gro_result ret;
4480 if (!(skb->dev->features & NETIF_F_GRO))
4486 gro_list_prepare(napi, skb);
4489 list_for_each_entry_rcu(ptype, head, list) {
4490 if (ptype->type != type || !ptype->callbacks.gro_receive)
4493 skb_set_network_header(skb, skb_gro_offset(skb));
4494 skb_reset_mac_len(skb);
4495 NAPI_GRO_CB(skb)->same_flow = 0;
4496 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
4497 NAPI_GRO_CB(skb)->free = 0;
4498 NAPI_GRO_CB(skb)->encap_mark = 0;
4499 NAPI_GRO_CB(skb)->recursion_counter = 0;
4500 NAPI_GRO_CB(skb)->is_fou = 0;
4501 NAPI_GRO_CB(skb)->is_atomic = 1;
4502 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4504 /* Setup for GRO checksum validation */
4505 switch (skb->ip_summed) {
4506 case CHECKSUM_COMPLETE:
4507 NAPI_GRO_CB(skb)->csum = skb->csum;
4508 NAPI_GRO_CB(skb)->csum_valid = 1;
4509 NAPI_GRO_CB(skb)->csum_cnt = 0;
4511 case CHECKSUM_UNNECESSARY:
4512 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4513 NAPI_GRO_CB(skb)->csum_valid = 0;
4516 NAPI_GRO_CB(skb)->csum_cnt = 0;
4517 NAPI_GRO_CB(skb)->csum_valid = 0;
4520 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4525 if (&ptype->list == head)
4528 same_flow = NAPI_GRO_CB(skb)->same_flow;
4529 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4532 struct sk_buff *nskb = *pp;
4536 napi_gro_complete(nskb);
4543 if (NAPI_GRO_CB(skb)->flush)
4546 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4547 struct sk_buff *nskb = napi->gro_list;
4549 /* locate the end of the list to select the 'oldest' flow */
4550 while (nskb->next) {
4556 napi_gro_complete(nskb);
4560 NAPI_GRO_CB(skb)->count = 1;
4561 NAPI_GRO_CB(skb)->age = jiffies;
4562 NAPI_GRO_CB(skb)->last = skb;
4563 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4564 skb->next = napi->gro_list;
4565 napi->gro_list = skb;
4569 grow = skb_gro_offset(skb) - skb_headlen(skb);
4571 gro_pull_from_frag0(skb, grow);
4580 struct packet_offload *gro_find_receive_by_type(__be16 type)
4582 struct list_head *offload_head = &offload_base;
4583 struct packet_offload *ptype;
4585 list_for_each_entry_rcu(ptype, offload_head, list) {
4586 if (ptype->type != type || !ptype->callbacks.gro_receive)
4592 EXPORT_SYMBOL(gro_find_receive_by_type);
4594 struct packet_offload *gro_find_complete_by_type(__be16 type)
4596 struct list_head *offload_head = &offload_base;
4597 struct packet_offload *ptype;
4599 list_for_each_entry_rcu(ptype, offload_head, list) {
4600 if (ptype->type != type || !ptype->callbacks.gro_complete)
4606 EXPORT_SYMBOL(gro_find_complete_by_type);
4608 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4612 if (netif_receive_skb_internal(skb))
4620 case GRO_MERGED_FREE:
4621 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) {
4623 kmem_cache_free(skbuff_head_cache, skb);
4637 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4639 skb_mark_napi_id(skb, napi);
4640 trace_napi_gro_receive_entry(skb);
4642 skb_gro_reset_offset(skb);
4644 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4646 EXPORT_SYMBOL(napi_gro_receive);
4648 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4650 if (unlikely(skb->pfmemalloc)) {
4654 __skb_pull(skb, skb_headlen(skb));
4655 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4656 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4658 skb->dev = napi->dev;
4660 skb->encapsulation = 0;
4661 skb_shinfo(skb)->gso_type = 0;
4662 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4667 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4669 struct sk_buff *skb = napi->skb;
4672 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4675 skb_mark_napi_id(skb, napi);
4680 EXPORT_SYMBOL(napi_get_frags);
4682 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4683 struct sk_buff *skb,
4689 __skb_push(skb, ETH_HLEN);
4690 skb->protocol = eth_type_trans(skb, skb->dev);
4691 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4696 case GRO_MERGED_FREE:
4697 napi_reuse_skb(napi, skb);
4707 /* Upper GRO stack assumes network header starts at gro_offset=0
4708 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4709 * We copy ethernet header into skb->data to have a common layout.
4711 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4713 struct sk_buff *skb = napi->skb;
4714 const struct ethhdr *eth;
4715 unsigned int hlen = sizeof(*eth);
4719 skb_reset_mac_header(skb);
4720 skb_gro_reset_offset(skb);
4722 eth = skb_gro_header_fast(skb, 0);
4723 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4724 eth = skb_gro_header_slow(skb, hlen, 0);
4725 if (unlikely(!eth)) {
4726 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
4727 __func__, napi->dev->name);
4728 napi_reuse_skb(napi, skb);
4732 gro_pull_from_frag0(skb, hlen);
4733 NAPI_GRO_CB(skb)->frag0 += hlen;
4734 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4736 __skb_pull(skb, hlen);
4739 * This works because the only protocols we care about don't require
4741 * We'll fix it up properly in napi_frags_finish()
4743 skb->protocol = eth->h_proto;
4748 gro_result_t napi_gro_frags(struct napi_struct *napi)
4750 struct sk_buff *skb = napi_frags_skb(napi);
4755 trace_napi_gro_frags_entry(skb);
4757 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4759 EXPORT_SYMBOL(napi_gro_frags);
4761 /* Compute the checksum from gro_offset and return the folded value
4762 * after adding in any pseudo checksum.
4764 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4769 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4771 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4772 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4774 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4775 !skb->csum_complete_sw)
4776 netdev_rx_csum_fault(skb->dev);
4779 NAPI_GRO_CB(skb)->csum = wsum;
4780 NAPI_GRO_CB(skb)->csum_valid = 1;
4784 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4787 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4788 * Note: called with local irq disabled, but exits with local irq enabled.
4790 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4793 struct softnet_data *remsd = sd->rps_ipi_list;
4796 sd->rps_ipi_list = NULL;
4800 /* Send pending IPI's to kick RPS processing on remote cpus. */
4802 struct softnet_data *next = remsd->rps_ipi_next;
4804 if (cpu_online(remsd->cpu))
4805 smp_call_function_single_async(remsd->cpu,
4814 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4817 return sd->rps_ipi_list != NULL;
4823 static int process_backlog(struct napi_struct *napi, int quota)
4825 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4829 /* Check if we have pending ipi, its better to send them now,
4830 * not waiting net_rx_action() end.
4832 if (sd_has_rps_ipi_waiting(sd)) {
4833 local_irq_disable();
4834 net_rps_action_and_irq_enable(sd);
4837 napi->weight = weight_p;
4839 struct sk_buff *skb;
4841 while ((skb = __skb_dequeue(&sd->process_queue))) {
4843 __netif_receive_skb(skb);
4845 input_queue_head_incr(sd);
4846 if (++work >= quota)
4851 local_irq_disable();
4853 if (skb_queue_empty(&sd->input_pkt_queue)) {
4855 * Inline a custom version of __napi_complete().
4856 * only current cpu owns and manipulates this napi,
4857 * and NAPI_STATE_SCHED is the only possible flag set
4859 * We can use a plain write instead of clear_bit(),
4860 * and we dont need an smp_mb() memory barrier.
4865 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4866 &sd->process_queue);
4876 * __napi_schedule - schedule for receive
4877 * @n: entry to schedule
4879 * The entry's receive function will be scheduled to run.
4880 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4882 void __napi_schedule(struct napi_struct *n)
4884 unsigned long flags;
4886 local_irq_save(flags);
4887 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4888 local_irq_restore(flags);
4890 EXPORT_SYMBOL(__napi_schedule);
4893 * __napi_schedule_irqoff - schedule for receive
4894 * @n: entry to schedule
4896 * Variant of __napi_schedule() assuming hard irqs are masked
4898 void __napi_schedule_irqoff(struct napi_struct *n)
4900 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4902 EXPORT_SYMBOL(__napi_schedule_irqoff);
4904 bool __napi_complete(struct napi_struct *n)
4906 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4908 /* Some drivers call us directly, instead of calling
4909 * napi_complete_done().
4911 if (unlikely(test_bit(NAPI_STATE_IN_BUSY_POLL, &n->state)))
4914 list_del_init(&n->poll_list);
4915 smp_mb__before_atomic();
4916 clear_bit(NAPI_STATE_SCHED, &n->state);
4919 EXPORT_SYMBOL(__napi_complete);
4921 bool napi_complete_done(struct napi_struct *n, int work_done)
4923 unsigned long flags;
4926 * 1) Don't let napi dequeue from the cpu poll list
4927 * just in case its running on a different cpu.
4928 * 2) If we are busy polling, do nothing here, we have
4929 * the guarantee we will be called later.
4931 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
4932 NAPIF_STATE_IN_BUSY_POLL)))
4936 unsigned long timeout = 0;
4939 timeout = n->dev->gro_flush_timeout;
4942 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4943 HRTIMER_MODE_REL_PINNED);
4945 napi_gro_flush(n, false);
4947 if (likely(list_empty(&n->poll_list))) {
4948 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4950 /* If n->poll_list is not empty, we need to mask irqs */
4951 local_irq_save(flags);
4953 local_irq_restore(flags);
4957 EXPORT_SYMBOL(napi_complete_done);
4959 /* must be called under rcu_read_lock(), as we dont take a reference */
4960 static struct napi_struct *napi_by_id(unsigned int napi_id)
4962 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4963 struct napi_struct *napi;
4965 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4966 if (napi->napi_id == napi_id)
4972 #if defined(CONFIG_NET_RX_BUSY_POLL)
4974 #define BUSY_POLL_BUDGET 8
4976 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
4980 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
4984 /* All we really want here is to re-enable device interrupts.
4985 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
4987 rc = napi->poll(napi, BUSY_POLL_BUDGET);
4988 netpoll_poll_unlock(have_poll_lock);
4989 if (rc == BUSY_POLL_BUDGET)
4990 __napi_schedule(napi);
4992 if (local_softirq_pending())
4996 bool sk_busy_loop(struct sock *sk, int nonblock)
4998 unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0;
4999 int (*napi_poll)(struct napi_struct *napi, int budget);
5000 int (*busy_poll)(struct napi_struct *dev);
5001 void *have_poll_lock = NULL;
5002 struct napi_struct *napi;
5011 napi = napi_by_id(sk->sk_napi_id);
5015 /* Note: ndo_busy_poll method is optional in linux-4.5 */
5016 busy_poll = napi->dev->netdev_ops->ndo_busy_poll;
5023 rc = busy_poll(napi);
5027 unsigned long val = READ_ONCE(napi->state);
5029 /* If multiple threads are competing for this napi,
5030 * we avoid dirtying napi->state as much as we can.
