2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/dst_metadata.h>
103 #include <net/pkt_sched.h>
104 #include <net/checksum.h>
105 #include <net/xfrm.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/module.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #include <linux/delay.h>
112 #include <net/iw_handler.h>
113 #include <asm/current.h>
114 #include <linux/audit.h>
115 #include <linux/dmaengine.h>
116 #include <linux/err.h>
117 #include <linux/ctype.h>
118 #include <linux/if_arp.h>
119 #include <linux/if_vlan.h>
120 #include <linux/ip.h>
122 #include <net/mpls.h>
123 #include <linux/ipv6.h>
124 #include <linux/in.h>
125 #include <linux/jhash.h>
126 #include <linux/random.h>
127 #include <trace/events/napi.h>
128 #include <trace/events/net.h>
129 #include <trace/events/skb.h>
130 #include <linux/pci.h>
131 #include <linux/inetdevice.h>
132 #include <linux/cpu_rmap.h>
133 #include <linux/static_key.h>
134 #include <linux/hashtable.h>
135 #include <linux/vmalloc.h>
136 #include <linux/if_macvlan.h>
137 #include <linux/errqueue.h>
138 #include <linux/hrtimer.h>
139 #include <linux/netfilter_ingress.h>
141 #include "net-sysfs.h"
143 /* Instead of increasing this, you should create a hash table. */
144 #define MAX_GRO_SKBS 8
146 /* This should be increased if a protocol with a bigger head is added. */
147 #define GRO_MAX_HEAD (MAX_HEADER + 128)
149 static DEFINE_SPINLOCK(ptype_lock);
150 static DEFINE_SPINLOCK(offload_lock);
151 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
152 struct list_head ptype_all __read_mostly; /* Taps */
153 static struct list_head offload_base __read_mostly;
155 static int netif_rx_internal(struct sk_buff *skb);
156 static int call_netdevice_notifiers_info(unsigned long val,
157 struct net_device *dev,
158 struct netdev_notifier_info *info);
161 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
164 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
166 * Writers must hold the rtnl semaphore while they loop through the
167 * dev_base_head list, and hold dev_base_lock for writing when they do the
168 * actual updates. This allows pure readers to access the list even
169 * while a writer is preparing to update it.
171 * To put it another way, dev_base_lock is held for writing only to
172 * protect against pure readers; the rtnl semaphore provides the
173 * protection against other writers.
175 * See, for example usages, register_netdevice() and
176 * unregister_netdevice(), which must be called with the rtnl
179 DEFINE_RWLOCK(dev_base_lock);
180 EXPORT_SYMBOL(dev_base_lock);
182 /* protects napi_hash addition/deletion and napi_gen_id */
183 static DEFINE_SPINLOCK(napi_hash_lock);
185 static unsigned int napi_gen_id = NR_CPUS;
186 static DEFINE_HASHTABLE(napi_hash, 8);
188 static seqcount_t devnet_rename_seq;
190 static inline void dev_base_seq_inc(struct net *net)
192 while (++net->dev_base_seq == 0);
195 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
197 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
199 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
202 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
204 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
207 static inline void rps_lock(struct softnet_data *sd)
210 spin_lock(&sd->input_pkt_queue.lock);
214 static inline void rps_unlock(struct softnet_data *sd)
217 spin_unlock(&sd->input_pkt_queue.lock);
221 /* Device list insertion */
222 static void list_netdevice(struct net_device *dev)
224 struct net *net = dev_net(dev);
228 write_lock_bh(&dev_base_lock);
229 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
230 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
231 hlist_add_head_rcu(&dev->index_hlist,
232 dev_index_hash(net, dev->ifindex));
233 write_unlock_bh(&dev_base_lock);
235 dev_base_seq_inc(net);
238 /* Device list removal
239 * caller must respect a RCU grace period before freeing/reusing dev
241 static void unlist_netdevice(struct net_device *dev)
245 /* Unlink dev from the device chain */
246 write_lock_bh(&dev_base_lock);
247 list_del_rcu(&dev->dev_list);
248 hlist_del_rcu(&dev->name_hlist);
249 hlist_del_rcu(&dev->index_hlist);
250 write_unlock_bh(&dev_base_lock);
252 dev_base_seq_inc(dev_net(dev));
259 static RAW_NOTIFIER_HEAD(netdev_chain);
262 * Device drivers call our routines to queue packets here. We empty the
263 * queue in the local softnet handler.
266 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
267 EXPORT_PER_CPU_SYMBOL(softnet_data);
269 #ifdef CONFIG_LOCKDEP
271 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
272 * according to dev->type
274 static const unsigned short netdev_lock_type[] =
275 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
276 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
277 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
278 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
279 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
280 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
281 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
282 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
283 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
284 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
285 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
286 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
287 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
288 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
289 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
291 static const char *const netdev_lock_name[] =
292 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
293 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
294 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
295 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
296 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
297 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
298 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
299 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
300 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
301 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
302 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
303 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
304 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
305 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
306 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
308 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
309 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
311 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
315 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
316 if (netdev_lock_type[i] == dev_type)
318 /* the last key is used by default */
319 return ARRAY_SIZE(netdev_lock_type) - 1;
322 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
323 unsigned short dev_type)
327 i = netdev_lock_pos(dev_type);
328 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
329 netdev_lock_name[i]);
332 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
336 i = netdev_lock_pos(dev->type);
337 lockdep_set_class_and_name(&dev->addr_list_lock,
338 &netdev_addr_lock_key[i],
339 netdev_lock_name[i]);
342 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
343 unsigned short dev_type)
346 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
351 /*******************************************************************************
353 Protocol management and registration routines
355 *******************************************************************************/
358 * Add a protocol ID to the list. Now that the input handler is
359 * smarter we can dispense with all the messy stuff that used to be
362 * BEWARE!!! Protocol handlers, mangling input packets,
363 * MUST BE last in hash buckets and checking protocol handlers
364 * MUST start from promiscuous ptype_all chain in net_bh.
365 * It is true now, do not change it.
366 * Explanation follows: if protocol handler, mangling packet, will
367 * be the first on list, it is not able to sense, that packet
368 * is cloned and should be copied-on-write, so that it will
369 * change it and subsequent readers will get broken packet.
373 static inline struct list_head *ptype_head(const struct packet_type *pt)
375 if (pt->type == htons(ETH_P_ALL))
376 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
378 return pt->dev ? &pt->dev->ptype_specific :
379 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
383 * dev_add_pack - add packet handler
384 * @pt: packet type declaration
386 * Add a protocol handler to the networking stack. The passed &packet_type
387 * is linked into kernel lists and may not be freed until it has been
388 * removed from the kernel lists.
390 * This call does not sleep therefore it can not
391 * guarantee all CPU's that are in middle of receiving packets
392 * will see the new packet type (until the next received packet).
395 void dev_add_pack(struct packet_type *pt)
397 struct list_head *head = ptype_head(pt);
399 spin_lock(&ptype_lock);
400 list_add_rcu(&pt->list, head);
401 spin_unlock(&ptype_lock);
403 EXPORT_SYMBOL(dev_add_pack);
406 * __dev_remove_pack - remove packet handler
407 * @pt: packet type declaration
409 * Remove a protocol handler that was previously added to the kernel
410 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
411 * from the kernel lists and can be freed or reused once this function
414 * The packet type might still be in use by receivers
415 * and must not be freed until after all the CPU's have gone
416 * through a quiescent state.
418 void __dev_remove_pack(struct packet_type *pt)
420 struct list_head *head = ptype_head(pt);
421 struct packet_type *pt1;
423 spin_lock(&ptype_lock);
425 list_for_each_entry(pt1, head, list) {
427 list_del_rcu(&pt->list);
432 pr_warn("dev_remove_pack: %p not found\n", pt);
434 spin_unlock(&ptype_lock);
436 EXPORT_SYMBOL(__dev_remove_pack);
439 * dev_remove_pack - remove packet handler
440 * @pt: packet type declaration
442 * Remove a protocol handler that was previously added to the kernel
443 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
444 * from the kernel lists and can be freed or reused once this function
447 * This call sleeps to guarantee that no CPU is looking at the packet
450 void dev_remove_pack(struct packet_type *pt)
452 __dev_remove_pack(pt);
456 EXPORT_SYMBOL(dev_remove_pack);
460 * dev_add_offload - register offload handlers
461 * @po: protocol offload declaration
463 * Add protocol offload handlers to the networking stack. The passed
464 * &proto_offload is linked into kernel lists and may not be freed until
465 * it has been removed from the kernel lists.
467 * This call does not sleep therefore it can not
468 * guarantee all CPU's that are in middle of receiving packets
469 * will see the new offload handlers (until the next received packet).
471 void dev_add_offload(struct packet_offload *po)
473 struct packet_offload *elem;
475 spin_lock(&offload_lock);
476 list_for_each_entry(elem, &offload_base, list) {
477 if (po->priority < elem->priority)
480 list_add_rcu(&po->list, elem->list.prev);
481 spin_unlock(&offload_lock);
483 EXPORT_SYMBOL(dev_add_offload);
486 * __dev_remove_offload - remove offload handler
487 * @po: packet offload declaration
489 * Remove a protocol offload handler that was previously added to the
490 * kernel offload handlers by dev_add_offload(). The passed &offload_type
491 * is removed from the kernel lists and can be freed or reused once this
494 * The packet type might still be in use by receivers
495 * and must not be freed until after all the CPU's have gone
496 * through a quiescent state.
498 static void __dev_remove_offload(struct packet_offload *po)
500 struct list_head *head = &offload_base;
501 struct packet_offload *po1;
503 spin_lock(&offload_lock);
505 list_for_each_entry(po1, head, list) {
507 list_del_rcu(&po->list);
512 pr_warn("dev_remove_offload: %p not found\n", po);
514 spin_unlock(&offload_lock);
518 * dev_remove_offload - remove packet offload handler
519 * @po: packet offload declaration
521 * Remove a packet offload handler that was previously added to the kernel
522 * offload handlers by dev_add_offload(). The passed &offload_type is
523 * removed from the kernel lists and can be freed or reused once this
526 * This call sleeps to guarantee that no CPU is looking at the packet
529 void dev_remove_offload(struct packet_offload *po)
531 __dev_remove_offload(po);
535 EXPORT_SYMBOL(dev_remove_offload);
537 /******************************************************************************
539 Device Boot-time Settings Routines
541 *******************************************************************************/
543 /* Boot time configuration table */
544 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
547 * netdev_boot_setup_add - add new setup entry
548 * @name: name of the device
549 * @map: configured settings for the device
551 * Adds new setup entry to the dev_boot_setup list. The function
552 * returns 0 on error and 1 on success. This is a generic routine to
555 static int netdev_boot_setup_add(char *name, struct ifmap *map)
557 struct netdev_boot_setup *s;
561 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
562 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
563 memset(s[i].name, 0, sizeof(s[i].name));
564 strlcpy(s[i].name, name, IFNAMSIZ);
565 memcpy(&s[i].map, map, sizeof(s[i].map));
570 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
574 * netdev_boot_setup_check - check boot time settings
575 * @dev: the netdevice
577 * Check boot time settings for the device.
578 * The found settings are set for the device to be used
579 * later in the device probing.
580 * Returns 0 if no settings found, 1 if they are.
582 int netdev_boot_setup_check(struct net_device *dev)
584 struct netdev_boot_setup *s = dev_boot_setup;
587 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
588 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
589 !strcmp(dev->name, s[i].name)) {
590 dev->irq = s[i].map.irq;
591 dev->base_addr = s[i].map.base_addr;
592 dev->mem_start = s[i].map.mem_start;
593 dev->mem_end = s[i].map.mem_end;
599 EXPORT_SYMBOL(netdev_boot_setup_check);
603 * netdev_boot_base - get address from boot time settings
604 * @prefix: prefix for network device
605 * @unit: id for network device
607 * Check boot time settings for the base address of device.
608 * The found settings are set for the device to be used
609 * later in the device probing.
610 * Returns 0 if no settings found.
612 unsigned long netdev_boot_base(const char *prefix, int unit)
614 const struct netdev_boot_setup *s = dev_boot_setup;
618 sprintf(name, "%s%d", prefix, unit);
621 * If device already registered then return base of 1
622 * to indicate not to probe for this interface
624 if (__dev_get_by_name(&init_net, name))
627 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
628 if (!strcmp(name, s[i].name))
629 return s[i].map.base_addr;
634 * Saves at boot time configured settings for any netdevice.
636 int __init netdev_boot_setup(char *str)
641 str = get_options(str, ARRAY_SIZE(ints), ints);
646 memset(&map, 0, sizeof(map));
650 map.base_addr = ints[2];
652 map.mem_start = ints[3];
654 map.mem_end = ints[4];
656 /* Add new entry to the list */
657 return netdev_boot_setup_add(str, &map);
660 __setup("netdev=", netdev_boot_setup);
662 /*******************************************************************************
664 Device Interface Subroutines
666 *******************************************************************************/
669 * dev_get_iflink - get 'iflink' value of a interface
670 * @dev: targeted interface
672 * Indicates the ifindex the interface is linked to.
673 * Physical interfaces have the same 'ifindex' and 'iflink' values.
676 int dev_get_iflink(const struct net_device *dev)
678 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
679 return dev->netdev_ops->ndo_get_iflink(dev);
683 EXPORT_SYMBOL(dev_get_iflink);
686 * dev_fill_metadata_dst - Retrieve tunnel egress information.
687 * @dev: targeted interface
690 * For better visibility of tunnel traffic OVS needs to retrieve
691 * egress tunnel information for a packet. Following API allows
692 * user to get this info.
694 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
696 struct ip_tunnel_info *info;
698 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
701 info = skb_tunnel_info_unclone(skb);
704 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
707 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
709 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
712 * __dev_get_by_name - find a device by its name
713 * @net: the applicable net namespace
714 * @name: name to find
716 * Find an interface by name. Must be called under RTNL semaphore
717 * or @dev_base_lock. If the name is found a pointer to the device
718 * is returned. If the name is not found then %NULL is returned. The
719 * reference counters are not incremented so the caller must be
720 * careful with locks.
723 struct net_device *__dev_get_by_name(struct net *net, const char *name)
725 struct net_device *dev;
726 struct hlist_head *head = dev_name_hash(net, name);
728 hlist_for_each_entry(dev, head, name_hlist)
729 if (!strncmp(dev->name, name, IFNAMSIZ))
734 EXPORT_SYMBOL(__dev_get_by_name);
737 * dev_get_by_name_rcu - find a device by its name
738 * @net: the applicable net namespace
739 * @name: name to find
741 * Find an interface by name.
742 * If the name is found a pointer to the device is returned.
743 * If the name is not found then %NULL is returned.
744 * The reference counters are not incremented so the caller must be
745 * careful with locks. The caller must hold RCU lock.
748 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
750 struct net_device *dev;
751 struct hlist_head *head = dev_name_hash(net, name);
753 hlist_for_each_entry_rcu(dev, head, name_hlist)
754 if (!strncmp(dev->name, name, IFNAMSIZ))
759 EXPORT_SYMBOL(dev_get_by_name_rcu);
762 * dev_get_by_name - find a device by its name
763 * @net: the applicable net namespace
764 * @name: name to find
766 * Find an interface by name. This can be called from any
767 * context and does its own locking. The returned handle has
768 * the usage count incremented and the caller must use dev_put() to
769 * release it when it is no longer needed. %NULL is returned if no
770 * matching device is found.
773 struct net_device *dev_get_by_name(struct net *net, const char *name)
775 struct net_device *dev;
778 dev = dev_get_by_name_rcu(net, name);
784 EXPORT_SYMBOL(dev_get_by_name);
787 * __dev_get_by_index - find a device by its ifindex
788 * @net: the applicable net namespace
789 * @ifindex: index of device
791 * Search for an interface by index. Returns %NULL if the device
792 * is not found or a pointer to the device. The device has not
793 * had its reference counter increased so the caller must be careful
794 * about locking. The caller must hold either the RTNL semaphore
798 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
800 struct net_device *dev;
801 struct hlist_head *head = dev_index_hash(net, ifindex);
803 hlist_for_each_entry(dev, head, index_hlist)
804 if (dev->ifindex == ifindex)
809 EXPORT_SYMBOL(__dev_get_by_index);
812 * dev_get_by_index_rcu - find a device by its ifindex
813 * @net: the applicable net namespace
814 * @ifindex: index of device
816 * Search for an interface by index. Returns %NULL if the device
817 * is not found or a pointer to the device. The device has not
818 * had its reference counter increased so the caller must be careful
819 * about locking. The caller must hold RCU lock.
822 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
824 struct net_device *dev;
825 struct hlist_head *head = dev_index_hash(net, ifindex);
827 hlist_for_each_entry_rcu(dev, head, index_hlist)
828 if (dev->ifindex == ifindex)
833 EXPORT_SYMBOL(dev_get_by_index_rcu);
837 * dev_get_by_index - find a device by its ifindex
838 * @net: the applicable net namespace
839 * @ifindex: index of device
841 * Search for an interface by index. Returns NULL if the device
842 * is not found or a pointer to the device. The device returned has
843 * had a reference added and the pointer is safe until the user calls
844 * dev_put to indicate they have finished with it.
847 struct net_device *dev_get_by_index(struct net *net, int ifindex)
849 struct net_device *dev;
852 dev = dev_get_by_index_rcu(net, ifindex);
858 EXPORT_SYMBOL(dev_get_by_index);
861 * netdev_get_name - get a netdevice name, knowing its ifindex.
862 * @net: network namespace
863 * @name: a pointer to the buffer where the name will be stored.
864 * @ifindex: the ifindex of the interface to get the name from.
866 * The use of raw_seqcount_begin() and cond_resched() before
867 * retrying is required as we want to give the writers a chance
868 * to complete when CONFIG_PREEMPT is not set.
870 int netdev_get_name(struct net *net, char *name, int ifindex)
872 struct net_device *dev;
876 seq = raw_seqcount_begin(&devnet_rename_seq);
878 dev = dev_get_by_index_rcu(net, ifindex);
884 strcpy(name, dev->name);
886 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
895 * dev_getbyhwaddr_rcu - find a device by its hardware address
896 * @net: the applicable net namespace
897 * @type: media type of device
898 * @ha: hardware address
900 * Search for an interface by MAC address. Returns NULL if the device
901 * is not found or a pointer to the device.
902 * The caller must hold RCU or RTNL.
903 * The returned device has not had its ref count increased
904 * and the caller must therefore be careful about locking
908 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
911 struct net_device *dev;
913 for_each_netdev_rcu(net, dev)
914 if (dev->type == type &&
915 !memcmp(dev->dev_addr, ha, dev->addr_len))
920 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
922 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
924 struct net_device *dev;
927 for_each_netdev(net, dev)
928 if (dev->type == type)
933 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
935 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
937 struct net_device *dev, *ret = NULL;
940 for_each_netdev_rcu(net, dev)
941 if (dev->type == type) {
949 EXPORT_SYMBOL(dev_getfirstbyhwtype);
952 * __dev_get_by_flags - find any device with given flags
953 * @net: the applicable net namespace
954 * @if_flags: IFF_* values
955 * @mask: bitmask of bits in if_flags to check
957 * Search for any interface with the given flags. Returns NULL if a device
958 * is not found or a pointer to the device. Must be called inside
959 * rtnl_lock(), and result refcount is unchanged.
962 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
965 struct net_device *dev, *ret;
970 for_each_netdev(net, dev) {
971 if (((dev->flags ^ if_flags) & mask) == 0) {
978 EXPORT_SYMBOL(__dev_get_by_flags);
981 * dev_valid_name - check if name is okay for network device
984 * Network device names need to be valid file names to
985 * to allow sysfs to work. We also disallow any kind of
988 bool dev_valid_name(const char *name)
992 if (strlen(name) >= IFNAMSIZ)
994 if (!strcmp(name, ".") || !strcmp(name, ".."))
998 if (*name == '/' || *name == ':' || isspace(*name))
1004 EXPORT_SYMBOL(dev_valid_name);
1007 * __dev_alloc_name - allocate a name for a device
1008 * @net: network namespace to allocate the device name in
1009 * @name: name format string
1010 * @buf: scratch buffer and result name string
1012 * Passed a format string - eg "lt%d" it will try and find a suitable
1013 * id. It scans list of devices to build up a free map, then chooses
1014 * the first empty slot. The caller must hold the dev_base or rtnl lock
1015 * while allocating the name and adding the device in order to avoid
1017 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1018 * Returns the number of the unit assigned or a negative errno code.
