2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <linux/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <linux/bpf.h>
98 #include <net/net_namespace.h>
100 #include <net/busy_poll.h>
101 #include <linux/rtnetlink.h>
102 #include <linux/stat.h>
104 #include <net/dst_metadata.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/iw_handler.h>
115 #include <asm/current.h>
116 #include <linux/audit.h>
117 #include <linux/dmaengine.h>
118 #include <linux/err.h>
119 #include <linux/ctype.h>
120 #include <linux/if_arp.h>
121 #include <linux/if_vlan.h>
122 #include <linux/ip.h>
124 #include <net/mpls.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/static_key.h>
136 #include <linux/hashtable.h>
137 #include <linux/vmalloc.h>
138 #include <linux/if_macvlan.h>
139 #include <linux/errqueue.h>
140 #include <linux/hrtimer.h>
141 #include <linux/netfilter_ingress.h>
142 #include <linux/crash_dump.h>
144 #include "net-sysfs.h"
146 /* Instead of increasing this, you should create a hash table. */
147 #define MAX_GRO_SKBS 8
149 /* This should be increased if a protocol with a bigger head is added. */
150 #define GRO_MAX_HEAD (MAX_HEADER + 128)
152 static DEFINE_SPINLOCK(ptype_lock);
153 static DEFINE_SPINLOCK(offload_lock);
154 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
155 struct list_head ptype_all __read_mostly; /* Taps */
156 static struct list_head offload_base __read_mostly;
158 static int netif_rx_internal(struct sk_buff *skb);
159 static int call_netdevice_notifiers_info(unsigned long val,
160 struct net_device *dev,
161 struct netdev_notifier_info *info);
164 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
167 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
169 * Writers must hold the rtnl semaphore while they loop through the
170 * dev_base_head list, and hold dev_base_lock for writing when they do the
171 * actual updates. This allows pure readers to access the list even
172 * while a writer is preparing to update it.
174 * To put it another way, dev_base_lock is held for writing only to
175 * protect against pure readers; the rtnl semaphore provides the
176 * protection against other writers.
178 * See, for example usages, register_netdevice() and
179 * unregister_netdevice(), which must be called with the rtnl
182 DEFINE_RWLOCK(dev_base_lock);
183 EXPORT_SYMBOL(dev_base_lock);
185 /* protects napi_hash addition/deletion and napi_gen_id */
186 static DEFINE_SPINLOCK(napi_hash_lock);
188 static unsigned int napi_gen_id = NR_CPUS;
189 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
191 static seqcount_t devnet_rename_seq;
193 static inline void dev_base_seq_inc(struct net *net)
195 while (++net->dev_base_seq == 0)
199 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
201 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
203 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
206 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
208 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
211 static inline void rps_lock(struct softnet_data *sd)
214 spin_lock(&sd->input_pkt_queue.lock);
218 static inline void rps_unlock(struct softnet_data *sd)
221 spin_unlock(&sd->input_pkt_queue.lock);
225 /* Device list insertion */
226 static void list_netdevice(struct net_device *dev)
228 struct net *net = dev_net(dev);
232 write_lock_bh(&dev_base_lock);
233 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
234 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
235 hlist_add_head_rcu(&dev->index_hlist,
236 dev_index_hash(net, dev->ifindex));
237 write_unlock_bh(&dev_base_lock);
239 dev_base_seq_inc(net);
242 /* Device list removal
243 * caller must respect a RCU grace period before freeing/reusing dev
245 static void unlist_netdevice(struct net_device *dev)
249 /* Unlink dev from the device chain */
250 write_lock_bh(&dev_base_lock);
251 list_del_rcu(&dev->dev_list);
252 hlist_del_rcu(&dev->name_hlist);
253 hlist_del_rcu(&dev->index_hlist);
254 write_unlock_bh(&dev_base_lock);
256 dev_base_seq_inc(dev_net(dev));
263 static RAW_NOTIFIER_HEAD(netdev_chain);
266 * Device drivers call our routines to queue packets here. We empty the
267 * queue in the local softnet handler.
270 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
271 EXPORT_PER_CPU_SYMBOL(softnet_data);
273 #ifdef CONFIG_LOCKDEP
275 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
276 * according to dev->type
278 static const unsigned short netdev_lock_type[] = {
279 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
280 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
281 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
282 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
283 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
284 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
285 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
286 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
287 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
288 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
289 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
290 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
291 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
292 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
293 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
295 static const char *const netdev_lock_name[] = {
296 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
297 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
298 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
299 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
300 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
301 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
302 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
303 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
304 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
305 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
306 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
307 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
308 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
309 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
310 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
312 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
313 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
315 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
319 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
320 if (netdev_lock_type[i] == dev_type)
322 /* the last key is used by default */
323 return ARRAY_SIZE(netdev_lock_type) - 1;
326 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
327 unsigned short dev_type)
331 i = netdev_lock_pos(dev_type);
332 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
333 netdev_lock_name[i]);
336 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
340 i = netdev_lock_pos(dev->type);
341 lockdep_set_class_and_name(&dev->addr_list_lock,
342 &netdev_addr_lock_key[i],
343 netdev_lock_name[i]);
346 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
347 unsigned short dev_type)
350 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
355 /*******************************************************************************
357 * Protocol management and registration routines
359 *******************************************************************************/
363 * Add a protocol ID to the list. Now that the input handler is
364 * smarter we can dispense with all the messy stuff that used to be
367 * BEWARE!!! Protocol handlers, mangling input packets,
368 * MUST BE last in hash buckets and checking protocol handlers
369 * MUST start from promiscuous ptype_all chain in net_bh.
370 * It is true now, do not change it.
371 * Explanation follows: if protocol handler, mangling packet, will
372 * be the first on list, it is not able to sense, that packet
373 * is cloned and should be copied-on-write, so that it will
374 * change it and subsequent readers will get broken packet.
378 static inline struct list_head *ptype_head(const struct packet_type *pt)
380 if (pt->type == htons(ETH_P_ALL))
381 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
383 return pt->dev ? &pt->dev->ptype_specific :
384 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
388 * dev_add_pack - add packet handler
389 * @pt: packet type declaration
391 * Add a protocol handler to the networking stack. The passed &packet_type
392 * is linked into kernel lists and may not be freed until it has been
393 * removed from the kernel lists.
395 * This call does not sleep therefore it can not
396 * guarantee all CPU's that are in middle of receiving packets
397 * will see the new packet type (until the next received packet).
400 void dev_add_pack(struct packet_type *pt)
402 struct list_head *head = ptype_head(pt);
404 spin_lock(&ptype_lock);
405 list_add_rcu(&pt->list, head);
406 spin_unlock(&ptype_lock);
408 EXPORT_SYMBOL(dev_add_pack);
411 * __dev_remove_pack - remove packet handler
412 * @pt: packet type declaration
414 * Remove a protocol handler that was previously added to the kernel
415 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
416 * from the kernel lists and can be freed or reused once this function
419 * The packet type might still be in use by receivers
420 * and must not be freed until after all the CPU's have gone
421 * through a quiescent state.
423 void __dev_remove_pack(struct packet_type *pt)
425 struct list_head *head = ptype_head(pt);
426 struct packet_type *pt1;
428 spin_lock(&ptype_lock);
430 list_for_each_entry(pt1, head, list) {
432 list_del_rcu(&pt->list);
437 pr_warn("dev_remove_pack: %p not found\n", pt);
439 spin_unlock(&ptype_lock);
441 EXPORT_SYMBOL(__dev_remove_pack);
444 * dev_remove_pack - remove packet handler
445 * @pt: packet type declaration
447 * Remove a protocol handler that was previously added to the kernel
448 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
449 * from the kernel lists and can be freed or reused once this function
452 * This call sleeps to guarantee that no CPU is looking at the packet
455 void dev_remove_pack(struct packet_type *pt)
457 __dev_remove_pack(pt);
461 EXPORT_SYMBOL(dev_remove_pack);
465 * dev_add_offload - register offload handlers
466 * @po: protocol offload declaration
468 * Add protocol offload handlers to the networking stack. The passed
469 * &proto_offload is linked into kernel lists and may not be freed until
470 * it has been removed from the kernel lists.
472 * This call does not sleep therefore it can not
473 * guarantee all CPU's that are in middle of receiving packets
474 * will see the new offload handlers (until the next received packet).
476 void dev_add_offload(struct packet_offload *po)
478 struct packet_offload *elem;
480 spin_lock(&offload_lock);
481 list_for_each_entry(elem, &offload_base, list) {
482 if (po->priority < elem->priority)
485 list_add_rcu(&po->list, elem->list.prev);
486 spin_unlock(&offload_lock);
488 EXPORT_SYMBOL(dev_add_offload);
491 * __dev_remove_offload - remove offload handler
492 * @po: packet offload declaration
494 * Remove a protocol offload handler that was previously added to the
495 * kernel offload handlers by dev_add_offload(). The passed &offload_type
496 * is removed from the kernel lists and can be freed or reused once this
499 * The packet type might still be in use by receivers
500 * and must not be freed until after all the CPU's have gone
501 * through a quiescent state.
503 static void __dev_remove_offload(struct packet_offload *po)
505 struct list_head *head = &offload_base;
506 struct packet_offload *po1;
508 spin_lock(&offload_lock);
510 list_for_each_entry(po1, head, list) {
512 list_del_rcu(&po->list);
517 pr_warn("dev_remove_offload: %p not found\n", po);
519 spin_unlock(&offload_lock);
523 * dev_remove_offload - remove packet offload handler
524 * @po: packet offload declaration
526 * Remove a packet offload handler that was previously added to the kernel
527 * offload handlers by dev_add_offload(). The passed &offload_type is
528 * removed from the kernel lists and can be freed or reused once this
531 * This call sleeps to guarantee that no CPU is looking at the packet
534 void dev_remove_offload(struct packet_offload *po)
536 __dev_remove_offload(po);
540 EXPORT_SYMBOL(dev_remove_offload);
542 /******************************************************************************
544 * Device Boot-time Settings Routines
546 ******************************************************************************/
548 /* Boot time configuration table */
549 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
552 * netdev_boot_setup_add - add new setup entry
553 * @name: name of the device
554 * @map: configured settings for the device
556 * Adds new setup entry to the dev_boot_setup list. The function
557 * returns 0 on error and 1 on success. This is a generic routine to
560 static int netdev_boot_setup_add(char *name, struct ifmap *map)
562 struct netdev_boot_setup *s;
566 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
567 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
568 memset(s[i].name, 0, sizeof(s[i].name));
569 strlcpy(s[i].name, name, IFNAMSIZ);
570 memcpy(&s[i].map, map, sizeof(s[i].map));
575 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
579 * netdev_boot_setup_check - check boot time settings
580 * @dev: the netdevice
582 * Check boot time settings for the device.
583 * The found settings are set for the device to be used
584 * later in the device probing.
585 * Returns 0 if no settings found, 1 if they are.
587 int netdev_boot_setup_check(struct net_device *dev)
589 struct netdev_boot_setup *s = dev_boot_setup;
592 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
593 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
594 !strcmp(dev->name, s[i].name)) {
595 dev->irq = s[i].map.irq;
596 dev->base_addr = s[i].map.base_addr;
597 dev->mem_start = s[i].map.mem_start;
598 dev->mem_end = s[i].map.mem_end;
604 EXPORT_SYMBOL(netdev_boot_setup_check);
608 * netdev_boot_base - get address from boot time settings
609 * @prefix: prefix for network device
610 * @unit: id for network device
612 * Check boot time settings for the base address of device.
613 * The found settings are set for the device to be used
614 * later in the device probing.
615 * Returns 0 if no settings found.
617 unsigned long netdev_boot_base(const char *prefix, int unit)
619 const struct netdev_boot_setup *s = dev_boot_setup;
623 sprintf(name, "%s%d", prefix, unit);
626 * If device already registered then return base of 1
627 * to indicate not to probe for this interface
629 if (__dev_get_by_name(&init_net, name))
632 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
633 if (!strcmp(name, s[i].name))
634 return s[i].map.base_addr;
639 * Saves at boot time configured settings for any netdevice.
641 int __init netdev_boot_setup(char *str)
646 str = get_options(str, ARRAY_SIZE(ints), ints);
651 memset(&map, 0, sizeof(map));
655 map.base_addr = ints[2];
657 map.mem_start = ints[3];
659 map.mem_end = ints[4];
661 /* Add new entry to the list */
662 return netdev_boot_setup_add(str, &map);
665 __setup("netdev=", netdev_boot_setup);
667 /*******************************************************************************
669 * Device Interface Subroutines
671 *******************************************************************************/
674 * dev_get_iflink - get 'iflink' value of a interface
675 * @dev: targeted interface
677 * Indicates the ifindex the interface is linked to.
678 * Physical interfaces have the same 'ifindex' and 'iflink' values.
681 int dev_get_iflink(const struct net_device *dev)
683 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
684 return dev->netdev_ops->ndo_get_iflink(dev);
688 EXPORT_SYMBOL(dev_get_iflink);
691 * dev_fill_metadata_dst - Retrieve tunnel egress information.
692 * @dev: targeted interface
695 * For better visibility of tunnel traffic OVS needs to retrieve
696 * egress tunnel information for a packet. Following API allows
697 * user to get this info.
699 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
701 struct ip_tunnel_info *info;
703 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
706 info = skb_tunnel_info_unclone(skb);
709 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
712 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
714 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
717 * __dev_get_by_name - find a device by its name
718 * @net: the applicable net namespace
719 * @name: name to find
721 * Find an interface by name. Must be called under RTNL semaphore
722 * or @dev_base_lock. If the name is found a pointer to the device
723 * is returned. If the name is not found then %NULL is returned. The
724 * reference counters are not incremented so the caller must be
725 * careful with locks.
728 struct net_device *__dev_get_by_name(struct net *net, const char *name)
730 struct net_device *dev;
731 struct hlist_head *head = dev_name_hash(net, name);
733 hlist_for_each_entry(dev, head, name_hlist)
734 if (!strncmp(dev->name, name, IFNAMSIZ))
739 EXPORT_SYMBOL(__dev_get_by_name);
742 * dev_get_by_name_rcu - find a device by its name
743 * @net: the applicable net namespace
744 * @name: name to find
746 * Find an interface by name.
747 * If the name is found a pointer to the device is returned.
748 * If the name is not found then %NULL is returned.
749 * The reference counters are not incremented so the caller must be
750 * careful with locks. The caller must hold RCU lock.
753 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
755 struct net_device *dev;
756 struct hlist_head *head = dev_name_hash(net, name);
758 hlist_for_each_entry_rcu(dev, head, name_hlist)
759 if (!strncmp(dev->name, name, IFNAMSIZ))
764 EXPORT_SYMBOL(dev_get_by_name_rcu);
767 * dev_get_by_name - find a device by its name
768 * @net: the applicable net namespace
769 * @name: name to find
771 * Find an interface by name. This can be called from any
772 * context and does its own locking. The returned handle has
773 * the usage count incremented and the caller must use dev_put() to
774 * release it when it is no longer needed. %NULL is returned if no
775 * matching device is found.
778 struct net_device *dev_get_by_name(struct net *net, const char *name)
780 struct net_device *dev;
783 dev = dev_get_by_name_rcu(net, name);
789 EXPORT_SYMBOL(dev_get_by_name);
792 * __dev_get_by_index - find a device by its ifindex
793 * @net: the applicable net namespace
794 * @ifindex: index of device
796 * Search for an interface by index. Returns %NULL if the device
797 * is not found or a pointer to the device. The device has not
798 * had its reference counter increased so the caller must be careful
799 * about locking. The caller must hold either the RTNL semaphore
803 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
805 struct net_device *dev;
806 struct hlist_head *head = dev_index_hash(net, ifindex);
808 hlist_for_each_entry(dev, head, index_hlist)
809 if (dev->ifindex == ifindex)
814 EXPORT_SYMBOL(__dev_get_by_index);
817 * dev_get_by_index_rcu - find a device by its ifindex
818 * @net: the applicable net namespace
819 * @ifindex: index of device
821 * Search for an interface by index. Returns %NULL if the device
822 * is not found or a pointer to the device. The device has not
823 * had its reference counter increased so the caller must be careful
824 * about locking. The caller must hold RCU lock.
827 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
829 struct net_device *dev;
830 struct hlist_head *head = dev_index_hash(net, ifindex);
832 hlist_for_each_entry_rcu(dev, head, index_hlist)
833 if (dev->ifindex == ifindex)
838 EXPORT_SYMBOL(dev_get_by_index_rcu);
842 * dev_get_by_index - find a device by its ifindex
843 * @net: the applicable net namespace
844 * @ifindex: index of device
846 * Search for an interface by index. Returns NULL if the device
847 * is not found or a pointer to the device. The device returned has
848 * had a reference added and the pointer is safe until the user calls
849 * dev_put to indicate they have finished with it.
852 struct net_device *dev_get_by_index(struct net *net, int ifindex)
854 struct net_device *dev;
857 dev = dev_get_by_index_rcu(net, ifindex);
863 EXPORT_SYMBOL(dev_get_by_index);
866 * netdev_get_name - get a netdevice name, knowing its ifindex.
867 * @net: network namespace
868 * @name: a pointer to the buffer where the name will be stored.
869 * @ifindex: the ifindex of the interface to get the name from.
871 * The use of raw_seqcount_begin() and cond_resched() before
872 * retrying is required as we want to give the writers a chance
873 * to complete when CONFIG_PREEMPT is not set.
875 int netdev_get_name(struct net *net, char *name, int ifindex)
877 struct net_device *dev;
881 seq = raw_seqcount_begin(&devnet_rename_seq);
883 dev = dev_get_by_index_rcu(net, ifindex);
889 strcpy(name, dev->name);
891 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
900 * dev_getbyhwaddr_rcu - find a device by its hardware address
901 * @net: the applicable net namespace
902 * @type: media type of device
903 * @ha: hardware address
905 * Search for an interface by MAC address. Returns NULL if the device
906 * is not found or a pointer to the device.
907 * The caller must hold RCU or RTNL.
908 * The returned device has not had its ref count increased
909 * and the caller must therefore be careful about locking
913 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
916 struct net_device *dev;
918 for_each_netdev_rcu(net, dev)
919 if (dev->type == type &&
920 !memcmp(dev->dev_addr, ha, dev->addr_len))
925 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
927 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
929 struct net_device *dev;
932 for_each_netdev(net, dev)
933 if (dev->type == type)
938 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
940 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
942 struct net_device *dev, *ret = NULL;
945 for_each_netdev_rcu(net, dev)
946 if (dev->type == type) {
954 EXPORT_SYMBOL(dev_getfirstbyhwtype);
957 * __dev_get_by_flags - find any device with given flags
958 * @net: the applicable net namespace
959 * @if_flags: IFF_* values
960 * @mask: bitmask of bits in if_flags to check
962 * Search for any interface with the given flags. Returns NULL if a device
963 * is not found or a pointer to the device. Must be called inside
964 * rtnl_lock(), and result refcount is unchanged.
967 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
970 struct net_device *dev, *ret;
975 for_each_netdev(net, dev) {
976 if (((dev->flags ^ if_flags) & mask) == 0) {
983 EXPORT_SYMBOL(__dev_get_by_flags);
986 * dev_valid_name - check if name is okay for network device
989 * Network device names need to be valid file names to
990 * to allow sysfs to work. We also disallow any kind of
993 bool dev_valid_name(const char *name)
997 if (strlen(name) >= IFNAMSIZ)
999 if (!strcmp(name, ".") || !strcmp(name, ".."))
1003 if (*name == '/' || *name == ':' || isspace(*name))
1009 EXPORT_SYMBOL(dev_valid_name);
1012 * __dev_alloc_name - allocate a name for a device
1013 * @net: network namespace to allocate the device name in
1014 * @name: name format string
1015 * @buf: scratch buffer and result name string
1017 * Passed a format string - eg "lt%d" it will try and find a suitable
1018 * id. It scans list of devices to build up a free map, then chooses
1019 * the first empty slot. The caller must hold the dev_base or rtnl lock
1020 * while allocating the name and adding the device in order to avoid
1022 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1023 * Returns the number of the unit assigned or a negative errno code.
1026 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1030 const int max_netdevices = 8*PAGE_SIZE;
1031 unsigned long *inuse;
1032 struct net_device *d;
1034 p = strnchr(name, IFNAMSIZ-1, '%');
1037 * Verify the string as this thing may have come from
1038 * the user. There must be either one "%d" and no other "%"
1041 if (p[1] != 'd' || strchr(p + 2, '%'))
1044 /* Use one page as a bit array of possible slots */
1045 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1049 for_each_netdev(net, d) {
1050 if (!sscanf(d->name, name, &i))
1052 if (i < 0 || i >= max_netdevices)
1055 /* avoid cases where sscanf is not exact inverse of printf */
1056 snprintf(buf, IFNAMSIZ, name, i);
1057 if (!strncmp(buf, d->name, IFNAMSIZ))
1061 i = find_first_zero_bit(inuse, max_netdevices);
1062 free_page((unsigned long) inuse);
1066 snprintf(buf, IFNAMSIZ, name, i);
1067 if (!__dev_get_by_name(net, buf))
1070 /* It is possible to run out of possible slots
1071 * when the name is long and there isn't enough space left
1072 * for the digits, or if all bits are used.
1078 * dev_alloc_name - allocate a name for a device
1080 * @name: name format string
1082 * Passed a format string - eg "lt%d" it will try and find a suitable
1083 * id. It scans list of devices to build up a free map, then chooses
1084 * the first empty slot. The caller must hold the dev_base or rtnl lock
1085 * while allocating the name and adding the device in order to avoid
1087 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1088 * Returns the number of the unit assigned or a negative errno code.
1091 int dev_alloc_name(struct net_device *dev, const char *name)
1097 BUG_ON(!dev_net(dev));
1099 ret = __dev_alloc_name(net, name, buf);
1101 strlcpy(dev->name, buf, IFNAMSIZ);
1104 EXPORT_SYMBOL(dev_alloc_name);
1106 static int dev_alloc_name_ns(struct net *net,
1107 struct net_device *dev,
1113 ret = __dev_alloc_name(net, name, buf);
1115 strlcpy(dev->name, buf, IFNAMSIZ);
1119 static int dev_get_valid_name(struct net *net,
1120 struct net_device *dev,
1125 if (!dev_valid_name(name))
1128 if (strchr(name, '%'))
1129 return dev_alloc_name_ns(net, dev, name);
1130 else if (__dev_get_by_name(net, name))
1132 else if (dev->name != name)
1133 strlcpy(dev->name, name, IFNAMSIZ);
1139 * dev_change_name - change name of a device
1141 * @newname: name (or format string) must be at least IFNAMSIZ
1143 * Change name of a device, can pass format strings "eth%d".