5032 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
5033 NAPIF_STATE_IN_BUSY_POLL))
5035 if (cmpxchg(&napi->state, val,
5036 val | NAPIF_STATE_IN_BUSY_POLL |
5037 NAPIF_STATE_SCHED) != val)
5039 have_poll_lock = netpoll_poll_lock(napi);
5040 napi_poll = napi->poll;
5042 rc = napi_poll(napi, BUSY_POLL_BUDGET);
5043 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
5046 __NET_ADD_STATS(sock_net(sk),
5047 LINUX_MIB_BUSYPOLLRXPACKETS, rc);
5050 if (rc == LL_FLUSH_FAILED)
5051 break; /* permanent failure */
5053 if (nonblock || !skb_queue_empty(&sk->sk_receive_queue) ||
5054 busy_loop_timeout(end_time))
5057 if (unlikely(need_resched())) {
5059 busy_poll_stop(napi, have_poll_lock);
5063 rc = !skb_queue_empty(&sk->sk_receive_queue);
5064 if (rc || busy_loop_timeout(end_time))
5071 busy_poll_stop(napi, have_poll_lock);
5073 rc = !skb_queue_empty(&sk->sk_receive_queue);
5078 EXPORT_SYMBOL(sk_busy_loop);
5080 #endif /* CONFIG_NET_RX_BUSY_POLL */
5082 static void napi_hash_add(struct napi_struct *napi)
5084 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
5085 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
5088 spin_lock(&napi_hash_lock);
5090 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */
5092 if (unlikely(++napi_gen_id < NR_CPUS + 1))
5093 napi_gen_id = NR_CPUS + 1;
5094 } while (napi_by_id(napi_gen_id));
5095 napi->napi_id = napi_gen_id;
5097 hlist_add_head_rcu(&napi->napi_hash_node,
5098 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
5100 spin_unlock(&napi_hash_lock);
5103 /* Warning : caller is responsible to make sure rcu grace period
5104 * is respected before freeing memory containing @napi
5106 bool napi_hash_del(struct napi_struct *napi)
5108 bool rcu_sync_needed = false;
5110 spin_lock(&napi_hash_lock);
5112 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
5113 rcu_sync_needed = true;
5114 hlist_del_rcu(&napi->napi_hash_node);
5116 spin_unlock(&napi_hash_lock);
5117 return rcu_sync_needed;
5119 EXPORT_SYMBOL_GPL(napi_hash_del);
5121 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
5123 struct napi_struct *napi;
5125 napi = container_of(timer, struct napi_struct, timer);
5127 napi_schedule(napi);
5129 return HRTIMER_NORESTART;
5132 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
5133 int (*poll)(struct napi_struct *, int), int weight)
5135 INIT_LIST_HEAD(&napi->poll_list);
5136 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
5137 napi->timer.function = napi_watchdog;
5138 napi->gro_count = 0;
5139 napi->gro_list = NULL;
5142 if (weight > NAPI_POLL_WEIGHT)
5143 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5145 napi->weight = weight;
5146 list_add(&napi->dev_list, &dev->napi_list);
5148 #ifdef CONFIG_NETPOLL
5149 napi->poll_owner = -1;
5151 set_bit(NAPI_STATE_SCHED, &napi->state);
5152 napi_hash_add(napi);
5154 EXPORT_SYMBOL(netif_napi_add);
5156 void napi_disable(struct napi_struct *n)
5159 set_bit(NAPI_STATE_DISABLE, &n->state);
5161 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
5163 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
5166 hrtimer_cancel(&n->timer);
5168 clear_bit(NAPI_STATE_DISABLE, &n->state);
5170 EXPORT_SYMBOL(napi_disable);
5172 /* Must be called in process context */
5173 void netif_napi_del(struct napi_struct *napi)
5176 if (napi_hash_del(napi))
5178 list_del_init(&napi->dev_list);
5179 napi_free_frags(napi);
5181 kfree_skb_list(napi->gro_list);
5182 napi->gro_list = NULL;
5183 napi->gro_count = 0;
5185 EXPORT_SYMBOL(netif_napi_del);
5187 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
5192 list_del_init(&n->poll_list);
5194 have = netpoll_poll_lock(n);
5198 /* This NAPI_STATE_SCHED test is for avoiding a race
5199 * with netpoll's poll_napi(). Only the entity which
5200 * obtains the lock and sees NAPI_STATE_SCHED set will
5201 * actually make the ->poll() call. Therefore we avoid
5202 * accidentally calling ->poll() when NAPI is not scheduled.
5205 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
5206 work = n->poll(n, weight);
5207 trace_napi_poll(n, work, weight);
5210 WARN_ON_ONCE(work > weight);
5212 if (likely(work < weight))
5215 /* Drivers must not modify the NAPI state if they
5216 * consume the entire weight. In such cases this code
5217 * still "owns" the NAPI instance and therefore can
5218 * move the instance around on the list at-will.
5220 if (unlikely(napi_disable_pending(n))) {
5226 /* flush too old packets
5227 * If HZ < 1000, flush all packets.
5229 napi_gro_flush(n, HZ >= 1000);
5232 /* Some drivers may have called napi_schedule
5233 * prior to exhausting their budget.
5235 if (unlikely(!list_empty(&n->poll_list))) {
5236 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5237 n->dev ? n->dev->name : "backlog");
5241 list_add_tail(&n->poll_list, repoll);
5244 netpoll_poll_unlock(have);
5249 static __latent_entropy void net_rx_action(struct softirq_action *h)
5251 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5252 unsigned long time_limit = jiffies + 2;
5253 int budget = netdev_budget;
5257 local_irq_disable();
5258 list_splice_init(&sd->poll_list, &list);
5262 struct napi_struct *n;
5264 if (list_empty(&list)) {
5265 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
5270 n = list_first_entry(&list, struct napi_struct, poll_list);
5271 budget -= napi_poll(n, &repoll);
5273 /* If softirq window is exhausted then punt.
5274 * Allow this to run for 2 jiffies since which will allow
5275 * an average latency of 1.5/HZ.
5277 if (unlikely(budget <= 0 ||
5278 time_after_eq(jiffies, time_limit))) {
5284 local_irq_disable();
5286 list_splice_tail_init(&sd->poll_list, &list);
5287 list_splice_tail(&repoll, &list);
5288 list_splice(&list, &sd->poll_list);
5289 if (!list_empty(&sd->poll_list))
5290 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5292 net_rps_action_and_irq_enable(sd);
5294 __kfree_skb_flush();
5297 struct netdev_adjacent {
5298 struct net_device *dev;
5300 /* upper master flag, there can only be one master device per list */
5303 /* counter for the number of times this device was added to us */
5306 /* private field for the users */
5309 struct list_head list;
5310 struct rcu_head rcu;
5313 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
5314 struct list_head *adj_list)
5316 struct netdev_adjacent *adj;
5318 list_for_each_entry(adj, adj_list, list) {
5319 if (adj->dev == adj_dev)
5325 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
5327 struct net_device *dev = data;
5329 return upper_dev == dev;
5333 * netdev_has_upper_dev - Check if device is linked to an upper device
5335 * @upper_dev: upper device to check
5337 * Find out if a device is linked to specified upper device and return true
5338 * in case it is. Note that this checks only immediate upper device,
5339 * not through a complete stack of devices. The caller must hold the RTNL lock.
5341 bool netdev_has_upper_dev(struct net_device *dev,
5342 struct net_device *upper_dev)
5346 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5349 EXPORT_SYMBOL(netdev_has_upper_dev);
5352 * netdev_has_upper_dev_all - Check if device is linked to an upper device
5354 * @upper_dev: upper device to check
5356 * Find out if a device is linked to specified upper device and return true
5357 * in case it is. Note that this checks the entire upper device chain.
5358 * The caller must hold rcu lock.
5361 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
5362 struct net_device *upper_dev)
5364 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5367 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
5370 * netdev_has_any_upper_dev - Check if device is linked to some device
5373 * Find out if a device is linked to an upper device and return true in case
5374 * it is. The caller must hold the RTNL lock.
5376 static bool netdev_has_any_upper_dev(struct net_device *dev)
5380 return !list_empty(&dev->adj_list.upper);
5384 * netdev_master_upper_dev_get - Get master upper device
5387 * Find a master upper device and return pointer to it or NULL in case
5388 * it's not there. The caller must hold the RTNL lock.
5390 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5392 struct netdev_adjacent *upper;
5396 if (list_empty(&dev->adj_list.upper))
5399 upper = list_first_entry(&dev->adj_list.upper,
5400 struct netdev_adjacent, list);
5401 if (likely(upper->master))
5405 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5408 * netdev_has_any_lower_dev - Check if device is linked to some device
5411 * Find out if a device is linked to a lower device and return true in case
5412 * it is. The caller must hold the RTNL lock.
5414 static bool netdev_has_any_lower_dev(struct net_device *dev)
5418 return !list_empty(&dev->adj_list.lower);
5421 void *netdev_adjacent_get_private(struct list_head *adj_list)
5423 struct netdev_adjacent *adj;
5425 adj = list_entry(adj_list, struct netdev_adjacent, list);
5427 return adj->private;
5429 EXPORT_SYMBOL(netdev_adjacent_get_private);
5432 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5434 * @iter: list_head ** of the current position
5436 * Gets the next device from the dev's upper list, starting from iter
5437 * position. The caller must hold RCU read lock.
5439 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5440 struct list_head **iter)
5442 struct netdev_adjacent *upper;
5444 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5446 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5448 if (&upper->list == &dev->adj_list.upper)
5451 *iter = &upper->list;
5455 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5457 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
5458 struct list_head **iter)
5460 struct netdev_adjacent *upper;
5462 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5464 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5466 if (&upper->list == &dev->adj_list.upper)
5469 *iter = &upper->list;
5474 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
5475 int (*fn)(struct net_device *dev,
5479 struct net_device *udev;
5480 struct list_head *iter;
5483 for (iter = &dev->adj_list.upper,
5484 udev = netdev_next_upper_dev_rcu(dev, &iter);
5486 udev = netdev_next_upper_dev_rcu(dev, &iter)) {
5487 /* first is the upper device itself */
5488 ret = fn(udev, data);
5492 /* then look at all of its upper devices */
5493 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data);
5500 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
5503 * netdev_lower_get_next_private - Get the next ->private from the
5504 * lower neighbour list
5506 * @iter: list_head ** of the current position
5508 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5509 * list, starting from iter position. The caller must hold either hold the
5510 * RTNL lock or its own locking that guarantees that the neighbour lower
5511 * list will remain unchanged.
5513 void *netdev_lower_get_next_private(struct net_device *dev,
5514 struct list_head **iter)
5516 struct netdev_adjacent *lower;
5518 lower = list_entry(*iter, struct netdev_adjacent, list);
5520 if (&lower->list == &dev->adj_list.lower)
5523 *iter = lower->list.next;
5525 return lower->private;
5527 EXPORT_SYMBOL(netdev_lower_get_next_private);
5530 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5531 * lower neighbour list, RCU
5534 * @iter: list_head ** of the current position
5536 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5537 * list, starting from iter position. The caller must hold RCU read lock.