1021 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1025 const int max_netdevices = 8*PAGE_SIZE;
1026 unsigned long *inuse;
1027 struct net_device *d;
1029 p = strnchr(name, IFNAMSIZ-1, '%');
1032 * Verify the string as this thing may have come from
1033 * the user. There must be either one "%d" and no other "%"
1036 if (p[1] != 'd' || strchr(p + 2, '%'))
1039 /* Use one page as a bit array of possible slots */
1040 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1044 for_each_netdev(net, d) {
1045 if (!sscanf(d->name, name, &i))
1047 if (i < 0 || i >= max_netdevices)
1050 /* avoid cases where sscanf is not exact inverse of printf */
1051 snprintf(buf, IFNAMSIZ, name, i);
1052 if (!strncmp(buf, d->name, IFNAMSIZ))
1056 i = find_first_zero_bit(inuse, max_netdevices);
1057 free_page((unsigned long) inuse);
1061 snprintf(buf, IFNAMSIZ, name, i);
1062 if (!__dev_get_by_name(net, buf))
1065 /* It is possible to run out of possible slots
1066 * when the name is long and there isn't enough space left
1067 * for the digits, or if all bits are used.
1073 * dev_alloc_name - allocate a name for a device
1075 * @name: name format string
1077 * Passed a format string - eg "lt%d" it will try and find a suitable
1078 * id. It scans list of devices to build up a free map, then chooses
1079 * the first empty slot. The caller must hold the dev_base or rtnl lock
1080 * while allocating the name and adding the device in order to avoid
1082 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1083 * Returns the number of the unit assigned or a negative errno code.
1086 int dev_alloc_name(struct net_device *dev, const char *name)
1092 BUG_ON(!dev_net(dev));
1094 ret = __dev_alloc_name(net, name, buf);
1096 strlcpy(dev->name, buf, IFNAMSIZ);
1099 EXPORT_SYMBOL(dev_alloc_name);
1101 static int dev_alloc_name_ns(struct net *net,
1102 struct net_device *dev,
1108 ret = __dev_alloc_name(net, name, buf);
1110 strlcpy(dev->name, buf, IFNAMSIZ);
1114 int dev_get_valid_name(struct net *net, struct net_device *dev,
1119 if (!dev_valid_name(name))
1122 if (strchr(name, '%'))
1123 return dev_alloc_name_ns(net, dev, name);
1124 else if (__dev_get_by_name(net, name))
1126 else if (dev->name != name)
1127 strlcpy(dev->name, name, IFNAMSIZ);
1131 EXPORT_SYMBOL(dev_get_valid_name);
1134 * dev_change_name - change name of a device
1136 * @newname: name (or format string) must be at least IFNAMSIZ
1138 * Change name of a device, can pass format strings "eth%d".
1141 int dev_change_name(struct net_device *dev, const char *newname)
1143 unsigned char old_assign_type;
1144 char oldname[IFNAMSIZ];
1150 BUG_ON(!dev_net(dev));
1153 if (dev->flags & IFF_UP)
1156 write_seqcount_begin(&devnet_rename_seq);
1158 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1159 write_seqcount_end(&devnet_rename_seq);
1163 memcpy(oldname, dev->name, IFNAMSIZ);
1165 err = dev_get_valid_name(net, dev, newname);
1167 write_seqcount_end(&devnet_rename_seq);
1171 if (oldname[0] && !strchr(oldname, '%'))
1172 netdev_info(dev, "renamed from %s\n", oldname);
1174 old_assign_type = dev->name_assign_type;
1175 dev->name_assign_type = NET_NAME_RENAMED;
1178 ret = device_rename(&dev->dev, dev->name);
1180 memcpy(dev->name, oldname, IFNAMSIZ);
1181 dev->name_assign_type = old_assign_type;
1182 write_seqcount_end(&devnet_rename_seq);
1186 write_seqcount_end(&devnet_rename_seq);
1188 netdev_adjacent_rename_links(dev, oldname);
1190 write_lock_bh(&dev_base_lock);
1191 hlist_del_rcu(&dev->name_hlist);
1192 write_unlock_bh(&dev_base_lock);
1196 write_lock_bh(&dev_base_lock);
1197 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1198 write_unlock_bh(&dev_base_lock);
1200 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1201 ret = notifier_to_errno(ret);
1204 /* err >= 0 after dev_alloc_name() or stores the first errno */
1207 write_seqcount_begin(&devnet_rename_seq);
1208 memcpy(dev->name, oldname, IFNAMSIZ);
1209 memcpy(oldname, newname, IFNAMSIZ);
1210 dev->name_assign_type = old_assign_type;
1211 old_assign_type = NET_NAME_RENAMED;
1214 pr_err("%s: name change rollback failed: %d\n",
1223 * dev_set_alias - change ifalias of a device
1225 * @alias: name up to IFALIASZ
1226 * @len: limit of bytes to copy from info
1228 * Set ifalias for a device,
1230 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1236 if (len >= IFALIASZ)
1240 kfree(dev->ifalias);
1241 dev->ifalias = NULL;
1245 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1248 dev->ifalias = new_ifalias;
1249 memcpy(dev->ifalias, alias, len);
1250 dev->ifalias[len] = 0;
1257 * netdev_features_change - device changes features
1258 * @dev: device to cause notification
1260 * Called to indicate a device has changed features.
1262 void netdev_features_change(struct net_device *dev)
1264 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1266 EXPORT_SYMBOL(netdev_features_change);
1269 * netdev_state_change - device changes state
1270 * @dev: device to cause notification
1272 * Called to indicate a device has changed state. This function calls
1273 * the notifier chains for netdev_chain and sends a NEWLINK message
1274 * to the routing socket.
1276 void netdev_state_change(struct net_device *dev)
1278 if (dev->flags & IFF_UP) {
1279 struct netdev_notifier_change_info change_info;
1281 change_info.flags_changed = 0;
1282 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1284 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1287 EXPORT_SYMBOL(netdev_state_change);
1290 * netdev_notify_peers - notify network peers about existence of @dev
1291 * @dev: network device
1293 * Generate traffic such that interested network peers are aware of
1294 * @dev, such as by generating a gratuitous ARP. This may be used when
1295 * a device wants to inform the rest of the network about some sort of
1296 * reconfiguration such as a failover event or virtual machine
1299 void netdev_notify_peers(struct net_device *dev)
1302 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1303 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1306 EXPORT_SYMBOL(netdev_notify_peers);
1308 static int __dev_open(struct net_device *dev)
1310 const struct net_device_ops *ops = dev->netdev_ops;
1315 if (!netif_device_present(dev))
1318 /* Block netpoll from trying to do any rx path servicing.
1319 * If we don't do this there is a chance ndo_poll_controller
1320 * or ndo_poll may be running while we open the device
1322 netpoll_poll_disable(dev);
1324 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1325 ret = notifier_to_errno(ret);
1329 set_bit(__LINK_STATE_START, &dev->state);
1331 if (ops->ndo_validate_addr)
1332 ret = ops->ndo_validate_addr(dev);
1334 if (!ret && ops->ndo_open)
1335 ret = ops->ndo_open(dev);
1337 netpoll_poll_enable(dev);
1340 clear_bit(__LINK_STATE_START, &dev->state);
1342 dev->flags |= IFF_UP;
1343 dev_set_rx_mode(dev);
1345 add_device_randomness(dev->dev_addr, dev->addr_len);
1352 * dev_open - prepare an interface for use.
1353 * @dev: device to open
1355 * Takes a device from down to up state. The device's private open
1356 * function is invoked and then the multicast lists are loaded. Finally
1357 * the device is moved into the up state and a %NETDEV_UP message is
1358 * sent to the netdev notifier chain.
1360 * Calling this function on an active interface is a nop. On a failure
1361 * a negative errno code is returned.
1363 int dev_open(struct net_device *dev)
1367 if (dev->flags & IFF_UP)
1370 ret = __dev_open(dev);
1374 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1375 call_netdevice_notifiers(NETDEV_UP, dev);
1379 EXPORT_SYMBOL(dev_open);
1381 static int __dev_close_many(struct list_head *head)
1383 struct net_device *dev;
1388 list_for_each_entry(dev, head, close_list) {
1389 /* Temporarily disable netpoll until the interface is down */
1390 netpoll_poll_disable(dev);
1392 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1394 clear_bit(__LINK_STATE_START, &dev->state);
1396 /* Synchronize to scheduled poll. We cannot touch poll list, it
1397 * can be even on different cpu. So just clear netif_running().
1399 * dev->stop() will invoke napi_disable() on all of it's
1400 * napi_struct instances on this device.
1402 smp_mb__after_atomic(); /* Commit netif_running(). */
1405 dev_deactivate_many(head);
1407 list_for_each_entry(dev, head, close_list) {
1408 const struct net_device_ops *ops = dev->netdev_ops;
1411 * Call the device specific close. This cannot fail.
1412 * Only if device is UP
1414 * We allow it to be called even after a DETACH hot-plug
1420 dev->flags &= ~IFF_UP;
1421 netpoll_poll_enable(dev);
1427 static int __dev_close(struct net_device *dev)
1432 list_add(&dev->close_list, &single);
1433 retval = __dev_close_many(&single);
1439 int dev_close_many(struct list_head *head, bool unlink)
1441 struct net_device *dev, *tmp;
1443 /* Remove the devices that don't need to be closed */
1444 list_for_each_entry_safe(dev, tmp, head, close_list)
1445 if (!(dev->flags & IFF_UP))
1446 list_del_init(&dev->close_list);
1448 __dev_close_many(head);
1450 list_for_each_entry_safe(dev, tmp, head, close_list) {
1451 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1452 call_netdevice_notifiers(NETDEV_DOWN, dev);
1454 list_del_init(&dev->close_list);
1459 EXPORT_SYMBOL(dev_close_many);
1462 * dev_close - shutdown an interface.
1463 * @dev: device to shutdown
1465 * This function moves an active device into down state. A
1466 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1467 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1470 int dev_close(struct net_device *dev)
1472 if (dev->flags & IFF_UP) {
1475 list_add(&dev->close_list, &single);
1476 dev_close_many(&single, true);
1481 EXPORT_SYMBOL(dev_close);
1485 * dev_disable_lro - disable Large Receive Offload on a device
1488 * Disable Large Receive Offload (LRO) on a net device. Must be
1489 * called under RTNL. This is needed if received packets may be
1490 * forwarded to another interface.
1492 void dev_disable_lro(struct net_device *dev)
1494 struct net_device *lower_dev;
1495 struct list_head *iter;
1497 dev->wanted_features &= ~NETIF_F_LRO;
1498 netdev_update_features(dev);
1500 if (unlikely(dev->features & NETIF_F_LRO))
1501 netdev_WARN(dev, "failed to disable LRO!\n");
1503 netdev_for_each_lower_dev(dev, lower_dev, iter)
1504 dev_disable_lro(lower_dev);
1506 EXPORT_SYMBOL(dev_disable_lro);
1508 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1509 struct net_device *dev)
1511 struct netdev_notifier_info info;
1513 netdev_notifier_info_init(&info, dev);
1514 return nb->notifier_call(nb, val, &info);
1517 static int dev_boot_phase = 1;
1520 * register_netdevice_notifier - register a network notifier block
1523 * Register a notifier to be called when network device events occur.
1524 * The notifier passed is linked into the kernel structures and must
1525 * not be reused until it has been unregistered. A negative errno code
1526 * is returned on a failure.
1528 * When registered all registration and up events are replayed
1529 * to the new notifier to allow device to have a race free
1530 * view of the network device list.
1533 int register_netdevice_notifier(struct notifier_block *nb)
1535 struct net_device *dev;
1536 struct net_device *last;
1541 err = raw_notifier_chain_register(&netdev_chain, nb);
1547 for_each_netdev(net, dev) {
1548 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1549 err = notifier_to_errno(err);
1553 if (!(dev->flags & IFF_UP))
1556 call_netdevice_notifier(nb, NETDEV_UP, dev);
1567 for_each_netdev(net, dev) {
1571 if (dev->flags & IFF_UP) {
1572 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1574 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1576 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1581 raw_notifier_chain_unregister(&netdev_chain, nb);
1584 EXPORT_SYMBOL(register_netdevice_notifier);
1587 * unregister_netdevice_notifier - unregister a network notifier block
1590 * Unregister a notifier previously registered by
1591 * register_netdevice_notifier(). The notifier is unlinked into the
1592 * kernel structures and may then be reused. A negative errno code
1593 * is returned on a failure.
1595 * After unregistering unregister and down device events are synthesized
1596 * for all devices on the device list to the removed notifier to remove
1597 * the need for special case cleanup code.
1600 int unregister_netdevice_notifier(struct notifier_block *nb)
1602 struct net_device *dev;
1607 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1612 for_each_netdev(net, dev) {
1613 if (dev->flags & IFF_UP) {
1614 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1616 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1618 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1625 EXPORT_SYMBOL(unregister_netdevice_notifier);
1628 * call_netdevice_notifiers_info - call all network notifier blocks
1629 * @val: value passed unmodified to notifier function
1630 * @dev: net_device pointer passed unmodified to notifier function
1631 * @info: notifier information data
1633 * Call all network notifier blocks. Parameters and return value
1634 * are as for raw_notifier_call_chain().
1637 static int call_netdevice_notifiers_info(unsigned long val,
1638 struct net_device *dev,
1639 struct netdev_notifier_info *info)
1642 netdev_notifier_info_init(info, dev);
1643 return raw_notifier_call_chain(&netdev_chain, val, info);
1647 * call_netdevice_notifiers - call all network notifier blocks
1648 * @val: value passed unmodified to notifier function
1649 * @dev: net_device pointer passed unmodified to notifier function
1651 * Call all network notifier blocks. Parameters and return value
1652 * are as for raw_notifier_call_chain().
1655 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1657 struct netdev_notifier_info info;
1659 return call_netdevice_notifiers_info(val, dev, &info);
1661 EXPORT_SYMBOL(call_netdevice_notifiers);
1663 #ifdef CONFIG_NET_INGRESS
1664 static struct static_key ingress_needed __read_mostly;
1666 void net_inc_ingress_queue(void)
1668 static_key_slow_inc(&ingress_needed);
1670 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1672 void net_dec_ingress_queue(void)
1674 static_key_slow_dec(&ingress_needed);
1676 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1679 static struct static_key netstamp_needed __read_mostly;
1680 #ifdef HAVE_JUMP_LABEL
1681 static atomic_t netstamp_needed_deferred;
1682 static atomic_t netstamp_wanted;
1683 static void netstamp_clear(struct work_struct *work)
1685 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1688 wanted = atomic_add_return(deferred, &netstamp_wanted);
1690 static_key_enable(&netstamp_needed);
1692 static_key_disable(&netstamp_needed);
1694 static DECLARE_WORK(netstamp_work, netstamp_clear);
1697 void net_enable_timestamp(void)
1699 #ifdef HAVE_JUMP_LABEL
1703 wanted = atomic_read(&netstamp_wanted);
1706 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1709 atomic_inc(&netstamp_needed_deferred);
1710 schedule_work(&netstamp_work);
1712 static_key_slow_inc(&netstamp_needed);
1715 EXPORT_SYMBOL(net_enable_timestamp);
1717 void net_disable_timestamp(void)
1719 #ifdef HAVE_JUMP_LABEL
1723 wanted = atomic_read(&netstamp_wanted);
1726 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
1729 atomic_dec(&netstamp_needed_deferred);
1730 schedule_work(&netstamp_work);
1732 static_key_slow_dec(&netstamp_needed);
1735 EXPORT_SYMBOL(net_disable_timestamp);
1737 static inline void net_timestamp_set(struct sk_buff *skb)
1739 skb->tstamp.tv64 = 0;
1740 if (static_key_false(&netstamp_needed))
1741 __net_timestamp(skb);
1744 #define net_timestamp_check(COND, SKB) \
1745 if (static_key_false(&netstamp_needed)) { \
1746 if ((COND) && !(SKB)->tstamp.tv64) \
1747 __net_timestamp(SKB); \
1750 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1754 if (!(dev->flags & IFF_UP))
1757 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1758 if (skb->len <= len)
1761 /* if TSO is enabled, we don't care about the length as the packet
1762 * could be forwarded without being segmented before
1764 if (skb_is_gso(skb))
1769 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1771 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1773 if (skb_orphan_frags(skb, GFP_ATOMIC) ||
1774 unlikely(!is_skb_forwardable(dev, skb))) {
1775 atomic_long_inc(&dev->rx_dropped);
1780 skb_scrub_packet(skb, true);
1782 skb->protocol = eth_type_trans(skb, dev);
1783 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1787 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1790 * dev_forward_skb - loopback an skb to another netif
1792 * @dev: destination network device
1793 * @skb: buffer to forward
1796 * NET_RX_SUCCESS (no congestion)
1797 * NET_RX_DROP (packet was dropped, but freed)
1799 * dev_forward_skb can be used for injecting an skb from the
1800 * start_xmit function of one device into the receive queue
1801 * of another device.
1803 * The receiving device may be in another namespace, so
1804 * we have to clear all information in the skb that could
1805 * impact namespace isolation.
1807 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1809 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1811 EXPORT_SYMBOL_GPL(dev_forward_skb);
1813 static inline int deliver_skb(struct sk_buff *skb,
1814 struct packet_type *pt_prev,
1815 struct net_device *orig_dev)
1817 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1819 atomic_inc(&skb->users);
1820 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1823 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1824 struct packet_type **pt,
1825 struct net_device *orig_dev,
1827 struct list_head *ptype_list)
1829 struct packet_type *ptype, *pt_prev = *pt;
1831 list_for_each_entry_rcu(ptype, ptype_list, list) {
1832 if (ptype->type != type)
1835 deliver_skb(skb, pt_prev, orig_dev);
1841 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1843 if (!ptype->af_packet_priv || !skb->sk)
1846 if (ptype->id_match)
1847 return ptype->id_match(ptype, skb->sk);
1848 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1855 * Support routine. Sends outgoing frames to any network
1856 * taps currently in use.
1859 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1861 struct packet_type *ptype;
1862 struct sk_buff *skb2 = NULL;
1863 struct packet_type *pt_prev = NULL;
1864 struct list_head *ptype_list = &ptype_all;
1868 list_for_each_entry_rcu(ptype, ptype_list, list) {
1869 /* Never send packets back to the socket
1870 * they originated from - MvS (miquels@drinkel.ow.org)
1872 if (skb_loop_sk(ptype, skb))
1876 deliver_skb(skb2, pt_prev, skb->dev);
1881 /* need to clone skb, done only once */
1882 skb2 = skb_clone(skb, GFP_ATOMIC);
1886 net_timestamp_set(skb2);
1888 /* skb->nh should be correctly
1889 * set by sender, so that the second statement is
1890 * just protection against buggy protocols.
1892 skb_reset_mac_header(skb2);
1894 if (skb_network_header(skb2) < skb2->data ||
1895 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1896 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1897 ntohs(skb2->protocol),
1899 skb_reset_network_header(skb2);
1902 skb2->transport_header = skb2->network_header;
1903 skb2->pkt_type = PACKET_OUTGOING;
1907 if (ptype_list == &ptype_all) {
1908 ptype_list = &dev->ptype_all;
1913 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1918 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1919 * @dev: Network device
1920 * @txq: number of queues available
1922 * If real_num_tx_queues is changed the tc mappings may no longer be
1923 * valid. To resolve this verify the tc mapping remains valid and if
1924 * not NULL the mapping. With no priorities mapping to this
1925 * offset/count pair it will no longer be used. In the worst case TC0
1926 * is invalid nothing can be done so disable priority mappings. If is
1927 * expected that drivers will fix this mapping if they can before
1928 * calling netif_set_real_num_tx_queues.
