2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <linux/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <linux/bpf.h>
98 #include <net/net_namespace.h>
100 #include <net/busy_poll.h>
101 #include <linux/rtnetlink.h>
102 #include <linux/stat.h>
104 #include <net/dst_metadata.h>
105 #include <net/pkt_sched.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/iw_handler.h>
115 #include <asm/current.h>
116 #include <linux/audit.h>
117 #include <linux/dmaengine.h>
118 #include <linux/err.h>
119 #include <linux/ctype.h>
120 #include <linux/if_arp.h>
121 #include <linux/if_vlan.h>
122 #include <linux/ip.h>
124 #include <net/mpls.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/static_key.h>
136 #include <linux/hashtable.h>
137 #include <linux/vmalloc.h>
138 #include <linux/if_macvlan.h>
139 #include <linux/errqueue.h>
140 #include <linux/hrtimer.h>
141 #include <linux/netfilter_ingress.h>
142 #include <linux/crash_dump.h>
144 #include "net-sysfs.h"
146 /* Instead of increasing this, you should create a hash table. */
147 #define MAX_GRO_SKBS 8
149 /* This should be increased if a protocol with a bigger head is added. */
150 #define GRO_MAX_HEAD (MAX_HEADER + 128)
152 static DEFINE_SPINLOCK(ptype_lock);
153 static DEFINE_SPINLOCK(offload_lock);
154 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
155 struct list_head ptype_all __read_mostly; /* Taps */
156 static struct list_head offload_base __read_mostly;
158 static int netif_rx_internal(struct sk_buff *skb);
159 static int call_netdevice_notifiers_info(unsigned long val,
160 struct net_device *dev,
161 struct netdev_notifier_info *info);
164 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
167 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
169 * Writers must hold the rtnl semaphore while they loop through the
170 * dev_base_head list, and hold dev_base_lock for writing when they do the
171 * actual updates. This allows pure readers to access the list even
172 * while a writer is preparing to update it.
174 * To put it another way, dev_base_lock is held for writing only to
175 * protect against pure readers; the rtnl semaphore provides the
176 * protection against other writers.
178 * See, for example usages, register_netdevice() and
179 * unregister_netdevice(), which must be called with the rtnl
182 DEFINE_RWLOCK(dev_base_lock);
183 EXPORT_SYMBOL(dev_base_lock);
185 /* protects napi_hash addition/deletion and napi_gen_id */
186 static DEFINE_SPINLOCK(napi_hash_lock);
188 static unsigned int napi_gen_id = NR_CPUS;
189 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
191 static seqcount_t devnet_rename_seq;
193 static inline void dev_base_seq_inc(struct net *net)
195 while (++net->dev_base_seq == 0);
198 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
200 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
202 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
205 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
207 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
210 static inline void rps_lock(struct softnet_data *sd)
213 spin_lock(&sd->input_pkt_queue.lock);
217 static inline void rps_unlock(struct softnet_data *sd)
220 spin_unlock(&sd->input_pkt_queue.lock);
224 /* Device list insertion */
225 static void list_netdevice(struct net_device *dev)
227 struct net *net = dev_net(dev);
231 write_lock_bh(&dev_base_lock);
232 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
233 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
234 hlist_add_head_rcu(&dev->index_hlist,
235 dev_index_hash(net, dev->ifindex));
236 write_unlock_bh(&dev_base_lock);
238 dev_base_seq_inc(net);
241 /* Device list removal
242 * caller must respect a RCU grace period before freeing/reusing dev
244 static void unlist_netdevice(struct net_device *dev)
248 /* Unlink dev from the device chain */
249 write_lock_bh(&dev_base_lock);
250 list_del_rcu(&dev->dev_list);
251 hlist_del_rcu(&dev->name_hlist);
252 hlist_del_rcu(&dev->index_hlist);
253 write_unlock_bh(&dev_base_lock);
255 dev_base_seq_inc(dev_net(dev));
262 static RAW_NOTIFIER_HEAD(netdev_chain);
265 * Device drivers call our routines to queue packets here. We empty the
266 * queue in the local softnet handler.
269 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
270 EXPORT_PER_CPU_SYMBOL(softnet_data);
272 #ifdef CONFIG_LOCKDEP
274 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
275 * according to dev->type
277 static const unsigned short netdev_lock_type[] =
278 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
279 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
280 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
281 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
282 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
283 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
284 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
285 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
286 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
287 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
288 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
289 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
290 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
291 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
292 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
294 static const char *const netdev_lock_name[] =
295 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
296 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
297 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
298 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
299 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
300 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
301 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
302 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
303 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
304 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
305 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
306 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
307 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
308 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
309 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
311 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
312 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
318 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
319 if (netdev_lock_type[i] == dev_type)
321 /* the last key is used by default */
322 return ARRAY_SIZE(netdev_lock_type) - 1;
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
326 unsigned short dev_type)
330 i = netdev_lock_pos(dev_type);
331 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
332 netdev_lock_name[i]);
335 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
339 i = netdev_lock_pos(dev->type);
340 lockdep_set_class_and_name(&dev->addr_list_lock,
341 &netdev_addr_lock_key[i],
342 netdev_lock_name[i]);
345 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
346 unsigned short dev_type)
349 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
354 /*******************************************************************************
356 Protocol management and registration routines
358 *******************************************************************************/
361 * Add a protocol ID to the list. Now that the input handler is
362 * smarter we can dispense with all the messy stuff that used to be
365 * BEWARE!!! Protocol handlers, mangling input packets,
366 * MUST BE last in hash buckets and checking protocol handlers
367 * MUST start from promiscuous ptype_all chain in net_bh.
368 * It is true now, do not change it.
369 * Explanation follows: if protocol handler, mangling packet, will
370 * be the first on list, it is not able to sense, that packet
371 * is cloned and should be copied-on-write, so that it will
372 * change it and subsequent readers will get broken packet.
376 static inline struct list_head *ptype_head(const struct packet_type *pt)
378 if (pt->type == htons(ETH_P_ALL))
379 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
381 return pt->dev ? &pt->dev->ptype_specific :
382 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
386 * dev_add_pack - add packet handler
387 * @pt: packet type declaration
389 * Add a protocol handler to the networking stack. The passed &packet_type
390 * is linked into kernel lists and may not be freed until it has been
391 * removed from the kernel lists.
393 * This call does not sleep therefore it can not
394 * guarantee all CPU's that are in middle of receiving packets
395 * will see the new packet type (until the next received packet).
398 void dev_add_pack(struct packet_type *pt)
400 struct list_head *head = ptype_head(pt);
402 spin_lock(&ptype_lock);
403 list_add_rcu(&pt->list, head);
404 spin_unlock(&ptype_lock);
406 EXPORT_SYMBOL(dev_add_pack);
409 * __dev_remove_pack - remove packet handler
410 * @pt: packet type declaration
412 * Remove a protocol handler that was previously added to the kernel
413 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
414 * from the kernel lists and can be freed or reused once this function
417 * The packet type might still be in use by receivers
418 * and must not be freed until after all the CPU's have gone
419 * through a quiescent state.
421 void __dev_remove_pack(struct packet_type *pt)
423 struct list_head *head = ptype_head(pt);
424 struct packet_type *pt1;
426 spin_lock(&ptype_lock);
428 list_for_each_entry(pt1, head, list) {
430 list_del_rcu(&pt->list);
435 pr_warn("dev_remove_pack: %p not found\n", pt);
437 spin_unlock(&ptype_lock);
439 EXPORT_SYMBOL(__dev_remove_pack);
442 * dev_remove_pack - remove packet handler
443 * @pt: packet type declaration
445 * Remove a protocol handler that was previously added to the kernel
446 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
447 * from the kernel lists and can be freed or reused once this function
450 * This call sleeps to guarantee that no CPU is looking at the packet
453 void dev_remove_pack(struct packet_type *pt)
455 __dev_remove_pack(pt);
459 EXPORT_SYMBOL(dev_remove_pack);
463 * dev_add_offload - register offload handlers
464 * @po: protocol offload declaration
466 * Add protocol offload handlers to the networking stack. The passed
467 * &proto_offload is linked into kernel lists and may not be freed until
468 * it has been removed from the kernel lists.
470 * This call does not sleep therefore it can not
471 * guarantee all CPU's that are in middle of receiving packets
472 * will see the new offload handlers (until the next received packet).
474 void dev_add_offload(struct packet_offload *po)
476 struct packet_offload *elem;
478 spin_lock(&offload_lock);
479 list_for_each_entry(elem, &offload_base, list) {
480 if (po->priority < elem->priority)
483 list_add_rcu(&po->list, elem->list.prev);
484 spin_unlock(&offload_lock);
486 EXPORT_SYMBOL(dev_add_offload);
489 * __dev_remove_offload - remove offload handler
490 * @po: packet offload declaration
492 * Remove a protocol offload handler that was previously added to the
493 * kernel offload handlers by dev_add_offload(). The passed &offload_type
494 * is removed from the kernel lists and can be freed or reused once this
497 * The packet type might still be in use by receivers
498 * and must not be freed until after all the CPU's have gone
499 * through a quiescent state.
501 static void __dev_remove_offload(struct packet_offload *po)
503 struct list_head *head = &offload_base;
504 struct packet_offload *po1;
506 spin_lock(&offload_lock);
508 list_for_each_entry(po1, head, list) {
510 list_del_rcu(&po->list);
515 pr_warn("dev_remove_offload: %p not found\n", po);
517 spin_unlock(&offload_lock);
521 * dev_remove_offload - remove packet offload handler
522 * @po: packet offload declaration
524 * Remove a packet offload handler that was previously added to the kernel
525 * offload handlers by dev_add_offload(). The passed &offload_type is
526 * removed from the kernel lists and can be freed or reused once this
529 * This call sleeps to guarantee that no CPU is looking at the packet
532 void dev_remove_offload(struct packet_offload *po)
534 __dev_remove_offload(po);
538 EXPORT_SYMBOL(dev_remove_offload);
540 /******************************************************************************
542 Device Boot-time Settings Routines
544 *******************************************************************************/
546 /* Boot time configuration table */
547 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
550 * netdev_boot_setup_add - add new setup entry
551 * @name: name of the device
552 * @map: configured settings for the device
554 * Adds new setup entry to the dev_boot_setup list. The function
555 * returns 0 on error and 1 on success. This is a generic routine to
558 static int netdev_boot_setup_add(char *name, struct ifmap *map)
560 struct netdev_boot_setup *s;
564 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
565 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
566 memset(s[i].name, 0, sizeof(s[i].name));
567 strlcpy(s[i].name, name, IFNAMSIZ);
568 memcpy(&s[i].map, map, sizeof(s[i].map));
573 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
577 * netdev_boot_setup_check - check boot time settings
578 * @dev: the netdevice
580 * Check boot time settings for the device.
581 * The found settings are set for the device to be used
582 * later in the device probing.
583 * Returns 0 if no settings found, 1 if they are.
585 int netdev_boot_setup_check(struct net_device *dev)
587 struct netdev_boot_setup *s = dev_boot_setup;
590 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
591 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
592 !strcmp(dev->name, s[i].name)) {
593 dev->irq = s[i].map.irq;
594 dev->base_addr = s[i].map.base_addr;
595 dev->mem_start = s[i].map.mem_start;
596 dev->mem_end = s[i].map.mem_end;
602 EXPORT_SYMBOL(netdev_boot_setup_check);
606 * netdev_boot_base - get address from boot time settings
607 * @prefix: prefix for network device
608 * @unit: id for network device
610 * Check boot time settings for the base address of device.
611 * The found settings are set for the device to be used
612 * later in the device probing.
613 * Returns 0 if no settings found.
615 unsigned long netdev_boot_base(const char *prefix, int unit)
617 const struct netdev_boot_setup *s = dev_boot_setup;
621 sprintf(name, "%s%d", prefix, unit);
624 * If device already registered then return base of 1
625 * to indicate not to probe for this interface
627 if (__dev_get_by_name(&init_net, name))
630 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
631 if (!strcmp(name, s[i].name))
632 return s[i].map.base_addr;
637 * Saves at boot time configured settings for any netdevice.
639 int __init netdev_boot_setup(char *str)
644 str = get_options(str, ARRAY_SIZE(ints), ints);
649 memset(&map, 0, sizeof(map));
653 map.base_addr = ints[2];
655 map.mem_start = ints[3];
657 map.mem_end = ints[4];
659 /* Add new entry to the list */
660 return netdev_boot_setup_add(str, &map);
663 __setup("netdev=", netdev_boot_setup);
665 /*******************************************************************************
667 Device Interface Subroutines
669 *******************************************************************************/
672 * dev_get_iflink - get 'iflink' value of a interface
673 * @dev: targeted interface
675 * Indicates the ifindex the interface is linked to.
676 * Physical interfaces have the same 'ifindex' and 'iflink' values.
679 int dev_get_iflink(const struct net_device *dev)
681 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
682 return dev->netdev_ops->ndo_get_iflink(dev);
686 EXPORT_SYMBOL(dev_get_iflink);
689 * dev_fill_metadata_dst - Retrieve tunnel egress information.
690 * @dev: targeted interface
693 * For better visibility of tunnel traffic OVS needs to retrieve
694 * egress tunnel information for a packet. Following API allows
695 * user to get this info.
697 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
699 struct ip_tunnel_info *info;
701 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
704 info = skb_tunnel_info_unclone(skb);
707 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
710 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
712 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
715 * __dev_get_by_name - find a device by its name
716 * @net: the applicable net namespace
717 * @name: name to find
719 * Find an interface by name. Must be called under RTNL semaphore
720 * or @dev_base_lock. If the name is found a pointer to the device
721 * is returned. If the name is not found then %NULL is returned. The
722 * reference counters are not incremented so the caller must be
723 * careful with locks.
726 struct net_device *__dev_get_by_name(struct net *net, const char *name)
728 struct net_device *dev;
729 struct hlist_head *head = dev_name_hash(net, name);
731 hlist_for_each_entry(dev, head, name_hlist)
732 if (!strncmp(dev->name, name, IFNAMSIZ))
737 EXPORT_SYMBOL(__dev_get_by_name);
740 * dev_get_by_name_rcu - find a device by its name
741 * @net: the applicable net namespace
742 * @name: name to find
744 * Find an interface by name.
745 * If the name is found a pointer to the device is returned.
746 * If the name is not found then %NULL is returned.
747 * The reference counters are not incremented so the caller must be
748 * careful with locks. The caller must hold RCU lock.
751 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
753 struct net_device *dev;
754 struct hlist_head *head = dev_name_hash(net, name);
756 hlist_for_each_entry_rcu(dev, head, name_hlist)
757 if (!strncmp(dev->name, name, IFNAMSIZ))
762 EXPORT_SYMBOL(dev_get_by_name_rcu);
765 * dev_get_by_name - find a device by its name
766 * @net: the applicable net namespace
767 * @name: name to find
769 * Find an interface by name. This can be called from any
770 * context and does its own locking. The returned handle has
771 * the usage count incremented and the caller must use dev_put() to
772 * release it when it is no longer needed. %NULL is returned if no
773 * matching device is found.
776 struct net_device *dev_get_by_name(struct net *net, const char *name)
778 struct net_device *dev;
781 dev = dev_get_by_name_rcu(net, name);
787 EXPORT_SYMBOL(dev_get_by_name);
790 * __dev_get_by_index - find a device by its ifindex
791 * @net: the applicable net namespace
792 * @ifindex: index of device
794 * Search for an interface by index. Returns %NULL if the device
795 * is not found or a pointer to the device. The device has not
796 * had its reference counter increased so the caller must be careful
797 * about locking. The caller must hold either the RTNL semaphore
801 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
803 struct net_device *dev;
804 struct hlist_head *head = dev_index_hash(net, ifindex);
806 hlist_for_each_entry(dev, head, index_hlist)
807 if (dev->ifindex == ifindex)
812 EXPORT_SYMBOL(__dev_get_by_index);
815 * dev_get_by_index_rcu - find a device by its ifindex
816 * @net: the applicable net namespace
817 * @ifindex: index of device
819 * Search for an interface by index. Returns %NULL if the device
820 * is not found or a pointer to the device. The device has not
821 * had its reference counter increased so the caller must be careful
822 * about locking. The caller must hold RCU lock.
825 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
827 struct net_device *dev;
828 struct hlist_head *head = dev_index_hash(net, ifindex);
830 hlist_for_each_entry_rcu(dev, head, index_hlist)
831 if (dev->ifindex == ifindex)
836 EXPORT_SYMBOL(dev_get_by_index_rcu);
840 * dev_get_by_index - find a device by its ifindex
841 * @net: the applicable net namespace
842 * @ifindex: index of device
844 * Search for an interface by index. Returns NULL if the device
845 * is not found or a pointer to the device. The device returned has
846 * had a reference added and the pointer is safe until the user calls
847 * dev_put to indicate they have finished with it.
850 struct net_device *dev_get_by_index(struct net *net, int ifindex)
852 struct net_device *dev;
855 dev = dev_get_by_index_rcu(net, ifindex);
861 EXPORT_SYMBOL(dev_get_by_index);
864 * netdev_get_name - get a netdevice name, knowing its ifindex.
865 * @net: network namespace
866 * @name: a pointer to the buffer where the name will be stored.
867 * @ifindex: the ifindex of the interface to get the name from.
869 * The use of raw_seqcount_begin() and cond_resched() before
870 * retrying is required as we want to give the writers a chance
871 * to complete when CONFIG_PREEMPT is not set.
873 int netdev_get_name(struct net *net, char *name, int ifindex)
875 struct net_device *dev;
879 seq = raw_seqcount_begin(&devnet_rename_seq);
881 dev = dev_get_by_index_rcu(net, ifindex);
887 strcpy(name, dev->name);
889 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
898 * dev_getbyhwaddr_rcu - find a device by its hardware address
899 * @net: the applicable net namespace
900 * @type: media type of device
901 * @ha: hardware address
903 * Search for an interface by MAC address. Returns NULL if the device
904 * is not found or a pointer to the device.
905 * The caller must hold RCU or RTNL.
906 * The returned device has not had its ref count increased
907 * and the caller must therefore be careful about locking
911 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
914 struct net_device *dev;
916 for_each_netdev_rcu(net, dev)
917 if (dev->type == type &&
918 !memcmp(dev->dev_addr, ha, dev->addr_len))
923 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
925 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
927 struct net_device *dev;
930 for_each_netdev(net, dev)
931 if (dev->type == type)
936 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
938 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
940 struct net_device *dev, *ret = NULL;
943 for_each_netdev_rcu(net, dev)
944 if (dev->type == type) {
952 EXPORT_SYMBOL(dev_getfirstbyhwtype);
955 * __dev_get_by_flags - find any device with given flags
956 * @net: the applicable net namespace
957 * @if_flags: IFF_* values
958 * @mask: bitmask of bits in if_flags to check
960 * Search for any interface with the given flags. Returns NULL if a device
961 * is not found or a pointer to the device. Must be called inside
962 * rtnl_lock(), and result refcount is unchanged.
965 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
968 struct net_device *dev, *ret;
973 for_each_netdev(net, dev) {
974 if (((dev->flags ^ if_flags) & mask) == 0) {
981 EXPORT_SYMBOL(__dev_get_by_flags);
984 * dev_valid_name - check if name is okay for network device
987 * Network device names need to be valid file names to
988 * to allow sysfs to work. We also disallow any kind of
991 bool dev_valid_name(const char *name)
995 if (strlen(name) >= IFNAMSIZ)
997 if (!strcmp(name, ".") || !strcmp(name, ".."))
1001 if (*name == '/' || *name == ':' || isspace(*name))
1007 EXPORT_SYMBOL(dev_valid_name);
1010 * __dev_alloc_name - allocate a name for a device
1011 * @net: network namespace to allocate the device name in
1012 * @name: name format string
1013 * @buf: scratch buffer and result name string
1015 * Passed a format string - eg "lt%d" it will try and find a suitable
1016 * id. It scans list of devices to build up a free map, then chooses
1017 * the first empty slot. The caller must hold the dev_base or rtnl lock
1018 * while allocating the name and adding the device in order to avoid
1020 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1021 * Returns the number of the unit assigned or a negative errno code.
1024 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1028 const int max_netdevices = 8*PAGE_SIZE;
1029 unsigned long *inuse;
1030 struct net_device *d;
1032 p = strnchr(name, IFNAMSIZ-1, '%');
1035 * Verify the string as this thing may have come from
1036 * the user. There must be either one "%d" and no other "%"
1039 if (p[1] != 'd' || strchr(p + 2, '%'))
1042 /* Use one page as a bit array of possible slots */
1043 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1047 for_each_netdev(net, d) {
1048 if (!sscanf(d->name, name, &i))
1050 if (i < 0 || i >= max_netdevices)
1053 /* avoid cases where sscanf is not exact inverse of printf */
1054 snprintf(buf, IFNAMSIZ, name, i);
1055 if (!strncmp(buf, d->name, IFNAMSIZ))
1059 i = find_first_zero_bit(inuse, max_netdevices);
1060 free_page((unsigned long) inuse);
1064 snprintf(buf, IFNAMSIZ, name, i);
1065 if (!__dev_get_by_name(net, buf))
1068 /* It is possible to run out of possible slots
1069 * when the name is long and there isn't enough space left
1070 * for the digits, or if all bits are used.