5539 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5540 struct list_head **iter)
5542 struct netdev_adjacent *lower;
5544 WARN_ON_ONCE(!rcu_read_lock_held());
5546 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5548 if (&lower->list == &dev->adj_list.lower)
5551 *iter = &lower->list;
5553 return lower->private;
5555 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5558 * netdev_lower_get_next - Get the next device from the lower neighbour
5561 * @iter: list_head ** of the current position
5563 * Gets the next netdev_adjacent from the dev's lower neighbour
5564 * list, starting from iter position. The caller must hold RTNL lock or
5565 * its own locking that guarantees that the neighbour lower
5566 * list will remain unchanged.
5568 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5570 struct netdev_adjacent *lower;
5572 lower = list_entry(*iter, struct netdev_adjacent, list);
5574 if (&lower->list == &dev->adj_list.lower)
5577 *iter = lower->list.next;
5581 EXPORT_SYMBOL(netdev_lower_get_next);
5583 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
5584 struct list_head **iter)
5586 struct netdev_adjacent *lower;
5588 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5590 if (&lower->list == &dev->adj_list.lower)
5593 *iter = &lower->list;
5598 int netdev_walk_all_lower_dev(struct net_device *dev,
5599 int (*fn)(struct net_device *dev,
5603 struct net_device *ldev;
5604 struct list_head *iter;
5607 for (iter = &dev->adj_list.lower,
5608 ldev = netdev_next_lower_dev(dev, &iter);
5610 ldev = netdev_next_lower_dev(dev, &iter)) {
5611 /* first is the lower device itself */
5612 ret = fn(ldev, data);
5616 /* then look at all of its lower devices */
5617 ret = netdev_walk_all_lower_dev(ldev, fn, data);
5624 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
5626 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
5627 struct list_head **iter)
5629 struct netdev_adjacent *lower;
5631 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5632 if (&lower->list == &dev->adj_list.lower)
5635 *iter = &lower->list;
5640 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
5641 int (*fn)(struct net_device *dev,
5645 struct net_device *ldev;
5646 struct list_head *iter;
5649 for (iter = &dev->adj_list.lower,
5650 ldev = netdev_next_lower_dev_rcu(dev, &iter);
5652 ldev = netdev_next_lower_dev_rcu(dev, &iter)) {
5653 /* first is the lower device itself */
5654 ret = fn(ldev, data);
5658 /* then look at all of its lower devices */
5659 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data);
5666 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
5669 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5670 * lower neighbour list, RCU
5674 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5675 * list. The caller must hold RCU read lock.
5677 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5679 struct netdev_adjacent *lower;
5681 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5682 struct netdev_adjacent, list);
5684 return lower->private;
5687 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5690 * netdev_master_upper_dev_get_rcu - Get master upper device
5693 * Find a master upper device and return pointer to it or NULL in case
5694 * it's not there. The caller must hold the RCU read lock.
5696 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5698 struct netdev_adjacent *upper;
5700 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5701 struct netdev_adjacent, list);
5702 if (upper && likely(upper->master))
5706 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5708 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5709 struct net_device *adj_dev,
5710 struct list_head *dev_list)
5712 char linkname[IFNAMSIZ+7];
5713 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5714 "upper_%s" : "lower_%s", adj_dev->name);
5715 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5718 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5720 struct list_head *dev_list)
5722 char linkname[IFNAMSIZ+7];
5723 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5724 "upper_%s" : "lower_%s", name);
5725 sysfs_remove_link(&(dev->dev.kobj), linkname);
5728 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5729 struct net_device *adj_dev,
5730 struct list_head *dev_list)
5732 return (dev_list == &dev->adj_list.upper ||
5733 dev_list == &dev->adj_list.lower) &&
5734 net_eq(dev_net(dev), dev_net(adj_dev));
5737 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5738 struct net_device *adj_dev,
5739 struct list_head *dev_list,
5740 void *private, bool master)
5742 struct netdev_adjacent *adj;
5745 adj = __netdev_find_adj(adj_dev, dev_list);
5749 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
5750 dev->name, adj_dev->name, adj->ref_nr);
5755 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5760 adj->master = master;
5762 adj->private = private;
5765 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
5766 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
5768 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5769 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5774 /* Ensure that master link is always the first item in list. */
5776 ret = sysfs_create_link(&(dev->dev.kobj),
5777 &(adj_dev->dev.kobj), "master");
5779 goto remove_symlinks;
5781 list_add_rcu(&adj->list, dev_list);
5783 list_add_tail_rcu(&adj->list, dev_list);
5789 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5790 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5798 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5799 struct net_device *adj_dev,
5801 struct list_head *dev_list)
5803 struct netdev_adjacent *adj;
5805 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
5806 dev->name, adj_dev->name, ref_nr);
5808 adj = __netdev_find_adj(adj_dev, dev_list);
5811 pr_err("Adjacency does not exist for device %s from %s\n",
5812 dev->name, adj_dev->name);
5817 if (adj->ref_nr > ref_nr) {
5818 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
5819 dev->name, adj_dev->name, ref_nr,
5820 adj->ref_nr - ref_nr);
5821 adj->ref_nr -= ref_nr;
5826 sysfs_remove_link(&(dev->dev.kobj), "master");
5828 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5829 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5831 list_del_rcu(&adj->list);
5832 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
5833 adj_dev->name, dev->name, adj_dev->name);
5835 kfree_rcu(adj, rcu);
5838 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5839 struct net_device *upper_dev,
5840 struct list_head *up_list,
5841 struct list_head *down_list,
5842 void *private, bool master)
5846 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
5851 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
5854 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
5861 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5862 struct net_device *upper_dev,
5864 struct list_head *up_list,
5865 struct list_head *down_list)
5867 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
5868 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
5871 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5872 struct net_device *upper_dev,
5873 void *private, bool master)
5875 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5876 &dev->adj_list.upper,
5877 &upper_dev->adj_list.lower,
5881 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5882 struct net_device *upper_dev)
5884 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
5885 &dev->adj_list.upper,
5886 &upper_dev->adj_list.lower);
5889 static int __netdev_upper_dev_link(struct net_device *dev,
5890 struct net_device *upper_dev, bool master,
5891 void *upper_priv, void *upper_info)
5893 struct netdev_notifier_changeupper_info changeupper_info;
5898 if (dev == upper_dev)
5901 /* To prevent loops, check if dev is not upper device to upper_dev. */
5902 if (netdev_has_upper_dev(upper_dev, dev))
5905 if (netdev_has_upper_dev(dev, upper_dev))
5908 if (master && netdev_master_upper_dev_get(dev))
5911 changeupper_info.upper_dev = upper_dev;
5912 changeupper_info.master = master;
5913 changeupper_info.linking = true;
5914 changeupper_info.upper_info = upper_info;
5916 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5917 &changeupper_info.info);
5918 ret = notifier_to_errno(ret);
5922 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
5927 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5928 &changeupper_info.info);
5929 ret = notifier_to_errno(ret);
5936 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5942 * netdev_upper_dev_link - Add a link to the upper device
5944 * @upper_dev: new upper device
5946 * Adds a link to device which is upper to this one. The caller must hold
5947 * the RTNL lock. On a failure a negative errno code is returned.
5948 * On success the reference counts are adjusted and the function
5951 int netdev_upper_dev_link(struct net_device *dev,
5952 struct net_device *upper_dev)
5954 return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL);
5956 EXPORT_SYMBOL(netdev_upper_dev_link);
5959 * netdev_master_upper_dev_link - Add a master link to the upper device
5961 * @upper_dev: new upper device
5962 * @upper_priv: upper device private
5963 * @upper_info: upper info to be passed down via notifier
5965 * Adds a link to device which is upper to this one. In this case, only
5966 * one master upper device can be linked, although other non-master devices
5967 * might be linked as well. The caller must hold the RTNL lock.
5968 * On a failure a negative errno code is returned. On success the reference
5969 * counts are adjusted and the function returns zero.
5971 int netdev_master_upper_dev_link(struct net_device *dev,
5972 struct net_device *upper_dev,
5973 void *upper_priv, void *upper_info)
5975 return __netdev_upper_dev_link(dev, upper_dev, true,
5976 upper_priv, upper_info);
5978 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5981 * netdev_upper_dev_unlink - Removes a link to upper device
5983 * @upper_dev: new upper device
5985 * Removes a link to device which is upper to this one. The caller must hold
5988 void netdev_upper_dev_unlink(struct net_device *dev,
5989 struct net_device *upper_dev)
5991 struct netdev_notifier_changeupper_info changeupper_info;
5994 changeupper_info.upper_dev = upper_dev;
5995 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
5996 changeupper_info.linking = false;
5998 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5999 &changeupper_info.info);
6001 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
6003 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
6004 &changeupper_info.info);
6006 EXPORT_SYMBOL(netdev_upper_dev_unlink);
6009 * netdev_bonding_info_change - Dispatch event about slave change
6011 * @bonding_info: info to dispatch
6013 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
6014 * The caller must hold the RTNL lock.
6016 void netdev_bonding_info_change(struct net_device *dev,
6017 struct netdev_bonding_info *bonding_info)
6019 struct netdev_notifier_bonding_info info;
6021 memcpy(&info.bonding_info, bonding_info,
6022 sizeof(struct netdev_bonding_info));
6023 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
6026 EXPORT_SYMBOL(netdev_bonding_info_change);
6028 static void netdev_adjacent_add_links(struct net_device *dev)
6030 struct netdev_adjacent *iter;
6032 struct net *net = dev_net(dev);
6034 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6035 if (!net_eq(net, dev_net(iter->dev)))
6037 netdev_adjacent_sysfs_add(iter->dev, dev,
6038 &iter->dev->adj_list.lower);
6039 netdev_adjacent_sysfs_add(dev, iter->dev,
6040 &dev->adj_list.upper);
6043 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6044 if (!net_eq(net, dev_net(iter->dev)))
6046 netdev_adjacent_sysfs_add(iter->dev, dev,
6047 &iter->dev->adj_list.upper);
6048 netdev_adjacent_sysfs_add(dev, iter->dev,
6049 &dev->adj_list.lower);
6053 static void netdev_adjacent_del_links(struct net_device *dev)
6055 struct netdev_adjacent *iter;
6057 struct net *net = dev_net(dev);
6059 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6060 if (!net_eq(net, dev_net(iter->dev)))
6062 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6063 &iter->dev->adj_list.lower);
6064 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6065 &dev->adj_list.upper);
6068 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6069 if (!net_eq(net, dev_net(iter->dev)))
6071 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6072 &iter->dev->adj_list.upper);
6073 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6074 &dev->adj_list.lower);
6078 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
6080 struct netdev_adjacent *iter;
6082 struct net *net = dev_net(dev);
6084 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6085 if (!net_eq(net, dev_net(iter->dev)))
6087 netdev_adjacent_sysfs_del(iter->dev, oldname,
6088 &iter->dev->adj_list.lower);
6089 netdev_adjacent_sysfs_add(iter->dev, dev,
6090 &iter->dev->adj_list.lower);
6093 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6094 if (!net_eq(net, dev_net(iter->dev)))
6096 netdev_adjacent_sysfs_del(iter->dev, oldname,
6097 &iter->dev->adj_list.upper);
6098 netdev_adjacent_sysfs_add(iter->dev, dev,
6099 &iter->dev->adj_list.upper);
6103 void *netdev_lower_dev_get_private(struct net_device *dev,
6104 struct net_device *lower_dev)
6106 struct netdev_adjacent *lower;
6110 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
6114 return lower->private;
6116 EXPORT_SYMBOL(netdev_lower_dev_get_private);
6119 int dev_get_nest_level(struct net_device *dev)
6121 struct net_device *lower = NULL;
6122 struct list_head *iter;
6128 netdev_for_each_lower_dev(dev, lower, iter) {
6129 nest = dev_get_nest_level(lower);
6130 if (max_nest < nest)
6134 return max_nest + 1;
6136 EXPORT_SYMBOL(dev_get_nest_level);
6139 * netdev_lower_change - Dispatch event about lower device state change
6140 * @lower_dev: device
6141 * @lower_state_info: state to dispatch
6143 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
6144 * The caller must hold the RTNL lock.