1146 int dev_change_name(struct net_device *dev, const char *newname)
1148 unsigned char old_assign_type;
1149 char oldname[IFNAMSIZ];
1155 BUG_ON(!dev_net(dev));
1158 if (dev->flags & IFF_UP)
1161 write_seqcount_begin(&devnet_rename_seq);
1163 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1164 write_seqcount_end(&devnet_rename_seq);
1168 memcpy(oldname, dev->name, IFNAMSIZ);
1170 err = dev_get_valid_name(net, dev, newname);
1172 write_seqcount_end(&devnet_rename_seq);
1176 if (oldname[0] && !strchr(oldname, '%'))
1177 netdev_info(dev, "renamed from %s\n", oldname);
1179 old_assign_type = dev->name_assign_type;
1180 dev->name_assign_type = NET_NAME_RENAMED;
1183 ret = device_rename(&dev->dev, dev->name);
1185 memcpy(dev->name, oldname, IFNAMSIZ);
1186 dev->name_assign_type = old_assign_type;
1187 write_seqcount_end(&devnet_rename_seq);
1191 write_seqcount_end(&devnet_rename_seq);
1193 netdev_adjacent_rename_links(dev, oldname);
1195 write_lock_bh(&dev_base_lock);
1196 hlist_del_rcu(&dev->name_hlist);
1197 write_unlock_bh(&dev_base_lock);
1201 write_lock_bh(&dev_base_lock);
1202 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1203 write_unlock_bh(&dev_base_lock);
1205 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1206 ret = notifier_to_errno(ret);
1209 /* err >= 0 after dev_alloc_name() or stores the first errno */
1212 write_seqcount_begin(&devnet_rename_seq);
1213 memcpy(dev->name, oldname, IFNAMSIZ);
1214 memcpy(oldname, newname, IFNAMSIZ);
1215 dev->name_assign_type = old_assign_type;
1216 old_assign_type = NET_NAME_RENAMED;
1219 pr_err("%s: name change rollback failed: %d\n",
1228 * dev_set_alias - change ifalias of a device
1230 * @alias: name up to IFALIASZ
1231 * @len: limit of bytes to copy from info
1233 * Set ifalias for a device,
1235 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1241 if (len >= IFALIASZ)
1245 kfree(dev->ifalias);
1246 dev->ifalias = NULL;
1250 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1253 dev->ifalias = new_ifalias;
1255 strlcpy(dev->ifalias, alias, len+1);
1261 * netdev_features_change - device changes features
1262 * @dev: device to cause notification
1264 * Called to indicate a device has changed features.
1266 void netdev_features_change(struct net_device *dev)
1268 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1270 EXPORT_SYMBOL(netdev_features_change);
1273 * netdev_state_change - device changes state
1274 * @dev: device to cause notification
1276 * Called to indicate a device has changed state. This function calls
1277 * the notifier chains for netdev_chain and sends a NEWLINK message
1278 * to the routing socket.
1280 void netdev_state_change(struct net_device *dev)
1282 if (dev->flags & IFF_UP) {
1283 struct netdev_notifier_change_info change_info;
1285 change_info.flags_changed = 0;
1286 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1288 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1291 EXPORT_SYMBOL(netdev_state_change);
1294 * netdev_notify_peers - notify network peers about existence of @dev
1295 * @dev: network device
1297 * Generate traffic such that interested network peers are aware of
1298 * @dev, such as by generating a gratuitous ARP. This may be used when
1299 * a device wants to inform the rest of the network about some sort of
1300 * reconfiguration such as a failover event or virtual machine
1303 void netdev_notify_peers(struct net_device *dev)
1306 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1309 EXPORT_SYMBOL(netdev_notify_peers);
1311 static int __dev_open(struct net_device *dev)
1313 const struct net_device_ops *ops = dev->netdev_ops;
1318 if (!netif_device_present(dev))
1321 /* Block netpoll from trying to do any rx path servicing.
1322 * If we don't do this there is a chance ndo_poll_controller
1323 * or ndo_poll may be running while we open the device
1325 netpoll_poll_disable(dev);
1327 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1328 ret = notifier_to_errno(ret);
1332 set_bit(__LINK_STATE_START, &dev->state);
1334 if (ops->ndo_validate_addr)
1335 ret = ops->ndo_validate_addr(dev);
1337 if (!ret && ops->ndo_open)
1338 ret = ops->ndo_open(dev);
1340 netpoll_poll_enable(dev);
1343 clear_bit(__LINK_STATE_START, &dev->state);
1345 dev->flags |= IFF_UP;
1346 dev_set_rx_mode(dev);
1348 add_device_randomness(dev->dev_addr, dev->addr_len);
1355 * dev_open - prepare an interface for use.
1356 * @dev: device to open
1358 * Takes a device from down to up state. The device's private open
1359 * function is invoked and then the multicast lists are loaded. Finally
1360 * the device is moved into the up state and a %NETDEV_UP message is
1361 * sent to the netdev notifier chain.
1363 * Calling this function on an active interface is a nop. On a failure
1364 * a negative errno code is returned.
1366 int dev_open(struct net_device *dev)
1370 if (dev->flags & IFF_UP)
1373 ret = __dev_open(dev);
1377 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1378 call_netdevice_notifiers(NETDEV_UP, dev);
1382 EXPORT_SYMBOL(dev_open);
1384 static int __dev_close_many(struct list_head *head)
1386 struct net_device *dev;
1391 list_for_each_entry(dev, head, close_list) {
1392 /* Temporarily disable netpoll until the interface is down */
1393 netpoll_poll_disable(dev);
1395 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1397 clear_bit(__LINK_STATE_START, &dev->state);
1399 /* Synchronize to scheduled poll. We cannot touch poll list, it
1400 * can be even on different cpu. So just clear netif_running().
1402 * dev->stop() will invoke napi_disable() on all of it's
1403 * napi_struct instances on this device.
1405 smp_mb__after_atomic(); /* Commit netif_running(). */
1408 dev_deactivate_many(head);
1410 list_for_each_entry(dev, head, close_list) {
1411 const struct net_device_ops *ops = dev->netdev_ops;
1414 * Call the device specific close. This cannot fail.
1415 * Only if device is UP
1417 * We allow it to be called even after a DETACH hot-plug
1423 dev->flags &= ~IFF_UP;
1424 netpoll_poll_enable(dev);
1430 static int __dev_close(struct net_device *dev)
1435 list_add(&dev->close_list, &single);
1436 retval = __dev_close_many(&single);
1442 int dev_close_many(struct list_head *head, bool unlink)
1444 struct net_device *dev, *tmp;
1446 /* Remove the devices that don't need to be closed */
1447 list_for_each_entry_safe(dev, tmp, head, close_list)
1448 if (!(dev->flags & IFF_UP))
1449 list_del_init(&dev->close_list);
1451 __dev_close_many(head);
1453 list_for_each_entry_safe(dev, tmp, head, close_list) {
1454 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1455 call_netdevice_notifiers(NETDEV_DOWN, dev);
1457 list_del_init(&dev->close_list);
1462 EXPORT_SYMBOL(dev_close_many);
1465 * dev_close - shutdown an interface.
1466 * @dev: device to shutdown
1468 * This function moves an active device into down state. A
1469 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1470 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1473 int dev_close(struct net_device *dev)
1475 if (dev->flags & IFF_UP) {
1478 list_add(&dev->close_list, &single);
1479 dev_close_many(&single, true);
1484 EXPORT_SYMBOL(dev_close);
1488 * dev_disable_lro - disable Large Receive Offload on a device
1491 * Disable Large Receive Offload (LRO) on a net device. Must be
1492 * called under RTNL. This is needed if received packets may be
1493 * forwarded to another interface.
1495 void dev_disable_lro(struct net_device *dev)
1497 struct net_device *lower_dev;
1498 struct list_head *iter;
1500 dev->wanted_features &= ~NETIF_F_LRO;
1501 netdev_update_features(dev);
1503 if (unlikely(dev->features & NETIF_F_LRO))
1504 netdev_WARN(dev, "failed to disable LRO!\n");
1506 netdev_for_each_lower_dev(dev, lower_dev, iter)
1507 dev_disable_lro(lower_dev);
1509 EXPORT_SYMBOL(dev_disable_lro);
1511 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1512 struct net_device *dev)
1514 struct netdev_notifier_info info;
1516 netdev_notifier_info_init(&info, dev);
1517 return nb->notifier_call(nb, val, &info);
1520 static int dev_boot_phase = 1;
1523 * register_netdevice_notifier - register a network notifier block
1526 * Register a notifier to be called when network device events occur.
1527 * The notifier passed is linked into the kernel structures and must
1528 * not be reused until it has been unregistered. A negative errno code
1529 * is returned on a failure.
1531 * When registered all registration and up events are replayed
1532 * to the new notifier to allow device to have a race free
1533 * view of the network device list.
1536 int register_netdevice_notifier(struct notifier_block *nb)
1538 struct net_device *dev;
1539 struct net_device *last;
1544 err = raw_notifier_chain_register(&netdev_chain, nb);
1550 for_each_netdev(net, dev) {
1551 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1552 err = notifier_to_errno(err);
1556 if (!(dev->flags & IFF_UP))
1559 call_netdevice_notifier(nb, NETDEV_UP, dev);
1570 for_each_netdev(net, dev) {
1574 if (dev->flags & IFF_UP) {
1575 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1577 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1579 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1584 raw_notifier_chain_unregister(&netdev_chain, nb);
1587 EXPORT_SYMBOL(register_netdevice_notifier);
1590 * unregister_netdevice_notifier - unregister a network notifier block
1593 * Unregister a notifier previously registered by
1594 * register_netdevice_notifier(). The notifier is unlinked into the
1595 * kernel structures and may then be reused. A negative errno code
1596 * is returned on a failure.
1598 * After unregistering unregister and down device events are synthesized
1599 * for all devices on the device list to the removed notifier to remove
1600 * the need for special case cleanup code.
1603 int unregister_netdevice_notifier(struct notifier_block *nb)
1605 struct net_device *dev;
1610 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1615 for_each_netdev(net, dev) {
1616 if (dev->flags & IFF_UP) {
1617 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1619 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1621 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1628 EXPORT_SYMBOL(unregister_netdevice_notifier);
1631 * call_netdevice_notifiers_info - call all network notifier blocks
1632 * @val: value passed unmodified to notifier function
1633 * @dev: net_device pointer passed unmodified to notifier function
1634 * @info: notifier information data
1636 * Call all network notifier blocks. Parameters and return value
1637 * are as for raw_notifier_call_chain().
1640 static int call_netdevice_notifiers_info(unsigned long val,
1641 struct net_device *dev,
1642 struct netdev_notifier_info *info)
1645 netdev_notifier_info_init(info, dev);
1646 return raw_notifier_call_chain(&netdev_chain, val, info);
1650 * call_netdevice_notifiers - call all network notifier blocks
1651 * @val: value passed unmodified to notifier function
1652 * @dev: net_device pointer passed unmodified to notifier function
1654 * Call all network notifier blocks. Parameters and return value
1655 * are as for raw_notifier_call_chain().
1658 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1660 struct netdev_notifier_info info;
1662 return call_netdevice_notifiers_info(val, dev, &info);
1664 EXPORT_SYMBOL(call_netdevice_notifiers);
1666 #ifdef CONFIG_NET_INGRESS
1667 static struct static_key ingress_needed __read_mostly;
1669 void net_inc_ingress_queue(void)
1671 static_key_slow_inc(&ingress_needed);
1673 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1675 void net_dec_ingress_queue(void)
1677 static_key_slow_dec(&ingress_needed);
1679 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1682 #ifdef CONFIG_NET_EGRESS
1683 static struct static_key egress_needed __read_mostly;
1685 void net_inc_egress_queue(void)
1687 static_key_slow_inc(&egress_needed);
1689 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1691 void net_dec_egress_queue(void)
1693 static_key_slow_dec(&egress_needed);
1695 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1698 static struct static_key netstamp_needed __read_mostly;
1699 #ifdef HAVE_JUMP_LABEL
1700 static atomic_t netstamp_needed_deferred;
1701 static void netstamp_clear(struct work_struct *work)
1703 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1706 static_key_slow_dec(&netstamp_needed);
1708 static DECLARE_WORK(netstamp_work, netstamp_clear);
1711 void net_enable_timestamp(void)
1713 static_key_slow_inc(&netstamp_needed);
1715 EXPORT_SYMBOL(net_enable_timestamp);
1717 void net_disable_timestamp(void)
1719 #ifdef HAVE_JUMP_LABEL
1720 /* net_disable_timestamp() can be called from non process context */
1721 atomic_inc(&netstamp_needed_deferred);
1722 schedule_work(&netstamp_work);
1724 static_key_slow_dec(&netstamp_needed);
1727 EXPORT_SYMBOL(net_disable_timestamp);
1729 static inline void net_timestamp_set(struct sk_buff *skb)
1732 if (static_key_false(&netstamp_needed))
1733 __net_timestamp(skb);
1736 #define net_timestamp_check(COND, SKB) \
1737 if (static_key_false(&netstamp_needed)) { \
1738 if ((COND) && !(SKB)->tstamp) \
1739 __net_timestamp(SKB); \
1742 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1746 if (!(dev->flags & IFF_UP))
1749 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1750 if (skb->len <= len)
1753 /* if TSO is enabled, we don't care about the length as the packet
1754 * could be forwarded without being segmented before
1756 if (skb_is_gso(skb))
1761 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1763 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1765 int ret = ____dev_forward_skb(dev, skb);
1768 skb->protocol = eth_type_trans(skb, dev);
1769 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1774 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1777 * dev_forward_skb - loopback an skb to another netif
1779 * @dev: destination network device
1780 * @skb: buffer to forward
1783 * NET_RX_SUCCESS (no congestion)
1784 * NET_RX_DROP (packet was dropped, but freed)
1786 * dev_forward_skb can be used for injecting an skb from the
1787 * start_xmit function of one device into the receive queue
1788 * of another device.
1790 * The receiving device may be in another namespace, so
1791 * we have to clear all information in the skb that could
1792 * impact namespace isolation.
1794 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1796 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1798 EXPORT_SYMBOL_GPL(dev_forward_skb);
1800 static inline int deliver_skb(struct sk_buff *skb,
1801 struct packet_type *pt_prev,
1802 struct net_device *orig_dev)
1804 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1806 atomic_inc(&skb->users);
1807 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1810 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1811 struct packet_type **pt,
1812 struct net_device *orig_dev,
1814 struct list_head *ptype_list)
1816 struct packet_type *ptype, *pt_prev = *pt;
1818 list_for_each_entry_rcu(ptype, ptype_list, list) {
1819 if (ptype->type != type)
1822 deliver_skb(skb, pt_prev, orig_dev);
1828 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1830 if (!ptype->af_packet_priv || !skb->sk)
1833 if (ptype->id_match)
1834 return ptype->id_match(ptype, skb->sk);
1835 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1842 * Support routine. Sends outgoing frames to any network
1843 * taps currently in use.
1846 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1848 struct packet_type *ptype;
1849 struct sk_buff *skb2 = NULL;
1850 struct packet_type *pt_prev = NULL;
1851 struct list_head *ptype_list = &ptype_all;
1855 list_for_each_entry_rcu(ptype, ptype_list, list) {
1856 /* Never send packets back to the socket
1857 * they originated from - MvS (miquels@drinkel.ow.org)
1859 if (skb_loop_sk(ptype, skb))
1863 deliver_skb(skb2, pt_prev, skb->dev);
1868 /* need to clone skb, done only once */
1869 skb2 = skb_clone(skb, GFP_ATOMIC);
1873 net_timestamp_set(skb2);
1875 /* skb->nh should be correctly
1876 * set by sender, so that the second statement is
1877 * just protection against buggy protocols.
1879 skb_reset_mac_header(skb2);
1881 if (skb_network_header(skb2) < skb2->data ||
1882 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1883 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1884 ntohs(skb2->protocol),
1886 skb_reset_network_header(skb2);
1889 skb2->transport_header = skb2->network_header;
1890 skb2->pkt_type = PACKET_OUTGOING;
1894 if (ptype_list == &ptype_all) {
1895 ptype_list = &dev->ptype_all;
1900 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1903 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
1906 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1907 * @dev: Network device
1908 * @txq: number of queues available
1910 * If real_num_tx_queues is changed the tc mappings may no longer be
1911 * valid. To resolve this verify the tc mapping remains valid and if
1912 * not NULL the mapping. With no priorities mapping to this
1913 * offset/count pair it will no longer be used. In the worst case TC0
1914 * is invalid nothing can be done so disable priority mappings. If is
1915 * expected that drivers will fix this mapping if they can before
1916 * calling netif_set_real_num_tx_queues.
1918 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1921 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1923 /* If TC0 is invalidated disable TC mapping */
1924 if (tc->offset + tc->count > txq) {
1925 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1930 /* Invalidated prio to tc mappings set to TC0 */
1931 for (i = 1; i < TC_BITMASK + 1; i++) {
1932 int q = netdev_get_prio_tc_map(dev, i);
1934 tc = &dev->tc_to_txq[q];
1935 if (tc->offset + tc->count > txq) {
1936 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1938 netdev_set_prio_tc_map(dev, i, 0);
1943 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
1946 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1949 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
1950 if ((txq - tc->offset) < tc->count)
1961 static DEFINE_MUTEX(xps_map_mutex);
1962 #define xmap_dereference(P) \
1963 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1965 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
1968 struct xps_map *map = NULL;
1972 map = xmap_dereference(dev_maps->cpu_map[tci]);
1976 for (pos = map->len; pos--;) {
1977 if (map->queues[pos] != index)
1981 map->queues[pos] = map->queues[--map->len];
1985 RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL);
1986 kfree_rcu(map, rcu);
1993 static bool remove_xps_queue_cpu(struct net_device *dev,
1994 struct xps_dev_maps *dev_maps,
1995 int cpu, u16 offset, u16 count)
1997 int num_tc = dev->num_tc ? : 1;
1998 bool active = false;
2001 for (tci = cpu * num_tc; num_tc--; tci++) {
2004 for (i = count, j = offset; i--; j++) {
2005 if (!remove_xps_queue(dev_maps, cpu, j))
2015 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2018 struct xps_dev_maps *dev_maps;
2020 bool active = false;
2022 mutex_lock(&xps_map_mutex);
2023 dev_maps = xmap_dereference(dev->xps_maps);
2028 for_each_possible_cpu(cpu)
2029 active |= remove_xps_queue_cpu(dev, dev_maps, cpu,
2033 RCU_INIT_POINTER(dev->xps_maps, NULL);
2034 kfree_rcu(dev_maps, rcu);
2037 for (i = offset + (count - 1); count--; i--)
2038 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2042 mutex_unlock(&xps_map_mutex);
2045 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2047 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2050 static struct xps_map *expand_xps_map(struct xps_map *map,
2053 struct xps_map *new_map;
2054 int alloc_len = XPS_MIN_MAP_ALLOC;
2057 for (pos = 0; map && pos < map->len; pos++) {
2058 if (map->queues[pos] != index)
2063 /* Need to add queue to this CPU's existing map */
2065 if (pos < map->alloc_len)
2068 alloc_len = map->alloc_len * 2;
2071 /* Need to allocate new map to store queue on this CPU's map */
2072 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2077 for (i = 0; i < pos; i++)
2078 new_map->queues[i] = map->queues[i];
2079 new_map->alloc_len = alloc_len;
2085 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2088 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2089 int i, cpu, tci, numa_node_id = -2;
2090 int maps_sz, num_tc = 1, tc = 0;
2091 struct xps_map *map, *new_map;
2092 bool active = false;
2095 num_tc = dev->num_tc;
2096 tc = netdev_txq_to_tc(dev, index);
2101 maps_sz = XPS_DEV_MAPS_SIZE(num_tc);
2102 if (maps_sz < L1_CACHE_BYTES)
2103 maps_sz = L1_CACHE_BYTES;
2105 mutex_lock(&xps_map_mutex);
2107 dev_maps = xmap_dereference(dev->xps_maps);
2109 /* allocate memory for queue storage */
2110 for_each_cpu_and(cpu, cpu_online_mask, mask) {
2112 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2113 if (!new_dev_maps) {
2114 mutex_unlock(&xps_map_mutex);
2118 tci = cpu * num_tc + tc;
2119 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) :
2122 map = expand_xps_map(map, cpu, index);
2126 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2130 goto out_no_new_maps;
2132 for_each_possible_cpu(cpu) {
2133 /* copy maps belonging to foreign traffic classes */
2134 for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) {
2135 /* fill in the new device map from the old device map */
2136 map = xmap_dereference(dev_maps->cpu_map[tci]);
2137 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2140 /* We need to explicitly update tci as prevous loop
2141 * could break out early if dev_maps is NULL.