1076 * dev_alloc_name - allocate a name for a device
1078 * @name: name format string
1080 * Passed a format string - eg "lt%d" it will try and find a suitable
1081 * id. It scans list of devices to build up a free map, then chooses
1082 * the first empty slot. The caller must hold the dev_base or rtnl lock
1083 * while allocating the name and adding the device in order to avoid
1085 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1086 * Returns the number of the unit assigned or a negative errno code.
1089 int dev_alloc_name(struct net_device *dev, const char *name)
1095 BUG_ON(!dev_net(dev));
1097 ret = __dev_alloc_name(net, name, buf);
1099 strlcpy(dev->name, buf, IFNAMSIZ);
1102 EXPORT_SYMBOL(dev_alloc_name);
1104 static int dev_alloc_name_ns(struct net *net,
1105 struct net_device *dev,
1111 ret = __dev_alloc_name(net, name, buf);
1113 strlcpy(dev->name, buf, IFNAMSIZ);
1117 static int dev_get_valid_name(struct net *net,
1118 struct net_device *dev,
1123 if (!dev_valid_name(name))
1126 if (strchr(name, '%'))
1127 return dev_alloc_name_ns(net, dev, name);
1128 else if (__dev_get_by_name(net, name))
1130 else if (dev->name != name)
1131 strlcpy(dev->name, name, IFNAMSIZ);
1137 * dev_change_name - change name of a device
1139 * @newname: name (or format string) must be at least IFNAMSIZ
1141 * Change name of a device, can pass format strings "eth%d".
1144 int dev_change_name(struct net_device *dev, const char *newname)
1146 unsigned char old_assign_type;
1147 char oldname[IFNAMSIZ];
1153 BUG_ON(!dev_net(dev));
1156 if (dev->flags & IFF_UP)
1159 write_seqcount_begin(&devnet_rename_seq);
1161 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1162 write_seqcount_end(&devnet_rename_seq);
1166 memcpy(oldname, dev->name, IFNAMSIZ);
1168 err = dev_get_valid_name(net, dev, newname);
1170 write_seqcount_end(&devnet_rename_seq);
1174 if (oldname[0] && !strchr(oldname, '%'))
1175 netdev_info(dev, "renamed from %s\n", oldname);
1177 old_assign_type = dev->name_assign_type;
1178 dev->name_assign_type = NET_NAME_RENAMED;
1181 ret = device_rename(&dev->dev, dev->name);
1183 memcpy(dev->name, oldname, IFNAMSIZ);
1184 dev->name_assign_type = old_assign_type;
1185 write_seqcount_end(&devnet_rename_seq);
1189 write_seqcount_end(&devnet_rename_seq);
1191 netdev_adjacent_rename_links(dev, oldname);
1193 write_lock_bh(&dev_base_lock);
1194 hlist_del_rcu(&dev->name_hlist);
1195 write_unlock_bh(&dev_base_lock);
1199 write_lock_bh(&dev_base_lock);
1200 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1201 write_unlock_bh(&dev_base_lock);
1203 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1204 ret = notifier_to_errno(ret);
1207 /* err >= 0 after dev_alloc_name() or stores the first errno */
1210 write_seqcount_begin(&devnet_rename_seq);
1211 memcpy(dev->name, oldname, IFNAMSIZ);
1212 memcpy(oldname, newname, IFNAMSIZ);
1213 dev->name_assign_type = old_assign_type;
1214 old_assign_type = NET_NAME_RENAMED;
1217 pr_err("%s: name change rollback failed: %d\n",
1226 * dev_set_alias - change ifalias of a device
1228 * @alias: name up to IFALIASZ
1229 * @len: limit of bytes to copy from info
1231 * Set ifalias for a device,
1233 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1239 if (len >= IFALIASZ)
1243 kfree(dev->ifalias);
1244 dev->ifalias = NULL;
1248 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1251 dev->ifalias = new_ifalias;
1253 strlcpy(dev->ifalias, alias, len+1);
1259 * netdev_features_change - device changes features
1260 * @dev: device to cause notification
1262 * Called to indicate a device has changed features.
1264 void netdev_features_change(struct net_device *dev)
1266 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1268 EXPORT_SYMBOL(netdev_features_change);
1271 * netdev_state_change - device changes state
1272 * @dev: device to cause notification
1274 * Called to indicate a device has changed state. This function calls
1275 * the notifier chains for netdev_chain and sends a NEWLINK message
1276 * to the routing socket.
1278 void netdev_state_change(struct net_device *dev)
1280 if (dev->flags & IFF_UP) {
1281 struct netdev_notifier_change_info change_info;
1283 change_info.flags_changed = 0;
1284 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1286 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1289 EXPORT_SYMBOL(netdev_state_change);
1292 * netdev_notify_peers - notify network peers about existence of @dev
1293 * @dev: network device
1295 * Generate traffic such that interested network peers are aware of
1296 * @dev, such as by generating a gratuitous ARP. This may be used when
1297 * a device wants to inform the rest of the network about some sort of
1298 * reconfiguration such as a failover event or virtual machine
1301 void netdev_notify_peers(struct net_device *dev)
1304 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1307 EXPORT_SYMBOL(netdev_notify_peers);
1309 static int __dev_open(struct net_device *dev)
1311 const struct net_device_ops *ops = dev->netdev_ops;
1316 if (!netif_device_present(dev))
1319 /* Block netpoll from trying to do any rx path servicing.
1320 * If we don't do this there is a chance ndo_poll_controller
1321 * or ndo_poll may be running while we open the device
1323 netpoll_poll_disable(dev);
1325 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1326 ret = notifier_to_errno(ret);
1330 set_bit(__LINK_STATE_START, &dev->state);
1332 if (ops->ndo_validate_addr)
1333 ret = ops->ndo_validate_addr(dev);
1335 if (!ret && ops->ndo_open)
1336 ret = ops->ndo_open(dev);
1338 netpoll_poll_enable(dev);
1341 clear_bit(__LINK_STATE_START, &dev->state);
1343 dev->flags |= IFF_UP;
1344 dev_set_rx_mode(dev);
1346 add_device_randomness(dev->dev_addr, dev->addr_len);
1353 * dev_open - prepare an interface for use.
1354 * @dev: device to open
1356 * Takes a device from down to up state. The device's private open
1357 * function is invoked and then the multicast lists are loaded. Finally
1358 * the device is moved into the up state and a %NETDEV_UP message is
1359 * sent to the netdev notifier chain.
1361 * Calling this function on an active interface is a nop. On a failure
1362 * a negative errno code is returned.
1364 int dev_open(struct net_device *dev)
1368 if (dev->flags & IFF_UP)
1371 ret = __dev_open(dev);
1375 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1376 call_netdevice_notifiers(NETDEV_UP, dev);
1380 EXPORT_SYMBOL(dev_open);
1382 static int __dev_close_many(struct list_head *head)
1384 struct net_device *dev;
1389 list_for_each_entry(dev, head, close_list) {
1390 /* Temporarily disable netpoll until the interface is down */
1391 netpoll_poll_disable(dev);
1393 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1395 clear_bit(__LINK_STATE_START, &dev->state);
1397 /* Synchronize to scheduled poll. We cannot touch poll list, it
1398 * can be even on different cpu. So just clear netif_running().
1400 * dev->stop() will invoke napi_disable() on all of it's
1401 * napi_struct instances on this device.
1403 smp_mb__after_atomic(); /* Commit netif_running(). */
1406 dev_deactivate_many(head);
1408 list_for_each_entry(dev, head, close_list) {
1409 const struct net_device_ops *ops = dev->netdev_ops;
1412 * Call the device specific close. This cannot fail.
1413 * Only if device is UP
1415 * We allow it to be called even after a DETACH hot-plug
1421 dev->flags &= ~IFF_UP;
1422 netpoll_poll_enable(dev);
1428 static int __dev_close(struct net_device *dev)
1433 list_add(&dev->close_list, &single);
1434 retval = __dev_close_many(&single);
1440 int dev_close_many(struct list_head *head, bool unlink)
1442 struct net_device *dev, *tmp;
1444 /* Remove the devices that don't need to be closed */
1445 list_for_each_entry_safe(dev, tmp, head, close_list)
1446 if (!(dev->flags & IFF_UP))
1447 list_del_init(&dev->close_list);
1449 __dev_close_many(head);
1451 list_for_each_entry_safe(dev, tmp, head, close_list) {
1452 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1453 call_netdevice_notifiers(NETDEV_DOWN, dev);
1455 list_del_init(&dev->close_list);
1460 EXPORT_SYMBOL(dev_close_many);
1463 * dev_close - shutdown an interface.
1464 * @dev: device to shutdown
1466 * This function moves an active device into down state. A
1467 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1468 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1471 int dev_close(struct net_device *dev)
1473 if (dev->flags & IFF_UP) {
1476 list_add(&dev->close_list, &single);
1477 dev_close_many(&single, true);
1482 EXPORT_SYMBOL(dev_close);
1486 * dev_disable_lro - disable Large Receive Offload on a device
1489 * Disable Large Receive Offload (LRO) on a net device. Must be
1490 * called under RTNL. This is needed if received packets may be
1491 * forwarded to another interface.
1493 void dev_disable_lro(struct net_device *dev)
1495 struct net_device *lower_dev;
1496 struct list_head *iter;
1498 dev->wanted_features &= ~NETIF_F_LRO;
1499 netdev_update_features(dev);
1501 if (unlikely(dev->features & NETIF_F_LRO))
1502 netdev_WARN(dev, "failed to disable LRO!\n");
1504 netdev_for_each_lower_dev(dev, lower_dev, iter)
1505 dev_disable_lro(lower_dev);
1507 EXPORT_SYMBOL(dev_disable_lro);
1509 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1510 struct net_device *dev)
1512 struct netdev_notifier_info info;
1514 netdev_notifier_info_init(&info, dev);
1515 return nb->notifier_call(nb, val, &info);
1518 static int dev_boot_phase = 1;
1521 * register_netdevice_notifier - register a network notifier block
1524 * Register a notifier to be called when network device events occur.
1525 * The notifier passed is linked into the kernel structures and must
1526 * not be reused until it has been unregistered. A negative errno code
1527 * is returned on a failure.
1529 * When registered all registration and up events are replayed
1530 * to the new notifier to allow device to have a race free
1531 * view of the network device list.
1534 int register_netdevice_notifier(struct notifier_block *nb)
1536 struct net_device *dev;
1537 struct net_device *last;
1542 err = raw_notifier_chain_register(&netdev_chain, nb);
1548 for_each_netdev(net, dev) {
1549 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1550 err = notifier_to_errno(err);
1554 if (!(dev->flags & IFF_UP))
1557 call_netdevice_notifier(nb, NETDEV_UP, dev);
1568 for_each_netdev(net, dev) {
1572 if (dev->flags & IFF_UP) {
1573 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1575 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1577 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1582 raw_notifier_chain_unregister(&netdev_chain, nb);
1585 EXPORT_SYMBOL(register_netdevice_notifier);
1588 * unregister_netdevice_notifier - unregister a network notifier block
1591 * Unregister a notifier previously registered by
1592 * register_netdevice_notifier(). The notifier is unlinked into the
1593 * kernel structures and may then be reused. A negative errno code
1594 * is returned on a failure.
1596 * After unregistering unregister and down device events are synthesized
1597 * for all devices on the device list to the removed notifier to remove
1598 * the need for special case cleanup code.
1601 int unregister_netdevice_notifier(struct notifier_block *nb)
1603 struct net_device *dev;
1608 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1613 for_each_netdev(net, dev) {
1614 if (dev->flags & IFF_UP) {
1615 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1617 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1619 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1626 EXPORT_SYMBOL(unregister_netdevice_notifier);
1629 * call_netdevice_notifiers_info - call all network notifier blocks
1630 * @val: value passed unmodified to notifier function
1631 * @dev: net_device pointer passed unmodified to notifier function
1632 * @info: notifier information data
1634 * Call all network notifier blocks. Parameters and return value
1635 * are as for raw_notifier_call_chain().
1638 static int call_netdevice_notifiers_info(unsigned long val,
1639 struct net_device *dev,
1640 struct netdev_notifier_info *info)
1643 netdev_notifier_info_init(info, dev);
1644 return raw_notifier_call_chain(&netdev_chain, val, info);
1648 * call_netdevice_notifiers - call all network notifier blocks
1649 * @val: value passed unmodified to notifier function
1650 * @dev: net_device pointer passed unmodified to notifier function
1652 * Call all network notifier blocks. Parameters and return value
1653 * are as for raw_notifier_call_chain().
1656 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1658 struct netdev_notifier_info info;
1660 return call_netdevice_notifiers_info(val, dev, &info);
1662 EXPORT_SYMBOL(call_netdevice_notifiers);
1664 #ifdef CONFIG_NET_INGRESS
1665 static struct static_key ingress_needed __read_mostly;
1667 void net_inc_ingress_queue(void)
1669 static_key_slow_inc(&ingress_needed);
1671 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1673 void net_dec_ingress_queue(void)
1675 static_key_slow_dec(&ingress_needed);
1677 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1680 #ifdef CONFIG_NET_EGRESS
1681 static struct static_key egress_needed __read_mostly;
1683 void net_inc_egress_queue(void)
1685 static_key_slow_inc(&egress_needed);
1687 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1689 void net_dec_egress_queue(void)
1691 static_key_slow_dec(&egress_needed);
1693 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1696 static struct static_key netstamp_needed __read_mostly;
1697 #ifdef HAVE_JUMP_LABEL
1698 static atomic_t netstamp_needed_deferred;
1699 static void netstamp_clear(struct work_struct *work)
1701 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1704 static_key_slow_dec(&netstamp_needed);
1706 static DECLARE_WORK(netstamp_work, netstamp_clear);
1709 void net_enable_timestamp(void)
1711 static_key_slow_inc(&netstamp_needed);
1713 EXPORT_SYMBOL(net_enable_timestamp);
1715 void net_disable_timestamp(void)
1717 #ifdef HAVE_JUMP_LABEL
1718 /* net_disable_timestamp() can be called from non process context */
1719 atomic_inc(&netstamp_needed_deferred);
1720 schedule_work(&netstamp_work);
1722 static_key_slow_dec(&netstamp_needed);
1725 EXPORT_SYMBOL(net_disable_timestamp);
1727 static inline void net_timestamp_set(struct sk_buff *skb)
1730 if (static_key_false(&netstamp_needed))
1731 __net_timestamp(skb);
1734 #define net_timestamp_check(COND, SKB) \
1735 if (static_key_false(&netstamp_needed)) { \
1736 if ((COND) && !(SKB)->tstamp) \
1737 __net_timestamp(SKB); \
1740 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1744 if (!(dev->flags & IFF_UP))
1747 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1748 if (skb->len <= len)
1751 /* if TSO is enabled, we don't care about the length as the packet
1752 * could be forwarded without being segmented before
1754 if (skb_is_gso(skb))
1759 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1761 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1763 int ret = ____dev_forward_skb(dev, skb);
1766 skb->protocol = eth_type_trans(skb, dev);
1767 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1772 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1775 * dev_forward_skb - loopback an skb to another netif
1777 * @dev: destination network device
1778 * @skb: buffer to forward
1781 * NET_RX_SUCCESS (no congestion)
1782 * NET_RX_DROP (packet was dropped, but freed)
1784 * dev_forward_skb can be used for injecting an skb from the
1785 * start_xmit function of one device into the receive queue
1786 * of another device.
1788 * The receiving device may be in another namespace, so
1789 * we have to clear all information in the skb that could
1790 * impact namespace isolation.
1792 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1794 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1796 EXPORT_SYMBOL_GPL(dev_forward_skb);
1798 static inline int deliver_skb(struct sk_buff *skb,
1799 struct packet_type *pt_prev,
1800 struct net_device *orig_dev)
1802 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1804 atomic_inc(&skb->users);
1805 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1808 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1809 struct packet_type **pt,
1810 struct net_device *orig_dev,
1812 struct list_head *ptype_list)
1814 struct packet_type *ptype, *pt_prev = *pt;
1816 list_for_each_entry_rcu(ptype, ptype_list, list) {
1817 if (ptype->type != type)
1820 deliver_skb(skb, pt_prev, orig_dev);
1826 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1828 if (!ptype->af_packet_priv || !skb->sk)
1831 if (ptype->id_match)
1832 return ptype->id_match(ptype, skb->sk);
1833 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1840 * Support routine. Sends outgoing frames to any network
1841 * taps currently in use.
1844 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1846 struct packet_type *ptype;
1847 struct sk_buff *skb2 = NULL;
1848 struct packet_type *pt_prev = NULL;
1849 struct list_head *ptype_list = &ptype_all;
1853 list_for_each_entry_rcu(ptype, ptype_list, list) {
1854 /* Never send packets back to the socket
1855 * they originated from - MvS (miquels@drinkel.ow.org)
1857 if (skb_loop_sk(ptype, skb))
1861 deliver_skb(skb2, pt_prev, skb->dev);
1866 /* need to clone skb, done only once */
1867 skb2 = skb_clone(skb, GFP_ATOMIC);
1871 net_timestamp_set(skb2);
1873 /* skb->nh should be correctly
1874 * set by sender, so that the second statement is
1875 * just protection against buggy protocols.
1877 skb_reset_mac_header(skb2);
1879 if (skb_network_header(skb2) < skb2->data ||
1880 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1881 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1882 ntohs(skb2->protocol),
1884 skb_reset_network_header(skb2);
1887 skb2->transport_header = skb2->network_header;
1888 skb2->pkt_type = PACKET_OUTGOING;
1892 if (ptype_list == &ptype_all) {
1893 ptype_list = &dev->ptype_all;
1898 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1901 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
1904 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1905 * @dev: Network device
1906 * @txq: number of queues available
1908 * If real_num_tx_queues is changed the tc mappings may no longer be
1909 * valid. To resolve this verify the tc mapping remains valid and if
1910 * not NULL the mapping. With no priorities mapping to this
1911 * offset/count pair it will no longer be used. In the worst case TC0
1912 * is invalid nothing can be done so disable priority mappings. If is
1913 * expected that drivers will fix this mapping if they can before
1914 * calling netif_set_real_num_tx_queues.
1916 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1919 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1921 /* If TC0 is invalidated disable TC mapping */
1922 if (tc->offset + tc->count > txq) {
1923 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1928 /* Invalidated prio to tc mappings set to TC0 */
1929 for (i = 1; i < TC_BITMASK + 1; i++) {
1930 int q = netdev_get_prio_tc_map(dev, i);
1932 tc = &dev->tc_to_txq[q];
1933 if (tc->offset + tc->count > txq) {
1934 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1936 netdev_set_prio_tc_map(dev, i, 0);
1941 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
1944 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1947 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
1948 if ((txq - tc->offset) < tc->count)
1959 static DEFINE_MUTEX(xps_map_mutex);
1960 #define xmap_dereference(P) \
1961 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1963 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
1966 struct xps_map *map = NULL;
1970 map = xmap_dereference(dev_maps->cpu_map[tci]);
1974 for (pos = map->len; pos--;) {
1975 if (map->queues[pos] != index)
1979 map->queues[pos] = map->queues[--map->len];
1983 RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL);
1984 kfree_rcu(map, rcu);
1991 static bool remove_xps_queue_cpu(struct net_device *dev,
1992 struct xps_dev_maps *dev_maps,
1993 int cpu, u16 offset, u16 count)
1995 int num_tc = dev->num_tc ? : 1;
1996 bool active = false;
1999 for (tci = cpu * num_tc; num_tc--; tci++) {
2002 for (i = count, j = offset; i--; j++) {
2003 if (!remove_xps_queue(dev_maps, cpu, j))
2013 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2016 struct xps_dev_maps *dev_maps;
2018 bool active = false;
2020 mutex_lock(&xps_map_mutex);
2021 dev_maps = xmap_dereference(dev->xps_maps);
2026 for_each_possible_cpu(cpu)
2027 active |= remove_xps_queue_cpu(dev, dev_maps, cpu,
2031 RCU_INIT_POINTER(dev->xps_maps, NULL);
2032 kfree_rcu(dev_maps, rcu);
2035 for (i = offset + (count - 1); count--; i--)
2036 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2040 mutex_unlock(&xps_map_mutex);
2043 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2045 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2048 static struct xps_map *expand_xps_map(struct xps_map *map,
2051 struct xps_map *new_map;
2052 int alloc_len = XPS_MIN_MAP_ALLOC;
2055 for (pos = 0; map && pos < map->len; pos++) {
2056 if (map->queues[pos] != index)
2061 /* Need to add queue to this CPU's existing map */
2063 if (pos < map->alloc_len)
2066 alloc_len = map->alloc_len * 2;
2069 /* Need to allocate new map to store queue on this CPU's map */
2070 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2075 for (i = 0; i < pos; i++)
2076 new_map->queues[i] = map->queues[i];
2077 new_map->alloc_len = alloc_len;
2083 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2086 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2087 int i, cpu, tci, numa_node_id = -2;
2088 int maps_sz, num_tc = 1, tc = 0;
2089 struct xps_map *map, *new_map;
2090 bool active = false;
2093 num_tc = dev->num_tc;
2094 tc = netdev_txq_to_tc(dev, index);
2099 maps_sz = XPS_DEV_MAPS_SIZE(num_tc);
2100 if (maps_sz < L1_CACHE_BYTES)
2101 maps_sz = L1_CACHE_BYTES;
2103 mutex_lock(&xps_map_mutex);
2105 dev_maps = xmap_dereference(dev->xps_maps);
2107 /* allocate memory for queue storage */
2108 for_each_cpu_and(cpu, cpu_online_mask, mask) {
2110 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2111 if (!new_dev_maps) {
2112 mutex_unlock(&xps_map_mutex);
2116 tci = cpu * num_tc + tc;
2117 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) :
2120 map = expand_xps_map(map, cpu, index);
2124 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2128 goto out_no_new_maps;
2130 for_each_possible_cpu(cpu) {
2131 /* copy maps belonging to foreign traffic classes */
2132 for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) {
2133 /* fill in the new device map from the old device map */
2134 map = xmap_dereference(dev_maps->cpu_map[tci]);
2135 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2138 /* We need to explicitly update tci as prevous loop
2139 * could break out early if dev_maps is NULL.