6146 void netdev_lower_state_changed(struct net_device *lower_dev,
6147 void *lower_state_info)
6149 struct netdev_notifier_changelowerstate_info changelowerstate_info;
6152 changelowerstate_info.lower_state_info = lower_state_info;
6153 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev,
6154 &changelowerstate_info.info);
6156 EXPORT_SYMBOL(netdev_lower_state_changed);
6158 int netdev_default_l2upper_neigh_construct(struct net_device *dev,
6159 struct neighbour *n)
6161 struct net_device *lower_dev, *stop_dev;
6162 struct list_head *iter;
6165 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6166 if (!lower_dev->netdev_ops->ndo_neigh_construct)
6168 err = lower_dev->netdev_ops->ndo_neigh_construct(lower_dev, n);
6170 stop_dev = lower_dev;
6177 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6178 if (lower_dev == stop_dev)
6180 if (!lower_dev->netdev_ops->ndo_neigh_destroy)
6182 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n);
6186 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_construct);
6188 void netdev_default_l2upper_neigh_destroy(struct net_device *dev,
6189 struct neighbour *n)
6191 struct net_device *lower_dev;
6192 struct list_head *iter;
6194 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6195 if (!lower_dev->netdev_ops->ndo_neigh_destroy)
6197 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n);
6200 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_destroy);
6202 static void dev_change_rx_flags(struct net_device *dev, int flags)
6204 const struct net_device_ops *ops = dev->netdev_ops;
6206 if (ops->ndo_change_rx_flags)
6207 ops->ndo_change_rx_flags(dev, flags);
6210 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
6212 unsigned int old_flags = dev->flags;
6218 dev->flags |= IFF_PROMISC;
6219 dev->promiscuity += inc;
6220 if (dev->promiscuity == 0) {
6223 * If inc causes overflow, untouch promisc and return error.
6226 dev->flags &= ~IFF_PROMISC;
6228 dev->promiscuity -= inc;
6229 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6234 if (dev->flags != old_flags) {
6235 pr_info("device %s %s promiscuous mode\n",
6237 dev->flags & IFF_PROMISC ? "entered" : "left");
6238 if (audit_enabled) {
6239 current_uid_gid(&uid, &gid);
6240 audit_log(current->audit_context, GFP_ATOMIC,
6241 AUDIT_ANOM_PROMISCUOUS,
6242 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6243 dev->name, (dev->flags & IFF_PROMISC),
6244 (old_flags & IFF_PROMISC),
6245 from_kuid(&init_user_ns, audit_get_loginuid(current)),
6246 from_kuid(&init_user_ns, uid),
6247 from_kgid(&init_user_ns, gid),
6248 audit_get_sessionid(current));
6251 dev_change_rx_flags(dev, IFF_PROMISC);
6254 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
6259 * dev_set_promiscuity - update promiscuity count on a device
6263 * Add or remove promiscuity from a device. While the count in the device
6264 * remains above zero the interface remains promiscuous. Once it hits zero
6265 * the device reverts back to normal filtering operation. A negative inc
6266 * value is used to drop promiscuity on the device.
6267 * Return 0 if successful or a negative errno code on error.
6269 int dev_set_promiscuity(struct net_device *dev, int inc)
6271 unsigned int old_flags = dev->flags;
6274 err = __dev_set_promiscuity(dev, inc, true);
6277 if (dev->flags != old_flags)
6278 dev_set_rx_mode(dev);
6281 EXPORT_SYMBOL(dev_set_promiscuity);
6283 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
6285 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
6289 dev->flags |= IFF_ALLMULTI;
6290 dev->allmulti += inc;
6291 if (dev->allmulti == 0) {
6294 * If inc causes overflow, untouch allmulti and return error.
6297 dev->flags &= ~IFF_ALLMULTI;
6299 dev->allmulti -= inc;
6300 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6305 if (dev->flags ^ old_flags) {
6306 dev_change_rx_flags(dev, IFF_ALLMULTI);
6307 dev_set_rx_mode(dev);
6309 __dev_notify_flags(dev, old_flags,
6310 dev->gflags ^ old_gflags);
6316 * dev_set_allmulti - update allmulti count on a device
6320 * Add or remove reception of all multicast frames to a device. While the
6321 * count in the device remains above zero the interface remains listening
6322 * to all interfaces. Once it hits zero the device reverts back to normal
6323 * filtering operation. A negative @inc value is used to drop the counter
6324 * when releasing a resource needing all multicasts.
6325 * Return 0 if successful or a negative errno code on error.
6328 int dev_set_allmulti(struct net_device *dev, int inc)
6330 return __dev_set_allmulti(dev, inc, true);
6332 EXPORT_SYMBOL(dev_set_allmulti);
6335 * Upload unicast and multicast address lists to device and
6336 * configure RX filtering. When the device doesn't support unicast
6337 * filtering it is put in promiscuous mode while unicast addresses
6340 void __dev_set_rx_mode(struct net_device *dev)
6342 const struct net_device_ops *ops = dev->netdev_ops;
6344 /* dev_open will call this function so the list will stay sane. */
6345 if (!(dev->flags&IFF_UP))
6348 if (!netif_device_present(dev))
6351 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
6352 /* Unicast addresses changes may only happen under the rtnl,
6353 * therefore calling __dev_set_promiscuity here is safe.
6355 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
6356 __dev_set_promiscuity(dev, 1, false);
6357 dev->uc_promisc = true;
6358 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
6359 __dev_set_promiscuity(dev, -1, false);
6360 dev->uc_promisc = false;
6364 if (ops->ndo_set_rx_mode)
6365 ops->ndo_set_rx_mode(dev);
6368 void dev_set_rx_mode(struct net_device *dev)
6370 netif_addr_lock_bh(dev);
6371 __dev_set_rx_mode(dev);
6372 netif_addr_unlock_bh(dev);
6376 * dev_get_flags - get flags reported to userspace
6379 * Get the combination of flag bits exported through APIs to userspace.
6381 unsigned int dev_get_flags(const struct net_device *dev)
6385 flags = (dev->flags & ~(IFF_PROMISC |
6390 (dev->gflags & (IFF_PROMISC |
6393 if (netif_running(dev)) {
6394 if (netif_oper_up(dev))
6395 flags |= IFF_RUNNING;
6396 if (netif_carrier_ok(dev))
6397 flags |= IFF_LOWER_UP;
6398 if (netif_dormant(dev))
6399 flags |= IFF_DORMANT;
6404 EXPORT_SYMBOL(dev_get_flags);
6406 int __dev_change_flags(struct net_device *dev, unsigned int flags)
6408 unsigned int old_flags = dev->flags;
6414 * Set the flags on our device.
6417 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
6418 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
6420 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6424 * Load in the correct multicast list now the flags have changed.
6427 if ((old_flags ^ flags) & IFF_MULTICAST)
6428 dev_change_rx_flags(dev, IFF_MULTICAST);
6430 dev_set_rx_mode(dev);
6433 * Have we downed the interface. We handle IFF_UP ourselves
6434 * according to user attempts to set it, rather than blindly
6439 if ((old_flags ^ flags) & IFF_UP)
6440 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
6442 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6443 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6444 unsigned int old_flags = dev->flags;
6446 dev->gflags ^= IFF_PROMISC;
6448 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6449 if (dev->flags != old_flags)
6450 dev_set_rx_mode(dev);
6453 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6454 is important. Some (broken) drivers set IFF_PROMISC, when
6455 IFF_ALLMULTI is requested not asking us and not reporting.
6457 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6458 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6460 dev->gflags ^= IFF_ALLMULTI;
6461 __dev_set_allmulti(dev, inc, false);
6467 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6468 unsigned int gchanges)
6470 unsigned int changes = dev->flags ^ old_flags;
6473 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6475 if (changes & IFF_UP) {
6476 if (dev->flags & IFF_UP)
6477 call_netdevice_notifiers(NETDEV_UP, dev);
6479 call_netdevice_notifiers(NETDEV_DOWN, dev);
6482 if (dev->flags & IFF_UP &&
6483 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6484 struct netdev_notifier_change_info change_info;
6486 change_info.flags_changed = changes;
6487 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6493 * dev_change_flags - change device settings
6495 * @flags: device state flags
6497 * Change settings on device based state flags. The flags are
6498 * in the userspace exported format.
6500 int dev_change_flags(struct net_device *dev, unsigned int flags)
6503 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6505 ret = __dev_change_flags(dev, flags);
6509 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6510 __dev_notify_flags(dev, old_flags, changes);
6513 EXPORT_SYMBOL(dev_change_flags);
6515 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6517 const struct net_device_ops *ops = dev->netdev_ops;
6519 if (ops->ndo_change_mtu)
6520 return ops->ndo_change_mtu(dev, new_mtu);
6527 * dev_set_mtu - Change maximum transfer unit
6529 * @new_mtu: new transfer unit
6531 * Change the maximum transfer size of the network device.
6533 int dev_set_mtu(struct net_device *dev, int new_mtu)
6537 if (new_mtu == dev->mtu)
6540 /* MTU must be positive, and in range */
6541 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
6542 net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n",
6543 dev->name, new_mtu, dev->min_mtu);
6547 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
6548 net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n",
6549 dev->name, new_mtu, dev->max_mtu);
6553 if (!netif_device_present(dev))
6556 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6557 err = notifier_to_errno(err);
6561 orig_mtu = dev->mtu;
6562 err = __dev_set_mtu(dev, new_mtu);
6565 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6566 err = notifier_to_errno(err);
6568 /* setting mtu back and notifying everyone again,
6569 * so that they have a chance to revert changes.
