2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <linux/bpf.h>
98 #include <net/net_namespace.h>
100 #include <net/busy_poll.h>
101 #include <linux/rtnetlink.h>
102 #include <linux/stat.h>
104 #include <net/dst_metadata.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/iw_handler.h>
115 #include <asm/current.h>
116 #include <linux/audit.h>
117 #include <linux/dmaengine.h>
118 #include <linux/err.h>
119 #include <linux/ctype.h>
120 #include <linux/if_arp.h>
121 #include <linux/if_vlan.h>
122 #include <linux/ip.h>
124 #include <net/mpls.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/static_key.h>
136 #include <linux/hashtable.h>
137 #include <linux/vmalloc.h>
138 #include <linux/if_macvlan.h>
139 #include <linux/errqueue.h>
140 #include <linux/hrtimer.h>
141 #include <linux/netfilter_ingress.h>
142 #include <linux/sctp.h>
143 #include <linux/crash_dump.h>
145 #include "net-sysfs.h"
147 /* Instead of increasing this, you should create a hash table. */
148 #define MAX_GRO_SKBS 8
150 /* This should be increased if a protocol with a bigger head is added. */
151 #define GRO_MAX_HEAD (MAX_HEADER + 128)
153 static DEFINE_SPINLOCK(ptype_lock);
154 static DEFINE_SPINLOCK(offload_lock);
155 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
156 struct list_head ptype_all __read_mostly; /* Taps */
157 static struct list_head offload_base __read_mostly;
159 static int netif_rx_internal(struct sk_buff *skb);
160 static int call_netdevice_notifiers_info(unsigned long val,
161 struct net_device *dev,
162 struct netdev_notifier_info *info);
165 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
168 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
170 * Writers must hold the rtnl semaphore while they loop through the
171 * dev_base_head list, and hold dev_base_lock for writing when they do the
172 * actual updates. This allows pure readers to access the list even
173 * while a writer is preparing to update it.
175 * To put it another way, dev_base_lock is held for writing only to
176 * protect against pure readers; the rtnl semaphore provides the
177 * protection against other writers.
179 * See, for example usages, register_netdevice() and
180 * unregister_netdevice(), which must be called with the rtnl
183 DEFINE_RWLOCK(dev_base_lock);
184 EXPORT_SYMBOL(dev_base_lock);
186 /* protects napi_hash addition/deletion and napi_gen_id */
187 static DEFINE_SPINLOCK(napi_hash_lock);
189 static unsigned int napi_gen_id = NR_CPUS;
190 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
192 static seqcount_t devnet_rename_seq;
194 static inline void dev_base_seq_inc(struct net *net)
196 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 *******************************************************************************/
362 * Add a protocol ID to the list. Now that the input handler is
363 * smarter we can dispense with all the messy stuff that used to be
366 * BEWARE!!! Protocol handlers, mangling input packets,
367 * MUST BE last in hash buckets and checking protocol handlers
368 * MUST start from promiscuous ptype_all chain in net_bh.
369 * It is true now, do not change it.
370 * Explanation follows: if protocol handler, mangling packet, will
371 * be the first on list, it is not able to sense, that packet
372 * is cloned and should be copied-on-write, so that it will
373 * change it and subsequent readers will get broken packet.
377 static inline struct list_head *ptype_head(const struct packet_type *pt)
379 if (pt->type == htons(ETH_P_ALL))
380 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
382 return pt->dev ? &pt->dev->ptype_specific :
383 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
387 * dev_add_pack - add packet handler
388 * @pt: packet type declaration
390 * Add a protocol handler to the networking stack. The passed &packet_type
391 * is linked into kernel lists and may not be freed until it has been
392 * removed from the kernel lists.
394 * This call does not sleep therefore it can not
395 * guarantee all CPU's that are in middle of receiving packets
396 * will see the new packet type (until the next received packet).
399 void dev_add_pack(struct packet_type *pt)
401 struct list_head *head = ptype_head(pt);
403 spin_lock(&ptype_lock);
404 list_add_rcu(&pt->list, head);
405 spin_unlock(&ptype_lock);
407 EXPORT_SYMBOL(dev_add_pack);
410 * __dev_remove_pack - remove packet handler
411 * @pt: packet type declaration
413 * Remove a protocol handler that was previously added to the kernel
414 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
415 * from the kernel lists and can be freed or reused once this function
418 * The packet type might still be in use by receivers
419 * and must not be freed until after all the CPU's have gone
420 * through a quiescent state.
422 void __dev_remove_pack(struct packet_type *pt)
424 struct list_head *head = ptype_head(pt);
425 struct packet_type *pt1;
427 spin_lock(&ptype_lock);
429 list_for_each_entry(pt1, head, list) {
431 list_del_rcu(&pt->list);
436 pr_warn("dev_remove_pack: %p not found\n", pt);
438 spin_unlock(&ptype_lock);
440 EXPORT_SYMBOL(__dev_remove_pack);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type *pt)
456 __dev_remove_pack(pt);
460 EXPORT_SYMBOL(dev_remove_pack);
464 * dev_add_offload - register offload handlers
465 * @po: protocol offload declaration
467 * Add protocol offload handlers to the networking stack. The passed
468 * &proto_offload is linked into kernel lists and may not be freed until
469 * it has been removed from the kernel lists.
471 * This call does not sleep therefore it can not
472 * guarantee all CPU's that are in middle of receiving packets
473 * will see the new offload handlers (until the next received packet).
475 void dev_add_offload(struct packet_offload *po)
477 struct packet_offload *elem;
479 spin_lock(&offload_lock);
480 list_for_each_entry(elem, &offload_base, list) {
481 if (po->priority < elem->priority)
484 list_add_rcu(&po->list, elem->list.prev);
485 spin_unlock(&offload_lock);
487 EXPORT_SYMBOL(dev_add_offload);
490 * __dev_remove_offload - remove offload handler
491 * @po: packet offload declaration
493 * Remove a protocol offload handler that was previously added to the
494 * kernel offload handlers by dev_add_offload(). The passed &offload_type
495 * is removed from the kernel lists and can be freed or reused once this
498 * The packet type might still be in use by receivers
499 * and must not be freed until after all the CPU's have gone
500 * through a quiescent state.
502 static void __dev_remove_offload(struct packet_offload *po)
504 struct list_head *head = &offload_base;
505 struct packet_offload *po1;
507 spin_lock(&offload_lock);
509 list_for_each_entry(po1, head, list) {
511 list_del_rcu(&po->list);
516 pr_warn("dev_remove_offload: %p not found\n", po);
518 spin_unlock(&offload_lock);
522 * dev_remove_offload - remove packet offload handler
523 * @po: packet offload declaration
525 * Remove a packet offload handler that was previously added to the kernel
526 * offload handlers by dev_add_offload(). The passed &offload_type is
527 * removed from the kernel lists and can be freed or reused once this
530 * This call sleeps to guarantee that no CPU is looking at the packet
533 void dev_remove_offload(struct packet_offload *po)
535 __dev_remove_offload(po);
539 EXPORT_SYMBOL(dev_remove_offload);
541 /******************************************************************************
543 Device Boot-time Settings Routines
545 *******************************************************************************/
547 /* Boot time configuration table */
548 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
551 * netdev_boot_setup_add - add new setup entry
552 * @name: name of the device
553 * @map: configured settings for the device
555 * Adds new setup entry to the dev_boot_setup list. The function
556 * returns 0 on error and 1 on success. This is a generic routine to
559 static int netdev_boot_setup_add(char *name, struct ifmap *map)
561 struct netdev_boot_setup *s;
565 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
566 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
567 memset(s[i].name, 0, sizeof(s[i].name));
568 strlcpy(s[i].name, name, IFNAMSIZ);
569 memcpy(&s[i].map, map, sizeof(s[i].map));
574 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
578 * netdev_boot_setup_check - check boot time settings
579 * @dev: the netdevice
581 * Check boot time settings for the device.
582 * The found settings are set for the device to be used
583 * later in the device probing.
584 * Returns 0 if no settings found, 1 if they are.
586 int netdev_boot_setup_check(struct net_device *dev)
588 struct netdev_boot_setup *s = dev_boot_setup;
591 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
592 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
593 !strcmp(dev->name, s[i].name)) {
594 dev->irq = s[i].map.irq;
595 dev->base_addr = s[i].map.base_addr;
596 dev->mem_start = s[i].map.mem_start;
597 dev->mem_end = s[i].map.mem_end;
603 EXPORT_SYMBOL(netdev_boot_setup_check);
607 * netdev_boot_base - get address from boot time settings
608 * @prefix: prefix for network device
609 * @unit: id for network device
611 * Check boot time settings for the base address of device.
612 * The found settings are set for the device to be used
613 * later in the device probing.
614 * Returns 0 if no settings found.
616 unsigned long netdev_boot_base(const char *prefix, int unit)
618 const struct netdev_boot_setup *s = dev_boot_setup;
622 sprintf(name, "%s%d", prefix, unit);
625 * If device already registered then return base of 1
626 * to indicate not to probe for this interface
628 if (__dev_get_by_name(&init_net, name))
631 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
632 if (!strcmp(name, s[i].name))
633 return s[i].map.base_addr;
638 * Saves at boot time configured settings for any netdevice.
640 int __init netdev_boot_setup(char *str)
645 str = get_options(str, ARRAY_SIZE(ints), ints);
650 memset(&map, 0, sizeof(map));
654 map.base_addr = ints[2];
656 map.mem_start = ints[3];
658 map.mem_end = ints[4];
660 /* Add new entry to the list */
661 return netdev_boot_setup_add(str, &map);
664 __setup("netdev=", netdev_boot_setup);
666 /*******************************************************************************
668 Device Interface Subroutines
670 *******************************************************************************/
673 * dev_get_iflink - get 'iflink' value of a interface
674 * @dev: targeted interface
676 * Indicates the ifindex the interface is linked to.
677 * Physical interfaces have the same 'ifindex' and 'iflink' values.
680 int dev_get_iflink(const struct net_device *dev)
682 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
683 return dev->netdev_ops->ndo_get_iflink(dev);
687 EXPORT_SYMBOL(dev_get_iflink);
690 * dev_fill_metadata_dst - Retrieve tunnel egress information.
691 * @dev: targeted interface
694 * For better visibility of tunnel traffic OVS needs to retrieve
695 * egress tunnel information for a packet. Following API allows
696 * user to get this info.
698 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
700 struct ip_tunnel_info *info;
702 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
705 info = skb_tunnel_info_unclone(skb);
708 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
711 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
713 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
716 * __dev_get_by_name - find a device by its name
717 * @net: the applicable net namespace
718 * @name: name to find
720 * Find an interface by name. Must be called under RTNL semaphore
721 * or @dev_base_lock. If the name is found a pointer to the device
722 * is returned. If the name is not found then %NULL is returned. The
723 * reference counters are not incremented so the caller must be
724 * careful with locks.
727 struct net_device *__dev_get_by_name(struct net *net, const char *name)
729 struct net_device *dev;
730 struct hlist_head *head = dev_name_hash(net, name);
732 hlist_for_each_entry(dev, head, name_hlist)
733 if (!strncmp(dev->name, name, IFNAMSIZ))
738 EXPORT_SYMBOL(__dev_get_by_name);
741 * dev_get_by_name_rcu - find a device by its name
742 * @net: the applicable net namespace
743 * @name: name to find
745 * Find an interface by name.
746 * If the name is found a pointer to the device is returned.
747 * If the name is not found then %NULL is returned.
748 * The reference counters are not incremented so the caller must be
749 * careful with locks. The caller must hold RCU lock.
752 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
754 struct net_device *dev;
755 struct hlist_head *head = dev_name_hash(net, name);
757 hlist_for_each_entry_rcu(dev, head, name_hlist)
758 if (!strncmp(dev->name, name, IFNAMSIZ))
763 EXPORT_SYMBOL(dev_get_by_name_rcu);
766 * dev_get_by_name - find a device by its name
767 * @net: the applicable net namespace
768 * @name: name to find
770 * Find an interface by name. This can be called from any
771 * context and does its own locking. The returned handle has
772 * the usage count incremented and the caller must use dev_put() to
773 * release it when it is no longer needed. %NULL is returned if no
774 * matching device is found.
777 struct net_device *dev_get_by_name(struct net *net, const char *name)
779 struct net_device *dev;
782 dev = dev_get_by_name_rcu(net, name);
788 EXPORT_SYMBOL(dev_get_by_name);
791 * __dev_get_by_index - find a device by its ifindex
792 * @net: the applicable net namespace
793 * @ifindex: index of device
795 * Search for an interface by index. Returns %NULL if the device
796 * is not found or a pointer to the device. The device has not
797 * had its reference counter increased so the caller must be careful
798 * about locking. The caller must hold either the RTNL semaphore
802 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
804 struct net_device *dev;
805 struct hlist_head *head = dev_index_hash(net, ifindex);
807 hlist_for_each_entry(dev, head, index_hlist)
808 if (dev->ifindex == ifindex)
813 EXPORT_SYMBOL(__dev_get_by_index);
816 * dev_get_by_index_rcu - find a device by its ifindex
817 * @net: the applicable net namespace
818 * @ifindex: index of device
820 * Search for an interface by index. Returns %NULL if the device
821 * is not found or a pointer to the device. The device has not
822 * had its reference counter increased so the caller must be careful
823 * about locking. The caller must hold RCU lock.
826 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
828 struct net_device *dev;
829 struct hlist_head *head = dev_index_hash(net, ifindex);
831 hlist_for_each_entry_rcu(dev, head, index_hlist)
832 if (dev->ifindex == ifindex)
837 EXPORT_SYMBOL(dev_get_by_index_rcu);
841 * dev_get_by_index - find a device by its ifindex
842 * @net: the applicable net namespace
843 * @ifindex: index of device
845 * Search for an interface by index. Returns NULL if the device
846 * is not found or a pointer to the device. The device returned has
847 * had a reference added and the pointer is safe until the user calls
848 * dev_put to indicate they have finished with it.
851 struct net_device *dev_get_by_index(struct net *net, int ifindex)
853 struct net_device *dev;
856 dev = dev_get_by_index_rcu(net, ifindex);
862 EXPORT_SYMBOL(dev_get_by_index);
865 * netdev_get_name - get a netdevice name, knowing its ifindex.
866 * @net: network namespace
867 * @name: a pointer to the buffer where the name will be stored.
868 * @ifindex: the ifindex of the interface to get the name from.
870 * The use of raw_seqcount_begin() and cond_resched() before
871 * retrying is required as we want to give the writers a chance
872 * to complete when CONFIG_PREEMPT is not set.
874 int netdev_get_name(struct net *net, char *name, int ifindex)
876 struct net_device *dev;
880 seq = raw_seqcount_begin(&devnet_rename_seq);
882 dev = dev_get_by_index_rcu(net, ifindex);
888 strcpy(name, dev->name);
890 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
899 * dev_getbyhwaddr_rcu - find a device by its hardware address
900 * @net: the applicable net namespace
901 * @type: media type of device
902 * @ha: hardware address
904 * Search for an interface by MAC address. Returns NULL if the device
905 * is not found or a pointer to the device.
906 * The caller must hold RCU or RTNL.
907 * The returned device has not had its ref count increased
908 * and the caller must therefore be careful about locking
912 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
915 struct net_device *dev;
917 for_each_netdev_rcu(net, dev)
918 if (dev->type == type &&
919 !memcmp(dev->dev_addr, ha, dev->addr_len))
924 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
926 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
928 struct net_device *dev;
931 for_each_netdev(net, dev)
932 if (dev->type == type)
937 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
939 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
941 struct net_device *dev, *ret = NULL;
944 for_each_netdev_rcu(net, dev)
945 if (dev->type == type) {
953 EXPORT_SYMBOL(dev_getfirstbyhwtype);
956 * __dev_get_by_flags - find any device with given flags
957 * @net: the applicable net namespace
958 * @if_flags: IFF_* values
959 * @mask: bitmask of bits in if_flags to check
961 * Search for any interface with the given flags. Returns NULL if a device
962 * is not found or a pointer to the device. Must be called inside
963 * rtnl_lock(), and result refcount is unchanged.
966 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
969 struct net_device *dev, *ret;
974 for_each_netdev(net, dev) {
975 if (((dev->flags ^ if_flags) & mask) == 0) {
982 EXPORT_SYMBOL(__dev_get_by_flags);
985 * dev_valid_name - check if name is okay for network device
988 * Network device names need to be valid file names to
989 * to allow sysfs to work. We also disallow any kind of
992 bool dev_valid_name(const char *name)
996 if (strlen(name) >= IFNAMSIZ)
998 if (!strcmp(name, ".") || !strcmp(name, ".."))
1002 if (*name == '/' || *name == ':' || isspace(*name))
1008 EXPORT_SYMBOL(dev_valid_name);
1011 * __dev_alloc_name - allocate a name for a device
1012 * @net: network namespace to allocate the device name in
1013 * @name: name format string
1014 * @buf: scratch buffer and result name string
1016 * Passed a format string - eg "lt%d" it will try and find a suitable
1017 * id. It scans list of devices to build up a free map, then chooses
1018 * the first empty slot. The caller must hold the dev_base or rtnl lock
1019 * while allocating the name and adding the device in order to avoid
1021 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1022 * Returns the number of the unit assigned or a negative errno code.
1025 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1029 const int max_netdevices = 8*PAGE_SIZE;
1030 unsigned long *inuse;
1031 struct net_device *d;
1033 p = strnchr(name, IFNAMSIZ-1, '%');
1036 * Verify the string as this thing may have come from
1037 * the user. There must be either one "%d" and no other "%"
1040 if (p[1] != 'd' || strchr(p + 2, '%'))
1043 /* Use one page as a bit array of possible slots */
1044 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1048 for_each_netdev(net, d) {
1049 if (!sscanf(d->name, name, &i))
1051 if (i < 0 || i >= max_netdevices)
1054 /* avoid cases where sscanf is not exact inverse of printf */
1055 snprintf(buf, IFNAMSIZ, name, i);
1056 if (!strncmp(buf, d->name, IFNAMSIZ))
1060 i = find_first_zero_bit(inuse, max_netdevices);
1061 free_page((unsigned long) inuse);
1065 snprintf(buf, IFNAMSIZ, name, i);
1066 if (!__dev_get_by_name(net, buf))
1069 /* It is possible to run out of possible slots
1070 * when the name is long and there isn't enough space left
1071 * for the digits, or if all bits are used.
1077 * dev_alloc_name - allocate a name for a device
1079 * @name: name format string
1081 * Passed a format string - eg "lt%d" it will try and find a suitable
1082 * id. It scans list of devices to build up a free map, then chooses
1083 * the first empty slot. The caller must hold the dev_base or rtnl lock
1084 * while allocating the name and adding the device in order to avoid
1086 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1087 * Returns the number of the unit assigned or a negative errno code.
1090 int dev_alloc_name(struct net_device *dev, const char *name)
1096 BUG_ON(!dev_net(dev));
1098 ret = __dev_alloc_name(net, name, buf);
1100 strlcpy(dev->name, buf, IFNAMSIZ);
1103 EXPORT_SYMBOL(dev_alloc_name);
1105 static int dev_alloc_name_ns(struct net *net,
1106 struct net_device *dev,
1112 ret = __dev_alloc_name(net, name, buf);
1114 strlcpy(dev->name, buf, IFNAMSIZ);
1118 static int dev_get_valid_name(struct net *net,
1119 struct net_device *dev,
1124 if (!dev_valid_name(name))
1127 if (strchr(name, '%'))
1128 return dev_alloc_name_ns(net, dev, name);
1129 else if (__dev_get_by_name(net, name))
1131 else if (dev->name != name)
1132 strlcpy(dev->name, name, IFNAMSIZ);
1138 * dev_change_name - change name of a device
1140 * @newname: name (or format string) must be at least IFNAMSIZ
1142 * Change name of a device, can pass format strings "eth%d".
1145 int dev_change_name(struct net_device *dev, const char *newname)
1147 unsigned char old_assign_type;
1148 char oldname[IFNAMSIZ];
1154 BUG_ON(!dev_net(dev));
1157 if (dev->flags & IFF_UP)
1160 write_seqcount_begin(&devnet_rename_seq);
1162 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1163 write_seqcount_end(&devnet_rename_seq);
1167 memcpy(oldname, dev->name, IFNAMSIZ);
1169 err = dev_get_valid_name(net, dev, newname);
1171 write_seqcount_end(&devnet_rename_seq);
1175 if (oldname[0] && !strchr(oldname, '%'))
1176 netdev_info(dev, "renamed from %s\n", oldname);
1178 old_assign_type = dev->name_assign_type;
1179 dev->name_assign_type = NET_NAME_RENAMED;
1182 ret = device_rename(&dev->dev, dev->name);
1184 memcpy(dev->name, oldname, IFNAMSIZ);
1185 dev->name_assign_type = old_assign_type;
1186 write_seqcount_end(&devnet_rename_seq);
1190 write_seqcount_end(&devnet_rename_seq);
1192 netdev_adjacent_rename_links(dev, oldname);
1194 write_lock_bh(&dev_base_lock);
1195 hlist_del_rcu(&dev->name_hlist);
1196 write_unlock_bh(&dev_base_lock);
1200 write_lock_bh(&dev_base_lock);
1201 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1202 write_unlock_bh(&dev_base_lock);
1204 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1205 ret = notifier_to_errno(ret);
1208 /* err >= 0 after dev_alloc_name() or stores the first errno */
1211 write_seqcount_begin(&devnet_rename_seq);
1212 memcpy(dev->name, oldname, IFNAMSIZ);
1213 memcpy(oldname, newname, IFNAMSIZ);
1214 dev->name_assign_type = old_assign_type;
1215 old_assign_type = NET_NAME_RENAMED;
1218 pr_err("%s: name change rollback failed: %d\n",
1227 * dev_set_alias - change ifalias of a device
1229 * @alias: name up to IFALIASZ
1230 * @len: limit of bytes to copy from info
1232 * Set ifalias for a device,
1234 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1240 if (len >= IFALIASZ)
1244 kfree(dev->ifalias);
1245 dev->ifalias = NULL;
1249 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1252 dev->ifalias = new_ifalias;
1254 strlcpy(dev->ifalias, alias, len+1);
1260 * netdev_features_change - device changes features
1261 * @dev: device to cause notification
1263 * Called to indicate a device has changed features.