2141 tci = cpu * num_tc + tc;
2143 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2144 /* add queue to CPU maps */
2147 map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2148 while ((pos < map->len) && (map->queues[pos] != index))
2151 if (pos == map->len)
2152 map->queues[map->len++] = index;
2154 if (numa_node_id == -2)
2155 numa_node_id = cpu_to_node(cpu);
2156 else if (numa_node_id != cpu_to_node(cpu))
2159 } else if (dev_maps) {
2160 /* fill in the new device map from the old device map */
2161 map = xmap_dereference(dev_maps->cpu_map[tci]);
2162 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2165 /* copy maps belonging to foreign traffic classes */
2166 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2167 /* fill in the new device map from the old device map */
2168 map = xmap_dereference(dev_maps->cpu_map[tci]);
2169 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2173 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2175 /* Cleanup old maps */
2177 goto out_no_old_maps;
2179 for_each_possible_cpu(cpu) {
2180 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2181 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2182 map = xmap_dereference(dev_maps->cpu_map[tci]);
2183 if (map && map != new_map)
2184 kfree_rcu(map, rcu);
2188 kfree_rcu(dev_maps, rcu);
2191 dev_maps = new_dev_maps;
2195 /* update Tx queue numa node */
2196 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2197 (numa_node_id >= 0) ? numa_node_id :
2203 /* removes queue from unused CPUs */
2204 for_each_possible_cpu(cpu) {
2205 for (i = tc, tci = cpu * num_tc; i--; tci++)
2206 active |= remove_xps_queue(dev_maps, tci, index);
2207 if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu))
2208 active |= remove_xps_queue(dev_maps, tci, index);
2209 for (i = num_tc - tc, tci++; --i; tci++)
2210 active |= remove_xps_queue(dev_maps, tci, index);
2213 /* free map if not active */
2215 RCU_INIT_POINTER(dev->xps_maps, NULL);
2216 kfree_rcu(dev_maps, rcu);
2220 mutex_unlock(&xps_map_mutex);
2224 /* remove any maps that we added */
2225 for_each_possible_cpu(cpu) {
2226 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2227 new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2229 xmap_dereference(dev_maps->cpu_map[tci]) :
2231 if (new_map && new_map != map)
2236 mutex_unlock(&xps_map_mutex);
2238 kfree(new_dev_maps);
2241 EXPORT_SYMBOL(netif_set_xps_queue);
2244 void netdev_reset_tc(struct net_device *dev)
2247 netif_reset_xps_queues_gt(dev, 0);
2250 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2251 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2253 EXPORT_SYMBOL(netdev_reset_tc);
2255 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2257 if (tc >= dev->num_tc)
2261 netif_reset_xps_queues(dev, offset, count);
2263 dev->tc_to_txq[tc].count = count;
2264 dev->tc_to_txq[tc].offset = offset;
2267 EXPORT_SYMBOL(netdev_set_tc_queue);
2269 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2271 if (num_tc > TC_MAX_QUEUE)
2275 netif_reset_xps_queues_gt(dev, 0);
2277 dev->num_tc = num_tc;
2280 EXPORT_SYMBOL(netdev_set_num_tc);
2283 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2284 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2286 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2290 if (txq < 1 || txq > dev->num_tx_queues)
2293 if (dev->reg_state == NETREG_REGISTERED ||
2294 dev->reg_state == NETREG_UNREGISTERING) {
2297 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2303 netif_setup_tc(dev, txq);
2305 if (txq < dev->real_num_tx_queues) {
2306 qdisc_reset_all_tx_gt(dev, txq);
2308 netif_reset_xps_queues_gt(dev, txq);
2313 dev->real_num_tx_queues = txq;
2316 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2320 * netif_set_real_num_rx_queues - set actual number of RX queues used
2321 * @dev: Network device
2322 * @rxq: Actual number of RX queues
2324 * This must be called either with the rtnl_lock held or before
2325 * registration of the net device. Returns 0 on success, or a
2326 * negative error code. If called before registration, it always
2329 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2333 if (rxq < 1 || rxq > dev->num_rx_queues)
2336 if (dev->reg_state == NETREG_REGISTERED) {
2339 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2345 dev->real_num_rx_queues = rxq;
2348 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2352 * netif_get_num_default_rss_queues - default number of RSS queues
2354 * This routine should set an upper limit on the number of RSS queues
2355 * used by default by multiqueue devices.
2357 int netif_get_num_default_rss_queues(void)
2359 return is_kdump_kernel() ?
2360 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2362 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2364 static void __netif_reschedule(struct Qdisc *q)
2366 struct softnet_data *sd;
2367 unsigned long flags;
2369 local_irq_save(flags);
2370 sd = this_cpu_ptr(&softnet_data);
2371 q->next_sched = NULL;
2372 *sd->output_queue_tailp = q;
2373 sd->output_queue_tailp = &q->next_sched;
2374 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2375 local_irq_restore(flags);
2378 void __netif_schedule(struct Qdisc *q)
2380 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2381 __netif_reschedule(q);
2383 EXPORT_SYMBOL(__netif_schedule);
2385 struct dev_kfree_skb_cb {
2386 enum skb_free_reason reason;
2389 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2391 return (struct dev_kfree_skb_cb *)skb->cb;
2394 void netif_schedule_queue(struct netdev_queue *txq)
2397 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2398 struct Qdisc *q = rcu_dereference(txq->qdisc);
2400 __netif_schedule(q);
2404 EXPORT_SYMBOL(netif_schedule_queue);
2407 * netif_wake_subqueue - allow sending packets on subqueue
2408 * @dev: network device
2409 * @queue_index: sub queue index
2411 * Resume individual transmit queue of a device with multiple transmit queues.
2413 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2415 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2417 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2421 q = rcu_dereference(txq->qdisc);
2422 __netif_schedule(q);
2426 EXPORT_SYMBOL(netif_wake_subqueue);
2428 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2430 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2434 q = rcu_dereference(dev_queue->qdisc);
2435 __netif_schedule(q);
2439 EXPORT_SYMBOL(netif_tx_wake_queue);
2441 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2443 unsigned long flags;
2445 if (likely(atomic_read(&skb->users) == 1)) {
2447 atomic_set(&skb->users, 0);
2448 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2451 get_kfree_skb_cb(skb)->reason = reason;
2452 local_irq_save(flags);
2453 skb->next = __this_cpu_read(softnet_data.completion_queue);
2454 __this_cpu_write(softnet_data.completion_queue, skb);
2455 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2456 local_irq_restore(flags);
2458 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2460 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2462 if (in_irq() || irqs_disabled())
2463 __dev_kfree_skb_irq(skb, reason);
2467 EXPORT_SYMBOL(__dev_kfree_skb_any);
2471 * netif_device_detach - mark device as removed
2472 * @dev: network device
2474 * Mark device as removed from system and therefore no longer available.
2476 void netif_device_detach(struct net_device *dev)
2478 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2479 netif_running(dev)) {
2480 netif_tx_stop_all_queues(dev);
2483 EXPORT_SYMBOL(netif_device_detach);
2486 * netif_device_attach - mark device as attached
2487 * @dev: network device
2489 * Mark device as attached from system and restart if needed.
2491 void netif_device_attach(struct net_device *dev)
2493 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2494 netif_running(dev)) {
2495 netif_tx_wake_all_queues(dev);
2496 __netdev_watchdog_up(dev);
2499 EXPORT_SYMBOL(netif_device_attach);
2502 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2503 * to be used as a distribution range.
2505 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2506 unsigned int num_tx_queues)
2510 u16 qcount = num_tx_queues;
2512 if (skb_rx_queue_recorded(skb)) {
2513 hash = skb_get_rx_queue(skb);
2514 while (unlikely(hash >= num_tx_queues))
2515 hash -= num_tx_queues;
2520 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2521 qoffset = dev->tc_to_txq[tc].offset;
2522 qcount = dev->tc_to_txq[tc].count;
2525 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2527 EXPORT_SYMBOL(__skb_tx_hash);
2529 static void skb_warn_bad_offload(const struct sk_buff *skb)
2531 static const netdev_features_t null_features;
2532 struct net_device *dev = skb->dev;
2533 const char *name = "";
2535 if (!net_ratelimit())
2539 if (dev->dev.parent)
2540 name = dev_driver_string(dev->dev.parent);
2542 name = netdev_name(dev);
2544 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2545 "gso_type=%d ip_summed=%d\n",
2546 name, dev ? &dev->features : &null_features,
2547 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2548 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2549 skb_shinfo(skb)->gso_type, skb->ip_summed);
2553 * Invalidate hardware checksum when packet is to be mangled, and
2554 * complete checksum manually on outgoing path.
2556 int skb_checksum_help(struct sk_buff *skb)
2559 int ret = 0, offset;
2561 if (skb->ip_summed == CHECKSUM_COMPLETE)
2562 goto out_set_summed;
2564 if (unlikely(skb_shinfo(skb)->gso_size)) {
2565 skb_warn_bad_offload(skb);
2569 /* Before computing a checksum, we should make sure no frag could
2570 * be modified by an external entity : checksum could be wrong.
2572 if (skb_has_shared_frag(skb)) {
2573 ret = __skb_linearize(skb);
2578 offset = skb_checksum_start_offset(skb);
2579 BUG_ON(offset >= skb_headlen(skb));
2580 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2582 offset += skb->csum_offset;
2583 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2585 if (skb_cloned(skb) &&
2586 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2587 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2592 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2594 skb->ip_summed = CHECKSUM_NONE;
2598 EXPORT_SYMBOL(skb_checksum_help);
2600 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2602 __be16 type = skb->protocol;
2604 /* Tunnel gso handlers can set protocol to ethernet. */
2605 if (type == htons(ETH_P_TEB)) {
2608 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2611 eth = (struct ethhdr *)skb_mac_header(skb);
2612 type = eth->h_proto;
2615 return __vlan_get_protocol(skb, type, depth);
2619 * skb_mac_gso_segment - mac layer segmentation handler.
2620 * @skb: buffer to segment
2621 * @features: features for the output path (see dev->features)
2623 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2624 netdev_features_t features)
2626 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2627 struct packet_offload *ptype;
2628 int vlan_depth = skb->mac_len;
2629 __be16 type = skb_network_protocol(skb, &vlan_depth);
2631 if (unlikely(!type))
2632 return ERR_PTR(-EINVAL);
2634 __skb_pull(skb, vlan_depth);
2637 list_for_each_entry_rcu(ptype, &offload_base, list) {
2638 if (ptype->type == type && ptype->callbacks.gso_segment) {
2639 segs = ptype->callbacks.gso_segment(skb, features);
2645 __skb_push(skb, skb->data - skb_mac_header(skb));
2649 EXPORT_SYMBOL(skb_mac_gso_segment);
2652 /* openvswitch calls this on rx path, so we need a different check.
2654 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2657 return skb->ip_summed != CHECKSUM_PARTIAL;
2659 return skb->ip_summed == CHECKSUM_NONE;
2663 * __skb_gso_segment - Perform segmentation on skb.
2664 * @skb: buffer to segment
2665 * @features: features for the output path (see dev->features)
2666 * @tx_path: whether it is called in TX path
2668 * This function segments the given skb and returns a list of segments.
2670 * It may return NULL if the skb requires no segmentation. This is
2671 * only possible when GSO is used for verifying header integrity.
2673 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2675 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2676 netdev_features_t features, bool tx_path)
2678 if (unlikely(skb_needs_check(skb, tx_path))) {
2681 skb_warn_bad_offload(skb);
2683 err = skb_cow_head(skb, 0);
2685 return ERR_PTR(err);
2688 /* Only report GSO partial support if it will enable us to
2689 * support segmentation on this frame without needing additional
2692 if (features & NETIF_F_GSO_PARTIAL) {
2693 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
2694 struct net_device *dev = skb->dev;
2696 partial_features |= dev->features & dev->gso_partial_features;
2697 if (!skb_gso_ok(skb, features | partial_features))
2698 features &= ~NETIF_F_GSO_PARTIAL;
2701 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2702 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2704 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2705 SKB_GSO_CB(skb)->encap_level = 0;
2707 skb_reset_mac_header(skb);
2708 skb_reset_mac_len(skb);
2710 return skb_mac_gso_segment(skb, features);
2712 EXPORT_SYMBOL(__skb_gso_segment);
2714 /* Take action when hardware reception checksum errors are detected. */
2716 void netdev_rx_csum_fault(struct net_device *dev)
2718 if (net_ratelimit()) {
2719 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2723 EXPORT_SYMBOL(netdev_rx_csum_fault);
2726 /* Actually, we should eliminate this check as soon as we know, that:
2727 * 1. IOMMU is present and allows to map all the memory.
2728 * 2. No high memory really exists on this machine.
2731 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2733 #ifdef CONFIG_HIGHMEM
2735 if (!(dev->features & NETIF_F_HIGHDMA)) {
2736 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2737 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2738 if (PageHighMem(skb_frag_page(frag)))
2743 if (PCI_DMA_BUS_IS_PHYS) {
2744 struct device *pdev = dev->dev.parent;
2748 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2749 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2750 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2751 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2759 /* If MPLS offload request, verify we are testing hardware MPLS features
2760 * instead of standard features for the netdev.
2762 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2763 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2764 netdev_features_t features,
2767 if (eth_p_mpls(type))
2768 features &= skb->dev->mpls_features;
2773 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2774 netdev_features_t features,
2781 static netdev_features_t harmonize_features(struct sk_buff *skb,
2782 netdev_features_t features)
2787 type = skb_network_protocol(skb, &tmp);
2788 features = net_mpls_features(skb, features, type);
2790 if (skb->ip_summed != CHECKSUM_NONE &&
2791 !can_checksum_protocol(features, type)) {
2792 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2794 if (illegal_highdma(skb->dev, skb))
2795 features &= ~NETIF_F_SG;
2800 netdev_features_t passthru_features_check(struct sk_buff *skb,
2801 struct net_device *dev,
2802 netdev_features_t features)
2806 EXPORT_SYMBOL(passthru_features_check);
2808 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2809 struct net_device *dev,
2810 netdev_features_t features)
2812 return vlan_features_check(skb, features);
2815 static netdev_features_t gso_features_check(const struct sk_buff *skb,
2816 struct net_device *dev,
2817 netdev_features_t features)
2819 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2821 if (gso_segs > dev->gso_max_segs)
2822 return features & ~NETIF_F_GSO_MASK;
2824 /* Support for GSO partial features requires software
2825 * intervention before we can actually process the packets
2826 * so we need to strip support for any partial features now
2827 * and we can pull them back in after we have partially
2828 * segmented the frame.
2830 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
2831 features &= ~dev->gso_partial_features;
2833 /* Make sure to clear the IPv4 ID mangling feature if the
2834 * IPv4 header has the potential to be fragmented.
2836 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2837 struct iphdr *iph = skb->encapsulation ?
2838 inner_ip_hdr(skb) : ip_hdr(skb);
2840 if (!(iph->frag_off & htons(IP_DF)))
2841 features &= ~NETIF_F_TSO_MANGLEID;
2847 netdev_features_t netif_skb_features(struct sk_buff *skb)
2849 struct net_device *dev = skb->dev;
2850 netdev_features_t features = dev->features;
2852 if (skb_is_gso(skb))
2853 features = gso_features_check(skb, dev, features);
2855 /* If encapsulation offload request, verify we are testing
2856 * hardware encapsulation features instead of standard
2857 * features for the netdev
2859 if (skb->encapsulation)
2860 features &= dev->hw_enc_features;
2862 if (skb_vlan_tagged(skb))
2863 features = netdev_intersect_features(features,
2864 dev->vlan_features |
2865 NETIF_F_HW_VLAN_CTAG_TX |
2866 NETIF_F_HW_VLAN_STAG_TX);
2868 if (dev->netdev_ops->ndo_features_check)
2869 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2872 features &= dflt_features_check(skb, dev, features);
2874 return harmonize_features(skb, features);
2876 EXPORT_SYMBOL(netif_skb_features);
2878 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2879 struct netdev_queue *txq, bool more)
2884 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2885 dev_queue_xmit_nit(skb, dev);
2888 trace_net_dev_start_xmit(skb, dev);
2889 rc = netdev_start_xmit(skb, dev, txq, more);
2890 trace_net_dev_xmit(skb, rc, dev, len);
2895 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2896 struct netdev_queue *txq, int *ret)
2898 struct sk_buff *skb = first;
2899 int rc = NETDEV_TX_OK;
2902 struct sk_buff *next = skb->next;
2905 rc = xmit_one(skb, dev, txq, next != NULL);
2906 if (unlikely(!dev_xmit_complete(rc))) {
2912 if (netif_xmit_stopped(txq) && skb) {
2913 rc = NETDEV_TX_BUSY;
2923 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2924 netdev_features_t features)
2926 if (skb_vlan_tag_present(skb) &&
2927 !vlan_hw_offload_capable(features, skb->vlan_proto))
2928 skb = __vlan_hwaccel_push_inside(skb);
2932 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2934 netdev_features_t features;
2936 features = netif_skb_features(skb);
2937 skb = validate_xmit_vlan(skb, features);
2941 if (netif_needs_gso(skb, features)) {
2942 struct sk_buff *segs;
2944 segs = skb_gso_segment(skb, features);
2952 if (skb_needs_linearize(skb, features) &&
2953 __skb_linearize(skb))
2956 /* If packet is not checksummed and device does not
2957 * support checksumming for this protocol, complete
2958 * checksumming here.
2960 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2961 if (skb->encapsulation)
2962 skb_set_inner_transport_header(skb,
2963 skb_checksum_start_offset(skb));
2965 skb_set_transport_header(skb,
2966 skb_checksum_start_offset(skb));
2967 if (!(features & NETIF_F_CSUM_MASK) &&
2968 skb_checksum_help(skb))
2978 atomic_long_inc(&dev->tx_dropped);
2982 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2984 struct sk_buff *next, *head = NULL, *tail;
2986 for (; skb != NULL; skb = next) {
2990 /* in case skb wont be segmented, point to itself */
2993 skb = validate_xmit_skb(skb, dev);
3001 /* If skb was segmented, skb->prev points to
3002 * the last segment. If not, it still contains skb.
3008 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3010 static void qdisc_pkt_len_init(struct sk_buff *skb)
3012 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3014 qdisc_skb_cb(skb)->pkt_len = skb->len;
3016 /* To get more precise estimation of bytes sent on wire,
3017 * we add to pkt_len the headers size of all segments
3019 if (shinfo->gso_size) {
3020 unsigned int hdr_len;
3021 u16 gso_segs = shinfo->gso_segs;
3023 /* mac layer + network layer */
3024 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3026 /* + transport layer */
3027 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
3028 hdr_len += tcp_hdrlen(skb);
3030 hdr_len += sizeof(struct udphdr);
3032 if (shinfo->gso_type & SKB_GSO_DODGY)
3033 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3036 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3040 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3041 struct net_device *dev,
3042 struct netdev_queue *txq)
3044 spinlock_t *root_lock = qdisc_lock(q);
3045 struct sk_buff *to_free = NULL;
3049 qdisc_calculate_pkt_len(skb, q);
3051 * Heuristic to force contended enqueues to serialize on a
3052 * separate lock before trying to get qdisc main lock.
3053 * This permits qdisc->running owner to get the lock more
3054 * often and dequeue packets faster.
3056 contended = qdisc_is_running(q);
3057 if (unlikely(contended))
3058 spin_lock(&q->busylock);
3060 spin_lock(root_lock);
3061 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3062 __qdisc_drop(skb, &to_free);
3064 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3065 qdisc_run_begin(q)) {
3067 * This is a work-conserving queue; there are no old skbs
3068 * waiting to be sent out; and the qdisc is not running -
3069 * xmit the skb directly.