1930 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1933 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1935 /* If TC0 is invalidated disable TC mapping */
1936 if (tc->offset + tc->count > txq) {
1937 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1942 /* Invalidated prio to tc mappings set to TC0 */
1943 for (i = 1; i < TC_BITMASK + 1; i++) {
1944 int q = netdev_get_prio_tc_map(dev, i);
1946 tc = &dev->tc_to_txq[q];
1947 if (tc->offset + tc->count > txq) {
1948 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1950 netdev_set_prio_tc_map(dev, i, 0);
1956 static DEFINE_MUTEX(xps_map_mutex);
1957 #define xmap_dereference(P) \
1958 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1960 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1963 struct xps_map *map = NULL;
1967 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1969 for (pos = 0; map && pos < map->len; pos++) {
1970 if (map->queues[pos] == index) {
1972 map->queues[pos] = map->queues[--map->len];
1974 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1975 kfree_rcu(map, rcu);
1985 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1987 struct xps_dev_maps *dev_maps;
1989 bool active = false;
1991 mutex_lock(&xps_map_mutex);
1992 dev_maps = xmap_dereference(dev->xps_maps);
1997 for_each_possible_cpu(cpu) {
1998 for (i = index; i < dev->num_tx_queues; i++) {
1999 if (!remove_xps_queue(dev_maps, cpu, i))
2002 if (i == dev->num_tx_queues)
2007 RCU_INIT_POINTER(dev->xps_maps, NULL);
2008 kfree_rcu(dev_maps, rcu);
2011 for (i = index; i < dev->num_tx_queues; i++)
2012 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2016 mutex_unlock(&xps_map_mutex);
2019 static struct xps_map *expand_xps_map(struct xps_map *map,
2022 struct xps_map *new_map;
2023 int alloc_len = XPS_MIN_MAP_ALLOC;
2026 for (pos = 0; map && pos < map->len; pos++) {
2027 if (map->queues[pos] != index)
2032 /* Need to add queue to this CPU's existing map */
2034 if (pos < map->alloc_len)
2037 alloc_len = map->alloc_len * 2;
2040 /* Need to allocate new map to store queue on this CPU's map */
2041 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2046 for (i = 0; i < pos; i++)
2047 new_map->queues[i] = map->queues[i];
2048 new_map->alloc_len = alloc_len;
2054 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2057 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2058 struct xps_map *map, *new_map;
2059 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2060 int cpu, numa_node_id = -2;
2061 bool active = false;
2063 mutex_lock(&xps_map_mutex);
2065 dev_maps = xmap_dereference(dev->xps_maps);
2067 /* allocate memory for queue storage */
2068 for_each_online_cpu(cpu) {
2069 if (!cpumask_test_cpu(cpu, mask))
2073 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2074 if (!new_dev_maps) {
2075 mutex_unlock(&xps_map_mutex);
2079 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2082 map = expand_xps_map(map, cpu, index);
2086 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2090 goto out_no_new_maps;
2092 for_each_possible_cpu(cpu) {
2093 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2094 /* add queue to CPU maps */
2097 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2098 while ((pos < map->len) && (map->queues[pos] != index))
2101 if (pos == map->len)
2102 map->queues[map->len++] = index;
2104 if (numa_node_id == -2)
2105 numa_node_id = cpu_to_node(cpu);
2106 else if (numa_node_id != cpu_to_node(cpu))
2109 } else if (dev_maps) {
2110 /* fill in the new device map from the old device map */
2111 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2112 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2117 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2119 /* Cleanup old maps */
2121 for_each_possible_cpu(cpu) {
2122 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2123 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2124 if (map && map != new_map)
2125 kfree_rcu(map, rcu);
2128 kfree_rcu(dev_maps, rcu);
2131 dev_maps = new_dev_maps;
2135 /* update Tx queue numa node */
2136 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2137 (numa_node_id >= 0) ? numa_node_id :
2143 /* removes queue from unused CPUs */
2144 for_each_possible_cpu(cpu) {
2145 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2148 if (remove_xps_queue(dev_maps, cpu, index))
2152 /* free map if not active */
2154 RCU_INIT_POINTER(dev->xps_maps, NULL);
2155 kfree_rcu(dev_maps, rcu);
2159 mutex_unlock(&xps_map_mutex);
2163 /* remove any maps that we added */
2164 for_each_possible_cpu(cpu) {
2165 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2166 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2168 if (new_map && new_map != map)
2172 mutex_unlock(&xps_map_mutex);
2174 kfree(new_dev_maps);
2177 EXPORT_SYMBOL(netif_set_xps_queue);
2181 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2182 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2184 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2188 if (txq < 1 || txq > dev->num_tx_queues)
2191 if (dev->reg_state == NETREG_REGISTERED ||
2192 dev->reg_state == NETREG_UNREGISTERING) {
2195 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2201 netif_setup_tc(dev, txq);
2203 if (txq < dev->real_num_tx_queues) {
2204 qdisc_reset_all_tx_gt(dev, txq);
2206 netif_reset_xps_queues_gt(dev, txq);
2211 dev->real_num_tx_queues = txq;
2214 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2218 * netif_set_real_num_rx_queues - set actual number of RX queues used
2219 * @dev: Network device
2220 * @rxq: Actual number of RX queues
2222 * This must be called either with the rtnl_lock held or before
2223 * registration of the net device. Returns 0 on success, or a
2224 * negative error code. If called before registration, it always
2227 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2231 if (rxq < 1 || rxq > dev->num_rx_queues)
2234 if (dev->reg_state == NETREG_REGISTERED) {
2237 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2243 dev->real_num_rx_queues = rxq;
2246 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2250 * netif_get_num_default_rss_queues - default number of RSS queues
2252 * This routine should set an upper limit on the number of RSS queues
2253 * used by default by multiqueue devices.
2255 int netif_get_num_default_rss_queues(void)
2257 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2259 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2261 static inline void __netif_reschedule(struct Qdisc *q)
2263 struct softnet_data *sd;
2264 unsigned long flags;
2266 local_irq_save(flags);
2267 sd = this_cpu_ptr(&softnet_data);
2268 q->next_sched = NULL;
2269 *sd->output_queue_tailp = q;
2270 sd->output_queue_tailp = &q->next_sched;
2271 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2272 local_irq_restore(flags);
2275 void __netif_schedule(struct Qdisc *q)
2277 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2278 __netif_reschedule(q);
2280 EXPORT_SYMBOL(__netif_schedule);
2282 struct dev_kfree_skb_cb {
2283 enum skb_free_reason reason;
2286 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2288 return (struct dev_kfree_skb_cb *)skb->cb;
2291 void netif_schedule_queue(struct netdev_queue *txq)
2294 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2295 struct Qdisc *q = rcu_dereference(txq->qdisc);
2297 __netif_schedule(q);
2301 EXPORT_SYMBOL(netif_schedule_queue);
2304 * netif_wake_subqueue - allow sending packets on subqueue
2305 * @dev: network device
2306 * @queue_index: sub queue index
2308 * Resume individual transmit queue of a device with multiple transmit queues.
2310 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2312 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2314 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2318 q = rcu_dereference(txq->qdisc);
2319 __netif_schedule(q);
2323 EXPORT_SYMBOL(netif_wake_subqueue);
2325 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2327 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2331 q = rcu_dereference(dev_queue->qdisc);
2332 __netif_schedule(q);
2336 EXPORT_SYMBOL(netif_tx_wake_queue);
2338 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2340 unsigned long flags;
2345 if (likely(atomic_read(&skb->users) == 1)) {
2347 atomic_set(&skb->users, 0);
2348 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2351 get_kfree_skb_cb(skb)->reason = reason;
2352 local_irq_save(flags);
2353 skb->next = __this_cpu_read(softnet_data.completion_queue);
2354 __this_cpu_write(softnet_data.completion_queue, skb);
2355 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2356 local_irq_restore(flags);
2358 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2360 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2362 if (in_irq() || irqs_disabled())
2363 __dev_kfree_skb_irq(skb, reason);
2367 EXPORT_SYMBOL(__dev_kfree_skb_any);
2371 * netif_device_detach - mark device as removed
2372 * @dev: network device
2374 * Mark device as removed from system and therefore no longer available.
2376 void netif_device_detach(struct net_device *dev)
2378 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2379 netif_running(dev)) {
2380 netif_tx_stop_all_queues(dev);
2383 EXPORT_SYMBOL(netif_device_detach);
2386 * netif_device_attach - mark device as attached
2387 * @dev: network device
2389 * Mark device as attached from system and restart if needed.
2391 void netif_device_attach(struct net_device *dev)
2393 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2394 netif_running(dev)) {
2395 netif_tx_wake_all_queues(dev);
2396 __netdev_watchdog_up(dev);
2399 EXPORT_SYMBOL(netif_device_attach);
2402 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2403 * to be used as a distribution range.
2405 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2406 unsigned int num_tx_queues)
2410 u16 qcount = num_tx_queues;
2412 if (skb_rx_queue_recorded(skb)) {
2413 hash = skb_get_rx_queue(skb);
2414 while (unlikely(hash >= num_tx_queues))
2415 hash -= num_tx_queues;
2420 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2421 qoffset = dev->tc_to_txq[tc].offset;
2422 qcount = dev->tc_to_txq[tc].count;
2425 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2427 EXPORT_SYMBOL(__skb_tx_hash);
2429 static void skb_warn_bad_offload(const struct sk_buff *skb)
2431 static const netdev_features_t null_features = 0;
2432 struct net_device *dev = skb->dev;
2433 const char *name = "";
2435 if (!net_ratelimit())
2439 if (dev->dev.parent)
2440 name = dev_driver_string(dev->dev.parent);
2442 name = netdev_name(dev);
2444 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2445 "gso_type=%d ip_summed=%d\n",
2446 name, dev ? &dev->features : &null_features,
2447 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2448 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2449 skb_shinfo(skb)->gso_type, skb->ip_summed);
2453 * Invalidate hardware checksum when packet is to be mangled, and
2454 * complete checksum manually on outgoing path.
2456 int skb_checksum_help(struct sk_buff *skb)
2459 int ret = 0, offset;
2461 if (skb->ip_summed == CHECKSUM_COMPLETE)
2462 goto out_set_summed;
2464 if (unlikely(skb_shinfo(skb)->gso_size)) {
2465 skb_warn_bad_offload(skb);
2469 /* Before computing a checksum, we should make sure no frag could
2470 * be modified by an external entity : checksum could be wrong.
2472 if (skb_has_shared_frag(skb)) {
2473 ret = __skb_linearize(skb);
2478 offset = skb_checksum_start_offset(skb);
2479 BUG_ON(offset >= skb_headlen(skb));
2480 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2482 offset += skb->csum_offset;
2483 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2485 if (skb_cloned(skb) &&
2486 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2487 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2492 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2494 skb->ip_summed = CHECKSUM_NONE;
2498 EXPORT_SYMBOL(skb_checksum_help);
2500 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2502 __be16 type = skb->protocol;
2504 /* Tunnel gso handlers can set protocol to ethernet. */
2505 if (type == htons(ETH_P_TEB)) {
2508 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2511 eth = (struct ethhdr *)skb_mac_header(skb);
2512 type = eth->h_proto;
2515 return __vlan_get_protocol(skb, type, depth);
2519 * skb_mac_gso_segment - mac layer segmentation handler.
2520 * @skb: buffer to segment
2521 * @features: features for the output path (see dev->features)
2523 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2524 netdev_features_t features)
2526 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2527 struct packet_offload *ptype;
2528 int vlan_depth = skb->mac_len;
2529 __be16 type = skb_network_protocol(skb, &vlan_depth);
2531 if (unlikely(!type))
2532 return ERR_PTR(-EINVAL);
2534 __skb_pull(skb, vlan_depth);
2537 list_for_each_entry_rcu(ptype, &offload_base, list) {
2538 if (ptype->type == type && ptype->callbacks.gso_segment) {
2539 segs = ptype->callbacks.gso_segment(skb, features);
2545 __skb_push(skb, skb->data - skb_mac_header(skb));
2549 EXPORT_SYMBOL(skb_mac_gso_segment);
2552 /* openvswitch calls this on rx path, so we need a different check.
2554 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2557 return skb->ip_summed != CHECKSUM_PARTIAL &&
2558 skb->ip_summed != CHECKSUM_UNNECESSARY;
2560 return skb->ip_summed == CHECKSUM_NONE;
2564 * __skb_gso_segment - Perform segmentation on skb.
2565 * @skb: buffer to segment
2566 * @features: features for the output path (see dev->features)
2567 * @tx_path: whether it is called in TX path
2569 * This function segments the given skb and returns a list of segments.
2571 * It may return NULL if the skb requires no segmentation. This is
2572 * only possible when GSO is used for verifying header integrity.
2574 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2576 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2577 netdev_features_t features, bool tx_path)
2579 struct sk_buff *segs;
2581 if (unlikely(skb_needs_check(skb, tx_path))) {
2584 /* We're going to init ->check field in TCP or UDP header */
2585 err = skb_cow_head(skb, 0);
2587 return ERR_PTR(err);
2590 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2591 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2593 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2594 SKB_GSO_CB(skb)->encap_level = 0;
2596 skb_reset_mac_header(skb);
2597 skb_reset_mac_len(skb);
2599 segs = skb_mac_gso_segment(skb, features);
2601 if (unlikely(skb_needs_check(skb, tx_path)))
2602 skb_warn_bad_offload(skb);
2606 EXPORT_SYMBOL(__skb_gso_segment);
2608 /* Take action when hardware reception checksum errors are detected. */
2610 void netdev_rx_csum_fault(struct net_device *dev)
2612 if (net_ratelimit()) {
2613 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2617 EXPORT_SYMBOL(netdev_rx_csum_fault);
2620 /* Actually, we should eliminate this check as soon as we know, that:
2621 * 1. IOMMU is present and allows to map all the memory.
2622 * 2. No high memory really exists on this machine.
2625 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2627 #ifdef CONFIG_HIGHMEM
2629 if (!(dev->features & NETIF_F_HIGHDMA)) {
2630 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2631 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2632 if (PageHighMem(skb_frag_page(frag)))
2637 if (PCI_DMA_BUS_IS_PHYS) {
2638 struct device *pdev = dev->dev.parent;
2642 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2643 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2644 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2645 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2653 /* If MPLS offload request, verify we are testing hardware MPLS features
2654 * instead of standard features for the netdev.
2656 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2657 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2658 netdev_features_t features,
2661 if (eth_p_mpls(type))
2662 features &= skb->dev->mpls_features;
2667 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2668 netdev_features_t features,
2675 static netdev_features_t harmonize_features(struct sk_buff *skb,
2676 netdev_features_t features)
2681 type = skb_network_protocol(skb, &tmp);
2682 features = net_mpls_features(skb, features, type);
2684 if (skb->ip_summed != CHECKSUM_NONE &&
2685 !can_checksum_protocol(features, type)) {
2686 features &= ~NETIF_F_ALL_CSUM;
2688 if (illegal_highdma(skb->dev, skb))
2689 features &= ~NETIF_F_SG;
2694 netdev_features_t passthru_features_check(struct sk_buff *skb,
2695 struct net_device *dev,
2696 netdev_features_t features)
2700 EXPORT_SYMBOL(passthru_features_check);
2702 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2703 struct net_device *dev,
2704 netdev_features_t features)
2706 return vlan_features_check(skb, features);
2709 netdev_features_t netif_skb_features(struct sk_buff *skb)
2711 struct net_device *dev = skb->dev;
2712 netdev_features_t features = dev->features;
2713 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2715 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2716 features &= ~NETIF_F_GSO_MASK;
2718 /* If encapsulation offload request, verify we are testing
2719 * hardware encapsulation features instead of standard
2720 * features for the netdev
2722 if (skb->encapsulation)
2723 features &= dev->hw_enc_features;
2725 if (skb_vlan_tagged(skb))
2726 features = netdev_intersect_features(features,
2727 dev->vlan_features |
2728 NETIF_F_HW_VLAN_CTAG_TX |
2729 NETIF_F_HW_VLAN_STAG_TX);
2731 if (dev->netdev_ops->ndo_features_check)
2732 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2735 features &= dflt_features_check(skb, dev, features);
2737 return harmonize_features(skb, features);
2739 EXPORT_SYMBOL(netif_skb_features);
2741 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2742 struct netdev_queue *txq, bool more)
2747 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2748 dev_queue_xmit_nit(skb, dev);
2751 trace_net_dev_start_xmit(skb, dev);
2752 rc = netdev_start_xmit(skb, dev, txq, more);
2753 trace_net_dev_xmit(skb, rc, dev, len);
2758 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2759 struct netdev_queue *txq, int *ret)
2761 struct sk_buff *skb = first;
2762 int rc = NETDEV_TX_OK;
2765 struct sk_buff *next = skb->next;
2768 rc = xmit_one(skb, dev, txq, next != NULL);
2769 if (unlikely(!dev_xmit_complete(rc))) {
2775 if (netif_xmit_stopped(txq) && skb) {
2776 rc = NETDEV_TX_BUSY;
2786 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2787 netdev_features_t features)
2789 if (skb_vlan_tag_present(skb) &&
2790 !vlan_hw_offload_capable(features, skb->vlan_proto))
2791 skb = __vlan_hwaccel_push_inside(skb);
2795 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2797 netdev_features_t features;
2802 features = netif_skb_features(skb);
2803 skb = validate_xmit_vlan(skb, features);
2807 if (netif_needs_gso(skb, features)) {
2808 struct sk_buff *segs;
2810 segs = skb_gso_segment(skb, features);
2818 if (skb_needs_linearize(skb, features) &&
2819 __skb_linearize(skb))
2822 /* If packet is not checksummed and device does not
2823 * support checksumming for this protocol, complete
2824 * checksumming here.
2826 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2827 if (skb->encapsulation)
2828 skb_set_inner_transport_header(skb,
2829 skb_checksum_start_offset(skb));
2831 skb_set_transport_header(skb,
2832 skb_checksum_start_offset(skb));
2833 if (!(features & NETIF_F_ALL_CSUM) &&
2834 skb_checksum_help(skb))
2847 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2849 struct sk_buff *next, *head = NULL, *tail;
2851 for (; skb != NULL; skb = next) {
2855 /* in case skb wont be segmented, point to itself */
2858 skb = validate_xmit_skb(skb, dev);
2866 /* If skb was segmented, skb->prev points to
2867 * the last segment. If not, it still contains skb.
2873 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
2875 static void qdisc_pkt_len_init(struct sk_buff *skb)
2877 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2879 qdisc_skb_cb(skb)->pkt_len = skb->len;
2881 /* To get more precise estimation of bytes sent on wire,
2882 * we add to pkt_len the headers size of all segments
2884 if (shinfo->gso_size) {
2885 unsigned int hdr_len;
2886 u16 gso_segs = shinfo->gso_segs;
2888 /* mac layer + network layer */
2889 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2891 /* + transport layer */
2892 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2893 hdr_len += tcp_hdrlen(skb);
2895 hdr_len += sizeof(struct udphdr);
2897 if (shinfo->gso_type & SKB_GSO_DODGY)
2898 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2901 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2905 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2906 struct net_device *dev,
2907 struct netdev_queue *txq)
2909 spinlock_t *root_lock = qdisc_lock(q);
2913 qdisc_pkt_len_init(skb);
2914 qdisc_calculate_pkt_len(skb, q);
2916 * Heuristic to force contended enqueues to serialize on a
2917 * separate lock before trying to get qdisc main lock.
2918 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2919 * often and dequeue packets faster.
2921 contended = qdisc_is_running(q);
2922 if (unlikely(contended))
2923 spin_lock(&q->busylock);
2925 spin_lock(root_lock);
2926 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2929 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2930 qdisc_run_begin(q)) {
2932 * This is a work-conserving queue; there are no old skbs
2933 * waiting to be sent out; and the qdisc is not running -
2934 * xmit the skb directly.
2937 qdisc_bstats_update(q, skb);
2939 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2940 if (unlikely(contended)) {
2941 spin_unlock(&q->busylock);
2948 rc = NET_XMIT_SUCCESS;
2950 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2951 if (qdisc_run_begin(q)) {
2952 if (unlikely(contended)) {
2953 spin_unlock(&q->busylock);
2959 spin_unlock(root_lock);
2960 if (unlikely(contended))
2961 spin_unlock(&q->busylock);
2965 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2966 static void skb_update_prio(struct sk_buff *skb)
2968 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2970 if (!skb->priority && skb->sk && map) {
2971 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2973 if (prioidx < map->priomap_len)
2974 skb->priority = map->priomap[prioidx];
2978 #define skb_update_prio(skb)
2981 DEFINE_PER_CPU(int, xmit_recursion);
2982 EXPORT_SYMBOL(xmit_recursion);
2984 #define RECURSION_LIMIT 10
2987 * dev_loopback_xmit - loop back @skb
2988 * @net: network namespace this loopback is happening in
2989 * @sk: sk needed to be a netfilter okfn
2990 * @skb: buffer to transmit
2992 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
2994 skb_reset_mac_header(skb);
2995 __skb_pull(skb, skb_network_offset(skb));
2996 skb->pkt_type = PACKET_LOOPBACK;
2997 skb->ip_summed = CHECKSUM_UNNECESSARY;
2998 WARN_ON(!skb_dst(skb));
3003 EXPORT_SYMBOL(dev_loopback_xmit);
3005 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3008 struct xps_dev_maps *dev_maps;
3009 struct xps_map *map;
3010 int queue_index = -1;
3013 dev_maps = rcu_dereference(dev->xps_maps);
3015 map = rcu_dereference(
3016 dev_maps->cpu_map[skb->sender_cpu - 1]);
3019 queue_index = map->queues[0];
3021 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3023 if (unlikely(queue_index >= dev->real_num_tx_queues))
3035 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3037 struct sock *sk = skb->sk;
3038 int queue_index = sk_tx_queue_get(sk);
3040 if (queue_index < 0 || skb->ooo_okay ||
3041 queue_index >= dev->real_num_tx_queues) {
3042 int new_index = get_xps_queue(dev, skb);
3044 new_index = skb_tx_hash(dev, skb);
3046 if (queue_index != new_index && sk &&
3048 rcu_access_pointer(sk->sk_dst_cache))
3049 sk_tx_queue_set(sk, new_index);
3051 queue_index = new_index;
3057 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3058 struct sk_buff *skb,
3061 int queue_index = 0;
3064 u32 sender_cpu = skb->sender_cpu - 1;
3066 if (sender_cpu >= (u32)NR_CPUS)
3067 skb->sender_cpu = raw_smp_processor_id() + 1;
3070 if (dev->real_num_tx_queues != 1) {
3071 const struct net_device_ops *ops = dev->netdev_ops;
3072 if (ops->ndo_select_queue)
3073 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3076 queue_index = __netdev_pick_tx(dev, skb);
3079 queue_index = netdev_cap_txqueue(dev, queue_index);
3082 skb_set_queue_mapping(skb, queue_index);
3083 return netdev_get_tx_queue(dev, queue_index);
3087 * __dev_queue_xmit - transmit a buffer
3088 * @skb: buffer to transmit
3089 * @accel_priv: private data used for L2 forwarding offload
3091 * Queue a buffer for transmission to a network device. The caller must
3092 * have set the device and priority and built the buffer before calling
3093 * this function. The function can be called from an interrupt.