1265 void netdev_features_change(struct net_device *dev)
1267 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1269 EXPORT_SYMBOL(netdev_features_change);
1272 * netdev_state_change - device changes state
1273 * @dev: device to cause notification
1275 * Called to indicate a device has changed state. This function calls
1276 * the notifier chains for netdev_chain and sends a NEWLINK message
1277 * to the routing socket.
1279 void netdev_state_change(struct net_device *dev)
1281 if (dev->flags & IFF_UP) {
1282 struct netdev_notifier_change_info change_info;
1284 change_info.flags_changed = 0;
1285 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1287 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1290 EXPORT_SYMBOL(netdev_state_change);
1293 * netdev_notify_peers - notify network peers about existence of @dev
1294 * @dev: network device
1296 * Generate traffic such that interested network peers are aware of
1297 * @dev, such as by generating a gratuitous ARP. This may be used when
1298 * a device wants to inform the rest of the network about some sort of
1299 * reconfiguration such as a failover event or virtual machine
1302 void netdev_notify_peers(struct net_device *dev)
1305 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1308 EXPORT_SYMBOL(netdev_notify_peers);
1310 static int __dev_open(struct net_device *dev)
1312 const struct net_device_ops *ops = dev->netdev_ops;
1317 if (!netif_device_present(dev))
1320 /* Block netpoll from trying to do any rx path servicing.
1321 * If we don't do this there is a chance ndo_poll_controller
1322 * or ndo_poll may be running while we open the device
1324 netpoll_poll_disable(dev);
1326 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1327 ret = notifier_to_errno(ret);
1331 set_bit(__LINK_STATE_START, &dev->state);
1333 if (ops->ndo_validate_addr)
1334 ret = ops->ndo_validate_addr(dev);
1336 if (!ret && ops->ndo_open)
1337 ret = ops->ndo_open(dev);
1339 netpoll_poll_enable(dev);
1342 clear_bit(__LINK_STATE_START, &dev->state);
1344 dev->flags |= IFF_UP;
1345 dev_set_rx_mode(dev);
1347 add_device_randomness(dev->dev_addr, dev->addr_len);
1354 * dev_open - prepare an interface for use.
1355 * @dev: device to open
1357 * Takes a device from down to up state. The device's private open
1358 * function is invoked and then the multicast lists are loaded. Finally
1359 * the device is moved into the up state and a %NETDEV_UP message is
1360 * sent to the netdev notifier chain.
1362 * Calling this function on an active interface is a nop. On a failure
1363 * a negative errno code is returned.
1365 int dev_open(struct net_device *dev)
1369 if (dev->flags & IFF_UP)
1372 ret = __dev_open(dev);
1376 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1377 call_netdevice_notifiers(NETDEV_UP, dev);
1381 EXPORT_SYMBOL(dev_open);
1383 static int __dev_close_many(struct list_head *head)
1385 struct net_device *dev;
1390 list_for_each_entry(dev, head, close_list) {
1391 /* Temporarily disable netpoll until the interface is down */
1392 netpoll_poll_disable(dev);
1394 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1396 clear_bit(__LINK_STATE_START, &dev->state);
1398 /* Synchronize to scheduled poll. We cannot touch poll list, it
1399 * can be even on different cpu. So just clear netif_running().
1401 * dev->stop() will invoke napi_disable() on all of it's
1402 * napi_struct instances on this device.
1404 smp_mb__after_atomic(); /* Commit netif_running(). */
1407 dev_deactivate_many(head);
1409 list_for_each_entry(dev, head, close_list) {
1410 const struct net_device_ops *ops = dev->netdev_ops;
1413 * Call the device specific close. This cannot fail.
1414 * Only if device is UP
1416 * We allow it to be called even after a DETACH hot-plug
1422 dev->flags &= ~IFF_UP;
1423 netpoll_poll_enable(dev);
1429 static int __dev_close(struct net_device *dev)
1434 list_add(&dev->close_list, &single);
1435 retval = __dev_close_many(&single);
1441 int dev_close_many(struct list_head *head, bool unlink)
1443 struct net_device *dev, *tmp;
1445 /* Remove the devices that don't need to be closed */
1446 list_for_each_entry_safe(dev, tmp, head, close_list)
1447 if (!(dev->flags & IFF_UP))
1448 list_del_init(&dev->close_list);
1450 __dev_close_many(head);
1452 list_for_each_entry_safe(dev, tmp, head, close_list) {
1453 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1454 call_netdevice_notifiers(NETDEV_DOWN, dev);
1456 list_del_init(&dev->close_list);
1461 EXPORT_SYMBOL(dev_close_many);
1464 * dev_close - shutdown an interface.
1465 * @dev: device to shutdown
1467 * This function moves an active device into down state. A
1468 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1469 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1472 int dev_close(struct net_device *dev)
1474 if (dev->flags & IFF_UP) {
1477 list_add(&dev->close_list, &single);
1478 dev_close_many(&single, true);
1483 EXPORT_SYMBOL(dev_close);
1487 * dev_disable_lro - disable Large Receive Offload on a device
1490 * Disable Large Receive Offload (LRO) on a net device. Must be
1491 * called under RTNL. This is needed if received packets may be
1492 * forwarded to another interface.
1494 void dev_disable_lro(struct net_device *dev)
1496 struct net_device *lower_dev;
1497 struct list_head *iter;
1499 dev->wanted_features &= ~NETIF_F_LRO;
1500 netdev_update_features(dev);
1502 if (unlikely(dev->features & NETIF_F_LRO))
1503 netdev_WARN(dev, "failed to disable LRO!\n");
1505 netdev_for_each_lower_dev(dev, lower_dev, iter)
1506 dev_disable_lro(lower_dev);
1508 EXPORT_SYMBOL(dev_disable_lro);
1510 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1511 struct net_device *dev)
1513 struct netdev_notifier_info info;
1515 netdev_notifier_info_init(&info, dev);
1516 return nb->notifier_call(nb, val, &info);
1519 static int dev_boot_phase = 1;
1522 * register_netdevice_notifier - register a network notifier block
1525 * Register a notifier to be called when network device events occur.
1526 * The notifier passed is linked into the kernel structures and must
1527 * not be reused until it has been unregistered. A negative errno code
1528 * is returned on a failure.
1530 * When registered all registration and up events are replayed
1531 * to the new notifier to allow device to have a race free
1532 * view of the network device list.
1535 int register_netdevice_notifier(struct notifier_block *nb)
1537 struct net_device *dev;
1538 struct net_device *last;
1543 err = raw_notifier_chain_register(&netdev_chain, nb);
1549 for_each_netdev(net, dev) {
1550 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1551 err = notifier_to_errno(err);
1555 if (!(dev->flags & IFF_UP))
1558 call_netdevice_notifier(nb, NETDEV_UP, dev);
1569 for_each_netdev(net, dev) {
1573 if (dev->flags & IFF_UP) {
1574 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1576 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1578 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1583 raw_notifier_chain_unregister(&netdev_chain, nb);
1586 EXPORT_SYMBOL(register_netdevice_notifier);
1589 * unregister_netdevice_notifier - unregister a network notifier block
1592 * Unregister a notifier previously registered by
1593 * register_netdevice_notifier(). The notifier is unlinked into the
1594 * kernel structures and may then be reused. A negative errno code
1595 * is returned on a failure.
1597 * After unregistering unregister and down device events are synthesized
1598 * for all devices on the device list to the removed notifier to remove
1599 * the need for special case cleanup code.
1602 int unregister_netdevice_notifier(struct notifier_block *nb)
1604 struct net_device *dev;
1609 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1614 for_each_netdev(net, dev) {
1615 if (dev->flags & IFF_UP) {
1616 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1618 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1620 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1627 EXPORT_SYMBOL(unregister_netdevice_notifier);
1630 * call_netdevice_notifiers_info - call all network notifier blocks
1631 * @val: value passed unmodified to notifier function
1632 * @dev: net_device pointer passed unmodified to notifier function
1633 * @info: notifier information data
1635 * Call all network notifier blocks. Parameters and return value
1636 * are as for raw_notifier_call_chain().
1639 static int call_netdevice_notifiers_info(unsigned long val,
1640 struct net_device *dev,
1641 struct netdev_notifier_info *info)
1644 netdev_notifier_info_init(info, dev);
1645 return raw_notifier_call_chain(&netdev_chain, val, info);
1649 * call_netdevice_notifiers - call all network notifier blocks
1650 * @val: value passed unmodified to notifier function
1651 * @dev: net_device pointer passed unmodified to notifier function
1653 * Call all network notifier blocks. Parameters and return value
1654 * are as for raw_notifier_call_chain().
1657 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1659 struct netdev_notifier_info info;
1661 return call_netdevice_notifiers_info(val, dev, &info);
1663 EXPORT_SYMBOL(call_netdevice_notifiers);
1665 #ifdef CONFIG_NET_INGRESS
1666 static struct static_key ingress_needed __read_mostly;
1668 void net_inc_ingress_queue(void)
1670 static_key_slow_inc(&ingress_needed);
1672 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1674 void net_dec_ingress_queue(void)
1676 static_key_slow_dec(&ingress_needed);
1678 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1681 #ifdef CONFIG_NET_EGRESS
1682 static struct static_key egress_needed __read_mostly;
1684 void net_inc_egress_queue(void)
1686 static_key_slow_inc(&egress_needed);
1688 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1690 void net_dec_egress_queue(void)
1692 static_key_slow_dec(&egress_needed);
1694 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1697 static struct static_key netstamp_needed __read_mostly;
1698 #ifdef HAVE_JUMP_LABEL
1699 /* We are not allowed to call static_key_slow_dec() from irq context
1700 * If net_disable_timestamp() is called from irq context, defer the
1701 * static_key_slow_dec() calls.
1703 static atomic_t netstamp_needed_deferred;
1706 void net_enable_timestamp(void)
1708 #ifdef HAVE_JUMP_LABEL
1709 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1713 static_key_slow_dec(&netstamp_needed);
1717 static_key_slow_inc(&netstamp_needed);
1719 EXPORT_SYMBOL(net_enable_timestamp);
1721 void net_disable_timestamp(void)
1723 #ifdef HAVE_JUMP_LABEL
1724 if (in_interrupt()) {
1725 atomic_inc(&netstamp_needed_deferred);
1729 static_key_slow_dec(&netstamp_needed);
1731 EXPORT_SYMBOL(net_disable_timestamp);
1733 static inline void net_timestamp_set(struct sk_buff *skb)
1735 skb->tstamp.tv64 = 0;
1736 if (static_key_false(&netstamp_needed))
1737 __net_timestamp(skb);
1740 #define net_timestamp_check(COND, SKB) \
1741 if (static_key_false(&netstamp_needed)) { \
1742 if ((COND) && !(SKB)->tstamp.tv64) \
1743 __net_timestamp(SKB); \
1746 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1750 if (!(dev->flags & IFF_UP))
1753 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1754 if (skb->len <= len)
1757 /* if TSO is enabled, we don't care about the length as the packet
1758 * could be forwarded without being segmented before
1760 if (skb_is_gso(skb))
1765 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1767 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1769 if (skb_orphan_frags(skb, GFP_ATOMIC) ||
1770 unlikely(!is_skb_forwardable(dev, skb))) {
1771 atomic_long_inc(&dev->rx_dropped);
1776 skb_scrub_packet(skb, true);
1778 skb->protocol = eth_type_trans(skb, dev);
1779 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1783 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1786 * dev_forward_skb - loopback an skb to another netif
1788 * @dev: destination network device
1789 * @skb: buffer to forward
1792 * NET_RX_SUCCESS (no congestion)
1793 * NET_RX_DROP (packet was dropped, but freed)
1795 * dev_forward_skb can be used for injecting an skb from the
1796 * start_xmit function of one device into the receive queue
1797 * of another device.
1799 * The receiving device may be in another namespace, so
1800 * we have to clear all information in the skb that could
1801 * impact namespace isolation.
1803 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1805 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1807 EXPORT_SYMBOL_GPL(dev_forward_skb);
1809 static inline int deliver_skb(struct sk_buff *skb,
1810 struct packet_type *pt_prev,
1811 struct net_device *orig_dev)
1813 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1815 atomic_inc(&skb->users);
1816 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1819 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1820 struct packet_type **pt,
1821 struct net_device *orig_dev,
1823 struct list_head *ptype_list)
1825 struct packet_type *ptype, *pt_prev = *pt;
1827 list_for_each_entry_rcu(ptype, ptype_list, list) {
1828 if (ptype->type != type)
1831 deliver_skb(skb, pt_prev, orig_dev);
1837 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1839 if (!ptype->af_packet_priv || !skb->sk)
1842 if (ptype->id_match)
1843 return ptype->id_match(ptype, skb->sk);
1844 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1851 * Support routine. Sends outgoing frames to any network
1852 * taps currently in use.
1855 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1857 struct packet_type *ptype;
1858 struct sk_buff *skb2 = NULL;
1859 struct packet_type *pt_prev = NULL;
1860 struct list_head *ptype_list = &ptype_all;
1864 list_for_each_entry_rcu(ptype, ptype_list, list) {
1865 /* Never send packets back to the socket
1866 * they originated from - MvS (miquels@drinkel.ow.org)
1868 if (skb_loop_sk(ptype, skb))
1872 deliver_skb(skb2, pt_prev, skb->dev);
1877 /* need to clone skb, done only once */
1878 skb2 = skb_clone(skb, GFP_ATOMIC);
1882 net_timestamp_set(skb2);
1884 /* skb->nh should be correctly
1885 * set by sender, so that the second statement is
1886 * just protection against buggy protocols.
1888 skb_reset_mac_header(skb2);
1890 if (skb_network_header(skb2) < skb2->data ||
1891 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1892 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1893 ntohs(skb2->protocol),
1895 skb_reset_network_header(skb2);
1898 skb2->transport_header = skb2->network_header;
1899 skb2->pkt_type = PACKET_OUTGOING;
1903 if (ptype_list == &ptype_all) {
1904 ptype_list = &dev->ptype_all;
1909 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1912 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
1915 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1916 * @dev: Network device
1917 * @txq: number of queues available
1919 * If real_num_tx_queues is changed the tc mappings may no longer be
1920 * valid. To resolve this verify the tc mapping remains valid and if
1921 * not NULL the mapping. With no priorities mapping to this
1922 * offset/count pair it will no longer be used. In the worst case TC0
1923 * is invalid nothing can be done so disable priority mappings. If is
1924 * expected that drivers will fix this mapping if they can before
1925 * calling netif_set_real_num_tx_queues.
1927 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1930 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1932 /* If TC0 is invalidated disable TC mapping */
1933 if (tc->offset + tc->count > txq) {
1934 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1939 /* Invalidated prio to tc mappings set to TC0 */
1940 for (i = 1; i < TC_BITMASK + 1; i++) {
1941 int q = netdev_get_prio_tc_map(dev, i);
1943 tc = &dev->tc_to_txq[q];
1944 if (tc->offset + tc->count > txq) {
1945 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1947 netdev_set_prio_tc_map(dev, i, 0);
1953 static DEFINE_MUTEX(xps_map_mutex);
1954 #define xmap_dereference(P) \
1955 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1957 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1960 struct xps_map *map = NULL;
1964 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1966 for (pos = 0; map && pos < map->len; pos++) {
1967 if (map->queues[pos] == index) {
1969 map->queues[pos] = map->queues[--map->len];
1971 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1972 kfree_rcu(map, rcu);
1982 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1984 struct xps_dev_maps *dev_maps;
1986 bool active = false;
1988 mutex_lock(&xps_map_mutex);
1989 dev_maps = xmap_dereference(dev->xps_maps);
1994 for_each_possible_cpu(cpu) {
1995 for (i = index; i < dev->num_tx_queues; i++) {
1996 if (!remove_xps_queue(dev_maps, cpu, i))
1999 if (i == dev->num_tx_queues)
2004 RCU_INIT_POINTER(dev->xps_maps, NULL);
2005 kfree_rcu(dev_maps, rcu);
2008 for (i = index; i < dev->num_tx_queues; i++)
2009 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2013 mutex_unlock(&xps_map_mutex);
2016 static struct xps_map *expand_xps_map(struct xps_map *map,
2019 struct xps_map *new_map;
2020 int alloc_len = XPS_MIN_MAP_ALLOC;
2023 for (pos = 0; map && pos < map->len; pos++) {
2024 if (map->queues[pos] != index)
2029 /* Need to add queue to this CPU's existing map */
2031 if (pos < map->alloc_len)
2034 alloc_len = map->alloc_len * 2;
2037 /* Need to allocate new map to store queue on this CPU's map */
2038 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2043 for (i = 0; i < pos; i++)
2044 new_map->queues[i] = map->queues[i];
2045 new_map->alloc_len = alloc_len;
2051 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2054 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2055 struct xps_map *map, *new_map;
2056 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2057 int cpu, numa_node_id = -2;
2058 bool active = false;
2060 mutex_lock(&xps_map_mutex);
2062 dev_maps = xmap_dereference(dev->xps_maps);
2064 /* allocate memory for queue storage */
2065 for_each_online_cpu(cpu) {
2066 if (!cpumask_test_cpu(cpu, mask))
2070 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2071 if (!new_dev_maps) {
2072 mutex_unlock(&xps_map_mutex);
2076 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2079 map = expand_xps_map(map, cpu, index);
2083 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2087 goto out_no_new_maps;
2089 for_each_possible_cpu(cpu) {
2090 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2091 /* add queue to CPU maps */
2094 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2095 while ((pos < map->len) && (map->queues[pos] != index))
2098 if (pos == map->len)
2099 map->queues[map->len++] = index;
2101 if (numa_node_id == -2)
2102 numa_node_id = cpu_to_node(cpu);
2103 else if (numa_node_id != cpu_to_node(cpu))
2106 } else if (dev_maps) {
2107 /* fill in the new device map from the old device map */
2108 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2109 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2114 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2116 /* Cleanup old maps */
2118 for_each_possible_cpu(cpu) {
2119 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2120 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2121 if (map && map != new_map)
2122 kfree_rcu(map, rcu);
2125 kfree_rcu(dev_maps, rcu);
2128 dev_maps = new_dev_maps;
2132 /* update Tx queue numa node */
2133 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2134 (numa_node_id >= 0) ? numa_node_id :
2140 /* removes queue from unused CPUs */
2141 for_each_possible_cpu(cpu) {
2142 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2145 if (remove_xps_queue(dev_maps, cpu, index))
2149 /* free map if not active */
2151 RCU_INIT_POINTER(dev->xps_maps, NULL);
2152 kfree_rcu(dev_maps, rcu);
2156 mutex_unlock(&xps_map_mutex);
2160 /* remove any maps that we added */
2161 for_each_possible_cpu(cpu) {
2162 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2163 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2165 if (new_map && new_map != map)
2169 mutex_unlock(&xps_map_mutex);
2171 kfree(new_dev_maps);
2174 EXPORT_SYMBOL(netif_set_xps_queue);
2178 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2179 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2181 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2185 if (txq < 1 || txq > dev->num_tx_queues)
2188 if (dev->reg_state == NETREG_REGISTERED ||
2189 dev->reg_state == NETREG_UNREGISTERING) {
2192 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2198 netif_setup_tc(dev, txq);
2200 if (txq < dev->real_num_tx_queues) {
2201 qdisc_reset_all_tx_gt(dev, txq);
2203 netif_reset_xps_queues_gt(dev, txq);
2208 dev->real_num_tx_queues = txq;
2211 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2215 * netif_set_real_num_rx_queues - set actual number of RX queues used
2216 * @dev: Network device
2217 * @rxq: Actual number of RX queues
2219 * This must be called either with the rtnl_lock held or before
2220 * registration of the net device. Returns 0 on success, or a
2221 * negative error code. If called before registration, it always
2224 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2228 if (rxq < 1 || rxq > dev->num_rx_queues)
2231 if (dev->reg_state == NETREG_REGISTERED) {
2234 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2240 dev->real_num_rx_queues = rxq;
2243 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2247 * netif_get_num_default_rss_queues - default number of RSS queues
2249 * This routine should set an upper limit on the number of RSS queues
2250 * used by default by multiqueue devices.
2252 int netif_get_num_default_rss_queues(void)
2254 return is_kdump_kernel() ?
2255 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2257 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2259 static void __netif_reschedule(struct Qdisc *q)
2261 struct softnet_data *sd;
2262 unsigned long flags;
2264 local_irq_save(flags);
2265 sd = this_cpu_ptr(&softnet_data);
2266 q->next_sched = NULL;
2267 *sd->output_queue_tailp = q;
2268 sd->output_queue_tailp = &q->next_sched;
2269 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2270 local_irq_restore(flags);
2273 void __netif_schedule(struct Qdisc *q)
2275 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2276 __netif_reschedule(q);
2278 EXPORT_SYMBOL(__netif_schedule);
2280 struct dev_kfree_skb_cb {
2281 enum skb_free_reason reason;
2284 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2286 return (struct dev_kfree_skb_cb *)skb->cb;
2289 void netif_schedule_queue(struct netdev_queue *txq)
2292 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2293 struct Qdisc *q = rcu_dereference(txq->qdisc);
2295 __netif_schedule(q);
2299 EXPORT_SYMBOL(netif_schedule_queue);
2302 * netif_wake_subqueue - allow sending packets on subqueue
2303 * @dev: network device
2304 * @queue_index: sub queue index
2306 * Resume individual transmit queue of a device with multiple transmit queues.
2308 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2310 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2312 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2316 q = rcu_dereference(txq->qdisc);
2317 __netif_schedule(q);
2321 EXPORT_SYMBOL(netif_wake_subqueue);
2323 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2325 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2329 q = rcu_dereference(dev_queue->qdisc);
2330 __netif_schedule(q);
2334 EXPORT_SYMBOL(netif_tx_wake_queue);
2336 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2338 unsigned long flags;
2340 if (likely(atomic_read(&skb->users) == 1)) {
2342 atomic_set(&skb->users, 0);
2343 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2346 get_kfree_skb_cb(skb)->reason = reason;
2347 local_irq_save(flags);
2348 skb->next = __this_cpu_read(softnet_data.completion_queue);
2349 __this_cpu_write(softnet_data.completion_queue, skb);
2350 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2351 local_irq_restore(flags);
2353 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2355 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2357 if (in_irq() || irqs_disabled())
2358 __dev_kfree_skb_irq(skb, reason);
2362 EXPORT_SYMBOL(__dev_kfree_skb_any);
2366 * netif_device_detach - mark device as removed
2367 * @dev: network device
2369 * Mark device as removed from system and therefore no longer available.