3072 qdisc_bstats_update(q, skb);
3074 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3075 if (unlikely(contended)) {
3076 spin_unlock(&q->busylock);
3083 rc = NET_XMIT_SUCCESS;
3085 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3086 if (qdisc_run_begin(q)) {
3087 if (unlikely(contended)) {
3088 spin_unlock(&q->busylock);
3094 spin_unlock(root_lock);
3095 if (unlikely(to_free))
3096 kfree_skb_list(to_free);
3097 if (unlikely(contended))
3098 spin_unlock(&q->busylock);
3102 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3103 static void skb_update_prio(struct sk_buff *skb)
3105 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
3107 if (!skb->priority && skb->sk && map) {
3108 unsigned int prioidx =
3109 sock_cgroup_prioidx(&skb->sk->sk_cgrp_data);
3111 if (prioidx < map->priomap_len)
3112 skb->priority = map->priomap[prioidx];
3116 #define skb_update_prio(skb)
3119 DEFINE_PER_CPU(int, xmit_recursion);
3120 EXPORT_SYMBOL(xmit_recursion);
3123 * dev_loopback_xmit - loop back @skb
3124 * @net: network namespace this loopback is happening in
3125 * @sk: sk needed to be a netfilter okfn
3126 * @skb: buffer to transmit
3128 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3130 skb_reset_mac_header(skb);
3131 __skb_pull(skb, skb_network_offset(skb));
3132 skb->pkt_type = PACKET_LOOPBACK;
3133 skb->ip_summed = CHECKSUM_UNNECESSARY;
3134 WARN_ON(!skb_dst(skb));
3139 EXPORT_SYMBOL(dev_loopback_xmit);
3141 #ifdef CONFIG_NET_EGRESS
3142 static struct sk_buff *
3143 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3145 struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list);
3146 struct tcf_result cl_res;
3151 /* skb->tc_verd and qdisc_skb_cb(skb)->pkt_len were already set
3152 * earlier by the caller.
3154 qdisc_bstats_cpu_update(cl->q, skb);
3156 switch (tc_classify(skb, cl, &cl_res, false)) {
3158 case TC_ACT_RECLASSIFY:
3159 skb->tc_index = TC_H_MIN(cl_res.classid);
3162 qdisc_qstats_cpu_drop(cl->q);
3163 *ret = NET_XMIT_DROP;
3168 *ret = NET_XMIT_SUCCESS;
3171 case TC_ACT_REDIRECT:
3172 /* No need to push/pop skb's mac_header here on egress! */
3173 skb_do_redirect(skb);
3174 *ret = NET_XMIT_SUCCESS;
3182 #endif /* CONFIG_NET_EGRESS */
3184 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3187 struct xps_dev_maps *dev_maps;
3188 struct xps_map *map;
3189 int queue_index = -1;
3192 dev_maps = rcu_dereference(dev->xps_maps);
3194 unsigned int tci = skb->sender_cpu - 1;
3198 tci += netdev_get_prio_tc_map(dev, skb->priority);
3201 map = rcu_dereference(dev_maps->cpu_map[tci]);
3204 queue_index = map->queues[0];
3206 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3208 if (unlikely(queue_index >= dev->real_num_tx_queues))
3220 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3222 struct sock *sk = skb->sk;
3223 int queue_index = sk_tx_queue_get(sk);
3225 if (queue_index < 0 || skb->ooo_okay ||
3226 queue_index >= dev->real_num_tx_queues) {
3227 int new_index = get_xps_queue(dev, skb);
3229 new_index = skb_tx_hash(dev, skb);
3231 if (queue_index != new_index && sk &&
3233 rcu_access_pointer(sk->sk_dst_cache))
3234 sk_tx_queue_set(sk, new_index);
3236 queue_index = new_index;
3242 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3243 struct sk_buff *skb,
3246 int queue_index = 0;
3249 u32 sender_cpu = skb->sender_cpu - 1;
3251 if (sender_cpu >= (u32)NR_CPUS)
3252 skb->sender_cpu = raw_smp_processor_id() + 1;
3255 if (dev->real_num_tx_queues != 1) {
3256 const struct net_device_ops *ops = dev->netdev_ops;
3257 if (ops->ndo_select_queue)
3258 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3261 queue_index = __netdev_pick_tx(dev, skb);
3264 queue_index = netdev_cap_txqueue(dev, queue_index);
3267 skb_set_queue_mapping(skb, queue_index);
3268 return netdev_get_tx_queue(dev, queue_index);
3272 * __dev_queue_xmit - transmit a buffer
3273 * @skb: buffer to transmit
3274 * @accel_priv: private data used for L2 forwarding offload
3276 * Queue a buffer for transmission to a network device. The caller must
3277 * have set the device and priority and built the buffer before calling
3278 * this function. The function can be called from an interrupt.
3280 * A negative errno code is returned on a failure. A success does not
3281 * guarantee the frame will be transmitted as it may be dropped due
3282 * to congestion or traffic shaping.
3284 * -----------------------------------------------------------------------------------
3285 * I notice this method can also return errors from the queue disciplines,
3286 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3289 * Regardless of the return value, the skb is consumed, so it is currently
3290 * difficult to retry a send to this method. (You can bump the ref count
3291 * before sending to hold a reference for retry if you are careful.)
3293 * When calling this method, interrupts MUST be enabled. This is because
3294 * the BH enable code must have IRQs enabled so that it will not deadlock.
3297 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3299 struct net_device *dev = skb->dev;
3300 struct netdev_queue *txq;
3304 skb_reset_mac_header(skb);
3306 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3307 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3309 /* Disable soft irqs for various locks below. Also
3310 * stops preemption for RCU.
3314 skb_update_prio(skb);
3316 qdisc_pkt_len_init(skb);
3317 #ifdef CONFIG_NET_CLS_ACT
3318 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3319 # ifdef CONFIG_NET_EGRESS
3320 if (static_key_false(&egress_needed)) {
3321 skb = sch_handle_egress(skb, &rc, dev);
3327 /* If device/qdisc don't need skb->dst, release it right now while
3328 * its hot in this cpu cache.
3330 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3335 txq = netdev_pick_tx(dev, skb, accel_priv);
3336 q = rcu_dereference_bh(txq->qdisc);
3338 trace_net_dev_queue(skb);
3340 rc = __dev_xmit_skb(skb, q, dev, txq);
3344 /* The device has no queue. Common case for software devices:
3345 loopback, all the sorts of tunnels...
3347 Really, it is unlikely that netif_tx_lock protection is necessary
3348 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3350 However, it is possible, that they rely on protection
3353 Check this and shot the lock. It is not prone from deadlocks.
3354 Either shot noqueue qdisc, it is even simpler 8)
3356 if (dev->flags & IFF_UP) {
3357 int cpu = smp_processor_id(); /* ok because BHs are off */
3359 if (txq->xmit_lock_owner != cpu) {
3360 if (unlikely(__this_cpu_read(xmit_recursion) >
3361 XMIT_RECURSION_LIMIT))
3362 goto recursion_alert;
3364 skb = validate_xmit_skb(skb, dev);
3368 HARD_TX_LOCK(dev, txq, cpu);
3370 if (!netif_xmit_stopped(txq)) {
3371 __this_cpu_inc(xmit_recursion);
3372 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3373 __this_cpu_dec(xmit_recursion);
3374 if (dev_xmit_complete(rc)) {
3375 HARD_TX_UNLOCK(dev, txq);
3379 HARD_TX_UNLOCK(dev, txq);
3380 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3383 /* Recursion is detected! It is possible,
3387 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3393 rcu_read_unlock_bh();
3395 atomic_long_inc(&dev->tx_dropped);
3396 kfree_skb_list(skb);
3399 rcu_read_unlock_bh();
3403 int dev_queue_xmit(struct sk_buff *skb)
3405 return __dev_queue_xmit(skb, NULL);
3407 EXPORT_SYMBOL(dev_queue_xmit);
3409 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3411 return __dev_queue_xmit(skb, accel_priv);
3413 EXPORT_SYMBOL(dev_queue_xmit_accel);
3416 /*=======================================================================
3418 =======================================================================*/
3420 int netdev_max_backlog __read_mostly = 1000;
3421 EXPORT_SYMBOL(netdev_max_backlog);
3423 int netdev_tstamp_prequeue __read_mostly = 1;
3424 int netdev_budget __read_mostly = 300;
3425 int weight_p __read_mostly = 64; /* old backlog weight */
3427 /* Called with irq disabled */
3428 static inline void ____napi_schedule(struct softnet_data *sd,
3429 struct napi_struct *napi)
3431 list_add_tail(&napi->poll_list, &sd->poll_list);
3432 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3437 /* One global table that all flow-based protocols share. */
3438 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3439 EXPORT_SYMBOL(rps_sock_flow_table);
3440 u32 rps_cpu_mask __read_mostly;
3441 EXPORT_SYMBOL(rps_cpu_mask);
3443 struct static_key rps_needed __read_mostly;
3444 EXPORT_SYMBOL(rps_needed);
3445 struct static_key rfs_needed __read_mostly;
3446 EXPORT_SYMBOL(rfs_needed);
3448 static struct rps_dev_flow *
3449 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3450 struct rps_dev_flow *rflow, u16 next_cpu)
3452 if (next_cpu < nr_cpu_ids) {
3453 #ifdef CONFIG_RFS_ACCEL
3454 struct netdev_rx_queue *rxqueue;
3455 struct rps_dev_flow_table *flow_table;
3456 struct rps_dev_flow *old_rflow;
3461 /* Should we steer this flow to a different hardware queue? */
3462 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3463 !(dev->features & NETIF_F_NTUPLE))
3465 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3466 if (rxq_index == skb_get_rx_queue(skb))
3469 rxqueue = dev->_rx + rxq_index;
3470 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3473 flow_id = skb_get_hash(skb) & flow_table->mask;
3474 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3475 rxq_index, flow_id);
3479 rflow = &flow_table->flows[flow_id];
3481 if (old_rflow->filter == rflow->filter)
3482 old_rflow->filter = RPS_NO_FILTER;
3486 per_cpu(softnet_data, next_cpu).input_queue_head;
3489 rflow->cpu = next_cpu;
3494 * get_rps_cpu is called from netif_receive_skb and returns the target
3495 * CPU from the RPS map of the receiving queue for a given skb.
3496 * rcu_read_lock must be held on entry.
3498 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3499 struct rps_dev_flow **rflowp)
3501 const struct rps_sock_flow_table *sock_flow_table;
3502 struct netdev_rx_queue *rxqueue = dev->_rx;
3503 struct rps_dev_flow_table *flow_table;
3504 struct rps_map *map;
3509 if (skb_rx_queue_recorded(skb)) {
3510 u16 index = skb_get_rx_queue(skb);
3512 if (unlikely(index >= dev->real_num_rx_queues)) {
3513 WARN_ONCE(dev->real_num_rx_queues > 1,
3514 "%s received packet on queue %u, but number "
3515 "of RX queues is %u\n",
3516 dev->name, index, dev->real_num_rx_queues);
3522 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3524 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3525 map = rcu_dereference(rxqueue->rps_map);
3526 if (!flow_table && !map)
3529 skb_reset_network_header(skb);
3530 hash = skb_get_hash(skb);
3534 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3535 if (flow_table && sock_flow_table) {
3536 struct rps_dev_flow *rflow;
3540 /* First check into global flow table if there is a match */
3541 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3542 if ((ident ^ hash) & ~rps_cpu_mask)
3545 next_cpu = ident & rps_cpu_mask;
3547 /* OK, now we know there is a match,
3548 * we can look at the local (per receive queue) flow table
3550 rflow = &flow_table->flows[hash & flow_table->mask];
3554 * If the desired CPU (where last recvmsg was done) is
3555 * different from current CPU (one in the rx-queue flow
3556 * table entry), switch if one of the following holds:
3557 * - Current CPU is unset (>= nr_cpu_ids).
3558 * - Current CPU is offline.
3559 * - The current CPU's queue tail has advanced beyond the
3560 * last packet that was enqueued using this table entry.
3561 * This guarantees that all previous packets for the flow
3562 * have been dequeued, thus preserving in order delivery.
3564 if (unlikely(tcpu != next_cpu) &&
3565 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3566 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3567 rflow->last_qtail)) >= 0)) {
3569 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3572 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3582 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3583 if (cpu_online(tcpu)) {
3593 #ifdef CONFIG_RFS_ACCEL
3596 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3597 * @dev: Device on which the filter was set
3598 * @rxq_index: RX queue index
3599 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3600 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3602 * Drivers that implement ndo_rx_flow_steer() should periodically call
3603 * this function for each installed filter and remove the filters for
3604 * which it returns %true.
3606 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3607 u32 flow_id, u16 filter_id)
3609 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3610 struct rps_dev_flow_table *flow_table;
3611 struct rps_dev_flow *rflow;
3616 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3617 if (flow_table && flow_id <= flow_table->mask) {
3618 rflow = &flow_table->flows[flow_id];
3619 cpu = ACCESS_ONCE(rflow->cpu);
3620 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3621 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3622 rflow->last_qtail) <
3623 (int)(10 * flow_table->mask)))
3629 EXPORT_SYMBOL(rps_may_expire_flow);
3631 #endif /* CONFIG_RFS_ACCEL */
3633 /* Called from hardirq (IPI) context */
3634 static void rps_trigger_softirq(void *data)
3636 struct softnet_data *sd = data;
3638 ____napi_schedule(sd, &sd->backlog);
3642 #endif /* CONFIG_RPS */
3645 * Check if this softnet_data structure is another cpu one
3646 * If yes, queue it to our IPI list and return 1
3649 static int rps_ipi_queued(struct softnet_data *sd)
3652 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3655 sd->rps_ipi_next = mysd->rps_ipi_list;
3656 mysd->rps_ipi_list = sd;
3658 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3661 #endif /* CONFIG_RPS */
3665 #ifdef CONFIG_NET_FLOW_LIMIT
3666 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3669 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3671 #ifdef CONFIG_NET_FLOW_LIMIT
3672 struct sd_flow_limit *fl;
3673 struct softnet_data *sd;
3674 unsigned int old_flow, new_flow;
3676 if (qlen < (netdev_max_backlog >> 1))
3679 sd = this_cpu_ptr(&softnet_data);
3682 fl = rcu_dereference(sd->flow_limit);
3684 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3685 old_flow = fl->history[fl->history_head];
3686 fl->history[fl->history_head] = new_flow;
3689 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3691 if (likely(fl->buckets[old_flow]))
3692 fl->buckets[old_flow]--;
3694 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3706 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3707 * queue (may be a remote CPU queue).
3709 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3710 unsigned int *qtail)
3712 struct softnet_data *sd;
3713 unsigned long flags;
3716 sd = &per_cpu(softnet_data, cpu);
3718 local_irq_save(flags);
3721 if (!netif_running(skb->dev))
3723 qlen = skb_queue_len(&sd->input_pkt_queue);
3724 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3727 __skb_queue_tail(&sd->input_pkt_queue, skb);
3728 input_queue_tail_incr_save(sd, qtail);
3730 local_irq_restore(flags);
3731 return NET_RX_SUCCESS;
3734 /* Schedule NAPI for backlog device
3735 * We can use non atomic operation since we own the queue lock
3737 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3738 if (!rps_ipi_queued(sd))
3739 ____napi_schedule(sd, &sd->backlog);
3748 local_irq_restore(flags);
3750 atomic_long_inc(&skb->dev->rx_dropped);
3755 static int netif_rx_internal(struct sk_buff *skb)
3759 net_timestamp_check(netdev_tstamp_prequeue, skb);
3761 trace_netif_rx(skb);
3763 if (static_key_false(&rps_needed)) {
3764 struct rps_dev_flow voidflow, *rflow = &voidflow;
3770 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3772 cpu = smp_processor_id();
3774 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3782 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3789 * netif_rx - post buffer to the network code
3790 * @skb: buffer to post
3792 * This function receives a packet from a device driver and queues it for
3793 * the upper (protocol) levels to process. It always succeeds. The buffer
3794 * may be dropped during processing for congestion control or by the
3798 * NET_RX_SUCCESS (no congestion)
3799 * NET_RX_DROP (packet was dropped)
3803 int netif_rx(struct sk_buff *skb)
3805 trace_netif_rx_entry(skb);
3807 return netif_rx_internal(skb);
3809 EXPORT_SYMBOL(netif_rx);
3811 int netif_rx_ni(struct sk_buff *skb)
3815 trace_netif_rx_ni_entry(skb);
3818 err = netif_rx_internal(skb);
3819 if (local_softirq_pending())
3825 EXPORT_SYMBOL(netif_rx_ni);
3827 static __latent_entropy void net_tx_action(struct softirq_action *h)
3829 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3831 if (sd->completion_queue) {
3832 struct sk_buff *clist;
3834 local_irq_disable();
3835 clist = sd->completion_queue;
3836 sd->completion_queue = NULL;
3840 struct sk_buff *skb = clist;
3841 clist = clist->next;
3843 WARN_ON(atomic_read(&skb->users));
3844 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3845 trace_consume_skb(skb);
3847 trace_kfree_skb(skb, net_tx_action);
3849 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
3852 __kfree_skb_defer(skb);
3855 __kfree_skb_flush();
3858 if (sd->output_queue) {
3861 local_irq_disable();
3862 head = sd->output_queue;
3863 sd->output_queue = NULL;
3864 sd->output_queue_tailp = &sd->output_queue;
3868 struct Qdisc *q = head;
3869 spinlock_t *root_lock;
3871 head = head->next_sched;
3873 root_lock = qdisc_lock(q);
3874 spin_lock(root_lock);
3875 /* We need to make sure head->next_sched is read
3876 * before clearing __QDISC_STATE_SCHED
3878 smp_mb__before_atomic();
3879 clear_bit(__QDISC_STATE_SCHED, &q->state);
3881 spin_unlock(root_lock);
3886 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
3887 /* This hook is defined here for ATM LANE */
3888 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3889 unsigned char *addr) __read_mostly;
3890 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3893 static inline struct sk_buff *
3894 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
3895 struct net_device *orig_dev)
3897 #ifdef CONFIG_NET_CLS_ACT
3898 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3899 struct tcf_result cl_res;
3901 /* If there's at least one ingress present somewhere (so
3902 * we get here via enabled static key), remaining devices
3903 * that are not configured with an ingress qdisc will bail
3909 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3913 qdisc_skb_cb(skb)->pkt_len = skb->len;
3914 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3915 qdisc_bstats_cpu_update(cl->q, skb);
3917 switch (tc_classify(skb, cl, &cl_res, false)) {
3919 case TC_ACT_RECLASSIFY:
3920 skb->tc_index = TC_H_MIN(cl_res.classid);
3923 qdisc_qstats_cpu_drop(cl->q);
3930 case TC_ACT_REDIRECT:
3931 /* skb_mac_header check was done by cls/act_bpf, so
3932 * we can safely push the L2 header back before
3933 * redirecting to another netdev
3935 __skb_push(skb, skb->mac_len);
3936 skb_do_redirect(skb);
3941 #endif /* CONFIG_NET_CLS_ACT */
3946 * netdev_is_rx_handler_busy - check if receive handler is registered
3947 * @dev: device to check
3949 * Check if a receive handler is already registered for a given device.
3950 * Return true if there one.
3952 * The caller must hold the rtnl_mutex.
3954 bool netdev_is_rx_handler_busy(struct net_device *dev)
3957 return dev && rtnl_dereference(dev->rx_handler);
3959 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
3962 * netdev_rx_handler_register - register receive handler
3963 * @dev: device to register a handler for
3964 * @rx_handler: receive handler to register
3965 * @rx_handler_data: data pointer that is used by rx handler
3967 * Register a receive handler for a device. This handler will then be
3968 * called from __netif_receive_skb. A negative errno code is returned
3971 * The caller must hold the rtnl_mutex.
3973 * For a general description of rx_handler, see enum rx_handler_result.
3975 int netdev_rx_handler_register(struct net_device *dev,
3976 rx_handler_func_t *rx_handler,
3977 void *rx_handler_data)
3981 if (dev->rx_handler)
3984 /* Note: rx_handler_data must be set before rx_handler */
3985 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3986 rcu_assign_pointer(dev->rx_handler, rx_handler);
3990 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3993 * netdev_rx_handler_unregister - unregister receive handler
3994 * @dev: device to unregister a handler from
3996 * Unregister a receive handler from a device.
3998 * The caller must hold the rtnl_mutex.
4000 void netdev_rx_handler_unregister(struct net_device *dev)
4004 RCU_INIT_POINTER(dev->rx_handler, NULL);
4005 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4006 * section has a guarantee to see a non NULL rx_handler_data
4010 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4012 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4015 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4016 * the special handling of PFMEMALLOC skbs.