2143 tci = cpu * num_tc + tc;
2145 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2146 /* add queue to CPU maps */
2149 map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2150 while ((pos < map->len) && (map->queues[pos] != index))
2153 if (pos == map->len)
2154 map->queues[map->len++] = index;
2156 if (numa_node_id == -2)
2157 numa_node_id = cpu_to_node(cpu);
2158 else if (numa_node_id != cpu_to_node(cpu))
2161 } else if (dev_maps) {
2162 /* fill in the new device map from the old device map */
2163 map = xmap_dereference(dev_maps->cpu_map[tci]);
2164 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2167 /* copy maps belonging to foreign traffic classes */
2168 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2169 /* fill in the new device map from the old device map */
2170 map = xmap_dereference(dev_maps->cpu_map[tci]);
2171 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2175 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2177 /* Cleanup old maps */
2179 goto out_no_old_maps;
2181 for_each_possible_cpu(cpu) {
2182 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2183 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2184 map = xmap_dereference(dev_maps->cpu_map[tci]);
2185 if (map && map != new_map)
2186 kfree_rcu(map, rcu);
2190 kfree_rcu(dev_maps, rcu);
2193 dev_maps = new_dev_maps;
2197 /* update Tx queue numa node */
2198 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2199 (numa_node_id >= 0) ? numa_node_id :
2205 /* removes queue from unused CPUs */
2206 for_each_possible_cpu(cpu) {
2207 for (i = tc, tci = cpu * num_tc; i--; tci++)
2208 active |= remove_xps_queue(dev_maps, tci, index);
2209 if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu))
2210 active |= remove_xps_queue(dev_maps, tci, index);
2211 for (i = num_tc - tc, tci++; --i; tci++)
2212 active |= remove_xps_queue(dev_maps, tci, index);
2215 /* free map if not active */
2217 RCU_INIT_POINTER(dev->xps_maps, NULL);
2218 kfree_rcu(dev_maps, rcu);
2222 mutex_unlock(&xps_map_mutex);
2226 /* remove any maps that we added */
2227 for_each_possible_cpu(cpu) {
2228 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2229 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2231 xmap_dereference(dev_maps->cpu_map[tci]) :
2233 if (new_map && new_map != map)
2238 mutex_unlock(&xps_map_mutex);
2240 kfree(new_dev_maps);
2243 EXPORT_SYMBOL(netif_set_xps_queue);
2246 void netdev_reset_tc(struct net_device *dev)
2249 netif_reset_xps_queues_gt(dev, 0);
2252 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2253 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2255 EXPORT_SYMBOL(netdev_reset_tc);
2257 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2259 if (tc >= dev->num_tc)
2263 netif_reset_xps_queues(dev, offset, count);
2265 dev->tc_to_txq[tc].count = count;
2266 dev->tc_to_txq[tc].offset = offset;
2269 EXPORT_SYMBOL(netdev_set_tc_queue);
2271 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2273 if (num_tc > TC_MAX_QUEUE)
2277 netif_reset_xps_queues_gt(dev, 0);
2279 dev->num_tc = num_tc;
2282 EXPORT_SYMBOL(netdev_set_num_tc);
2285 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2286 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2288 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2292 if (txq < 1 || txq > dev->num_tx_queues)
2295 if (dev->reg_state == NETREG_REGISTERED ||
2296 dev->reg_state == NETREG_UNREGISTERING) {
2299 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2305 netif_setup_tc(dev, txq);
2307 if (txq < dev->real_num_tx_queues) {
2308 qdisc_reset_all_tx_gt(dev, txq);
2310 netif_reset_xps_queues_gt(dev, txq);
2315 dev->real_num_tx_queues = txq;
2318 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2322 * netif_set_real_num_rx_queues - set actual number of RX queues used
2323 * @dev: Network device
2324 * @rxq: Actual number of RX queues
2326 * This must be called either with the rtnl_lock held or before
2327 * registration of the net device. Returns 0 on success, or a
2328 * negative error code. If called before registration, it always
2331 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2335 if (rxq < 1 || rxq > dev->num_rx_queues)
2338 if (dev->reg_state == NETREG_REGISTERED) {
2341 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2347 dev->real_num_rx_queues = rxq;
2350 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2354 * netif_get_num_default_rss_queues - default number of RSS queues
2356 * This routine should set an upper limit on the number of RSS queues
2357 * used by default by multiqueue devices.
2359 int netif_get_num_default_rss_queues(void)
2361 return is_kdump_kernel() ?
2362 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2364 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2366 static void __netif_reschedule(struct Qdisc *q)
2368 struct softnet_data *sd;
2369 unsigned long flags;
2371 local_irq_save(flags);
2372 sd = this_cpu_ptr(&softnet_data);
2373 q->next_sched = NULL;
2374 *sd->output_queue_tailp = q;
2375 sd->output_queue_tailp = &q->next_sched;
2376 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2377 local_irq_restore(flags);
2380 void __netif_schedule(struct Qdisc *q)
2382 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2383 __netif_reschedule(q);
2385 EXPORT_SYMBOL(__netif_schedule);
2387 struct dev_kfree_skb_cb {
2388 enum skb_free_reason reason;
2391 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2393 return (struct dev_kfree_skb_cb *)skb->cb;
2396 void netif_schedule_queue(struct netdev_queue *txq)
2399 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2400 struct Qdisc *q = rcu_dereference(txq->qdisc);
2402 __netif_schedule(q);
2406 EXPORT_SYMBOL(netif_schedule_queue);
2408 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2410 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2414 q = rcu_dereference(dev_queue->qdisc);
2415 __netif_schedule(q);
2419 EXPORT_SYMBOL(netif_tx_wake_queue);
2421 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2423 unsigned long flags;
2425 if (likely(atomic_read(&skb->users) == 1)) {
2427 atomic_set(&skb->users, 0);
2428 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2431 get_kfree_skb_cb(skb)->reason = reason;
2432 local_irq_save(flags);
2433 skb->next = __this_cpu_read(softnet_data.completion_queue);
2434 __this_cpu_write(softnet_data.completion_queue, skb);
2435 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2436 local_irq_restore(flags);
2438 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2440 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2442 if (in_irq() || irqs_disabled())
2443 __dev_kfree_skb_irq(skb, reason);
2447 EXPORT_SYMBOL(__dev_kfree_skb_any);
2451 * netif_device_detach - mark device as removed
2452 * @dev: network device
2454 * Mark device as removed from system and therefore no longer available.
2456 void netif_device_detach(struct net_device *dev)
2458 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2459 netif_running(dev)) {
2460 netif_tx_stop_all_queues(dev);
2463 EXPORT_SYMBOL(netif_device_detach);
2466 * netif_device_attach - mark device as attached
2467 * @dev: network device
2469 * Mark device as attached from system and restart if needed.
2471 void netif_device_attach(struct net_device *dev)
2473 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2474 netif_running(dev)) {
2475 netif_tx_wake_all_queues(dev);
2476 __netdev_watchdog_up(dev);
2479 EXPORT_SYMBOL(netif_device_attach);
2482 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2483 * to be used as a distribution range.
2485 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2486 unsigned int num_tx_queues)
2490 u16 qcount = num_tx_queues;
2492 if (skb_rx_queue_recorded(skb)) {
2493 hash = skb_get_rx_queue(skb);
2494 while (unlikely(hash >= num_tx_queues))
2495 hash -= num_tx_queues;
2500 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2502 qoffset = dev->tc_to_txq[tc].offset;
2503 qcount = dev->tc_to_txq[tc].count;
2506 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2508 EXPORT_SYMBOL(__skb_tx_hash);
2510 static void skb_warn_bad_offload(const struct sk_buff *skb)
2512 static const netdev_features_t null_features;
2513 struct net_device *dev = skb->dev;
2514 const char *name = "";
2516 if (!net_ratelimit())
2520 if (dev->dev.parent)
2521 name = dev_driver_string(dev->dev.parent);
2523 name = netdev_name(dev);
2525 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2526 "gso_type=%d ip_summed=%d\n",
2527 name, dev ? &dev->features : &null_features,
2528 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2529 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2530 skb_shinfo(skb)->gso_type, skb->ip_summed);
2534 * Invalidate hardware checksum when packet is to be mangled, and
2535 * complete checksum manually on outgoing path.
2537 int skb_checksum_help(struct sk_buff *skb)
2540 int ret = 0, offset;
2542 if (skb->ip_summed == CHECKSUM_COMPLETE)
2543 goto out_set_summed;
2545 if (unlikely(skb_shinfo(skb)->gso_size)) {
2546 skb_warn_bad_offload(skb);
2550 /* Before computing a checksum, we should make sure no frag could
2551 * be modified by an external entity : checksum could be wrong.
2553 if (skb_has_shared_frag(skb)) {
2554 ret = __skb_linearize(skb);
2559 offset = skb_checksum_start_offset(skb);
2560 BUG_ON(offset >= skb_headlen(skb));
2561 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2563 offset += skb->csum_offset;
2564 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2566 if (skb_cloned(skb) &&
2567 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2568 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2573 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2575 skb->ip_summed = CHECKSUM_NONE;
2579 EXPORT_SYMBOL(skb_checksum_help);
2581 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2583 __be16 type = skb->protocol;
2585 /* Tunnel gso handlers can set protocol to ethernet. */
2586 if (type == htons(ETH_P_TEB)) {
2589 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2592 eth = (struct ethhdr *)skb_mac_header(skb);
2593 type = eth->h_proto;
2596 return __vlan_get_protocol(skb, type, depth);
2600 * skb_mac_gso_segment - mac layer segmentation handler.
2601 * @skb: buffer to segment
2602 * @features: features for the output path (see dev->features)
2604 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2605 netdev_features_t features)
2607 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2608 struct packet_offload *ptype;
2609 int vlan_depth = skb->mac_len;
2610 __be16 type = skb_network_protocol(skb, &vlan_depth);
2612 if (unlikely(!type))
2613 return ERR_PTR(-EINVAL);
2615 __skb_pull(skb, vlan_depth);
2618 list_for_each_entry_rcu(ptype, &offload_base, list) {
2619 if (ptype->type == type && ptype->callbacks.gso_segment) {
2620 segs = ptype->callbacks.gso_segment(skb, features);
2626 __skb_push(skb, skb->data - skb_mac_header(skb));
2630 EXPORT_SYMBOL(skb_mac_gso_segment);
2633 /* openvswitch calls this on rx path, so we need a different check.
2635 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2638 return skb->ip_summed != CHECKSUM_PARTIAL &&
2639 skb->ip_summed != CHECKSUM_NONE;
2641 return skb->ip_summed == CHECKSUM_NONE;
2645 * __skb_gso_segment - Perform segmentation on skb.
2646 * @skb: buffer to segment
2647 * @features: features for the output path (see dev->features)
2648 * @tx_path: whether it is called in TX path
2650 * This function segments the given skb and returns a list of segments.
2652 * It may return NULL if the skb requires no segmentation. This is
2653 * only possible when GSO is used for verifying header integrity.
2655 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2657 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2658 netdev_features_t features, bool tx_path)
2660 struct sk_buff *segs;
2662 if (unlikely(skb_needs_check(skb, tx_path))) {
2665 /* We're going to init ->check field in TCP or UDP header */
2666 err = skb_cow_head(skb, 0);
2668 return ERR_PTR(err);
2671 /* Only report GSO partial support if it will enable us to
2672 * support segmentation on this frame without needing additional
2675 if (features & NETIF_F_GSO_PARTIAL) {
2676 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
2677 struct net_device *dev = skb->dev;
2679 partial_features |= dev->features & dev->gso_partial_features;
2680 if (!skb_gso_ok(skb, features | partial_features))
2681 features &= ~NETIF_F_GSO_PARTIAL;
2684 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2685 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2687 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2688 SKB_GSO_CB(skb)->encap_level = 0;
2690 skb_reset_mac_header(skb);
2691 skb_reset_mac_len(skb);
2693 segs = skb_mac_gso_segment(skb, features);
2695 if (unlikely(skb_needs_check(skb, tx_path)))
2696 skb_warn_bad_offload(skb);
2700 EXPORT_SYMBOL(__skb_gso_segment);
2702 /* Take action when hardware reception checksum errors are detected. */
2704 void netdev_rx_csum_fault(struct net_device *dev)
2706 if (net_ratelimit()) {
2707 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2711 EXPORT_SYMBOL(netdev_rx_csum_fault);
2714 /* Actually, we should eliminate this check as soon as we know, that:
2715 * 1. IOMMU is present and allows to map all the memory.
2716 * 2. No high memory really exists on this machine.
2719 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2721 #ifdef CONFIG_HIGHMEM
2724 if (!(dev->features & NETIF_F_HIGHDMA)) {
2725 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2726 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2728 if (PageHighMem(skb_frag_page(frag)))
2733 if (PCI_DMA_BUS_IS_PHYS) {
2734 struct device *pdev = dev->dev.parent;
2738 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2739 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2740 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2742 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2750 /* If MPLS offload request, verify we are testing hardware MPLS features
2751 * instead of standard features for the netdev.
2753 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2754 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2755 netdev_features_t features,
2758 if (eth_p_mpls(type))
2759 features &= skb->dev->mpls_features;
2764 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2765 netdev_features_t features,
2772 static netdev_features_t harmonize_features(struct sk_buff *skb,
2773 netdev_features_t features)
2778 type = skb_network_protocol(skb, &tmp);
2779 features = net_mpls_features(skb, features, type);
2781 if (skb->ip_summed != CHECKSUM_NONE &&
2782 !can_checksum_protocol(features, type)) {
2783 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2785 if (illegal_highdma(skb->dev, skb))
2786 features &= ~NETIF_F_SG;
2791 netdev_features_t passthru_features_check(struct sk_buff *skb,
2792 struct net_device *dev,
2793 netdev_features_t features)
2797 EXPORT_SYMBOL(passthru_features_check);
2799 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2800 struct net_device *dev,
2801 netdev_features_t features)
2803 return vlan_features_check(skb, features);
2806 static netdev_features_t gso_features_check(const struct sk_buff *skb,
2807 struct net_device *dev,
2808 netdev_features_t features)
2810 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2812 if (gso_segs > dev->gso_max_segs)
2813 return features & ~NETIF_F_GSO_MASK;
2815 /* Support for GSO partial features requires software
2816 * intervention before we can actually process the packets
2817 * so we need to strip support for any partial features now
2818 * and we can pull them back in after we have partially
2819 * segmented the frame.
2821 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
2822 features &= ~dev->gso_partial_features;
2824 /* Make sure to clear the IPv4 ID mangling feature if the
2825 * IPv4 header has the potential to be fragmented.
2827 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2828 struct iphdr *iph = skb->encapsulation ?
2829 inner_ip_hdr(skb) : ip_hdr(skb);
2831 if (!(iph->frag_off & htons(IP_DF)))
2832 features &= ~NETIF_F_TSO_MANGLEID;
2838 netdev_features_t netif_skb_features(struct sk_buff *skb)
2840 struct net_device *dev = skb->dev;
2841 netdev_features_t features = dev->features;
2843 if (skb_is_gso(skb))
2844 features = gso_features_check(skb, dev, features);
2846 /* If encapsulation offload request, verify we are testing
2847 * hardware encapsulation features instead of standard
2848 * features for the netdev
2850 if (skb->encapsulation)
2851 features &= dev->hw_enc_features;
2853 if (skb_vlan_tagged(skb))
2854 features = netdev_intersect_features(features,
2855 dev->vlan_features |
2856 NETIF_F_HW_VLAN_CTAG_TX |
2857 NETIF_F_HW_VLAN_STAG_TX);
2859 if (dev->netdev_ops->ndo_features_check)
2860 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2863 features &= dflt_features_check(skb, dev, features);
2865 return harmonize_features(skb, features);
2867 EXPORT_SYMBOL(netif_skb_features);
2869 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2870 struct netdev_queue *txq, bool more)
2875 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2876 dev_queue_xmit_nit(skb, dev);
2879 trace_net_dev_start_xmit(skb, dev);
2880 rc = netdev_start_xmit(skb, dev, txq, more);
2881 trace_net_dev_xmit(skb, rc, dev, len);
2886 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2887 struct netdev_queue *txq, int *ret)
2889 struct sk_buff *skb = first;
2890 int rc = NETDEV_TX_OK;
2893 struct sk_buff *next = skb->next;
2896 rc = xmit_one(skb, dev, txq, next != NULL);
2897 if (unlikely(!dev_xmit_complete(rc))) {
2903 if (netif_xmit_stopped(txq) && skb) {
2904 rc = NETDEV_TX_BUSY;
2914 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2915 netdev_features_t features)
2917 if (skb_vlan_tag_present(skb) &&
2918 !vlan_hw_offload_capable(features, skb->vlan_proto))
2919 skb = __vlan_hwaccel_push_inside(skb);
2923 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2925 netdev_features_t features;
2927 features = netif_skb_features(skb);
2928 skb = validate_xmit_vlan(skb, features);
2932 if (netif_needs_gso(skb, features)) {
2933 struct sk_buff *segs;
2935 segs = skb_gso_segment(skb, features);
2943 if (skb_needs_linearize(skb, features) &&
2944 __skb_linearize(skb))
2947 /* If packet is not checksummed and device does not
2948 * support checksumming for this protocol, complete
2949 * checksumming here.
2951 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2952 if (skb->encapsulation)
2953 skb_set_inner_transport_header(skb,
2954 skb_checksum_start_offset(skb));
2956 skb_set_transport_header(skb,
2957 skb_checksum_start_offset(skb));
2958 if (!(features & NETIF_F_CSUM_MASK) &&
2959 skb_checksum_help(skb))
2969 atomic_long_inc(&dev->tx_dropped);
2973 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2975 struct sk_buff *next, *head = NULL, *tail;
2977 for (; skb != NULL; skb = next) {
2981 /* in case skb wont be segmented, point to itself */
2984 skb = validate_xmit_skb(skb, dev);
2992 /* If skb was segmented, skb->prev points to
2993 * the last segment. If not, it still contains skb.
2999 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3001 static void qdisc_pkt_len_init(struct sk_buff *skb)
3003 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3005 qdisc_skb_cb(skb)->pkt_len = skb->len;
3007 /* To get more precise estimation of bytes sent on wire,
3008 * we add to pkt_len the headers size of all segments
3010 if (shinfo->gso_size) {
3011 unsigned int hdr_len;
3012 u16 gso_segs = shinfo->gso_segs;
3014 /* mac layer + network layer */
3015 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3017 /* + transport layer */
3018 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
3019 hdr_len += tcp_hdrlen(skb);
3021 hdr_len += sizeof(struct udphdr);
3023 if (shinfo->gso_type & SKB_GSO_DODGY)
3024 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3027 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3031 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3032 struct net_device *dev,
3033 struct netdev_queue *txq)
3035 spinlock_t *root_lock = qdisc_lock(q);
3036 struct sk_buff *to_free = NULL;
3040 qdisc_calculate_pkt_len(skb, q);
3042 * Heuristic to force contended enqueues to serialize on a
3043 * separate lock before trying to get qdisc main lock.
3044 * This permits qdisc->running owner to get the lock more
3045 * often and dequeue packets faster.
3047 contended = qdisc_is_running(q);
3048 if (unlikely(contended))
3049 spin_lock(&q->busylock);
3051 spin_lock(root_lock);
3052 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3053 __qdisc_drop(skb, &to_free);
3055 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3056 qdisc_run_begin(q)) {
3058 * This is a work-conserving queue; there are no old skbs
3059 * waiting to be sent out; and the qdisc is not running -
3060 * xmit the skb directly.
3063 qdisc_bstats_update(q, skb);
3065 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3066 if (unlikely(contended)) {
3067 spin_unlock(&q->busylock);
3074 rc = NET_XMIT_SUCCESS;
3076 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3077 if (qdisc_run_begin(q)) {
3078 if (unlikely(contended)) {
3079 spin_unlock(&q->busylock);
3085 spin_unlock(root_lock);
3086 if (unlikely(to_free))
3087 kfree_skb_list(to_free);
3088 if (unlikely(contended))
3089 spin_unlock(&q->busylock);
3093 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3094 static void skb_update_prio(struct sk_buff *skb)
3096 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
3098 if (!skb->priority && skb->sk && map) {
3099 unsigned int prioidx =
3100 sock_cgroup_prioidx(&skb->sk->sk_cgrp_data);
3102 if (prioidx < map->priomap_len)
3103 skb->priority = map->priomap[prioidx];
3107 #define skb_update_prio(skb)
3110 DEFINE_PER_CPU(int, xmit_recursion);
3111 EXPORT_SYMBOL(xmit_recursion);
3114 * dev_loopback_xmit - loop back @skb
3115 * @net: network namespace this loopback is happening in
3116 * @sk: sk needed to be a netfilter okfn
3117 * @skb: buffer to transmit
3119 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3121 skb_reset_mac_header(skb);
3122 __skb_pull(skb, skb_network_offset(skb));
3123 skb->pkt_type = PACKET_LOOPBACK;
3124 skb->ip_summed = CHECKSUM_UNNECESSARY;
3125 WARN_ON(!skb_dst(skb));
3130 EXPORT_SYMBOL(dev_loopback_xmit);
3132 #ifdef CONFIG_NET_EGRESS
3133 static struct sk_buff *
3134 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3136 struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list);
3137 struct tcf_result cl_res;
3142 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3143 qdisc_bstats_cpu_update(cl->q, skb);
3145 switch (tc_classify(skb, cl, &cl_res, false)) {
3147 case TC_ACT_RECLASSIFY:
3148 skb->tc_index = TC_H_MIN(cl_res.classid);
3151 qdisc_qstats_cpu_drop(cl->q);
3152 *ret = NET_XMIT_DROP;
3157 *ret = NET_XMIT_SUCCESS;
3160 case TC_ACT_REDIRECT:
3161 /* No need to push/pop skb's mac_header here on egress! */
3162 skb_do_redirect(skb);
3163 *ret = NET_XMIT_SUCCESS;
3171 #endif /* CONFIG_NET_EGRESS */
3173 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3176 struct xps_dev_maps *dev_maps;
3177 struct xps_map *map;
3178 int queue_index = -1;
3181 dev_maps = rcu_dereference(dev->xps_maps);
3183 unsigned int tci = skb->sender_cpu - 1;
3187 tci += netdev_get_prio_tc_map(dev, skb->priority);
3190 map = rcu_dereference(dev_maps->cpu_map[tci]);
3193 queue_index = map->queues[0];
3195 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3197 if (unlikely(queue_index >= dev->real_num_tx_queues))
3209 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3211 struct sock *sk = skb->sk;
3212 int queue_index = sk_tx_queue_get(sk);
3214 if (queue_index < 0 || skb->ooo_okay ||
3215 queue_index >= dev->real_num_tx_queues) {
3216 int new_index = get_xps_queue(dev, skb);
3219 new_index = skb_tx_hash(dev, skb);
3221 if (queue_index != new_index && sk &&
3223 rcu_access_pointer(sk->sk_dst_cache))
3224 sk_tx_queue_set(sk, new_index);
3226 queue_index = new_index;
3232 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3233 struct sk_buff *skb,
3236 int queue_index = 0;
3239 u32 sender_cpu = skb->sender_cpu - 1;
3241 if (sender_cpu >= (u32)NR_CPUS)
3242 skb->sender_cpu = raw_smp_processor_id() + 1;
3245 if (dev->real_num_tx_queues != 1) {
3246 const struct net_device_ops *ops = dev->netdev_ops;
3248 if (ops->ndo_select_queue)
3249 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3252 queue_index = __netdev_pick_tx(dev, skb);
3255 queue_index = netdev_cap_txqueue(dev, queue_index);
3258 skb_set_queue_mapping(skb, queue_index);
3259 return netdev_get_tx_queue(dev, queue_index);
3263 * __dev_queue_xmit - transmit a buffer
3264 * @skb: buffer to transmit
3265 * @accel_priv: private data used for L2 forwarding offload
3267 * Queue a buffer for transmission to a network device. The caller must
3268 * have set the device and priority and built the buffer before calling
3269 * this function. The function can be called from an interrupt.