2371 void netif_device_detach(struct net_device *dev)
2373 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2374 netif_running(dev)) {
2375 netif_tx_stop_all_queues(dev);
2378 EXPORT_SYMBOL(netif_device_detach);
2381 * netif_device_attach - mark device as attached
2382 * @dev: network device
2384 * Mark device as attached from system and restart if needed.
2386 void netif_device_attach(struct net_device *dev)
2388 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2389 netif_running(dev)) {
2390 netif_tx_wake_all_queues(dev);
2391 __netdev_watchdog_up(dev);
2394 EXPORT_SYMBOL(netif_device_attach);
2397 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2398 * to be used as a distribution range.
2400 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2401 unsigned int num_tx_queues)
2405 u16 qcount = num_tx_queues;
2407 if (skb_rx_queue_recorded(skb)) {
2408 hash = skb_get_rx_queue(skb);
2409 while (unlikely(hash >= num_tx_queues))
2410 hash -= num_tx_queues;
2415 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2416 qoffset = dev->tc_to_txq[tc].offset;
2417 qcount = dev->tc_to_txq[tc].count;
2420 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2422 EXPORT_SYMBOL(__skb_tx_hash);
2424 static void skb_warn_bad_offload(const struct sk_buff *skb)
2426 static const netdev_features_t null_features;
2427 struct net_device *dev = skb->dev;
2428 const char *name = "";
2430 if (!net_ratelimit())
2434 if (dev->dev.parent)
2435 name = dev_driver_string(dev->dev.parent);
2437 name = netdev_name(dev);
2439 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2440 "gso_type=%d ip_summed=%d\n",
2441 name, dev ? &dev->features : &null_features,
2442 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2443 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2444 skb_shinfo(skb)->gso_type, skb->ip_summed);
2448 * Invalidate hardware checksum when packet is to be mangled, and
2449 * complete checksum manually on outgoing path.
2451 int skb_checksum_help(struct sk_buff *skb)
2454 int ret = 0, offset;
2456 if (skb->ip_summed == CHECKSUM_COMPLETE)
2457 goto out_set_summed;
2459 if (unlikely(skb_shinfo(skb)->gso_size)) {
2460 skb_warn_bad_offload(skb);
2464 /* Before computing a checksum, we should make sure no frag could
2465 * be modified by an external entity : checksum could be wrong.
2467 if (skb_has_shared_frag(skb)) {
2468 ret = __skb_linearize(skb);
2473 offset = skb_checksum_start_offset(skb);
2474 BUG_ON(offset >= skb_headlen(skb));
2475 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2477 offset += skb->csum_offset;
2478 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2480 if (skb_cloned(skb) &&
2481 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2482 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2487 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2489 skb->ip_summed = CHECKSUM_NONE;
2493 EXPORT_SYMBOL(skb_checksum_help);
2495 /* skb_csum_offload_check - Driver helper function to determine if a device
2496 * with limited checksum offload capabilities is able to offload the checksum
2497 * for a given packet.
2500 * skb - sk_buff for the packet in question
2501 * spec - contains the description of what device can offload
2502 * csum_encapped - returns true if the checksum being offloaded is
2503 * encpasulated. That is it is checksum for the transport header
2504 * in the inner headers.
2505 * checksum_help - when set indicates that helper function should
2506 * call skb_checksum_help if offload checks fail
2509 * true: Packet has passed the checksum checks and should be offloadable to
2510 * the device (a driver may still need to check for additional
2511 * restrictions of its device)
2512 * false: Checksum is not offloadable. If checksum_help was set then
2513 * skb_checksum_help was called to resolve checksum for non-GSO
2514 * packets and when IP protocol is not SCTP
2516 bool __skb_csum_offload_chk(struct sk_buff *skb,
2517 const struct skb_csum_offl_spec *spec,
2518 bool *csum_encapped,
2522 struct ipv6hdr *ipv6;
2527 if (skb->protocol == htons(ETH_P_8021Q) ||
2528 skb->protocol == htons(ETH_P_8021AD)) {
2529 if (!spec->vlan_okay)
2533 /* We check whether the checksum refers to a transport layer checksum in
2534 * the outermost header or an encapsulated transport layer checksum that
2535 * corresponds to the inner headers of the skb. If the checksum is for
2536 * something else in the packet we need help.
2538 if (skb_checksum_start_offset(skb) == skb_transport_offset(skb)) {
2539 /* Non-encapsulated checksum */
2540 protocol = eproto_to_ipproto(vlan_get_protocol(skb));
2541 nhdr = skb_network_header(skb);
2542 *csum_encapped = false;
2543 if (spec->no_not_encapped)
2545 } else if (skb->encapsulation && spec->encap_okay &&
2546 skb_checksum_start_offset(skb) ==
2547 skb_inner_transport_offset(skb)) {
2548 /* Encapsulated checksum */
2549 *csum_encapped = true;
2550 switch (skb->inner_protocol_type) {
2551 case ENCAP_TYPE_ETHER:
2552 protocol = eproto_to_ipproto(skb->inner_protocol);
2554 case ENCAP_TYPE_IPPROTO:
2555 protocol = skb->inner_protocol;
2558 nhdr = skb_inner_network_header(skb);
2565 if (!spec->ipv4_okay)
2568 ip_proto = iph->protocol;
2569 if (iph->ihl != 5 && !spec->ip_options_okay)
2573 if (!spec->ipv6_okay)
2575 if (spec->no_encapped_ipv6 && *csum_encapped)
2578 nhdr += sizeof(*ipv6);
2579 ip_proto = ipv6->nexthdr;
2588 if (!spec->tcp_okay ||
2589 skb->csum_offset != offsetof(struct tcphdr, check))
2593 if (!spec->udp_okay ||
2594 skb->csum_offset != offsetof(struct udphdr, check))
2598 if (!spec->sctp_okay ||
2599 skb->csum_offset != offsetof(struct sctphdr, checksum))
2603 case NEXTHDR_ROUTING:
2604 case NEXTHDR_DEST: {
2607 if (protocol != IPPROTO_IPV6 || !spec->ext_hdrs_okay)
2610 ip_proto = opthdr[0];
2611 nhdr += (opthdr[1] + 1) << 3;
2613 goto ip_proto_again;
2619 /* Passed the tests for offloading checksum */
2623 if (csum_help && !skb_shinfo(skb)->gso_size)
2624 skb_checksum_help(skb);
2628 EXPORT_SYMBOL(__skb_csum_offload_chk);
2630 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2632 __be16 type = skb->protocol;
2634 /* Tunnel gso handlers can set protocol to ethernet. */
2635 if (type == htons(ETH_P_TEB)) {
2638 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2641 eth = (struct ethhdr *)skb_mac_header(skb);
2642 type = eth->h_proto;
2645 return __vlan_get_protocol(skb, type, depth);
2649 * skb_mac_gso_segment - mac layer segmentation handler.
2650 * @skb: buffer to segment
2651 * @features: features for the output path (see dev->features)
2653 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2654 netdev_features_t features)
2656 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2657 struct packet_offload *ptype;
2658 int vlan_depth = skb->mac_len;
2659 __be16 type = skb_network_protocol(skb, &vlan_depth);
2661 if (unlikely(!type))
2662 return ERR_PTR(-EINVAL);
2664 __skb_pull(skb, vlan_depth);
2667 list_for_each_entry_rcu(ptype, &offload_base, list) {
2668 if (ptype->type == type && ptype->callbacks.gso_segment) {
2669 segs = ptype->callbacks.gso_segment(skb, features);
2675 __skb_push(skb, skb->data - skb_mac_header(skb));
2679 EXPORT_SYMBOL(skb_mac_gso_segment);
2682 /* openvswitch calls this on rx path, so we need a different check.
2684 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2687 return skb->ip_summed != CHECKSUM_PARTIAL;
2689 return skb->ip_summed == CHECKSUM_NONE;
2693 * __skb_gso_segment - Perform segmentation on skb.
2694 * @skb: buffer to segment
2695 * @features: features for the output path (see dev->features)
2696 * @tx_path: whether it is called in TX path
2698 * This function segments the given skb and returns a list of segments.
2700 * It may return NULL if the skb requires no segmentation. This is
2701 * only possible when GSO is used for verifying header integrity.
2703 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2705 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2706 netdev_features_t features, bool tx_path)
2708 if (unlikely(skb_needs_check(skb, tx_path))) {
2711 skb_warn_bad_offload(skb);
2713 err = skb_cow_head(skb, 0);
2715 return ERR_PTR(err);
2718 /* Only report GSO partial support if it will enable us to
2719 * support segmentation on this frame without needing additional
2722 if (features & NETIF_F_GSO_PARTIAL) {
2723 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
2724 struct net_device *dev = skb->dev;
2726 partial_features |= dev->features & dev->gso_partial_features;
2727 if (!skb_gso_ok(skb, features | partial_features))
2728 features &= ~NETIF_F_GSO_PARTIAL;
2731 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2732 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2734 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2735 SKB_GSO_CB(skb)->encap_level = 0;
2737 skb_reset_mac_header(skb);
2738 skb_reset_mac_len(skb);
2740 return skb_mac_gso_segment(skb, features);
2742 EXPORT_SYMBOL(__skb_gso_segment);
2744 /* Take action when hardware reception checksum errors are detected. */
2746 void netdev_rx_csum_fault(struct net_device *dev)
2748 if (net_ratelimit()) {
2749 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2753 EXPORT_SYMBOL(netdev_rx_csum_fault);
2756 /* Actually, we should eliminate this check as soon as we know, that:
2757 * 1. IOMMU is present and allows to map all the memory.
2758 * 2. No high memory really exists on this machine.
2761 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2763 #ifdef CONFIG_HIGHMEM
2765 if (!(dev->features & NETIF_F_HIGHDMA)) {
2766 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2767 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2768 if (PageHighMem(skb_frag_page(frag)))
2773 if (PCI_DMA_BUS_IS_PHYS) {
2774 struct device *pdev = dev->dev.parent;
2778 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2779 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2780 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2781 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2789 /* If MPLS offload request, verify we are testing hardware MPLS features
2790 * instead of standard features for the netdev.
2792 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2793 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2794 netdev_features_t features,
2797 if (eth_p_mpls(type))
2798 features &= skb->dev->mpls_features;
2803 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2804 netdev_features_t features,
2811 static netdev_features_t harmonize_features(struct sk_buff *skb,
2812 netdev_features_t features)
2817 type = skb_network_protocol(skb, &tmp);
2818 features = net_mpls_features(skb, features, type);
2820 if (skb->ip_summed != CHECKSUM_NONE &&
2821 !can_checksum_protocol(features, type)) {
2822 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2823 } else if (illegal_highdma(skb->dev, skb)) {
2824 features &= ~NETIF_F_SG;
2830 netdev_features_t passthru_features_check(struct sk_buff *skb,
2831 struct net_device *dev,
2832 netdev_features_t features)
2836 EXPORT_SYMBOL(passthru_features_check);
2838 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2839 struct net_device *dev,
2840 netdev_features_t features)
2842 return vlan_features_check(skb, features);
2845 static netdev_features_t gso_features_check(const struct sk_buff *skb,
2846 struct net_device *dev,
2847 netdev_features_t features)
2849 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2851 if (gso_segs > dev->gso_max_segs)
2852 return features & ~NETIF_F_GSO_MASK;
2854 /* Support for GSO partial features requires software
2855 * intervention before we can actually process the packets
2856 * so we need to strip support for any partial features now
2857 * and we can pull them back in after we have partially
2858 * segmented the frame.
2860 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
2861 features &= ~dev->gso_partial_features;
2863 /* Make sure to clear the IPv4 ID mangling feature if the
2864 * IPv4 header has the potential to be fragmented.
2866 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2867 struct iphdr *iph = skb->encapsulation ?
2868 inner_ip_hdr(skb) : ip_hdr(skb);
2870 if (!(iph->frag_off & htons(IP_DF)))
2871 features &= ~NETIF_F_TSO_MANGLEID;
2877 netdev_features_t netif_skb_features(struct sk_buff *skb)
2879 struct net_device *dev = skb->dev;
2880 netdev_features_t features = dev->features;
2882 if (skb_is_gso(skb))
2883 features = gso_features_check(skb, dev, features);
2885 /* If encapsulation offload request, verify we are testing
2886 * hardware encapsulation features instead of standard
2887 * features for the netdev
2889 if (skb->encapsulation)
2890 features &= dev->hw_enc_features;
2892 if (skb_vlan_tagged(skb))
2893 features = netdev_intersect_features(features,
2894 dev->vlan_features |
2895 NETIF_F_HW_VLAN_CTAG_TX |
2896 NETIF_F_HW_VLAN_STAG_TX);
2898 if (dev->netdev_ops->ndo_features_check)
2899 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2902 features &= dflt_features_check(skb, dev, features);
2904 return harmonize_features(skb, features);
2906 EXPORT_SYMBOL(netif_skb_features);
2908 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2909 struct netdev_queue *txq, bool more)
2914 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2915 dev_queue_xmit_nit(skb, dev);
2918 trace_net_dev_start_xmit(skb, dev);
2919 rc = netdev_start_xmit(skb, dev, txq, more);
2920 trace_net_dev_xmit(skb, rc, dev, len);
2925 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2926 struct netdev_queue *txq, int *ret)
2928 struct sk_buff *skb = first;
2929 int rc = NETDEV_TX_OK;
2932 struct sk_buff *next = skb->next;
2935 rc = xmit_one(skb, dev, txq, next != NULL);
2936 if (unlikely(!dev_xmit_complete(rc))) {
2942 if (netif_xmit_stopped(txq) && skb) {
2943 rc = NETDEV_TX_BUSY;
2953 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2954 netdev_features_t features)
2956 if (skb_vlan_tag_present(skb) &&
2957 !vlan_hw_offload_capable(features, skb->vlan_proto))
2958 skb = __vlan_hwaccel_push_inside(skb);
2962 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2964 netdev_features_t features;
2966 features = netif_skb_features(skb);
2967 skb = validate_xmit_vlan(skb, features);
2971 if (netif_needs_gso(skb, features)) {
2972 struct sk_buff *segs;
2974 segs = skb_gso_segment(skb, features);
2982 if (skb_needs_linearize(skb, features) &&
2983 __skb_linearize(skb))
2986 /* If packet is not checksummed and device does not
2987 * support checksumming for this protocol, complete
2988 * checksumming here.
2990 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2991 if (skb->encapsulation)
2992 skb_set_inner_transport_header(skb,
2993 skb_checksum_start_offset(skb));
2995 skb_set_transport_header(skb,
2996 skb_checksum_start_offset(skb));
2997 if (!(features & NETIF_F_CSUM_MASK) &&
2998 skb_checksum_help(skb))
3008 atomic_long_inc(&dev->tx_dropped);
3012 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
3014 struct sk_buff *next, *head = NULL, *tail;
3016 for (; skb != NULL; skb = next) {
3020 /* in case skb wont be segmented, point to itself */
3023 skb = validate_xmit_skb(skb, dev);
3031 /* If skb was segmented, skb->prev points to
3032 * the last segment. If not, it still contains skb.
3039 static void qdisc_pkt_len_init(struct sk_buff *skb)
3041 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3043 qdisc_skb_cb(skb)->pkt_len = skb->len;
3045 /* To get more precise estimation of bytes sent on wire,
3046 * we add to pkt_len the headers size of all segments
3048 if (shinfo->gso_size) {
3049 unsigned int hdr_len;
3050 u16 gso_segs = shinfo->gso_segs;
3052 /* mac layer + network layer */
3053 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3055 /* + transport layer */
3056 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
3057 hdr_len += tcp_hdrlen(skb);
3059 hdr_len += sizeof(struct udphdr);
3061 if (shinfo->gso_type & SKB_GSO_DODGY)
3062 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3065 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3069 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3070 struct net_device *dev,
3071 struct netdev_queue *txq)
3073 spinlock_t *root_lock = qdisc_lock(q);
3074 struct sk_buff *to_free = NULL;
3078 qdisc_calculate_pkt_len(skb, q);
3080 * Heuristic to force contended enqueues to serialize on a
3081 * separate lock before trying to get qdisc main lock.
3082 * This permits qdisc->running owner to get the lock more
3083 * often and dequeue packets faster.
3085 contended = qdisc_is_running(q);
3086 if (unlikely(contended))
3087 spin_lock(&q->busylock);
3089 spin_lock(root_lock);
3090 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3091 __qdisc_drop(skb, &to_free);
3093 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3094 qdisc_run_begin(q)) {
3096 * This is a work-conserving queue; there are no old skbs
3097 * waiting to be sent out; and the qdisc is not running -
3098 * xmit the skb directly.
3101 qdisc_bstats_update(q, skb);
3103 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3104 if (unlikely(contended)) {
3105 spin_unlock(&q->busylock);
3112 rc = NET_XMIT_SUCCESS;
3114 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3115 if (qdisc_run_begin(q)) {
3116 if (unlikely(contended)) {
3117 spin_unlock(&q->busylock);
3123 spin_unlock(root_lock);
3124 if (unlikely(to_free))
3125 kfree_skb_list(to_free);
3126 if (unlikely(contended))
3127 spin_unlock(&q->busylock);
3131 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3132 static void skb_update_prio(struct sk_buff *skb)
3134 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
3136 if (!skb->priority && skb->sk && map) {
3137 unsigned int prioidx =
3138 sock_cgroup_prioidx(&skb->sk->sk_cgrp_data);
3140 if (prioidx < map->priomap_len)
3141 skb->priority = map->priomap[prioidx];
3145 #define skb_update_prio(skb)
3148 DEFINE_PER_CPU(int, xmit_recursion);
3149 EXPORT_SYMBOL(xmit_recursion);
3152 * dev_loopback_xmit - loop back @skb
3153 * @net: network namespace this loopback is happening in
3154 * @sk: sk needed to be a netfilter okfn
3155 * @skb: buffer to transmit
3157 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3159 skb_reset_mac_header(skb);
3160 __skb_pull(skb, skb_network_offset(skb));
3161 skb->pkt_type = PACKET_LOOPBACK;
3162 skb->ip_summed = CHECKSUM_UNNECESSARY;
3163 WARN_ON(!skb_dst(skb));
3168 EXPORT_SYMBOL(dev_loopback_xmit);
3170 #ifdef CONFIG_NET_EGRESS
3171 static struct sk_buff *
3172 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3174 struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list);
3175 struct tcf_result cl_res;
3180 /* skb->tc_verd and qdisc_skb_cb(skb)->pkt_len were already set
3181 * earlier by the caller.
3183 qdisc_bstats_cpu_update(cl->q, skb);
3185 switch (tc_classify(skb, cl, &cl_res, false)) {
3187 case TC_ACT_RECLASSIFY:
3188 skb->tc_index = TC_H_MIN(cl_res.classid);
3191 qdisc_qstats_cpu_drop(cl->q);
3192 *ret = NET_XMIT_DROP;
3197 *ret = NET_XMIT_SUCCESS;
3200 case TC_ACT_REDIRECT:
3201 /* No need to push/pop skb's mac_header here on egress! */
3202 skb_do_redirect(skb);
3203 *ret = NET_XMIT_SUCCESS;
3211 #endif /* CONFIG_NET_EGRESS */
3213 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3216 struct xps_dev_maps *dev_maps;
3217 struct xps_map *map;
3218 int queue_index = -1;
3221 dev_maps = rcu_dereference(dev->xps_maps);
3223 map = rcu_dereference(
3224 dev_maps->cpu_map[skb->sender_cpu - 1]);
3227 queue_index = map->queues[0];
3229 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3231 if (unlikely(queue_index >= dev->real_num_tx_queues))
3243 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3245 struct sock *sk = skb->sk;
3246 int queue_index = sk_tx_queue_get(sk);
3248 if (queue_index < 0 || skb->ooo_okay ||
3249 queue_index >= dev->real_num_tx_queues) {
3250 int new_index = get_xps_queue(dev, skb);
3252 new_index = skb_tx_hash(dev, skb);
3254 if (queue_index != new_index && sk &&
3256 rcu_access_pointer(sk->sk_dst_cache))
3257 sk_tx_queue_set(sk, new_index);
3259 queue_index = new_index;
3265 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3266 struct sk_buff *skb,
3269 int queue_index = 0;
3272 u32 sender_cpu = skb->sender_cpu - 1;
3274 if (sender_cpu >= (u32)NR_CPUS)
3275 skb->sender_cpu = raw_smp_processor_id() + 1;
3278 if (dev->real_num_tx_queues != 1) {
3279 const struct net_device_ops *ops = dev->netdev_ops;
3280 if (ops->ndo_select_queue)
3281 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3284 queue_index = __netdev_pick_tx(dev, skb);
3287 queue_index = netdev_cap_txqueue(dev, queue_index);
3290 skb_set_queue_mapping(skb, queue_index);
3291 return netdev_get_tx_queue(dev, queue_index);
3295 * __dev_queue_xmit - transmit a buffer
3296 * @skb: buffer to transmit
3297 * @accel_priv: private data used for L2 forwarding offload
3299 * Queue a buffer for transmission to a network device. The caller must
3300 * have set the device and priority and built the buffer before calling
3301 * this function. The function can be called from an interrupt.
3303 * A negative errno code is returned on a failure. A success does not
3304 * guarantee the frame will be transmitted as it may be dropped due
3305 * to congestion or traffic shaping.
3307 * -----------------------------------------------------------------------------------
3308 * I notice this method can also return errors from the queue disciplines,
3309 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3312 * Regardless of the return value, the skb is consumed, so it is currently
3313 * difficult to retry a send to this method. (You can bump the ref count
3314 * before sending to hold a reference for retry if you are careful.)
3316 * When calling this method, interrupts MUST be enabled. This is because
3317 * the BH enable code must have IRQs enabled so that it will not deadlock.
3320 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3322 struct net_device *dev = skb->dev;
3323 struct netdev_queue *txq;
3327 skb_reset_mac_header(skb);
3329 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3330 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3332 /* Disable soft irqs for various locks below. Also
3333 * stops preemption for RCU.