4018 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4020 switch (skb->protocol) {
4021 case htons(ETH_P_ARP):
4022 case htons(ETH_P_IP):
4023 case htons(ETH_P_IPV6):
4024 case htons(ETH_P_8021Q):
4025 case htons(ETH_P_8021AD):
4032 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4033 int *ret, struct net_device *orig_dev)
4035 #ifdef CONFIG_NETFILTER_INGRESS
4036 if (nf_hook_ingress_active(skb)) {
4040 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4045 ingress_retval = nf_hook_ingress(skb);
4047 return ingress_retval;
4049 #endif /* CONFIG_NETFILTER_INGRESS */
4053 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
4055 struct packet_type *ptype, *pt_prev;
4056 rx_handler_func_t *rx_handler;
4057 struct net_device *orig_dev;
4058 bool deliver_exact = false;
4059 int ret = NET_RX_DROP;
4062 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4064 trace_netif_receive_skb(skb);
4066 orig_dev = skb->dev;
4068 skb_reset_network_header(skb);
4069 if (!skb_transport_header_was_set(skb))
4070 skb_reset_transport_header(skb);
4071 skb_reset_mac_len(skb);
4076 skb->skb_iif = skb->dev->ifindex;
4078 __this_cpu_inc(softnet_data.processed);
4080 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4081 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4082 skb = skb_vlan_untag(skb);
4087 #ifdef CONFIG_NET_CLS_ACT
4088 if (skb->tc_verd & TC_NCLS) {
4089 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
4097 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4099 ret = deliver_skb(skb, pt_prev, orig_dev);
4103 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4105 ret = deliver_skb(skb, pt_prev, orig_dev);
4110 #ifdef CONFIG_NET_INGRESS
4111 if (static_key_false(&ingress_needed)) {
4112 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4116 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4120 #ifdef CONFIG_NET_CLS_ACT
4124 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4127 if (skb_vlan_tag_present(skb)) {
4129 ret = deliver_skb(skb, pt_prev, orig_dev);
4132 if (vlan_do_receive(&skb))
4134 else if (unlikely(!skb))
4138 rx_handler = rcu_dereference(skb->dev->rx_handler);
4141 ret = deliver_skb(skb, pt_prev, orig_dev);
4144 switch (rx_handler(&skb)) {
4145 case RX_HANDLER_CONSUMED:
4146 ret = NET_RX_SUCCESS;
4148 case RX_HANDLER_ANOTHER:
4150 case RX_HANDLER_EXACT:
4151 deliver_exact = true;
4152 case RX_HANDLER_PASS:
4159 if (unlikely(skb_vlan_tag_present(skb))) {
4160 if (skb_vlan_tag_get_id(skb))
4161 skb->pkt_type = PACKET_OTHERHOST;
4162 /* Note: we might in the future use prio bits
4163 * and set skb->priority like in vlan_do_receive()
4164 * For the time being, just ignore Priority Code Point
4169 type = skb->protocol;
4171 /* deliver only exact match when indicated */
4172 if (likely(!deliver_exact)) {
4173 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4174 &ptype_base[ntohs(type) &
4178 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4179 &orig_dev->ptype_specific);
4181 if (unlikely(skb->dev != orig_dev)) {
4182 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4183 &skb->dev->ptype_specific);
4187 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
4190 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4194 atomic_long_inc(&skb->dev->rx_dropped);
4196 atomic_long_inc(&skb->dev->rx_nohandler);
4198 /* Jamal, now you will not able to escape explaining
4199 * me how you were going to use this. :-)
4208 static int __netif_receive_skb(struct sk_buff *skb)
4212 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4213 unsigned long pflags = current->flags;
4216 * PFMEMALLOC skbs are special, they should
4217 * - be delivered to SOCK_MEMALLOC sockets only
4218 * - stay away from userspace
4219 * - have bounded memory usage
4221 * Use PF_MEMALLOC as this saves us from propagating the allocation
4222 * context down to all allocation sites.
4224 current->flags |= PF_MEMALLOC;
4225 ret = __netif_receive_skb_core(skb, true);
4226 tsk_restore_flags(current, pflags, PF_MEMALLOC);
4228 ret = __netif_receive_skb_core(skb, false);
4233 static int netif_receive_skb_internal(struct sk_buff *skb)
4237 net_timestamp_check(netdev_tstamp_prequeue, skb);
4239 if (skb_defer_rx_timestamp(skb))
4240 return NET_RX_SUCCESS;
4245 if (static_key_false(&rps_needed)) {
4246 struct rps_dev_flow voidflow, *rflow = &voidflow;
4247 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4250 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4256 ret = __netif_receive_skb(skb);
4262 * netif_receive_skb - process receive buffer from network
4263 * @skb: buffer to process
4265 * netif_receive_skb() is the main receive data processing function.
4266 * It always succeeds. The buffer may be dropped during processing
4267 * for congestion control or by the protocol layers.
4269 * This function may only be called from softirq context and interrupts
4270 * should be enabled.
4272 * Return values (usually ignored):
4273 * NET_RX_SUCCESS: no congestion
4274 * NET_RX_DROP: packet was dropped
4276 int netif_receive_skb(struct sk_buff *skb)
4278 trace_netif_receive_skb_entry(skb);
4280 return netif_receive_skb_internal(skb);
4282 EXPORT_SYMBOL(netif_receive_skb);
4284 DEFINE_PER_CPU(struct work_struct, flush_works);
4286 /* Network device is going away, flush any packets still pending */
4287 static void flush_backlog(struct work_struct *work)
4289 struct sk_buff *skb, *tmp;
4290 struct softnet_data *sd;
4293 sd = this_cpu_ptr(&softnet_data);
4295 local_irq_disable();
4297 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4298 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4299 __skb_unlink(skb, &sd->input_pkt_queue);
4301 input_queue_head_incr(sd);
4307 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4308 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4309 __skb_unlink(skb, &sd->process_queue);
4311 input_queue_head_incr(sd);
4317 static void flush_all_backlogs(void)
4323 for_each_online_cpu(cpu)
4324 queue_work_on(cpu, system_highpri_wq,
4325 per_cpu_ptr(&flush_works, cpu));
4327 for_each_online_cpu(cpu)
4328 flush_work(per_cpu_ptr(&flush_works, cpu));
4333 static int napi_gro_complete(struct sk_buff *skb)
4335 struct packet_offload *ptype;
4336 __be16 type = skb->protocol;
4337 struct list_head *head = &offload_base;
4340 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4342 if (NAPI_GRO_CB(skb)->count == 1) {
4343 skb_shinfo(skb)->gso_size = 0;
4348 list_for_each_entry_rcu(ptype, head, list) {
4349 if (ptype->type != type || !ptype->callbacks.gro_complete)
4352 err = ptype->callbacks.gro_complete(skb, 0);
4358 WARN_ON(&ptype->list == head);
4360 return NET_RX_SUCCESS;
4364 return netif_receive_skb_internal(skb);
4367 /* napi->gro_list contains packets ordered by age.
4368 * youngest packets at the head of it.
4369 * Complete skbs in reverse order to reduce latencies.
4371 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4373 struct sk_buff *skb, *prev = NULL;
4375 /* scan list and build reverse chain */
4376 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4381 for (skb = prev; skb; skb = prev) {
4384 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4388 napi_gro_complete(skb);
4392 napi->gro_list = NULL;
4394 EXPORT_SYMBOL(napi_gro_flush);
4396 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4399 unsigned int maclen = skb->dev->hard_header_len;
4400 u32 hash = skb_get_hash_raw(skb);
4402 for (p = napi->gro_list; p; p = p->next) {
4403 unsigned long diffs;
4405 NAPI_GRO_CB(p)->flush = 0;
4407 if (hash != skb_get_hash_raw(p)) {
4408 NAPI_GRO_CB(p)->same_flow = 0;
4412 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4413 diffs |= p->vlan_tci ^ skb->vlan_tci;
4414 diffs |= skb_metadata_dst_cmp(p, skb);
4415 if (maclen == ETH_HLEN)
4416 diffs |= compare_ether_header(skb_mac_header(p),
4417 skb_mac_header(skb));
4419 diffs = memcmp(skb_mac_header(p),
4420 skb_mac_header(skb),
4422 NAPI_GRO_CB(p)->same_flow = !diffs;
4426 static void skb_gro_reset_offset(struct sk_buff *skb)
4428 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4429 const skb_frag_t *frag0 = &pinfo->frags[0];
4431 NAPI_GRO_CB(skb)->data_offset = 0;
4432 NAPI_GRO_CB(skb)->frag0 = NULL;
4433 NAPI_GRO_CB(skb)->frag0_len = 0;
4435 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4437 !PageHighMem(skb_frag_page(frag0))) {
4438 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4439 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
4440 skb_frag_size(frag0),
4441 skb->end - skb->tail);
4445 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4447 struct skb_shared_info *pinfo = skb_shinfo(skb);
4449 BUG_ON(skb->end - skb->tail < grow);
4451 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4453 skb->data_len -= grow;
4456 pinfo->frags[0].page_offset += grow;
4457 skb_frag_size_sub(&pinfo->frags[0], grow);
4459 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4460 skb_frag_unref(skb, 0);
4461 memmove(pinfo->frags, pinfo->frags + 1,
4462 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4466 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4468 struct sk_buff **pp = NULL;
4469 struct packet_offload *ptype;
4470 __be16 type = skb->protocol;
4471 struct list_head *head = &offload_base;
4473 enum gro_result ret;
4476 if (!(skb->dev->features & NETIF_F_GRO))
4482 gro_list_prepare(napi, skb);
4485 list_for_each_entry_rcu(ptype, head, list) {
4486 if (ptype->type != type || !ptype->callbacks.gro_receive)
4489 skb_set_network_header(skb, skb_gro_offset(skb));
4490 skb_reset_mac_len(skb);
4491 NAPI_GRO_CB(skb)->same_flow = 0;
4492 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
4493 NAPI_GRO_CB(skb)->free = 0;
4494 NAPI_GRO_CB(skb)->encap_mark = 0;
4495 NAPI_GRO_CB(skb)->recursion_counter = 0;
4496 NAPI_GRO_CB(skb)->is_fou = 0;
4497 NAPI_GRO_CB(skb)->is_atomic = 1;
4498 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4500 /* Setup for GRO checksum validation */
4501 switch (skb->ip_summed) {
4502 case CHECKSUM_COMPLETE:
4503 NAPI_GRO_CB(skb)->csum = skb->csum;
4504 NAPI_GRO_CB(skb)->csum_valid = 1;
4505 NAPI_GRO_CB(skb)->csum_cnt = 0;
4507 case CHECKSUM_UNNECESSARY:
4508 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4509 NAPI_GRO_CB(skb)->csum_valid = 0;
4512 NAPI_GRO_CB(skb)->csum_cnt = 0;
4513 NAPI_GRO_CB(skb)->csum_valid = 0;
4516 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4521 if (&ptype->list == head)
4524 same_flow = NAPI_GRO_CB(skb)->same_flow;
4525 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4528 struct sk_buff *nskb = *pp;
4532 napi_gro_complete(nskb);
4539 if (NAPI_GRO_CB(skb)->flush)
4542 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4543 struct sk_buff *nskb = napi->gro_list;
4545 /* locate the end of the list to select the 'oldest' flow */
4546 while (nskb->next) {
4552 napi_gro_complete(nskb);
4556 NAPI_GRO_CB(skb)->count = 1;
4557 NAPI_GRO_CB(skb)->age = jiffies;
4558 NAPI_GRO_CB(skb)->last = skb;
4559 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4560 skb->next = napi->gro_list;
4561 napi->gro_list = skb;
4565 grow = skb_gro_offset(skb) - skb_headlen(skb);
4567 gro_pull_from_frag0(skb, grow);
4576 struct packet_offload *gro_find_receive_by_type(__be16 type)
4578 struct list_head *offload_head = &offload_base;
4579 struct packet_offload *ptype;
4581 list_for_each_entry_rcu(ptype, offload_head, list) {
4582 if (ptype->type != type || !ptype->callbacks.gro_receive)
4588 EXPORT_SYMBOL(gro_find_receive_by_type);
4590 struct packet_offload *gro_find_complete_by_type(__be16 type)
4592 struct list_head *offload_head = &offload_base;
4593 struct packet_offload *ptype;
4595 list_for_each_entry_rcu(ptype, offload_head, list) {
4596 if (ptype->type != type || !ptype->callbacks.gro_complete)
4602 EXPORT_SYMBOL(gro_find_complete_by_type);
4604 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4608 if (netif_receive_skb_internal(skb))
4616 case GRO_MERGED_FREE:
4617 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) {
4619 kmem_cache_free(skbuff_head_cache, skb);
4633 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4635 skb_mark_napi_id(skb, napi);
4636 trace_napi_gro_receive_entry(skb);
4638 skb_gro_reset_offset(skb);
4640 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4642 EXPORT_SYMBOL(napi_gro_receive);
4644 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4646 if (unlikely(skb->pfmemalloc)) {
4650 __skb_pull(skb, skb_headlen(skb));
4651 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4652 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4654 skb->dev = napi->dev;
4656 skb->encapsulation = 0;
4657 skb_shinfo(skb)->gso_type = 0;
4658 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4663 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4665 struct sk_buff *skb = napi->skb;
4668 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4671 skb_mark_napi_id(skb, napi);
4676 EXPORT_SYMBOL(napi_get_frags);
4678 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4679 struct sk_buff *skb,
4685 __skb_push(skb, ETH_HLEN);
4686 skb->protocol = eth_type_trans(skb, skb->dev);
4687 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4692 case GRO_MERGED_FREE:
4693 napi_reuse_skb(napi, skb);
4703 /* Upper GRO stack assumes network header starts at gro_offset=0
4704 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4705 * We copy ethernet header into skb->data to have a common layout.
4707 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4709 struct sk_buff *skb = napi->skb;
4710 const struct ethhdr *eth;
4711 unsigned int hlen = sizeof(*eth);
4715 skb_reset_mac_header(skb);
4716 skb_gro_reset_offset(skb);
4718 eth = skb_gro_header_fast(skb, 0);
4719 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4720 eth = skb_gro_header_slow(skb, hlen, 0);
4721 if (unlikely(!eth)) {
4722 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
4723 __func__, napi->dev->name);
4724 napi_reuse_skb(napi, skb);
4728 gro_pull_from_frag0(skb, hlen);
4729 NAPI_GRO_CB(skb)->frag0 += hlen;
4730 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4732 __skb_pull(skb, hlen);
4735 * This works because the only protocols we care about don't require
4737 * We'll fix it up properly in napi_frags_finish()
4739 skb->protocol = eth->h_proto;
4744 gro_result_t napi_gro_frags(struct napi_struct *napi)
4746 struct sk_buff *skb = napi_frags_skb(napi);
4751 trace_napi_gro_frags_entry(skb);
4753 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4755 EXPORT_SYMBOL(napi_gro_frags);
4757 /* Compute the checksum from gro_offset and return the folded value
4758 * after adding in any pseudo checksum.
4760 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4765 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4767 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4768 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4770 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4771 !skb->csum_complete_sw)
4772 netdev_rx_csum_fault(skb->dev);
4775 NAPI_GRO_CB(skb)->csum = wsum;
4776 NAPI_GRO_CB(skb)->csum_valid = 1;
4780 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4783 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4784 * Note: called with local irq disabled, but exits with local irq enabled.
4786 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4789 struct softnet_data *remsd = sd->rps_ipi_list;
4792 sd->rps_ipi_list = NULL;
4796 /* Send pending IPI's to kick RPS processing on remote cpus. */
4798 struct softnet_data *next = remsd->rps_ipi_next;
4800 if (cpu_online(remsd->cpu))
4801 smp_call_function_single_async(remsd->cpu,
4810 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4813 return sd->rps_ipi_list != NULL;
4819 static int process_backlog(struct napi_struct *napi, int quota)
4821 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4825 /* Check if we have pending ipi, its better to send them now,
4826 * not waiting net_rx_action() end.
4828 if (sd_has_rps_ipi_waiting(sd)) {
4829 local_irq_disable();
4830 net_rps_action_and_irq_enable(sd);
4833 napi->weight = weight_p;
4835 struct sk_buff *skb;
4837 while ((skb = __skb_dequeue(&sd->process_queue))) {
4839 __netif_receive_skb(skb);
4841 input_queue_head_incr(sd);
4842 if (++work >= quota)
4847 local_irq_disable();
4849 if (skb_queue_empty(&sd->input_pkt_queue)) {
4851 * Inline a custom version of __napi_complete().
4852 * only current cpu owns and manipulates this napi,
4853 * and NAPI_STATE_SCHED is the only possible flag set
4855 * We can use a plain write instead of clear_bit(),
4856 * and we dont need an smp_mb() memory barrier.
4861 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4862 &sd->process_queue);
4872 * __napi_schedule - schedule for receive
4873 * @n: entry to schedule
4875 * The entry's receive function will be scheduled to run.
4876 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4878 void __napi_schedule(struct napi_struct *n)
4880 unsigned long flags;
4882 local_irq_save(flags);
4883 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4884 local_irq_restore(flags);
4886 EXPORT_SYMBOL(__napi_schedule);
4889 * __napi_schedule_irqoff - schedule for receive
4890 * @n: entry to schedule
4892 * Variant of __napi_schedule() assuming hard irqs are masked
4894 void __napi_schedule_irqoff(struct napi_struct *n)
4896 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4898 EXPORT_SYMBOL(__napi_schedule_irqoff);
4900 bool __napi_complete(struct napi_struct *n)
4902 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4904 /* Some drivers call us directly, instead of calling
4905 * napi_complete_done().
4907 if (unlikely(test_bit(NAPI_STATE_IN_BUSY_POLL, &n->state)))
4910 list_del_init(&n->poll_list);
4911 smp_mb__before_atomic();
4912 clear_bit(NAPI_STATE_SCHED, &n->state);
4915 EXPORT_SYMBOL(__napi_complete);
4917 bool napi_complete_done(struct napi_struct *n, int work_done)
4919 unsigned long flags;
4922 * 1) Don't let napi dequeue from the cpu poll list
4923 * just in case its running on a different cpu.
4924 * 2) If we are busy polling, do nothing here, we have
4925 * the guarantee we will be called later.
4927 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
4928 NAPIF_STATE_IN_BUSY_POLL)))
4932 unsigned long timeout = 0;
4935 timeout = n->dev->gro_flush_timeout;
4938 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4939 HRTIMER_MODE_REL_PINNED);
4941 napi_gro_flush(n, false);
4943 if (likely(list_empty(&n->poll_list))) {
4944 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4946 /* If n->poll_list is not empty, we need to mask irqs */
4947 local_irq_save(flags);
4949 local_irq_restore(flags);
4953 EXPORT_SYMBOL(napi_complete_done);
4955 /* must be called under rcu_read_lock(), as we dont take a reference */
4956 static struct napi_struct *napi_by_id(unsigned int napi_id)
4958 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4959 struct napi_struct *napi;
4961 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4962 if (napi->napi_id == napi_id)
4968 #if defined(CONFIG_NET_RX_BUSY_POLL)
4970 #define BUSY_POLL_BUDGET 8
4972 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
4976 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
4980 /* All we really want here is to re-enable device interrupts.
4981 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
4983 rc = napi->poll(napi, BUSY_POLL_BUDGET);
4984 netpoll_poll_unlock(have_poll_lock);
4985 if (rc == BUSY_POLL_BUDGET)
4986 __napi_schedule(napi);
4988 if (local_softirq_pending())
4992 bool sk_busy_loop(struct sock *sk, int nonblock)
4994 unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0;
4995 int (*napi_poll)(struct napi_struct *napi, int budget);
4996 int (*busy_poll)(struct napi_struct *dev);
4997 void *have_poll_lock = NULL;
4998 struct napi_struct *napi;
5007 napi = napi_by_id(sk->sk_napi_id);
5011 /* Note: ndo_busy_poll method is optional in linux-4.5 */
5012 busy_poll = napi->dev->netdev_ops->ndo_busy_poll;
5019 rc = busy_poll(napi);
5023 unsigned long val = READ_ONCE(napi->state);
5025 /* If multiple threads are competing for this napi,
5026 * we avoid dirtying napi->state as much as we can.