3271 * A negative errno code is returned on a failure. A success does not
3272 * guarantee the frame will be transmitted as it may be dropped due
3273 * to congestion or traffic shaping.
3275 * -----------------------------------------------------------------------------------
3276 * I notice this method can also return errors from the queue disciplines,
3277 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3280 * Regardless of the return value, the skb is consumed, so it is currently
3281 * difficult to retry a send to this method. (You can bump the ref count
3282 * before sending to hold a reference for retry if you are careful.)
3284 * When calling this method, interrupts MUST be enabled. This is because
3285 * the BH enable code must have IRQs enabled so that it will not deadlock.
3288 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3290 struct net_device *dev = skb->dev;
3291 struct netdev_queue *txq;
3295 skb_reset_mac_header(skb);
3297 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3298 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3300 /* Disable soft irqs for various locks below. Also
3301 * stops preemption for RCU.
3305 skb_update_prio(skb);
3307 qdisc_pkt_len_init(skb);
3308 #ifdef CONFIG_NET_CLS_ACT
3309 skb->tc_at_ingress = 0;
3310 # ifdef CONFIG_NET_EGRESS
3311 if (static_key_false(&egress_needed)) {
3312 skb = sch_handle_egress(skb, &rc, dev);
3318 /* If device/qdisc don't need skb->dst, release it right now while
3319 * its hot in this cpu cache.
3321 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3326 txq = netdev_pick_tx(dev, skb, accel_priv);
3327 q = rcu_dereference_bh(txq->qdisc);
3329 trace_net_dev_queue(skb);
3331 rc = __dev_xmit_skb(skb, q, dev, txq);
3335 /* The device has no queue. Common case for software devices:
3336 * loopback, all the sorts of tunnels...
3338 * Really, it is unlikely that netif_tx_lock protection is necessary
3339 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
3341 * However, it is possible, that they rely on protection
3344 * Check this and shot the lock. It is not prone from deadlocks.
3345 *Either shot noqueue qdisc, it is even simpler 8)
3347 if (dev->flags & IFF_UP) {
3348 int cpu = smp_processor_id(); /* ok because BHs are off */
3350 if (txq->xmit_lock_owner != cpu) {
3351 if (unlikely(__this_cpu_read(xmit_recursion) >
3352 XMIT_RECURSION_LIMIT))
3353 goto recursion_alert;
3355 skb = validate_xmit_skb(skb, dev);
3359 HARD_TX_LOCK(dev, txq, cpu);
3361 if (!netif_xmit_stopped(txq)) {
3362 __this_cpu_inc(xmit_recursion);
3363 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3364 __this_cpu_dec(xmit_recursion);
3365 if (dev_xmit_complete(rc)) {
3366 HARD_TX_UNLOCK(dev, txq);
3370 HARD_TX_UNLOCK(dev, txq);
3371 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3374 /* Recursion is detected! It is possible,
3378 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3384 rcu_read_unlock_bh();
3386 atomic_long_inc(&dev->tx_dropped);
3387 kfree_skb_list(skb);
3390 rcu_read_unlock_bh();
3394 int dev_queue_xmit(struct sk_buff *skb)
3396 return __dev_queue_xmit(skb, NULL);
3398 EXPORT_SYMBOL(dev_queue_xmit);
3400 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3402 return __dev_queue_xmit(skb, accel_priv);
3404 EXPORT_SYMBOL(dev_queue_xmit_accel);
3407 /*************************************************************************
3409 *************************************************************************/
3411 int netdev_max_backlog __read_mostly = 1000;
3412 EXPORT_SYMBOL(netdev_max_backlog);
3414 int netdev_tstamp_prequeue __read_mostly = 1;
3415 int netdev_budget __read_mostly = 300;
3416 int weight_p __read_mostly = 64; /* old backlog weight */
3417 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
3418 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
3419 int dev_rx_weight __read_mostly = 64;
3420 int dev_tx_weight __read_mostly = 64;
3422 /* Called with irq disabled */
3423 static inline void ____napi_schedule(struct softnet_data *sd,
3424 struct napi_struct *napi)
3426 list_add_tail(&napi->poll_list, &sd->poll_list);
3427 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3432 /* One global table that all flow-based protocols share. */
3433 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3434 EXPORT_SYMBOL(rps_sock_flow_table);
3435 u32 rps_cpu_mask __read_mostly;
3436 EXPORT_SYMBOL(rps_cpu_mask);
3438 struct static_key rps_needed __read_mostly;
3439 EXPORT_SYMBOL(rps_needed);
3440 struct static_key rfs_needed __read_mostly;
3441 EXPORT_SYMBOL(rfs_needed);
3443 static struct rps_dev_flow *
3444 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3445 struct rps_dev_flow *rflow, u16 next_cpu)
3447 if (next_cpu < nr_cpu_ids) {
3448 #ifdef CONFIG_RFS_ACCEL
3449 struct netdev_rx_queue *rxqueue;
3450 struct rps_dev_flow_table *flow_table;
3451 struct rps_dev_flow *old_rflow;
3456 /* Should we steer this flow to a different hardware queue? */
3457 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3458 !(dev->features & NETIF_F_NTUPLE))
3460 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3461 if (rxq_index == skb_get_rx_queue(skb))
3464 rxqueue = dev->_rx + rxq_index;
3465 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3468 flow_id = skb_get_hash(skb) & flow_table->mask;
3469 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3470 rxq_index, flow_id);
3474 rflow = &flow_table->flows[flow_id];
3476 if (old_rflow->filter == rflow->filter)
3477 old_rflow->filter = RPS_NO_FILTER;
3481 per_cpu(softnet_data, next_cpu).input_queue_head;
3484 rflow->cpu = next_cpu;
3489 * get_rps_cpu is called from netif_receive_skb and returns the target
3490 * CPU from the RPS map of the receiving queue for a given skb.
3491 * rcu_read_lock must be held on entry.
3493 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3494 struct rps_dev_flow **rflowp)
3496 const struct rps_sock_flow_table *sock_flow_table;
3497 struct netdev_rx_queue *rxqueue = dev->_rx;
3498 struct rps_dev_flow_table *flow_table;
3499 struct rps_map *map;
3504 if (skb_rx_queue_recorded(skb)) {
3505 u16 index = skb_get_rx_queue(skb);
3507 if (unlikely(index >= dev->real_num_rx_queues)) {
3508 WARN_ONCE(dev->real_num_rx_queues > 1,
3509 "%s received packet on queue %u, but number "
3510 "of RX queues is %u\n",
3511 dev->name, index, dev->real_num_rx_queues);
3517 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3519 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3520 map = rcu_dereference(rxqueue->rps_map);
3521 if (!flow_table && !map)
3524 skb_reset_network_header(skb);
3525 hash = skb_get_hash(skb);
3529 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3530 if (flow_table && sock_flow_table) {
3531 struct rps_dev_flow *rflow;
3535 /* First check into global flow table if there is a match */
3536 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3537 if ((ident ^ hash) & ~rps_cpu_mask)
3540 next_cpu = ident & rps_cpu_mask;
3542 /* OK, now we know there is a match,
3543 * we can look at the local (per receive queue) flow table
3545 rflow = &flow_table->flows[hash & flow_table->mask];
3549 * If the desired CPU (where last recvmsg was done) is
3550 * different from current CPU (one in the rx-queue flow
3551 * table entry), switch if one of the following holds:
3552 * - Current CPU is unset (>= nr_cpu_ids).
3553 * - Current CPU is offline.
3554 * - The current CPU's queue tail has advanced beyond the
3555 * last packet that was enqueued using this table entry.
3556 * This guarantees that all previous packets for the flow
3557 * have been dequeued, thus preserving in order delivery.
3559 if (unlikely(tcpu != next_cpu) &&
3560 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3561 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3562 rflow->last_qtail)) >= 0)) {
3564 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3567 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3577 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3578 if (cpu_online(tcpu)) {
3588 #ifdef CONFIG_RFS_ACCEL
3591 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3592 * @dev: Device on which the filter was set
3593 * @rxq_index: RX queue index
3594 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3595 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3597 * Drivers that implement ndo_rx_flow_steer() should periodically call
3598 * this function for each installed filter and remove the filters for
3599 * which it returns %true.
3601 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3602 u32 flow_id, u16 filter_id)
3604 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3605 struct rps_dev_flow_table *flow_table;
3606 struct rps_dev_flow *rflow;
3611 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3612 if (flow_table && flow_id <= flow_table->mask) {
3613 rflow = &flow_table->flows[flow_id];
3614 cpu = ACCESS_ONCE(rflow->cpu);
3615 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3616 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3617 rflow->last_qtail) <
3618 (int)(10 * flow_table->mask)))
3624 EXPORT_SYMBOL(rps_may_expire_flow);
3626 #endif /* CONFIG_RFS_ACCEL */
3628 /* Called from hardirq (IPI) context */
3629 static void rps_trigger_softirq(void *data)
3631 struct softnet_data *sd = data;
3633 ____napi_schedule(sd, &sd->backlog);
3637 #endif /* CONFIG_RPS */
3640 * Check if this softnet_data structure is another cpu one
3641 * If yes, queue it to our IPI list and return 1
3644 static int rps_ipi_queued(struct softnet_data *sd)
3647 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3650 sd->rps_ipi_next = mysd->rps_ipi_list;
3651 mysd->rps_ipi_list = sd;
3653 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3656 #endif /* CONFIG_RPS */
3660 #ifdef CONFIG_NET_FLOW_LIMIT
3661 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3664 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3666 #ifdef CONFIG_NET_FLOW_LIMIT
3667 struct sd_flow_limit *fl;
3668 struct softnet_data *sd;
3669 unsigned int old_flow, new_flow;
3671 if (qlen < (netdev_max_backlog >> 1))
3674 sd = this_cpu_ptr(&softnet_data);
3677 fl = rcu_dereference(sd->flow_limit);
3679 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3680 old_flow = fl->history[fl->history_head];
3681 fl->history[fl->history_head] = new_flow;
3684 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3686 if (likely(fl->buckets[old_flow]))
3687 fl->buckets[old_flow]--;
3689 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3701 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3702 * queue (may be a remote CPU queue).
3704 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3705 unsigned int *qtail)
3707 struct softnet_data *sd;
3708 unsigned long flags;
3711 sd = &per_cpu(softnet_data, cpu);
3713 local_irq_save(flags);
3716 if (!netif_running(skb->dev))
3718 qlen = skb_queue_len(&sd->input_pkt_queue);
3719 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3722 __skb_queue_tail(&sd->input_pkt_queue, skb);
3723 input_queue_tail_incr_save(sd, qtail);
3725 local_irq_restore(flags);
3726 return NET_RX_SUCCESS;
3729 /* Schedule NAPI for backlog device
3730 * We can use non atomic operation since we own the queue lock
3732 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3733 if (!rps_ipi_queued(sd))
3734 ____napi_schedule(sd, &sd->backlog);
3743 local_irq_restore(flags);
3745 atomic_long_inc(&skb->dev->rx_dropped);
3750 static int netif_rx_internal(struct sk_buff *skb)
3754 net_timestamp_check(netdev_tstamp_prequeue, skb);
3756 trace_netif_rx(skb);
3758 if (static_key_false(&rps_needed)) {
3759 struct rps_dev_flow voidflow, *rflow = &voidflow;
3765 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3767 cpu = smp_processor_id();
3769 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3778 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3785 * netif_rx - post buffer to the network code
3786 * @skb: buffer to post
3788 * This function receives a packet from a device driver and queues it for
3789 * the upper (protocol) levels to process. It always succeeds. The buffer
3790 * may be dropped during processing for congestion control or by the
3794 * NET_RX_SUCCESS (no congestion)
3795 * NET_RX_DROP (packet was dropped)
3799 int netif_rx(struct sk_buff *skb)
3801 trace_netif_rx_entry(skb);
3803 return netif_rx_internal(skb);
3805 EXPORT_SYMBOL(netif_rx);
3807 int netif_rx_ni(struct sk_buff *skb)
3811 trace_netif_rx_ni_entry(skb);
3814 err = netif_rx_internal(skb);
3815 if (local_softirq_pending())
3821 EXPORT_SYMBOL(netif_rx_ni);
3823 static __latent_entropy void net_tx_action(struct softirq_action *h)
3825 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3827 if (sd->completion_queue) {
3828 struct sk_buff *clist;
3830 local_irq_disable();
3831 clist = sd->completion_queue;
3832 sd->completion_queue = NULL;
3836 struct sk_buff *skb = clist;
3838 clist = clist->next;
3840 WARN_ON(atomic_read(&skb->users));
3841 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3842 trace_consume_skb(skb);
3844 trace_kfree_skb(skb, net_tx_action);
3846 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
3849 __kfree_skb_defer(skb);
3852 __kfree_skb_flush();
3855 if (sd->output_queue) {
3858 local_irq_disable();
3859 head = sd->output_queue;
3860 sd->output_queue = NULL;
3861 sd->output_queue_tailp = &sd->output_queue;
3865 struct Qdisc *q = head;
3866 spinlock_t *root_lock;
3868 head = head->next_sched;
3870 root_lock = qdisc_lock(q);
3871 spin_lock(root_lock);
3872 /* We need to make sure head->next_sched is read
3873 * before clearing __QDISC_STATE_SCHED
3875 smp_mb__before_atomic();
3876 clear_bit(__QDISC_STATE_SCHED, &q->state);
3878 spin_unlock(root_lock);
3883 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
3884 /* This hook is defined here for ATM LANE */
3885 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3886 unsigned char *addr) __read_mostly;
3887 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3890 static inline struct sk_buff *
3891 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
3892 struct net_device *orig_dev)
3894 #ifdef CONFIG_NET_CLS_ACT
3895 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3896 struct tcf_result cl_res;
3898 /* If there's at least one ingress present somewhere (so
3899 * we get here via enabled static key), remaining devices
3900 * that are not configured with an ingress qdisc will bail
3906 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3910 qdisc_skb_cb(skb)->pkt_len = skb->len;
3911 skb->tc_at_ingress = 1;
3912 qdisc_bstats_cpu_update(cl->q, skb);
3914 switch (tc_classify(skb, cl, &cl_res, false)) {
3916 case TC_ACT_RECLASSIFY:
3917 skb->tc_index = TC_H_MIN(cl_res.classid);
3920 qdisc_qstats_cpu_drop(cl->q);
3927 case TC_ACT_REDIRECT:
3928 /* skb_mac_header check was done by cls/act_bpf, so
3929 * we can safely push the L2 header back before
3930 * redirecting to another netdev
3932 __skb_push(skb, skb->mac_len);
3933 skb_do_redirect(skb);
3938 #endif /* CONFIG_NET_CLS_ACT */
3943 * netdev_is_rx_handler_busy - check if receive handler is registered
3944 * @dev: device to check
3946 * Check if a receive handler is already registered for a given device.
3947 * Return true if there one.
3949 * The caller must hold the rtnl_mutex.
3951 bool netdev_is_rx_handler_busy(struct net_device *dev)
3954 return dev && rtnl_dereference(dev->rx_handler);
3956 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
3959 * netdev_rx_handler_register - register receive handler
3960 * @dev: device to register a handler for
3961 * @rx_handler: receive handler to register
3962 * @rx_handler_data: data pointer that is used by rx handler
3964 * Register a receive handler for a device. This handler will then be
3965 * called from __netif_receive_skb. A negative errno code is returned
3968 * The caller must hold the rtnl_mutex.
3970 * For a general description of rx_handler, see enum rx_handler_result.
3972 int netdev_rx_handler_register(struct net_device *dev,
3973 rx_handler_func_t *rx_handler,
3974 void *rx_handler_data)
3976 if (netdev_is_rx_handler_busy(dev))
3979 /* Note: rx_handler_data must be set before rx_handler */
3980 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3981 rcu_assign_pointer(dev->rx_handler, rx_handler);
3985 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3988 * netdev_rx_handler_unregister - unregister receive handler
3989 * @dev: device to unregister a handler from
3991 * Unregister a receive handler from a device.
3993 * The caller must hold the rtnl_mutex.
3995 void netdev_rx_handler_unregister(struct net_device *dev)
3999 RCU_INIT_POINTER(dev->rx_handler, NULL);
4000 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4001 * section has a guarantee to see a non NULL rx_handler_data
4005 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4007 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4010 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4011 * the special handling of PFMEMALLOC skbs.
4013 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4015 switch (skb->protocol) {
4016 case htons(ETH_P_ARP):
4017 case htons(ETH_P_IP):
4018 case htons(ETH_P_IPV6):
4019 case htons(ETH_P_8021Q):
4020 case htons(ETH_P_8021AD):
4027 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4028 int *ret, struct net_device *orig_dev)
4030 #ifdef CONFIG_NETFILTER_INGRESS
4031 if (nf_hook_ingress_active(skb)) {
4035 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4040 ingress_retval = nf_hook_ingress(skb);
4042 return ingress_retval;
4044 #endif /* CONFIG_NETFILTER_INGRESS */
4048 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
4050 struct packet_type *ptype, *pt_prev;
4051 rx_handler_func_t *rx_handler;
4052 struct net_device *orig_dev;
4053 bool deliver_exact = false;
4054 int ret = NET_RX_DROP;
4057 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4059 trace_netif_receive_skb(skb);
4061 orig_dev = skb->dev;
4063 skb_reset_network_header(skb);
4064 if (!skb_transport_header_was_set(skb))
4065 skb_reset_transport_header(skb);
4066 skb_reset_mac_len(skb);
4071 skb->skb_iif = skb->dev->ifindex;
4073 __this_cpu_inc(softnet_data.processed);
4075 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4076 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4077 skb = skb_vlan_untag(skb);
4082 if (skb_skip_tc_classify(skb))
4088 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4090 ret = deliver_skb(skb, pt_prev, orig_dev);
4094 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4096 ret = deliver_skb(skb, pt_prev, orig_dev);
4101 #ifdef CONFIG_NET_INGRESS
4102 if (static_key_false(&ingress_needed)) {
4103 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4107 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4113 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4116 if (skb_vlan_tag_present(skb)) {
4118 ret = deliver_skb(skb, pt_prev, orig_dev);
4121 if (vlan_do_receive(&skb))
4123 else if (unlikely(!skb))
4127 rx_handler = rcu_dereference(skb->dev->rx_handler);
4130 ret = deliver_skb(skb, pt_prev, orig_dev);
4133 switch (rx_handler(&skb)) {
4134 case RX_HANDLER_CONSUMED:
4135 ret = NET_RX_SUCCESS;
4137 case RX_HANDLER_ANOTHER:
4139 case RX_HANDLER_EXACT:
4140 deliver_exact = true;
4141 case RX_HANDLER_PASS:
4148 if (unlikely(skb_vlan_tag_present(skb))) {
4149 if (skb_vlan_tag_get_id(skb))
4150 skb->pkt_type = PACKET_OTHERHOST;
4151 /* Note: we might in the future use prio bits
4152 * and set skb->priority like in vlan_do_receive()
4153 * For the time being, just ignore Priority Code Point
4158 type = skb->protocol;
4160 /* deliver only exact match when indicated */
4161 if (likely(!deliver_exact)) {
4162 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4163 &ptype_base[ntohs(type) &
4167 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4168 &orig_dev->ptype_specific);
4170 if (unlikely(skb->dev != orig_dev)) {
4171 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4172 &skb->dev->ptype_specific);
4176 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
4179 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4183 atomic_long_inc(&skb->dev->rx_dropped);
4185 atomic_long_inc(&skb->dev->rx_nohandler);
4187 /* Jamal, now you will not able to escape explaining
4188 * me how you were going to use this. :-)
4197 static int __netif_receive_skb(struct sk_buff *skb)
4201 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4202 unsigned long pflags = current->flags;
4205 * PFMEMALLOC skbs are special, they should
4206 * - be delivered to SOCK_MEMALLOC sockets only
4207 * - stay away from userspace
4208 * - have bounded memory usage
4210 * Use PF_MEMALLOC as this saves us from propagating the allocation
4211 * context down to all allocation sites.
4213 current->flags |= PF_MEMALLOC;
4214 ret = __netif_receive_skb_core(skb, true);
4215 tsk_restore_flags(current, pflags, PF_MEMALLOC);
4217 ret = __netif_receive_skb_core(skb, false);
4222 static int netif_receive_skb_internal(struct sk_buff *skb)
4226 net_timestamp_check(netdev_tstamp_prequeue, skb);
4228 if (skb_defer_rx_timestamp(skb))
4229 return NET_RX_SUCCESS;
4234 if (static_key_false(&rps_needed)) {
4235 struct rps_dev_flow voidflow, *rflow = &voidflow;
4236 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4239 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4245 ret = __netif_receive_skb(skb);
4251 * netif_receive_skb - process receive buffer from network
4252 * @skb: buffer to process
4254 * netif_receive_skb() is the main receive data processing function.
4255 * It always succeeds. The buffer may be dropped during processing
4256 * for congestion control or by the protocol layers.
4258 * This function may only be called from softirq context and interrupts
4259 * should be enabled.
4261 * Return values (usually ignored):
4262 * NET_RX_SUCCESS: no congestion
4263 * NET_RX_DROP: packet was dropped
4265 int netif_receive_skb(struct sk_buff *skb)
4267 trace_netif_receive_skb_entry(skb);
4269 return netif_receive_skb_internal(skb);
4271 EXPORT_SYMBOL(netif_receive_skb);
4273 DEFINE_PER_CPU(struct work_struct, flush_works);
4275 /* Network device is going away, flush any packets still pending */
4276 static void flush_backlog(struct work_struct *work)
4278 struct sk_buff *skb, *tmp;
4279 struct softnet_data *sd;
4282 sd = this_cpu_ptr(&softnet_data);
4284 local_irq_disable();
4286 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4287 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4288 __skb_unlink(skb, &sd->input_pkt_queue);
4290 input_queue_head_incr(sd);
4296 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4297 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4298 __skb_unlink(skb, &sd->process_queue);
4300 input_queue_head_incr(sd);
4306 static void flush_all_backlogs(void)
4312 for_each_online_cpu(cpu)
4313 queue_work_on(cpu, system_highpri_wq,
4314 per_cpu_ptr(&flush_works, cpu));
4316 for_each_online_cpu(cpu)
4317 flush_work(per_cpu_ptr(&flush_works, cpu));
4322 static int napi_gro_complete(struct sk_buff *skb)
4324 struct packet_offload *ptype;
4325 __be16 type = skb->protocol;
4326 struct list_head *head = &offload_base;
4329 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4331 if (NAPI_GRO_CB(skb)->count == 1) {
4332 skb_shinfo(skb)->gso_size = 0;
4337 list_for_each_entry_rcu(ptype, head, list) {
4338 if (ptype->type != type || !ptype->callbacks.gro_complete)
4341 err = ptype->callbacks.gro_complete(skb, 0);
4347 WARN_ON(&ptype->list == head);
4349 return NET_RX_SUCCESS;
4353 return netif_receive_skb_internal(skb);
4356 /* napi->gro_list contains packets ordered by age.