3337 skb_update_prio(skb);
3339 qdisc_pkt_len_init(skb);
3340 #ifdef CONFIG_NET_CLS_ACT
3341 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3342 # ifdef CONFIG_NET_EGRESS
3343 if (static_key_false(&egress_needed)) {
3344 skb = sch_handle_egress(skb, &rc, dev);
3350 /* If device/qdisc don't need skb->dst, release it right now while
3351 * its hot in this cpu cache.
3353 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3358 txq = netdev_pick_tx(dev, skb, accel_priv);
3359 q = rcu_dereference_bh(txq->qdisc);
3361 trace_net_dev_queue(skb);
3363 rc = __dev_xmit_skb(skb, q, dev, txq);
3367 /* The device has no queue. Common case for software devices:
3368 loopback, all the sorts of tunnels...
3370 Really, it is unlikely that netif_tx_lock protection is necessary
3371 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3373 However, it is possible, that they rely on protection
3376 Check this and shot the lock. It is not prone from deadlocks.
3377 Either shot noqueue qdisc, it is even simpler 8)
3379 if (dev->flags & IFF_UP) {
3380 int cpu = smp_processor_id(); /* ok because BHs are off */
3382 if (txq->xmit_lock_owner != cpu) {
3383 if (unlikely(__this_cpu_read(xmit_recursion) >
3384 XMIT_RECURSION_LIMIT))
3385 goto recursion_alert;
3387 skb = validate_xmit_skb(skb, dev);
3391 HARD_TX_LOCK(dev, txq, cpu);
3393 if (!netif_xmit_stopped(txq)) {
3394 __this_cpu_inc(xmit_recursion);
3395 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3396 __this_cpu_dec(xmit_recursion);
3397 if (dev_xmit_complete(rc)) {
3398 HARD_TX_UNLOCK(dev, txq);
3402 HARD_TX_UNLOCK(dev, txq);
3403 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3406 /* Recursion is detected! It is possible,
3410 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3416 rcu_read_unlock_bh();
3418 atomic_long_inc(&dev->tx_dropped);
3419 kfree_skb_list(skb);
3422 rcu_read_unlock_bh();
3426 int dev_queue_xmit(struct sk_buff *skb)
3428 return __dev_queue_xmit(skb, NULL);
3430 EXPORT_SYMBOL(dev_queue_xmit);
3432 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3434 return __dev_queue_xmit(skb, accel_priv);
3436 EXPORT_SYMBOL(dev_queue_xmit_accel);
3439 /*=======================================================================
3441 =======================================================================*/
3443 int netdev_max_backlog __read_mostly = 1000;
3444 EXPORT_SYMBOL(netdev_max_backlog);
3446 int netdev_tstamp_prequeue __read_mostly = 1;
3447 int netdev_budget __read_mostly = 300;
3448 int weight_p __read_mostly = 64; /* old backlog weight */
3450 /* Called with irq disabled */
3451 static inline void ____napi_schedule(struct softnet_data *sd,
3452 struct napi_struct *napi)
3454 list_add_tail(&napi->poll_list, &sd->poll_list);
3455 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3460 /* One global table that all flow-based protocols share. */
3461 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3462 EXPORT_SYMBOL(rps_sock_flow_table);
3463 u32 rps_cpu_mask __read_mostly;
3464 EXPORT_SYMBOL(rps_cpu_mask);
3466 struct static_key rps_needed __read_mostly;
3467 EXPORT_SYMBOL(rps_needed);
3469 static struct rps_dev_flow *
3470 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3471 struct rps_dev_flow *rflow, u16 next_cpu)
3473 if (next_cpu < nr_cpu_ids) {
3474 #ifdef CONFIG_RFS_ACCEL
3475 struct netdev_rx_queue *rxqueue;
3476 struct rps_dev_flow_table *flow_table;
3477 struct rps_dev_flow *old_rflow;
3482 /* Should we steer this flow to a different hardware queue? */
3483 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3484 !(dev->features & NETIF_F_NTUPLE))
3486 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3487 if (rxq_index == skb_get_rx_queue(skb))
3490 rxqueue = dev->_rx + rxq_index;
3491 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3494 flow_id = skb_get_hash(skb) & flow_table->mask;
3495 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3496 rxq_index, flow_id);
3500 rflow = &flow_table->flows[flow_id];
3502 if (old_rflow->filter == rflow->filter)
3503 old_rflow->filter = RPS_NO_FILTER;
3507 per_cpu(softnet_data, next_cpu).input_queue_head;
3510 rflow->cpu = next_cpu;
3515 * get_rps_cpu is called from netif_receive_skb and returns the target
3516 * CPU from the RPS map of the receiving queue for a given skb.
3517 * rcu_read_lock must be held on entry.
3519 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3520 struct rps_dev_flow **rflowp)
3522 const struct rps_sock_flow_table *sock_flow_table;
3523 struct netdev_rx_queue *rxqueue = dev->_rx;
3524 struct rps_dev_flow_table *flow_table;
3525 struct rps_map *map;
3530 if (skb_rx_queue_recorded(skb)) {
3531 u16 index = skb_get_rx_queue(skb);
3533 if (unlikely(index >= dev->real_num_rx_queues)) {
3534 WARN_ONCE(dev->real_num_rx_queues > 1,
3535 "%s received packet on queue %u, but number "
3536 "of RX queues is %u\n",
3537 dev->name, index, dev->real_num_rx_queues);
3543 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3545 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3546 map = rcu_dereference(rxqueue->rps_map);
3547 if (!flow_table && !map)
3550 skb_reset_network_header(skb);
3551 hash = skb_get_hash(skb);
3555 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3556 if (flow_table && sock_flow_table) {
3557 struct rps_dev_flow *rflow;
3561 /* First check into global flow table if there is a match */
3562 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3563 if ((ident ^ hash) & ~rps_cpu_mask)
3566 next_cpu = ident & rps_cpu_mask;
3568 /* OK, now we know there is a match,
3569 * we can look at the local (per receive queue) flow table
3571 rflow = &flow_table->flows[hash & flow_table->mask];
3575 * If the desired CPU (where last recvmsg was done) is
3576 * different from current CPU (one in the rx-queue flow
3577 * table entry), switch if one of the following holds:
3578 * - Current CPU is unset (>= nr_cpu_ids).
3579 * - Current CPU is offline.
3580 * - The current CPU's queue tail has advanced beyond the
3581 * last packet that was enqueued using this table entry.
3582 * This guarantees that all previous packets for the flow
3583 * have been dequeued, thus preserving in order delivery.
3585 if (unlikely(tcpu != next_cpu) &&
3586 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3587 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3588 rflow->last_qtail)) >= 0)) {
3590 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3593 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3603 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3604 if (cpu_online(tcpu)) {
3614 #ifdef CONFIG_RFS_ACCEL
3617 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3618 * @dev: Device on which the filter was set
3619 * @rxq_index: RX queue index
3620 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3621 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3623 * Drivers that implement ndo_rx_flow_steer() should periodically call
3624 * this function for each installed filter and remove the filters for
3625 * which it returns %true.
3627 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3628 u32 flow_id, u16 filter_id)
3630 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3631 struct rps_dev_flow_table *flow_table;
3632 struct rps_dev_flow *rflow;
3637 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3638 if (flow_table && flow_id <= flow_table->mask) {
3639 rflow = &flow_table->flows[flow_id];
3640 cpu = ACCESS_ONCE(rflow->cpu);
3641 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3642 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3643 rflow->last_qtail) <
3644 (int)(10 * flow_table->mask)))
3650 EXPORT_SYMBOL(rps_may_expire_flow);
3652 #endif /* CONFIG_RFS_ACCEL */
3654 /* Called from hardirq (IPI) context */
3655 static void rps_trigger_softirq(void *data)
3657 struct softnet_data *sd = data;
3659 ____napi_schedule(sd, &sd->backlog);
3663 #endif /* CONFIG_RPS */
3666 * Check if this softnet_data structure is another cpu one
3667 * If yes, queue it to our IPI list and return 1
3670 static int rps_ipi_queued(struct softnet_data *sd)
3673 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3676 sd->rps_ipi_next = mysd->rps_ipi_list;
3677 mysd->rps_ipi_list = sd;
3679 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3682 #endif /* CONFIG_RPS */
3686 #ifdef CONFIG_NET_FLOW_LIMIT
3687 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3690 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3692 #ifdef CONFIG_NET_FLOW_LIMIT
3693 struct sd_flow_limit *fl;
3694 struct softnet_data *sd;
3695 unsigned int old_flow, new_flow;
3697 if (qlen < (netdev_max_backlog >> 1))
3700 sd = this_cpu_ptr(&softnet_data);
3703 fl = rcu_dereference(sd->flow_limit);
3705 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3706 old_flow = fl->history[fl->history_head];
3707 fl->history[fl->history_head] = new_flow;
3710 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3712 if (likely(fl->buckets[old_flow]))
3713 fl->buckets[old_flow]--;
3715 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3727 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3728 * queue (may be a remote CPU queue).
3730 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3731 unsigned int *qtail)
3733 struct softnet_data *sd;
3734 unsigned long flags;
3737 sd = &per_cpu(softnet_data, cpu);
3739 local_irq_save(flags);
3742 if (!netif_running(skb->dev))
3744 qlen = skb_queue_len(&sd->input_pkt_queue);
3745 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3748 __skb_queue_tail(&sd->input_pkt_queue, skb);
3749 input_queue_tail_incr_save(sd, qtail);
3751 local_irq_restore(flags);
3752 return NET_RX_SUCCESS;
3755 /* Schedule NAPI for backlog device
3756 * We can use non atomic operation since we own the queue lock
3758 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3759 if (!rps_ipi_queued(sd))
3760 ____napi_schedule(sd, &sd->backlog);
3769 local_irq_restore(flags);
3771 atomic_long_inc(&skb->dev->rx_dropped);
3776 static int netif_rx_internal(struct sk_buff *skb)
3780 net_timestamp_check(netdev_tstamp_prequeue, skb);
3782 trace_netif_rx(skb);
3784 if (static_key_false(&rps_needed)) {
3785 struct rps_dev_flow voidflow, *rflow = &voidflow;
3791 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3793 cpu = smp_processor_id();
3795 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3803 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3810 * netif_rx - post buffer to the network code
3811 * @skb: buffer to post
3813 * This function receives a packet from a device driver and queues it for
3814 * the upper (protocol) levels to process. It always succeeds. The buffer
3815 * may be dropped during processing for congestion control or by the
3819 * NET_RX_SUCCESS (no congestion)
3820 * NET_RX_DROP (packet was dropped)
3824 int netif_rx(struct sk_buff *skb)
3826 trace_netif_rx_entry(skb);
3828 return netif_rx_internal(skb);
3830 EXPORT_SYMBOL(netif_rx);
3832 int netif_rx_ni(struct sk_buff *skb)
3836 trace_netif_rx_ni_entry(skb);
3839 err = netif_rx_internal(skb);
3840 if (local_softirq_pending())
3846 EXPORT_SYMBOL(netif_rx_ni);
3848 static void net_tx_action(struct softirq_action *h)
3850 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3852 if (sd->completion_queue) {
3853 struct sk_buff *clist;
3855 local_irq_disable();
3856 clist = sd->completion_queue;
3857 sd->completion_queue = NULL;
3861 struct sk_buff *skb = clist;
3862 clist = clist->next;
3864 WARN_ON(atomic_read(&skb->users));
3865 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3866 trace_consume_skb(skb);
3868 trace_kfree_skb(skb, net_tx_action);
3870 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
3873 __kfree_skb_defer(skb);
3876 __kfree_skb_flush();
3879 if (sd->output_queue) {
3882 local_irq_disable();
3883 head = sd->output_queue;
3884 sd->output_queue = NULL;
3885 sd->output_queue_tailp = &sd->output_queue;
3889 struct Qdisc *q = head;
3890 spinlock_t *root_lock;
3892 head = head->next_sched;
3894 root_lock = qdisc_lock(q);
3895 spin_lock(root_lock);
3896 /* We need to make sure head->next_sched is read
3897 * before clearing __QDISC_STATE_SCHED
3899 smp_mb__before_atomic();
3900 clear_bit(__QDISC_STATE_SCHED, &q->state);
3902 spin_unlock(root_lock);
3907 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
3908 /* This hook is defined here for ATM LANE */
3909 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3910 unsigned char *addr) __read_mostly;
3911 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3914 static inline struct sk_buff *
3915 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
3916 struct net_device *orig_dev)
3918 #ifdef CONFIG_NET_CLS_ACT
3919 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3920 struct tcf_result cl_res;
3922 /* If there's at least one ingress present somewhere (so
3923 * we get here via enabled static key), remaining devices
3924 * that are not configured with an ingress qdisc will bail
3930 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3934 qdisc_skb_cb(skb)->pkt_len = skb->len;
3935 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3936 qdisc_bstats_cpu_update(cl->q, skb);
3938 switch (tc_classify(skb, cl, &cl_res, false)) {
3940 case TC_ACT_RECLASSIFY:
3941 skb->tc_index = TC_H_MIN(cl_res.classid);
3944 qdisc_qstats_cpu_drop(cl->q);
3951 case TC_ACT_REDIRECT:
3952 /* skb_mac_header check was done by cls/act_bpf, so
3953 * we can safely push the L2 header back before
3954 * redirecting to another netdev
3956 __skb_push(skb, skb->mac_len);
3957 skb_do_redirect(skb);
3962 #endif /* CONFIG_NET_CLS_ACT */
3967 * netdev_is_rx_handler_busy - check if receive handler is registered
3968 * @dev: device to check
3970 * Check if a receive handler is already registered for a given device.
3971 * Return true if there one.
3973 * The caller must hold the rtnl_mutex.
3975 bool netdev_is_rx_handler_busy(struct net_device *dev)
3978 return dev && rtnl_dereference(dev->rx_handler);
3980 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
3983 * netdev_rx_handler_register - register receive handler
3984 * @dev: device to register a handler for
3985 * @rx_handler: receive handler to register
3986 * @rx_handler_data: data pointer that is used by rx handler
3988 * Register a receive handler for a device. This handler will then be
3989 * called from __netif_receive_skb. A negative errno code is returned
3992 * The caller must hold the rtnl_mutex.
3994 * For a general description of rx_handler, see enum rx_handler_result.
3996 int netdev_rx_handler_register(struct net_device *dev,
3997 rx_handler_func_t *rx_handler,
3998 void *rx_handler_data)
4002 if (dev->rx_handler)
4005 /* Note: rx_handler_data must be set before rx_handler */
4006 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4007 rcu_assign_pointer(dev->rx_handler, rx_handler);
4011 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4014 * netdev_rx_handler_unregister - unregister receive handler
4015 * @dev: device to unregister a handler from
4017 * Unregister a receive handler from a device.
4019 * The caller must hold the rtnl_mutex.
4021 void netdev_rx_handler_unregister(struct net_device *dev)
4025 RCU_INIT_POINTER(dev->rx_handler, NULL);
4026 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4027 * section has a guarantee to see a non NULL rx_handler_data
4031 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4033 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4036 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4037 * the special handling of PFMEMALLOC skbs.
4039 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4041 switch (skb->protocol) {
4042 case htons(ETH_P_ARP):
4043 case htons(ETH_P_IP):
4044 case htons(ETH_P_IPV6):
4045 case htons(ETH_P_8021Q):
4046 case htons(ETH_P_8021AD):
4053 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4054 int *ret, struct net_device *orig_dev)
4056 #ifdef CONFIG_NETFILTER_INGRESS
4057 if (nf_hook_ingress_active(skb)) {
4061 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4066 ingress_retval = nf_hook_ingress(skb);
4068 return ingress_retval;
4070 #endif /* CONFIG_NETFILTER_INGRESS */
4074 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
4076 struct packet_type *ptype, *pt_prev;
4077 rx_handler_func_t *rx_handler;
4078 struct net_device *orig_dev;
4079 bool deliver_exact = false;
4080 int ret = NET_RX_DROP;
4083 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4085 trace_netif_receive_skb(skb);
4087 orig_dev = skb->dev;
4089 skb_reset_network_header(skb);
4090 if (!skb_transport_header_was_set(skb))
4091 skb_reset_transport_header(skb);
4092 skb_reset_mac_len(skb);
4097 skb->skb_iif = skb->dev->ifindex;
4099 __this_cpu_inc(softnet_data.processed);
4101 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4102 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4103 skb = skb_vlan_untag(skb);
4108 #ifdef CONFIG_NET_CLS_ACT
4109 if (skb->tc_verd & TC_NCLS) {
4110 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
4118 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4120 ret = deliver_skb(skb, pt_prev, orig_dev);
4124 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4126 ret = deliver_skb(skb, pt_prev, orig_dev);
4131 #ifdef CONFIG_NET_INGRESS
4132 if (static_key_false(&ingress_needed)) {
4133 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4137 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4141 #ifdef CONFIG_NET_CLS_ACT
4145 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4148 if (skb_vlan_tag_present(skb)) {
4150 ret = deliver_skb(skb, pt_prev, orig_dev);
4153 if (vlan_do_receive(&skb))
4155 else if (unlikely(!skb))
4159 rx_handler = rcu_dereference(skb->dev->rx_handler);
4162 ret = deliver_skb(skb, pt_prev, orig_dev);
4165 switch (rx_handler(&skb)) {
4166 case RX_HANDLER_CONSUMED:
4167 ret = NET_RX_SUCCESS;
4169 case RX_HANDLER_ANOTHER:
4171 case RX_HANDLER_EXACT:
4172 deliver_exact = true;
4173 case RX_HANDLER_PASS:
4180 if (unlikely(skb_vlan_tag_present(skb))) {
4181 if (skb_vlan_tag_get_id(skb))
4182 skb->pkt_type = PACKET_OTHERHOST;
4183 /* Note: we might in the future use prio bits
4184 * and set skb->priority like in vlan_do_receive()
4185 * For the time being, just ignore Priority Code Point
4190 type = skb->protocol;
4192 /* deliver only exact match when indicated */
4193 if (likely(!deliver_exact)) {
4194 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4195 &ptype_base[ntohs(type) &
4199 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4200 &orig_dev->ptype_specific);
4202 if (unlikely(skb->dev != orig_dev)) {
4203 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4204 &skb->dev->ptype_specific);
4208 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
4211 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4215 atomic_long_inc(&skb->dev->rx_dropped);
4217 atomic_long_inc(&skb->dev->rx_nohandler);
4219 /* Jamal, now you will not able to escape explaining
4220 * me how you were going to use this. :-)
4229 static int __netif_receive_skb(struct sk_buff *skb)
4233 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4234 unsigned long pflags = current->flags;
4237 * PFMEMALLOC skbs are special, they should
4238 * - be delivered to SOCK_MEMALLOC sockets only
4239 * - stay away from userspace
4240 * - have bounded memory usage
4242 * Use PF_MEMALLOC as this saves us from propagating the allocation
4243 * context down to all allocation sites.
4245 current->flags |= PF_MEMALLOC;
4246 ret = __netif_receive_skb_core(skb, true);
4247 tsk_restore_flags(current, pflags, PF_MEMALLOC);
4249 ret = __netif_receive_skb_core(skb, false);
4254 static int netif_receive_skb_internal(struct sk_buff *skb)
4258 net_timestamp_check(netdev_tstamp_prequeue, skb);
4260 if (skb_defer_rx_timestamp(skb))
4261 return NET_RX_SUCCESS;
4266 if (static_key_false(&rps_needed)) {
4267 struct rps_dev_flow voidflow, *rflow = &voidflow;
4268 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4271 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4277 ret = __netif_receive_skb(skb);
4283 * netif_receive_skb - process receive buffer from network
4284 * @skb: buffer to process
4286 * netif_receive_skb() is the main receive data processing function.
4287 * It always succeeds. The buffer may be dropped during processing
4288 * for congestion control or by the protocol layers.
4290 * This function may only be called from softirq context and interrupts
4291 * should be enabled.
4293 * Return values (usually ignored):
4294 * NET_RX_SUCCESS: no congestion
4295 * NET_RX_DROP: packet was dropped
4297 int netif_receive_skb(struct sk_buff *skb)
4299 trace_netif_receive_skb_entry(skb);
4301 return netif_receive_skb_internal(skb);
4303 EXPORT_SYMBOL(netif_receive_skb);
4305 DEFINE_PER_CPU(struct work_struct, flush_works);
4307 /* Network device is going away, flush any packets still pending */
4308 static void flush_backlog(struct work_struct *work)
4310 struct sk_buff *skb, *tmp;
4311 struct softnet_data *sd;
4314 sd = this_cpu_ptr(&softnet_data);
4316 local_irq_disable();
4318 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4319 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4320 __skb_unlink(skb, &sd->input_pkt_queue);
4322 input_queue_head_incr(sd);
4328 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4329 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4330 __skb_unlink(skb, &sd->process_queue);
4332 input_queue_head_incr(sd);
4338 static void flush_all_backlogs(void)
4344 for_each_online_cpu(cpu)
4345 queue_work_on(cpu, system_highpri_wq,
4346 per_cpu_ptr(&flush_works, cpu));
4348 for_each_online_cpu(cpu)
4349 flush_work(per_cpu_ptr(&flush_works, cpu));
4354 static int napi_gro_complete(struct sk_buff *skb)
4356 struct packet_offload *ptype;
4357 __be16 type = skb->protocol;
4358 struct list_head *head = &offload_base;
4361 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4363 if (NAPI_GRO_CB(skb)->count == 1) {
4364 skb_shinfo(skb)->gso_size = 0;
4369 list_for_each_entry_rcu(ptype, head, list) {
4370 if (ptype->type != type || !ptype->callbacks.gro_complete)
4373 err = ptype->callbacks.gro_complete(skb, 0);
4379 WARN_ON(&ptype->list == head);
4381 return NET_RX_SUCCESS;
4385 return netif_receive_skb_internal(skb);
4388 /* napi->gro_list contains packets ordered by age.
4389 * youngest packets at the head of it.
4390 * Complete skbs in reverse order to reduce latencies.