5028 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
5029 NAPIF_STATE_IN_BUSY_POLL))
5031 if (cmpxchg(&napi->state, val,
5032 val | NAPIF_STATE_IN_BUSY_POLL |
5033 NAPIF_STATE_SCHED) != val)
5035 have_poll_lock = netpoll_poll_lock(napi);
5036 napi_poll = napi->poll;
5038 rc = napi_poll(napi, BUSY_POLL_BUDGET);
5039 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
5042 __NET_ADD_STATS(sock_net(sk),
5043 LINUX_MIB_BUSYPOLLRXPACKETS, rc);
5046 if (rc == LL_FLUSH_FAILED)
5047 break; /* permanent failure */
5049 if (nonblock || !skb_queue_empty(&sk->sk_receive_queue) ||
5050 busy_loop_timeout(end_time))
5053 if (unlikely(need_resched())) {
5055 busy_poll_stop(napi, have_poll_lock);
5059 rc = !skb_queue_empty(&sk->sk_receive_queue);
5060 if (rc || busy_loop_timeout(end_time))
5067 busy_poll_stop(napi, have_poll_lock);
5069 rc = !skb_queue_empty(&sk->sk_receive_queue);
5074 EXPORT_SYMBOL(sk_busy_loop);
5076 #endif /* CONFIG_NET_RX_BUSY_POLL */
5078 static void napi_hash_add(struct napi_struct *napi)
5080 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
5081 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
5084 spin_lock(&napi_hash_lock);
5086 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */
5088 if (unlikely(++napi_gen_id < NR_CPUS + 1))
5089 napi_gen_id = NR_CPUS + 1;
5090 } while (napi_by_id(napi_gen_id));
5091 napi->napi_id = napi_gen_id;
5093 hlist_add_head_rcu(&napi->napi_hash_node,
5094 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
5096 spin_unlock(&napi_hash_lock);
5099 /* Warning : caller is responsible to make sure rcu grace period
5100 * is respected before freeing memory containing @napi
5102 bool napi_hash_del(struct napi_struct *napi)
5104 bool rcu_sync_needed = false;
5106 spin_lock(&napi_hash_lock);
5108 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
5109 rcu_sync_needed = true;
5110 hlist_del_rcu(&napi->napi_hash_node);
5112 spin_unlock(&napi_hash_lock);
5113 return rcu_sync_needed;
5115 EXPORT_SYMBOL_GPL(napi_hash_del);
5117 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
5119 struct napi_struct *napi;
5121 napi = container_of(timer, struct napi_struct, timer);
5123 napi_schedule(napi);
5125 return HRTIMER_NORESTART;
5128 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
5129 int (*poll)(struct napi_struct *, int), int weight)
5131 INIT_LIST_HEAD(&napi->poll_list);
5132 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
5133 napi->timer.function = napi_watchdog;
5134 napi->gro_count = 0;
5135 napi->gro_list = NULL;
5138 if (weight > NAPI_POLL_WEIGHT)
5139 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5141 napi->weight = weight;
5142 list_add(&napi->dev_list, &dev->napi_list);
5144 #ifdef CONFIG_NETPOLL
5145 napi->poll_owner = -1;
5147 set_bit(NAPI_STATE_SCHED, &napi->state);
5148 napi_hash_add(napi);
5150 EXPORT_SYMBOL(netif_napi_add);
5152 void napi_disable(struct napi_struct *n)
5155 set_bit(NAPI_STATE_DISABLE, &n->state);
5157 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
5159 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
5162 hrtimer_cancel(&n->timer);
5164 clear_bit(NAPI_STATE_DISABLE, &n->state);
5166 EXPORT_SYMBOL(napi_disable);
5168 /* Must be called in process context */
5169 void netif_napi_del(struct napi_struct *napi)
5172 if (napi_hash_del(napi))
5174 list_del_init(&napi->dev_list);
5175 napi_free_frags(napi);
5177 kfree_skb_list(napi->gro_list);
5178 napi->gro_list = NULL;
5179 napi->gro_count = 0;
5181 EXPORT_SYMBOL(netif_napi_del);
5183 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
5188 list_del_init(&n->poll_list);
5190 have = netpoll_poll_lock(n);
5194 /* This NAPI_STATE_SCHED test is for avoiding a race
5195 * with netpoll's poll_napi(). Only the entity which
5196 * obtains the lock and sees NAPI_STATE_SCHED set will
5197 * actually make the ->poll() call. Therefore we avoid
5198 * accidentally calling ->poll() when NAPI is not scheduled.
5201 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
5202 work = n->poll(n, weight);
5203 trace_napi_poll(n, work, weight);
5206 WARN_ON_ONCE(work > weight);
5208 if (likely(work < weight))
5211 /* Drivers must not modify the NAPI state if they
5212 * consume the entire weight. In such cases this code
5213 * still "owns" the NAPI instance and therefore can
5214 * move the instance around on the list at-will.
5216 if (unlikely(napi_disable_pending(n))) {
5222 /* flush too old packets
5223 * If HZ < 1000, flush all packets.
5225 napi_gro_flush(n, HZ >= 1000);
5228 /* Some drivers may have called napi_schedule
5229 * prior to exhausting their budget.
5231 if (unlikely(!list_empty(&n->poll_list))) {
5232 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5233 n->dev ? n->dev->name : "backlog");
5237 list_add_tail(&n->poll_list, repoll);
5240 netpoll_poll_unlock(have);
5245 static __latent_entropy void net_rx_action(struct softirq_action *h)
5247 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5248 unsigned long time_limit = jiffies + 2;
5249 int budget = netdev_budget;
5253 local_irq_disable();
5254 list_splice_init(&sd->poll_list, &list);
5258 struct napi_struct *n;
5260 if (list_empty(&list)) {
5261 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
5266 n = list_first_entry(&list, struct napi_struct, poll_list);
5267 budget -= napi_poll(n, &repoll);
5269 /* If softirq window is exhausted then punt.
5270 * Allow this to run for 2 jiffies since which will allow
5271 * an average latency of 1.5/HZ.
5273 if (unlikely(budget <= 0 ||
5274 time_after_eq(jiffies, time_limit))) {
5280 local_irq_disable();
5282 list_splice_tail_init(&sd->poll_list, &list);
5283 list_splice_tail(&repoll, &list);
5284 list_splice(&list, &sd->poll_list);
5285 if (!list_empty(&sd->poll_list))
5286 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5288 net_rps_action_and_irq_enable(sd);
5290 __kfree_skb_flush();
5293 struct netdev_adjacent {
5294 struct net_device *dev;
5296 /* upper master flag, there can only be one master device per list */
5299 /* counter for the number of times this device was added to us */
5302 /* private field for the users */
5305 struct list_head list;
5306 struct rcu_head rcu;
5309 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
5310 struct list_head *adj_list)
5312 struct netdev_adjacent *adj;
5314 list_for_each_entry(adj, adj_list, list) {
5315 if (adj->dev == adj_dev)
5321 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
5323 struct net_device *dev = data;
5325 return upper_dev == dev;
5329 * netdev_has_upper_dev - Check if device is linked to an upper device
5331 * @upper_dev: upper device to check
5333 * Find out if a device is linked to specified upper device and return true
5334 * in case it is. Note that this checks only immediate upper device,
5335 * not through a complete stack of devices. The caller must hold the RTNL lock.
5337 bool netdev_has_upper_dev(struct net_device *dev,
5338 struct net_device *upper_dev)
5342 return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5345 EXPORT_SYMBOL(netdev_has_upper_dev);
5348 * netdev_has_upper_dev_all - Check if device is linked to an upper device
5350 * @upper_dev: upper device to check
5352 * Find out if a device is linked to specified upper device and return true
5353 * in case it is. Note that this checks the entire upper device chain.
5354 * The caller must hold rcu lock.
5357 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
5358 struct net_device *upper_dev)
5360 return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5363 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
5366 * netdev_has_any_upper_dev - Check if device is linked to some device
5369 * Find out if a device is linked to an upper device and return true in case
5370 * it is. The caller must hold the RTNL lock.
5372 static bool netdev_has_any_upper_dev(struct net_device *dev)
5376 return !list_empty(&dev->adj_list.upper);
5380 * netdev_master_upper_dev_get - Get master upper device
5383 * Find a master upper device and return pointer to it or NULL in case
5384 * it's not there. The caller must hold the RTNL lock.
5386 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5388 struct netdev_adjacent *upper;
5392 if (list_empty(&dev->adj_list.upper))
5395 upper = list_first_entry(&dev->adj_list.upper,
5396 struct netdev_adjacent, list);
5397 if (likely(upper->master))
5401 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5404 * netdev_has_any_lower_dev - Check if device is linked to some device
5407 * Find out if a device is linked to a lower device and return true in case
5408 * it is. The caller must hold the RTNL lock.
5410 static bool netdev_has_any_lower_dev(struct net_device *dev)
5414 return !list_empty(&dev->adj_list.lower);
5417 void *netdev_adjacent_get_private(struct list_head *adj_list)
5419 struct netdev_adjacent *adj;
5421 adj = list_entry(adj_list, struct netdev_adjacent, list);
5423 return adj->private;
5425 EXPORT_SYMBOL(netdev_adjacent_get_private);
5428 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5430 * @iter: list_head ** of the current position
5432 * Gets the next device from the dev's upper list, starting from iter
5433 * position. The caller must hold RCU read lock.
5435 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5436 struct list_head **iter)
5438 struct netdev_adjacent *upper;
5440 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5442 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5444 if (&upper->list == &dev->adj_list.upper)
5447 *iter = &upper->list;
5451 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5453 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
5454 struct list_head **iter)
5456 struct netdev_adjacent *upper;
5458 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5460 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5462 if (&upper->list == &dev->adj_list.upper)
5465 *iter = &upper->list;
5470 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
5471 int (*fn)(struct net_device *dev,
5475 struct net_device *udev;
5476 struct list_head *iter;
5479 for (iter = &dev->adj_list.upper,
5480 udev = netdev_next_upper_dev_rcu(dev, &iter);
5482 udev = netdev_next_upper_dev_rcu(dev, &iter)) {
5483 /* first is the upper device itself */
5484 ret = fn(udev, data);
5488 /* then look at all of its upper devices */
5489 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data);
5496 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
5499 * netdev_lower_get_next_private - Get the next ->private from the
5500 * lower neighbour list
5502 * @iter: list_head ** of the current position
5504 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5505 * list, starting from iter position. The caller must hold either hold the
5506 * RTNL lock or its own locking that guarantees that the neighbour lower
5507 * list will remain unchanged.
5509 void *netdev_lower_get_next_private(struct net_device *dev,
5510 struct list_head **iter)
5512 struct netdev_adjacent *lower;
5514 lower = list_entry(*iter, struct netdev_adjacent, list);
5516 if (&lower->list == &dev->adj_list.lower)
5519 *iter = lower->list.next;
5521 return lower->private;
5523 EXPORT_SYMBOL(netdev_lower_get_next_private);
5526 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5527 * lower neighbour list, RCU
5530 * @iter: list_head ** of the current position
5532 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5533 * list, starting from iter position. The caller must hold RCU read lock.
5535 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5536 struct list_head **iter)
5538 struct netdev_adjacent *lower;
5540 WARN_ON_ONCE(!rcu_read_lock_held());
5542 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5544 if (&lower->list == &dev->adj_list.lower)
5547 *iter = &lower->list;
5549 return lower->private;
5551 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5554 * netdev_lower_get_next - Get the next device from the lower neighbour
5557 * @iter: list_head ** of the current position
5559 * Gets the next netdev_adjacent from the dev's lower neighbour
5560 * list, starting from iter position. The caller must hold RTNL lock or
5561 * its own locking that guarantees that the neighbour lower
5562 * list will remain unchanged.
5564 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5566 struct netdev_adjacent *lower;
5568 lower = list_entry(*iter, struct netdev_adjacent, list);
5570 if (&lower->list == &dev->adj_list.lower)
5573 *iter = lower->list.next;
5577 EXPORT_SYMBOL(netdev_lower_get_next);
5579 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
5580 struct list_head **iter)
5582 struct netdev_adjacent *lower;
5584 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5586 if (&lower->list == &dev->adj_list.lower)
5589 *iter = &lower->list;
5594 int netdev_walk_all_lower_dev(struct net_device *dev,
5595 int (*fn)(struct net_device *dev,
5599 struct net_device *ldev;
5600 struct list_head *iter;
5603 for (iter = &dev->adj_list.lower,
5604 ldev = netdev_next_lower_dev(dev, &iter);
5606 ldev = netdev_next_lower_dev(dev, &iter)) {
5607 /* first is the lower device itself */
5608 ret = fn(ldev, data);
5612 /* then look at all of its lower devices */
5613 ret = netdev_walk_all_lower_dev(ldev, fn, data);
5620 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
5622 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
5623 struct list_head **iter)
5625 struct netdev_adjacent *lower;
5627 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5628 if (&lower->list == &dev->adj_list.lower)
5631 *iter = &lower->list;
5636 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
5637 int (*fn)(struct net_device *dev,
5641 struct net_device *ldev;
5642 struct list_head *iter;
5645 for (iter = &dev->adj_list.lower,
5646 ldev = netdev_next_lower_dev_rcu(dev, &iter);
5648 ldev = netdev_next_lower_dev_rcu(dev, &iter)) {
5649 /* first is the lower device itself */
5650 ret = fn(ldev, data);
5654 /* then look at all of its lower devices */
5655 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data);
5662 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
5665 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5666 * lower neighbour list, RCU
5670 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5671 * list. The caller must hold RCU read lock.
5673 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5675 struct netdev_adjacent *lower;
5677 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5678 struct netdev_adjacent, list);
5680 return lower->private;
5683 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5686 * netdev_master_upper_dev_get_rcu - Get master upper device
5689 * Find a master upper device and return pointer to it or NULL in case
5690 * it's not there. The caller must hold the RCU read lock.
5692 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5694 struct netdev_adjacent *upper;
5696 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5697 struct netdev_adjacent, list);
5698 if (upper && likely(upper->master))
5702 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5704 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5705 struct net_device *adj_dev,
5706 struct list_head *dev_list)
5708 char linkname[IFNAMSIZ+7];
5709 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5710 "upper_%s" : "lower_%s", adj_dev->name);
5711 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5714 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5716 struct list_head *dev_list)
5718 char linkname[IFNAMSIZ+7];
5719 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5720 "upper_%s" : "lower_%s", name);
5721 sysfs_remove_link(&(dev->dev.kobj), linkname);
5724 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5725 struct net_device *adj_dev,
5726 struct list_head *dev_list)
5728 return (dev_list == &dev->adj_list.upper ||
5729 dev_list == &dev->adj_list.lower) &&
5730 net_eq(dev_net(dev), dev_net(adj_dev));
5733 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5734 struct net_device *adj_dev,
5735 struct list_head *dev_list,
5736 void *private, bool master)
5738 struct netdev_adjacent *adj;
5741 adj = __netdev_find_adj(adj_dev, dev_list);
5745 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
5746 dev->name, adj_dev->name, adj->ref_nr);
5751 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5756 adj->master = master;
5758 adj->private = private;
5761 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
5762 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
5764 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5765 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5770 /* Ensure that master link is always the first item in list. */
5772 ret = sysfs_create_link(&(dev->dev.kobj),
5773 &(adj_dev->dev.kobj), "master");
5775 goto remove_symlinks;
5777 list_add_rcu(&adj->list, dev_list);
5779 list_add_tail_rcu(&adj->list, dev_list);
5785 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5786 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5794 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5795 struct net_device *adj_dev,
5797 struct list_head *dev_list)
5799 struct netdev_adjacent *adj;
5801 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
5802 dev->name, adj_dev->name, ref_nr);
5804 adj = __netdev_find_adj(adj_dev, dev_list);
5807 pr_err("Adjacency does not exist for device %s from %s\n",
5808 dev->name, adj_dev->name);
5813 if (adj->ref_nr > ref_nr) {
5814 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
5815 dev->name, adj_dev->name, ref_nr,
5816 adj->ref_nr - ref_nr);
5817 adj->ref_nr -= ref_nr;
5822 sysfs_remove_link(&(dev->dev.kobj), "master");
5824 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5825 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5827 list_del_rcu(&adj->list);
5828 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
5829 adj_dev->name, dev->name, adj_dev->name);
5831 kfree_rcu(adj, rcu);
5834 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5835 struct net_device *upper_dev,
5836 struct list_head *up_list,
5837 struct list_head *down_list,
5838 void *private, bool master)
5842 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
5847 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
5850 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
5857 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5858 struct net_device *upper_dev,
5860 struct list_head *up_list,
5861 struct list_head *down_list)
5863 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
5864 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
5867 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5868 struct net_device *upper_dev,
5869 void *private, bool master)
5871 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5872 &dev->adj_list.upper,
5873 &upper_dev->adj_list.lower,
5877 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5878 struct net_device *upper_dev)
5880 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
5881 &dev->adj_list.upper,
5882 &upper_dev->adj_list.lower);
5885 static int __netdev_upper_dev_link(struct net_device *dev,
5886 struct net_device *upper_dev, bool master,
5887 void *upper_priv, void *upper_info)
5889 struct netdev_notifier_changeupper_info changeupper_info;
5894 if (dev == upper_dev)
5897 /* To prevent loops, check if dev is not upper device to upper_dev. */
5898 if (netdev_has_upper_dev(upper_dev, dev))
5901 if (netdev_has_upper_dev(dev, upper_dev))
5904 if (master && netdev_master_upper_dev_get(dev))
5907 changeupper_info.upper_dev = upper_dev;
5908 changeupper_info.master = master;
5909 changeupper_info.linking = true;
5910 changeupper_info.upper_info = upper_info;
5912 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5913 &changeupper_info.info);
5914 ret = notifier_to_errno(ret);
5918 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
5923 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5924 &changeupper_info.info);
5925 ret = notifier_to_errno(ret);
5932 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5938 * netdev_upper_dev_link - Add a link to the upper device
5940 * @upper_dev: new upper device
5942 * Adds a link to device which is upper to this one. The caller must hold
5943 * the RTNL lock. On a failure a negative errno code is returned.
5944 * On success the reference counts are adjusted and the function
5947 int netdev_upper_dev_link(struct net_device *dev,
5948 struct net_device *upper_dev)
5950 return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL);
5952 EXPORT_SYMBOL(netdev_upper_dev_link);
5955 * netdev_master_upper_dev_link - Add a master link to the upper device
5957 * @upper_dev: new upper device
5958 * @upper_priv: upper device private
5959 * @upper_info: upper info to be passed down via notifier
5961 * Adds a link to device which is upper to this one. In this case, only
5962 * one master upper device can be linked, although other non-master devices
5963 * might be linked as well. The caller must hold the RTNL lock.
5964 * On a failure a negative errno code is returned. On success the reference
5965 * counts are adjusted and the function returns zero.
5967 int netdev_master_upper_dev_link(struct net_device *dev,
5968 struct net_device *upper_dev,
5969 void *upper_priv, void *upper_info)
5971 return __netdev_upper_dev_link(dev, upper_dev, true,
5972 upper_priv, upper_info);
5974 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5977 * netdev_upper_dev_unlink - Removes a link to upper device
5979 * @upper_dev: new upper device
5981 * Removes a link to device which is upper to this one. The caller must hold
5984 void netdev_upper_dev_unlink(struct net_device *dev,
5985 struct net_device *upper_dev)
5987 struct netdev_notifier_changeupper_info changeupper_info;
5990 changeupper_info.upper_dev = upper_dev;
5991 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
5992 changeupper_info.linking = false;
5994 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5995 &changeupper_info.info);
5997 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5999 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
6000 &changeupper_info.info);
6002 EXPORT_SYMBOL(netdev_upper_dev_unlink);
6005 * netdev_bonding_info_change - Dispatch event about slave change
6007 * @bonding_info: info to dispatch
6009 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
6010 * The caller must hold the RTNL lock.
6012 void netdev_bonding_info_change(struct net_device *dev,
6013 struct netdev_bonding_info *bonding_info)
6015 struct netdev_notifier_bonding_info info;
6017 memcpy(&info.bonding_info, bonding_info,
6018 sizeof(struct netdev_bonding_info));
6019 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
6022 EXPORT_SYMBOL(netdev_bonding_info_change);
6024 static void netdev_adjacent_add_links(struct net_device *dev)
6026 struct netdev_adjacent *iter;
6028 struct net *net = dev_net(dev);
6030 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6031 if (!net_eq(net, dev_net(iter->dev)))
6033 netdev_adjacent_sysfs_add(iter->dev, dev,
6034 &iter->dev->adj_list.lower);
6035 netdev_adjacent_sysfs_add(dev, iter->dev,
6036 &dev->adj_list.upper);
6039 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6040 if (!net_eq(net, dev_net(iter->dev)))
6042 netdev_adjacent_sysfs_add(iter->dev, dev,
6043 &iter->dev->adj_list.upper);
6044 netdev_adjacent_sysfs_add(dev, iter->dev,
6045 &dev->adj_list.lower);
6049 static void netdev_adjacent_del_links(struct net_device *dev)
6051 struct netdev_adjacent *iter;
6053 struct net *net = dev_net(dev);
6055 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6056 if (!net_eq(net, dev_net(iter->dev)))
6058 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6059 &iter->dev->adj_list.lower);
6060 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6061 &dev->adj_list.upper);
6064 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6065 if (!net_eq(net, dev_net(iter->dev)))
6067 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6068 &iter->dev->adj_list.upper);
6069 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6070 &dev->adj_list.lower);
6074 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
6076 struct netdev_adjacent *iter;
6078 struct net *net = dev_net(dev);
6080 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6081 if (!net_eq(net, dev_net(iter->dev)))
6083 netdev_adjacent_sysfs_del(iter->dev, oldname,
6084 &iter->dev->adj_list.lower);
6085 netdev_adjacent_sysfs_add(iter->dev, dev,
6086 &iter->dev->adj_list.lower);
6089 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6090 if (!net_eq(net, dev_net(iter->dev)))
6092 netdev_adjacent_sysfs_del(iter->dev, oldname,
6093 &iter->dev->adj_list.upper);
6094 netdev_adjacent_sysfs_add(iter->dev, dev,
6095 &iter->dev->adj_list.upper);
6099 void *netdev_lower_dev_get_private(struct net_device *dev,
6100 struct net_device *lower_dev)
6102 struct netdev_adjacent *lower;
6106 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
6110 return lower->private;
6112 EXPORT_SYMBOL(netdev_lower_dev_get_private);
6115 int dev_get_nest_level(struct net_device *dev)
6117 struct net_device *lower = NULL;
6118 struct list_head *iter;
6124 netdev_for_each_lower_dev(dev, lower, iter) {
6125 nest = dev_get_nest_level(lower);
6126 if (max_nest < nest)
6130 return max_nest + 1;
6132 EXPORT_SYMBOL(dev_get_nest_level);
6135 * netdev_lower_change - Dispatch event about lower device state change
6136 * @lower_dev: device
6137 * @lower_state_info: state to dispatch
6139 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
6140 * The caller must hold the RTNL lock.