6571 __dev_set_mtu(dev, orig_mtu);
6572 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6577 EXPORT_SYMBOL(dev_set_mtu);
6580 * dev_set_group - Change group this device belongs to
6582 * @new_group: group this device should belong to
6584 void dev_set_group(struct net_device *dev, int new_group)
6586 dev->group = new_group;
6588 EXPORT_SYMBOL(dev_set_group);
6591 * dev_set_mac_address - Change Media Access Control Address
6595 * Change the hardware (MAC) address of the device
6597 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6599 const struct net_device_ops *ops = dev->netdev_ops;
6602 if (!ops->ndo_set_mac_address)
6604 if (sa->sa_family != dev->type)
6606 if (!netif_device_present(dev))
6608 err = ops->ndo_set_mac_address(dev, sa);
6611 dev->addr_assign_type = NET_ADDR_SET;
6612 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6613 add_device_randomness(dev->dev_addr, dev->addr_len);
6616 EXPORT_SYMBOL(dev_set_mac_address);
6619 * dev_change_carrier - Change device carrier
6621 * @new_carrier: new value
6623 * Change device carrier
6625 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6627 const struct net_device_ops *ops = dev->netdev_ops;
6629 if (!ops->ndo_change_carrier)
6631 if (!netif_device_present(dev))
6633 return ops->ndo_change_carrier(dev, new_carrier);
6635 EXPORT_SYMBOL(dev_change_carrier);
6638 * dev_get_phys_port_id - Get device physical port ID
6642 * Get device physical port ID
6644 int dev_get_phys_port_id(struct net_device *dev,
6645 struct netdev_phys_item_id *ppid)
6647 const struct net_device_ops *ops = dev->netdev_ops;
6649 if (!ops->ndo_get_phys_port_id)
6651 return ops->ndo_get_phys_port_id(dev, ppid);
6653 EXPORT_SYMBOL(dev_get_phys_port_id);
6656 * dev_get_phys_port_name - Get device physical port name
6659 * @len: limit of bytes to copy to name
6661 * Get device physical port name
6663 int dev_get_phys_port_name(struct net_device *dev,
6664 char *name, size_t len)
6666 const struct net_device_ops *ops = dev->netdev_ops;
6668 if (!ops->ndo_get_phys_port_name)
6670 return ops->ndo_get_phys_port_name(dev, name, len);
6672 EXPORT_SYMBOL(dev_get_phys_port_name);
6675 * dev_change_proto_down - update protocol port state information
6677 * @proto_down: new value
6679 * This info can be used by switch drivers to set the phys state of the
6682 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6684 const struct net_device_ops *ops = dev->netdev_ops;
6686 if (!ops->ndo_change_proto_down)
6688 if (!netif_device_present(dev))
6690 return ops->ndo_change_proto_down(dev, proto_down);
6692 EXPORT_SYMBOL(dev_change_proto_down);
6695 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
6697 * @fd: new program fd or negative value to clear
6698 * @flags: xdp-related flags
6700 * Set or clear a bpf program for a device
6702 int dev_change_xdp_fd(struct net_device *dev, int fd, u32 flags)
6704 const struct net_device_ops *ops = dev->netdev_ops;
6705 struct bpf_prog *prog = NULL;
6706 struct netdev_xdp xdp;
6714 if (flags & XDP_FLAGS_UPDATE_IF_NOEXIST) {
6715 memset(&xdp, 0, sizeof(xdp));
6716 xdp.command = XDP_QUERY_PROG;
6718 err = ops->ndo_xdp(dev, &xdp);
6721 if (xdp.prog_attached)
6725 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP);
6727 return PTR_ERR(prog);
6730 memset(&xdp, 0, sizeof(xdp));
6731 xdp.command = XDP_SETUP_PROG;
6734 err = ops->ndo_xdp(dev, &xdp);
6735 if (err < 0 && prog)
6740 EXPORT_SYMBOL(dev_change_xdp_fd);
6743 * dev_new_index - allocate an ifindex
6744 * @net: the applicable net namespace
6746 * Returns a suitable unique value for a new device interface
6747 * number. The caller must hold the rtnl semaphore or the
6748 * dev_base_lock to be sure it remains unique.
6750 static int dev_new_index(struct net *net)
6752 int ifindex = net->ifindex;
6756 if (!__dev_get_by_index(net, ifindex))
6757 return net->ifindex = ifindex;
6761 /* Delayed registration/unregisteration */
6762 static LIST_HEAD(net_todo_list);
6763 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6765 static void net_set_todo(struct net_device *dev)
6767 list_add_tail(&dev->todo_list, &net_todo_list);
6768 dev_net(dev)->dev_unreg_count++;
6771 static void rollback_registered_many(struct list_head *head)
6773 struct net_device *dev, *tmp;
6774 LIST_HEAD(close_head);
6776 BUG_ON(dev_boot_phase);
6779 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6780 /* Some devices call without registering
6781 * for initialization unwind. Remove those
6782 * devices and proceed with the remaining.
6784 if (dev->reg_state == NETREG_UNINITIALIZED) {
6785 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6789 list_del(&dev->unreg_list);
6792 dev->dismantle = true;
6793 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6796 /* If device is running, close it first. */
6797 list_for_each_entry(dev, head, unreg_list)
6798 list_add_tail(&dev->close_list, &close_head);
6799 dev_close_many(&close_head, true);
6801 list_for_each_entry(dev, head, unreg_list) {
6802 /* And unlink it from device chain. */
6803 unlist_netdevice(dev);
6805 dev->reg_state = NETREG_UNREGISTERING;
6807 flush_all_backlogs();
6811 list_for_each_entry(dev, head, unreg_list) {
6812 struct sk_buff *skb = NULL;
6814 /* Shutdown queueing discipline. */
6818 /* Notify protocols, that we are about to destroy
6819 this device. They should clean all the things.
6821 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6823 if (!dev->rtnl_link_ops ||
6824 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6825 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6829 * Flush the unicast and multicast chains
6834 if (dev->netdev_ops->ndo_uninit)
6835 dev->netdev_ops->ndo_uninit(dev);
6838 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6840 /* Notifier chain MUST detach us all upper devices. */
6841 WARN_ON(netdev_has_any_upper_dev(dev));
6842 WARN_ON(netdev_has_any_lower_dev(dev));
6844 /* Remove entries from kobject tree */
6845 netdev_unregister_kobject(dev);
6847 /* Remove XPS queueing entries */
6848 netif_reset_xps_queues_gt(dev, 0);
6854 list_for_each_entry(dev, head, unreg_list)
6858 static void rollback_registered(struct net_device *dev)
6862 list_add(&dev->unreg_list, &single);
6863 rollback_registered_many(&single);
6867 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6868 struct net_device *upper, netdev_features_t features)
6870 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6871 netdev_features_t feature;
6874 for_each_netdev_feature(&upper_disables, feature_bit) {
6875 feature = __NETIF_F_BIT(feature_bit);
6876 if (!(upper->wanted_features & feature)
6877 && (features & feature)) {
6878 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6879 &feature, upper->name);
6880 features &= ~feature;
6887 static void netdev_sync_lower_features(struct net_device *upper,
6888 struct net_device *lower, netdev_features_t features)
6890 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6891 netdev_features_t feature;
6894 for_each_netdev_feature(&upper_disables, feature_bit) {
6895 feature = __NETIF_F_BIT(feature_bit);
6896 if (!(features & feature) && (lower->features & feature)) {
6897 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6898 &feature, lower->name);
6899 lower->wanted_features &= ~feature;
6900 netdev_update_features(lower);
6902 if (unlikely(lower->features & feature))
6903 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6904 &feature, lower->name);
6909 static netdev_features_t netdev_fix_features(struct net_device *dev,
6910 netdev_features_t features)
6912 /* Fix illegal checksum combinations */
6913 if ((features & NETIF_F_HW_CSUM) &&
6914 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6915 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6916 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6919 /* TSO requires that SG is present as well. */
6920 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6921 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6922 features &= ~NETIF_F_ALL_TSO;
6925 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6926 !(features & NETIF_F_IP_CSUM)) {
6927 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6928 features &= ~NETIF_F_TSO;
6929 features &= ~NETIF_F_TSO_ECN;
6932 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6933 !(features & NETIF_F_IPV6_CSUM)) {
6934 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6935 features &= ~NETIF_F_TSO6;
6938 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
6939 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
6940 features &= ~NETIF_F_TSO_MANGLEID;
6942 /* TSO ECN requires that TSO is present as well. */
6943 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6944 features &= ~NETIF_F_TSO_ECN;
6946 /* Software GSO depends on SG. */
6947 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6948 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6949 features &= ~NETIF_F_GSO;
6952 /* UFO needs SG and checksumming */
6953 if (features & NETIF_F_UFO) {
6954 /* maybe split UFO into V4 and V6? */
6955 if (!(features & NETIF_F_HW_CSUM) &&
6956 ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) !=
6957 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) {
6959 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6960 features &= ~NETIF_F_UFO;
6963 if (!(features & NETIF_F_SG)) {
6965 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6966 features &= ~NETIF_F_UFO;
6970 /* GSO partial features require GSO partial be set */
6971 if ((features & dev->gso_partial_features) &&
6972 !(features & NETIF_F_GSO_PARTIAL)) {
6974 "Dropping partially supported GSO features since no GSO partial.\n");
6975 features &= ~dev->gso_partial_features;
6978 #ifdef CONFIG_NET_RX_BUSY_POLL
6979 if (dev->netdev_ops->ndo_busy_poll)
6980 features |= NETIF_F_BUSY_POLL;
6983 features &= ~NETIF_F_BUSY_POLL;
6988 int __netdev_update_features(struct net_device *dev)
6990 struct net_device *upper, *lower;
6991 netdev_features_t features;
6992 struct list_head *iter;
6997 features = netdev_get_wanted_features(dev);
6999 if (dev->netdev_ops->ndo_fix_features)
7000 features = dev->netdev_ops->ndo_fix_features(dev, features);
7002 /* driver might be less strict about feature dependencies */
7003 features = netdev_fix_features(dev, features);
7005 /* some features can't be enabled if they're off an an upper device */
7006 netdev_for_each_upper_dev_rcu(dev, upper, iter)
7007 features = netdev_sync_upper_features(dev, upper, features);
7009 if (dev->features == features)
7012 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
7013 &dev->features, &features);
7015 if (dev->netdev_ops->ndo_set_features)
7016 err = dev->netdev_ops->ndo_set_features(dev, features);
7020 if (unlikely(err < 0)) {
7022 "set_features() failed (%d); wanted %pNF, left %pNF\n",
7023 err, &features, &dev->features);
7024 /* return non-0 since some features might have changed and
7025 * it's better to fire a spurious notification than miss it
7031 /* some features must be disabled on lower devices when disabled
7032 * on an upper device (think: bonding master or bridge)
7034 netdev_for_each_lower_dev(dev, lower, iter)
7035 netdev_sync_lower_features(dev, lower, features);
7038 dev->features = features;
7040 return err < 0 ? 0 : 1;
7044 * netdev_update_features - recalculate device features
7045 * @dev: the device to check
7047 * Recalculate dev->features set and send notifications if it
7048 * has changed. Should be called after driver or hardware dependent
7049 * conditions might have changed that influence the features.