4392 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4394 struct sk_buff *skb, *prev = NULL;
4396 /* scan list and build reverse chain */
4397 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4402 for (skb = prev; skb; skb = prev) {
4405 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4409 napi_gro_complete(skb);
4413 napi->gro_list = NULL;
4415 EXPORT_SYMBOL(napi_gro_flush);
4417 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4420 unsigned int maclen = skb->dev->hard_header_len;
4421 u32 hash = skb_get_hash_raw(skb);
4423 for (p = napi->gro_list; p; p = p->next) {
4424 unsigned long diffs;
4426 NAPI_GRO_CB(p)->flush = 0;
4428 if (hash != skb_get_hash_raw(p)) {
4429 NAPI_GRO_CB(p)->same_flow = 0;
4433 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4434 diffs |= p->vlan_tci ^ skb->vlan_tci;
4435 diffs |= skb_metadata_dst_cmp(p, skb);
4436 if (maclen == ETH_HLEN)
4437 diffs |= compare_ether_header(skb_mac_header(p),
4438 skb_mac_header(skb));
4440 diffs = memcmp(skb_mac_header(p),
4441 skb_mac_header(skb),
4443 NAPI_GRO_CB(p)->same_flow = !diffs;
4447 static void skb_gro_reset_offset(struct sk_buff *skb)
4449 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4450 const skb_frag_t *frag0 = &pinfo->frags[0];
4452 NAPI_GRO_CB(skb)->data_offset = 0;
4453 NAPI_GRO_CB(skb)->frag0 = NULL;
4454 NAPI_GRO_CB(skb)->frag0_len = 0;
4456 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4458 !PageHighMem(skb_frag_page(frag0))) {
4459 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4460 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4464 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4466 struct skb_shared_info *pinfo = skb_shinfo(skb);
4468 BUG_ON(skb->end - skb->tail < grow);
4470 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4472 skb->data_len -= grow;
4475 pinfo->frags[0].page_offset += grow;
4476 skb_frag_size_sub(&pinfo->frags[0], grow);
4478 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4479 skb_frag_unref(skb, 0);
4480 memmove(pinfo->frags, pinfo->frags + 1,
4481 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4485 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4487 struct sk_buff **pp = NULL;
4488 struct packet_offload *ptype;
4489 __be16 type = skb->protocol;
4490 struct list_head *head = &offload_base;
4492 enum gro_result ret;
4495 if (!(skb->dev->features & NETIF_F_GRO))
4498 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4501 gro_list_prepare(napi, skb);
4504 list_for_each_entry_rcu(ptype, head, list) {
4505 if (ptype->type != type || !ptype->callbacks.gro_receive)
4508 skb_set_network_header(skb, skb_gro_offset(skb));
4509 skb_reset_mac_len(skb);
4510 NAPI_GRO_CB(skb)->same_flow = 0;
4511 NAPI_GRO_CB(skb)->flush = 0;
4512 NAPI_GRO_CB(skb)->free = 0;
4513 NAPI_GRO_CB(skb)->encap_mark = 0;
4514 NAPI_GRO_CB(skb)->is_fou = 0;
4515 NAPI_GRO_CB(skb)->is_atomic = 1;
4516 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4518 /* Setup for GRO checksum validation */
4519 switch (skb->ip_summed) {
4520 case CHECKSUM_COMPLETE:
4521 NAPI_GRO_CB(skb)->csum = skb->csum;
4522 NAPI_GRO_CB(skb)->csum_valid = 1;
4523 NAPI_GRO_CB(skb)->csum_cnt = 0;
4525 case CHECKSUM_UNNECESSARY:
4526 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4527 NAPI_GRO_CB(skb)->csum_valid = 0;
4530 NAPI_GRO_CB(skb)->csum_cnt = 0;
4531 NAPI_GRO_CB(skb)->csum_valid = 0;
4534 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4539 if (&ptype->list == head)
4542 same_flow = NAPI_GRO_CB(skb)->same_flow;
4543 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4546 struct sk_buff *nskb = *pp;
4550 napi_gro_complete(nskb);
4557 if (NAPI_GRO_CB(skb)->flush)
4560 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4561 struct sk_buff *nskb = napi->gro_list;
4563 /* locate the end of the list to select the 'oldest' flow */
4564 while (nskb->next) {
4570 napi_gro_complete(nskb);
4574 NAPI_GRO_CB(skb)->count = 1;
4575 NAPI_GRO_CB(skb)->age = jiffies;
4576 NAPI_GRO_CB(skb)->last = skb;
4577 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4578 skb->next = napi->gro_list;
4579 napi->gro_list = skb;
4583 grow = skb_gro_offset(skb) - skb_headlen(skb);
4585 gro_pull_from_frag0(skb, grow);
4594 struct packet_offload *gro_find_receive_by_type(__be16 type)
4596 struct list_head *offload_head = &offload_base;
4597 struct packet_offload *ptype;
4599 list_for_each_entry_rcu(ptype, offload_head, list) {
4600 if (ptype->type != type || !ptype->callbacks.gro_receive)
4606 EXPORT_SYMBOL(gro_find_receive_by_type);
4608 struct packet_offload *gro_find_complete_by_type(__be16 type)
4610 struct list_head *offload_head = &offload_base;
4611 struct packet_offload *ptype;
4613 list_for_each_entry_rcu(ptype, offload_head, list) {
4614 if (ptype->type != type || !ptype->callbacks.gro_complete)
4620 EXPORT_SYMBOL(gro_find_complete_by_type);
4622 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4626 if (netif_receive_skb_internal(skb))
4634 case GRO_MERGED_FREE:
4635 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) {
4637 kmem_cache_free(skbuff_head_cache, skb);
4651 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4653 skb_mark_napi_id(skb, napi);
4654 trace_napi_gro_receive_entry(skb);
4656 skb_gro_reset_offset(skb);
4658 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4660 EXPORT_SYMBOL(napi_gro_receive);
4662 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4664 if (unlikely(skb->pfmemalloc)) {
4668 __skb_pull(skb, skb_headlen(skb));
4669 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4670 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4672 skb->dev = napi->dev;
4674 skb->encapsulation = 0;
4675 skb_shinfo(skb)->gso_type = 0;
4676 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4681 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4683 struct sk_buff *skb = napi->skb;
4686 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4689 skb_mark_napi_id(skb, napi);
4694 EXPORT_SYMBOL(napi_get_frags);
4696 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4697 struct sk_buff *skb,
4703 __skb_push(skb, ETH_HLEN);
4704 skb->protocol = eth_type_trans(skb, skb->dev);
4705 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4710 case GRO_MERGED_FREE:
4711 napi_reuse_skb(napi, skb);
4721 /* Upper GRO stack assumes network header starts at gro_offset=0
4722 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4723 * We copy ethernet header into skb->data to have a common layout.
4725 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4727 struct sk_buff *skb = napi->skb;
4728 const struct ethhdr *eth;
4729 unsigned int hlen = sizeof(*eth);
4733 skb_reset_mac_header(skb);
4734 skb_gro_reset_offset(skb);
4736 eth = skb_gro_header_fast(skb, 0);
4737 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4738 eth = skb_gro_header_slow(skb, hlen, 0);
4739 if (unlikely(!eth)) {
4740 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
4741 __func__, napi->dev->name);
4742 napi_reuse_skb(napi, skb);
4746 gro_pull_from_frag0(skb, hlen);
4747 NAPI_GRO_CB(skb)->frag0 += hlen;
4748 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4750 __skb_pull(skb, hlen);
4753 * This works because the only protocols we care about don't require
4755 * We'll fix it up properly in napi_frags_finish()
4757 skb->protocol = eth->h_proto;
4762 gro_result_t napi_gro_frags(struct napi_struct *napi)
4764 struct sk_buff *skb = napi_frags_skb(napi);
4769 trace_napi_gro_frags_entry(skb);
4771 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4773 EXPORT_SYMBOL(napi_gro_frags);
4775 /* Compute the checksum from gro_offset and return the folded value
4776 * after adding in any pseudo checksum.
4778 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4783 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4785 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4786 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4788 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4789 !skb->csum_complete_sw)
4790 netdev_rx_csum_fault(skb->dev);
4793 NAPI_GRO_CB(skb)->csum = wsum;
4794 NAPI_GRO_CB(skb)->csum_valid = 1;
4798 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4801 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4802 * Note: called with local irq disabled, but exits with local irq enabled.
4804 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4807 struct softnet_data *remsd = sd->rps_ipi_list;
4810 sd->rps_ipi_list = NULL;
4814 /* Send pending IPI's to kick RPS processing on remote cpus. */
4816 struct softnet_data *next = remsd->rps_ipi_next;
4818 if (cpu_online(remsd->cpu))
4819 smp_call_function_single_async(remsd->cpu,
4828 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4831 return sd->rps_ipi_list != NULL;
4837 static int process_backlog(struct napi_struct *napi, int quota)
4839 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4843 /* Check if we have pending ipi, its better to send them now,
4844 * not waiting net_rx_action() end.
4846 if (sd_has_rps_ipi_waiting(sd)) {
4847 local_irq_disable();
4848 net_rps_action_and_irq_enable(sd);
4851 napi->weight = weight_p;
4853 struct sk_buff *skb;
4855 while ((skb = __skb_dequeue(&sd->process_queue))) {
4857 __netif_receive_skb(skb);
4859 input_queue_head_incr(sd);
4860 if (++work >= quota)
4865 local_irq_disable();
4867 if (skb_queue_empty(&sd->input_pkt_queue)) {
4869 * Inline a custom version of __napi_complete().
4870 * only current cpu owns and manipulates this napi,
4871 * and NAPI_STATE_SCHED is the only possible flag set
4873 * We can use a plain write instead of clear_bit(),
4874 * and we dont need an smp_mb() memory barrier.
4879 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4880 &sd->process_queue);
4890 * __napi_schedule - schedule for receive
4891 * @n: entry to schedule
4893 * The entry's receive function will be scheduled to run.
4894 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4896 void __napi_schedule(struct napi_struct *n)
4898 unsigned long flags;
4900 local_irq_save(flags);
4901 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4902 local_irq_restore(flags);
4904 EXPORT_SYMBOL(__napi_schedule);
4907 * __napi_schedule_irqoff - schedule for receive
4908 * @n: entry to schedule
4910 * Variant of __napi_schedule() assuming hard irqs are masked
4912 void __napi_schedule_irqoff(struct napi_struct *n)
4914 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4916 EXPORT_SYMBOL(__napi_schedule_irqoff);
4918 void __napi_complete(struct napi_struct *n)
4920 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4922 list_del_init(&n->poll_list);
4923 smp_mb__before_atomic();
4924 clear_bit(NAPI_STATE_SCHED, &n->state);
4926 EXPORT_SYMBOL(__napi_complete);
4928 void napi_complete_done(struct napi_struct *n, int work_done)
4930 unsigned long flags;
4933 * don't let napi dequeue from the cpu poll list
4934 * just in case its running on a different cpu
4936 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4940 unsigned long timeout = 0;
4943 timeout = n->dev->gro_flush_timeout;
4946 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4947 HRTIMER_MODE_REL_PINNED);
4949 napi_gro_flush(n, false);
4951 if (likely(list_empty(&n->poll_list))) {
4952 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4954 /* If n->poll_list is not empty, we need to mask irqs */
4955 local_irq_save(flags);
4957 local_irq_restore(flags);
4960 EXPORT_SYMBOL(napi_complete_done);
4962 /* must be called under rcu_read_lock(), as we dont take a reference */
4963 static struct napi_struct *napi_by_id(unsigned int napi_id)
4965 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4966 struct napi_struct *napi;
4968 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4969 if (napi->napi_id == napi_id)
4975 #if defined(CONFIG_NET_RX_BUSY_POLL)
4976 #define BUSY_POLL_BUDGET 8
4977 bool sk_busy_loop(struct sock *sk, int nonblock)
4979 unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0;
4980 int (*busy_poll)(struct napi_struct *dev);
4981 struct napi_struct *napi;
4986 napi = napi_by_id(sk->sk_napi_id);
4990 /* Note: ndo_busy_poll method is optional in linux-4.5 */
4991 busy_poll = napi->dev->netdev_ops->ndo_busy_poll;
4997 rc = busy_poll(napi);
4998 } else if (napi_schedule_prep(napi)) {
4999 void *have = netpoll_poll_lock(napi);
5001 if (test_bit(NAPI_STATE_SCHED, &napi->state)) {
5002 rc = napi->poll(napi, BUSY_POLL_BUDGET);
5003 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
5004 if (rc == BUSY_POLL_BUDGET) {
5005 napi_complete_done(napi, rc);
5006 napi_schedule(napi);
5009 netpoll_poll_unlock(have);
5012 __NET_ADD_STATS(sock_net(sk),
5013 LINUX_MIB_BUSYPOLLRXPACKETS, rc);
5016 if (rc == LL_FLUSH_FAILED)
5017 break; /* permanent failure */
5020 } while (!nonblock && skb_queue_empty(&sk->sk_receive_queue) &&
5021 !need_resched() && !busy_loop_timeout(end_time));
5023 rc = !skb_queue_empty(&sk->sk_receive_queue);
5028 EXPORT_SYMBOL(sk_busy_loop);
5030 #endif /* CONFIG_NET_RX_BUSY_POLL */
5032 void napi_hash_add(struct napi_struct *napi)
5034 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
5035 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
5038 spin_lock(&napi_hash_lock);
5040 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */
5042 if (unlikely(++napi_gen_id < NR_CPUS + 1))
5043 napi_gen_id = NR_CPUS + 1;
5044 } while (napi_by_id(napi_gen_id));
5045 napi->napi_id = napi_gen_id;
5047 hlist_add_head_rcu(&napi->napi_hash_node,
5048 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
5050 spin_unlock(&napi_hash_lock);
5052 EXPORT_SYMBOL_GPL(napi_hash_add);
5054 /* Warning : caller is responsible to make sure rcu grace period
5055 * is respected before freeing memory containing @napi
5057 bool napi_hash_del(struct napi_struct *napi)
5059 bool rcu_sync_needed = false;
5061 spin_lock(&napi_hash_lock);
5063 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
5064 rcu_sync_needed = true;
5065 hlist_del_rcu(&napi->napi_hash_node);
5067 spin_unlock(&napi_hash_lock);
5068 return rcu_sync_needed;
5070 EXPORT_SYMBOL_GPL(napi_hash_del);
5072 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
5074 struct napi_struct *napi;
5076 napi = container_of(timer, struct napi_struct, timer);
5078 napi_schedule(napi);
5080 return HRTIMER_NORESTART;
5083 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
5084 int (*poll)(struct napi_struct *, int), int weight)
5086 INIT_LIST_HEAD(&napi->poll_list);
5087 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
5088 napi->timer.function = napi_watchdog;
5089 napi->gro_count = 0;
5090 napi->gro_list = NULL;
5093 if (weight > NAPI_POLL_WEIGHT)
5094 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5096 napi->weight = weight;
5097 list_add(&napi->dev_list, &dev->napi_list);
5099 #ifdef CONFIG_NETPOLL
5100 spin_lock_init(&napi->poll_lock);
5101 napi->poll_owner = -1;
5103 set_bit(NAPI_STATE_SCHED, &napi->state);
5104 napi_hash_add(napi);
5106 EXPORT_SYMBOL(netif_napi_add);
5108 void napi_disable(struct napi_struct *n)
5111 set_bit(NAPI_STATE_DISABLE, &n->state);
5113 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
5115 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
5118 hrtimer_cancel(&n->timer);
5120 clear_bit(NAPI_STATE_DISABLE, &n->state);
5122 EXPORT_SYMBOL(napi_disable);
5124 /* Must be called in process context */
5125 void netif_napi_del(struct napi_struct *napi)
5128 if (napi_hash_del(napi))
5130 list_del_init(&napi->dev_list);
5131 napi_free_frags(napi);
5133 kfree_skb_list(napi->gro_list);
5134 napi->gro_list = NULL;
5135 napi->gro_count = 0;
5137 EXPORT_SYMBOL(netif_napi_del);
5139 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
5144 list_del_init(&n->poll_list);
5146 have = netpoll_poll_lock(n);
5150 /* This NAPI_STATE_SCHED test is for avoiding a race
5151 * with netpoll's poll_napi(). Only the entity which
5152 * obtains the lock and sees NAPI_STATE_SCHED set will
5153 * actually make the ->poll() call. Therefore we avoid
5154 * accidentally calling ->poll() when NAPI is not scheduled.
5157 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
5158 work = n->poll(n, weight);
5159 trace_napi_poll(n, work, weight);
5162 WARN_ON_ONCE(work > weight);
5164 if (likely(work < weight))
5167 /* Drivers must not modify the NAPI state if they
5168 * consume the entire weight. In such cases this code
5169 * still "owns" the NAPI instance and therefore can
5170 * move the instance around on the list at-will.
5172 if (unlikely(napi_disable_pending(n))) {
5178 /* flush too old packets
5179 * If HZ < 1000, flush all packets.
5181 napi_gro_flush(n, HZ >= 1000);
5184 /* Some drivers may have called napi_schedule
5185 * prior to exhausting their budget.
5187 if (unlikely(!list_empty(&n->poll_list))) {
5188 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5189 n->dev ? n->dev->name : "backlog");
5193 list_add_tail(&n->poll_list, repoll);
5196 netpoll_poll_unlock(have);
5201 static void net_rx_action(struct softirq_action *h)
5203 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5204 unsigned long time_limit = jiffies + 2;
5205 int budget = netdev_budget;
5209 local_irq_disable();
5210 list_splice_init(&sd->poll_list, &list);
5214 struct napi_struct *n;
5216 if (list_empty(&list)) {
5217 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
5222 n = list_first_entry(&list, struct napi_struct, poll_list);
5223 budget -= napi_poll(n, &repoll);
5225 /* If softirq window is exhausted then punt.
5226 * Allow this to run for 2 jiffies since which will allow
5227 * an average latency of 1.5/HZ.
5229 if (unlikely(budget <= 0 ||
5230 time_after_eq(jiffies, time_limit))) {
5236 __kfree_skb_flush();
5237 local_irq_disable();
5239 list_splice_tail_init(&sd->poll_list, &list);
5240 list_splice_tail(&repoll, &list);
5241 list_splice(&list, &sd->poll_list);
5242 if (!list_empty(&sd->poll_list))
5243 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5245 net_rps_action_and_irq_enable(sd);
5248 struct netdev_adjacent {
5249 struct net_device *dev;
5251 /* upper master flag, there can only be one master device per list */
5254 /* counter for the number of times this device was added to us */
5257 /* private field for the users */
5260 struct list_head list;
5261 struct rcu_head rcu;
5264 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
5265 struct list_head *adj_list)
5267 struct netdev_adjacent *adj;
5269 list_for_each_entry(adj, adj_list, list) {
5270 if (adj->dev == adj_dev)
5277 * netdev_has_upper_dev - Check if device is linked to an upper device
5279 * @upper_dev: upper device to check
5281 * Find out if a device is linked to specified upper device and return true
5282 * in case it is. Note that this checks only immediate upper device,
5283 * not through a complete stack of devices. The caller must hold the RTNL lock.
5285 bool netdev_has_upper_dev(struct net_device *dev,
5286 struct net_device *upper_dev)
5290 return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper);
5292 EXPORT_SYMBOL(netdev_has_upper_dev);
5295 * netdev_has_any_upper_dev - Check if device is linked to some device
5298 * Find out if a device is linked to an upper device and return true in case
5299 * it is. The caller must hold the RTNL lock.
5301 static bool netdev_has_any_upper_dev(struct net_device *dev)
5305 return !list_empty(&dev->all_adj_list.upper);
5309 * netdev_master_upper_dev_get - Get master upper device
5312 * Find a master upper device and return pointer to it or NULL in case
5313 * it's not there. The caller must hold the RTNL lock.
5315 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5317 struct netdev_adjacent *upper;
5321 if (list_empty(&dev->adj_list.upper))
5324 upper = list_first_entry(&dev->adj_list.upper,
5325 struct netdev_adjacent, list);
5326 if (likely(upper->master))
5330 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5332 void *netdev_adjacent_get_private(struct list_head *adj_list)
5334 struct netdev_adjacent *adj;
5336 adj = list_entry(adj_list, struct netdev_adjacent, list);
5338 return adj->private;
5340 EXPORT_SYMBOL(netdev_adjacent_get_private);
5343 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5345 * @iter: list_head ** of the current position
5347 * Gets the next device from the dev's upper list, starting from iter
5348 * position. The caller must hold RCU read lock.
5350 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5351 struct list_head **iter)
5353 struct netdev_adjacent *upper;
5355 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5357 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5359 if (&upper->list == &dev->adj_list.upper)
5362 *iter = &upper->list;
5366 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5369 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
5371 * @iter: list_head ** of the current position
5373 * Gets the next device from the dev's upper list, starting from iter
5374 * position. The caller must hold RCU read lock.
5376 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
5377 struct list_head **iter)
5379 struct netdev_adjacent *upper;
5381 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5383 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5385 if (&upper->list == &dev->all_adj_list.upper)
5388 *iter = &upper->list;
5392 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
5395 * netdev_lower_get_next_private - Get the next ->private from the
5396 * lower neighbour list
5398 * @iter: list_head ** of the current position
5400 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5401 * list, starting from iter position. The caller must hold either hold the
5402 * RTNL lock or its own locking that guarantees that the neighbour lower
5403 * list will remain unchanged.
5405 void *netdev_lower_get_next_private(struct net_device *dev,
5406 struct list_head **iter)
5408 struct netdev_adjacent *lower;
5410 lower = list_entry(*iter, struct netdev_adjacent, list);
5412 if (&lower->list == &dev->adj_list.lower)
5415 *iter = lower->list.next;
5417 return lower->private;
5419 EXPORT_SYMBOL(netdev_lower_get_next_private);
5422 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5423 * lower neighbour list, RCU
5426 * @iter: list_head ** of the current position
5428 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5429 * list, starting from iter position. The caller must hold RCU read lock.
5431 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5432 struct list_head **iter)
5434 struct netdev_adjacent *lower;
5436 WARN_ON_ONCE(!rcu_read_lock_held());
5438 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5440 if (&lower->list == &dev->adj_list.lower)
5443 *iter = &lower->list;
5445 return lower->private;
5447 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5450 * netdev_lower_get_next - Get the next device from the lower neighbour
5453 * @iter: list_head ** of the current position
5455 * Gets the next netdev_adjacent from the dev's lower neighbour
5456 * list, starting from iter position. The caller must hold RTNL lock or
5457 * its own locking that guarantees that the neighbour lower
5458 * list will remain unchanged.