6142 void netdev_lower_state_changed(struct net_device *lower_dev,
6143 void *lower_state_info)
6145 struct netdev_notifier_changelowerstate_info changelowerstate_info;
6148 changelowerstate_info.lower_state_info = lower_state_info;
6149 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev,
6150 &changelowerstate_info.info);
6152 EXPORT_SYMBOL(netdev_lower_state_changed);
6154 int netdev_default_l2upper_neigh_construct(struct net_device *dev,
6155 struct neighbour *n)
6157 struct net_device *lower_dev, *stop_dev;
6158 struct list_head *iter;
6161 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6162 if (!lower_dev->netdev_ops->ndo_neigh_construct)
6164 err = lower_dev->netdev_ops->ndo_neigh_construct(lower_dev, n);
6166 stop_dev = lower_dev;
6173 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6174 if (lower_dev == stop_dev)
6176 if (!lower_dev->netdev_ops->ndo_neigh_destroy)
6178 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n);
6182 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_construct);
6184 void netdev_default_l2upper_neigh_destroy(struct net_device *dev,
6185 struct neighbour *n)
6187 struct net_device *lower_dev;
6188 struct list_head *iter;
6190 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6191 if (!lower_dev->netdev_ops->ndo_neigh_destroy)
6193 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n);
6196 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_destroy);
6198 static void dev_change_rx_flags(struct net_device *dev, int flags)
6200 const struct net_device_ops *ops = dev->netdev_ops;
6202 if (ops->ndo_change_rx_flags)
6203 ops->ndo_change_rx_flags(dev, flags);
6206 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
6208 unsigned int old_flags = dev->flags;
6214 dev->flags |= IFF_PROMISC;
6215 dev->promiscuity += inc;
6216 if (dev->promiscuity == 0) {
6219 * If inc causes overflow, untouch promisc and return error.
6222 dev->flags &= ~IFF_PROMISC;
6224 dev->promiscuity -= inc;
6225 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6230 if (dev->flags != old_flags) {
6231 pr_info("device %s %s promiscuous mode\n",
6233 dev->flags & IFF_PROMISC ? "entered" : "left");
6234 if (audit_enabled) {
6235 current_uid_gid(&uid, &gid);
6236 audit_log(current->audit_context, GFP_ATOMIC,
6237 AUDIT_ANOM_PROMISCUOUS,
6238 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6239 dev->name, (dev->flags & IFF_PROMISC),
6240 (old_flags & IFF_PROMISC),
6241 from_kuid(&init_user_ns, audit_get_loginuid(current)),
6242 from_kuid(&init_user_ns, uid),
6243 from_kgid(&init_user_ns, gid),
6244 audit_get_sessionid(current));
6247 dev_change_rx_flags(dev, IFF_PROMISC);
6250 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
6255 * dev_set_promiscuity - update promiscuity count on a device
6259 * Add or remove promiscuity from a device. While the count in the device
6260 * remains above zero the interface remains promiscuous. Once it hits zero
6261 * the device reverts back to normal filtering operation. A negative inc
6262 * value is used to drop promiscuity on the device.
6263 * Return 0 if successful or a negative errno code on error.
6265 int dev_set_promiscuity(struct net_device *dev, int inc)
6267 unsigned int old_flags = dev->flags;
6270 err = __dev_set_promiscuity(dev, inc, true);
6273 if (dev->flags != old_flags)
6274 dev_set_rx_mode(dev);
6277 EXPORT_SYMBOL(dev_set_promiscuity);
6279 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
6281 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
6285 dev->flags |= IFF_ALLMULTI;
6286 dev->allmulti += inc;
6287 if (dev->allmulti == 0) {
6290 * If inc causes overflow, untouch allmulti and return error.
6293 dev->flags &= ~IFF_ALLMULTI;
6295 dev->allmulti -= inc;
6296 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6301 if (dev->flags ^ old_flags) {
6302 dev_change_rx_flags(dev, IFF_ALLMULTI);
6303 dev_set_rx_mode(dev);
6305 __dev_notify_flags(dev, old_flags,
6306 dev->gflags ^ old_gflags);
6312 * dev_set_allmulti - update allmulti count on a device
6316 * Add or remove reception of all multicast frames to a device. While the
6317 * count in the device remains above zero the interface remains listening
6318 * to all interfaces. Once it hits zero the device reverts back to normal
6319 * filtering operation. A negative @inc value is used to drop the counter
6320 * when releasing a resource needing all multicasts.
6321 * Return 0 if successful or a negative errno code on error.
6324 int dev_set_allmulti(struct net_device *dev, int inc)
6326 return __dev_set_allmulti(dev, inc, true);
6328 EXPORT_SYMBOL(dev_set_allmulti);
6331 * Upload unicast and multicast address lists to device and
6332 * configure RX filtering. When the device doesn't support unicast
6333 * filtering it is put in promiscuous mode while unicast addresses
6336 void __dev_set_rx_mode(struct net_device *dev)
6338 const struct net_device_ops *ops = dev->netdev_ops;
6340 /* dev_open will call this function so the list will stay sane. */
6341 if (!(dev->flags&IFF_UP))
6344 if (!netif_device_present(dev))
6347 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
6348 /* Unicast addresses changes may only happen under the rtnl,
6349 * therefore calling __dev_set_promiscuity here is safe.
6351 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
6352 __dev_set_promiscuity(dev, 1, false);
6353 dev->uc_promisc = true;
6354 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
6355 __dev_set_promiscuity(dev, -1, false);
6356 dev->uc_promisc = false;
6360 if (ops->ndo_set_rx_mode)
6361 ops->ndo_set_rx_mode(dev);
6364 void dev_set_rx_mode(struct net_device *dev)
6366 netif_addr_lock_bh(dev);
6367 __dev_set_rx_mode(dev);
6368 netif_addr_unlock_bh(dev);
6372 * dev_get_flags - get flags reported to userspace
6375 * Get the combination of flag bits exported through APIs to userspace.
6377 unsigned int dev_get_flags(const struct net_device *dev)
6381 flags = (dev->flags & ~(IFF_PROMISC |
6386 (dev->gflags & (IFF_PROMISC |
6389 if (netif_running(dev)) {
6390 if (netif_oper_up(dev))
6391 flags |= IFF_RUNNING;
6392 if (netif_carrier_ok(dev))
6393 flags |= IFF_LOWER_UP;
6394 if (netif_dormant(dev))
6395 flags |= IFF_DORMANT;
6400 EXPORT_SYMBOL(dev_get_flags);
6402 int __dev_change_flags(struct net_device *dev, unsigned int flags)
6404 unsigned int old_flags = dev->flags;
6410 * Set the flags on our device.
6413 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
6414 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
6416 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6420 * Load in the correct multicast list now the flags have changed.
6423 if ((old_flags ^ flags) & IFF_MULTICAST)
6424 dev_change_rx_flags(dev, IFF_MULTICAST);
6426 dev_set_rx_mode(dev);
6429 * Have we downed the interface. We handle IFF_UP ourselves
6430 * according to user attempts to set it, rather than blindly
6435 if ((old_flags ^ flags) & IFF_UP)
6436 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
6438 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6439 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6440 unsigned int old_flags = dev->flags;
6442 dev->gflags ^= IFF_PROMISC;
6444 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6445 if (dev->flags != old_flags)
6446 dev_set_rx_mode(dev);
6449 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6450 is important. Some (broken) drivers set IFF_PROMISC, when
6451 IFF_ALLMULTI is requested not asking us and not reporting.
6453 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6454 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6456 dev->gflags ^= IFF_ALLMULTI;
6457 __dev_set_allmulti(dev, inc, false);
6463 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6464 unsigned int gchanges)
6466 unsigned int changes = dev->flags ^ old_flags;
6469 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6471 if (changes & IFF_UP) {
6472 if (dev->flags & IFF_UP)
6473 call_netdevice_notifiers(NETDEV_UP, dev);
6475 call_netdevice_notifiers(NETDEV_DOWN, dev);
6478 if (dev->flags & IFF_UP &&
6479 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6480 struct netdev_notifier_change_info change_info;
6482 change_info.flags_changed = changes;
6483 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6489 * dev_change_flags - change device settings
6491 * @flags: device state flags
6493 * Change settings on device based state flags. The flags are
6494 * in the userspace exported format.
6496 int dev_change_flags(struct net_device *dev, unsigned int flags)
6499 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6501 ret = __dev_change_flags(dev, flags);
6505 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6506 __dev_notify_flags(dev, old_flags, changes);
6509 EXPORT_SYMBOL(dev_change_flags);
6511 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6513 const struct net_device_ops *ops = dev->netdev_ops;
6515 if (ops->ndo_change_mtu)
6516 return ops->ndo_change_mtu(dev, new_mtu);
6523 * dev_set_mtu - Change maximum transfer unit
6525 * @new_mtu: new transfer unit
6527 * Change the maximum transfer size of the network device.
6529 int dev_set_mtu(struct net_device *dev, int new_mtu)
6533 if (new_mtu == dev->mtu)
6536 /* MTU must be positive, and in range */
6537 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
6538 net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n",
6539 dev->name, new_mtu, dev->min_mtu);
6543 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
6544 net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n",
6545 dev->name, new_mtu, dev->max_mtu);
6549 if (!netif_device_present(dev))
6552 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6553 err = notifier_to_errno(err);
6557 orig_mtu = dev->mtu;
6558 err = __dev_set_mtu(dev, new_mtu);
6561 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6562 err = notifier_to_errno(err);
6564 /* setting mtu back and notifying everyone again,
6565 * so that they have a chance to revert changes.
6567 __dev_set_mtu(dev, orig_mtu);
6568 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6573 EXPORT_SYMBOL(dev_set_mtu);
6576 * dev_set_group - Change group this device belongs to
6578 * @new_group: group this device should belong to
6580 void dev_set_group(struct net_device *dev, int new_group)
6582 dev->group = new_group;
6584 EXPORT_SYMBOL(dev_set_group);
6587 * dev_set_mac_address - Change Media Access Control Address
6591 * Change the hardware (MAC) address of the device
6593 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6595 const struct net_device_ops *ops = dev->netdev_ops;
6598 if (!ops->ndo_set_mac_address)
6600 if (sa->sa_family != dev->type)
6602 if (!netif_device_present(dev))
6604 err = ops->ndo_set_mac_address(dev, sa);
6607 dev->addr_assign_type = NET_ADDR_SET;
6608 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6609 add_device_randomness(dev->dev_addr, dev->addr_len);
6612 EXPORT_SYMBOL(dev_set_mac_address);
6615 * dev_change_carrier - Change device carrier
6617 * @new_carrier: new value
6619 * Change device carrier
6621 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6623 const struct net_device_ops *ops = dev->netdev_ops;
6625 if (!ops->ndo_change_carrier)
6627 if (!netif_device_present(dev))
6629 return ops->ndo_change_carrier(dev, new_carrier);
6631 EXPORT_SYMBOL(dev_change_carrier);
6634 * dev_get_phys_port_id - Get device physical port ID
6638 * Get device physical port ID
6640 int dev_get_phys_port_id(struct net_device *dev,
6641 struct netdev_phys_item_id *ppid)
6643 const struct net_device_ops *ops = dev->netdev_ops;
6645 if (!ops->ndo_get_phys_port_id)
6647 return ops->ndo_get_phys_port_id(dev, ppid);
6649 EXPORT_SYMBOL(dev_get_phys_port_id);
6652 * dev_get_phys_port_name - Get device physical port name
6655 * @len: limit of bytes to copy to name
6657 * Get device physical port name
6659 int dev_get_phys_port_name(struct net_device *dev,
6660 char *name, size_t len)
6662 const struct net_device_ops *ops = dev->netdev_ops;
6664 if (!ops->ndo_get_phys_port_name)
6666 return ops->ndo_get_phys_port_name(dev, name, len);
6668 EXPORT_SYMBOL(dev_get_phys_port_name);
6671 * dev_change_proto_down - update protocol port state information
6673 * @proto_down: new value
6675 * This info can be used by switch drivers to set the phys state of the
6678 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6680 const struct net_device_ops *ops = dev->netdev_ops;
6682 if (!ops->ndo_change_proto_down)
6684 if (!netif_device_present(dev))
6686 return ops->ndo_change_proto_down(dev, proto_down);
6688 EXPORT_SYMBOL(dev_change_proto_down);
6691 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
6693 * @fd: new program fd or negative value to clear
6694 * @flags: xdp-related flags
6696 * Set or clear a bpf program for a device
6698 int dev_change_xdp_fd(struct net_device *dev, int fd, u32 flags)
6700 const struct net_device_ops *ops = dev->netdev_ops;
6701 struct bpf_prog *prog = NULL;
6702 struct netdev_xdp xdp;
6710 if (flags & XDP_FLAGS_UPDATE_IF_NOEXIST) {
6711 memset(&xdp, 0, sizeof(xdp));
6712 xdp.command = XDP_QUERY_PROG;
6714 err = ops->ndo_xdp(dev, &xdp);
6717 if (xdp.prog_attached)
6721 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP);
6723 return PTR_ERR(prog);
6726 memset(&xdp, 0, sizeof(xdp));
6727 xdp.command = XDP_SETUP_PROG;
6730 err = ops->ndo_xdp(dev, &xdp);
6731 if (err < 0 && prog)
6736 EXPORT_SYMBOL(dev_change_xdp_fd);
6739 * dev_new_index - allocate an ifindex
6740 * @net: the applicable net namespace
6742 * Returns a suitable unique value for a new device interface
6743 * number. The caller must hold the rtnl semaphore or the
6744 * dev_base_lock to be sure it remains unique.
6746 static int dev_new_index(struct net *net)
6748 int ifindex = net->ifindex;
6752 if (!__dev_get_by_index(net, ifindex))
6753 return net->ifindex = ifindex;
6757 /* Delayed registration/unregisteration */
6758 static LIST_HEAD(net_todo_list);
6759 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6761 static void net_set_todo(struct net_device *dev)
6763 list_add_tail(&dev->todo_list, &net_todo_list);
6764 dev_net(dev)->dev_unreg_count++;
6767 static void rollback_registered_many(struct list_head *head)
6769 struct net_device *dev, *tmp;
6770 LIST_HEAD(close_head);
6772 BUG_ON(dev_boot_phase);
6775 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6776 /* Some devices call without registering
6777 * for initialization unwind. Remove those
6778 * devices and proceed with the remaining.
6780 if (dev->reg_state == NETREG_UNINITIALIZED) {
6781 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6785 list_del(&dev->unreg_list);
6788 dev->dismantle = true;
6789 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6792 /* If device is running, close it first. */
6793 list_for_each_entry(dev, head, unreg_list)
6794 list_add_tail(&dev->close_list, &close_head);
6795 dev_close_many(&close_head, true);
6797 list_for_each_entry(dev, head, unreg_list) {
6798 /* And unlink it from device chain. */
6799 unlist_netdevice(dev);
6801 dev->reg_state = NETREG_UNREGISTERING;
6803 flush_all_backlogs();
6807 list_for_each_entry(dev, head, unreg_list) {
6808 struct sk_buff *skb = NULL;
6810 /* Shutdown queueing discipline. */
6814 /* Notify protocols, that we are about to destroy
6815 this device. They should clean all the things.
6817 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6819 if (!dev->rtnl_link_ops ||
6820 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6821 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6825 * Flush the unicast and multicast chains
6830 if (dev->netdev_ops->ndo_uninit)
6831 dev->netdev_ops->ndo_uninit(dev);
6834 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6836 /* Notifier chain MUST detach us all upper devices. */
6837 WARN_ON(netdev_has_any_upper_dev(dev));
6838 WARN_ON(netdev_has_any_lower_dev(dev));
6840 /* Remove entries from kobject tree */
6841 netdev_unregister_kobject(dev);
6843 /* Remove XPS queueing entries */
6844 netif_reset_xps_queues_gt(dev, 0);
6850 list_for_each_entry(dev, head, unreg_list)
6854 static void rollback_registered(struct net_device *dev)
6858 list_add(&dev->unreg_list, &single);
6859 rollback_registered_many(&single);
6863 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6864 struct net_device *upper, netdev_features_t features)
6866 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6867 netdev_features_t feature;
6870 for_each_netdev_feature(&upper_disables, feature_bit) {
6871 feature = __NETIF_F_BIT(feature_bit);
6872 if (!(upper->wanted_features & feature)
6873 && (features & feature)) {
6874 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6875 &feature, upper->name);
6876 features &= ~feature;
6883 static void netdev_sync_lower_features(struct net_device *upper,
6884 struct net_device *lower, netdev_features_t features)
6886 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6887 netdev_features_t feature;
6890 for_each_netdev_feature(&upper_disables, feature_bit) {
6891 feature = __NETIF_F_BIT(feature_bit);
6892 if (!(features & feature) && (lower->features & feature)) {
6893 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6894 &feature, lower->name);
6895 lower->wanted_features &= ~feature;
6896 netdev_update_features(lower);
6898 if (unlikely(lower->features & feature))
6899 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6900 &feature, lower->name);
6905 static netdev_features_t netdev_fix_features(struct net_device *dev,
6906 netdev_features_t features)
6908 /* Fix illegal checksum combinations */
6909 if ((features & NETIF_F_HW_CSUM) &&
6910 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6911 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6912 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6915 /* TSO requires that SG is present as well. */
6916 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6917 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6918 features &= ~NETIF_F_ALL_TSO;
6921 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6922 !(features & NETIF_F_IP_CSUM)) {
6923 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6924 features &= ~NETIF_F_TSO;
6925 features &= ~NETIF_F_TSO_ECN;
6928 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6929 !(features & NETIF_F_IPV6_CSUM)) {
6930 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6931 features &= ~NETIF_F_TSO6;
6934 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
6935 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
6936 features &= ~NETIF_F_TSO_MANGLEID;
6938 /* TSO ECN requires that TSO is present as well. */
6939 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6940 features &= ~NETIF_F_TSO_ECN;
6942 /* Software GSO depends on SG. */
6943 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6944 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6945 features &= ~NETIF_F_GSO;
6948 /* UFO needs SG and checksumming */
6949 if (features & NETIF_F_UFO) {
6950 /* maybe split UFO into V4 and V6? */
6951 if (!(features & NETIF_F_HW_CSUM) &&
6952 ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) !=
6953 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) {
6955 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6956 features &= ~NETIF_F_UFO;
6959 if (!(features & NETIF_F_SG)) {
6961 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6962 features &= ~NETIF_F_UFO;
6966 /* GSO partial features require GSO partial be set */
6967 if ((features & dev->gso_partial_features) &&
6968 !(features & NETIF_F_GSO_PARTIAL)) {
6970 "Dropping partially supported GSO features since no GSO partial.\n");
6971 features &= ~dev->gso_partial_features;
6974 #ifdef CONFIG_NET_RX_BUSY_POLL
6975 if (dev->netdev_ops->ndo_busy_poll)
6976 features |= NETIF_F_BUSY_POLL;
6979 features &= ~NETIF_F_BUSY_POLL;
6984 int __netdev_update_features(struct net_device *dev)
6986 struct net_device *upper, *lower;
6987 netdev_features_t features;
6988 struct list_head *iter;
6993 features = netdev_get_wanted_features(dev);
6995 if (dev->netdev_ops->ndo_fix_features)
6996 features = dev->netdev_ops->ndo_fix_features(dev, features);
6998 /* driver might be less strict about feature dependencies */
6999 features = netdev_fix_features(dev, features);
7001 /* some features can't be enabled if they're off an an upper device */
7002 netdev_for_each_upper_dev_rcu(dev, upper, iter)
7003 features = netdev_sync_upper_features(dev, upper, features);
7005 if (dev->features == features)
7008 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
7009 &dev->features, &features);
7011 if (dev->netdev_ops->ndo_set_features)
7012 err = dev->netdev_ops->ndo_set_features(dev, features);
7016 if (unlikely(err < 0)) {
7018 "set_features() failed (%d); wanted %pNF, left %pNF\n",
7019 err, &features, &dev->features);
7020 /* return non-0 since some features might have changed and
7021 * it's better to fire a spurious notification than miss it
7027 /* some features must be disabled on lower devices when disabled
7028 * on an upper device (think: bonding master or bridge)
7030 netdev_for_each_lower_dev(dev, lower, iter)
7031 netdev_sync_lower_features(dev, lower, features);
7034 dev->features = features;
7036 return err < 0 ? 0 : 1;
7040 * netdev_update_features - recalculate device features
7041 * @dev: the device to check
7043 * Recalculate dev->features set and send notifications if it
7044 * has changed. Should be called after driver or hardware dependent
7045 * conditions might have changed that influence the features.