4357 * youngest packets at the head of it.
4358 * Complete skbs in reverse order to reduce latencies.
4360 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4362 struct sk_buff *skb, *prev = NULL;
4364 /* scan list and build reverse chain */
4365 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4370 for (skb = prev; skb; skb = prev) {
4373 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4377 napi_gro_complete(skb);
4381 napi->gro_list = NULL;
4383 EXPORT_SYMBOL(napi_gro_flush);
4385 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4388 unsigned int maclen = skb->dev->hard_header_len;
4389 u32 hash = skb_get_hash_raw(skb);
4391 for (p = napi->gro_list; p; p = p->next) {
4392 unsigned long diffs;
4394 NAPI_GRO_CB(p)->flush = 0;
4396 if (hash != skb_get_hash_raw(p)) {
4397 NAPI_GRO_CB(p)->same_flow = 0;
4401 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4402 diffs |= p->vlan_tci ^ skb->vlan_tci;
4403 diffs |= skb_metadata_dst_cmp(p, skb);
4404 if (maclen == ETH_HLEN)
4405 diffs |= compare_ether_header(skb_mac_header(p),
4406 skb_mac_header(skb));
4408 diffs = memcmp(skb_mac_header(p),
4409 skb_mac_header(skb),
4411 NAPI_GRO_CB(p)->same_flow = !diffs;
4415 static void skb_gro_reset_offset(struct sk_buff *skb)
4417 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4418 const skb_frag_t *frag0 = &pinfo->frags[0];
4420 NAPI_GRO_CB(skb)->data_offset = 0;
4421 NAPI_GRO_CB(skb)->frag0 = NULL;
4422 NAPI_GRO_CB(skb)->frag0_len = 0;
4424 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4426 !PageHighMem(skb_frag_page(frag0))) {
4427 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4428 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
4429 skb_frag_size(frag0),
4430 skb->end - skb->tail);
4434 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4436 struct skb_shared_info *pinfo = skb_shinfo(skb);
4438 BUG_ON(skb->end - skb->tail < grow);
4440 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4442 skb->data_len -= grow;
4445 pinfo->frags[0].page_offset += grow;
4446 skb_frag_size_sub(&pinfo->frags[0], grow);
4448 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4449 skb_frag_unref(skb, 0);
4450 memmove(pinfo->frags, pinfo->frags + 1,
4451 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4455 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4457 struct sk_buff **pp = NULL;
4458 struct packet_offload *ptype;
4459 __be16 type = skb->protocol;
4460 struct list_head *head = &offload_base;
4462 enum gro_result ret;
4465 if (!(skb->dev->features & NETIF_F_GRO))
4471 gro_list_prepare(napi, skb);
4474 list_for_each_entry_rcu(ptype, head, list) {
4475 if (ptype->type != type || !ptype->callbacks.gro_receive)
4478 skb_set_network_header(skb, skb_gro_offset(skb));
4479 skb_reset_mac_len(skb);
4480 NAPI_GRO_CB(skb)->same_flow = 0;
4481 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
4482 NAPI_GRO_CB(skb)->free = 0;
4483 NAPI_GRO_CB(skb)->encap_mark = 0;
4484 NAPI_GRO_CB(skb)->recursion_counter = 0;
4485 NAPI_GRO_CB(skb)->is_fou = 0;
4486 NAPI_GRO_CB(skb)->is_atomic = 1;
4487 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4489 /* Setup for GRO checksum validation */
4490 switch (skb->ip_summed) {
4491 case CHECKSUM_COMPLETE:
4492 NAPI_GRO_CB(skb)->csum = skb->csum;
4493 NAPI_GRO_CB(skb)->csum_valid = 1;
4494 NAPI_GRO_CB(skb)->csum_cnt = 0;
4496 case CHECKSUM_UNNECESSARY:
4497 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4498 NAPI_GRO_CB(skb)->csum_valid = 0;
4501 NAPI_GRO_CB(skb)->csum_cnt = 0;
4502 NAPI_GRO_CB(skb)->csum_valid = 0;
4505 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4510 if (&ptype->list == head)
4513 if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
4518 same_flow = NAPI_GRO_CB(skb)->same_flow;
4519 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4522 struct sk_buff *nskb = *pp;
4526 napi_gro_complete(nskb);
4533 if (NAPI_GRO_CB(skb)->flush)
4536 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4537 struct sk_buff *nskb = napi->gro_list;
4539 /* locate the end of the list to select the 'oldest' flow */
4540 while (nskb->next) {
4546 napi_gro_complete(nskb);
4550 NAPI_GRO_CB(skb)->count = 1;
4551 NAPI_GRO_CB(skb)->age = jiffies;
4552 NAPI_GRO_CB(skb)->last = skb;
4553 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4554 skb->next = napi->gro_list;
4555 napi->gro_list = skb;
4559 grow = skb_gro_offset(skb) - skb_headlen(skb);
4561 gro_pull_from_frag0(skb, grow);
4570 struct packet_offload *gro_find_receive_by_type(__be16 type)
4572 struct list_head *offload_head = &offload_base;
4573 struct packet_offload *ptype;
4575 list_for_each_entry_rcu(ptype, offload_head, list) {
4576 if (ptype->type != type || !ptype->callbacks.gro_receive)
4582 EXPORT_SYMBOL(gro_find_receive_by_type);
4584 struct packet_offload *gro_find_complete_by_type(__be16 type)
4586 struct list_head *offload_head = &offload_base;
4587 struct packet_offload *ptype;
4589 list_for_each_entry_rcu(ptype, offload_head, list) {
4590 if (ptype->type != type || !ptype->callbacks.gro_complete)
4596 EXPORT_SYMBOL(gro_find_complete_by_type);
4598 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4602 if (netif_receive_skb_internal(skb))
4610 case GRO_MERGED_FREE:
4611 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) {
4614 kmem_cache_free(skbuff_head_cache, skb);
4629 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4631 skb_mark_napi_id(skb, napi);
4632 trace_napi_gro_receive_entry(skb);
4634 skb_gro_reset_offset(skb);
4636 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4638 EXPORT_SYMBOL(napi_gro_receive);
4640 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4642 if (unlikely(skb->pfmemalloc)) {
4646 __skb_pull(skb, skb_headlen(skb));
4647 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4648 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4650 skb->dev = napi->dev;
4652 skb->encapsulation = 0;
4653 skb_shinfo(skb)->gso_type = 0;
4654 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4660 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4662 struct sk_buff *skb = napi->skb;
4665 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4668 skb_mark_napi_id(skb, napi);
4673 EXPORT_SYMBOL(napi_get_frags);
4675 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4676 struct sk_buff *skb,
4682 __skb_push(skb, ETH_HLEN);
4683 skb->protocol = eth_type_trans(skb, skb->dev);
4684 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4689 case GRO_MERGED_FREE:
4690 napi_reuse_skb(napi, skb);
4701 /* Upper GRO stack assumes network header starts at gro_offset=0
4702 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4703 * We copy ethernet header into skb->data to have a common layout.
4705 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4707 struct sk_buff *skb = napi->skb;
4708 const struct ethhdr *eth;
4709 unsigned int hlen = sizeof(*eth);
4713 skb_reset_mac_header(skb);
4714 skb_gro_reset_offset(skb);
4716 eth = skb_gro_header_fast(skb, 0);
4717 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4718 eth = skb_gro_header_slow(skb, hlen, 0);
4719 if (unlikely(!eth)) {
4720 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
4721 __func__, napi->dev->name);
4722 napi_reuse_skb(napi, skb);
4726 gro_pull_from_frag0(skb, hlen);
4727 NAPI_GRO_CB(skb)->frag0 += hlen;
4728 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4730 __skb_pull(skb, hlen);
4733 * This works because the only protocols we care about don't require
4735 * We'll fix it up properly in napi_frags_finish()
4737 skb->protocol = eth->h_proto;
4742 gro_result_t napi_gro_frags(struct napi_struct *napi)
4744 struct sk_buff *skb = napi_frags_skb(napi);
4749 trace_napi_gro_frags_entry(skb);
4751 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4753 EXPORT_SYMBOL(napi_gro_frags);
4755 /* Compute the checksum from gro_offset and return the folded value
4756 * after adding in any pseudo checksum.
4758 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4763 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4765 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4766 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4768 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4769 !skb->csum_complete_sw)
4770 netdev_rx_csum_fault(skb->dev);
4773 NAPI_GRO_CB(skb)->csum = wsum;
4774 NAPI_GRO_CB(skb)->csum_valid = 1;
4778 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4781 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4782 * Note: called with local irq disabled, but exits with local irq enabled.
4784 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4787 struct softnet_data *remsd = sd->rps_ipi_list;
4790 sd->rps_ipi_list = NULL;
4794 /* Send pending IPI's to kick RPS processing on remote cpus. */
4796 struct softnet_data *next = remsd->rps_ipi_next;
4798 if (cpu_online(remsd->cpu))
4799 smp_call_function_single_async(remsd->cpu,
4808 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4811 return sd->rps_ipi_list != NULL;
4817 static int process_backlog(struct napi_struct *napi, int quota)
4819 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4823 /* Check if we have pending ipi, its better to send them now,
4824 * not waiting net_rx_action() end.
4826 if (sd_has_rps_ipi_waiting(sd)) {
4827 local_irq_disable();
4828 net_rps_action_and_irq_enable(sd);
4831 napi->weight = dev_rx_weight;
4833 struct sk_buff *skb;
4835 while ((skb = __skb_dequeue(&sd->process_queue))) {
4837 __netif_receive_skb(skb);
4839 input_queue_head_incr(sd);
4840 if (++work >= quota)
4845 local_irq_disable();
4847 if (skb_queue_empty(&sd->input_pkt_queue)) {
4849 * Inline a custom version of __napi_complete().
4850 * only current cpu owns and manipulates this napi,
4851 * and NAPI_STATE_SCHED is the only possible flag set
4853 * We can use a plain write instead of clear_bit(),
4854 * and we dont need an smp_mb() memory barrier.
4859 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4860 &sd->process_queue);
4870 * __napi_schedule - schedule for receive
4871 * @n: entry to schedule
4873 * The entry's receive function will be scheduled to run.
4874 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4876 void __napi_schedule(struct napi_struct *n)
4878 unsigned long flags;
4880 local_irq_save(flags);
4881 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4882 local_irq_restore(flags);
4884 EXPORT_SYMBOL(__napi_schedule);
4887 * napi_schedule_prep - check if napi can be scheduled
4890 * Test if NAPI routine is already running, and if not mark
4891 * it as running. This is used as a condition variable
4892 * insure only one NAPI poll instance runs. We also make
4893 * sure there is no pending NAPI disable.
4895 bool napi_schedule_prep(struct napi_struct *n)
4897 unsigned long val, new;
4900 val = READ_ONCE(n->state);
4901 if (unlikely(val & NAPIF_STATE_DISABLE))
4903 new = val | NAPIF_STATE_SCHED;
4905 /* Sets STATE_MISSED bit if STATE_SCHED was already set
4906 * This was suggested by Alexander Duyck, as compiler
4907 * emits better code than :
4908 * if (val & NAPIF_STATE_SCHED)
4909 * new |= NAPIF_STATE_MISSED;
4911 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
4913 } while (cmpxchg(&n->state, val, new) != val);
4915 return !(val & NAPIF_STATE_SCHED);
4917 EXPORT_SYMBOL(napi_schedule_prep);
4920 * __napi_schedule_irqoff - schedule for receive
4921 * @n: entry to schedule
4923 * Variant of __napi_schedule() assuming hard irqs are masked
4925 void __napi_schedule_irqoff(struct napi_struct *n)
4927 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4929 EXPORT_SYMBOL(__napi_schedule_irqoff);
4931 bool napi_complete_done(struct napi_struct *n, int work_done)
4933 unsigned long flags, val, new;
4936 * 1) Don't let napi dequeue from the cpu poll list
4937 * just in case its running on a different cpu.
4938 * 2) If we are busy polling, do nothing here, we have
4939 * the guarantee we will be called later.
4941 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
4942 NAPIF_STATE_IN_BUSY_POLL)))
4946 unsigned long timeout = 0;
4949 timeout = n->dev->gro_flush_timeout;
4952 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4953 HRTIMER_MODE_REL_PINNED);
4955 napi_gro_flush(n, false);
4957 if (unlikely(!list_empty(&n->poll_list))) {
4958 /* If n->poll_list is not empty, we need to mask irqs */
4959 local_irq_save(flags);
4960 list_del_init(&n->poll_list);
4961 local_irq_restore(flags);
4965 val = READ_ONCE(n->state);
4967 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
4969 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
4971 /* If STATE_MISSED was set, leave STATE_SCHED set,
4972 * because we will call napi->poll() one more time.
4973 * This C code was suggested by Alexander Duyck to help gcc.
4975 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
4977 } while (cmpxchg(&n->state, val, new) != val);
4979 if (unlikely(val & NAPIF_STATE_MISSED)) {
4986 EXPORT_SYMBOL(napi_complete_done);
4988 /* must be called under rcu_read_lock(), as we dont take a reference */
4989 static struct napi_struct *napi_by_id(unsigned int napi_id)
4991 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4992 struct napi_struct *napi;
4994 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4995 if (napi->napi_id == napi_id)
5001 #if defined(CONFIG_NET_RX_BUSY_POLL)
5003 #define BUSY_POLL_BUDGET 8
5005 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
5009 /* Busy polling means there is a high chance device driver hard irq
5010 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
5011 * set in napi_schedule_prep().
5012 * Since we are about to call napi->poll() once more, we can safely
5013 * clear NAPI_STATE_MISSED.
5015 * Note: x86 could use a single "lock and ..." instruction
5016 * to perform these two clear_bit()
5018 clear_bit(NAPI_STATE_MISSED, &napi->state);
5019 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
5023 /* All we really want here is to re-enable device interrupts.
5024 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
5026 rc = napi->poll(napi, BUSY_POLL_BUDGET);
5027 netpoll_poll_unlock(have_poll_lock);
5028 if (rc == BUSY_POLL_BUDGET)
5029 __napi_schedule(napi);
5031 if (local_softirq_pending())
5035 bool sk_busy_loop(struct sock *sk, int nonblock)
5037 unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0;
5038 int (*napi_poll)(struct napi_struct *napi, int budget);
5039 void *have_poll_lock = NULL;
5040 struct napi_struct *napi;
5049 napi = napi_by_id(sk->sk_napi_id);
5058 unsigned long val = READ_ONCE(napi->state);
5060 /* If multiple threads are competing for this napi,
5061 * we avoid dirtying napi->state as much as we can.
5063 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
5064 NAPIF_STATE_IN_BUSY_POLL))
5066 if (cmpxchg(&napi->state, val,
5067 val | NAPIF_STATE_IN_BUSY_POLL |
5068 NAPIF_STATE_SCHED) != val)
5070 have_poll_lock = netpoll_poll_lock(napi);
5071 napi_poll = napi->poll;
5073 rc = napi_poll(napi, BUSY_POLL_BUDGET);
5074 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
5077 __NET_ADD_STATS(sock_net(sk),
5078 LINUX_MIB_BUSYPOLLRXPACKETS, rc);
5081 if (nonblock || !skb_queue_empty(&sk->sk_receive_queue) ||
5082 busy_loop_timeout(end_time))
5085 if (unlikely(need_resched())) {
5087 busy_poll_stop(napi, have_poll_lock);
5091 rc = !skb_queue_empty(&sk->sk_receive_queue);
5092 if (rc || busy_loop_timeout(end_time))
5099 busy_poll_stop(napi, have_poll_lock);
5101 rc = !skb_queue_empty(&sk->sk_receive_queue);
5106 EXPORT_SYMBOL(sk_busy_loop);
5108 #endif /* CONFIG_NET_RX_BUSY_POLL */
5110 static void napi_hash_add(struct napi_struct *napi)
5112 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
5113 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
5116 spin_lock(&napi_hash_lock);
5118 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */
5120 if (unlikely(++napi_gen_id < NR_CPUS + 1))
5121 napi_gen_id = NR_CPUS + 1;
5122 } while (napi_by_id(napi_gen_id));
5123 napi->napi_id = napi_gen_id;
5125 hlist_add_head_rcu(&napi->napi_hash_node,
5126 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
5128 spin_unlock(&napi_hash_lock);
5131 /* Warning : caller is responsible to make sure rcu grace period
5132 * is respected before freeing memory containing @napi
5134 bool napi_hash_del(struct napi_struct *napi)
5136 bool rcu_sync_needed = false;
5138 spin_lock(&napi_hash_lock);
5140 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
5141 rcu_sync_needed = true;
5142 hlist_del_rcu(&napi->napi_hash_node);
5144 spin_unlock(&napi_hash_lock);
5145 return rcu_sync_needed;
5147 EXPORT_SYMBOL_GPL(napi_hash_del);
5149 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
5151 struct napi_struct *napi;
5153 napi = container_of(timer, struct napi_struct, timer);
5155 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
5156 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
5158 if (napi->gro_list && !napi_disable_pending(napi) &&
5159 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
5160 __napi_schedule_irqoff(napi);
5162 return HRTIMER_NORESTART;
5165 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
5166 int (*poll)(struct napi_struct *, int), int weight)
5168 INIT_LIST_HEAD(&napi->poll_list);
5169 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
5170 napi->timer.function = napi_watchdog;
5171 napi->gro_count = 0;
5172 napi->gro_list = NULL;
5175 if (weight > NAPI_POLL_WEIGHT)
5176 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5178 napi->weight = weight;
5179 list_add(&napi->dev_list, &dev->napi_list);
5181 #ifdef CONFIG_NETPOLL
5182 napi->poll_owner = -1;
5184 set_bit(NAPI_STATE_SCHED, &napi->state);
5185 napi_hash_add(napi);
5187 EXPORT_SYMBOL(netif_napi_add);
5189 void napi_disable(struct napi_struct *n)
5192 set_bit(NAPI_STATE_DISABLE, &n->state);
5194 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
5196 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
5199 hrtimer_cancel(&n->timer);
5201 clear_bit(NAPI_STATE_DISABLE, &n->state);
5203 EXPORT_SYMBOL(napi_disable);
5205 /* Must be called in process context */
5206 void netif_napi_del(struct napi_struct *napi)
5209 if (napi_hash_del(napi))
5211 list_del_init(&napi->dev_list);
5212 napi_free_frags(napi);
5214 kfree_skb_list(napi->gro_list);
5215 napi->gro_list = NULL;
5216 napi->gro_count = 0;
5218 EXPORT_SYMBOL(netif_napi_del);
5220 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
5225 list_del_init(&n->poll_list);
5227 have = netpoll_poll_lock(n);
5231 /* This NAPI_STATE_SCHED test is for avoiding a race
5232 * with netpoll's poll_napi(). Only the entity which
5233 * obtains the lock and sees NAPI_STATE_SCHED set will
5234 * actually make the ->poll() call. Therefore we avoid
5235 * accidentally calling ->poll() when NAPI is not scheduled.
5238 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
5239 work = n->poll(n, weight);
5240 trace_napi_poll(n, work, weight);
5243 WARN_ON_ONCE(work > weight);
5245 if (likely(work < weight))
5248 /* Drivers must not modify the NAPI state if they
5249 * consume the entire weight. In such cases this code
5250 * still "owns" the NAPI instance and therefore can
5251 * move the instance around on the list at-will.
5253 if (unlikely(napi_disable_pending(n))) {
5259 /* flush too old packets
5260 * If HZ < 1000, flush all packets.
5262 napi_gro_flush(n, HZ >= 1000);
5265 /* Some drivers may have called napi_schedule
5266 * prior to exhausting their budget.
5268 if (unlikely(!list_empty(&n->poll_list))) {
5269 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5270 n->dev ? n->dev->name : "backlog");
5274 list_add_tail(&n->poll_list, repoll);
5277 netpoll_poll_unlock(have);
5282 static __latent_entropy void net_rx_action(struct softirq_action *h)
5284 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5285 unsigned long time_limit = jiffies + 2;
5286 int budget = netdev_budget;
5290 local_irq_disable();
5291 list_splice_init(&sd->poll_list, &list);
5295 struct napi_struct *n;
5297 if (list_empty(&list)) {
5298 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
5303 n = list_first_entry(&list, struct napi_struct, poll_list);
5304 budget -= napi_poll(n, &repoll);
5306 /* If softirq window is exhausted then punt.
5307 * Allow this to run for 2 jiffies since which will allow
5308 * an average latency of 1.5/HZ.
5310 if (unlikely(budget <= 0 ||
5311 time_after_eq(jiffies, time_limit))) {
5317 local_irq_disable();
5319 list_splice_tail_init(&sd->poll_list, &list);
5320 list_splice_tail(&repoll, &list);
5321 list_splice(&list, &sd->poll_list);
5322 if (!list_empty(&sd->poll_list))
5323 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5325 net_rps_action_and_irq_enable(sd);
5327 __kfree_skb_flush();
5330 struct netdev_adjacent {
5331 struct net_device *dev;
5333 /* upper master flag, there can only be one master device per list */
5336 /* counter for the number of times this device was added to us */
5339 /* private field for the users */
5342 struct list_head list;
5343 struct rcu_head rcu;
5346 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
5347 struct list_head *adj_list)
5349 struct netdev_adjacent *adj;
5351 list_for_each_entry(adj, adj_list, list) {
5352 if (adj->dev == adj_dev)
5358 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
5360 struct net_device *dev = data;
5362 return upper_dev == dev;
5366 * netdev_has_upper_dev - Check if device is linked to an upper device
5368 * @upper_dev: upper device to check
5370 * Find out if a device is linked to specified upper device and return true
5371 * in case it is. Note that this checks only immediate upper device,
5372 * not through a complete stack of devices. The caller must hold the RTNL lock.
5374 bool netdev_has_upper_dev(struct net_device *dev,
5375 struct net_device *upper_dev)
5379 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5382 EXPORT_SYMBOL(netdev_has_upper_dev);
5385 * netdev_has_upper_dev_all - Check if device is linked to an upper device
5387 * @upper_dev: upper device to check
5389 * Find out if a device is linked to specified upper device and return true
5390 * in case it is. Note that this checks the entire upper device chain.