3095 * A negative errno code is returned on a failure. A success does not
3096 * guarantee the frame will be transmitted as it may be dropped due
3097 * to congestion or traffic shaping.
3099 * -----------------------------------------------------------------------------------
3100 * I notice this method can also return errors from the queue disciplines,
3101 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3104 * Regardless of the return value, the skb is consumed, so it is currently
3105 * difficult to retry a send to this method. (You can bump the ref count
3106 * before sending to hold a reference for retry if you are careful.)
3108 * When calling this method, interrupts MUST be enabled. This is because
3109 * the BH enable code must have IRQs enabled so that it will not deadlock.
3112 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3114 struct net_device *dev = skb->dev;
3115 struct netdev_queue *txq;
3119 skb_reset_mac_header(skb);
3121 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3122 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3124 /* Disable soft irqs for various locks below. Also
3125 * stops preemption for RCU.
3129 skb_update_prio(skb);
3131 /* If device/qdisc don't need skb->dst, release it right now while
3132 * its hot in this cpu cache.
3134 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3139 #ifdef CONFIG_NET_SWITCHDEV
3140 /* Don't forward if offload device already forwarded */
3141 if (skb->offload_fwd_mark &&
3142 skb->offload_fwd_mark == dev->offload_fwd_mark) {
3144 rc = NET_XMIT_SUCCESS;
3149 txq = netdev_pick_tx(dev, skb, accel_priv);
3150 q = rcu_dereference_bh(txq->qdisc);
3152 #ifdef CONFIG_NET_CLS_ACT
3153 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3155 trace_net_dev_queue(skb);
3157 rc = __dev_xmit_skb(skb, q, dev, txq);
3161 /* The device has no queue. Common case for software devices:
3162 loopback, all the sorts of tunnels...
3164 Really, it is unlikely that netif_tx_lock protection is necessary
3165 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3167 However, it is possible, that they rely on protection
3170 Check this and shot the lock. It is not prone from deadlocks.
3171 Either shot noqueue qdisc, it is even simpler 8)
3173 if (dev->flags & IFF_UP) {
3174 int cpu = smp_processor_id(); /* ok because BHs are off */
3176 if (txq->xmit_lock_owner != cpu) {
3178 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
3179 goto recursion_alert;
3181 skb = validate_xmit_skb(skb, dev);
3185 HARD_TX_LOCK(dev, txq, cpu);
3187 if (!netif_xmit_stopped(txq)) {
3188 __this_cpu_inc(xmit_recursion);
3189 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3190 __this_cpu_dec(xmit_recursion);
3191 if (dev_xmit_complete(rc)) {
3192 HARD_TX_UNLOCK(dev, txq);
3196 HARD_TX_UNLOCK(dev, txq);
3197 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3200 /* Recursion is detected! It is possible,
3204 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3211 rcu_read_unlock_bh();
3213 atomic_long_inc(&dev->tx_dropped);
3214 kfree_skb_list(skb);
3217 rcu_read_unlock_bh();
3221 int dev_queue_xmit(struct sk_buff *skb)
3223 return __dev_queue_xmit(skb, NULL);
3225 EXPORT_SYMBOL(dev_queue_xmit);
3227 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3229 return __dev_queue_xmit(skb, accel_priv);
3231 EXPORT_SYMBOL(dev_queue_xmit_accel);
3234 /*=======================================================================
3236 =======================================================================*/
3238 int netdev_max_backlog __read_mostly = 1000;
3239 EXPORT_SYMBOL(netdev_max_backlog);
3241 int netdev_tstamp_prequeue __read_mostly = 1;
3242 int netdev_budget __read_mostly = 300;
3243 int weight_p __read_mostly = 64; /* old backlog weight */
3245 /* Called with irq disabled */
3246 static inline void ____napi_schedule(struct softnet_data *sd,
3247 struct napi_struct *napi)
3249 list_add_tail(&napi->poll_list, &sd->poll_list);
3250 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3255 /* One global table that all flow-based protocols share. */
3256 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3257 EXPORT_SYMBOL(rps_sock_flow_table);
3258 u32 rps_cpu_mask __read_mostly;
3259 EXPORT_SYMBOL(rps_cpu_mask);
3261 struct static_key rps_needed __read_mostly;
3263 static struct rps_dev_flow *
3264 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3265 struct rps_dev_flow *rflow, u16 next_cpu)
3267 if (next_cpu < nr_cpu_ids) {
3268 #ifdef CONFIG_RFS_ACCEL
3269 struct netdev_rx_queue *rxqueue;
3270 struct rps_dev_flow_table *flow_table;
3271 struct rps_dev_flow *old_rflow;
3276 /* Should we steer this flow to a different hardware queue? */
3277 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3278 !(dev->features & NETIF_F_NTUPLE))
3280 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3281 if (rxq_index == skb_get_rx_queue(skb))
3284 rxqueue = dev->_rx + rxq_index;
3285 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3288 flow_id = skb_get_hash(skb) & flow_table->mask;
3289 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3290 rxq_index, flow_id);
3294 rflow = &flow_table->flows[flow_id];
3296 if (old_rflow->filter == rflow->filter)
3297 old_rflow->filter = RPS_NO_FILTER;
3301 per_cpu(softnet_data, next_cpu).input_queue_head;
3304 rflow->cpu = next_cpu;
3309 * get_rps_cpu is called from netif_receive_skb and returns the target
3310 * CPU from the RPS map of the receiving queue for a given skb.
3311 * rcu_read_lock must be held on entry.
3313 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3314 struct rps_dev_flow **rflowp)
3316 const struct rps_sock_flow_table *sock_flow_table;
3317 struct netdev_rx_queue *rxqueue = dev->_rx;
3318 struct rps_dev_flow_table *flow_table;
3319 struct rps_map *map;
3324 if (skb_rx_queue_recorded(skb)) {
3325 u16 index = skb_get_rx_queue(skb);
3327 if (unlikely(index >= dev->real_num_rx_queues)) {
3328 WARN_ONCE(dev->real_num_rx_queues > 1,
3329 "%s received packet on queue %u, but number "
3330 "of RX queues is %u\n",
3331 dev->name, index, dev->real_num_rx_queues);
3337 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3339 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3340 map = rcu_dereference(rxqueue->rps_map);
3341 if (!flow_table && !map)
3344 skb_reset_network_header(skb);
3345 hash = skb_get_hash(skb);
3349 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3350 if (flow_table && sock_flow_table) {
3351 struct rps_dev_flow *rflow;
3355 /* First check into global flow table if there is a match */
3356 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3357 if ((ident ^ hash) & ~rps_cpu_mask)
3360 next_cpu = ident & rps_cpu_mask;
3362 /* OK, now we know there is a match,
3363 * we can look at the local (per receive queue) flow table
3365 rflow = &flow_table->flows[hash & flow_table->mask];
3369 * If the desired CPU (where last recvmsg was done) is
3370 * different from current CPU (one in the rx-queue flow
3371 * table entry), switch if one of the following holds:
3372 * - Current CPU is unset (>= nr_cpu_ids).
3373 * - Current CPU is offline.
3374 * - The current CPU's queue tail has advanced beyond the
3375 * last packet that was enqueued using this table entry.
3376 * This guarantees that all previous packets for the flow
3377 * have been dequeued, thus preserving in order delivery.
3379 if (unlikely(tcpu != next_cpu) &&
3380 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3381 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3382 rflow->last_qtail)) >= 0)) {
3384 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3387 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3397 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3398 if (cpu_online(tcpu)) {
3408 #ifdef CONFIG_RFS_ACCEL
3411 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3412 * @dev: Device on which the filter was set
3413 * @rxq_index: RX queue index
3414 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3415 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3417 * Drivers that implement ndo_rx_flow_steer() should periodically call
3418 * this function for each installed filter and remove the filters for
3419 * which it returns %true.
3421 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3422 u32 flow_id, u16 filter_id)
3424 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3425 struct rps_dev_flow_table *flow_table;
3426 struct rps_dev_flow *rflow;
3431 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3432 if (flow_table && flow_id <= flow_table->mask) {
3433 rflow = &flow_table->flows[flow_id];
3434 cpu = ACCESS_ONCE(rflow->cpu);
3435 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3436 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3437 rflow->last_qtail) <
3438 (int)(10 * flow_table->mask)))
3444 EXPORT_SYMBOL(rps_may_expire_flow);
3446 #endif /* CONFIG_RFS_ACCEL */
3448 /* Called from hardirq (IPI) context */
3449 static void rps_trigger_softirq(void *data)
3451 struct softnet_data *sd = data;
3453 ____napi_schedule(sd, &sd->backlog);
3457 #endif /* CONFIG_RPS */
3460 * Check if this softnet_data structure is another cpu one
3461 * If yes, queue it to our IPI list and return 1
3464 static int rps_ipi_queued(struct softnet_data *sd)
3467 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3470 sd->rps_ipi_next = mysd->rps_ipi_list;
3471 mysd->rps_ipi_list = sd;
3473 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3476 #endif /* CONFIG_RPS */
3480 #ifdef CONFIG_NET_FLOW_LIMIT
3481 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3484 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3486 #ifdef CONFIG_NET_FLOW_LIMIT
3487 struct sd_flow_limit *fl;
3488 struct softnet_data *sd;
3489 unsigned int old_flow, new_flow;
3491 if (qlen < (netdev_max_backlog >> 1))
3494 sd = this_cpu_ptr(&softnet_data);
3497 fl = rcu_dereference(sd->flow_limit);
3499 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3500 old_flow = fl->history[fl->history_head];
3501 fl->history[fl->history_head] = new_flow;
3504 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3506 if (likely(fl->buckets[old_flow]))
3507 fl->buckets[old_flow]--;
3509 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3521 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3522 * queue (may be a remote CPU queue).
3524 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3525 unsigned int *qtail)
3527 struct softnet_data *sd;
3528 unsigned long flags;
3531 sd = &per_cpu(softnet_data, cpu);
3533 local_irq_save(flags);
3536 if (!netif_running(skb->dev))
3538 qlen = skb_queue_len(&sd->input_pkt_queue);
3539 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3542 __skb_queue_tail(&sd->input_pkt_queue, skb);
3543 input_queue_tail_incr_save(sd, qtail);
3545 local_irq_restore(flags);
3546 return NET_RX_SUCCESS;
3549 /* Schedule NAPI for backlog device
3550 * We can use non atomic operation since we own the queue lock
3552 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3553 if (!rps_ipi_queued(sd))
3554 ____napi_schedule(sd, &sd->backlog);
3563 local_irq_restore(flags);
3565 atomic_long_inc(&skb->dev->rx_dropped);
3570 static int netif_rx_internal(struct sk_buff *skb)
3574 net_timestamp_check(netdev_tstamp_prequeue, skb);
3576 trace_netif_rx(skb);
3578 if (static_key_false(&rps_needed)) {
3579 struct rps_dev_flow voidflow, *rflow = &voidflow;
3585 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3587 cpu = smp_processor_id();
3589 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3597 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3604 * netif_rx - post buffer to the network code
3605 * @skb: buffer to post
3607 * This function receives a packet from a device driver and queues it for
3608 * the upper (protocol) levels to process. It always succeeds. The buffer
3609 * may be dropped during processing for congestion control or by the
3613 * NET_RX_SUCCESS (no congestion)
3614 * NET_RX_DROP (packet was dropped)
3618 int netif_rx(struct sk_buff *skb)
3620 trace_netif_rx_entry(skb);
3622 return netif_rx_internal(skb);
3624 EXPORT_SYMBOL(netif_rx);
3626 int netif_rx_ni(struct sk_buff *skb)
3630 trace_netif_rx_ni_entry(skb);
3633 err = netif_rx_internal(skb);
3634 if (local_softirq_pending())
3640 EXPORT_SYMBOL(netif_rx_ni);
3642 static void net_tx_action(struct softirq_action *h)
3644 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3646 if (sd->completion_queue) {
3647 struct sk_buff *clist;
3649 local_irq_disable();
3650 clist = sd->completion_queue;
3651 sd->completion_queue = NULL;
3655 struct sk_buff *skb = clist;
3656 clist = clist->next;
3658 WARN_ON(atomic_read(&skb->users));
3659 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3660 trace_consume_skb(skb);
3662 trace_kfree_skb(skb, net_tx_action);
3667 if (sd->output_queue) {
3670 local_irq_disable();
3671 head = sd->output_queue;
3672 sd->output_queue = NULL;
3673 sd->output_queue_tailp = &sd->output_queue;
3677 struct Qdisc *q = head;
3678 spinlock_t *root_lock;
3680 head = head->next_sched;
3682 root_lock = qdisc_lock(q);
3683 if (spin_trylock(root_lock)) {
3684 smp_mb__before_atomic();
3685 clear_bit(__QDISC_STATE_SCHED,
3688 spin_unlock(root_lock);
3690 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3692 __netif_reschedule(q);
3694 smp_mb__before_atomic();
3695 clear_bit(__QDISC_STATE_SCHED,
3703 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3704 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3705 /* This hook is defined here for ATM LANE */
3706 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3707 unsigned char *addr) __read_mostly;
3708 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3711 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3712 struct packet_type **pt_prev,
3713 int *ret, struct net_device *orig_dev)
3715 #ifdef CONFIG_NET_CLS_ACT
3716 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3717 struct tcf_result cl_res;
3719 /* If there's at least one ingress present somewhere (so
3720 * we get here via enabled static key), remaining devices
3721 * that are not configured with an ingress qdisc will bail
3727 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3731 qdisc_skb_cb(skb)->pkt_len = skb->len;
3732 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3733 qdisc_bstats_cpu_update(cl->q, skb);
3735 switch (tc_classify(skb, cl, &cl_res, false)) {
3737 case TC_ACT_RECLASSIFY:
3738 skb->tc_index = TC_H_MIN(cl_res.classid);
3741 qdisc_qstats_cpu_drop(cl->q);
3746 case TC_ACT_REDIRECT:
3747 /* skb_mac_header check was done by cls/act_bpf, so
3748 * we can safely push the L2 header back before
3749 * redirecting to another netdev
3751 __skb_push(skb, skb->mac_len);
3752 skb_do_redirect(skb);
3757 #endif /* CONFIG_NET_CLS_ACT */
3762 * netdev_is_rx_handler_busy - check if receive handler is registered
3763 * @dev: device to check
3765 * Check if a receive handler is already registered for a given device.
3766 * Return true if there one.
3768 * The caller must hold the rtnl_mutex.
3770 bool netdev_is_rx_handler_busy(struct net_device *dev)
3773 return dev && rtnl_dereference(dev->rx_handler);
3775 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
3778 * netdev_rx_handler_register - register receive handler
3779 * @dev: device to register a handler for
3780 * @rx_handler: receive handler to register
3781 * @rx_handler_data: data pointer that is used by rx handler
3783 * Register a receive handler for a device. This handler will then be
3784 * called from __netif_receive_skb. A negative errno code is returned
3787 * The caller must hold the rtnl_mutex.
3789 * For a general description of rx_handler, see enum rx_handler_result.
3791 int netdev_rx_handler_register(struct net_device *dev,
3792 rx_handler_func_t *rx_handler,
3793 void *rx_handler_data)
3797 if (dev->rx_handler)
3800 /* Note: rx_handler_data must be set before rx_handler */
3801 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3802 rcu_assign_pointer(dev->rx_handler, rx_handler);
3806 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3809 * netdev_rx_handler_unregister - unregister receive handler
3810 * @dev: device to unregister a handler from
3812 * Unregister a receive handler from a device.
3814 * The caller must hold the rtnl_mutex.
3816 void netdev_rx_handler_unregister(struct net_device *dev)
3820 RCU_INIT_POINTER(dev->rx_handler, NULL);
3821 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3822 * section has a guarantee to see a non NULL rx_handler_data
3826 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3828 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3831 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3832 * the special handling of PFMEMALLOC skbs.
3834 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3836 switch (skb->protocol) {
3837 case htons(ETH_P_ARP):
3838 case htons(ETH_P_IP):
3839 case htons(ETH_P_IPV6):
3840 case htons(ETH_P_8021Q):
3841 case htons(ETH_P_8021AD):
3848 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
3849 int *ret, struct net_device *orig_dev)
3851 #ifdef CONFIG_NETFILTER_INGRESS
3852 if (nf_hook_ingress_active(skb)) {
3854 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3858 return nf_hook_ingress(skb);
3860 #endif /* CONFIG_NETFILTER_INGRESS */
3864 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3866 struct packet_type *ptype, *pt_prev;
3867 rx_handler_func_t *rx_handler;
3868 struct net_device *orig_dev;
3869 bool deliver_exact = false;
3870 int ret = NET_RX_DROP;
3873 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3875 trace_netif_receive_skb(skb);
3877 orig_dev = skb->dev;
3879 skb_reset_network_header(skb);
3880 if (!skb_transport_header_was_set(skb))
3881 skb_reset_transport_header(skb);
3882 skb_reset_mac_len(skb);
3887 skb->skb_iif = skb->dev->ifindex;
3889 __this_cpu_inc(softnet_data.processed);
3891 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3892 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3893 skb = skb_vlan_untag(skb);
3898 #ifdef CONFIG_NET_CLS_ACT
3899 if (skb->tc_verd & TC_NCLS) {
3900 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3908 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3910 ret = deliver_skb(skb, pt_prev, orig_dev);
3914 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3916 ret = deliver_skb(skb, pt_prev, orig_dev);
3921 #ifdef CONFIG_NET_INGRESS
3922 if (static_key_false(&ingress_needed)) {
3923 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3927 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
3931 #ifdef CONFIG_NET_CLS_ACT
3935 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3938 if (skb_vlan_tag_present(skb)) {
3940 ret = deliver_skb(skb, pt_prev, orig_dev);
3943 if (vlan_do_receive(&skb))
3945 else if (unlikely(!skb))
3949 rx_handler = rcu_dereference(skb->dev->rx_handler);
3952 ret = deliver_skb(skb, pt_prev, orig_dev);
3955 switch (rx_handler(&skb)) {
3956 case RX_HANDLER_CONSUMED:
3957 ret = NET_RX_SUCCESS;
3959 case RX_HANDLER_ANOTHER:
3961 case RX_HANDLER_EXACT:
3962 deliver_exact = true;
3963 case RX_HANDLER_PASS:
3970 if (unlikely(skb_vlan_tag_present(skb))) {
3971 if (skb_vlan_tag_get_id(skb))
3972 skb->pkt_type = PACKET_OTHERHOST;
3973 /* Note: we might in the future use prio bits
3974 * and set skb->priority like in vlan_do_receive()
3975 * For the time being, just ignore Priority Code Point
3980 type = skb->protocol;
3982 /* deliver only exact match when indicated */
3983 if (likely(!deliver_exact)) {
3984 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3985 &ptype_base[ntohs(type) &
3989 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3990 &orig_dev->ptype_specific);
3992 if (unlikely(skb->dev != orig_dev)) {
3993 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3994 &skb->dev->ptype_specific);
3998 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
4001 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4004 atomic_long_inc(&skb->dev->rx_dropped);
4006 /* Jamal, now you will not able to escape explaining
4007 * me how you were going to use this. :-)
4016 static int __netif_receive_skb(struct sk_buff *skb)
4020 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4021 unsigned long pflags = current->flags;
4024 * PFMEMALLOC skbs are special, they should
4025 * - be delivered to SOCK_MEMALLOC sockets only
4026 * - stay away from userspace
4027 * - have bounded memory usage
4029 * Use PF_MEMALLOC as this saves us from propagating the allocation
4030 * context down to all allocation sites.