5460 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5462 struct netdev_adjacent *lower;
5464 lower = list_entry(*iter, struct netdev_adjacent, list);
5466 if (&lower->list == &dev->adj_list.lower)
5469 *iter = lower->list.next;
5473 EXPORT_SYMBOL(netdev_lower_get_next);
5476 * netdev_all_lower_get_next - Get the next device from all lower neighbour list
5478 * @iter: list_head ** of the current position
5480 * Gets the next netdev_adjacent from the dev's all lower neighbour
5481 * list, starting from iter position. The caller must hold RTNL lock or
5482 * its own locking that guarantees that the neighbour all lower
5483 * list will remain unchanged.
5485 struct net_device *netdev_all_lower_get_next(struct net_device *dev, struct list_head **iter)
5487 struct netdev_adjacent *lower;
5489 lower = list_entry(*iter, struct netdev_adjacent, list);
5491 if (&lower->list == &dev->all_adj_list.lower)
5494 *iter = lower->list.next;
5498 EXPORT_SYMBOL(netdev_all_lower_get_next);
5501 * netdev_all_lower_get_next_rcu - Get the next device from all
5502 * lower neighbour list, RCU variant
5504 * @iter: list_head ** of the current position
5506 * Gets the next netdev_adjacent from the dev's all lower neighbour
5507 * list, starting from iter position. The caller must hold RCU read lock.
5509 struct net_device *netdev_all_lower_get_next_rcu(struct net_device *dev,
5510 struct list_head **iter)
5512 struct netdev_adjacent *lower;
5514 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5516 if (&lower->list == &dev->all_adj_list.lower)
5519 *iter = &lower->list;
5523 EXPORT_SYMBOL(netdev_all_lower_get_next_rcu);
5526 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5527 * lower neighbour list, RCU
5531 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5532 * list. The caller must hold RCU read lock.
5534 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5536 struct netdev_adjacent *lower;
5538 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5539 struct netdev_adjacent, list);
5541 return lower->private;
5544 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5547 * netdev_master_upper_dev_get_rcu - Get master upper device
5550 * Find a master upper device and return pointer to it or NULL in case
5551 * it's not there. The caller must hold the RCU read lock.
5553 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5555 struct netdev_adjacent *upper;
5557 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5558 struct netdev_adjacent, list);
5559 if (upper && likely(upper->master))
5563 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5565 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5566 struct net_device *adj_dev,
5567 struct list_head *dev_list)
5569 char linkname[IFNAMSIZ+7];
5570 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5571 "upper_%s" : "lower_%s", adj_dev->name);
5572 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5575 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5577 struct list_head *dev_list)
5579 char linkname[IFNAMSIZ+7];
5580 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5581 "upper_%s" : "lower_%s", name);
5582 sysfs_remove_link(&(dev->dev.kobj), linkname);
5585 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5586 struct net_device *adj_dev,
5587 struct list_head *dev_list)
5589 return (dev_list == &dev->adj_list.upper ||
5590 dev_list == &dev->adj_list.lower) &&
5591 net_eq(dev_net(dev), dev_net(adj_dev));
5594 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5595 struct net_device *adj_dev,
5597 struct list_head *dev_list,
5598 void *private, bool master)
5600 struct netdev_adjacent *adj;
5603 adj = __netdev_find_adj(adj_dev, dev_list);
5606 adj->ref_nr += ref_nr;
5610 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5615 adj->master = master;
5616 adj->ref_nr = ref_nr;
5617 adj->private = private;
5620 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5621 adj_dev->name, dev->name, adj_dev->name);
5623 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5624 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5629 /* Ensure that master link is always the first item in list. */
5631 ret = sysfs_create_link(&(dev->dev.kobj),
5632 &(adj_dev->dev.kobj), "master");
5634 goto remove_symlinks;
5636 list_add_rcu(&adj->list, dev_list);
5638 list_add_tail_rcu(&adj->list, dev_list);
5644 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5645 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5653 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5654 struct net_device *adj_dev,
5656 struct list_head *dev_list)
5658 struct netdev_adjacent *adj;
5660 adj = __netdev_find_adj(adj_dev, dev_list);
5663 pr_err("tried to remove device %s from %s\n",
5664 dev->name, adj_dev->name);
5668 if (adj->ref_nr > ref_nr) {
5669 pr_debug("%s to %s ref_nr-%d = %d\n", dev->name, adj_dev->name,
5670 ref_nr, adj->ref_nr-ref_nr);
5671 adj->ref_nr -= ref_nr;
5676 sysfs_remove_link(&(dev->dev.kobj), "master");
5678 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5679 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5681 list_del_rcu(&adj->list);
5682 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5683 adj_dev->name, dev->name, adj_dev->name);
5685 kfree_rcu(adj, rcu);
5688 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5689 struct net_device *upper_dev,
5691 struct list_head *up_list,
5692 struct list_head *down_list,
5693 void *private, bool master)
5697 ret = __netdev_adjacent_dev_insert(dev, upper_dev, ref_nr, up_list,
5702 ret = __netdev_adjacent_dev_insert(upper_dev, dev, ref_nr, down_list,
5705 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
5712 static int __netdev_adjacent_dev_link(struct net_device *dev,
5713 struct net_device *upper_dev,
5716 return __netdev_adjacent_dev_link_lists(dev, upper_dev, ref_nr,
5717 &dev->all_adj_list.upper,
5718 &upper_dev->all_adj_list.lower,
5722 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5723 struct net_device *upper_dev,
5725 struct list_head *up_list,
5726 struct list_head *down_list)
5728 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
5729 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
5732 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5733 struct net_device *upper_dev,
5736 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, ref_nr,
5737 &dev->all_adj_list.upper,
5738 &upper_dev->all_adj_list.lower);
5741 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5742 struct net_device *upper_dev,
5743 void *private, bool master)
5745 int ret = __netdev_adjacent_dev_link(dev, upper_dev, 1);
5750 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, 1,
5751 &dev->adj_list.upper,
5752 &upper_dev->adj_list.lower,
5755 __netdev_adjacent_dev_unlink(dev, upper_dev, 1);
5762 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5763 struct net_device *upper_dev)
5765 __netdev_adjacent_dev_unlink(dev, upper_dev, 1);
5766 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
5767 &dev->adj_list.upper,
5768 &upper_dev->adj_list.lower);
5771 static int __netdev_upper_dev_link(struct net_device *dev,
5772 struct net_device *upper_dev, bool master,
5773 void *upper_priv, void *upper_info)
5775 struct netdev_notifier_changeupper_info changeupper_info;
5776 struct netdev_adjacent *i, *j, *to_i, *to_j;
5781 if (dev == upper_dev)
5784 /* To prevent loops, check if dev is not upper device to upper_dev. */
5785 if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper))
5788 if (__netdev_find_adj(upper_dev, &dev->adj_list.upper))
5791 if (master && netdev_master_upper_dev_get(dev))
5794 changeupper_info.upper_dev = upper_dev;
5795 changeupper_info.master = master;
5796 changeupper_info.linking = true;
5797 changeupper_info.upper_info = upper_info;
5799 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5800 &changeupper_info.info);
5801 ret = notifier_to_errno(ret);
5805 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
5810 /* Now that we linked these devs, make all the upper_dev's
5811 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5812 * versa, and don't forget the devices itself. All of these
5813 * links are non-neighbours.
5815 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5816 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5817 pr_debug("Interlinking %s with %s, non-neighbour\n",
5818 i->dev->name, j->dev->name);
5819 ret = __netdev_adjacent_dev_link(i->dev, j->dev, i->ref_nr);
5825 /* add dev to every upper_dev's upper device */
5826 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5827 pr_debug("linking %s's upper device %s with %s\n",
5828 upper_dev->name, i->dev->name, dev->name);
5829 ret = __netdev_adjacent_dev_link(dev, i->dev, i->ref_nr);
5831 goto rollback_upper_mesh;
5834 /* add upper_dev to every dev's lower device */
5835 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5836 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5837 i->dev->name, upper_dev->name);
5838 ret = __netdev_adjacent_dev_link(i->dev, upper_dev, i->ref_nr);
5840 goto rollback_lower_mesh;
5843 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5844 &changeupper_info.info);
5845 ret = notifier_to_errno(ret);
5847 goto rollback_lower_mesh;
5851 rollback_lower_mesh:
5853 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5856 __netdev_adjacent_dev_unlink(i->dev, upper_dev, i->ref_nr);
5861 rollback_upper_mesh:
5863 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5866 __netdev_adjacent_dev_unlink(dev, i->dev, i->ref_nr);
5874 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5875 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5876 if (i == to_i && j == to_j)
5878 __netdev_adjacent_dev_unlink(i->dev, j->dev, i->ref_nr);
5884 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5890 * netdev_upper_dev_link - Add a link to the upper device
5892 * @upper_dev: new upper device
5894 * Adds a link to device which is upper to this one. The caller must hold
5895 * the RTNL lock. On a failure a negative errno code is returned.
5896 * On success the reference counts are adjusted and the function
5899 int netdev_upper_dev_link(struct net_device *dev,
5900 struct net_device *upper_dev)
5902 return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL);
5904 EXPORT_SYMBOL(netdev_upper_dev_link);
5907 * netdev_master_upper_dev_link - Add a master link to the upper device
5909 * @upper_dev: new upper device
5910 * @upper_priv: upper device private
5911 * @upper_info: upper info to be passed down via notifier
5913 * Adds a link to device which is upper to this one. In this case, only
5914 * one master upper device can be linked, although other non-master devices
5915 * might be linked as well. The caller must hold the RTNL lock.
5916 * On a failure a negative errno code is returned. On success the reference
5917 * counts are adjusted and the function returns zero.
5919 int netdev_master_upper_dev_link(struct net_device *dev,
5920 struct net_device *upper_dev,
5921 void *upper_priv, void *upper_info)
5923 return __netdev_upper_dev_link(dev, upper_dev, true,
5924 upper_priv, upper_info);
5926 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5929 * netdev_upper_dev_unlink - Removes a link to upper device
5931 * @upper_dev: new upper device
5933 * Removes a link to device which is upper to this one. The caller must hold
5936 void netdev_upper_dev_unlink(struct net_device *dev,
5937 struct net_device *upper_dev)
5939 struct netdev_notifier_changeupper_info changeupper_info;
5940 struct netdev_adjacent *i, *j;
5943 changeupper_info.upper_dev = upper_dev;
5944 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
5945 changeupper_info.linking = false;
5947 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5948 &changeupper_info.info);
5950 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5952 /* Here is the tricky part. We must remove all dev's lower
5953 * devices from all upper_dev's upper devices and vice
5954 * versa, to maintain the graph relationship.
5956 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5957 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5958 __netdev_adjacent_dev_unlink(i->dev, j->dev, i->ref_nr);
5960 /* remove also the devices itself from lower/upper device
5963 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5964 __netdev_adjacent_dev_unlink(i->dev, upper_dev, i->ref_nr);
5966 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5967 __netdev_adjacent_dev_unlink(dev, i->dev, i->ref_nr);
5969 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5970 &changeupper_info.info);
5972 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5975 * netdev_bonding_info_change - Dispatch event about slave change
5977 * @bonding_info: info to dispatch
5979 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5980 * The caller must hold the RTNL lock.
5982 void netdev_bonding_info_change(struct net_device *dev,
5983 struct netdev_bonding_info *bonding_info)
5985 struct netdev_notifier_bonding_info info;
5987 memcpy(&info.bonding_info, bonding_info,
5988 sizeof(struct netdev_bonding_info));
5989 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5992 EXPORT_SYMBOL(netdev_bonding_info_change);
5994 static void netdev_adjacent_add_links(struct net_device *dev)
5996 struct netdev_adjacent *iter;
5998 struct net *net = dev_net(dev);
6000 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6001 if (!net_eq(net, dev_net(iter->dev)))
6003 netdev_adjacent_sysfs_add(iter->dev, dev,
6004 &iter->dev->adj_list.lower);
6005 netdev_adjacent_sysfs_add(dev, iter->dev,
6006 &dev->adj_list.upper);
6009 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6010 if (!net_eq(net, dev_net(iter->dev)))
6012 netdev_adjacent_sysfs_add(iter->dev, dev,
6013 &iter->dev->adj_list.upper);
6014 netdev_adjacent_sysfs_add(dev, iter->dev,
6015 &dev->adj_list.lower);
6019 static void netdev_adjacent_del_links(struct net_device *dev)
6021 struct netdev_adjacent *iter;
6023 struct net *net = dev_net(dev);
6025 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6026 if (!net_eq(net, dev_net(iter->dev)))
6028 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6029 &iter->dev->adj_list.lower);
6030 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6031 &dev->adj_list.upper);
6034 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6035 if (!net_eq(net, dev_net(iter->dev)))
6037 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6038 &iter->dev->adj_list.upper);
6039 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6040 &dev->adj_list.lower);
6044 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
6046 struct netdev_adjacent *iter;
6048 struct net *net = dev_net(dev);
6050 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6051 if (!net_eq(net, dev_net(iter->dev)))
6053 netdev_adjacent_sysfs_del(iter->dev, oldname,
6054 &iter->dev->adj_list.lower);
6055 netdev_adjacent_sysfs_add(iter->dev, dev,
6056 &iter->dev->adj_list.lower);
6059 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6060 if (!net_eq(net, dev_net(iter->dev)))
6062 netdev_adjacent_sysfs_del(iter->dev, oldname,
6063 &iter->dev->adj_list.upper);
6064 netdev_adjacent_sysfs_add(iter->dev, dev,
6065 &iter->dev->adj_list.upper);
6069 void *netdev_lower_dev_get_private(struct net_device *dev,
6070 struct net_device *lower_dev)
6072 struct netdev_adjacent *lower;
6076 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
6080 return lower->private;
6082 EXPORT_SYMBOL(netdev_lower_dev_get_private);
6085 int dev_get_nest_level(struct net_device *dev)
6087 struct net_device *lower = NULL;
6088 struct list_head *iter;
6094 netdev_for_each_lower_dev(dev, lower, iter) {
6095 nest = dev_get_nest_level(lower);
6096 if (max_nest < nest)
6100 return max_nest + 1;
6102 EXPORT_SYMBOL(dev_get_nest_level);
6105 * netdev_lower_change - Dispatch event about lower device state change
6106 * @lower_dev: device
6107 * @lower_state_info: state to dispatch
6109 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
6110 * The caller must hold the RTNL lock.
6112 void netdev_lower_state_changed(struct net_device *lower_dev,
6113 void *lower_state_info)
6115 struct netdev_notifier_changelowerstate_info changelowerstate_info;
6118 changelowerstate_info.lower_state_info = lower_state_info;
6119 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev,
6120 &changelowerstate_info.info);
6122 EXPORT_SYMBOL(netdev_lower_state_changed);
6124 int netdev_default_l2upper_neigh_construct(struct net_device *dev,
6125 struct neighbour *n)
6127 struct net_device *lower_dev, *stop_dev;
6128 struct list_head *iter;
6131 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6132 if (!lower_dev->netdev_ops->ndo_neigh_construct)
6134 err = lower_dev->netdev_ops->ndo_neigh_construct(lower_dev, n);
6136 stop_dev = lower_dev;
6143 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6144 if (lower_dev == stop_dev)
6146 if (!lower_dev->netdev_ops->ndo_neigh_destroy)
6148 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n);
6152 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_construct);
6154 void netdev_default_l2upper_neigh_destroy(struct net_device *dev,
6155 struct neighbour *n)
6157 struct net_device *lower_dev;
6158 struct list_head *iter;
6160 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6161 if (!lower_dev->netdev_ops->ndo_neigh_destroy)
6163 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n);
6166 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_destroy);
6168 static void dev_change_rx_flags(struct net_device *dev, int flags)
6170 const struct net_device_ops *ops = dev->netdev_ops;
6172 if (ops->ndo_change_rx_flags)
6173 ops->ndo_change_rx_flags(dev, flags);
6176 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
6178 unsigned int old_flags = dev->flags;
6184 dev->flags |= IFF_PROMISC;
6185 dev->promiscuity += inc;
6186 if (dev->promiscuity == 0) {
6189 * If inc causes overflow, untouch promisc and return error.
6192 dev->flags &= ~IFF_PROMISC;
6194 dev->promiscuity -= inc;
6195 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6200 if (dev->flags != old_flags) {
6201 pr_info("device %s %s promiscuous mode\n",
6203 dev->flags & IFF_PROMISC ? "entered" : "left");
6204 if (audit_enabled) {
6205 current_uid_gid(&uid, &gid);
6206 audit_log(current->audit_context, GFP_ATOMIC,
6207 AUDIT_ANOM_PROMISCUOUS,
6208 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6209 dev->name, (dev->flags & IFF_PROMISC),
6210 (old_flags & IFF_PROMISC),
6211 from_kuid(&init_user_ns, audit_get_loginuid(current)),
6212 from_kuid(&init_user_ns, uid),
6213 from_kgid(&init_user_ns, gid),
6214 audit_get_sessionid(current));
6217 dev_change_rx_flags(dev, IFF_PROMISC);
6220 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
6225 * dev_set_promiscuity - update promiscuity count on a device
6229 * Add or remove promiscuity from a device. While the count in the device
6230 * remains above zero the interface remains promiscuous. Once it hits zero
6231 * the device reverts back to normal filtering operation. A negative inc
6232 * value is used to drop promiscuity on the device.
6233 * Return 0 if successful or a negative errno code on error.
6235 int dev_set_promiscuity(struct net_device *dev, int inc)
6237 unsigned int old_flags = dev->flags;
6240 err = __dev_set_promiscuity(dev, inc, true);
6243 if (dev->flags != old_flags)
6244 dev_set_rx_mode(dev);
6247 EXPORT_SYMBOL(dev_set_promiscuity);
6249 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
6251 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
6255 dev->flags |= IFF_ALLMULTI;
6256 dev->allmulti += inc;
6257 if (dev->allmulti == 0) {
6260 * If inc causes overflow, untouch allmulti and return error.
6263 dev->flags &= ~IFF_ALLMULTI;
6265 dev->allmulti -= inc;
6266 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6271 if (dev->flags ^ old_flags) {
6272 dev_change_rx_flags(dev, IFF_ALLMULTI);
6273 dev_set_rx_mode(dev);
6275 __dev_notify_flags(dev, old_flags,
6276 dev->gflags ^ old_gflags);
6282 * dev_set_allmulti - update allmulti count on a device
6286 * Add or remove reception of all multicast frames to a device. While the
6287 * count in the device remains above zero the interface remains listening
6288 * to all interfaces. Once it hits zero the device reverts back to normal
6289 * filtering operation. A negative @inc value is used to drop the counter
6290 * when releasing a resource needing all multicasts.
6291 * Return 0 if successful or a negative errno code on error.
6294 int dev_set_allmulti(struct net_device *dev, int inc)
6296 return __dev_set_allmulti(dev, inc, true);
6298 EXPORT_SYMBOL(dev_set_allmulti);
6301 * Upload unicast and multicast address lists to device and
6302 * configure RX filtering. When the device doesn't support unicast
6303 * filtering it is put in promiscuous mode while unicast addresses
6306 void __dev_set_rx_mode(struct net_device *dev)
6308 const struct net_device_ops *ops = dev->netdev_ops;
6310 /* dev_open will call this function so the list will stay sane. */
6311 if (!(dev->flags&IFF_UP))
6314 if (!netif_device_present(dev))
6317 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
6318 /* Unicast addresses changes may only happen under the rtnl,
6319 * therefore calling __dev_set_promiscuity here is safe.
6321 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
6322 __dev_set_promiscuity(dev, 1, false);
6323 dev->uc_promisc = true;
6324 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
6325 __dev_set_promiscuity(dev, -1, false);
6326 dev->uc_promisc = false;
6330 if (ops->ndo_set_rx_mode)
6331 ops->ndo_set_rx_mode(dev);
6334 void dev_set_rx_mode(struct net_device *dev)
6336 netif_addr_lock_bh(dev);
6337 __dev_set_rx_mode(dev);
6338 netif_addr_unlock_bh(dev);
6342 * dev_get_flags - get flags reported to userspace
6345 * Get the combination of flag bits exported through APIs to userspace.
6347 unsigned int dev_get_flags(const struct net_device *dev)
6351 flags = (dev->flags & ~(IFF_PROMISC |
6356 (dev->gflags & (IFF_PROMISC |
6359 if (netif_running(dev)) {
6360 if (netif_oper_up(dev))
6361 flags |= IFF_RUNNING;
6362 if (netif_carrier_ok(dev))
6363 flags |= IFF_LOWER_UP;
6364 if (netif_dormant(dev))
6365 flags |= IFF_DORMANT;
6370 EXPORT_SYMBOL(dev_get_flags);
6372 int __dev_change_flags(struct net_device *dev, unsigned int flags)
6374 unsigned int old_flags = dev->flags;
6380 * Set the flags on our device.
6383 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
6384 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
6386 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6390 * Load in the correct multicast list now the flags have changed.
6393 if ((old_flags ^ flags) & IFF_MULTICAST)
6394 dev_change_rx_flags(dev, IFF_MULTICAST);
6396 dev_set_rx_mode(dev);
6399 * Have we downed the interface. We handle IFF_UP ourselves
6400 * according to user attempts to set it, rather than blindly
6405 if ((old_flags ^ flags) & IFF_UP)
6406 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
6408 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6409 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6410 unsigned int old_flags = dev->flags;
6412 dev->gflags ^= IFF_PROMISC;
6414 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6415 if (dev->flags != old_flags)
6416 dev_set_rx_mode(dev);
6419 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6420 is important. Some (broken) drivers set IFF_PROMISC, when
6421 IFF_ALLMULTI is requested not asking us and not reporting.
6423 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6424 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6426 dev->gflags ^= IFF_ALLMULTI;
6427 __dev_set_allmulti(dev, inc, false);
6433 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6434 unsigned int gchanges)
6436 unsigned int changes = dev->flags ^ old_flags;
6439 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6441 if (changes & IFF_UP) {
6442 if (dev->flags & IFF_UP)
6443 call_netdevice_notifiers(NETDEV_UP, dev);
6445 call_netdevice_notifiers(NETDEV_DOWN, dev);
6448 if (dev->flags & IFF_UP &&
6449 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6450 struct netdev_notifier_change_info change_info;
6452 change_info.flags_changed = changes;
6453 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6459 * dev_change_flags - change device settings
6461 * @flags: device state flags
6463 * Change settings on device based state flags. The flags are
6464 * in the userspace exported format.