5391 * The caller must hold rcu lock.
5394 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
5395 struct net_device *upper_dev)
5397 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5400 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
5403 * netdev_has_any_upper_dev - Check if device is linked to some device
5406 * Find out if a device is linked to an upper device and return true in case
5407 * it is. The caller must hold the RTNL lock.
5409 static bool netdev_has_any_upper_dev(struct net_device *dev)
5413 return !list_empty(&dev->adj_list.upper);
5417 * netdev_master_upper_dev_get - Get master upper device
5420 * Find a master upper device and return pointer to it or NULL in case
5421 * it's not there. The caller must hold the RTNL lock.
5423 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5425 struct netdev_adjacent *upper;
5429 if (list_empty(&dev->adj_list.upper))
5432 upper = list_first_entry(&dev->adj_list.upper,
5433 struct netdev_adjacent, list);
5434 if (likely(upper->master))
5438 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5441 * netdev_has_any_lower_dev - Check if device is linked to some device
5444 * Find out if a device is linked to a lower device and return true in case
5445 * it is. The caller must hold the RTNL lock.
5447 static bool netdev_has_any_lower_dev(struct net_device *dev)
5451 return !list_empty(&dev->adj_list.lower);
5454 void *netdev_adjacent_get_private(struct list_head *adj_list)
5456 struct netdev_adjacent *adj;
5458 adj = list_entry(adj_list, struct netdev_adjacent, list);
5460 return adj->private;
5462 EXPORT_SYMBOL(netdev_adjacent_get_private);
5465 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5467 * @iter: list_head ** of the current position
5469 * Gets the next device from the dev's upper list, starting from iter
5470 * position. The caller must hold RCU read lock.
5472 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5473 struct list_head **iter)
5475 struct netdev_adjacent *upper;
5477 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5479 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5481 if (&upper->list == &dev->adj_list.upper)
5484 *iter = &upper->list;
5488 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5490 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
5491 struct list_head **iter)
5493 struct netdev_adjacent *upper;
5495 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5497 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5499 if (&upper->list == &dev->adj_list.upper)
5502 *iter = &upper->list;
5507 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
5508 int (*fn)(struct net_device *dev,
5512 struct net_device *udev;
5513 struct list_head *iter;
5516 for (iter = &dev->adj_list.upper,
5517 udev = netdev_next_upper_dev_rcu(dev, &iter);
5519 udev = netdev_next_upper_dev_rcu(dev, &iter)) {
5520 /* first is the upper device itself */
5521 ret = fn(udev, data);
5525 /* then look at all of its upper devices */
5526 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data);
5533 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
5536 * netdev_lower_get_next_private - Get the next ->private from the
5537 * lower neighbour list
5539 * @iter: list_head ** of the current position
5541 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5542 * list, starting from iter position. The caller must hold either hold the
5543 * RTNL lock or its own locking that guarantees that the neighbour lower
5544 * list will remain unchanged.
5546 void *netdev_lower_get_next_private(struct net_device *dev,
5547 struct list_head **iter)
5549 struct netdev_adjacent *lower;
5551 lower = list_entry(*iter, struct netdev_adjacent, list);
5553 if (&lower->list == &dev->adj_list.lower)
5556 *iter = lower->list.next;
5558 return lower->private;
5560 EXPORT_SYMBOL(netdev_lower_get_next_private);
5563 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5564 * lower neighbour list, RCU
5567 * @iter: list_head ** of the current position
5569 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5570 * list, starting from iter position. The caller must hold RCU read lock.
5572 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5573 struct list_head **iter)
5575 struct netdev_adjacent *lower;
5577 WARN_ON_ONCE(!rcu_read_lock_held());
5579 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5581 if (&lower->list == &dev->adj_list.lower)
5584 *iter = &lower->list;
5586 return lower->private;
5588 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5591 * netdev_lower_get_next - Get the next device from the lower neighbour
5594 * @iter: list_head ** of the current position
5596 * Gets the next netdev_adjacent from the dev's lower neighbour
5597 * list, starting from iter position. The caller must hold RTNL lock or
5598 * its own locking that guarantees that the neighbour lower
5599 * list will remain unchanged.
5601 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5603 struct netdev_adjacent *lower;
5605 lower = list_entry(*iter, struct netdev_adjacent, list);
5607 if (&lower->list == &dev->adj_list.lower)
5610 *iter = lower->list.next;
5614 EXPORT_SYMBOL(netdev_lower_get_next);
5616 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
5617 struct list_head **iter)
5619 struct netdev_adjacent *lower;
5621 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5623 if (&lower->list == &dev->adj_list.lower)
5626 *iter = &lower->list;
5631 int netdev_walk_all_lower_dev(struct net_device *dev,
5632 int (*fn)(struct net_device *dev,
5636 struct net_device *ldev;
5637 struct list_head *iter;
5640 for (iter = &dev->adj_list.lower,
5641 ldev = netdev_next_lower_dev(dev, &iter);
5643 ldev = netdev_next_lower_dev(dev, &iter)) {
5644 /* first is the lower device itself */
5645 ret = fn(ldev, data);
5649 /* then look at all of its lower devices */
5650 ret = netdev_walk_all_lower_dev(ldev, fn, data);
5657 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
5659 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
5660 struct list_head **iter)
5662 struct netdev_adjacent *lower;
5664 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5665 if (&lower->list == &dev->adj_list.lower)
5668 *iter = &lower->list;
5673 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
5674 int (*fn)(struct net_device *dev,
5678 struct net_device *ldev;
5679 struct list_head *iter;
5682 for (iter = &dev->adj_list.lower,
5683 ldev = netdev_next_lower_dev_rcu(dev, &iter);
5685 ldev = netdev_next_lower_dev_rcu(dev, &iter)) {
5686 /* first is the lower device itself */
5687 ret = fn(ldev, data);
5691 /* then look at all of its lower devices */
5692 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data);
5699 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
5702 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5703 * lower neighbour list, RCU
5707 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5708 * list. The caller must hold RCU read lock.
5710 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5712 struct netdev_adjacent *lower;
5714 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5715 struct netdev_adjacent, list);
5717 return lower->private;
5720 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5723 * netdev_master_upper_dev_get_rcu - Get master upper device
5726 * Find a master upper device and return pointer to it or NULL in case
5727 * it's not there. The caller must hold the RCU read lock.
5729 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5731 struct netdev_adjacent *upper;
5733 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5734 struct netdev_adjacent, list);
5735 if (upper && likely(upper->master))
5739 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5741 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5742 struct net_device *adj_dev,
5743 struct list_head *dev_list)
5745 char linkname[IFNAMSIZ+7];
5747 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5748 "upper_%s" : "lower_%s", adj_dev->name);
5749 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5752 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5754 struct list_head *dev_list)
5756 char linkname[IFNAMSIZ+7];
5758 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5759 "upper_%s" : "lower_%s", name);
5760 sysfs_remove_link(&(dev->dev.kobj), linkname);
5763 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5764 struct net_device *adj_dev,
5765 struct list_head *dev_list)
5767 return (dev_list == &dev->adj_list.upper ||
5768 dev_list == &dev->adj_list.lower) &&
5769 net_eq(dev_net(dev), dev_net(adj_dev));
5772 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5773 struct net_device *adj_dev,
5774 struct list_head *dev_list,
5775 void *private, bool master)
5777 struct netdev_adjacent *adj;
5780 adj = __netdev_find_adj(adj_dev, dev_list);
5784 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
5785 dev->name, adj_dev->name, adj->ref_nr);
5790 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5795 adj->master = master;
5797 adj->private = private;
5800 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
5801 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
5803 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5804 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5809 /* Ensure that master link is always the first item in list. */
5811 ret = sysfs_create_link(&(dev->dev.kobj),
5812 &(adj_dev->dev.kobj), "master");
5814 goto remove_symlinks;
5816 list_add_rcu(&adj->list, dev_list);
5818 list_add_tail_rcu(&adj->list, dev_list);
5824 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5825 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5833 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5834 struct net_device *adj_dev,
5836 struct list_head *dev_list)
5838 struct netdev_adjacent *adj;
5840 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
5841 dev->name, adj_dev->name, ref_nr);
5843 adj = __netdev_find_adj(adj_dev, dev_list);
5846 pr_err("Adjacency does not exist for device %s from %s\n",
5847 dev->name, adj_dev->name);
5852 if (adj->ref_nr > ref_nr) {
5853 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
5854 dev->name, adj_dev->name, ref_nr,
5855 adj->ref_nr - ref_nr);
5856 adj->ref_nr -= ref_nr;
5861 sysfs_remove_link(&(dev->dev.kobj), "master");
5863 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5864 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5866 list_del_rcu(&adj->list);
5867 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
5868 adj_dev->name, dev->name, adj_dev->name);
5870 kfree_rcu(adj, rcu);
5873 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5874 struct net_device *upper_dev,
5875 struct list_head *up_list,
5876 struct list_head *down_list,
5877 void *private, bool master)
5881 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
5886 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
5889 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
5896 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5897 struct net_device *upper_dev,
5899 struct list_head *up_list,
5900 struct list_head *down_list)
5902 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
5903 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
5906 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5907 struct net_device *upper_dev,
5908 void *private, bool master)
5910 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5911 &dev->adj_list.upper,
5912 &upper_dev->adj_list.lower,
5916 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5917 struct net_device *upper_dev)
5919 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
5920 &dev->adj_list.upper,
5921 &upper_dev->adj_list.lower);
5924 static int __netdev_upper_dev_link(struct net_device *dev,
5925 struct net_device *upper_dev, bool master,
5926 void *upper_priv, void *upper_info)
5928 struct netdev_notifier_changeupper_info changeupper_info;
5933 if (dev == upper_dev)
5936 /* To prevent loops, check if dev is not upper device to upper_dev. */
5937 if (netdev_has_upper_dev(upper_dev, dev))
5940 if (netdev_has_upper_dev(dev, upper_dev))
5943 if (master && netdev_master_upper_dev_get(dev))
5946 changeupper_info.upper_dev = upper_dev;
5947 changeupper_info.master = master;
5948 changeupper_info.linking = true;
5949 changeupper_info.upper_info = upper_info;
5951 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5952 &changeupper_info.info);
5953 ret = notifier_to_errno(ret);
5957 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
5962 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5963 &changeupper_info.info);
5964 ret = notifier_to_errno(ret);
5971 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5977 * netdev_upper_dev_link - Add a link to the upper device
5979 * @upper_dev: new upper device
5981 * Adds a link to device which is upper to this one. The caller must hold
5982 * the RTNL lock. On a failure a negative errno code is returned.
5983 * On success the reference counts are adjusted and the function
5986 int netdev_upper_dev_link(struct net_device *dev,
5987 struct net_device *upper_dev)
5989 return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL);
5991 EXPORT_SYMBOL(netdev_upper_dev_link);
5994 * netdev_master_upper_dev_link - Add a master link to the upper device
5996 * @upper_dev: new upper device
5997 * @upper_priv: upper device private
5998 * @upper_info: upper info to be passed down via notifier
6000 * Adds a link to device which is upper to this one. In this case, only
6001 * one master upper device can be linked, although other non-master devices
6002 * might be linked as well. The caller must hold the RTNL lock.
6003 * On a failure a negative errno code is returned. On success the reference
6004 * counts are adjusted and the function returns zero.
6006 int netdev_master_upper_dev_link(struct net_device *dev,
6007 struct net_device *upper_dev,
6008 void *upper_priv, void *upper_info)
6010 return __netdev_upper_dev_link(dev, upper_dev, true,
6011 upper_priv, upper_info);
6013 EXPORT_SYMBOL(netdev_master_upper_dev_link);
6016 * netdev_upper_dev_unlink - Removes a link to upper device
6018 * @upper_dev: new upper device
6020 * Removes a link to device which is upper to this one. The caller must hold
6023 void netdev_upper_dev_unlink(struct net_device *dev,
6024 struct net_device *upper_dev)
6026 struct netdev_notifier_changeupper_info changeupper_info;
6030 changeupper_info.upper_dev = upper_dev;
6031 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
6032 changeupper_info.linking = false;
6034 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
6035 &changeupper_info.info);
6037 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
6039 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
6040 &changeupper_info.info);
6042 EXPORT_SYMBOL(netdev_upper_dev_unlink);
6045 * netdev_bonding_info_change - Dispatch event about slave change
6047 * @bonding_info: info to dispatch
6049 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
6050 * The caller must hold the RTNL lock.
6052 void netdev_bonding_info_change(struct net_device *dev,
6053 struct netdev_bonding_info *bonding_info)
6055 struct netdev_notifier_bonding_info info;
6057 memcpy(&info.bonding_info, bonding_info,
6058 sizeof(struct netdev_bonding_info));
6059 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
6062 EXPORT_SYMBOL(netdev_bonding_info_change);
6064 static void netdev_adjacent_add_links(struct net_device *dev)
6066 struct netdev_adjacent *iter;
6068 struct net *net = dev_net(dev);
6070 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6071 if (!net_eq(net, dev_net(iter->dev)))
6073 netdev_adjacent_sysfs_add(iter->dev, dev,
6074 &iter->dev->adj_list.lower);
6075 netdev_adjacent_sysfs_add(dev, iter->dev,
6076 &dev->adj_list.upper);
6079 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6080 if (!net_eq(net, dev_net(iter->dev)))
6082 netdev_adjacent_sysfs_add(iter->dev, dev,
6083 &iter->dev->adj_list.upper);
6084 netdev_adjacent_sysfs_add(dev, iter->dev,
6085 &dev->adj_list.lower);
6089 static void netdev_adjacent_del_links(struct net_device *dev)
6091 struct netdev_adjacent *iter;
6093 struct net *net = dev_net(dev);
6095 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6096 if (!net_eq(net, dev_net(iter->dev)))
6098 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6099 &iter->dev->adj_list.lower);
6100 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6101 &dev->adj_list.upper);
6104 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6105 if (!net_eq(net, dev_net(iter->dev)))
6107 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6108 &iter->dev->adj_list.upper);
6109 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6110 &dev->adj_list.lower);
6114 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
6116 struct netdev_adjacent *iter;
6118 struct net *net = dev_net(dev);
6120 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6121 if (!net_eq(net, dev_net(iter->dev)))
6123 netdev_adjacent_sysfs_del(iter->dev, oldname,
6124 &iter->dev->adj_list.lower);
6125 netdev_adjacent_sysfs_add(iter->dev, dev,
6126 &iter->dev->adj_list.lower);
6129 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6130 if (!net_eq(net, dev_net(iter->dev)))
6132 netdev_adjacent_sysfs_del(iter->dev, oldname,
6133 &iter->dev->adj_list.upper);
6134 netdev_adjacent_sysfs_add(iter->dev, dev,
6135 &iter->dev->adj_list.upper);
6139 void *netdev_lower_dev_get_private(struct net_device *dev,
6140 struct net_device *lower_dev)
6142 struct netdev_adjacent *lower;
6146 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
6150 return lower->private;
6152 EXPORT_SYMBOL(netdev_lower_dev_get_private);
6155 int dev_get_nest_level(struct net_device *dev)
6157 struct net_device *lower = NULL;
6158 struct list_head *iter;
6164 netdev_for_each_lower_dev(dev, lower, iter) {
6165 nest = dev_get_nest_level(lower);
6166 if (max_nest < nest)
6170 return max_nest + 1;
6172 EXPORT_SYMBOL(dev_get_nest_level);
6175 * netdev_lower_change - Dispatch event about lower device state change
6176 * @lower_dev: device
6177 * @lower_state_info: state to dispatch
6179 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
6180 * The caller must hold the RTNL lock.
6182 void netdev_lower_state_changed(struct net_device *lower_dev,
6183 void *lower_state_info)
6185 struct netdev_notifier_changelowerstate_info changelowerstate_info;
6188 changelowerstate_info.lower_state_info = lower_state_info;
6189 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev,
6190 &changelowerstate_info.info);
6192 EXPORT_SYMBOL(netdev_lower_state_changed);
6194 static void dev_change_rx_flags(struct net_device *dev, int flags)
6196 const struct net_device_ops *ops = dev->netdev_ops;
6198 if (ops->ndo_change_rx_flags)
6199 ops->ndo_change_rx_flags(dev, flags);
6202 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
6204 unsigned int old_flags = dev->flags;
6210 dev->flags |= IFF_PROMISC;
6211 dev->promiscuity += inc;
6212 if (dev->promiscuity == 0) {
6215 * If inc causes overflow, untouch promisc and return error.
6218 dev->flags &= ~IFF_PROMISC;
6220 dev->promiscuity -= inc;
6221 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6226 if (dev->flags != old_flags) {
6227 pr_info("device %s %s promiscuous mode\n",
6229 dev->flags & IFF_PROMISC ? "entered" : "left");
6230 if (audit_enabled) {
6231 current_uid_gid(&uid, &gid);
6232 audit_log(current->audit_context, GFP_ATOMIC,
6233 AUDIT_ANOM_PROMISCUOUS,
6234 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6235 dev->name, (dev->flags & IFF_PROMISC),
6236 (old_flags & IFF_PROMISC),
6237 from_kuid(&init_user_ns, audit_get_loginuid(current)),
6238 from_kuid(&init_user_ns, uid),
6239 from_kgid(&init_user_ns, gid),
6240 audit_get_sessionid(current));
6243 dev_change_rx_flags(dev, IFF_PROMISC);
6246 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
6251 * dev_set_promiscuity - update promiscuity count on a device
6255 * Add or remove promiscuity from a device. While the count in the device
6256 * remains above zero the interface remains promiscuous. Once it hits zero
6257 * the device reverts back to normal filtering operation. A negative inc
6258 * value is used to drop promiscuity on the device.
6259 * Return 0 if successful or a negative errno code on error.
6261 int dev_set_promiscuity(struct net_device *dev, int inc)
6263 unsigned int old_flags = dev->flags;
6266 err = __dev_set_promiscuity(dev, inc, true);
6269 if (dev->flags != old_flags)
6270 dev_set_rx_mode(dev);
6273 EXPORT_SYMBOL(dev_set_promiscuity);
6275 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
6277 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
6281 dev->flags |= IFF_ALLMULTI;
6282 dev->allmulti += inc;
6283 if (dev->allmulti == 0) {
6286 * If inc causes overflow, untouch allmulti and return error.
6289 dev->flags &= ~IFF_ALLMULTI;
6291 dev->allmulti -= inc;
6292 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6297 if (dev->flags ^ old_flags) {
6298 dev_change_rx_flags(dev, IFF_ALLMULTI);
6299 dev_set_rx_mode(dev);
6301 __dev_notify_flags(dev, old_flags,
6302 dev->gflags ^ old_gflags);
6308 * dev_set_allmulti - update allmulti count on a device
6312 * Add or remove reception of all multicast frames to a device. While the
6313 * count in the device remains above zero the interface remains listening
6314 * to all interfaces. Once it hits zero the device reverts back to normal
6315 * filtering operation. A negative @inc value is used to drop the counter
6316 * when releasing a resource needing all multicasts.
6317 * Return 0 if successful or a negative errno code on error.
6320 int dev_set_allmulti(struct net_device *dev, int inc)
6322 return __dev_set_allmulti(dev, inc, true);
6324 EXPORT_SYMBOL(dev_set_allmulti);
6327 * Upload unicast and multicast address lists to device and
6328 * configure RX filtering. When the device doesn't support unicast
6329 * filtering it is put in promiscuous mode while unicast addresses
6332 void __dev_set_rx_mode(struct net_device *dev)
6334 const struct net_device_ops *ops = dev->netdev_ops;
6336 /* dev_open will call this function so the list will stay sane. */
6337 if (!(dev->flags&IFF_UP))
6340 if (!netif_device_present(dev))
6343 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
6344 /* Unicast addresses changes may only happen under the rtnl,
6345 * therefore calling __dev_set_promiscuity here is safe.
6347 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
6348 __dev_set_promiscuity(dev, 1, false);
6349 dev->uc_promisc = true;
6350 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
6351 __dev_set_promiscuity(dev, -1, false);
6352 dev->uc_promisc = false;
6356 if (ops->ndo_set_rx_mode)
6357 ops->ndo_set_rx_mode(dev);
6360 void dev_set_rx_mode(struct net_device *dev)
6362 netif_addr_lock_bh(dev);
6363 __dev_set_rx_mode(dev);
6364 netif_addr_unlock_bh(dev);
6368 * dev_get_flags - get flags reported to userspace
6371 * Get the combination of flag bits exported through APIs to userspace.
6373 unsigned int dev_get_flags(const struct net_device *dev)
6377 flags = (dev->flags & ~(IFF_PROMISC |
6382 (dev->gflags & (IFF_PROMISC |
6385 if (netif_running(dev)) {
6386 if (netif_oper_up(dev))
6387 flags |= IFF_RUNNING;
6388 if (netif_carrier_ok(dev))
6389 flags |= IFF_LOWER_UP;
6390 if (netif_dormant(dev))
6391 flags |= IFF_DORMANT;
6396 EXPORT_SYMBOL(dev_get_flags);
6398 int __dev_change_flags(struct net_device *dev, unsigned int flags)
6400 unsigned int old_flags = dev->flags;
6406 * Set the flags on our device.
6409 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
6410 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
6412 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6416 * Load in the correct multicast list now the flags have changed.
6419 if ((old_flags ^ flags) & IFF_MULTICAST)
6420 dev_change_rx_flags(dev, IFF_MULTICAST);
6422 dev_set_rx_mode(dev);
6425 * Have we downed the interface. We handle IFF_UP ourselves
6426 * according to user attempts to set it, rather than blindly
6431 if ((old_flags ^ flags) & IFF_UP)
6432 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
6434 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6435 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6436 unsigned int old_flags = dev->flags;
6438 dev->gflags ^= IFF_PROMISC;
6440 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6441 if (dev->flags != old_flags)
6442 dev_set_rx_mode(dev);
6445 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6446 * is important. Some (broken) drivers set IFF_PROMISC, when
6447 * IFF_ALLMULTI is requested not asking us and not reporting.