7051 void netdev_update_features(struct net_device *dev)
7053 if (__netdev_update_features(dev))
7054 netdev_features_change(dev);
7056 EXPORT_SYMBOL(netdev_update_features);
7059 * netdev_change_features - recalculate device features
7060 * @dev: the device to check
7062 * Recalculate dev->features set and send notifications even
7063 * if they have not changed. Should be called instead of
7064 * netdev_update_features() if also dev->vlan_features might
7065 * have changed to allow the changes to be propagated to stacked
7068 void netdev_change_features(struct net_device *dev)
7070 __netdev_update_features(dev);
7071 netdev_features_change(dev);
7073 EXPORT_SYMBOL(netdev_change_features);
7076 * netif_stacked_transfer_operstate - transfer operstate
7077 * @rootdev: the root or lower level device to transfer state from
7078 * @dev: the device to transfer operstate to
7080 * Transfer operational state from root to device. This is normally
7081 * called when a stacking relationship exists between the root
7082 * device and the device(a leaf device).
7084 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
7085 struct net_device *dev)
7087 if (rootdev->operstate == IF_OPER_DORMANT)
7088 netif_dormant_on(dev);
7090 netif_dormant_off(dev);
7092 if (netif_carrier_ok(rootdev)) {
7093 if (!netif_carrier_ok(dev))
7094 netif_carrier_on(dev);
7096 if (netif_carrier_ok(dev))
7097 netif_carrier_off(dev);
7100 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
7103 static int netif_alloc_rx_queues(struct net_device *dev)
7105 unsigned int i, count = dev->num_rx_queues;
7106 struct netdev_rx_queue *rx;
7107 size_t sz = count * sizeof(*rx);
7111 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7119 for (i = 0; i < count; i++)
7125 static void netdev_init_one_queue(struct net_device *dev,
7126 struct netdev_queue *queue, void *_unused)
7128 /* Initialize queue lock */
7129 spin_lock_init(&queue->_xmit_lock);
7130 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
7131 queue->xmit_lock_owner = -1;
7132 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
7135 dql_init(&queue->dql, HZ);
7139 static void netif_free_tx_queues(struct net_device *dev)
7144 static int netif_alloc_netdev_queues(struct net_device *dev)
7146 unsigned int count = dev->num_tx_queues;
7147 struct netdev_queue *tx;
7148 size_t sz = count * sizeof(*tx);
7150 if (count < 1 || count > 0xffff)
7153 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7161 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
7162 spin_lock_init(&dev->tx_global_lock);
7167 void netif_tx_stop_all_queues(struct net_device *dev)
7171 for (i = 0; i < dev->num_tx_queues; i++) {
7172 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
7173 netif_tx_stop_queue(txq);
7176 EXPORT_SYMBOL(netif_tx_stop_all_queues);
7179 * register_netdevice - register a network device
7180 * @dev: device to register
7182 * Take a completed network device structure and add it to the kernel
7183 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7184 * chain. 0 is returned on success. A negative errno code is returned
7185 * on a failure to set up the device, or if the name is a duplicate.
7187 * Callers must hold the rtnl semaphore. You may want
7188 * register_netdev() instead of this.
7191 * The locking appears insufficient to guarantee two parallel registers
7192 * will not get the same name.
7195 int register_netdevice(struct net_device *dev)
7198 struct net *net = dev_net(dev);
7200 BUG_ON(dev_boot_phase);
7205 /* When net_device's are persistent, this will be fatal. */
7206 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
7209 spin_lock_init(&dev->addr_list_lock);
7210 netdev_set_addr_lockdep_class(dev);
7212 ret = dev_get_valid_name(net, dev, dev->name);
7216 /* Init, if this function is available */
7217 if (dev->netdev_ops->ndo_init) {
7218 ret = dev->netdev_ops->ndo_init(dev);
7226 if (((dev->hw_features | dev->features) &
7227 NETIF_F_HW_VLAN_CTAG_FILTER) &&
7228 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
7229 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
7230 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
7237 dev->ifindex = dev_new_index(net);
7238 else if (__dev_get_by_index(net, dev->ifindex))
7241 /* Transfer changeable features to wanted_features and enable
7242 * software offloads (GSO and GRO).
7244 dev->hw_features |= NETIF_F_SOFT_FEATURES;
7245 dev->features |= NETIF_F_SOFT_FEATURES;
7246 dev->wanted_features = dev->features & dev->hw_features;
7248 if (!(dev->flags & IFF_LOOPBACK))
7249 dev->hw_features |= NETIF_F_NOCACHE_COPY;
7251 /* If IPv4 TCP segmentation offload is supported we should also
7252 * allow the device to enable segmenting the frame with the option
7253 * of ignoring a static IP ID value. This doesn't enable the
7254 * feature itself but allows the user to enable it later.
7256 if (dev->hw_features & NETIF_F_TSO)
7257 dev->hw_features |= NETIF_F_TSO_MANGLEID;
7258 if (dev->vlan_features & NETIF_F_TSO)
7259 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
7260 if (dev->mpls_features & NETIF_F_TSO)
7261 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
7262 if (dev->hw_enc_features & NETIF_F_TSO)
7263 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
7265 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
7267 dev->vlan_features |= NETIF_F_HIGHDMA;
7269 /* Make NETIF_F_SG inheritable to tunnel devices.
7271 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
7273 /* Make NETIF_F_SG inheritable to MPLS.
7275 dev->mpls_features |= NETIF_F_SG;
7277 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
7278 ret = notifier_to_errno(ret);
7282 ret = netdev_register_kobject(dev);
7285 dev->reg_state = NETREG_REGISTERED;
7287 __netdev_update_features(dev);
7290 * Default initial state at registry is that the
7291 * device is present.
7294 set_bit(__LINK_STATE_PRESENT, &dev->state);
7296 linkwatch_init_dev(dev);
7298 dev_init_scheduler(dev);
7300 list_netdevice(dev);
7301 add_device_randomness(dev->dev_addr, dev->addr_len);
7303 /* If the device has permanent device address, driver should
7304 * set dev_addr and also addr_assign_type should be set to
7305 * NET_ADDR_PERM (default value).
7307 if (dev->addr_assign_type == NET_ADDR_PERM)
7308 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
7310 /* Notify protocols, that a new device appeared. */
7311 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
7312 ret = notifier_to_errno(ret);
7314 rollback_registered(dev);
7315 dev->reg_state = NETREG_UNREGISTERED;
7318 * Prevent userspace races by waiting until the network
7319 * device is fully setup before sending notifications.
7321 if (!dev->rtnl_link_ops ||
7322 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7323 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7329 if (dev->netdev_ops->ndo_uninit)
7330 dev->netdev_ops->ndo_uninit(dev);
7333 EXPORT_SYMBOL(register_netdevice);
7336 * init_dummy_netdev - init a dummy network device for NAPI
7337 * @dev: device to init
7339 * This takes a network device structure and initialize the minimum
7340 * amount of fields so it can be used to schedule NAPI polls without
7341 * registering a full blown interface. This is to be used by drivers
7342 * that need to tie several hardware interfaces to a single NAPI
7343 * poll scheduler due to HW limitations.
7345 int init_dummy_netdev(struct net_device *dev)
7347 /* Clear everything. Note we don't initialize spinlocks
7348 * are they aren't supposed to be taken by any of the
7349 * NAPI code and this dummy netdev is supposed to be
7350 * only ever used for NAPI polls
7352 memset(dev, 0, sizeof(struct net_device));
7354 /* make sure we BUG if trying to hit standard
7355 * register/unregister code path
7357 dev->reg_state = NETREG_DUMMY;
7359 /* NAPI wants this */
7360 INIT_LIST_HEAD(&dev->napi_list);
7362 /* a dummy interface is started by default */
7363 set_bit(__LINK_STATE_PRESENT, &dev->state);
7364 set_bit(__LINK_STATE_START, &dev->state);
7366 /* Note : We dont allocate pcpu_refcnt for dummy devices,
7367 * because users of this 'device' dont need to change
7373 EXPORT_SYMBOL_GPL(init_dummy_netdev);
7377 * register_netdev - register a network device
7378 * @dev: device to register
7380 * Take a completed network device structure and add it to the kernel
7381 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7382 * chain. 0 is returned on success. A negative errno code is returned
7383 * on a failure to set up the device, or if the name is a duplicate.
7385 * This is a wrapper around register_netdevice that takes the rtnl semaphore
7386 * and expands the device name if you passed a format string to
7389 int register_netdev(struct net_device *dev)
7394 err = register_netdevice(dev);
7398 EXPORT_SYMBOL(register_netdev);
7400 int netdev_refcnt_read(const struct net_device *dev)
7404 for_each_possible_cpu(i)
7405 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
7408 EXPORT_SYMBOL(netdev_refcnt_read);
7411 * netdev_wait_allrefs - wait until all references are gone.
7412 * @dev: target net_device
7414 * This is called when unregistering network devices.
7416 * Any protocol or device that holds a reference should register
7417 * for netdevice notification, and cleanup and put back the
7418 * reference if they receive an UNREGISTER event.
7419 * We can get stuck here if buggy protocols don't correctly
7422 static void netdev_wait_allrefs(struct net_device *dev)
7424 unsigned long rebroadcast_time, warning_time;
7427 linkwatch_forget_dev(dev);
7429 rebroadcast_time = warning_time = jiffies;
7430 refcnt = netdev_refcnt_read(dev);
7432 while (refcnt != 0) {
7433 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
7436 /* Rebroadcast unregister notification */
7437 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7443 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7444 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
7446 /* We must not have linkwatch events
7447 * pending on unregister. If this
7448 * happens, we simply run the queue
7449 * unscheduled, resulting in a noop
7452 linkwatch_run_queue();
7457 rebroadcast_time = jiffies;
7462 refcnt = netdev_refcnt_read(dev);
7464 if (time_after(jiffies, warning_time + 10 * HZ)) {
7465 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
7467 warning_time = jiffies;
7476 * register_netdevice(x1);
7477 * register_netdevice(x2);
7479 * unregister_netdevice(y1);
7480 * unregister_netdevice(y2);
7486 * We are invoked by rtnl_unlock().
7487 * This allows us to deal with problems:
7488 * 1) We can delete sysfs objects which invoke hotplug
7489 * without deadlocking with linkwatch via keventd.
7490 * 2) Since we run with the RTNL semaphore not held, we can sleep
7491 * safely in order to wait for the netdev refcnt to drop to zero.
7493 * We must not return until all unregister events added during
7494 * the interval the lock was held have been completed.