6466 int dev_change_flags(struct net_device *dev, unsigned int flags)
6469 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6471 ret = __dev_change_flags(dev, flags);
6475 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6476 __dev_notify_flags(dev, old_flags, changes);
6479 EXPORT_SYMBOL(dev_change_flags);
6481 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6483 const struct net_device_ops *ops = dev->netdev_ops;
6485 if (ops->ndo_change_mtu)
6486 return ops->ndo_change_mtu(dev, new_mtu);
6493 * dev_set_mtu - Change maximum transfer unit
6495 * @new_mtu: new transfer unit
6497 * Change the maximum transfer size of the network device.
6499 int dev_set_mtu(struct net_device *dev, int new_mtu)
6503 if (new_mtu == dev->mtu)
6506 /* MTU must be positive. */
6510 if (!netif_device_present(dev))
6513 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6514 err = notifier_to_errno(err);
6518 orig_mtu = dev->mtu;
6519 err = __dev_set_mtu(dev, new_mtu);
6522 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6523 err = notifier_to_errno(err);
6525 /* setting mtu back and notifying everyone again,
6526 * so that they have a chance to revert changes.
6528 __dev_set_mtu(dev, orig_mtu);
6529 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6534 EXPORT_SYMBOL(dev_set_mtu);
6537 * dev_set_group - Change group this device belongs to
6539 * @new_group: group this device should belong to
6541 void dev_set_group(struct net_device *dev, int new_group)
6543 dev->group = new_group;
6545 EXPORT_SYMBOL(dev_set_group);
6548 * dev_set_mac_address - Change Media Access Control Address
6552 * Change the hardware (MAC) address of the device
6554 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6556 const struct net_device_ops *ops = dev->netdev_ops;
6559 if (!ops->ndo_set_mac_address)
6561 if (sa->sa_family != dev->type)
6563 if (!netif_device_present(dev))
6565 err = ops->ndo_set_mac_address(dev, sa);
6568 dev->addr_assign_type = NET_ADDR_SET;
6569 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6570 add_device_randomness(dev->dev_addr, dev->addr_len);
6573 EXPORT_SYMBOL(dev_set_mac_address);
6576 * dev_change_carrier - Change device carrier
6578 * @new_carrier: new value
6580 * Change device carrier
6582 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6584 const struct net_device_ops *ops = dev->netdev_ops;
6586 if (!ops->ndo_change_carrier)
6588 if (!netif_device_present(dev))
6590 return ops->ndo_change_carrier(dev, new_carrier);
6592 EXPORT_SYMBOL(dev_change_carrier);
6595 * dev_get_phys_port_id - Get device physical port ID
6599 * Get device physical port ID
6601 int dev_get_phys_port_id(struct net_device *dev,
6602 struct netdev_phys_item_id *ppid)
6604 const struct net_device_ops *ops = dev->netdev_ops;
6606 if (!ops->ndo_get_phys_port_id)
6608 return ops->ndo_get_phys_port_id(dev, ppid);
6610 EXPORT_SYMBOL(dev_get_phys_port_id);
6613 * dev_get_phys_port_name - Get device physical port name
6616 * @len: limit of bytes to copy to name
6618 * Get device physical port name
6620 int dev_get_phys_port_name(struct net_device *dev,
6621 char *name, size_t len)
6623 const struct net_device_ops *ops = dev->netdev_ops;
6625 if (!ops->ndo_get_phys_port_name)
6627 return ops->ndo_get_phys_port_name(dev, name, len);
6629 EXPORT_SYMBOL(dev_get_phys_port_name);
6632 * dev_change_proto_down - update protocol port state information
6634 * @proto_down: new value
6636 * This info can be used by switch drivers to set the phys state of the
6639 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6641 const struct net_device_ops *ops = dev->netdev_ops;
6643 if (!ops->ndo_change_proto_down)
6645 if (!netif_device_present(dev))
6647 return ops->ndo_change_proto_down(dev, proto_down);
6649 EXPORT_SYMBOL(dev_change_proto_down);
6652 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
6654 * @fd: new program fd or negative value to clear
6656 * Set or clear a bpf program for a device
6658 int dev_change_xdp_fd(struct net_device *dev, int fd)
6660 const struct net_device_ops *ops = dev->netdev_ops;
6661 struct bpf_prog *prog = NULL;
6662 struct netdev_xdp xdp = {};
6668 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP);
6670 return PTR_ERR(prog);
6673 xdp.command = XDP_SETUP_PROG;
6675 err = ops->ndo_xdp(dev, &xdp);
6676 if (err < 0 && prog)
6681 EXPORT_SYMBOL(dev_change_xdp_fd);
6684 * dev_new_index - allocate an ifindex
6685 * @net: the applicable net namespace
6687 * Returns a suitable unique value for a new device interface
6688 * number. The caller must hold the rtnl semaphore or the
6689 * dev_base_lock to be sure it remains unique.
6691 static int dev_new_index(struct net *net)
6693 int ifindex = net->ifindex;
6697 if (!__dev_get_by_index(net, ifindex))
6698 return net->ifindex = ifindex;
6702 /* Delayed registration/unregisteration */
6703 static LIST_HEAD(net_todo_list);
6704 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6706 static void net_set_todo(struct net_device *dev)
6708 list_add_tail(&dev->todo_list, &net_todo_list);
6709 dev_net(dev)->dev_unreg_count++;
6712 static void rollback_registered_many(struct list_head *head)
6714 struct net_device *dev, *tmp;
6715 LIST_HEAD(close_head);
6717 BUG_ON(dev_boot_phase);
6720 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6721 /* Some devices call without registering
6722 * for initialization unwind. Remove those
6723 * devices and proceed with the remaining.
6725 if (dev->reg_state == NETREG_UNINITIALIZED) {
6726 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6730 list_del(&dev->unreg_list);
6733 dev->dismantle = true;
6734 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6737 /* If device is running, close it first. */
6738 list_for_each_entry(dev, head, unreg_list)
6739 list_add_tail(&dev->close_list, &close_head);
6740 dev_close_many(&close_head, true);
6742 list_for_each_entry(dev, head, unreg_list) {
6743 /* And unlink it from device chain. */
6744 unlist_netdevice(dev);
6746 dev->reg_state = NETREG_UNREGISTERING;
6748 flush_all_backlogs();
6752 list_for_each_entry(dev, head, unreg_list) {
6753 struct sk_buff *skb = NULL;
6755 /* Shutdown queueing discipline. */
6759 /* Notify protocols, that we are about to destroy
6760 this device. They should clean all the things.
6762 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6764 if (!dev->rtnl_link_ops ||
6765 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6766 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6770 * Flush the unicast and multicast chains
6775 if (dev->netdev_ops->ndo_uninit)
6776 dev->netdev_ops->ndo_uninit(dev);
6779 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6781 /* Notifier chain MUST detach us all upper devices. */
6782 WARN_ON(netdev_has_any_upper_dev(dev));
6784 /* Remove entries from kobject tree */
6785 netdev_unregister_kobject(dev);
6787 /* Remove XPS queueing entries */
6788 netif_reset_xps_queues_gt(dev, 0);
6794 list_for_each_entry(dev, head, unreg_list)
6798 static void rollback_registered(struct net_device *dev)
6802 list_add(&dev->unreg_list, &single);
6803 rollback_registered_many(&single);
6807 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6808 struct net_device *upper, netdev_features_t features)
6810 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6811 netdev_features_t feature;
6814 for_each_netdev_feature(&upper_disables, feature_bit) {
6815 feature = __NETIF_F_BIT(feature_bit);
6816 if (!(upper->wanted_features & feature)
6817 && (features & feature)) {
6818 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6819 &feature, upper->name);
6820 features &= ~feature;
6827 static void netdev_sync_lower_features(struct net_device *upper,
6828 struct net_device *lower, netdev_features_t features)
6830 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6831 netdev_features_t feature;
6834 for_each_netdev_feature(&upper_disables, feature_bit) {
6835 feature = __NETIF_F_BIT(feature_bit);
6836 if (!(features & feature) && (lower->features & feature)) {
6837 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6838 &feature, lower->name);
6839 lower->wanted_features &= ~feature;
6840 netdev_update_features(lower);
6842 if (unlikely(lower->features & feature))
6843 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6844 &feature, lower->name);
6849 static netdev_features_t netdev_fix_features(struct net_device *dev,
6850 netdev_features_t features)
6852 /* Fix illegal checksum combinations */
6853 if ((features & NETIF_F_HW_CSUM) &&
6854 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6855 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6856 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6859 /* TSO requires that SG is present as well. */
6860 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6861 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6862 features &= ~NETIF_F_ALL_TSO;
6865 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6866 !(features & NETIF_F_IP_CSUM)) {
6867 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6868 features &= ~NETIF_F_TSO;
6869 features &= ~NETIF_F_TSO_ECN;
6872 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6873 !(features & NETIF_F_IPV6_CSUM)) {
6874 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6875 features &= ~NETIF_F_TSO6;
6878 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
6879 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
6880 features &= ~NETIF_F_TSO_MANGLEID;
6882 /* TSO ECN requires that TSO is present as well. */
6883 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6884 features &= ~NETIF_F_TSO_ECN;
6886 /* Software GSO depends on SG. */
6887 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6888 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6889 features &= ~NETIF_F_GSO;
6892 /* UFO needs SG and checksumming */
6893 if (features & NETIF_F_UFO) {
6894 /* maybe split UFO into V4 and V6? */
6895 if (!(features & NETIF_F_HW_CSUM) &&
6896 ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) !=
6897 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) {
6899 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6900 features &= ~NETIF_F_UFO;
6903 if (!(features & NETIF_F_SG)) {
6905 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6906 features &= ~NETIF_F_UFO;
6910 /* GSO partial features require GSO partial be set */
6911 if ((features & dev->gso_partial_features) &&
6912 !(features & NETIF_F_GSO_PARTIAL)) {
6914 "Dropping partially supported GSO features since no GSO partial.\n");
6915 features &= ~dev->gso_partial_features;
6918 #ifdef CONFIG_NET_RX_BUSY_POLL
6919 if (dev->netdev_ops->ndo_busy_poll)
6920 features |= NETIF_F_BUSY_POLL;
6923 features &= ~NETIF_F_BUSY_POLL;
6928 int __netdev_update_features(struct net_device *dev)
6930 struct net_device *upper, *lower;
6931 netdev_features_t features;
6932 struct list_head *iter;
6937 features = netdev_get_wanted_features(dev);
6939 if (dev->netdev_ops->ndo_fix_features)
6940 features = dev->netdev_ops->ndo_fix_features(dev, features);
6942 /* driver might be less strict about feature dependencies */
6943 features = netdev_fix_features(dev, features);
6945 /* some features can't be enabled if they're off an an upper device */
6946 netdev_for_each_upper_dev_rcu(dev, upper, iter)
6947 features = netdev_sync_upper_features(dev, upper, features);
6949 if (dev->features == features)
6952 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6953 &dev->features, &features);
6955 if (dev->netdev_ops->ndo_set_features)
6956 err = dev->netdev_ops->ndo_set_features(dev, features);
6960 if (unlikely(err < 0)) {
6962 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6963 err, &features, &dev->features);
6964 /* return non-0 since some features might have changed and
6965 * it's better to fire a spurious notification than miss it
6971 /* some features must be disabled on lower devices when disabled
6972 * on an upper device (think: bonding master or bridge)
6974 netdev_for_each_lower_dev(dev, lower, iter)
6975 netdev_sync_lower_features(dev, lower, features);
6978 dev->features = features;
6980 return err < 0 ? 0 : 1;
6984 * netdev_update_features - recalculate device features
6985 * @dev: the device to check
6987 * Recalculate dev->features set and send notifications if it
6988 * has changed. Should be called after driver or hardware dependent
6989 * conditions might have changed that influence the features.
6991 void netdev_update_features(struct net_device *dev)
6993 if (__netdev_update_features(dev))
6994 netdev_features_change(dev);
6996 EXPORT_SYMBOL(netdev_update_features);
6999 * netdev_change_features - recalculate device features
7000 * @dev: the device to check
7002 * Recalculate dev->features set and send notifications even
7003 * if they have not changed. Should be called instead of
7004 * netdev_update_features() if also dev->vlan_features might
7005 * have changed to allow the changes to be propagated to stacked
7008 void netdev_change_features(struct net_device *dev)
7010 __netdev_update_features(dev);
7011 netdev_features_change(dev);
7013 EXPORT_SYMBOL(netdev_change_features);
7016 * netif_stacked_transfer_operstate - transfer operstate
7017 * @rootdev: the root or lower level device to transfer state from
7018 * @dev: the device to transfer operstate to
7020 * Transfer operational state from root to device. This is normally
7021 * called when a stacking relationship exists between the root
7022 * device and the device(a leaf device).
7024 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
7025 struct net_device *dev)
7027 if (rootdev->operstate == IF_OPER_DORMANT)
7028 netif_dormant_on(dev);
7030 netif_dormant_off(dev);
7032 if (netif_carrier_ok(rootdev)) {
7033 if (!netif_carrier_ok(dev))
7034 netif_carrier_on(dev);
7036 if (netif_carrier_ok(dev))
7037 netif_carrier_off(dev);
7040 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
7043 static int netif_alloc_rx_queues(struct net_device *dev)
7045 unsigned int i, count = dev->num_rx_queues;
7046 struct netdev_rx_queue *rx;
7047 size_t sz = count * sizeof(*rx);
7051 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7059 for (i = 0; i < count; i++)
7065 static void netdev_init_one_queue(struct net_device *dev,
7066 struct netdev_queue *queue, void *_unused)
7068 /* Initialize queue lock */
7069 spin_lock_init(&queue->_xmit_lock);
7070 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
7071 queue->xmit_lock_owner = -1;
7072 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
7075 dql_init(&queue->dql, HZ);
7079 static void netif_free_tx_queues(struct net_device *dev)
7084 static int netif_alloc_netdev_queues(struct net_device *dev)
7086 unsigned int count = dev->num_tx_queues;
7087 struct netdev_queue *tx;
7088 size_t sz = count * sizeof(*tx);
7090 if (count < 1 || count > 0xffff)
7093 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7101 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
7102 spin_lock_init(&dev->tx_global_lock);
7107 void netif_tx_stop_all_queues(struct net_device *dev)
7111 for (i = 0; i < dev->num_tx_queues; i++) {
7112 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
7113 netif_tx_stop_queue(txq);
7116 EXPORT_SYMBOL(netif_tx_stop_all_queues);
7119 * register_netdevice - register a network device
7120 * @dev: device to register
7122 * Take a completed network device structure and add it to the kernel
7123 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7124 * chain. 0 is returned on success. A negative errno code is returned
7125 * on a failure to set up the device, or if the name is a duplicate.
7127 * Callers must hold the rtnl semaphore. You may want
7128 * register_netdev() instead of this.
7131 * The locking appears insufficient to guarantee two parallel registers
7132 * will not get the same name.
7135 int register_netdevice(struct net_device *dev)
7138 struct net *net = dev_net(dev);
7140 BUG_ON(dev_boot_phase);
7145 /* When net_device's are persistent, this will be fatal. */
7146 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
7149 spin_lock_init(&dev->addr_list_lock);
7150 netdev_set_addr_lockdep_class(dev);
7152 ret = dev_get_valid_name(net, dev, dev->name);
7156 /* Init, if this function is available */
7157 if (dev->netdev_ops->ndo_init) {
7158 ret = dev->netdev_ops->ndo_init(dev);
7166 if (((dev->hw_features | dev->features) &
7167 NETIF_F_HW_VLAN_CTAG_FILTER) &&
7168 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
7169 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
7170 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
7177 dev->ifindex = dev_new_index(net);
7178 else if (__dev_get_by_index(net, dev->ifindex))
7181 /* Transfer changeable features to wanted_features and enable
7182 * software offloads (GSO and GRO).
7184 dev->hw_features |= NETIF_F_SOFT_FEATURES;
7185 dev->features |= NETIF_F_SOFT_FEATURES;
7186 dev->wanted_features = dev->features & dev->hw_features;
7188 if (!(dev->flags & IFF_LOOPBACK))
7189 dev->hw_features |= NETIF_F_NOCACHE_COPY;
7191 /* If IPv4 TCP segmentation offload is supported we should also
7192 * allow the device to enable segmenting the frame with the option
7193 * of ignoring a static IP ID value. This doesn't enable the
7194 * feature itself but allows the user to enable it later.
7196 if (dev->hw_features & NETIF_F_TSO)
7197 dev->hw_features |= NETIF_F_TSO_MANGLEID;
7198 if (dev->vlan_features & NETIF_F_TSO)
7199 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
7200 if (dev->mpls_features & NETIF_F_TSO)
7201 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
7202 if (dev->hw_enc_features & NETIF_F_TSO)
7203 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
7205 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
7207 dev->vlan_features |= NETIF_F_HIGHDMA;
7209 /* Make NETIF_F_SG inheritable to tunnel devices.
7211 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
7213 /* Make NETIF_F_SG inheritable to MPLS.
7215 dev->mpls_features |= NETIF_F_SG;
7217 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
7218 ret = notifier_to_errno(ret);
7222 ret = netdev_register_kobject(dev);
7225 dev->reg_state = NETREG_REGISTERED;
7227 __netdev_update_features(dev);
7230 * Default initial state at registry is that the
7231 * device is present.
7234 set_bit(__LINK_STATE_PRESENT, &dev->state);
7236 linkwatch_init_dev(dev);
7238 dev_init_scheduler(dev);
7240 list_netdevice(dev);
7241 add_device_randomness(dev->dev_addr, dev->addr_len);
7243 /* If the device has permanent device address, driver should
7244 * set dev_addr and also addr_assign_type should be set to
7245 * NET_ADDR_PERM (default value).
7247 if (dev->addr_assign_type == NET_ADDR_PERM)
7248 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
7250 /* Notify protocols, that a new device appeared. */
7251 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
7252 ret = notifier_to_errno(ret);
7254 rollback_registered(dev);
7255 dev->reg_state = NETREG_UNREGISTERED;
7258 * Prevent userspace races by waiting until the network
7259 * device is fully setup before sending notifications.
7261 if (!dev->rtnl_link_ops ||
7262 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7263 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7269 if (dev->netdev_ops->ndo_uninit)
7270 dev->netdev_ops->ndo_uninit(dev);
7273 EXPORT_SYMBOL(register_netdevice);
7276 * init_dummy_netdev - init a dummy network device for NAPI
7277 * @dev: device to init
7279 * This takes a network device structure and initialize the minimum
7280 * amount of fields so it can be used to schedule NAPI polls without
7281 * registering a full blown interface. This is to be used by drivers
7282 * that need to tie several hardware interfaces to a single NAPI
7283 * poll scheduler due to HW limitations.
7285 int init_dummy_netdev(struct net_device *dev)
7287 /* Clear everything. Note we don't initialize spinlocks
7288 * are they aren't supposed to be taken by any of the
7289 * NAPI code and this dummy netdev is supposed to be
7290 * only ever used for NAPI polls
7292 memset(dev, 0, sizeof(struct net_device));
7294 /* make sure we BUG if trying to hit standard
7295 * register/unregister code path
7297 dev->reg_state = NETREG_DUMMY;
7299 /* NAPI wants this */
7300 INIT_LIST_HEAD(&dev->napi_list);
7302 /* a dummy interface is started by default */
7303 set_bit(__LINK_STATE_PRESENT, &dev->state);
7304 set_bit(__LINK_STATE_START, &dev->state);
7306 /* Note : We dont allocate pcpu_refcnt for dummy devices,
7307 * because users of this 'device' dont need to change
7313 EXPORT_SYMBOL_GPL(init_dummy_netdev);
7317 * register_netdev - register a network device
7318 * @dev: device to register
7320 * Take a completed network device structure and add it to the kernel
7321 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7322 * chain. 0 is returned on success. A negative errno code is returned
7323 * on a failure to set up the device, or if the name is a duplicate.
7325 * This is a wrapper around register_netdevice that takes the rtnl semaphore
7326 * and expands the device name if you passed a format string to
7329 int register_netdev(struct net_device *dev)
7334 err = register_netdevice(dev);
7338 EXPORT_SYMBOL(register_netdev);
7340 int netdev_refcnt_read(const struct net_device *dev)
7344 for_each_possible_cpu(i)
7345 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
7348 EXPORT_SYMBOL(netdev_refcnt_read);
7351 * netdev_wait_allrefs - wait until all references are gone.
7352 * @dev: target net_device
7354 * This is called when unregistering network devices.
7356 * Any protocol or device that holds a reference should register
7357 * for netdevice notification, and cleanup and put back the
7358 * reference if they receive an UNREGISTER event.
7359 * We can get stuck here if buggy protocols don't correctly
7362 static void netdev_wait_allrefs(struct net_device *dev)
7364 unsigned long rebroadcast_time, warning_time;
7367 linkwatch_forget_dev(dev);
7369 rebroadcast_time = warning_time = jiffies;
7370 refcnt = netdev_refcnt_read(dev);
7372 while (refcnt != 0) {
7373 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
7376 /* Rebroadcast unregister notification */
7377 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7383 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7384 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
7386 /* We must not have linkwatch events
7387 * pending on unregister. If this
7388 * happens, we simply run the queue
7389 * unscheduled, resulting in a noop
7392 linkwatch_run_queue();
7397 rebroadcast_time = jiffies;
7402 refcnt = netdev_refcnt_read(dev);
7404 if (time_after(jiffies, warning_time + 10 * HZ)) {
7405 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
7407 warning_time = jiffies;
7416 * register_netdevice(x1);
7417 * register_netdevice(x2);
7419 * unregister_netdevice(y1);
7420 * unregister_netdevice(y2);
7426 * We are invoked by rtnl_unlock().
7427 * This allows us to deal with problems:
7428 * 1) We can delete sysfs objects which invoke hotplug
7429 * without deadlocking with linkwatch via keventd.
7430 * 2) Since we run with the RTNL semaphore not held, we can sleep
7431 * safely in order to wait for the netdev refcnt to drop to zero.