6449 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6450 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6452 dev->gflags ^= IFF_ALLMULTI;
6453 __dev_set_allmulti(dev, inc, false);
6459 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6460 unsigned int gchanges)
6462 unsigned int changes = dev->flags ^ old_flags;
6465 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6467 if (changes & IFF_UP) {
6468 if (dev->flags & IFF_UP)
6469 call_netdevice_notifiers(NETDEV_UP, dev);
6471 call_netdevice_notifiers(NETDEV_DOWN, dev);
6474 if (dev->flags & IFF_UP &&
6475 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6476 struct netdev_notifier_change_info change_info;
6478 change_info.flags_changed = changes;
6479 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6485 * dev_change_flags - change device settings
6487 * @flags: device state flags
6489 * Change settings on device based state flags. The flags are
6490 * in the userspace exported format.
6492 int dev_change_flags(struct net_device *dev, unsigned int flags)
6495 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6497 ret = __dev_change_flags(dev, flags);
6501 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6502 __dev_notify_flags(dev, old_flags, changes);
6505 EXPORT_SYMBOL(dev_change_flags);
6507 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6509 const struct net_device_ops *ops = dev->netdev_ops;
6511 if (ops->ndo_change_mtu)
6512 return ops->ndo_change_mtu(dev, new_mtu);
6519 * dev_set_mtu - Change maximum transfer unit
6521 * @new_mtu: new transfer unit
6523 * Change the maximum transfer size of the network device.
6525 int dev_set_mtu(struct net_device *dev, int new_mtu)
6529 if (new_mtu == dev->mtu)
6532 /* MTU must be positive, and in range */
6533 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
6534 net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n",
6535 dev->name, new_mtu, dev->min_mtu);
6539 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
6540 net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n",
6541 dev->name, new_mtu, dev->max_mtu);
6545 if (!netif_device_present(dev))
6548 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6549 err = notifier_to_errno(err);
6553 orig_mtu = dev->mtu;
6554 err = __dev_set_mtu(dev, new_mtu);
6557 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6558 err = notifier_to_errno(err);
6560 /* setting mtu back and notifying everyone again,
6561 * so that they have a chance to revert changes.
6563 __dev_set_mtu(dev, orig_mtu);
6564 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6569 EXPORT_SYMBOL(dev_set_mtu);
6572 * dev_set_group - Change group this device belongs to
6574 * @new_group: group this device should belong to
6576 void dev_set_group(struct net_device *dev, int new_group)
6578 dev->group = new_group;
6580 EXPORT_SYMBOL(dev_set_group);
6583 * dev_set_mac_address - Change Media Access Control Address
6587 * Change the hardware (MAC) address of the device
6589 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6591 const struct net_device_ops *ops = dev->netdev_ops;
6594 if (!ops->ndo_set_mac_address)
6596 if (sa->sa_family != dev->type)
6598 if (!netif_device_present(dev))
6600 err = ops->ndo_set_mac_address(dev, sa);
6603 dev->addr_assign_type = NET_ADDR_SET;
6604 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6605 add_device_randomness(dev->dev_addr, dev->addr_len);
6608 EXPORT_SYMBOL(dev_set_mac_address);
6611 * dev_change_carrier - Change device carrier
6613 * @new_carrier: new value
6615 * Change device carrier
6617 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6619 const struct net_device_ops *ops = dev->netdev_ops;
6621 if (!ops->ndo_change_carrier)
6623 if (!netif_device_present(dev))
6625 return ops->ndo_change_carrier(dev, new_carrier);
6627 EXPORT_SYMBOL(dev_change_carrier);
6630 * dev_get_phys_port_id - Get device physical port ID
6634 * Get device physical port ID
6636 int dev_get_phys_port_id(struct net_device *dev,
6637 struct netdev_phys_item_id *ppid)
6639 const struct net_device_ops *ops = dev->netdev_ops;
6641 if (!ops->ndo_get_phys_port_id)
6643 return ops->ndo_get_phys_port_id(dev, ppid);
6645 EXPORT_SYMBOL(dev_get_phys_port_id);
6648 * dev_get_phys_port_name - Get device physical port name
6651 * @len: limit of bytes to copy to name
6653 * Get device physical port name
6655 int dev_get_phys_port_name(struct net_device *dev,
6656 char *name, size_t len)
6658 const struct net_device_ops *ops = dev->netdev_ops;
6660 if (!ops->ndo_get_phys_port_name)
6662 return ops->ndo_get_phys_port_name(dev, name, len);
6664 EXPORT_SYMBOL(dev_get_phys_port_name);
6667 * dev_change_proto_down - update protocol port state information
6669 * @proto_down: new value
6671 * This info can be used by switch drivers to set the phys state of the
6674 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6676 const struct net_device_ops *ops = dev->netdev_ops;
6678 if (!ops->ndo_change_proto_down)
6680 if (!netif_device_present(dev))
6682 return ops->ndo_change_proto_down(dev, proto_down);
6684 EXPORT_SYMBOL(dev_change_proto_down);
6687 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
6689 * @fd: new program fd or negative value to clear
6690 * @flags: xdp-related flags
6692 * Set or clear a bpf program for a device
6694 int dev_change_xdp_fd(struct net_device *dev, int fd, u32 flags)
6696 const struct net_device_ops *ops = dev->netdev_ops;
6697 struct bpf_prog *prog = NULL;
6698 struct netdev_xdp xdp;
6706 if (flags & XDP_FLAGS_UPDATE_IF_NOEXIST) {
6707 memset(&xdp, 0, sizeof(xdp));
6708 xdp.command = XDP_QUERY_PROG;
6710 err = ops->ndo_xdp(dev, &xdp);
6713 if (xdp.prog_attached)
6717 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP);
6719 return PTR_ERR(prog);
6722 memset(&xdp, 0, sizeof(xdp));
6723 xdp.command = XDP_SETUP_PROG;
6726 err = ops->ndo_xdp(dev, &xdp);
6727 if (err < 0 && prog)
6732 EXPORT_SYMBOL(dev_change_xdp_fd);
6735 * dev_new_index - allocate an ifindex
6736 * @net: the applicable net namespace
6738 * Returns a suitable unique value for a new device interface
6739 * number. The caller must hold the rtnl semaphore or the
6740 * dev_base_lock to be sure it remains unique.
6742 static int dev_new_index(struct net *net)
6744 int ifindex = net->ifindex;
6749 if (!__dev_get_by_index(net, ifindex))
6750 return net->ifindex = ifindex;
6754 /* Delayed registration/unregisteration */
6755 static LIST_HEAD(net_todo_list);
6756 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6758 static void net_set_todo(struct net_device *dev)
6760 list_add_tail(&dev->todo_list, &net_todo_list);
6761 dev_net(dev)->dev_unreg_count++;
6764 static void rollback_registered_many(struct list_head *head)
6766 struct net_device *dev, *tmp;
6767 LIST_HEAD(close_head);
6769 BUG_ON(dev_boot_phase);
6772 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6773 /* Some devices call without registering
6774 * for initialization unwind. Remove those
6775 * devices and proceed with the remaining.
6777 if (dev->reg_state == NETREG_UNINITIALIZED) {
6778 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6782 list_del(&dev->unreg_list);
6785 dev->dismantle = true;
6786 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6789 /* If device is running, close it first. */
6790 list_for_each_entry(dev, head, unreg_list)
6791 list_add_tail(&dev->close_list, &close_head);
6792 dev_close_many(&close_head, true);
6794 list_for_each_entry(dev, head, unreg_list) {
6795 /* And unlink it from device chain. */
6796 unlist_netdevice(dev);
6798 dev->reg_state = NETREG_UNREGISTERING;
6800 flush_all_backlogs();
6804 list_for_each_entry(dev, head, unreg_list) {
6805 struct sk_buff *skb = NULL;
6807 /* Shutdown queueing discipline. */
6811 /* Notify protocols, that we are about to destroy
6812 * this device. They should clean all the things.
6814 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6816 if (!dev->rtnl_link_ops ||
6817 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6818 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6822 * Flush the unicast and multicast chains
6827 if (dev->netdev_ops->ndo_uninit)
6828 dev->netdev_ops->ndo_uninit(dev);
6831 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6833 /* Notifier chain MUST detach us all upper devices. */
6834 WARN_ON(netdev_has_any_upper_dev(dev));
6835 WARN_ON(netdev_has_any_lower_dev(dev));
6837 /* Remove entries from kobject tree */
6838 netdev_unregister_kobject(dev);
6840 /* Remove XPS queueing entries */
6841 netif_reset_xps_queues_gt(dev, 0);
6847 list_for_each_entry(dev, head, unreg_list)
6851 static void rollback_registered(struct net_device *dev)
6855 list_add(&dev->unreg_list, &single);
6856 rollback_registered_many(&single);
6860 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6861 struct net_device *upper, netdev_features_t features)
6863 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6864 netdev_features_t feature;
6867 for_each_netdev_feature(&upper_disables, feature_bit) {
6868 feature = __NETIF_F_BIT(feature_bit);
6869 if (!(upper->wanted_features & feature)
6870 && (features & feature)) {
6871 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6872 &feature, upper->name);
6873 features &= ~feature;
6880 static void netdev_sync_lower_features(struct net_device *upper,
6881 struct net_device *lower, netdev_features_t features)
6883 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6884 netdev_features_t feature;
6887 for_each_netdev_feature(&upper_disables, feature_bit) {
6888 feature = __NETIF_F_BIT(feature_bit);
6889 if (!(features & feature) && (lower->features & feature)) {
6890 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6891 &feature, lower->name);
6892 lower->wanted_features &= ~feature;
6893 netdev_update_features(lower);
6895 if (unlikely(lower->features & feature))
6896 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6897 &feature, lower->name);
6902 static netdev_features_t netdev_fix_features(struct net_device *dev,
6903 netdev_features_t features)
6905 /* Fix illegal checksum combinations */
6906 if ((features & NETIF_F_HW_CSUM) &&
6907 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6908 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6909 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6912 /* TSO requires that SG is present as well. */
6913 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6914 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6915 features &= ~NETIF_F_ALL_TSO;
6918 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6919 !(features & NETIF_F_IP_CSUM)) {
6920 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6921 features &= ~NETIF_F_TSO;
6922 features &= ~NETIF_F_TSO_ECN;
6925 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6926 !(features & NETIF_F_IPV6_CSUM)) {
6927 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6928 features &= ~NETIF_F_TSO6;
6931 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
6932 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
6933 features &= ~NETIF_F_TSO_MANGLEID;
6935 /* TSO ECN requires that TSO is present as well. */
6936 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6937 features &= ~NETIF_F_TSO_ECN;
6939 /* Software GSO depends on SG. */
6940 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6941 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6942 features &= ~NETIF_F_GSO;
6945 /* UFO needs SG and checksumming */
6946 if (features & NETIF_F_UFO) {
6947 /* maybe split UFO into V4 and V6? */
6948 if (!(features & NETIF_F_HW_CSUM) &&
6949 ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) !=
6950 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) {
6952 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6953 features &= ~NETIF_F_UFO;
6956 if (!(features & NETIF_F_SG)) {
6958 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6959 features &= ~NETIF_F_UFO;
6963 /* GSO partial features require GSO partial be set */
6964 if ((features & dev->gso_partial_features) &&
6965 !(features & NETIF_F_GSO_PARTIAL)) {
6967 "Dropping partially supported GSO features since no GSO partial.\n");
6968 features &= ~dev->gso_partial_features;
6974 int __netdev_update_features(struct net_device *dev)
6976 struct net_device *upper, *lower;
6977 netdev_features_t features;
6978 struct list_head *iter;
6983 features = netdev_get_wanted_features(dev);
6985 if (dev->netdev_ops->ndo_fix_features)
6986 features = dev->netdev_ops->ndo_fix_features(dev, features);
6988 /* driver might be less strict about feature dependencies */
6989 features = netdev_fix_features(dev, features);
6991 /* some features can't be enabled if they're off an an upper device */
6992 netdev_for_each_upper_dev_rcu(dev, upper, iter)
6993 features = netdev_sync_upper_features(dev, upper, features);
6995 if (dev->features == features)
6998 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6999 &dev->features, &features);
7001 if (dev->netdev_ops->ndo_set_features)
7002 err = dev->netdev_ops->ndo_set_features(dev, features);
7006 if (unlikely(err < 0)) {
7008 "set_features() failed (%d); wanted %pNF, left %pNF\n",
7009 err, &features, &dev->features);
7010 /* return non-0 since some features might have changed and
7011 * it's better to fire a spurious notification than miss it
7017 /* some features must be disabled on lower devices when disabled
7018 * on an upper device (think: bonding master or bridge)
7020 netdev_for_each_lower_dev(dev, lower, iter)
7021 netdev_sync_lower_features(dev, lower, features);
7024 dev->features = features;
7026 return err < 0 ? 0 : 1;
7030 * netdev_update_features - recalculate device features
7031 * @dev: the device to check
7033 * Recalculate dev->features set and send notifications if it
7034 * has changed. Should be called after driver or hardware dependent
7035 * conditions might have changed that influence the features.
7037 void netdev_update_features(struct net_device *dev)
7039 if (__netdev_update_features(dev))
7040 netdev_features_change(dev);
7042 EXPORT_SYMBOL(netdev_update_features);
7045 * netdev_change_features - recalculate device features
7046 * @dev: the device to check
7048 * Recalculate dev->features set and send notifications even
7049 * if they have not changed. Should be called instead of
7050 * netdev_update_features() if also dev->vlan_features might
7051 * have changed to allow the changes to be propagated to stacked
7054 void netdev_change_features(struct net_device *dev)
7056 __netdev_update_features(dev);
7057 netdev_features_change(dev);
7059 EXPORT_SYMBOL(netdev_change_features);
7062 * netif_stacked_transfer_operstate - transfer operstate
7063 * @rootdev: the root or lower level device to transfer state from
7064 * @dev: the device to transfer operstate to
7066 * Transfer operational state from root to device. This is normally
7067 * called when a stacking relationship exists between the root
7068 * device and the device(a leaf device).
7070 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
7071 struct net_device *dev)
7073 if (rootdev->operstate == IF_OPER_DORMANT)
7074 netif_dormant_on(dev);
7076 netif_dormant_off(dev);
7078 if (netif_carrier_ok(rootdev)) {
7079 if (!netif_carrier_ok(dev))
7080 netif_carrier_on(dev);
7082 if (netif_carrier_ok(dev))
7083 netif_carrier_off(dev);
7086 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
7089 static int netif_alloc_rx_queues(struct net_device *dev)
7091 unsigned int i, count = dev->num_rx_queues;
7092 struct netdev_rx_queue *rx;
7093 size_t sz = count * sizeof(*rx);
7097 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7105 for (i = 0; i < count; i++)
7111 static void netdev_init_one_queue(struct net_device *dev,
7112 struct netdev_queue *queue, void *_unused)
7114 /* Initialize queue lock */
7115 spin_lock_init(&queue->_xmit_lock);
7116 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
7117 queue->xmit_lock_owner = -1;
7118 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
7121 dql_init(&queue->dql, HZ);
7125 static void netif_free_tx_queues(struct net_device *dev)
7130 static int netif_alloc_netdev_queues(struct net_device *dev)
7132 unsigned int count = dev->num_tx_queues;
7133 struct netdev_queue *tx;
7134 size_t sz = count * sizeof(*tx);
7136 if (count < 1 || count > 0xffff)
7139 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7147 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
7148 spin_lock_init(&dev->tx_global_lock);
7153 void netif_tx_stop_all_queues(struct net_device *dev)
7157 for (i = 0; i < dev->num_tx_queues; i++) {
7158 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
7160 netif_tx_stop_queue(txq);
7163 EXPORT_SYMBOL(netif_tx_stop_all_queues);
7166 * register_netdevice - register a network device
7167 * @dev: device to register
7169 * Take a completed network device structure and add it to the kernel
7170 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7171 * chain. 0 is returned on success. A negative errno code is returned
7172 * on a failure to set up the device, or if the name is a duplicate.
7174 * Callers must hold the rtnl semaphore. You may want
7175 * register_netdev() instead of this.
7178 * The locking appears insufficient to guarantee two parallel registers
7179 * will not get the same name.
7182 int register_netdevice(struct net_device *dev)
7185 struct net *net = dev_net(dev);
7187 BUG_ON(dev_boot_phase);
7192 /* When net_device's are persistent, this will be fatal. */
7193 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
7196 spin_lock_init(&dev->addr_list_lock);
7197 netdev_set_addr_lockdep_class(dev);
7199 ret = dev_get_valid_name(net, dev, dev->name);
7203 /* Init, if this function is available */
7204 if (dev->netdev_ops->ndo_init) {
7205 ret = dev->netdev_ops->ndo_init(dev);
7213 if (((dev->hw_features | dev->features) &
7214 NETIF_F_HW_VLAN_CTAG_FILTER) &&
7215 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
7216 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
7217 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
7224 dev->ifindex = dev_new_index(net);
7225 else if (__dev_get_by_index(net, dev->ifindex))
7228 /* Transfer changeable features to wanted_features and enable
7229 * software offloads (GSO and GRO).
7231 dev->hw_features |= NETIF_F_SOFT_FEATURES;
7232 dev->features |= NETIF_F_SOFT_FEATURES;
7233 dev->wanted_features = dev->features & dev->hw_features;
7235 if (!(dev->flags & IFF_LOOPBACK))
7236 dev->hw_features |= NETIF_F_NOCACHE_COPY;
7238 /* If IPv4 TCP segmentation offload is supported we should also
7239 * allow the device to enable segmenting the frame with the option
7240 * of ignoring a static IP ID value. This doesn't enable the
7241 * feature itself but allows the user to enable it later.
7243 if (dev->hw_features & NETIF_F_TSO)
7244 dev->hw_features |= NETIF_F_TSO_MANGLEID;
7245 if (dev->vlan_features & NETIF_F_TSO)
7246 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
7247 if (dev->mpls_features & NETIF_F_TSO)
7248 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
7249 if (dev->hw_enc_features & NETIF_F_TSO)
7250 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
7252 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
7254 dev->vlan_features |= NETIF_F_HIGHDMA;
7256 /* Make NETIF_F_SG inheritable to tunnel devices.
7258 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
7260 /* Make NETIF_F_SG inheritable to MPLS.
7262 dev->mpls_features |= NETIF_F_SG;
7264 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
7265 ret = notifier_to_errno(ret);
7269 ret = netdev_register_kobject(dev);
7272 dev->reg_state = NETREG_REGISTERED;
7274 __netdev_update_features(dev);
7277 * Default initial state at registry is that the
7278 * device is present.
7281 set_bit(__LINK_STATE_PRESENT, &dev->state);
7283 linkwatch_init_dev(dev);
7285 dev_init_scheduler(dev);
7287 list_netdevice(dev);
7288 add_device_randomness(dev->dev_addr, dev->addr_len);
7290 /* If the device has permanent device address, driver should
7291 * set dev_addr and also addr_assign_type should be set to
7292 * NET_ADDR_PERM (default value).
7294 if (dev->addr_assign_type == NET_ADDR_PERM)
7295 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
7297 /* Notify protocols, that a new device appeared. */
7298 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
7299 ret = notifier_to_errno(ret);
7301 rollback_registered(dev);
7302 dev->reg_state = NETREG_UNREGISTERED;
7305 * Prevent userspace races by waiting until the network
7306 * device is fully setup before sending notifications.
7308 if (!dev->rtnl_link_ops ||
7309 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7310 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7316 if (dev->netdev_ops->ndo_uninit)
7317 dev->netdev_ops->ndo_uninit(dev);
7320 EXPORT_SYMBOL(register_netdevice);
7323 * init_dummy_netdev - init a dummy network device for NAPI
7324 * @dev: device to init
7326 * This takes a network device structure and initialize the minimum
7327 * amount of fields so it can be used to schedule NAPI polls without
7328 * registering a full blown interface. This is to be used by drivers
7329 * that need to tie several hardware interfaces to a single NAPI
7330 * poll scheduler due to HW limitations.
7332 int init_dummy_netdev(struct net_device *dev)
7334 /* Clear everything. Note we don't initialize spinlocks
7335 * are they aren't supposed to be taken by any of the
7336 * NAPI code and this dummy netdev is supposed to be
7337 * only ever used for NAPI polls
7339 memset(dev, 0, sizeof(struct net_device));
7341 /* make sure we BUG if trying to hit standard
7342 * register/unregister code path
7344 dev->reg_state = NETREG_DUMMY;
7346 /* NAPI wants this */
7347 INIT_LIST_HEAD(&dev->napi_list);
7349 /* a dummy interface is started by default */
7350 set_bit(__LINK_STATE_PRESENT, &dev->state);
7351 set_bit(__LINK_STATE_START, &dev->state);
7353 /* Note : We dont allocate pcpu_refcnt for dummy devices,
7354 * because users of this 'device' dont need to change
7360 EXPORT_SYMBOL_GPL(init_dummy_netdev);
7364 * register_netdev - register a network device
7365 * @dev: device to register
7367 * Take a completed network device structure and add it to the kernel
7368 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7369 * chain. 0 is returned on success. A negative errno code is returned
7370 * on a failure to set up the device, or if the name is a duplicate.
7372 * This is a wrapper around register_netdevice that takes the rtnl semaphore
7373 * and expands the device name if you passed a format string to
7376 int register_netdev(struct net_device *dev)
7381 err = register_netdevice(dev);
7385 EXPORT_SYMBOL(register_netdev);
7387 int netdev_refcnt_read(const struct net_device *dev)
7391 for_each_possible_cpu(i)
7392 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
7395 EXPORT_SYMBOL(netdev_refcnt_read);
7398 * netdev_wait_allrefs - wait until all references are gone.
7399 * @dev: target net_device
7401 * This is called when unregistering network devices.
7403 * Any protocol or device that holds a reference should register
7404 * for netdevice notification, and cleanup and put back the
7405 * reference if they receive an UNREGISTER event.
7406 * We can get stuck here if buggy protocols don't correctly
7409 static void netdev_wait_allrefs(struct net_device *dev)
7411 unsigned long rebroadcast_time, warning_time;
7414 linkwatch_forget_dev(dev);
7416 rebroadcast_time = warning_time = jiffies;
7417 refcnt = netdev_refcnt_read(dev);
7419 while (refcnt != 0) {
7420 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
7423 /* Rebroadcast unregister notification */
7424 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7430 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7431 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
7433 /* We must not have linkwatch events
7434 * pending on unregister. If this
7435 * happens, we simply run the queue
7436 * unscheduled, resulting in a noop
7439 linkwatch_run_queue();
7444 rebroadcast_time = jiffies;
7449 refcnt = netdev_refcnt_read(dev);
7451 if (time_after(jiffies, warning_time + 10 * HZ)) {
7452 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
7454 warning_time = jiffies;
7463 * register_netdevice(x1);
7464 * register_netdevice(x2);
7466 * unregister_netdevice(y1);
7467 * unregister_netdevice(y2);
7473 * We are invoked by rtnl_unlock().