4032 current->flags |= PF_MEMALLOC;
4033 ret = __netif_receive_skb_core(skb, true);
4034 tsk_restore_flags(current, pflags, PF_MEMALLOC);
4036 ret = __netif_receive_skb_core(skb, false);
4041 static int netif_receive_skb_internal(struct sk_buff *skb)
4045 net_timestamp_check(netdev_tstamp_prequeue, skb);
4047 if (skb_defer_rx_timestamp(skb))
4048 return NET_RX_SUCCESS;
4053 if (static_key_false(&rps_needed)) {
4054 struct rps_dev_flow voidflow, *rflow = &voidflow;
4055 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4058 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4064 ret = __netif_receive_skb(skb);
4070 * netif_receive_skb - process receive buffer from network
4071 * @skb: buffer to process
4073 * netif_receive_skb() is the main receive data processing function.
4074 * It always succeeds. The buffer may be dropped during processing
4075 * for congestion control or by the protocol layers.
4077 * This function may only be called from softirq context and interrupts
4078 * should be enabled.
4080 * Return values (usually ignored):
4081 * NET_RX_SUCCESS: no congestion
4082 * NET_RX_DROP: packet was dropped
4084 int netif_receive_skb(struct sk_buff *skb)
4086 trace_netif_receive_skb_entry(skb);
4088 return netif_receive_skb_internal(skb);
4090 EXPORT_SYMBOL(netif_receive_skb);
4092 /* Network device is going away, flush any packets still pending
4093 * Called with irqs disabled.
4095 static void flush_backlog(void *arg)
4097 struct net_device *dev = arg;
4098 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4099 struct sk_buff *skb, *tmp;
4102 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4103 if (skb->dev == dev) {
4104 __skb_unlink(skb, &sd->input_pkt_queue);
4106 input_queue_head_incr(sd);
4111 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4112 if (skb->dev == dev) {
4113 __skb_unlink(skb, &sd->process_queue);
4115 input_queue_head_incr(sd);
4120 static int napi_gro_complete(struct sk_buff *skb)
4122 struct packet_offload *ptype;
4123 __be16 type = skb->protocol;
4124 struct list_head *head = &offload_base;
4127 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4129 if (NAPI_GRO_CB(skb)->count == 1) {
4130 skb_shinfo(skb)->gso_size = 0;
4135 list_for_each_entry_rcu(ptype, head, list) {
4136 if (ptype->type != type || !ptype->callbacks.gro_complete)
4139 err = ptype->callbacks.gro_complete(skb, 0);
4145 WARN_ON(&ptype->list == head);
4147 return NET_RX_SUCCESS;
4151 return netif_receive_skb_internal(skb);
4154 /* napi->gro_list contains packets ordered by age.
4155 * youngest packets at the head of it.
4156 * Complete skbs in reverse order to reduce latencies.
4158 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4160 struct sk_buff *skb, *prev = NULL;
4162 /* scan list and build reverse chain */
4163 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4168 for (skb = prev; skb; skb = prev) {
4171 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4175 napi_gro_complete(skb);
4179 napi->gro_list = NULL;
4181 EXPORT_SYMBOL(napi_gro_flush);
4183 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4186 unsigned int maclen = skb->dev->hard_header_len;
4187 u32 hash = skb_get_hash_raw(skb);
4189 for (p = napi->gro_list; p; p = p->next) {
4190 unsigned long diffs;
4192 NAPI_GRO_CB(p)->flush = 0;
4194 if (hash != skb_get_hash_raw(p)) {
4195 NAPI_GRO_CB(p)->same_flow = 0;
4199 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4200 diffs |= p->vlan_tci ^ skb->vlan_tci;
4201 diffs |= skb_metadata_dst_cmp(p, skb);
4202 if (maclen == ETH_HLEN)
4203 diffs |= compare_ether_header(skb_mac_header(p),
4204 skb_mac_header(skb));
4206 diffs = memcmp(skb_mac_header(p),
4207 skb_mac_header(skb),
4209 NAPI_GRO_CB(p)->same_flow = !diffs;
4213 static void skb_gro_reset_offset(struct sk_buff *skb)
4215 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4216 const skb_frag_t *frag0 = &pinfo->frags[0];
4218 NAPI_GRO_CB(skb)->data_offset = 0;
4219 NAPI_GRO_CB(skb)->frag0 = NULL;
4220 NAPI_GRO_CB(skb)->frag0_len = 0;
4222 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4224 !PageHighMem(skb_frag_page(frag0))) {
4225 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4226 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
4227 skb_frag_size(frag0),
4228 skb->end - skb->tail);
4232 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4234 struct skb_shared_info *pinfo = skb_shinfo(skb);
4236 BUG_ON(skb->end - skb->tail < grow);
4238 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4240 skb->data_len -= grow;
4243 pinfo->frags[0].page_offset += grow;
4244 skb_frag_size_sub(&pinfo->frags[0], grow);
4246 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4247 skb_frag_unref(skb, 0);
4248 memmove(pinfo->frags, pinfo->frags + 1,
4249 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4253 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4255 struct sk_buff **pp = NULL;
4256 struct packet_offload *ptype;
4257 __be16 type = skb->protocol;
4258 struct list_head *head = &offload_base;
4260 enum gro_result ret;
4263 if (!(skb->dev->features & NETIF_F_GRO))
4266 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4269 gro_list_prepare(napi, skb);
4272 list_for_each_entry_rcu(ptype, head, list) {
4273 if (ptype->type != type || !ptype->callbacks.gro_receive)
4276 skb_set_network_header(skb, skb_gro_offset(skb));
4277 skb_reset_mac_len(skb);
4278 NAPI_GRO_CB(skb)->same_flow = 0;
4279 NAPI_GRO_CB(skb)->flush = 0;
4280 NAPI_GRO_CB(skb)->free = 0;
4281 NAPI_GRO_CB(skb)->encap_mark = 0;
4282 NAPI_GRO_CB(skb)->recursion_counter = 0;
4283 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4285 /* Setup for GRO checksum validation */
4286 switch (skb->ip_summed) {
4287 case CHECKSUM_COMPLETE:
4288 NAPI_GRO_CB(skb)->csum = skb->csum;
4289 NAPI_GRO_CB(skb)->csum_valid = 1;
4290 NAPI_GRO_CB(skb)->csum_cnt = 0;
4292 case CHECKSUM_UNNECESSARY:
4293 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4294 NAPI_GRO_CB(skb)->csum_valid = 0;
4297 NAPI_GRO_CB(skb)->csum_cnt = 0;
4298 NAPI_GRO_CB(skb)->csum_valid = 0;
4301 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4306 if (&ptype->list == head)
4309 same_flow = NAPI_GRO_CB(skb)->same_flow;
4310 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4313 struct sk_buff *nskb = *pp;
4317 napi_gro_complete(nskb);
4324 if (NAPI_GRO_CB(skb)->flush)
4327 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4328 struct sk_buff *nskb = napi->gro_list;
4330 /* locate the end of the list to select the 'oldest' flow */
4331 while (nskb->next) {
4337 napi_gro_complete(nskb);
4341 NAPI_GRO_CB(skb)->count = 1;
4342 NAPI_GRO_CB(skb)->age = jiffies;
4343 NAPI_GRO_CB(skb)->last = skb;
4344 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4345 skb->next = napi->gro_list;
4346 napi->gro_list = skb;
4350 grow = skb_gro_offset(skb) - skb_headlen(skb);
4352 gro_pull_from_frag0(skb, grow);
4361 struct packet_offload *gro_find_receive_by_type(__be16 type)
4363 struct list_head *offload_head = &offload_base;
4364 struct packet_offload *ptype;
4366 list_for_each_entry_rcu(ptype, offload_head, list) {
4367 if (ptype->type != type || !ptype->callbacks.gro_receive)
4373 EXPORT_SYMBOL(gro_find_receive_by_type);
4375 struct packet_offload *gro_find_complete_by_type(__be16 type)
4377 struct list_head *offload_head = &offload_base;
4378 struct packet_offload *ptype;
4380 list_for_each_entry_rcu(ptype, offload_head, list) {
4381 if (ptype->type != type || !ptype->callbacks.gro_complete)
4387 EXPORT_SYMBOL(gro_find_complete_by_type);
4389 static void napi_skb_free_stolen_head(struct sk_buff *skb)
4392 kmem_cache_free(skbuff_head_cache, skb);
4395 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4399 if (netif_receive_skb_internal(skb))
4407 case GRO_MERGED_FREE:
4408 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4409 napi_skb_free_stolen_head(skb);
4422 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4424 trace_napi_gro_receive_entry(skb);
4426 skb_gro_reset_offset(skb);
4428 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4430 EXPORT_SYMBOL(napi_gro_receive);
4432 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4434 if (unlikely(skb->pfmemalloc)) {
4438 __skb_pull(skb, skb_headlen(skb));
4439 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4440 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4442 skb->dev = napi->dev;
4444 skb->encapsulation = 0;
4445 skb_shinfo(skb)->gso_type = 0;
4446 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4451 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4453 struct sk_buff *skb = napi->skb;
4456 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4461 EXPORT_SYMBOL(napi_get_frags);
4463 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4464 struct sk_buff *skb,
4470 __skb_push(skb, ETH_HLEN);
4471 skb->protocol = eth_type_trans(skb, skb->dev);
4472 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4477 napi_reuse_skb(napi, skb);
4480 case GRO_MERGED_FREE:
4481 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4482 napi_skb_free_stolen_head(skb);
4484 napi_reuse_skb(napi, skb);
4494 /* Upper GRO stack assumes network header starts at gro_offset=0
4495 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4496 * We copy ethernet header into skb->data to have a common layout.
4498 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4500 struct sk_buff *skb = napi->skb;
4501 const struct ethhdr *eth;
4502 unsigned int hlen = sizeof(*eth);
4506 skb_reset_mac_header(skb);
4507 skb_gro_reset_offset(skb);
4509 eth = skb_gro_header_fast(skb, 0);
4510 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4511 eth = skb_gro_header_slow(skb, hlen, 0);
4512 if (unlikely(!eth)) {
4513 napi_reuse_skb(napi, skb);
4517 gro_pull_from_frag0(skb, hlen);
4518 NAPI_GRO_CB(skb)->frag0 += hlen;
4519 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4521 __skb_pull(skb, hlen);
4524 * This works because the only protocols we care about don't require
4526 * We'll fix it up properly in napi_frags_finish()
4528 skb->protocol = eth->h_proto;
4533 gro_result_t napi_gro_frags(struct napi_struct *napi)
4535 struct sk_buff *skb = napi_frags_skb(napi);
4540 trace_napi_gro_frags_entry(skb);
4542 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4544 EXPORT_SYMBOL(napi_gro_frags);
4546 /* Compute the checksum from gro_offset and return the folded value
4547 * after adding in any pseudo checksum.
4549 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4554 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4556 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4557 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4559 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4560 !skb->csum_complete_sw)
4561 netdev_rx_csum_fault(skb->dev);
4564 NAPI_GRO_CB(skb)->csum = wsum;
4565 NAPI_GRO_CB(skb)->csum_valid = 1;
4569 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4572 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4573 * Note: called with local irq disabled, but exits with local irq enabled.
4575 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4578 struct softnet_data *remsd = sd->rps_ipi_list;
4581 sd->rps_ipi_list = NULL;
4585 /* Send pending IPI's to kick RPS processing on remote cpus. */
4587 struct softnet_data *next = remsd->rps_ipi_next;
4589 if (cpu_online(remsd->cpu))
4590 smp_call_function_single_async(remsd->cpu,
4599 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4602 return sd->rps_ipi_list != NULL;
4608 static int process_backlog(struct napi_struct *napi, int quota)
4611 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4613 /* Check if we have pending ipi, its better to send them now,
4614 * not waiting net_rx_action() end.
4616 if (sd_has_rps_ipi_waiting(sd)) {
4617 local_irq_disable();
4618 net_rps_action_and_irq_enable(sd);
4621 napi->weight = weight_p;
4622 local_irq_disable();
4624 struct sk_buff *skb;
4626 while ((skb = __skb_dequeue(&sd->process_queue))) {
4629 __netif_receive_skb(skb);
4631 local_irq_disable();
4632 input_queue_head_incr(sd);
4633 if (++work >= quota) {
4640 if (skb_queue_empty(&sd->input_pkt_queue)) {
4642 * Inline a custom version of __napi_complete().
4643 * only current cpu owns and manipulates this napi,
4644 * and NAPI_STATE_SCHED is the only possible flag set
4646 * We can use a plain write instead of clear_bit(),
4647 * and we dont need an smp_mb() memory barrier.
4655 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4656 &sd->process_queue);
4665 * __napi_schedule - schedule for receive
4666 * @n: entry to schedule
4668 * The entry's receive function will be scheduled to run.
4669 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4671 void __napi_schedule(struct napi_struct *n)
4673 unsigned long flags;
4675 local_irq_save(flags);
4676 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4677 local_irq_restore(flags);
4679 EXPORT_SYMBOL(__napi_schedule);
4682 * __napi_schedule_irqoff - schedule for receive
4683 * @n: entry to schedule
4685 * Variant of __napi_schedule() assuming hard irqs are masked
4687 void __napi_schedule_irqoff(struct napi_struct *n)
4689 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4691 EXPORT_SYMBOL(__napi_schedule_irqoff);
4693 void __napi_complete(struct napi_struct *n)
4695 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4697 list_del_init(&n->poll_list);
4698 smp_mb__before_atomic();
4699 clear_bit(NAPI_STATE_SCHED, &n->state);
4701 EXPORT_SYMBOL(__napi_complete);
4703 void napi_complete_done(struct napi_struct *n, int work_done)
4705 unsigned long flags;
4708 * don't let napi dequeue from the cpu poll list
4709 * just in case its running on a different cpu
4711 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4715 unsigned long timeout = 0;
4718 timeout = n->dev->gro_flush_timeout;
4721 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4722 HRTIMER_MODE_REL_PINNED);
4724 napi_gro_flush(n, false);
4726 if (likely(list_empty(&n->poll_list))) {
4727 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4729 /* If n->poll_list is not empty, we need to mask irqs */
4730 local_irq_save(flags);
4732 local_irq_restore(flags);
4735 EXPORT_SYMBOL(napi_complete_done);
4737 /* must be called under rcu_read_lock(), as we dont take a reference */
4738 struct napi_struct *napi_by_id(unsigned int napi_id)
4740 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4741 struct napi_struct *napi;
4743 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4744 if (napi->napi_id == napi_id)
4749 EXPORT_SYMBOL_GPL(napi_by_id);
4751 void napi_hash_add(struct napi_struct *napi)
4753 if (test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
4756 spin_lock(&napi_hash_lock);
4758 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */
4760 if (unlikely(++napi_gen_id < NR_CPUS + 1))
4761 napi_gen_id = NR_CPUS + 1;
4762 } while (napi_by_id(napi_gen_id));
4763 napi->napi_id = napi_gen_id;
4765 hlist_add_head_rcu(&napi->napi_hash_node,
4766 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4768 spin_unlock(&napi_hash_lock);
4770 EXPORT_SYMBOL_GPL(napi_hash_add);
4772 /* Warning : caller is responsible to make sure rcu grace period
4773 * is respected before freeing memory containing @napi
4775 void napi_hash_del(struct napi_struct *napi)
4777 spin_lock(&napi_hash_lock);
4779 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4780 hlist_del_rcu(&napi->napi_hash_node);
4782 spin_unlock(&napi_hash_lock);
4784 EXPORT_SYMBOL_GPL(napi_hash_del);
4786 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4788 struct napi_struct *napi;
4790 napi = container_of(timer, struct napi_struct, timer);
4792 napi_schedule(napi);
4794 return HRTIMER_NORESTART;
4797 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4798 int (*poll)(struct napi_struct *, int), int weight)
4800 INIT_LIST_HEAD(&napi->poll_list);
4801 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4802 napi->timer.function = napi_watchdog;
4803 napi->gro_count = 0;
4804 napi->gro_list = NULL;
4807 if (weight > NAPI_POLL_WEIGHT)
4808 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4810 napi->weight = weight;
4811 list_add(&napi->dev_list, &dev->napi_list);
4813 #ifdef CONFIG_NETPOLL
4814 spin_lock_init(&napi->poll_lock);
4815 napi->poll_owner = -1;
4817 set_bit(NAPI_STATE_SCHED, &napi->state);
4819 EXPORT_SYMBOL(netif_napi_add);
4821 void napi_disable(struct napi_struct *n)
4824 set_bit(NAPI_STATE_DISABLE, &n->state);
4826 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4828 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
4831 hrtimer_cancel(&n->timer);
4833 clear_bit(NAPI_STATE_DISABLE, &n->state);
4835 EXPORT_SYMBOL(napi_disable);
4837 void netif_napi_del(struct napi_struct *napi)
4839 list_del_init(&napi->dev_list);
4840 napi_free_frags(napi);
4842 kfree_skb_list(napi->gro_list);
4843 napi->gro_list = NULL;
4844 napi->gro_count = 0;
4846 EXPORT_SYMBOL(netif_napi_del);
4848 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4853 list_del_init(&n->poll_list);
4855 have = netpoll_poll_lock(n);
4859 /* This NAPI_STATE_SCHED test is for avoiding a race
4860 * with netpoll's poll_napi(). Only the entity which
4861 * obtains the lock and sees NAPI_STATE_SCHED set will
4862 * actually make the ->poll() call. Therefore we avoid
4863 * accidentally calling ->poll() when NAPI is not scheduled.
4866 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4867 work = n->poll(n, weight);
4871 WARN_ON_ONCE(work > weight);
4873 if (likely(work < weight))
4876 /* Drivers must not modify the NAPI state if they
4877 * consume the entire weight. In such cases this code
4878 * still "owns" the NAPI instance and therefore can
4879 * move the instance around on the list at-will.
4881 if (unlikely(napi_disable_pending(n))) {
4887 /* flush too old packets
4888 * If HZ < 1000, flush all packets.
4890 napi_gro_flush(n, HZ >= 1000);
4893 /* Some drivers may have called napi_schedule
4894 * prior to exhausting their budget.
4896 if (unlikely(!list_empty(&n->poll_list))) {
4897 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4898 n->dev ? n->dev->name : "backlog");
4902 list_add_tail(&n->poll_list, repoll);
4905 netpoll_poll_unlock(have);
4910 static void net_rx_action(struct softirq_action *h)
4912 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4913 unsigned long time_limit = jiffies + 2;
4914 int budget = netdev_budget;
4918 local_irq_disable();
4919 list_splice_init(&sd->poll_list, &list);
4923 struct napi_struct *n;
4925 if (list_empty(&list)) {
4926 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4931 n = list_first_entry(&list, struct napi_struct, poll_list);
4932 budget -= napi_poll(n, &repoll);
4934 /* If softirq window is exhausted then punt.
4935 * Allow this to run for 2 jiffies since which will allow
4936 * an average latency of 1.5/HZ.
4938 if (unlikely(budget <= 0 ||
4939 time_after_eq(jiffies, time_limit))) {
4945 local_irq_disable();
4947 list_splice_tail_init(&sd->poll_list, &list);
4948 list_splice_tail(&repoll, &list);
4949 list_splice(&list, &sd->poll_list);
4950 if (!list_empty(&sd->poll_list))
4951 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4953 net_rps_action_and_irq_enable(sd);
4956 struct netdev_adjacent {
4957 struct net_device *dev;
4959 /* upper master flag, there can only be one master device per list */
4962 /* counter for the number of times this device was added to us */
4965 /* private field for the users */
4968 struct list_head list;
4969 struct rcu_head rcu;
4972 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
4973 struct list_head *adj_list)
4975 struct netdev_adjacent *adj;
4977 list_for_each_entry(adj, adj_list, list) {
4978 if (adj->dev == adj_dev)
4985 * netdev_has_upper_dev - Check if device is linked to an upper device
4987 * @upper_dev: upper device to check
4989 * Find out if a device is linked to specified upper device and return true
4990 * in case it is. Note that this checks only immediate upper device,
4991 * not through a complete stack of devices. The caller must hold the RTNL lock.
4993 bool netdev_has_upper_dev(struct net_device *dev,
4994 struct net_device *upper_dev)
4998 return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper);
5000 EXPORT_SYMBOL(netdev_has_upper_dev);
5003 * netdev_has_any_upper_dev - Check if device is linked to some device
5006 * Find out if a device is linked to an upper device and return true in case
5007 * it is. The caller must hold the RTNL lock.
5009 static bool netdev_has_any_upper_dev(struct net_device *dev)
5013 return !list_empty(&dev->all_adj_list.upper);
5017 * netdev_master_upper_dev_get - Get master upper device
5020 * Find a master upper device and return pointer to it or NULL in case
5021 * it's not there. The caller must hold the RTNL lock.
5023 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5025 struct netdev_adjacent *upper;
5029 if (list_empty(&dev->adj_list.upper))
5032 upper = list_first_entry(&dev->adj_list.upper,
5033 struct netdev_adjacent, list);
5034 if (likely(upper->master))
5038 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5040 void *netdev_adjacent_get_private(struct list_head *adj_list)
5042 struct netdev_adjacent *adj;
5044 adj = list_entry(adj_list, struct netdev_adjacent, list);
5046 return adj->private;
5048 EXPORT_SYMBOL(netdev_adjacent_get_private);
5051 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5053 * @iter: list_head ** of the current position
5055 * Gets the next device from the dev's upper list, starting from iter
5056 * position. The caller must hold RCU read lock.