7496 void netdev_run_todo(void)
7498 struct list_head list;
7500 /* Snapshot list, allow later requests */
7501 list_replace_init(&net_todo_list, &list);
7506 /* Wait for rcu callbacks to finish before next phase */
7507 if (!list_empty(&list))
7510 while (!list_empty(&list)) {
7511 struct net_device *dev
7512 = list_first_entry(&list, struct net_device, todo_list);
7513 list_del(&dev->todo_list);
7516 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7519 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
7520 pr_err("network todo '%s' but state %d\n",
7521 dev->name, dev->reg_state);
7526 dev->reg_state = NETREG_UNREGISTERED;
7528 netdev_wait_allrefs(dev);
7531 BUG_ON(netdev_refcnt_read(dev));
7532 BUG_ON(!list_empty(&dev->ptype_all));
7533 BUG_ON(!list_empty(&dev->ptype_specific));
7534 WARN_ON(rcu_access_pointer(dev->ip_ptr));
7535 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
7536 WARN_ON(dev->dn_ptr);
7538 if (dev->destructor)
7539 dev->destructor(dev);
7541 /* Report a network device has been unregistered */
7543 dev_net(dev)->dev_unreg_count--;
7545 wake_up(&netdev_unregistering_wq);
7547 /* Free network device */
7548 kobject_put(&dev->dev.kobj);
7552 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
7553 * all the same fields in the same order as net_device_stats, with only
7554 * the type differing, but rtnl_link_stats64 may have additional fields
7555 * at the end for newer counters.
7557 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7558 const struct net_device_stats *netdev_stats)
7560 #if BITS_PER_LONG == 64
7561 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
7562 memcpy(stats64, netdev_stats, sizeof(*stats64));
7563 /* zero out counters that only exist in rtnl_link_stats64 */
7564 memset((char *)stats64 + sizeof(*netdev_stats), 0,
7565 sizeof(*stats64) - sizeof(*netdev_stats));
7567 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
7568 const unsigned long *src = (const unsigned long *)netdev_stats;
7569 u64 *dst = (u64 *)stats64;
7571 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
7572 for (i = 0; i < n; i++)
7574 /* zero out counters that only exist in rtnl_link_stats64 */
7575 memset((char *)stats64 + n * sizeof(u64), 0,
7576 sizeof(*stats64) - n * sizeof(u64));
7579 EXPORT_SYMBOL(netdev_stats_to_stats64);
7582 * dev_get_stats - get network device statistics
7583 * @dev: device to get statistics from
7584 * @storage: place to store stats
7586 * Get network statistics from device. Return @storage.
7587 * The device driver may provide its own method by setting
7588 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7589 * otherwise the internal statistics structure is used.
7591 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
7592 struct rtnl_link_stats64 *storage)
7594 const struct net_device_ops *ops = dev->netdev_ops;
7596 if (ops->ndo_get_stats64) {
7597 memset(storage, 0, sizeof(*storage));
7598 ops->ndo_get_stats64(dev, storage);
7599 } else if (ops->ndo_get_stats) {
7600 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7602 netdev_stats_to_stats64(storage, &dev->stats);
7604 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
7605 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
7606 storage->rx_nohandler += atomic_long_read(&dev->rx_nohandler);
7609 EXPORT_SYMBOL(dev_get_stats);
7611 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7613 struct netdev_queue *queue = dev_ingress_queue(dev);
7615 #ifdef CONFIG_NET_CLS_ACT
7618 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7621 netdev_init_one_queue(dev, queue, NULL);
7622 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7623 queue->qdisc_sleeping = &noop_qdisc;
7624 rcu_assign_pointer(dev->ingress_queue, queue);
7629 static const struct ethtool_ops default_ethtool_ops;
7631 void netdev_set_default_ethtool_ops(struct net_device *dev,
7632 const struct ethtool_ops *ops)
7634 if (dev->ethtool_ops == &default_ethtool_ops)
7635 dev->ethtool_ops = ops;
7637 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7639 void netdev_freemem(struct net_device *dev)
7641 char *addr = (char *)dev - dev->padded;
7647 * alloc_netdev_mqs - allocate network device
7648 * @sizeof_priv: size of private data to allocate space for
7649 * @name: device name format string
7650 * @name_assign_type: origin of device name
7651 * @setup: callback to initialize device
7652 * @txqs: the number of TX subqueues to allocate
7653 * @rxqs: the number of RX subqueues to allocate
7655 * Allocates a struct net_device with private data area for driver use
7656 * and performs basic initialization. Also allocates subqueue structs
7657 * for each queue on the device.
7659 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7660 unsigned char name_assign_type,
7661 void (*setup)(struct net_device *),
7662 unsigned int txqs, unsigned int rxqs)
7664 struct net_device *dev;
7666 struct net_device *p;
7668 BUG_ON(strlen(name) >= sizeof(dev->name));
7671 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7677 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7682 alloc_size = sizeof(struct net_device);
7684 /* ensure 32-byte alignment of private area */
7685 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7686 alloc_size += sizeof_priv;
7688 /* ensure 32-byte alignment of whole construct */
7689 alloc_size += NETDEV_ALIGN - 1;
7691 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7693 p = vzalloc(alloc_size);
7697 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7698 dev->padded = (char *)dev - (char *)p;
7700 dev->pcpu_refcnt = alloc_percpu(int);
7701 if (!dev->pcpu_refcnt)
7704 if (dev_addr_init(dev))
7710 dev_net_set(dev, &init_net);
7712 dev->gso_max_size = GSO_MAX_SIZE;
7713 dev->gso_max_segs = GSO_MAX_SEGS;
7715 INIT_LIST_HEAD(&dev->napi_list);
7716 INIT_LIST_HEAD(&dev->unreg_list);
7717 INIT_LIST_HEAD(&dev->close_list);
7718 INIT_LIST_HEAD(&dev->link_watch_list);
7719 INIT_LIST_HEAD(&dev->adj_list.upper);
7720 INIT_LIST_HEAD(&dev->adj_list.lower);
7721 INIT_LIST_HEAD(&dev->ptype_all);
7722 INIT_LIST_HEAD(&dev->ptype_specific);
7723 #ifdef CONFIG_NET_SCHED
7724 hash_init(dev->qdisc_hash);
7726 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7729 if (!dev->tx_queue_len) {
7730 dev->priv_flags |= IFF_NO_QUEUE;
7731 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
7734 dev->num_tx_queues = txqs;
7735 dev->real_num_tx_queues = txqs;
7736 if (netif_alloc_netdev_queues(dev))
7740 dev->num_rx_queues = rxqs;
7741 dev->real_num_rx_queues = rxqs;
7742 if (netif_alloc_rx_queues(dev))
7746 strcpy(dev->name, name);
7747 dev->name_assign_type = name_assign_type;
7748 dev->group = INIT_NETDEV_GROUP;
7749 if (!dev->ethtool_ops)
7750 dev->ethtool_ops = &default_ethtool_ops;
7752 nf_hook_ingress_init(dev);
7761 free_percpu(dev->pcpu_refcnt);
7763 netdev_freemem(dev);
7766 EXPORT_SYMBOL(alloc_netdev_mqs);
7769 * free_netdev - free network device
7772 * This function does the last stage of destroying an allocated device
7773 * interface. The reference to the device object is released.
7774 * If this is the last reference then it will be freed.
7775 * Must be called in process context.
7777 void free_netdev(struct net_device *dev)
7779 struct napi_struct *p, *n;
7782 netif_free_tx_queues(dev);
7787 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7789 /* Flush device addresses */
7790 dev_addr_flush(dev);
7792 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7795 free_percpu(dev->pcpu_refcnt);
7796 dev->pcpu_refcnt = NULL;
7798 /* Compatibility with error handling in drivers */
7799 if (dev->reg_state == NETREG_UNINITIALIZED) {
7800 netdev_freemem(dev);
7804 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7805 dev->reg_state = NETREG_RELEASED;
7807 /* will free via device release */
7808 put_device(&dev->dev);
7810 EXPORT_SYMBOL(free_netdev);
7813 * synchronize_net - Synchronize with packet receive processing
7815 * Wait for packets currently being received to be done.
7816 * Does not block later packets from starting.
7818 void synchronize_net(void)
7821 if (rtnl_is_locked())
7822 synchronize_rcu_expedited();
7826 EXPORT_SYMBOL(synchronize_net);
7829 * unregister_netdevice_queue - remove device from the kernel
7833 * This function shuts down a device interface and removes it
7834 * from the kernel tables.
7835 * If head not NULL, device is queued to be unregistered later.
7837 * Callers must hold the rtnl semaphore. You may want
7838 * unregister_netdev() instead of this.
7841 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7846 list_move_tail(&dev->unreg_list, head);
7848 rollback_registered(dev);
7849 /* Finish processing unregister after unlock */
7853 EXPORT_SYMBOL(unregister_netdevice_queue);
7856 * unregister_netdevice_many - unregister many devices
7857 * @head: list of devices
7859 * Note: As most callers use a stack allocated list_head,
7860 * we force a list_del() to make sure stack wont be corrupted later.
7862 void unregister_netdevice_many(struct list_head *head)
7864 struct net_device *dev;
7866 if (!list_empty(head)) {
7867 rollback_registered_many(head);
7868 list_for_each_entry(dev, head, unreg_list)
7873 EXPORT_SYMBOL(unregister_netdevice_many);
7876 * unregister_netdev - remove device from the kernel
7879 * This function shuts down a device interface and removes it
7880 * from the kernel tables.
7882 * This is just a wrapper for unregister_netdevice that takes
7883 * the rtnl semaphore. In general you want to use this and not
7884 * unregister_netdevice.
7886 void unregister_netdev(struct net_device *dev)
7889 unregister_netdevice(dev);
7892 EXPORT_SYMBOL(unregister_netdev);
7895 * dev_change_net_namespace - move device to different nethost namespace
7897 * @net: network namespace
7898 * @pat: If not NULL name pattern to try if the current device name
7899 * is already taken in the destination network namespace.
7901 * This function shuts down a device interface and moves it
7902 * to a new network namespace. On success 0 is returned, on
7903 * a failure a netagive errno code is returned.
7905 * Callers must hold the rtnl semaphore.
7908 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7914 /* Don't allow namespace local devices to be moved. */
7916 if (dev->features & NETIF_F_NETNS_LOCAL)
7919 /* Ensure the device has been registrered */
7920 if (dev->reg_state != NETREG_REGISTERED)
7923 /* Get out if there is nothing todo */
7925 if (net_eq(dev_net(dev), net))
7928 /* Pick the destination device name, and ensure
7929 * we can use it in the destination network namespace.
7932 if (__dev_get_by_name(net, dev->name)) {
7933 /* We get here if we can't use the current device name */
7936 if (dev_get_valid_name(net, dev, pat) < 0)
7941 * And now a mini version of register_netdevice unregister_netdevice.
7944 /* If device is running close it first. */
7947 /* And unlink it from device chain */
7949 unlist_netdevice(dev);
7953 /* Shutdown queueing discipline. */
7956 /* Notify protocols, that we are about to destroy
7957 this device. They should clean all the things.
7959 Note that dev->reg_state stays at NETREG_REGISTERED.
7960 This is wanted because this way 8021q and macvlan know
7961 the device is just moving and can keep their slaves up.