7474 * This allows us to deal with problems:
7475 * 1) We can delete sysfs objects which invoke hotplug
7476 * without deadlocking with linkwatch via keventd.
7477 * 2) Since we run with the RTNL semaphore not held, we can sleep
7478 * safely in order to wait for the netdev refcnt to drop to zero.
7480 * We must not return until all unregister events added during
7481 * the interval the lock was held have been completed.
7483 void netdev_run_todo(void)
7485 struct list_head list;
7487 /* Snapshot list, allow later requests */
7488 list_replace_init(&net_todo_list, &list);
7493 /* Wait for rcu callbacks to finish before next phase */
7494 if (!list_empty(&list))
7497 while (!list_empty(&list)) {
7498 struct net_device *dev
7499 = list_first_entry(&list, struct net_device, todo_list);
7500 list_del(&dev->todo_list);
7503 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7506 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
7507 pr_err("network todo '%s' but state %d\n",
7508 dev->name, dev->reg_state);
7513 dev->reg_state = NETREG_UNREGISTERED;
7515 netdev_wait_allrefs(dev);
7518 BUG_ON(netdev_refcnt_read(dev));
7519 BUG_ON(!list_empty(&dev->ptype_all));
7520 BUG_ON(!list_empty(&dev->ptype_specific));
7521 WARN_ON(rcu_access_pointer(dev->ip_ptr));
7522 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
7523 WARN_ON(dev->dn_ptr);
7525 if (dev->destructor)
7526 dev->destructor(dev);
7528 /* Report a network device has been unregistered */
7530 dev_net(dev)->dev_unreg_count--;
7532 wake_up(&netdev_unregistering_wq);
7534 /* Free network device */
7535 kobject_put(&dev->dev.kobj);
7539 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
7540 * all the same fields in the same order as net_device_stats, with only
7541 * the type differing, but rtnl_link_stats64 may have additional fields
7542 * at the end for newer counters.
7544 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7545 const struct net_device_stats *netdev_stats)
7547 #if BITS_PER_LONG == 64
7548 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
7549 memcpy(stats64, netdev_stats, sizeof(*stats64));
7550 /* zero out counters that only exist in rtnl_link_stats64 */
7551 memset((char *)stats64 + sizeof(*netdev_stats), 0,
7552 sizeof(*stats64) - sizeof(*netdev_stats));
7554 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
7555 const unsigned long *src = (const unsigned long *)netdev_stats;
7556 u64 *dst = (u64 *)stats64;
7558 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
7559 for (i = 0; i < n; i++)
7561 /* zero out counters that only exist in rtnl_link_stats64 */
7562 memset((char *)stats64 + n * sizeof(u64), 0,
7563 sizeof(*stats64) - n * sizeof(u64));
7566 EXPORT_SYMBOL(netdev_stats_to_stats64);
7569 * dev_get_stats - get network device statistics
7570 * @dev: device to get statistics from
7571 * @storage: place to store stats
7573 * Get network statistics from device. Return @storage.
7574 * The device driver may provide its own method by setting
7575 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7576 * otherwise the internal statistics structure is used.
7578 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
7579 struct rtnl_link_stats64 *storage)
7581 const struct net_device_ops *ops = dev->netdev_ops;
7583 if (ops->ndo_get_stats64) {
7584 memset(storage, 0, sizeof(*storage));
7585 ops->ndo_get_stats64(dev, storage);
7586 } else if (ops->ndo_get_stats) {
7587 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7589 netdev_stats_to_stats64(storage, &dev->stats);
7591 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
7592 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
7593 storage->rx_nohandler += atomic_long_read(&dev->rx_nohandler);
7596 EXPORT_SYMBOL(dev_get_stats);
7598 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7600 struct netdev_queue *queue = dev_ingress_queue(dev);
7602 #ifdef CONFIG_NET_CLS_ACT
7605 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7608 netdev_init_one_queue(dev, queue, NULL);
7609 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7610 queue->qdisc_sleeping = &noop_qdisc;
7611 rcu_assign_pointer(dev->ingress_queue, queue);
7616 static const struct ethtool_ops default_ethtool_ops;
7618 void netdev_set_default_ethtool_ops(struct net_device *dev,
7619 const struct ethtool_ops *ops)
7621 if (dev->ethtool_ops == &default_ethtool_ops)
7622 dev->ethtool_ops = ops;
7624 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7626 void netdev_freemem(struct net_device *dev)
7628 char *addr = (char *)dev - dev->padded;
7634 * alloc_netdev_mqs - allocate network device
7635 * @sizeof_priv: size of private data to allocate space for
7636 * @name: device name format string
7637 * @name_assign_type: origin of device name
7638 * @setup: callback to initialize device
7639 * @txqs: the number of TX subqueues to allocate
7640 * @rxqs: the number of RX subqueues to allocate
7642 * Allocates a struct net_device with private data area for driver use
7643 * and performs basic initialization. Also allocates subqueue structs
7644 * for each queue on the device.
7646 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7647 unsigned char name_assign_type,
7648 void (*setup)(struct net_device *),
7649 unsigned int txqs, unsigned int rxqs)
7651 struct net_device *dev;
7653 struct net_device *p;
7655 BUG_ON(strlen(name) >= sizeof(dev->name));
7658 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7664 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7669 alloc_size = sizeof(struct net_device);
7671 /* ensure 32-byte alignment of private area */
7672 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7673 alloc_size += sizeof_priv;
7675 /* ensure 32-byte alignment of whole construct */
7676 alloc_size += NETDEV_ALIGN - 1;
7678 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7680 p = vzalloc(alloc_size);
7684 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7685 dev->padded = (char *)dev - (char *)p;
7687 dev->pcpu_refcnt = alloc_percpu(int);
7688 if (!dev->pcpu_refcnt)
7691 if (dev_addr_init(dev))
7697 dev_net_set(dev, &init_net);
7699 dev->gso_max_size = GSO_MAX_SIZE;
7700 dev->gso_max_segs = GSO_MAX_SEGS;
7702 INIT_LIST_HEAD(&dev->napi_list);
7703 INIT_LIST_HEAD(&dev->unreg_list);
7704 INIT_LIST_HEAD(&dev->close_list);
7705 INIT_LIST_HEAD(&dev->link_watch_list);
7706 INIT_LIST_HEAD(&dev->adj_list.upper);
7707 INIT_LIST_HEAD(&dev->adj_list.lower);
7708 INIT_LIST_HEAD(&dev->ptype_all);
7709 INIT_LIST_HEAD(&dev->ptype_specific);
7710 #ifdef CONFIG_NET_SCHED
7711 hash_init(dev->qdisc_hash);
7713 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7716 if (!dev->tx_queue_len) {
7717 dev->priv_flags |= IFF_NO_QUEUE;
7718 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
7721 dev->num_tx_queues = txqs;
7722 dev->real_num_tx_queues = txqs;
7723 if (netif_alloc_netdev_queues(dev))
7727 dev->num_rx_queues = rxqs;
7728 dev->real_num_rx_queues = rxqs;
7729 if (netif_alloc_rx_queues(dev))
7733 strcpy(dev->name, name);
7734 dev->name_assign_type = name_assign_type;
7735 dev->group = INIT_NETDEV_GROUP;
7736 if (!dev->ethtool_ops)
7737 dev->ethtool_ops = &default_ethtool_ops;
7739 nf_hook_ingress_init(dev);
7748 free_percpu(dev->pcpu_refcnt);
7750 netdev_freemem(dev);
7753 EXPORT_SYMBOL(alloc_netdev_mqs);
7756 * free_netdev - free network device
7759 * This function does the last stage of destroying an allocated device
7760 * interface. The reference to the device object is released. If this
7761 * is the last reference then it will be freed.Must be called in process
7764 void free_netdev(struct net_device *dev)
7766 struct napi_struct *p, *n;
7769 netif_free_tx_queues(dev);
7774 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7776 /* Flush device addresses */
7777 dev_addr_flush(dev);
7779 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7782 free_percpu(dev->pcpu_refcnt);
7783 dev->pcpu_refcnt = NULL;
7785 /* Compatibility with error handling in drivers */
7786 if (dev->reg_state == NETREG_UNINITIALIZED) {
7787 netdev_freemem(dev);
7791 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7792 dev->reg_state = NETREG_RELEASED;
7794 /* will free via device release */
7795 put_device(&dev->dev);
7797 EXPORT_SYMBOL(free_netdev);
7800 * synchronize_net - Synchronize with packet receive processing
7802 * Wait for packets currently being received to be done.
7803 * Does not block later packets from starting.
7805 void synchronize_net(void)
7808 if (rtnl_is_locked())
7809 synchronize_rcu_expedited();
7813 EXPORT_SYMBOL(synchronize_net);
7816 * unregister_netdevice_queue - remove device from the kernel
7820 * This function shuts down a device interface and removes it
7821 * from the kernel tables.
7822 * If head not NULL, device is queued to be unregistered later.
7824 * Callers must hold the rtnl semaphore. You may want
7825 * unregister_netdev() instead of this.
7828 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7833 list_move_tail(&dev->unreg_list, head);
7835 rollback_registered(dev);
7836 /* Finish processing unregister after unlock */
7840 EXPORT_SYMBOL(unregister_netdevice_queue);
7843 * unregister_netdevice_many - unregister many devices
7844 * @head: list of devices
7846 * Note: As most callers use a stack allocated list_head,
7847 * we force a list_del() to make sure stack wont be corrupted later.
7849 void unregister_netdevice_many(struct list_head *head)
7851 struct net_device *dev;
7853 if (!list_empty(head)) {
7854 rollback_registered_many(head);
7855 list_for_each_entry(dev, head, unreg_list)
7860 EXPORT_SYMBOL(unregister_netdevice_many);
7863 * unregister_netdev - remove device from the kernel
7866 * This function shuts down a device interface and removes it
7867 * from the kernel tables.
7869 * This is just a wrapper for unregister_netdevice that takes
7870 * the rtnl semaphore. In general you want to use this and not
7871 * unregister_netdevice.
7873 void unregister_netdev(struct net_device *dev)
7876 unregister_netdevice(dev);
7879 EXPORT_SYMBOL(unregister_netdev);
7882 * dev_change_net_namespace - move device to different nethost namespace
7884 * @net: network namespace
7885 * @pat: If not NULL name pattern to try if the current device name
7886 * is already taken in the destination network namespace.
7888 * This function shuts down a device interface and moves it
7889 * to a new network namespace. On success 0 is returned, on
7890 * a failure a netagive errno code is returned.
7892 * Callers must hold the rtnl semaphore.
7895 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7901 /* Don't allow namespace local devices to be moved. */
7903 if (dev->features & NETIF_F_NETNS_LOCAL)
7906 /* Ensure the device has been registrered */
7907 if (dev->reg_state != NETREG_REGISTERED)
7910 /* Get out if there is nothing todo */
7912 if (net_eq(dev_net(dev), net))
7915 /* Pick the destination device name, and ensure
7916 * we can use it in the destination network namespace.
7919 if (__dev_get_by_name(net, dev->name)) {
7920 /* We get here if we can't use the current device name */
7923 if (dev_get_valid_name(net, dev, pat) < 0)
7928 * And now a mini version of register_netdevice unregister_netdevice.
7931 /* If device is running close it first. */
7934 /* And unlink it from device chain */
7936 unlist_netdevice(dev);
7940 /* Shutdown queueing discipline. */
7943 /* Notify protocols, that we are about to destroy
7944 * this device. They should clean all the things.
7946 * Note that dev->reg_state stays at NETREG_REGISTERED.
7947 * This is wanted because this way 8021q and macvlan know
7948 * the device is just moving and can keep their slaves up.
7950 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7952 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7953 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7956 * Flush the unicast and multicast chains
7961 /* Send a netdev-removed uevent to the old namespace */
7962 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7963 netdev_adjacent_del_links(dev);
7965 /* Actually switch the network namespace */
7966 dev_net_set(dev, net);
7968 /* If there is an ifindex conflict assign a new one */
7969 if (__dev_get_by_index(net, dev->ifindex))
7970 dev->ifindex = dev_new_index(net);
7972 /* Send a netdev-add uevent to the new namespace */
7973 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7974 netdev_adjacent_add_links(dev);
7976 /* Fixup kobjects */
7977 err = device_rename(&dev->dev, dev->name);
7980 /* Add the device back in the hashes */
7981 list_netdevice(dev);
7983 /* Notify protocols, that a new device appeared. */
7984 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7987 * Prevent userspace races by waiting until the network
7988 * device is fully setup before sending notifications.
7990 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7997 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7999 static int dev_cpu_dead(unsigned int oldcpu)
8001 struct sk_buff **list_skb;
8002 struct sk_buff *skb;
8004 struct softnet_data *sd, *oldsd;
8006 local_irq_disable();
8007 cpu = smp_processor_id();
8008 sd = &per_cpu(softnet_data, cpu);
8009 oldsd = &per_cpu(softnet_data, oldcpu);
8011 /* Find end of our completion_queue. */
8012 list_skb = &sd->completion_queue;
8014 list_skb = &(*list_skb)->next;
8015 /* Append completion queue from offline CPU. */
8016 *list_skb = oldsd->completion_queue;
8017 oldsd->completion_queue = NULL;
8019 /* Append output queue from offline CPU. */
8020 if (oldsd->output_queue) {
8021 *sd->output_queue_tailp = oldsd->output_queue;
8022 sd->output_queue_tailp = oldsd->output_queue_tailp;
8023 oldsd->output_queue = NULL;
8024 oldsd->output_queue_tailp = &oldsd->output_queue;
8026 /* Append NAPI poll list from offline CPU, with one exception :
8027 * process_backlog() must be called by cpu owning percpu backlog.
8028 * We properly handle process_queue & input_pkt_queue later.
8030 while (!list_empty(&oldsd->poll_list)) {
8031 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
8035 list_del_init(&napi->poll_list);
8036 if (napi->poll == process_backlog)
8039 ____napi_schedule(sd, napi);
8042 raise_softirq_irqoff(NET_TX_SOFTIRQ);
8045 /* Process offline CPU's input_pkt_queue */
8046 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
8048 input_queue_head_incr(oldsd);
8050 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
8052 input_queue_head_incr(oldsd);
8059 * netdev_increment_features - increment feature set by one
8060 * @all: current feature set
8061 * @one: new feature set
8062 * @mask: mask feature set
8064 * Computes a new feature set after adding a device with feature set
8065 * @one to the master device with current feature set @all. Will not
8066 * enable anything that is off in @mask. Returns the new feature set.
8068 netdev_features_t netdev_increment_features(netdev_features_t all,
8069 netdev_features_t one, netdev_features_t mask)
8071 if (mask & NETIF_F_HW_CSUM)
8072 mask |= NETIF_F_CSUM_MASK;
8073 mask |= NETIF_F_VLAN_CHALLENGED;
8075 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
8076 all &= one | ~NETIF_F_ALL_FOR_ALL;
8078 /* If one device supports hw checksumming, set for all. */
8079 if (all & NETIF_F_HW_CSUM)
8080 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
8084 EXPORT_SYMBOL(netdev_increment_features);
8086 static struct hlist_head * __net_init netdev_create_hash(void)
8089 struct hlist_head *hash;
8091 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
8093 for (i = 0; i < NETDEV_HASHENTRIES; i++)
8094 INIT_HLIST_HEAD(&hash[i]);
8099 /* Initialize per network namespace state */
8100 static int __net_init netdev_init(struct net *net)
8102 if (net != &init_net)
8103 INIT_LIST_HEAD(&net->dev_base_head);
8105 net->dev_name_head = netdev_create_hash();
8106 if (net->dev_name_head == NULL)
8109 net->dev_index_head = netdev_create_hash();
8110 if (net->dev_index_head == NULL)
8116 kfree(net->dev_name_head);
8122 * netdev_drivername - network driver for the device
8123 * @dev: network device
8125 * Determine network driver for device.
8127 const char *netdev_drivername(const struct net_device *dev)
8129 const struct device_driver *driver;
8130 const struct device *parent;
8131 const char *empty = "";
8133 parent = dev->dev.parent;
8137 driver = parent->driver;
8138 if (driver && driver->name)
8139 return driver->name;
8143 static void __netdev_printk(const char *level, const struct net_device *dev,
8144 struct va_format *vaf)
8146 if (dev && dev->dev.parent) {
8147 dev_printk_emit(level[1] - '0',
8150 dev_driver_string(dev->dev.parent),
8151 dev_name(dev->dev.parent),
8152 netdev_name(dev), netdev_reg_state(dev),
8155 printk("%s%s%s: %pV",
8156 level, netdev_name(dev), netdev_reg_state(dev), vaf);
8158 printk("%s(NULL net_device): %pV", level, vaf);
8162 void netdev_printk(const char *level, const struct net_device *dev,
8163 const char *format, ...)
8165 struct va_format vaf;
8168 va_start(args, format);
8173 __netdev_printk(level, dev, &vaf);
8177 EXPORT_SYMBOL(netdev_printk);
8179 #define define_netdev_printk_level(func, level) \
8180 void func(const struct net_device *dev, const char *fmt, ...) \
8182 struct va_format vaf; \
8185 va_start(args, fmt); \
8190 __netdev_printk(level, dev, &vaf); \
8194 EXPORT_SYMBOL(func);
8196 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
8197 define_netdev_printk_level(netdev_alert, KERN_ALERT);
8198 define_netdev_printk_level(netdev_crit, KERN_CRIT);
8199 define_netdev_printk_level(netdev_err, KERN_ERR);
8200 define_netdev_printk_level(netdev_warn, KERN_WARNING);
8201 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
8202 define_netdev_printk_level(netdev_info, KERN_INFO);
8204 static void __net_exit netdev_exit(struct net *net)
8206 kfree(net->dev_name_head);
8207 kfree(net->dev_index_head);
8210 static struct pernet_operations __net_initdata netdev_net_ops = {
8211 .init = netdev_init,
8212 .exit = netdev_exit,
8215 static void __net_exit default_device_exit(struct net *net)
8217 struct net_device *dev, *aux;
8219 * Push all migratable network devices back to the
8220 * initial network namespace
8223 for_each_netdev_safe(net, dev, aux) {
8225 char fb_name[IFNAMSIZ];
8227 /* Ignore unmoveable devices (i.e. loopback) */
8228 if (dev->features & NETIF_F_NETNS_LOCAL)
8231 /* Leave virtual devices for the generic cleanup */
8232 if (dev->rtnl_link_ops)
8235 /* Push remaining network devices to init_net */
8236 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
8237 err = dev_change_net_namespace(dev, &init_net, fb_name);
8239 pr_emerg("%s: failed to move %s to init_net: %d\n",
8240 __func__, dev->name, err);
8247 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
8249 /* Return with the rtnl_lock held when there are no network
8250 * devices unregistering in any network namespace in net_list.
8254 DEFINE_WAIT_FUNC(wait, woken_wake_function);
8256 add_wait_queue(&netdev_unregistering_wq, &wait);
8258 unregistering = false;
8260 list_for_each_entry(net, net_list, exit_list) {
8261 if (net->dev_unreg_count > 0) {
8262 unregistering = true;
8270 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
8272 remove_wait_queue(&netdev_unregistering_wq, &wait);
8275 static void __net_exit default_device_exit_batch(struct list_head *net_list)
8277 /* At exit all network devices most be removed from a network
8278 * namespace. Do this in the reverse order of registration.
8279 * Do this across as many network namespaces as possible to
8280 * improve batching efficiency.
8282 struct net_device *dev;
8284 LIST_HEAD(dev_kill_list);
8286 /* To prevent network device cleanup code from dereferencing
8287 * loopback devices or network devices that have been freed
8288 * wait here for all pending unregistrations to complete,
8289 * before unregistring the loopback device and allowing the
8290 * network namespace be freed.
8292 * The netdev todo list containing all network devices
8293 * unregistrations that happen in default_device_exit_batch
8294 * will run in the rtnl_unlock() at the end of
8295 * default_device_exit_batch.
8297 rtnl_lock_unregistering(net_list);
8298 list_for_each_entry(net, net_list, exit_list) {
8299 for_each_netdev_reverse(net, dev) {
8300 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
8301 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
8303 unregister_netdevice_queue(dev, &dev_kill_list);
8306 unregister_netdevice_many(&dev_kill_list);
8310 static struct pernet_operations __net_initdata default_device_ops = {
8311 .exit = default_device_exit,
8312 .exit_batch = default_device_exit_batch,
8316 * Initialize the DEV module. At boot time this walks the device list and
8317 * unhooks any devices that fail to initialise (normally hardware not
8318 * present) and leaves us with a valid list of present and active devices.
8323 * This is called single threaded during boot, so no need
8324 * to take the rtnl semaphore.
8326 static int __init net_dev_init(void)
8328 int i, rc = -ENOMEM;
8330 BUG_ON(!dev_boot_phase);
8332 if (dev_proc_init())
8335 if (netdev_kobject_init())
8338 INIT_LIST_HEAD(&ptype_all);
8339 for (i = 0; i < PTYPE_HASH_SIZE; i++)
8340 INIT_LIST_HEAD(&ptype_base[i]);
8342 INIT_LIST_HEAD(&offload_base);
8344 if (register_pernet_subsys(&netdev_net_ops))
8348 * Initialise the packet receive queues.
8351 for_each_possible_cpu(i) {
8352 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
8353 struct softnet_data *sd = &per_cpu(softnet_data, i);
8355 INIT_WORK(flush, flush_backlog);
8357 skb_queue_head_init(&sd->input_pkt_queue);
8358 skb_queue_head_init(&sd->process_queue);
8359 INIT_LIST_HEAD(&sd->poll_list);
8360 sd->output_queue_tailp = &sd->output_queue;
8362 sd->csd.func = rps_trigger_softirq;
8367 sd->backlog.poll = process_backlog;
8368 sd->backlog.weight = weight_p;
8373 /* The loopback device is special if any other network devices
8374 * is present in a network namespace the loopback device must
8375 * be present. Since we now dynamically allocate and free the
8376 * loopback device ensure this invariant is maintained by
8377 * keeping the loopback device as the first device on the
8378 * list of network devices. Ensuring the loopback devices
8379 * is the first device that appears and the last network device
8382 if (register_pernet_device(&loopback_net_ops))
8385 if (register_pernet_device(&default_device_ops))
8388 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
8389 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
8391 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
8392 NULL, dev_cpu_dead);
8400 subsys_initcall(net_dev_init);