7433 * We must not return until all unregister events added during
7434 * the interval the lock was held have been completed.
7436 void netdev_run_todo(void)
7438 struct list_head list;
7440 /* Snapshot list, allow later requests */
7441 list_replace_init(&net_todo_list, &list);
7446 /* Wait for rcu callbacks to finish before next phase */
7447 if (!list_empty(&list))
7450 while (!list_empty(&list)) {
7451 struct net_device *dev
7452 = list_first_entry(&list, struct net_device, todo_list);
7453 list_del(&dev->todo_list);
7456 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7459 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
7460 pr_err("network todo '%s' but state %d\n",
7461 dev->name, dev->reg_state);
7466 dev->reg_state = NETREG_UNREGISTERED;
7468 netdev_wait_allrefs(dev);
7471 BUG_ON(netdev_refcnt_read(dev));
7472 BUG_ON(!list_empty(&dev->ptype_all));
7473 BUG_ON(!list_empty(&dev->ptype_specific));
7474 WARN_ON(rcu_access_pointer(dev->ip_ptr));
7475 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
7476 WARN_ON(dev->dn_ptr);
7478 if (dev->destructor)
7479 dev->destructor(dev);
7481 /* Report a network device has been unregistered */
7483 dev_net(dev)->dev_unreg_count--;
7485 wake_up(&netdev_unregistering_wq);
7487 /* Free network device */
7488 kobject_put(&dev->dev.kobj);
7492 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
7493 * all the same fields in the same order as net_device_stats, with only
7494 * the type differing, but rtnl_link_stats64 may have additional fields
7495 * at the end for newer counters.
7497 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7498 const struct net_device_stats *netdev_stats)
7500 #if BITS_PER_LONG == 64
7501 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
7502 memcpy(stats64, netdev_stats, sizeof(*stats64));
7503 /* zero out counters that only exist in rtnl_link_stats64 */
7504 memset((char *)stats64 + sizeof(*netdev_stats), 0,
7505 sizeof(*stats64) - sizeof(*netdev_stats));
7507 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
7508 const unsigned long *src = (const unsigned long *)netdev_stats;
7509 u64 *dst = (u64 *)stats64;
7511 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
7512 for (i = 0; i < n; i++)
7514 /* zero out counters that only exist in rtnl_link_stats64 */
7515 memset((char *)stats64 + n * sizeof(u64), 0,
7516 sizeof(*stats64) - n * sizeof(u64));
7519 EXPORT_SYMBOL(netdev_stats_to_stats64);
7522 * dev_get_stats - get network device statistics
7523 * @dev: device to get statistics from
7524 * @storage: place to store stats
7526 * Get network statistics from device. Return @storage.
7527 * The device driver may provide its own method by setting
7528 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7529 * otherwise the internal statistics structure is used.
7531 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
7532 struct rtnl_link_stats64 *storage)
7534 const struct net_device_ops *ops = dev->netdev_ops;
7536 if (ops->ndo_get_stats64) {
7537 memset(storage, 0, sizeof(*storage));
7538 ops->ndo_get_stats64(dev, storage);
7539 } else if (ops->ndo_get_stats) {
7540 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7542 netdev_stats_to_stats64(storage, &dev->stats);
7544 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
7545 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
7546 storage->rx_nohandler += atomic_long_read(&dev->rx_nohandler);
7549 EXPORT_SYMBOL(dev_get_stats);
7551 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7553 struct netdev_queue *queue = dev_ingress_queue(dev);
7555 #ifdef CONFIG_NET_CLS_ACT
7558 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7561 netdev_init_one_queue(dev, queue, NULL);
7562 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7563 queue->qdisc_sleeping = &noop_qdisc;
7564 rcu_assign_pointer(dev->ingress_queue, queue);
7569 static const struct ethtool_ops default_ethtool_ops;
7571 void netdev_set_default_ethtool_ops(struct net_device *dev,
7572 const struct ethtool_ops *ops)
7574 if (dev->ethtool_ops == &default_ethtool_ops)
7575 dev->ethtool_ops = ops;
7577 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7579 void netdev_freemem(struct net_device *dev)
7581 char *addr = (char *)dev - dev->padded;
7587 * alloc_netdev_mqs - allocate network device
7588 * @sizeof_priv: size of private data to allocate space for
7589 * @name: device name format string
7590 * @name_assign_type: origin of device name
7591 * @setup: callback to initialize device
7592 * @txqs: the number of TX subqueues to allocate
7593 * @rxqs: the number of RX subqueues to allocate
7595 * Allocates a struct net_device with private data area for driver use
7596 * and performs basic initialization. Also allocates subqueue structs
7597 * for each queue on the device.
7599 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7600 unsigned char name_assign_type,
7601 void (*setup)(struct net_device *),
7602 unsigned int txqs, unsigned int rxqs)
7604 struct net_device *dev;
7606 struct net_device *p;
7608 BUG_ON(strlen(name) >= sizeof(dev->name));
7611 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7617 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7622 alloc_size = sizeof(struct net_device);
7624 /* ensure 32-byte alignment of private area */
7625 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7626 alloc_size += sizeof_priv;
7628 /* ensure 32-byte alignment of whole construct */
7629 alloc_size += NETDEV_ALIGN - 1;
7631 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7633 p = vzalloc(alloc_size);
7637 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7638 dev->padded = (char *)dev - (char *)p;
7640 dev->pcpu_refcnt = alloc_percpu(int);
7641 if (!dev->pcpu_refcnt)
7644 if (dev_addr_init(dev))
7650 dev_net_set(dev, &init_net);
7652 dev->gso_max_size = GSO_MAX_SIZE;
7653 dev->gso_max_segs = GSO_MAX_SEGS;
7655 INIT_LIST_HEAD(&dev->napi_list);
7656 INIT_LIST_HEAD(&dev->unreg_list);
7657 INIT_LIST_HEAD(&dev->close_list);
7658 INIT_LIST_HEAD(&dev->link_watch_list);
7659 INIT_LIST_HEAD(&dev->adj_list.upper);
7660 INIT_LIST_HEAD(&dev->adj_list.lower);
7661 INIT_LIST_HEAD(&dev->all_adj_list.upper);
7662 INIT_LIST_HEAD(&dev->all_adj_list.lower);
7663 INIT_LIST_HEAD(&dev->ptype_all);
7664 INIT_LIST_HEAD(&dev->ptype_specific);
7665 #ifdef CONFIG_NET_SCHED
7666 hash_init(dev->qdisc_hash);
7668 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7671 if (!dev->tx_queue_len) {
7672 dev->priv_flags |= IFF_NO_QUEUE;
7673 dev->tx_queue_len = 1;
7676 dev->num_tx_queues = txqs;
7677 dev->real_num_tx_queues = txqs;
7678 if (netif_alloc_netdev_queues(dev))
7682 dev->num_rx_queues = rxqs;
7683 dev->real_num_rx_queues = rxqs;
7684 if (netif_alloc_rx_queues(dev))
7688 strcpy(dev->name, name);
7689 dev->name_assign_type = name_assign_type;
7690 dev->group = INIT_NETDEV_GROUP;
7691 if (!dev->ethtool_ops)
7692 dev->ethtool_ops = &default_ethtool_ops;
7694 nf_hook_ingress_init(dev);
7703 free_percpu(dev->pcpu_refcnt);
7705 netdev_freemem(dev);
7708 EXPORT_SYMBOL(alloc_netdev_mqs);
7711 * free_netdev - free network device
7714 * This function does the last stage of destroying an allocated device
7715 * interface. The reference to the device object is released.
7716 * If this is the last reference then it will be freed.
7717 * Must be called in process context.
7719 void free_netdev(struct net_device *dev)
7721 struct napi_struct *p, *n;
7724 netif_free_tx_queues(dev);
7729 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7731 /* Flush device addresses */
7732 dev_addr_flush(dev);
7734 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7737 free_percpu(dev->pcpu_refcnt);
7738 dev->pcpu_refcnt = NULL;
7740 /* Compatibility with error handling in drivers */
7741 if (dev->reg_state == NETREG_UNINITIALIZED) {
7742 netdev_freemem(dev);
7746 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7747 dev->reg_state = NETREG_RELEASED;
7749 /* will free via device release */
7750 put_device(&dev->dev);
7752 EXPORT_SYMBOL(free_netdev);
7755 * synchronize_net - Synchronize with packet receive processing
7757 * Wait for packets currently being received to be done.
7758 * Does not block later packets from starting.
7760 void synchronize_net(void)
7763 if (rtnl_is_locked())
7764 synchronize_rcu_expedited();
7768 EXPORT_SYMBOL(synchronize_net);
7771 * unregister_netdevice_queue - remove device from the kernel
7775 * This function shuts down a device interface and removes it
7776 * from the kernel tables.
7777 * If head not NULL, device is queued to be unregistered later.
7779 * Callers must hold the rtnl semaphore. You may want
7780 * unregister_netdev() instead of this.
7783 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7788 list_move_tail(&dev->unreg_list, head);
7790 rollback_registered(dev);
7791 /* Finish processing unregister after unlock */
7795 EXPORT_SYMBOL(unregister_netdevice_queue);
7798 * unregister_netdevice_many - unregister many devices
7799 * @head: list of devices
7801 * Note: As most callers use a stack allocated list_head,
7802 * we force a list_del() to make sure stack wont be corrupted later.
7804 void unregister_netdevice_many(struct list_head *head)
7806 struct net_device *dev;
7808 if (!list_empty(head)) {
7809 rollback_registered_many(head);
7810 list_for_each_entry(dev, head, unreg_list)
7815 EXPORT_SYMBOL(unregister_netdevice_many);
7818 * unregister_netdev - remove device from the kernel
7821 * This function shuts down a device interface and removes it
7822 * from the kernel tables.
7824 * This is just a wrapper for unregister_netdevice that takes
7825 * the rtnl semaphore. In general you want to use this and not
7826 * unregister_netdevice.
7828 void unregister_netdev(struct net_device *dev)
7831 unregister_netdevice(dev);
7834 EXPORT_SYMBOL(unregister_netdev);
7837 * dev_change_net_namespace - move device to different nethost namespace
7839 * @net: network namespace
7840 * @pat: If not NULL name pattern to try if the current device name
7841 * is already taken in the destination network namespace.
7843 * This function shuts down a device interface and moves it
7844 * to a new network namespace. On success 0 is returned, on
7845 * a failure a netagive errno code is returned.
7847 * Callers must hold the rtnl semaphore.
7850 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7856 /* Don't allow namespace local devices to be moved. */
7858 if (dev->features & NETIF_F_NETNS_LOCAL)
7861 /* Ensure the device has been registrered */
7862 if (dev->reg_state != NETREG_REGISTERED)
7865 /* Get out if there is nothing todo */
7867 if (net_eq(dev_net(dev), net))
7870 /* Pick the destination device name, and ensure
7871 * we can use it in the destination network namespace.
7874 if (__dev_get_by_name(net, dev->name)) {
7875 /* We get here if we can't use the current device name */
7878 if (dev_get_valid_name(net, dev, pat) < 0)
7883 * And now a mini version of register_netdevice unregister_netdevice.
7886 /* If device is running close it first. */
7889 /* And unlink it from device chain */
7891 unlist_netdevice(dev);
7895 /* Shutdown queueing discipline. */
7898 /* Notify protocols, that we are about to destroy
7899 this device. They should clean all the things.
7901 Note that dev->reg_state stays at NETREG_REGISTERED.
7902 This is wanted because this way 8021q and macvlan know
7903 the device is just moving and can keep their slaves up.
7905 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7907 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7908 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7911 * Flush the unicast and multicast chains
7916 /* Send a netdev-removed uevent to the old namespace */
7917 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7918 netdev_adjacent_del_links(dev);
7920 /* Actually switch the network namespace */
7921 dev_net_set(dev, net);
7923 /* If there is an ifindex conflict assign a new one */
7924 if (__dev_get_by_index(net, dev->ifindex))
7925 dev->ifindex = dev_new_index(net);
7927 /* Send a netdev-add uevent to the new namespace */
7928 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7929 netdev_adjacent_add_links(dev);
7931 /* Fixup kobjects */
7932 err = device_rename(&dev->dev, dev->name);
7935 /* Add the device back in the hashes */
7936 list_netdevice(dev);
7938 /* Notify protocols, that a new device appeared. */
7939 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7942 * Prevent userspace races by waiting until the network
7943 * device is fully setup before sending notifications.
7945 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7952 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7954 static int dev_cpu_callback(struct notifier_block *nfb,
7955 unsigned long action,
7958 struct sk_buff **list_skb;
7959 struct sk_buff *skb;
7960 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7961 struct softnet_data *sd, *oldsd;
7963 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7966 local_irq_disable();
7967 cpu = smp_processor_id();
7968 sd = &per_cpu(softnet_data, cpu);
7969 oldsd = &per_cpu(softnet_data, oldcpu);
7971 /* Find end of our completion_queue. */
7972 list_skb = &sd->completion_queue;
7974 list_skb = &(*list_skb)->next;
7975 /* Append completion queue from offline CPU. */
7976 *list_skb = oldsd->completion_queue;
7977 oldsd->completion_queue = NULL;
7979 /* Append output queue from offline CPU. */
7980 if (oldsd->output_queue) {
7981 *sd->output_queue_tailp = oldsd->output_queue;
7982 sd->output_queue_tailp = oldsd->output_queue_tailp;
7983 oldsd->output_queue = NULL;
7984 oldsd->output_queue_tailp = &oldsd->output_queue;
7986 /* Append NAPI poll list from offline CPU, with one exception :
7987 * process_backlog() must be called by cpu owning percpu backlog.
7988 * We properly handle process_queue & input_pkt_queue later.
7990 while (!list_empty(&oldsd->poll_list)) {
7991 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7995 list_del_init(&napi->poll_list);
7996 if (napi->poll == process_backlog)
7999 ____napi_schedule(sd, napi);
8002 raise_softirq_irqoff(NET_TX_SOFTIRQ);
8005 /* Process offline CPU's input_pkt_queue */
8006 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
8008 input_queue_head_incr(oldsd);
8010 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
8012 input_queue_head_incr(oldsd);
8020 * netdev_increment_features - increment feature set by one
8021 * @all: current feature set
8022 * @one: new feature set
8023 * @mask: mask feature set
8025 * Computes a new feature set after adding a device with feature set
8026 * @one to the master device with current feature set @all. Will not
8027 * enable anything that is off in @mask. Returns the new feature set.
8029 netdev_features_t netdev_increment_features(netdev_features_t all,
8030 netdev_features_t one, netdev_features_t mask)
8032 if (mask & NETIF_F_HW_CSUM)
8033 mask |= NETIF_F_CSUM_MASK;
8034 mask |= NETIF_F_VLAN_CHALLENGED;
8036 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
8037 all &= one | ~NETIF_F_ALL_FOR_ALL;
8039 /* If one device supports hw checksumming, set for all. */
8040 if (all & NETIF_F_HW_CSUM)
8041 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
8045 EXPORT_SYMBOL(netdev_increment_features);
8047 static struct hlist_head * __net_init netdev_create_hash(void)
8050 struct hlist_head *hash;
8052 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
8054 for (i = 0; i < NETDEV_HASHENTRIES; i++)
8055 INIT_HLIST_HEAD(&hash[i]);
8060 /* Initialize per network namespace state */
8061 static int __net_init netdev_init(struct net *net)
8063 if (net != &init_net)
8064 INIT_LIST_HEAD(&net->dev_base_head);
8066 net->dev_name_head = netdev_create_hash();
8067 if (net->dev_name_head == NULL)
8070 net->dev_index_head = netdev_create_hash();
8071 if (net->dev_index_head == NULL)
8077 kfree(net->dev_name_head);
8083 * netdev_drivername - network driver for the device
8084 * @dev: network device
8086 * Determine network driver for device.
8088 const char *netdev_drivername(const struct net_device *dev)
8090 const struct device_driver *driver;
8091 const struct device *parent;
8092 const char *empty = "";
8094 parent = dev->dev.parent;
8098 driver = parent->driver;
8099 if (driver && driver->name)
8100 return driver->name;
8104 static void __netdev_printk(const char *level, const struct net_device *dev,
8105 struct va_format *vaf)
8107 if (dev && dev->dev.parent) {
8108 dev_printk_emit(level[1] - '0',
8111 dev_driver_string(dev->dev.parent),
8112 dev_name(dev->dev.parent),
8113 netdev_name(dev), netdev_reg_state(dev),
8116 printk("%s%s%s: %pV",
8117 level, netdev_name(dev), netdev_reg_state(dev), vaf);
8119 printk("%s(NULL net_device): %pV", level, vaf);
8123 void netdev_printk(const char *level, const struct net_device *dev,
8124 const char *format, ...)
8126 struct va_format vaf;
8129 va_start(args, format);
8134 __netdev_printk(level, dev, &vaf);
8138 EXPORT_SYMBOL(netdev_printk);
8140 #define define_netdev_printk_level(func, level) \
8141 void func(const struct net_device *dev, const char *fmt, ...) \
8143 struct va_format vaf; \
8146 va_start(args, fmt); \
8151 __netdev_printk(level, dev, &vaf); \
8155 EXPORT_SYMBOL(func);
8157 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
8158 define_netdev_printk_level(netdev_alert, KERN_ALERT);
8159 define_netdev_printk_level(netdev_crit, KERN_CRIT);
8160 define_netdev_printk_level(netdev_err, KERN_ERR);
8161 define_netdev_printk_level(netdev_warn, KERN_WARNING);
8162 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
8163 define_netdev_printk_level(netdev_info, KERN_INFO);
8165 static void __net_exit netdev_exit(struct net *net)
8167 kfree(net->dev_name_head);
8168 kfree(net->dev_index_head);
8171 static struct pernet_operations __net_initdata netdev_net_ops = {
8172 .init = netdev_init,
8173 .exit = netdev_exit,
8176 static void __net_exit default_device_exit(struct net *net)
8178 struct net_device *dev, *aux;
8180 * Push all migratable network devices back to the
8181 * initial network namespace
8184 for_each_netdev_safe(net, dev, aux) {
8186 char fb_name[IFNAMSIZ];
8188 /* Ignore unmoveable devices (i.e. loopback) */
8189 if (dev->features & NETIF_F_NETNS_LOCAL)
8192 /* Leave virtual devices for the generic cleanup */
8193 if (dev->rtnl_link_ops)
8196 /* Push remaining network devices to init_net */
8197 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
8198 err = dev_change_net_namespace(dev, &init_net, fb_name);
8200 pr_emerg("%s: failed to move %s to init_net: %d\n",
8201 __func__, dev->name, err);
8208 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
8210 /* Return with the rtnl_lock held when there are no network
8211 * devices unregistering in any network namespace in net_list.
8215 DEFINE_WAIT_FUNC(wait, woken_wake_function);
8217 add_wait_queue(&netdev_unregistering_wq, &wait);
8219 unregistering = false;
8221 list_for_each_entry(net, net_list, exit_list) {
8222 if (net->dev_unreg_count > 0) {
8223 unregistering = true;
8231 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
8233 remove_wait_queue(&netdev_unregistering_wq, &wait);
8236 static void __net_exit default_device_exit_batch(struct list_head *net_list)
8238 /* At exit all network devices most be removed from a network
8239 * namespace. Do this in the reverse order of registration.
8240 * Do this across as many network namespaces as possible to
8241 * improve batching efficiency.
8243 struct net_device *dev;
8245 LIST_HEAD(dev_kill_list);
8247 /* To prevent network device cleanup code from dereferencing
8248 * loopback devices or network devices that have been freed
8249 * wait here for all pending unregistrations to complete,
8250 * before unregistring the loopback device and allowing the
8251 * network namespace be freed.
8253 * The netdev todo list containing all network devices
8254 * unregistrations that happen in default_device_exit_batch
8255 * will run in the rtnl_unlock() at the end of
8256 * default_device_exit_batch.
8258 rtnl_lock_unregistering(net_list);
8259 list_for_each_entry(net, net_list, exit_list) {
8260 for_each_netdev_reverse(net, dev) {
8261 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
8262 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
8264 unregister_netdevice_queue(dev, &dev_kill_list);
8267 unregister_netdevice_many(&dev_kill_list);
8271 static struct pernet_operations __net_initdata default_device_ops = {
8272 .exit = default_device_exit,
8273 .exit_batch = default_device_exit_batch,
8277 * Initialize the DEV module. At boot time this walks the device list and
8278 * unhooks any devices that fail to initialise (normally hardware not
8279 * present) and leaves us with a valid list of present and active devices.
8284 * This is called single threaded during boot, so no need
8285 * to take the rtnl semaphore.
8287 static int __init net_dev_init(void)
8289 int i, rc = -ENOMEM;
8291 BUG_ON(!dev_boot_phase);
8293 if (dev_proc_init())
8296 if (netdev_kobject_init())
8299 INIT_LIST_HEAD(&ptype_all);
8300 for (i = 0; i < PTYPE_HASH_SIZE; i++)
8301 INIT_LIST_HEAD(&ptype_base[i]);
8303 INIT_LIST_HEAD(&offload_base);
8305 if (register_pernet_subsys(&netdev_net_ops))
8309 * Initialise the packet receive queues.
8312 for_each_possible_cpu(i) {
8313 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
8314 struct softnet_data *sd = &per_cpu(softnet_data, i);
8316 INIT_WORK(flush, flush_backlog);
8318 skb_queue_head_init(&sd->input_pkt_queue);
8319 skb_queue_head_init(&sd->process_queue);
8320 INIT_LIST_HEAD(&sd->poll_list);
8321 sd->output_queue_tailp = &sd->output_queue;
8323 sd->csd.func = rps_trigger_softirq;
8328 sd->backlog.poll = process_backlog;
8329 sd->backlog.weight = weight_p;
8334 /* The loopback device is special if any other network devices
8335 * is present in a network namespace the loopback device must
8336 * be present. Since we now dynamically allocate and free the
8337 * loopback device ensure this invariant is maintained by
8338 * keeping the loopback device as the first device on the
8339 * list of network devices. Ensuring the loopback devices
8340 * is the first device that appears and the last network device
8343 if (register_pernet_device(&loopback_net_ops))
8346 if (register_pernet_device(&default_device_ops))
8349 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
8350 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
8352 hotcpu_notifier(dev_cpu_callback, 0);
8359 subsys_initcall(net_dev_init);