5058 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5059 struct list_head **iter)
5061 struct netdev_adjacent *upper;
5063 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5065 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5067 if (&upper->list == &dev->adj_list.upper)
5070 *iter = &upper->list;
5074 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5077 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
5079 * @iter: list_head ** of the current position
5081 * Gets the next device from the dev's upper list, starting from iter
5082 * position. The caller must hold RCU read lock.
5084 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
5085 struct list_head **iter)
5087 struct netdev_adjacent *upper;
5089 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5091 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5093 if (&upper->list == &dev->all_adj_list.upper)
5096 *iter = &upper->list;
5100 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
5103 * netdev_lower_get_next_private - Get the next ->private from the
5104 * lower neighbour list
5106 * @iter: list_head ** of the current position
5108 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5109 * list, starting from iter position. The caller must hold either hold the
5110 * RTNL lock or its own locking that guarantees that the neighbour lower
5111 * list will remain unchanged.
5113 void *netdev_lower_get_next_private(struct net_device *dev,
5114 struct list_head **iter)
5116 struct netdev_adjacent *lower;
5118 lower = list_entry(*iter, struct netdev_adjacent, list);
5120 if (&lower->list == &dev->adj_list.lower)
5123 *iter = lower->list.next;
5125 return lower->private;
5127 EXPORT_SYMBOL(netdev_lower_get_next_private);
5130 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5131 * lower neighbour list, RCU
5134 * @iter: list_head ** of the current position
5136 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5137 * list, starting from iter position. The caller must hold RCU read lock.
5139 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5140 struct list_head **iter)
5142 struct netdev_adjacent *lower;
5144 WARN_ON_ONCE(!rcu_read_lock_held());
5146 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5148 if (&lower->list == &dev->adj_list.lower)
5151 *iter = &lower->list;
5153 return lower->private;
5155 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5158 * netdev_lower_get_next - Get the next device from the lower neighbour
5161 * @iter: list_head ** of the current position
5163 * Gets the next netdev_adjacent from the dev's lower neighbour
5164 * list, starting from iter position. The caller must hold RTNL lock or
5165 * its own locking that guarantees that the neighbour lower
5166 * list will remain unchanged.
5168 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5170 struct netdev_adjacent *lower;
5172 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5174 if (&lower->list == &dev->adj_list.lower)
5177 *iter = &lower->list;
5181 EXPORT_SYMBOL(netdev_lower_get_next);
5184 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5185 * lower neighbour list, RCU
5189 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5190 * list. The caller must hold RCU read lock.
5192 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5194 struct netdev_adjacent *lower;
5196 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5197 struct netdev_adjacent, list);
5199 return lower->private;
5202 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5205 * netdev_master_upper_dev_get_rcu - Get master upper device
5208 * Find a master upper device and return pointer to it or NULL in case
5209 * it's not there. The caller must hold the RCU read lock.
5211 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5213 struct netdev_adjacent *upper;
5215 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5216 struct netdev_adjacent, list);
5217 if (upper && likely(upper->master))
5221 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5223 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5224 struct net_device *adj_dev,
5225 struct list_head *dev_list)
5227 char linkname[IFNAMSIZ+7];
5228 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5229 "upper_%s" : "lower_%s", adj_dev->name);
5230 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5233 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5235 struct list_head *dev_list)
5237 char linkname[IFNAMSIZ+7];
5238 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5239 "upper_%s" : "lower_%s", name);
5240 sysfs_remove_link(&(dev->dev.kobj), linkname);
5243 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5244 struct net_device *adj_dev,
5245 struct list_head *dev_list)
5247 return (dev_list == &dev->adj_list.upper ||
5248 dev_list == &dev->adj_list.lower) &&
5249 net_eq(dev_net(dev), dev_net(adj_dev));
5252 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5253 struct net_device *adj_dev,
5255 struct list_head *dev_list,
5256 void *private, bool master)
5258 struct netdev_adjacent *adj;
5261 adj = __netdev_find_adj(adj_dev, dev_list);
5264 adj->ref_nr += ref_nr;
5268 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5273 adj->master = master;
5274 adj->ref_nr = ref_nr;
5275 adj->private = private;
5278 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5279 adj_dev->name, dev->name, adj_dev->name);
5281 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5282 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5287 /* Ensure that master link is always the first item in list. */
5289 ret = sysfs_create_link(&(dev->dev.kobj),
5290 &(adj_dev->dev.kobj), "master");
5292 goto remove_symlinks;
5294 list_add_rcu(&adj->list, dev_list);
5296 list_add_tail_rcu(&adj->list, dev_list);
5302 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5303 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5311 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5312 struct net_device *adj_dev,
5314 struct list_head *dev_list)
5316 struct netdev_adjacent *adj;
5318 adj = __netdev_find_adj(adj_dev, dev_list);
5321 pr_err("tried to remove device %s from %s\n",
5322 dev->name, adj_dev->name);
5326 if (adj->ref_nr > ref_nr) {
5327 pr_debug("%s to %s ref_nr-%d = %d\n", dev->name, adj_dev->name,
5328 ref_nr, adj->ref_nr-ref_nr);
5329 adj->ref_nr -= ref_nr;
5334 sysfs_remove_link(&(dev->dev.kobj), "master");
5336 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5337 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5339 list_del_rcu(&adj->list);
5340 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5341 adj_dev->name, dev->name, adj_dev->name);
5343 kfree_rcu(adj, rcu);
5346 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5347 struct net_device *upper_dev,
5349 struct list_head *up_list,
5350 struct list_head *down_list,
5351 void *private, bool master)
5355 ret = __netdev_adjacent_dev_insert(dev, upper_dev, ref_nr, up_list,
5360 ret = __netdev_adjacent_dev_insert(upper_dev, dev, ref_nr, down_list,
5363 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
5370 static int __netdev_adjacent_dev_link(struct net_device *dev,
5371 struct net_device *upper_dev,
5374 return __netdev_adjacent_dev_link_lists(dev, upper_dev, ref_nr,
5375 &dev->all_adj_list.upper,
5376 &upper_dev->all_adj_list.lower,
5380 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5381 struct net_device *upper_dev,
5383 struct list_head *up_list,
5384 struct list_head *down_list)
5386 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
5387 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
5390 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5391 struct net_device *upper_dev,
5394 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, ref_nr,
5395 &dev->all_adj_list.upper,
5396 &upper_dev->all_adj_list.lower);
5399 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5400 struct net_device *upper_dev,
5401 void *private, bool master)
5403 int ret = __netdev_adjacent_dev_link(dev, upper_dev, 1);
5408 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, 1,
5409 &dev->adj_list.upper,
5410 &upper_dev->adj_list.lower,
5413 __netdev_adjacent_dev_unlink(dev, upper_dev, 1);
5420 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5421 struct net_device *upper_dev)
5423 __netdev_adjacent_dev_unlink(dev, upper_dev, 1);
5424 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
5425 &dev->adj_list.upper,
5426 &upper_dev->adj_list.lower);
5429 static int __netdev_upper_dev_link(struct net_device *dev,
5430 struct net_device *upper_dev, bool master,
5433 struct netdev_notifier_changeupper_info changeupper_info;
5434 struct netdev_adjacent *i, *j, *to_i, *to_j;
5439 if (dev == upper_dev)
5442 /* To prevent loops, check if dev is not upper device to upper_dev. */
5443 if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper))
5446 if (__netdev_find_adj(upper_dev, &dev->adj_list.upper))
5449 if (master && netdev_master_upper_dev_get(dev))
5452 changeupper_info.upper_dev = upper_dev;
5453 changeupper_info.master = master;
5454 changeupper_info.linking = true;
5456 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5457 &changeupper_info.info);
5458 ret = notifier_to_errno(ret);
5462 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5467 /* Now that we linked these devs, make all the upper_dev's
5468 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5469 * versa, and don't forget the devices itself. All of these
5470 * links are non-neighbours.
5472 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5473 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5474 pr_debug("Interlinking %s with %s, non-neighbour\n",
5475 i->dev->name, j->dev->name);
5476 ret = __netdev_adjacent_dev_link(i->dev, j->dev, i->ref_nr);
5482 /* add dev to every upper_dev's upper device */
5483 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5484 pr_debug("linking %s's upper device %s with %s\n",
5485 upper_dev->name, i->dev->name, dev->name);
5486 ret = __netdev_adjacent_dev_link(dev, i->dev, i->ref_nr);
5488 goto rollback_upper_mesh;
5491 /* add upper_dev to every dev's lower device */
5492 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5493 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5494 i->dev->name, upper_dev->name);
5495 ret = __netdev_adjacent_dev_link(i->dev, upper_dev, i->ref_nr);
5497 goto rollback_lower_mesh;
5500 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5501 &changeupper_info.info);
5504 rollback_lower_mesh:
5506 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5509 __netdev_adjacent_dev_unlink(i->dev, upper_dev, i->ref_nr);
5514 rollback_upper_mesh:
5516 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5519 __netdev_adjacent_dev_unlink(dev, i->dev, i->ref_nr);
5527 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5528 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5529 if (i == to_i && j == to_j)
5531 __netdev_adjacent_dev_unlink(i->dev, j->dev, i->ref_nr);
5537 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5543 * netdev_upper_dev_link - Add a link to the upper device
5545 * @upper_dev: new upper device
5547 * Adds a link to device which is upper to this one. The caller must hold
5548 * the RTNL lock. On a failure a negative errno code is returned.
5549 * On success the reference counts are adjusted and the function
5552 int netdev_upper_dev_link(struct net_device *dev,
5553 struct net_device *upper_dev)
5555 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5557 EXPORT_SYMBOL(netdev_upper_dev_link);
5560 * netdev_master_upper_dev_link - Add a master link to the upper device
5562 * @upper_dev: new upper device
5564 * Adds a link to device which is upper to this one. In this case, only
5565 * one master upper device can be linked, although other non-master devices
5566 * might be linked as well. The caller must hold the RTNL lock.
5567 * On a failure a negative errno code is returned. On success the reference
5568 * counts are adjusted and the function returns zero.
5570 int netdev_master_upper_dev_link(struct net_device *dev,
5571 struct net_device *upper_dev)
5573 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5575 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5577 int netdev_master_upper_dev_link_private(struct net_device *dev,
5578 struct net_device *upper_dev,
5581 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5583 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5586 * netdev_upper_dev_unlink - Removes a link to upper device
5588 * @upper_dev: new upper device
5590 * Removes a link to device which is upper to this one. The caller must hold
5593 void netdev_upper_dev_unlink(struct net_device *dev,
5594 struct net_device *upper_dev)
5596 struct netdev_notifier_changeupper_info changeupper_info;
5597 struct netdev_adjacent *i, *j;
5600 changeupper_info.upper_dev = upper_dev;
5601 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
5602 changeupper_info.linking = false;
5604 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5605 &changeupper_info.info);
5607 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5609 /* Here is the tricky part. We must remove all dev's lower
5610 * devices from all upper_dev's upper devices and vice
5611 * versa, to maintain the graph relationship.
5613 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5614 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5615 __netdev_adjacent_dev_unlink(i->dev, j->dev, i->ref_nr);
5617 /* remove also the devices itself from lower/upper device
5620 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5621 __netdev_adjacent_dev_unlink(i->dev, upper_dev, i->ref_nr);
5623 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5624 __netdev_adjacent_dev_unlink(dev, i->dev, i->ref_nr);
5626 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5627 &changeupper_info.info);
5629 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5632 * netdev_bonding_info_change - Dispatch event about slave change
5634 * @bonding_info: info to dispatch
5636 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5637 * The caller must hold the RTNL lock.
5639 void netdev_bonding_info_change(struct net_device *dev,
5640 struct netdev_bonding_info *bonding_info)
5642 struct netdev_notifier_bonding_info info;
5644 memcpy(&info.bonding_info, bonding_info,
5645 sizeof(struct netdev_bonding_info));
5646 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5649 EXPORT_SYMBOL(netdev_bonding_info_change);
5651 static void netdev_adjacent_add_links(struct net_device *dev)
5653 struct netdev_adjacent *iter;
5655 struct net *net = dev_net(dev);
5657 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5658 if (!net_eq(net,dev_net(iter->dev)))
5660 netdev_adjacent_sysfs_add(iter->dev, dev,
5661 &iter->dev->adj_list.lower);
5662 netdev_adjacent_sysfs_add(dev, iter->dev,
5663 &dev->adj_list.upper);
5666 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5667 if (!net_eq(net,dev_net(iter->dev)))
5669 netdev_adjacent_sysfs_add(iter->dev, dev,
5670 &iter->dev->adj_list.upper);
5671 netdev_adjacent_sysfs_add(dev, iter->dev,
5672 &dev->adj_list.lower);
5676 static void netdev_adjacent_del_links(struct net_device *dev)
5678 struct netdev_adjacent *iter;
5680 struct net *net = dev_net(dev);
5682 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5683 if (!net_eq(net,dev_net(iter->dev)))
5685 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5686 &iter->dev->adj_list.lower);
5687 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5688 &dev->adj_list.upper);
5691 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5692 if (!net_eq(net,dev_net(iter->dev)))
5694 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5695 &iter->dev->adj_list.upper);
5696 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5697 &dev->adj_list.lower);
5701 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5703 struct netdev_adjacent *iter;
5705 struct net *net = dev_net(dev);
5707 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5708 if (!net_eq(net,dev_net(iter->dev)))
5710 netdev_adjacent_sysfs_del(iter->dev, oldname,
5711 &iter->dev->adj_list.lower);
5712 netdev_adjacent_sysfs_add(iter->dev, dev,
5713 &iter->dev->adj_list.lower);
5716 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5717 if (!net_eq(net,dev_net(iter->dev)))
5719 netdev_adjacent_sysfs_del(iter->dev, oldname,
5720 &iter->dev->adj_list.upper);
5721 netdev_adjacent_sysfs_add(iter->dev, dev,
5722 &iter->dev->adj_list.upper);
5726 void *netdev_lower_dev_get_private(struct net_device *dev,
5727 struct net_device *lower_dev)
5729 struct netdev_adjacent *lower;
5733 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
5737 return lower->private;
5739 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5742 int dev_get_nest_level(struct net_device *dev,
5743 bool (*type_check)(struct net_device *dev))
5745 struct net_device *lower = NULL;
5746 struct list_head *iter;
5752 netdev_for_each_lower_dev(dev, lower, iter) {
5753 nest = dev_get_nest_level(lower, type_check);
5754 if (max_nest < nest)
5758 if (type_check(dev))
5763 EXPORT_SYMBOL(dev_get_nest_level);
5765 static void dev_change_rx_flags(struct net_device *dev, int flags)
5767 const struct net_device_ops *ops = dev->netdev_ops;
5769 if (ops->ndo_change_rx_flags)
5770 ops->ndo_change_rx_flags(dev, flags);
5773 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5775 unsigned int old_flags = dev->flags;
5781 dev->flags |= IFF_PROMISC;
5782 dev->promiscuity += inc;
5783 if (dev->promiscuity == 0) {
5786 * If inc causes overflow, untouch promisc and return error.
5789 dev->flags &= ~IFF_PROMISC;
5791 dev->promiscuity -= inc;
5792 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5797 if (dev->flags != old_flags) {
5798 pr_info("device %s %s promiscuous mode\n",
5800 dev->flags & IFF_PROMISC ? "entered" : "left");
5801 if (audit_enabled) {
5802 current_uid_gid(&uid, &gid);
5803 audit_log(current->audit_context, GFP_ATOMIC,
5804 AUDIT_ANOM_PROMISCUOUS,
5805 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5806 dev->name, (dev->flags & IFF_PROMISC),
5807 (old_flags & IFF_PROMISC),
5808 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5809 from_kuid(&init_user_ns, uid),
5810 from_kgid(&init_user_ns, gid),
5811 audit_get_sessionid(current));
5814 dev_change_rx_flags(dev, IFF_PROMISC);
5817 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5822 * dev_set_promiscuity - update promiscuity count on a device
5826 * Add or remove promiscuity from a device. While the count in the device
5827 * remains above zero the interface remains promiscuous. Once it hits zero
5828 * the device reverts back to normal filtering operation. A negative inc
5829 * value is used to drop promiscuity on the device.
5830 * Return 0 if successful or a negative errno code on error.
5832 int dev_set_promiscuity(struct net_device *dev, int inc)
5834 unsigned int old_flags = dev->flags;
5837 err = __dev_set_promiscuity(dev, inc, true);
5840 if (dev->flags != old_flags)
5841 dev_set_rx_mode(dev);
5844 EXPORT_SYMBOL(dev_set_promiscuity);
5846 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5848 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5852 dev->flags |= IFF_ALLMULTI;
5853 dev->allmulti += inc;
5854 if (dev->allmulti == 0) {
5857 * If inc causes overflow, untouch allmulti and return error.
5860 dev->flags &= ~IFF_ALLMULTI;
5862 dev->allmulti -= inc;
5863 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5868 if (dev->flags ^ old_flags) {
5869 dev_change_rx_flags(dev, IFF_ALLMULTI);
5870 dev_set_rx_mode(dev);
5872 __dev_notify_flags(dev, old_flags,
5873 dev->gflags ^ old_gflags);
5879 * dev_set_allmulti - update allmulti count on a device
5883 * Add or remove reception of all multicast frames to a device. While the
5884 * count in the device remains above zero the interface remains listening
5885 * to all interfaces. Once it hits zero the device reverts back to normal
5886 * filtering operation. A negative @inc value is used to drop the counter
5887 * when releasing a resource needing all multicasts.
5888 * Return 0 if successful or a negative errno code on error.
5891 int dev_set_allmulti(struct net_device *dev, int inc)
5893 return __dev_set_allmulti(dev, inc, true);
5895 EXPORT_SYMBOL(dev_set_allmulti);
5898 * Upload unicast and multicast address lists to device and
5899 * configure RX filtering. When the device doesn't support unicast
5900 * filtering it is put in promiscuous mode while unicast addresses
5903 void __dev_set_rx_mode(struct net_device *dev)
5905 const struct net_device_ops *ops = dev->netdev_ops;
5907 /* dev_open will call this function so the list will stay sane. */
5908 if (!(dev->flags&IFF_UP))
5911 if (!netif_device_present(dev))
5914 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5915 /* Unicast addresses changes may only happen under the rtnl,
5916 * therefore calling __dev_set_promiscuity here is safe.
5918 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5919 __dev_set_promiscuity(dev, 1, false);
5920 dev->uc_promisc = true;
5921 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5922 __dev_set_promiscuity(dev, -1, false);
5923 dev->uc_promisc = false;
5927 if (ops->ndo_set_rx_mode)
5928 ops->ndo_set_rx_mode(dev);
5931 void dev_set_rx_mode(struct net_device *dev)
5933 netif_addr_lock_bh(dev);
5934 __dev_set_rx_mode(dev);
5935 netif_addr_unlock_bh(dev);
5939 * dev_get_flags - get flags reported to userspace
5942 * Get the combination of flag bits exported through APIs to userspace.
5944 unsigned int dev_get_flags(const struct net_device *dev)
5948 flags = (dev->flags & ~(IFF_PROMISC |
5953 (dev->gflags & (IFF_PROMISC |
5956 if (netif_running(dev)) {
5957 if (netif_oper_up(dev))
5958 flags |= IFF_RUNNING;
5959 if (netif_carrier_ok(dev))
5960 flags |= IFF_LOWER_UP;
5961 if (netif_dormant(dev))
5962 flags |= IFF_DORMANT;
5967 EXPORT_SYMBOL(dev_get_flags);
5969 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5971 unsigned int old_flags = dev->flags;
5977 * Set the flags on our device.
5980 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5981 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5983 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5987 * Load in the correct multicast list now the flags have changed.
5990 if ((old_flags ^ flags) & IFF_MULTICAST)
5991 dev_change_rx_flags(dev, IFF_MULTICAST);
5993 dev_set_rx_mode(dev);
5996 * Have we downed the interface. We handle IFF_UP ourselves
5997 * according to user attempts to set it, rather than blindly
6002 if ((old_flags ^ flags) & IFF_UP)
6003 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
6005 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6006 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6007 unsigned int old_flags = dev->flags;
6009 dev->gflags ^= IFF_PROMISC;
6011 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6012 if (dev->flags != old_flags)
6013 dev_set_rx_mode(dev);
6016 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6017 is important. Some (broken) drivers set IFF_PROMISC, when
6018 IFF_ALLMULTI is requested not asking us and not reporting.
6020 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6021 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6023 dev->gflags ^= IFF_ALLMULTI;
6024 __dev_set_allmulti(dev, inc, false);
6030 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6031 unsigned int gchanges)
6033 unsigned int changes = dev->flags ^ old_flags;
6036 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6038 if (changes & IFF_UP) {
6039 if (dev->flags & IFF_UP)
6040 call_netdevice_notifiers(NETDEV_UP, dev);
6042 call_netdevice_notifiers(NETDEV_DOWN, dev);
6045 if (dev->flags & IFF_UP &&
6046 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6047 struct netdev_notifier_change_info change_info;
6049 change_info.flags_changed = changes;
6050 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6056 * dev_change_flags - change device settings
6058 * @flags: device state flags
6060 * Change settings on device based state flags. The flags are
6061 * in the userspace exported format.