7963 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7965 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7966 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7969 * Flush the unicast and multicast chains
7974 /* Send a netdev-removed uevent to the old namespace */
7975 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7976 netdev_adjacent_del_links(dev);
7978 /* Actually switch the network namespace */
7979 dev_net_set(dev, net);
7981 /* If there is an ifindex conflict assign a new one */
7982 if (__dev_get_by_index(net, dev->ifindex))
7983 dev->ifindex = dev_new_index(net);
7985 /* Send a netdev-add uevent to the new namespace */
7986 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7987 netdev_adjacent_add_links(dev);
7989 /* Fixup kobjects */
7990 err = device_rename(&dev->dev, dev->name);
7993 /* Add the device back in the hashes */
7994 list_netdevice(dev);
7996 /* Notify protocols, that a new device appeared. */
7997 call_netdevice_notifiers(NETDEV_REGISTER, dev);
8000 * Prevent userspace races by waiting until the network
8001 * device is fully setup before sending notifications.
8003 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
8010 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
8012 static int dev_cpu_dead(unsigned int oldcpu)
8014 struct sk_buff **list_skb;
8015 struct sk_buff *skb;
8017 struct softnet_data *sd, *oldsd;
8019 local_irq_disable();
8020 cpu = smp_processor_id();
8021 sd = &per_cpu(softnet_data, cpu);
8022 oldsd = &per_cpu(softnet_data, oldcpu);
8024 /* Find end of our completion_queue. */
8025 list_skb = &sd->completion_queue;
8027 list_skb = &(*list_skb)->next;
8028 /* Append completion queue from offline CPU. */
8029 *list_skb = oldsd->completion_queue;
8030 oldsd->completion_queue = NULL;
8032 /* Append output queue from offline CPU. */
8033 if (oldsd->output_queue) {
8034 *sd->output_queue_tailp = oldsd->output_queue;
8035 sd->output_queue_tailp = oldsd->output_queue_tailp;
8036 oldsd->output_queue = NULL;
8037 oldsd->output_queue_tailp = &oldsd->output_queue;
8039 /* Append NAPI poll list from offline CPU, with one exception :
8040 * process_backlog() must be called by cpu owning percpu backlog.
8041 * We properly handle process_queue & input_pkt_queue later.
8043 while (!list_empty(&oldsd->poll_list)) {
8044 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
8048 list_del_init(&napi->poll_list);
8049 if (napi->poll == process_backlog)
8052 ____napi_schedule(sd, napi);
8055 raise_softirq_irqoff(NET_TX_SOFTIRQ);
8058 /* Process offline CPU's input_pkt_queue */
8059 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
8061 input_queue_head_incr(oldsd);
8063 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
8065 input_queue_head_incr(oldsd);
8072 * netdev_increment_features - increment feature set by one
8073 * @all: current feature set
8074 * @one: new feature set
8075 * @mask: mask feature set
8077 * Computes a new feature set after adding a device with feature set
8078 * @one to the master device with current feature set @all. Will not
8079 * enable anything that is off in @mask. Returns the new feature set.
8081 netdev_features_t netdev_increment_features(netdev_features_t all,
8082 netdev_features_t one, netdev_features_t mask)
8084 if (mask & NETIF_F_HW_CSUM)
8085 mask |= NETIF_F_CSUM_MASK;
8086 mask |= NETIF_F_VLAN_CHALLENGED;
8088 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
8089 all &= one | ~NETIF_F_ALL_FOR_ALL;
8091 /* If one device supports hw checksumming, set for all. */
8092 if (all & NETIF_F_HW_CSUM)
8093 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
8097 EXPORT_SYMBOL(netdev_increment_features);
8099 static struct hlist_head * __net_init netdev_create_hash(void)
8102 struct hlist_head *hash;
8104 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
8106 for (i = 0; i < NETDEV_HASHENTRIES; i++)
8107 INIT_HLIST_HEAD(&hash[i]);
8112 /* Initialize per network namespace state */
8113 static int __net_init netdev_init(struct net *net)
8115 if (net != &init_net)
8116 INIT_LIST_HEAD(&net->dev_base_head);
8118 net->dev_name_head = netdev_create_hash();
8119 if (net->dev_name_head == NULL)
8122 net->dev_index_head = netdev_create_hash();
8123 if (net->dev_index_head == NULL)
8129 kfree(net->dev_name_head);
8135 * netdev_drivername - network driver for the device
8136 * @dev: network device
8138 * Determine network driver for device.
8140 const char *netdev_drivername(const struct net_device *dev)
8142 const struct device_driver *driver;
8143 const struct device *parent;
8144 const char *empty = "";
8146 parent = dev->dev.parent;
8150 driver = parent->driver;
8151 if (driver && driver->name)
8152 return driver->name;
8156 static void __netdev_printk(const char *level, const struct net_device *dev,
8157 struct va_format *vaf)
8159 if (dev && dev->dev.parent) {
8160 dev_printk_emit(level[1] - '0',
8163 dev_driver_string(dev->dev.parent),
8164 dev_name(dev->dev.parent),
8165 netdev_name(dev), netdev_reg_state(dev),
8168 printk("%s%s%s: %pV",
8169 level, netdev_name(dev), netdev_reg_state(dev), vaf);
8171 printk("%s(NULL net_device): %pV", level, vaf);
8175 void netdev_printk(const char *level, const struct net_device *dev,
8176 const char *format, ...)
8178 struct va_format vaf;
8181 va_start(args, format);
8186 __netdev_printk(level, dev, &vaf);
8190 EXPORT_SYMBOL(netdev_printk);
8192 #define define_netdev_printk_level(func, level) \
8193 void func(const struct net_device *dev, const char *fmt, ...) \
8195 struct va_format vaf; \
8198 va_start(args, fmt); \
8203 __netdev_printk(level, dev, &vaf); \
8207 EXPORT_SYMBOL(func);
8209 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
8210 define_netdev_printk_level(netdev_alert, KERN_ALERT);
8211 define_netdev_printk_level(netdev_crit, KERN_CRIT);
8212 define_netdev_printk_level(netdev_err, KERN_ERR);
8213 define_netdev_printk_level(netdev_warn, KERN_WARNING);
8214 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
8215 define_netdev_printk_level(netdev_info, KERN_INFO);
8217 static void __net_exit netdev_exit(struct net *net)
8219 kfree(net->dev_name_head);
8220 kfree(net->dev_index_head);
8223 static struct pernet_operations __net_initdata netdev_net_ops = {
8224 .init = netdev_init,
8225 .exit = netdev_exit,
8228 static void __net_exit default_device_exit(struct net *net)
8230 struct net_device *dev, *aux;
8232 * Push all migratable network devices back to the
8233 * initial network namespace
8236 for_each_netdev_safe(net, dev, aux) {
8238 char fb_name[IFNAMSIZ];
8240 /* Ignore unmoveable devices (i.e. loopback) */
8241 if (dev->features & NETIF_F_NETNS_LOCAL)
8244 /* Leave virtual devices for the generic cleanup */
8245 if (dev->rtnl_link_ops)
8248 /* Push remaining network devices to init_net */
8249 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
8250 err = dev_change_net_namespace(dev, &init_net, fb_name);
8252 pr_emerg("%s: failed to move %s to init_net: %d\n",
8253 __func__, dev->name, err);
8260 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
8262 /* Return with the rtnl_lock held when there are no network
8263 * devices unregistering in any network namespace in net_list.
8267 DEFINE_WAIT_FUNC(wait, woken_wake_function);
8269 add_wait_queue(&netdev_unregistering_wq, &wait);
8271 unregistering = false;
8273 list_for_each_entry(net, net_list, exit_list) {
8274 if (net->dev_unreg_count > 0) {
8275 unregistering = true;
8283 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
8285 remove_wait_queue(&netdev_unregistering_wq, &wait);
8288 static void __net_exit default_device_exit_batch(struct list_head *net_list)
8290 /* At exit all network devices most be removed from a network
8291 * namespace. Do this in the reverse order of registration.
8292 * Do this across as many network namespaces as possible to
8293 * improve batching efficiency.
8295 struct net_device *dev;
8297 LIST_HEAD(dev_kill_list);
8299 /* To prevent network device cleanup code from dereferencing
8300 * loopback devices or network devices that have been freed
8301 * wait here for all pending unregistrations to complete,
8302 * before unregistring the loopback device and allowing the
8303 * network namespace be freed.
8305 * The netdev todo list containing all network devices
8306 * unregistrations that happen in default_device_exit_batch
8307 * will run in the rtnl_unlock() at the end of
8308 * default_device_exit_batch.
8310 rtnl_lock_unregistering(net_list);
8311 list_for_each_entry(net, net_list, exit_list) {
8312 for_each_netdev_reverse(net, dev) {
8313 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
8314 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
8316 unregister_netdevice_queue(dev, &dev_kill_list);
8319 unregister_netdevice_many(&dev_kill_list);
8323 static struct pernet_operations __net_initdata default_device_ops = {
8324 .exit = default_device_exit,
8325 .exit_batch = default_device_exit_batch,
8329 * Initialize the DEV module. At boot time this walks the device list and
8330 * unhooks any devices that fail to initialise (normally hardware not
8331 * present) and leaves us with a valid list of present and active devices.
8336 * This is called single threaded during boot, so no need
8337 * to take the rtnl semaphore.
8339 static int __init net_dev_init(void)
8341 int i, rc = -ENOMEM;
8343 BUG_ON(!dev_boot_phase);
8345 if (dev_proc_init())
8348 if (netdev_kobject_init())
8351 INIT_LIST_HEAD(&ptype_all);
8352 for (i = 0; i < PTYPE_HASH_SIZE; i++)
8353 INIT_LIST_HEAD(&ptype_base[i]);
8355 INIT_LIST_HEAD(&offload_base);
8357 if (register_pernet_subsys(&netdev_net_ops))
8361 * Initialise the packet receive queues.
8364 for_each_possible_cpu(i) {
8365 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
8366 struct softnet_data *sd = &per_cpu(softnet_data, i);
8368 INIT_WORK(flush, flush_backlog);
8370 skb_queue_head_init(&sd->input_pkt_queue);
8371 skb_queue_head_init(&sd->process_queue);
8372 INIT_LIST_HEAD(&sd->poll_list);
8373 sd->output_queue_tailp = &sd->output_queue;
8375 sd->csd.func = rps_trigger_softirq;
8380 sd->backlog.poll = process_backlog;
8381 sd->backlog.weight = weight_p;
8386 /* The loopback device is special if any other network devices
8387 * is present in a network namespace the loopback device must
8388 * be present. Since we now dynamically allocate and free the
8389 * loopback device ensure this invariant is maintained by
8390 * keeping the loopback device as the first device on the
8391 * list of network devices. Ensuring the loopback devices
8392 * is the first device that appears and the last network device
8395 if (register_pernet_device(&loopback_net_ops))
8398 if (register_pernet_device(&default_device_ops))
8401 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
8402 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
8404 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
8405 NULL, dev_cpu_dead);
8413 subsys_initcall(net_dev_init);