6063 int dev_change_flags(struct net_device *dev, unsigned int flags)
6066 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6068 ret = __dev_change_flags(dev, flags);
6072 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6073 __dev_notify_flags(dev, old_flags, changes);
6076 EXPORT_SYMBOL(dev_change_flags);
6078 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6080 const struct net_device_ops *ops = dev->netdev_ops;
6082 if (ops->ndo_change_mtu)
6083 return ops->ndo_change_mtu(dev, new_mtu);
6090 * dev_set_mtu - Change maximum transfer unit
6092 * @new_mtu: new transfer unit
6094 * Change the maximum transfer size of the network device.
6096 int dev_set_mtu(struct net_device *dev, int new_mtu)
6100 if (new_mtu == dev->mtu)
6103 /* MTU must be positive. */
6107 if (!netif_device_present(dev))
6110 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6111 err = notifier_to_errno(err);
6115 orig_mtu = dev->mtu;
6116 err = __dev_set_mtu(dev, new_mtu);
6119 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6120 err = notifier_to_errno(err);
6122 /* setting mtu back and notifying everyone again,
6123 * so that they have a chance to revert changes.
6125 __dev_set_mtu(dev, orig_mtu);
6126 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6131 EXPORT_SYMBOL(dev_set_mtu);
6134 * dev_set_group - Change group this device belongs to
6136 * @new_group: group this device should belong to
6138 void dev_set_group(struct net_device *dev, int new_group)
6140 dev->group = new_group;
6142 EXPORT_SYMBOL(dev_set_group);
6145 * dev_set_mac_address - Change Media Access Control Address
6149 * Change the hardware (MAC) address of the device
6151 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6153 const struct net_device_ops *ops = dev->netdev_ops;
6156 if (!ops->ndo_set_mac_address)
6158 if (sa->sa_family != dev->type)
6160 if (!netif_device_present(dev))
6162 err = ops->ndo_set_mac_address(dev, sa);
6165 dev->addr_assign_type = NET_ADDR_SET;
6166 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6167 add_device_randomness(dev->dev_addr, dev->addr_len);
6170 EXPORT_SYMBOL(dev_set_mac_address);
6173 * dev_change_carrier - Change device carrier
6175 * @new_carrier: new value
6177 * Change device carrier
6179 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6181 const struct net_device_ops *ops = dev->netdev_ops;
6183 if (!ops->ndo_change_carrier)
6185 if (!netif_device_present(dev))
6187 return ops->ndo_change_carrier(dev, new_carrier);
6189 EXPORT_SYMBOL(dev_change_carrier);
6192 * dev_get_phys_port_id - Get device physical port ID
6196 * Get device physical port ID
6198 int dev_get_phys_port_id(struct net_device *dev,
6199 struct netdev_phys_item_id *ppid)
6201 const struct net_device_ops *ops = dev->netdev_ops;
6203 if (!ops->ndo_get_phys_port_id)
6205 return ops->ndo_get_phys_port_id(dev, ppid);
6207 EXPORT_SYMBOL(dev_get_phys_port_id);
6210 * dev_get_phys_port_name - Get device physical port name
6214 * Get device physical port name
6216 int dev_get_phys_port_name(struct net_device *dev,
6217 char *name, size_t len)
6219 const struct net_device_ops *ops = dev->netdev_ops;
6221 if (!ops->ndo_get_phys_port_name)
6223 return ops->ndo_get_phys_port_name(dev, name, len);
6225 EXPORT_SYMBOL(dev_get_phys_port_name);
6228 * dev_change_proto_down - update protocol port state information
6230 * @proto_down: new value
6232 * This info can be used by switch drivers to set the phys state of the
6235 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6237 const struct net_device_ops *ops = dev->netdev_ops;
6239 if (!ops->ndo_change_proto_down)
6241 if (!netif_device_present(dev))
6243 return ops->ndo_change_proto_down(dev, proto_down);
6245 EXPORT_SYMBOL(dev_change_proto_down);
6248 * dev_new_index - allocate an ifindex
6249 * @net: the applicable net namespace
6251 * Returns a suitable unique value for a new device interface
6252 * number. The caller must hold the rtnl semaphore or the
6253 * dev_base_lock to be sure it remains unique.
6255 static int dev_new_index(struct net *net)
6257 int ifindex = net->ifindex;
6261 if (!__dev_get_by_index(net, ifindex))
6262 return net->ifindex = ifindex;
6266 /* Delayed registration/unregisteration */
6267 static LIST_HEAD(net_todo_list);
6268 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6270 static void net_set_todo(struct net_device *dev)
6272 list_add_tail(&dev->todo_list, &net_todo_list);
6273 dev_net(dev)->dev_unreg_count++;
6276 static void rollback_registered_many(struct list_head *head)
6278 struct net_device *dev, *tmp;
6279 LIST_HEAD(close_head);
6281 BUG_ON(dev_boot_phase);
6284 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6285 /* Some devices call without registering
6286 * for initialization unwind. Remove those
6287 * devices and proceed with the remaining.
6289 if (dev->reg_state == NETREG_UNINITIALIZED) {
6290 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6294 list_del(&dev->unreg_list);
6297 dev->dismantle = true;
6298 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6301 /* If device is running, close it first. */
6302 list_for_each_entry(dev, head, unreg_list)
6303 list_add_tail(&dev->close_list, &close_head);
6304 dev_close_many(&close_head, true);
6306 list_for_each_entry(dev, head, unreg_list) {
6307 /* And unlink it from device chain. */
6308 unlist_netdevice(dev);
6310 dev->reg_state = NETREG_UNREGISTERING;
6311 on_each_cpu(flush_backlog, dev, 1);
6316 list_for_each_entry(dev, head, unreg_list) {
6317 struct sk_buff *skb = NULL;
6319 /* Shutdown queueing discipline. */
6323 /* Notify protocols, that we are about to destroy
6324 this device. They should clean all the things.
6326 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6328 if (!dev->rtnl_link_ops ||
6329 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6330 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6334 * Flush the unicast and multicast chains
6339 if (dev->netdev_ops->ndo_uninit)
6340 dev->netdev_ops->ndo_uninit(dev);
6343 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6345 /* Notifier chain MUST detach us all upper devices. */
6346 WARN_ON(netdev_has_any_upper_dev(dev));
6348 /* Remove entries from kobject tree */
6349 netdev_unregister_kobject(dev);
6351 /* Remove XPS queueing entries */
6352 netif_reset_xps_queues_gt(dev, 0);
6358 list_for_each_entry(dev, head, unreg_list)
6362 static void rollback_registered(struct net_device *dev)
6366 list_add(&dev->unreg_list, &single);
6367 rollback_registered_many(&single);
6371 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6372 struct net_device *upper, netdev_features_t features)
6374 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6375 netdev_features_t feature;
6378 for_each_netdev_feature(&upper_disables, feature_bit) {
6379 feature = __NETIF_F_BIT(feature_bit);
6380 if (!(upper->wanted_features & feature)
6381 && (features & feature)) {
6382 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6383 &feature, upper->name);
6384 features &= ~feature;
6391 static void netdev_sync_lower_features(struct net_device *upper,
6392 struct net_device *lower, netdev_features_t features)
6394 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6395 netdev_features_t feature;
6398 for_each_netdev_feature(&upper_disables, feature_bit) {
6399 feature = __NETIF_F_BIT(feature_bit);
6400 if (!(features & feature) && (lower->features & feature)) {
6401 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6402 &feature, lower->name);
6403 lower->wanted_features &= ~feature;
6404 netdev_update_features(lower);
6406 if (unlikely(lower->features & feature))
6407 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6408 &feature, lower->name);
6413 static netdev_features_t netdev_fix_features(struct net_device *dev,
6414 netdev_features_t features)
6416 /* Fix illegal checksum combinations */
6417 if ((features & NETIF_F_HW_CSUM) &&
6418 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6419 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6420 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6423 /* TSO requires that SG is present as well. */
6424 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6425 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6426 features &= ~NETIF_F_ALL_TSO;
6429 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6430 !(features & NETIF_F_IP_CSUM)) {
6431 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6432 features &= ~NETIF_F_TSO;
6433 features &= ~NETIF_F_TSO_ECN;
6436 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6437 !(features & NETIF_F_IPV6_CSUM)) {
6438 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6439 features &= ~NETIF_F_TSO6;
6442 /* TSO ECN requires that TSO is present as well. */
6443 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6444 features &= ~NETIF_F_TSO_ECN;
6446 /* Software GSO depends on SG. */
6447 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6448 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6449 features &= ~NETIF_F_GSO;
6452 /* UFO needs SG and checksumming */
6453 if (features & NETIF_F_UFO) {
6454 /* maybe split UFO into V4 and V6? */
6455 if (!((features & NETIF_F_GEN_CSUM) ||
6456 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6457 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6459 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6460 features &= ~NETIF_F_UFO;
6463 if (!(features & NETIF_F_SG)) {
6465 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6466 features &= ~NETIF_F_UFO;
6470 #ifdef CONFIG_NET_RX_BUSY_POLL
6471 if (dev->netdev_ops->ndo_busy_poll)
6472 features |= NETIF_F_BUSY_POLL;
6475 features &= ~NETIF_F_BUSY_POLL;
6480 int __netdev_update_features(struct net_device *dev)
6482 struct net_device *upper, *lower;
6483 netdev_features_t features;
6484 struct list_head *iter;
6489 features = netdev_get_wanted_features(dev);
6491 if (dev->netdev_ops->ndo_fix_features)
6492 features = dev->netdev_ops->ndo_fix_features(dev, features);
6494 /* driver might be less strict about feature dependencies */
6495 features = netdev_fix_features(dev, features);
6497 /* some features can't be enabled if they're off an an upper device */
6498 netdev_for_each_upper_dev_rcu(dev, upper, iter)
6499 features = netdev_sync_upper_features(dev, upper, features);
6501 if (dev->features == features)
6504 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6505 &dev->features, &features);
6507 if (dev->netdev_ops->ndo_set_features)
6508 err = dev->netdev_ops->ndo_set_features(dev, features);
6512 if (unlikely(err < 0)) {
6514 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6515 err, &features, &dev->features);
6516 /* return non-0 since some features might have changed and
6517 * it's better to fire a spurious notification than miss it
6523 /* some features must be disabled on lower devices when disabled
6524 * on an upper device (think: bonding master or bridge)
6526 netdev_for_each_lower_dev(dev, lower, iter)
6527 netdev_sync_lower_features(dev, lower, features);
6530 dev->features = features;
6532 return err < 0 ? 0 : 1;
6536 * netdev_update_features - recalculate device features
6537 * @dev: the device to check
6539 * Recalculate dev->features set and send notifications if it
6540 * has changed. Should be called after driver or hardware dependent
6541 * conditions might have changed that influence the features.
6543 void netdev_update_features(struct net_device *dev)
6545 if (__netdev_update_features(dev))
6546 netdev_features_change(dev);
6548 EXPORT_SYMBOL(netdev_update_features);
6551 * netdev_change_features - recalculate device features
6552 * @dev: the device to check
6554 * Recalculate dev->features set and send notifications even
6555 * if they have not changed. Should be called instead of
6556 * netdev_update_features() if also dev->vlan_features might
6557 * have changed to allow the changes to be propagated to stacked
6560 void netdev_change_features(struct net_device *dev)
6562 __netdev_update_features(dev);
6563 netdev_features_change(dev);
6565 EXPORT_SYMBOL(netdev_change_features);
6568 * netif_stacked_transfer_operstate - transfer operstate
6569 * @rootdev: the root or lower level device to transfer state from
6570 * @dev: the device to transfer operstate to
6572 * Transfer operational state from root to device. This is normally
6573 * called when a stacking relationship exists between the root
6574 * device and the device(a leaf device).
6576 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6577 struct net_device *dev)
6579 if (rootdev->operstate == IF_OPER_DORMANT)
6580 netif_dormant_on(dev);
6582 netif_dormant_off(dev);
6584 if (netif_carrier_ok(rootdev)) {
6585 if (!netif_carrier_ok(dev))
6586 netif_carrier_on(dev);
6588 if (netif_carrier_ok(dev))
6589 netif_carrier_off(dev);
6592 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6595 static int netif_alloc_rx_queues(struct net_device *dev)
6597 unsigned int i, count = dev->num_rx_queues;
6598 struct netdev_rx_queue *rx;
6599 size_t sz = count * sizeof(*rx);
6603 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6611 for (i = 0; i < count; i++)
6617 static void netdev_init_one_queue(struct net_device *dev,
6618 struct netdev_queue *queue, void *_unused)
6620 /* Initialize queue lock */
6621 spin_lock_init(&queue->_xmit_lock);
6622 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6623 queue->xmit_lock_owner = -1;
6624 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6627 dql_init(&queue->dql, HZ);
6631 static void netif_free_tx_queues(struct net_device *dev)
6636 static int netif_alloc_netdev_queues(struct net_device *dev)
6638 unsigned int count = dev->num_tx_queues;
6639 struct netdev_queue *tx;
6640 size_t sz = count * sizeof(*tx);
6642 if (count < 1 || count > 0xffff)
6645 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6653 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6654 spin_lock_init(&dev->tx_global_lock);
6659 void netif_tx_stop_all_queues(struct net_device *dev)
6663 for (i = 0; i < dev->num_tx_queues; i++) {
6664 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
6665 netif_tx_stop_queue(txq);
6668 EXPORT_SYMBOL(netif_tx_stop_all_queues);
6671 * register_netdevice - register a network device
6672 * @dev: device to register
6674 * Take a completed network device structure and add it to the kernel
6675 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6676 * chain. 0 is returned on success. A negative errno code is returned
6677 * on a failure to set up the device, or if the name is a duplicate.
6679 * Callers must hold the rtnl semaphore. You may want
6680 * register_netdev() instead of this.
6683 * The locking appears insufficient to guarantee two parallel registers
6684 * will not get the same name.
6687 int register_netdevice(struct net_device *dev)
6690 struct net *net = dev_net(dev);
6692 BUG_ON(dev_boot_phase);
6697 /* When net_device's are persistent, this will be fatal. */
6698 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6701 spin_lock_init(&dev->addr_list_lock);
6702 netdev_set_addr_lockdep_class(dev);
6704 ret = dev_get_valid_name(net, dev, dev->name);
6708 /* Init, if this function is available */
6709 if (dev->netdev_ops->ndo_init) {
6710 ret = dev->netdev_ops->ndo_init(dev);
6718 if (((dev->hw_features | dev->features) &
6719 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6720 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6721 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6722 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6729 dev->ifindex = dev_new_index(net);
6730 else if (__dev_get_by_index(net, dev->ifindex))
6733 /* Transfer changeable features to wanted_features and enable
6734 * software offloads (GSO and GRO).
6736 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6737 dev->features |= NETIF_F_SOFT_FEATURES;
6738 dev->wanted_features = dev->features & dev->hw_features;
6740 if (!(dev->flags & IFF_LOOPBACK)) {
6741 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6744 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6746 dev->vlan_features |= NETIF_F_HIGHDMA;
6748 /* Make NETIF_F_SG inheritable to tunnel devices.
6750 dev->hw_enc_features |= NETIF_F_SG;
6752 /* Make NETIF_F_SG inheritable to MPLS.
6754 dev->mpls_features |= NETIF_F_SG;
6756 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6757 ret = notifier_to_errno(ret);
6761 ret = netdev_register_kobject(dev);
6764 dev->reg_state = NETREG_REGISTERED;
6766 __netdev_update_features(dev);
6769 * Default initial state at registry is that the
6770 * device is present.
6773 set_bit(__LINK_STATE_PRESENT, &dev->state);
6775 linkwatch_init_dev(dev);
6777 dev_init_scheduler(dev);
6779 list_netdevice(dev);
6780 add_device_randomness(dev->dev_addr, dev->addr_len);
6782 /* If the device has permanent device address, driver should
6783 * set dev_addr and also addr_assign_type should be set to
6784 * NET_ADDR_PERM (default value).
6786 if (dev->addr_assign_type == NET_ADDR_PERM)
6787 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6789 /* Notify protocols, that a new device appeared. */
6790 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6791 ret = notifier_to_errno(ret);
6793 rollback_registered(dev);
6794 dev->reg_state = NETREG_UNREGISTERED;
6797 * Prevent userspace races by waiting until the network
6798 * device is fully setup before sending notifications.
6800 if (!dev->rtnl_link_ops ||
6801 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6802 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6808 if (dev->netdev_ops->ndo_uninit)
6809 dev->netdev_ops->ndo_uninit(dev);
6812 EXPORT_SYMBOL(register_netdevice);
6815 * init_dummy_netdev - init a dummy network device for NAPI
6816 * @dev: device to init
6818 * This takes a network device structure and initialize the minimum
6819 * amount of fields so it can be used to schedule NAPI polls without
6820 * registering a full blown interface. This is to be used by drivers
6821 * that need to tie several hardware interfaces to a single NAPI
6822 * poll scheduler due to HW limitations.
6824 int init_dummy_netdev(struct net_device *dev)
6826 /* Clear everything. Note we don't initialize spinlocks
6827 * are they aren't supposed to be taken by any of the
6828 * NAPI code and this dummy netdev is supposed to be
6829 * only ever used for NAPI polls
6831 memset(dev, 0, sizeof(struct net_device));
6833 /* make sure we BUG if trying to hit standard
6834 * register/unregister code path
6836 dev->reg_state = NETREG_DUMMY;
6838 /* NAPI wants this */
6839 INIT_LIST_HEAD(&dev->napi_list);
6841 /* a dummy interface is started by default */
6842 set_bit(__LINK_STATE_PRESENT, &dev->state);
6843 set_bit(__LINK_STATE_START, &dev->state);
6845 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6846 * because users of this 'device' dont need to change
6852 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6856 * register_netdev - register a network device
6857 * @dev: device to register
6859 * Take a completed network device structure and add it to the kernel
6860 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6861 * chain. 0 is returned on success. A negative errno code is returned
6862 * on a failure to set up the device, or if the name is a duplicate.
6864 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6865 * and expands the device name if you passed a format string to
6868 int register_netdev(struct net_device *dev)
6873 err = register_netdevice(dev);
6877 EXPORT_SYMBOL(register_netdev);
6879 int netdev_refcnt_read(const struct net_device *dev)
6883 for_each_possible_cpu(i)
6884 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6887 EXPORT_SYMBOL(netdev_refcnt_read);
6890 * netdev_wait_allrefs - wait until all references are gone.
6891 * @dev: target net_device
6893 * This is called when unregistering network devices.
6895 * Any protocol or device that holds a reference should register
6896 * for netdevice notification, and cleanup and put back the
6897 * reference if they receive an UNREGISTER event.
6898 * We can get stuck here if buggy protocols don't correctly
6901 static void netdev_wait_allrefs(struct net_device *dev)
6903 unsigned long rebroadcast_time, warning_time;
6906 linkwatch_forget_dev(dev);
6908 rebroadcast_time = warning_time = jiffies;
6909 refcnt = netdev_refcnt_read(dev);
6911 while (refcnt != 0) {
6912 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6915 /* Rebroadcast unregister notification */
6916 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6922 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6923 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6925 /* We must not have linkwatch events
6926 * pending on unregister. If this
6927 * happens, we simply run the queue
6928 * unscheduled, resulting in a noop
6931 linkwatch_run_queue();
6936 rebroadcast_time = jiffies;
6941 refcnt = netdev_refcnt_read(dev);
6943 if (time_after(jiffies, warning_time + 10 * HZ)) {
6944 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6946 warning_time = jiffies;
6955 * register_netdevice(x1);
6956 * register_netdevice(x2);
6958 * unregister_netdevice(y1);
6959 * unregister_netdevice(y2);
6965 * We are invoked by rtnl_unlock().
6966 * This allows us to deal with problems:
6967 * 1) We can delete sysfs objects which invoke hotplug
6968 * without deadlocking with linkwatch via keventd.
6969 * 2) Since we run with the RTNL semaphore not held, we can sleep
6970 * safely in order to wait for the netdev refcnt to drop to zero.
6972 * We must not return until all unregister events added during
6973 * the interval the lock was held have been completed.
6975 void netdev_run_todo(void)
6977 struct list_head list;
6979 /* Snapshot list, allow later requests */
6980 list_replace_init(&net_todo_list, &list);
6985 /* Wait for rcu callbacks to finish before next phase */
6986 if (!list_empty(&list))
6989 while (!list_empty(&list)) {
6990 struct net_device *dev
6991 = list_first_entry(&list, struct net_device, todo_list);
6992 list_del(&dev->todo_list);
6995 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6998 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6999 pr_err("network todo '%s' but state %d\n",
7000 dev->name, dev->reg_state);
7005 dev->reg_state = NETREG_UNREGISTERED;
7007 netdev_wait_allrefs(dev);
7010 BUG_ON(netdev_refcnt_read(dev));
7011 BUG_ON(!list_empty(&dev->ptype_all));
7012 BUG_ON(!list_empty(&dev->ptype_specific));
7013 WARN_ON(rcu_access_pointer(dev->ip_ptr));
7014 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
7015 WARN_ON(dev->dn_ptr);
7017 if (dev->destructor)
7018 dev->destructor(dev);
7020 /* Report a network device has been unregistered */
7022 dev_net(dev)->dev_unreg_count--;
7024 wake_up(&netdev_unregistering_wq);
7026 /* Free network device */
7027 kobject_put(&dev->dev.kobj);
7031 /* Convert net_device_stats to rtnl_link_stats64. They have the same
7032 * fields in the same order, with only the type differing.
7034 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7035 const struct net_device_stats *netdev_stats)
7037 #if BITS_PER_LONG == 64
7038 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
7039 memcpy(stats64, netdev_stats, sizeof(*stats64));
7041 size_t i, n = sizeof(*stats64) / sizeof(u64);
7042 const unsigned long *src = (const unsigned long *)netdev_stats;
7043 u64 *dst = (u64 *)stats64;
7045 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
7046 sizeof(*stats64) / sizeof(u64));
7047 for (i = 0; i < n; i++)
7051 EXPORT_SYMBOL(netdev_stats_to_stats64);
7054 * dev_get_stats - get network device statistics
7055 * @dev: device to get statistics from
7056 * @storage: place to store stats
7058 * Get network statistics from device. Return @storage.
7059 * The device driver may provide its own method by setting
7060 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7061 * otherwise the internal statistics structure is used.
7063 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
7064 struct rtnl_link_stats64 *storage)
7066 const struct net_device_ops *ops = dev->netdev_ops;
7068 if (ops->ndo_get_stats64) {
7069 memset(storage, 0, sizeof(*storage));
7070 ops->ndo_get_stats64(dev, storage);
7071 } else if (ops->ndo_get_stats) {
7072 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7074 netdev_stats_to_stats64(storage, &dev->stats);
7076 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
7077 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
7080 EXPORT_SYMBOL(dev_get_stats);
7082 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7084 struct netdev_queue *queue = dev_ingress_queue(dev);
7086 #ifdef CONFIG_NET_CLS_ACT
7089 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7092 netdev_init_one_queue(dev, queue, NULL);
7093 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7094 queue->qdisc_sleeping = &noop_qdisc;
7095 rcu_assign_pointer(dev->ingress_queue, queue);
7100 static const struct ethtool_ops default_ethtool_ops;
7102 void netdev_set_default_ethtool_ops(struct net_device *dev,
7103 const struct ethtool_ops *ops)
7105 if (dev->ethtool_ops == &default_ethtool_ops)
7106 dev->ethtool_ops = ops;
7108 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7110 void netdev_freemem(struct net_device *dev)
7112 char *addr = (char *)dev - dev->padded;
7118 * alloc_netdev_mqs - allocate network device
7119 * @sizeof_priv: size of private data to allocate space for
7120 * @name: device name format string
7121 * @name_assign_type: origin of device name
7122 * @setup: callback to initialize device
7123 * @txqs: the number of TX subqueues to allocate
7124 * @rxqs: the number of RX subqueues to allocate
7126 * Allocates a struct net_device with private data area for driver use
7127 * and performs basic initialization. Also allocates subqueue structs
7128 * for each queue on the device.
7130 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7131 unsigned char name_assign_type,
7132 void (*setup)(struct net_device *),
7133 unsigned int txqs, unsigned int rxqs)
7135 struct net_device *dev;
7137 struct net_device *p;
7139 BUG_ON(strlen(name) >= sizeof(dev->name));
7142 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7148 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7153 alloc_size = sizeof(struct net_device);
7155 /* ensure 32-byte alignment of private area */
7156 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7157 alloc_size += sizeof_priv;
7159 /* ensure 32-byte alignment of whole construct */
7160 alloc_size += NETDEV_ALIGN - 1;
7162 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7164 p = vzalloc(alloc_size);
7168 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7169 dev->padded = (char *)dev - (char *)p;
7171 dev->pcpu_refcnt = alloc_percpu(int);
7172 if (!dev->pcpu_refcnt)
7175 if (dev_addr_init(dev))
7181 dev_net_set(dev, &init_net);
7183 dev->gso_max_size = GSO_MAX_SIZE;
7184 dev->gso_max_segs = GSO_MAX_SEGS;
7185 dev->gso_min_segs = 0;
7187 INIT_LIST_HEAD(&dev->napi_list);
7188 INIT_LIST_HEAD(&dev->unreg_list);
7189 INIT_LIST_HEAD(&dev->close_list);
7190 INIT_LIST_HEAD(&dev->link_watch_list);
7191 INIT_LIST_HEAD(&dev->adj_list.upper);
7192 INIT_LIST_HEAD(&dev->adj_list.lower);
7193 INIT_LIST_HEAD(&dev->all_adj_list.upper);
7194 INIT_LIST_HEAD(&dev->all_adj_list.lower);
7195 INIT_LIST_HEAD(&dev->ptype_all);
7196 INIT_LIST_HEAD(&dev->ptype_specific);
7197 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7200 if (!dev->tx_queue_len) {
7201 dev->priv_flags |= IFF_NO_QUEUE;
7202 dev->tx_queue_len = 1;
7205 dev->num_tx_queues = txqs;
7206 dev->real_num_tx_queues = txqs;
7207 if (netif_alloc_netdev_queues(dev))
7211 dev->num_rx_queues = rxqs;
7212 dev->real_num_rx_queues = rxqs;
7213 if (netif_alloc_rx_queues(dev))
7217 strcpy(dev->name, name);
7218 dev->name_assign_type = name_assign_type;
7219 dev->group = INIT_NETDEV_GROUP;
7220 if (!dev->ethtool_ops)
7221 dev->ethtool_ops = &default_ethtool_ops;
7223 nf_hook_ingress_init(dev);
7232 free_percpu(dev->pcpu_refcnt);
7234 netdev_freemem(dev);
7237 EXPORT_SYMBOL(alloc_netdev_mqs);
7240 * free_netdev - free network device
7243 * This function does the last stage of destroying an allocated device
7244 * interface. The reference to the device object is released.
7245 * If this is the last reference then it will be freed.
7247 void free_netdev(struct net_device *dev)
7249 struct napi_struct *p, *n;
7251 netif_free_tx_queues(dev);
7256 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7258 /* Flush device addresses */
7259 dev_addr_flush(dev);
7261 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7264 free_percpu(dev->pcpu_refcnt);
7265 dev->pcpu_refcnt = NULL;
7267 /* Compatibility with error handling in drivers */
7268 if (dev->reg_state == NETREG_UNINITIALIZED) {
7269 netdev_freemem(dev);
7273 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7274 dev->reg_state = NETREG_RELEASED;
7276 /* will free via device release */
7277 put_device(&dev->dev);
7279 EXPORT_SYMBOL(free_netdev);
7282 * synchronize_net - Synchronize with packet receive processing
7284 * Wait for packets currently being received to be done.
7285 * Does not block later packets from starting.
7287 void synchronize_net(void)
7290 if (rtnl_is_locked())
7291 synchronize_rcu_expedited();
7295 EXPORT_SYMBOL(synchronize_net);
7298 * unregister_netdevice_queue - remove device from the kernel
7302 * This function shuts down a device interface and removes it
7303 * from the kernel tables.
7304 * If head not NULL, device is queued to be unregistered later.
7306 * Callers must hold the rtnl semaphore. You may want
7307 * unregister_netdev() instead of this.
7310 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7315 list_move_tail(&dev->unreg_list, head);
7317 rollback_registered(dev);
7318 /* Finish processing unregister after unlock */
7322 EXPORT_SYMBOL(unregister_netdevice_queue);
7325 * unregister_netdevice_many - unregister many devices
7326 * @head: list of devices
7328 * Note: As most callers use a stack allocated list_head,
7329 * we force a list_del() to make sure stack wont be corrupted later.
7331 void unregister_netdevice_many(struct list_head *head)
7333 struct net_device *dev;
7335 if (!list_empty(head)) {
7336 rollback_registered_many(head);
7337 list_for_each_entry(dev, head, unreg_list)
7342 EXPORT_SYMBOL(unregister_netdevice_many);
7345 * unregister_netdev - remove device from the kernel
7348 * This function shuts down a device interface and removes it
7349 * from the kernel tables.
7351 * This is just a wrapper for unregister_netdevice that takes
7352 * the rtnl semaphore. In general you want to use this and not
7353 * unregister_netdevice.
7355 void unregister_netdev(struct net_device *dev)
7358 unregister_netdevice(dev);
7361 EXPORT_SYMBOL(unregister_netdev);
7364 * dev_change_net_namespace - move device to different nethost namespace
7366 * @net: network namespace
7367 * @pat: If not NULL name pattern to try if the current device name
7368 * is already taken in the destination network namespace.
7370 * This function shuts down a device interface and moves it
7371 * to a new network namespace. On success 0 is returned, on
7372 * a failure a netagive errno code is returned.
7374 * Callers must hold the rtnl semaphore.
7377 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7383 /* Don't allow namespace local devices to be moved. */
7385 if (dev->features & NETIF_F_NETNS_LOCAL)
7388 /* Ensure the device has been registrered */
7389 if (dev->reg_state != NETREG_REGISTERED)
7392 /* Get out if there is nothing todo */
7394 if (net_eq(dev_net(dev), net))
7397 /* Pick the destination device name, and ensure
7398 * we can use it in the destination network namespace.
7401 if (__dev_get_by_name(net, dev->name)) {
7402 /* We get here if we can't use the current device name */
7405 if (dev_get_valid_name(net, dev, pat) < 0)
7410 * And now a mini version of register_netdevice unregister_netdevice.
7413 /* If device is running close it first. */
7416 /* And unlink it from device chain */
7418 unlist_netdevice(dev);
7422 /* Shutdown queueing discipline. */
7425 /* Notify protocols, that we are about to destroy
7426 this device. They should clean all the things.
7428 Note that dev->reg_state stays at NETREG_REGISTERED.
7429 This is wanted because this way 8021q and macvlan know
7430 the device is just moving and can keep their slaves up.
7432 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7434 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7435 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7438 * Flush the unicast and multicast chains
7443 /* Send a netdev-removed uevent to the old namespace */
7444 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7445 netdev_adjacent_del_links(dev);
7447 /* Actually switch the network namespace */
7448 dev_net_set(dev, net);
7450 /* If there is an ifindex conflict assign a new one */
7451 if (__dev_get_by_index(net, dev->ifindex))
7452 dev->ifindex = dev_new_index(net);
7454 /* Send a netdev-add uevent to the new namespace */
7455 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7456 netdev_adjacent_add_links(dev);
7458 /* Fixup kobjects */
7459 err = device_rename(&dev->dev, dev->name);
7462 /* Add the device back in the hashes */
7463 list_netdevice(dev);
7465 /* Notify protocols, that a new device appeared. */
7466 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7469 * Prevent userspace races by waiting until the network
7470 * device is fully setup before sending notifications.
7472 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7479 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7481 static int dev_cpu_callback(struct notifier_block *nfb,
7482 unsigned long action,
7485 struct sk_buff **list_skb;
7486 struct sk_buff *skb;
7487 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7488 struct softnet_data *sd, *oldsd;
7490 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7493 local_irq_disable();
7494 cpu = smp_processor_id();
7495 sd = &per_cpu(softnet_data, cpu);
7496 oldsd = &per_cpu(softnet_data, oldcpu);
7498 /* Find end of our completion_queue. */
7499 list_skb = &sd->completion_queue;
7501 list_skb = &(*list_skb)->next;
7502 /* Append completion queue from offline CPU. */
7503 *list_skb = oldsd->completion_queue;
7504 oldsd->completion_queue = NULL;
7506 /* Append output queue from offline CPU. */
7507 if (oldsd->output_queue) {
7508 *sd->output_queue_tailp = oldsd->output_queue;
7509 sd->output_queue_tailp = oldsd->output_queue_tailp;
7510 oldsd->output_queue = NULL;
7511 oldsd->output_queue_tailp = &oldsd->output_queue;
7513 /* Append NAPI poll list from offline CPU, with one exception :
7514 * process_backlog() must be called by cpu owning percpu backlog.
7515 * We properly handle process_queue & input_pkt_queue later.
7517 while (!list_empty(&oldsd->poll_list)) {
7518 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7522 list_del_init(&napi->poll_list);
7523 if (napi->poll == process_backlog)
7526 ____napi_schedule(sd, napi);
7529 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7532 /* Process offline CPU's input_pkt_queue */
7533 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7535 input_queue_head_incr(oldsd);
7537 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7539 input_queue_head_incr(oldsd);
7547 * netdev_increment_features - increment feature set by one
7548 * @all: current feature set
7549 * @one: new feature set
7550 * @mask: mask feature set
7552 * Computes a new feature set after adding a device with feature set
7553 * @one to the master device with current feature set @all. Will not
7554 * enable anything that is off in @mask. Returns the new feature set.
7556 netdev_features_t netdev_increment_features(netdev_features_t all,
7557 netdev_features_t one, netdev_features_t mask)
7559 if (mask & NETIF_F_GEN_CSUM)
7560 mask |= NETIF_F_ALL_CSUM;
7561 mask |= NETIF_F_VLAN_CHALLENGED;
7563 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7564 all &= one | ~NETIF_F_ALL_FOR_ALL;
7566 /* If one device supports hw checksumming, set for all. */
7567 if (all & NETIF_F_GEN_CSUM)
7568 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7572 EXPORT_SYMBOL(netdev_increment_features);
7574 static struct hlist_head * __net_init netdev_create_hash(void)
7577 struct hlist_head *hash;
7579 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7581 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7582 INIT_HLIST_HEAD(&hash[i]);
7587 /* Initialize per network namespace state */
7588 static int __net_init netdev_init(struct net *net)
7590 if (net != &init_net)
7591 INIT_LIST_HEAD(&net->dev_base_head);
7593 net->dev_name_head = netdev_create_hash();
7594 if (net->dev_name_head == NULL)
7597 net->dev_index_head = netdev_create_hash();
7598 if (net->dev_index_head == NULL)
7604 kfree(net->dev_name_head);
7610 * netdev_drivername - network driver for the device
7611 * @dev: network device
7613 * Determine network driver for device.
7615 const char *netdev_drivername(const struct net_device *dev)
7617 const struct device_driver *driver;
7618 const struct device *parent;
7619 const char *empty = "";
7621 parent = dev->dev.parent;
7625 driver = parent->driver;
7626 if (driver && driver->name)
7627 return driver->name;
7631 static void __netdev_printk(const char *level, const struct net_device *dev,
7632 struct va_format *vaf)
7634 if (dev && dev->dev.parent) {
7635 dev_printk_emit(level[1] - '0',
7638 dev_driver_string(dev->dev.parent),
7639 dev_name(dev->dev.parent),
7640 netdev_name(dev), netdev_reg_state(dev),
7643 printk("%s%s%s: %pV",
7644 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7646 printk("%s(NULL net_device): %pV", level, vaf);
7650 void netdev_printk(const char *level, const struct net_device *dev,
7651 const char *format, ...)
7653 struct va_format vaf;
7656 va_start(args, format);
7661 __netdev_printk(level, dev, &vaf);
7665 EXPORT_SYMBOL(netdev_printk);
7667 #define define_netdev_printk_level(func, level) \
7668 void func(const struct net_device *dev, const char *fmt, ...) \
7670 struct va_format vaf; \
7673 va_start(args, fmt); \
7678 __netdev_printk(level, dev, &vaf); \
7682 EXPORT_SYMBOL(func);
7684 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7685 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7686 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7687 define_netdev_printk_level(netdev_err, KERN_ERR);
7688 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7689 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7690 define_netdev_printk_level(netdev_info, KERN_INFO);
7692 static void __net_exit netdev_exit(struct net *net)
7694 kfree(net->dev_name_head);
7695 kfree(net->dev_index_head);
7698 static struct pernet_operations __net_initdata netdev_net_ops = {
7699 .init = netdev_init,
7700 .exit = netdev_exit,
7703 static void __net_exit default_device_exit(struct net *net)
7705 struct net_device *dev, *aux;
7707 * Push all migratable network devices back to the
7708 * initial network namespace
7711 for_each_netdev_safe(net, dev, aux) {
7713 char fb_name[IFNAMSIZ];
7715 /* Ignore unmoveable devices (i.e. loopback) */
7716 if (dev->features & NETIF_F_NETNS_LOCAL)
7719 /* Leave virtual devices for the generic cleanup */
7720 if (dev->rtnl_link_ops)
7723 /* Push remaining network devices to init_net */
7724 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7725 err = dev_change_net_namespace(dev, &init_net, fb_name);
7727 pr_emerg("%s: failed to move %s to init_net: %d\n",
7728 __func__, dev->name, err);
7735 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7737 /* Return with the rtnl_lock held when there are no network
7738 * devices unregistering in any network namespace in net_list.
7742 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7744 add_wait_queue(&netdev_unregistering_wq, &wait);
7746 unregistering = false;
7748 list_for_each_entry(net, net_list, exit_list) {
7749 if (net->dev_unreg_count > 0) {
7750 unregistering = true;
7758 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7760 remove_wait_queue(&netdev_unregistering_wq, &wait);
7763 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7765 /* At exit all network devices most be removed from a network
7766 * namespace. Do this in the reverse order of registration.
7767 * Do this across as many network namespaces as possible to
7768 * improve batching efficiency.
7770 struct net_device *dev;
7772 LIST_HEAD(dev_kill_list);
7774 /* To prevent network device cleanup code from dereferencing
7775 * loopback devices or network devices that have been freed
7776 * wait here for all pending unregistrations to complete,
7777 * before unregistring the loopback device and allowing the
7778 * network namespace be freed.
7780 * The netdev todo list containing all network devices
7781 * unregistrations that happen in default_device_exit_batch
7782 * will run in the rtnl_unlock() at the end of
7783 * default_device_exit_batch.
7785 rtnl_lock_unregistering(net_list);
7786 list_for_each_entry(net, net_list, exit_list) {
7787 for_each_netdev_reverse(net, dev) {
7788 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7789 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7791 unregister_netdevice_queue(dev, &dev_kill_list);
7794 unregister_netdevice_many(&dev_kill_list);
7798 static struct pernet_operations __net_initdata default_device_ops = {
7799 .exit = default_device_exit,
7800 .exit_batch = default_device_exit_batch,
7804 * Initialize the DEV module. At boot time this walks the device list and
7805 * unhooks any devices that fail to initialise (normally hardware not
7806 * present) and leaves us with a valid list of present and active devices.
7811 * This is called single threaded during boot, so no need
7812 * to take the rtnl semaphore.
7814 static int __init net_dev_init(void)
7816 int i, rc = -ENOMEM;
7818 BUG_ON(!dev_boot_phase);
7820 if (dev_proc_init())
7823 if (netdev_kobject_init())
7826 INIT_LIST_HEAD(&ptype_all);
7827 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7828 INIT_LIST_HEAD(&ptype_base[i]);
7830 INIT_LIST_HEAD(&offload_base);
7832 if (register_pernet_subsys(&netdev_net_ops))
7836 * Initialise the packet receive queues.
7839 for_each_possible_cpu(i) {
7840 struct softnet_data *sd = &per_cpu(softnet_data, i);
7842 skb_queue_head_init(&sd->input_pkt_queue);
7843 skb_queue_head_init(&sd->process_queue);
7844 INIT_LIST_HEAD(&sd->poll_list);
7845 sd->output_queue_tailp = &sd->output_queue;
7847 sd->csd.func = rps_trigger_softirq;
7852 sd->backlog.poll = process_backlog;
7853 sd->backlog.weight = weight_p;
7858 /* The loopback device is special if any other network devices
7859 * is present in a network namespace the loopback device must
7860 * be present. Since we now dynamically allocate and free the
7861 * loopback device ensure this invariant is maintained by
7862 * keeping the loopback device as the first device on the
7863 * list of network devices. Ensuring the loopback devices
7864 * is the first device that appears and the last network device
7867 if (register_pernet_device(&loopback_net_ops))
7870 if (register_pernet_device(&default_device_ops))
7873 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7874 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7876 hotcpu_notifier(dev_cpu_callback, 0);
7883 subsys_initcall(net_dev_init);