7047 void netdev_update_features(struct net_device *dev)
7049 if (__netdev_update_features(dev))
7050 netdev_features_change(dev);
7052 EXPORT_SYMBOL(netdev_update_features);
7055 * netdev_change_features - recalculate device features
7056 * @dev: the device to check
7058 * Recalculate dev->features set and send notifications even
7059 * if they have not changed. Should be called instead of
7060 * netdev_update_features() if also dev->vlan_features might
7061 * have changed to allow the changes to be propagated to stacked
7064 void netdev_change_features(struct net_device *dev)
7066 __netdev_update_features(dev);
7067 netdev_features_change(dev);
7069 EXPORT_SYMBOL(netdev_change_features);
7072 * netif_stacked_transfer_operstate - transfer operstate
7073 * @rootdev: the root or lower level device to transfer state from
7074 * @dev: the device to transfer operstate to
7076 * Transfer operational state from root to device. This is normally
7077 * called when a stacking relationship exists between the root
7078 * device and the device(a leaf device).
7080 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
7081 struct net_device *dev)
7083 if (rootdev->operstate == IF_OPER_DORMANT)
7084 netif_dormant_on(dev);
7086 netif_dormant_off(dev);
7088 if (netif_carrier_ok(rootdev)) {
7089 if (!netif_carrier_ok(dev))
7090 netif_carrier_on(dev);
7092 if (netif_carrier_ok(dev))
7093 netif_carrier_off(dev);
7096 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
7099 static int netif_alloc_rx_queues(struct net_device *dev)
7101 unsigned int i, count = dev->num_rx_queues;
7102 struct netdev_rx_queue *rx;
7103 size_t sz = count * sizeof(*rx);
7107 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7115 for (i = 0; i < count; i++)
7121 static void netdev_init_one_queue(struct net_device *dev,
7122 struct netdev_queue *queue, void *_unused)
7124 /* Initialize queue lock */
7125 spin_lock_init(&queue->_xmit_lock);
7126 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
7127 queue->xmit_lock_owner = -1;
7128 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
7131 dql_init(&queue->dql, HZ);
7135 static void netif_free_tx_queues(struct net_device *dev)
7140 static int netif_alloc_netdev_queues(struct net_device *dev)
7142 unsigned int count = dev->num_tx_queues;
7143 struct netdev_queue *tx;
7144 size_t sz = count * sizeof(*tx);
7146 if (count < 1 || count > 0xffff)
7149 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7157 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
7158 spin_lock_init(&dev->tx_global_lock);
7163 void netif_tx_stop_all_queues(struct net_device *dev)
7167 for (i = 0; i < dev->num_tx_queues; i++) {
7168 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
7169 netif_tx_stop_queue(txq);
7172 EXPORT_SYMBOL(netif_tx_stop_all_queues);
7175 * register_netdevice - register a network device
7176 * @dev: device to register
7178 * Take a completed network device structure and add it to the kernel
7179 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7180 * chain. 0 is returned on success. A negative errno code is returned
7181 * on a failure to set up the device, or if the name is a duplicate.
7183 * Callers must hold the rtnl semaphore. You may want
7184 * register_netdev() instead of this.
7187 * The locking appears insufficient to guarantee two parallel registers
7188 * will not get the same name.
7191 int register_netdevice(struct net_device *dev)
7194 struct net *net = dev_net(dev);
7196 BUG_ON(dev_boot_phase);
7201 /* When net_device's are persistent, this will be fatal. */
7202 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
7205 spin_lock_init(&dev->addr_list_lock);
7206 netdev_set_addr_lockdep_class(dev);
7208 ret = dev_get_valid_name(net, dev, dev->name);
7212 /* Init, if this function is available */
7213 if (dev->netdev_ops->ndo_init) {
7214 ret = dev->netdev_ops->ndo_init(dev);
7222 if (((dev->hw_features | dev->features) &
7223 NETIF_F_HW_VLAN_CTAG_FILTER) &&
7224 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
7225 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
7226 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
7233 dev->ifindex = dev_new_index(net);
7234 else if (__dev_get_by_index(net, dev->ifindex))
7237 /* Transfer changeable features to wanted_features and enable
7238 * software offloads (GSO and GRO).
7240 dev->hw_features |= NETIF_F_SOFT_FEATURES;
7241 dev->features |= NETIF_F_SOFT_FEATURES;
7242 dev->wanted_features = dev->features & dev->hw_features;
7244 if (!(dev->flags & IFF_LOOPBACK))
7245 dev->hw_features |= NETIF_F_NOCACHE_COPY;
7247 /* If IPv4 TCP segmentation offload is supported we should also
7248 * allow the device to enable segmenting the frame with the option
7249 * of ignoring a static IP ID value. This doesn't enable the
7250 * feature itself but allows the user to enable it later.
7252 if (dev->hw_features & NETIF_F_TSO)
7253 dev->hw_features |= NETIF_F_TSO_MANGLEID;
7254 if (dev->vlan_features & NETIF_F_TSO)
7255 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
7256 if (dev->mpls_features & NETIF_F_TSO)
7257 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
7258 if (dev->hw_enc_features & NETIF_F_TSO)
7259 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
7261 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
7263 dev->vlan_features |= NETIF_F_HIGHDMA;
7265 /* Make NETIF_F_SG inheritable to tunnel devices.
7267 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
7269 /* Make NETIF_F_SG inheritable to MPLS.
7271 dev->mpls_features |= NETIF_F_SG;
7273 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
7274 ret = notifier_to_errno(ret);
7278 ret = netdev_register_kobject(dev);
7281 dev->reg_state = NETREG_REGISTERED;
7283 __netdev_update_features(dev);
7286 * Default initial state at registry is that the
7287 * device is present.
7290 set_bit(__LINK_STATE_PRESENT, &dev->state);
7292 linkwatch_init_dev(dev);
7294 dev_init_scheduler(dev);
7296 list_netdevice(dev);
7297 add_device_randomness(dev->dev_addr, dev->addr_len);
7299 /* If the device has permanent device address, driver should
7300 * set dev_addr and also addr_assign_type should be set to
7301 * NET_ADDR_PERM (default value).
7303 if (dev->addr_assign_type == NET_ADDR_PERM)
7304 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
7306 /* Notify protocols, that a new device appeared. */
7307 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
7308 ret = notifier_to_errno(ret);
7310 rollback_registered(dev);
7311 dev->reg_state = NETREG_UNREGISTERED;
7314 * Prevent userspace races by waiting until the network
7315 * device is fully setup before sending notifications.
7317 if (!dev->rtnl_link_ops ||
7318 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7319 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7325 if (dev->netdev_ops->ndo_uninit)
7326 dev->netdev_ops->ndo_uninit(dev);
7329 EXPORT_SYMBOL(register_netdevice);
7332 * init_dummy_netdev - init a dummy network device for NAPI
7333 * @dev: device to init
7335 * This takes a network device structure and initialize the minimum
7336 * amount of fields so it can be used to schedule NAPI polls without
7337 * registering a full blown interface. This is to be used by drivers
7338 * that need to tie several hardware interfaces to a single NAPI
7339 * poll scheduler due to HW limitations.
7341 int init_dummy_netdev(struct net_device *dev)
7343 /* Clear everything. Note we don't initialize spinlocks
7344 * are they aren't supposed to be taken by any of the
7345 * NAPI code and this dummy netdev is supposed to be
7346 * only ever used for NAPI polls
7348 memset(dev, 0, sizeof(struct net_device));
7350 /* make sure we BUG if trying to hit standard
7351 * register/unregister code path
7353 dev->reg_state = NETREG_DUMMY;
7355 /* NAPI wants this */
7356 INIT_LIST_HEAD(&dev->napi_list);
7358 /* a dummy interface is started by default */
7359 set_bit(__LINK_STATE_PRESENT, &dev->state);
7360 set_bit(__LINK_STATE_START, &dev->state);
7362 /* Note : We dont allocate pcpu_refcnt for dummy devices,
7363 * because users of this 'device' dont need to change
7369 EXPORT_SYMBOL_GPL(init_dummy_netdev);
7373 * register_netdev - register a network device
7374 * @dev: device to register
7376 * Take a completed network device structure and add it to the kernel
7377 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7378 * chain. 0 is returned on success. A negative errno code is returned
7379 * on a failure to set up the device, or if the name is a duplicate.
7381 * This is a wrapper around register_netdevice that takes the rtnl semaphore
7382 * and expands the device name if you passed a format string to
7385 int register_netdev(struct net_device *dev)
7390 err = register_netdevice(dev);
7394 EXPORT_SYMBOL(register_netdev);
7396 int netdev_refcnt_read(const struct net_device *dev)
7400 for_each_possible_cpu(i)
7401 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
7404 EXPORT_SYMBOL(netdev_refcnt_read);
7407 * netdev_wait_allrefs - wait until all references are gone.
7408 * @dev: target net_device
7410 * This is called when unregistering network devices.
7412 * Any protocol or device that holds a reference should register
7413 * for netdevice notification, and cleanup and put back the
7414 * reference if they receive an UNREGISTER event.
7415 * We can get stuck here if buggy protocols don't correctly
7418 static void netdev_wait_allrefs(struct net_device *dev)
7420 unsigned long rebroadcast_time, warning_time;
7423 linkwatch_forget_dev(dev);
7425 rebroadcast_time = warning_time = jiffies;
7426 refcnt = netdev_refcnt_read(dev);
7428 while (refcnt != 0) {
7429 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
7432 /* Rebroadcast unregister notification */
7433 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7439 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7440 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
7442 /* We must not have linkwatch events
7443 * pending on unregister. If this
7444 * happens, we simply run the queue
7445 * unscheduled, resulting in a noop
7448 linkwatch_run_queue();
7453 rebroadcast_time = jiffies;
7458 refcnt = netdev_refcnt_read(dev);
7460 if (time_after(jiffies, warning_time + 10 * HZ)) {
7461 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
7463 warning_time = jiffies;
7472 * register_netdevice(x1);
7473 * register_netdevice(x2);
7475 * unregister_netdevice(y1);
7476 * unregister_netdevice(y2);
7482 * We are invoked by rtnl_unlock().
7483 * This allows us to deal with problems:
7484 * 1) We can delete sysfs objects which invoke hotplug
7485 * without deadlocking with linkwatch via keventd.
7486 * 2) Since we run with the RTNL semaphore not held, we can sleep
7487 * safely in order to wait for the netdev refcnt to drop to zero.
7489 * We must not return until all unregister events added during
7490 * the interval the lock was held have been completed.
7492 void netdev_run_todo(void)
7494 struct list_head list;
7496 /* Snapshot list, allow later requests */
7497 list_replace_init(&net_todo_list, &list);
7502 /* Wait for rcu callbacks to finish before next phase */
7503 if (!list_empty(&list))
7506 while (!list_empty(&list)) {
7507 struct net_device *dev
7508 = list_first_entry(&list, struct net_device, todo_list);
7509 list_del(&dev->todo_list);
7512 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7515 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
7516 pr_err("network todo '%s' but state %d\n",
7517 dev->name, dev->reg_state);
7522 dev->reg_state = NETREG_UNREGISTERED;
7524 netdev_wait_allrefs(dev);
7527 BUG_ON(netdev_refcnt_read(dev));
7528 BUG_ON(!list_empty(&dev->ptype_all));
7529 BUG_ON(!list_empty(&dev->ptype_specific));
7530 WARN_ON(rcu_access_pointer(dev->ip_ptr));
7531 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
7532 WARN_ON(dev->dn_ptr);
7534 if (dev->destructor)
7535 dev->destructor(dev);
7537 /* Report a network device has been unregistered */
7539 dev_net(dev)->dev_unreg_count--;
7541 wake_up(&netdev_unregistering_wq);
7543 /* Free network device */
7544 kobject_put(&dev->dev.kobj);
7548 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
7549 * all the same fields in the same order as net_device_stats, with only
7550 * the type differing, but rtnl_link_stats64 may have additional fields
7551 * at the end for newer counters.
7553 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7554 const struct net_device_stats *netdev_stats)
7556 #if BITS_PER_LONG == 64
7557 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
7558 memcpy(stats64, netdev_stats, sizeof(*stats64));
7559 /* zero out counters that only exist in rtnl_link_stats64 */
7560 memset((char *)stats64 + sizeof(*netdev_stats), 0,
7561 sizeof(*stats64) - sizeof(*netdev_stats));
7563 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
7564 const unsigned long *src = (const unsigned long *)netdev_stats;
7565 u64 *dst = (u64 *)stats64;
7567 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
7568 for (i = 0; i < n; i++)
7570 /* zero out counters that only exist in rtnl_link_stats64 */
7571 memset((char *)stats64 + n * sizeof(u64), 0,
7572 sizeof(*stats64) - n * sizeof(u64));
7575 EXPORT_SYMBOL(netdev_stats_to_stats64);
7578 * dev_get_stats - get network device statistics
7579 * @dev: device to get statistics from
7580 * @storage: place to store stats
7582 * Get network statistics from device. Return @storage.
7583 * The device driver may provide its own method by setting
7584 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7585 * otherwise the internal statistics structure is used.
7587 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
7588 struct rtnl_link_stats64 *storage)
7590 const struct net_device_ops *ops = dev->netdev_ops;
7592 if (ops->ndo_get_stats64) {
7593 memset(storage, 0, sizeof(*storage));
7594 ops->ndo_get_stats64(dev, storage);
7595 } else if (ops->ndo_get_stats) {
7596 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7598 netdev_stats_to_stats64(storage, &dev->stats);
7600 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
7601 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
7602 storage->rx_nohandler += atomic_long_read(&dev->rx_nohandler);
7605 EXPORT_SYMBOL(dev_get_stats);
7607 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7609 struct netdev_queue *queue = dev_ingress_queue(dev);
7611 #ifdef CONFIG_NET_CLS_ACT
7614 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7617 netdev_init_one_queue(dev, queue, NULL);
7618 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7619 queue->qdisc_sleeping = &noop_qdisc;
7620 rcu_assign_pointer(dev->ingress_queue, queue);
7625 static const struct ethtool_ops default_ethtool_ops;
7627 void netdev_set_default_ethtool_ops(struct net_device *dev,
7628 const struct ethtool_ops *ops)
7630 if (dev->ethtool_ops == &default_ethtool_ops)
7631 dev->ethtool_ops = ops;
7633 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7635 void netdev_freemem(struct net_device *dev)
7637 char *addr = (char *)dev - dev->padded;
7643 * alloc_netdev_mqs - allocate network device
7644 * @sizeof_priv: size of private data to allocate space for
7645 * @name: device name format string
7646 * @name_assign_type: origin of device name
7647 * @setup: callback to initialize device
7648 * @txqs: the number of TX subqueues to allocate
7649 * @rxqs: the number of RX subqueues to allocate
7651 * Allocates a struct net_device with private data area for driver use
7652 * and performs basic initialization. Also allocates subqueue structs
7653 * for each queue on the device.
7655 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7656 unsigned char name_assign_type,
7657 void (*setup)(struct net_device *),
7658 unsigned int txqs, unsigned int rxqs)
7660 struct net_device *dev;
7662 struct net_device *p;
7664 BUG_ON(strlen(name) >= sizeof(dev->name));
7667 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7673 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7678 alloc_size = sizeof(struct net_device);
7680 /* ensure 32-byte alignment of private area */
7681 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7682 alloc_size += sizeof_priv;
7684 /* ensure 32-byte alignment of whole construct */
7685 alloc_size += NETDEV_ALIGN - 1;
7687 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7689 p = vzalloc(alloc_size);
7693 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7694 dev->padded = (char *)dev - (char *)p;
7696 dev->pcpu_refcnt = alloc_percpu(int);
7697 if (!dev->pcpu_refcnt)
7700 if (dev_addr_init(dev))
7706 dev_net_set(dev, &init_net);
7708 dev->gso_max_size = GSO_MAX_SIZE;
7709 dev->gso_max_segs = GSO_MAX_SEGS;
7711 INIT_LIST_HEAD(&dev->napi_list);
7712 INIT_LIST_HEAD(&dev->unreg_list);
7713 INIT_LIST_HEAD(&dev->close_list);
7714 INIT_LIST_HEAD(&dev->link_watch_list);
7715 INIT_LIST_HEAD(&dev->adj_list.upper);
7716 INIT_LIST_HEAD(&dev->adj_list.lower);
7717 INIT_LIST_HEAD(&dev->ptype_all);
7718 INIT_LIST_HEAD(&dev->ptype_specific);
7719 #ifdef CONFIG_NET_SCHED
7720 hash_init(dev->qdisc_hash);
7722 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7725 if (!dev->tx_queue_len) {
7726 dev->priv_flags |= IFF_NO_QUEUE;
7727 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
7730 dev->num_tx_queues = txqs;
7731 dev->real_num_tx_queues = txqs;
7732 if (netif_alloc_netdev_queues(dev))
7736 dev->num_rx_queues = rxqs;
7737 dev->real_num_rx_queues = rxqs;
7738 if (netif_alloc_rx_queues(dev))
7742 strcpy(dev->name, name);
7743 dev->name_assign_type = name_assign_type;
7744 dev->group = INIT_NETDEV_GROUP;
7745 if (!dev->ethtool_ops)
7746 dev->ethtool_ops = &default_ethtool_ops;
7748 nf_hook_ingress_init(dev);
7757 free_percpu(dev->pcpu_refcnt);
7759 netdev_freemem(dev);
7762 EXPORT_SYMBOL(alloc_netdev_mqs);
7765 * free_netdev - free network device
7768 * This function does the last stage of destroying an allocated device
7769 * interface. The reference to the device object is released.
7770 * If this is the last reference then it will be freed.
7771 * Must be called in process context.
7773 void free_netdev(struct net_device *dev)
7775 struct napi_struct *p, *n;
7778 netif_free_tx_queues(dev);
7783 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7785 /* Flush device addresses */
7786 dev_addr_flush(dev);
7788 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7791 free_percpu(dev->pcpu_refcnt);
7792 dev->pcpu_refcnt = NULL;
7794 /* Compatibility with error handling in drivers */
7795 if (dev->reg_state == NETREG_UNINITIALIZED) {
7796 netdev_freemem(dev);
7800 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7801 dev->reg_state = NETREG_RELEASED;
7803 /* will free via device release */
7804 put_device(&dev->dev);
7806 EXPORT_SYMBOL(free_netdev);
7809 * synchronize_net - Synchronize with packet receive processing
7811 * Wait for packets currently being received to be done.
7812 * Does not block later packets from starting.
7814 void synchronize_net(void)
7817 if (rtnl_is_locked())
7818 synchronize_rcu_expedited();
7822 EXPORT_SYMBOL(synchronize_net);
7825 * unregister_netdevice_queue - remove device from the kernel
7829 * This function shuts down a device interface and removes it
7830 * from the kernel tables.
7831 * If head not NULL, device is queued to be unregistered later.
7833 * Callers must hold the rtnl semaphore. You may want
7834 * unregister_netdev() instead of this.
7837 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7842 list_move_tail(&dev->unreg_list, head);
7844 rollback_registered(dev);
7845 /* Finish processing unregister after unlock */
7849 EXPORT_SYMBOL(unregister_netdevice_queue);
7852 * unregister_netdevice_many - unregister many devices
7853 * @head: list of devices
7855 * Note: As most callers use a stack allocated list_head,
7856 * we force a list_del() to make sure stack wont be corrupted later.
7858 void unregister_netdevice_many(struct list_head *head)
7860 struct net_device *dev;
7862 if (!list_empty(head)) {
7863 rollback_registered_many(head);
7864 list_for_each_entry(dev, head, unreg_list)
7869 EXPORT_SYMBOL(unregister_netdevice_many);
7872 * unregister_netdev - remove device from the kernel
7875 * This function shuts down a device interface and removes it
7876 * from the kernel tables.
7878 * This is just a wrapper for unregister_netdevice that takes
7879 * the rtnl semaphore. In general you want to use this and not
7880 * unregister_netdevice.
7882 void unregister_netdev(struct net_device *dev)
7885 unregister_netdevice(dev);
7888 EXPORT_SYMBOL(unregister_netdev);
7891 * dev_change_net_namespace - move device to different nethost namespace
7893 * @net: network namespace
7894 * @pat: If not NULL name pattern to try if the current device name
7895 * is already taken in the destination network namespace.
7897 * This function shuts down a device interface and moves it
7898 * to a new network namespace. On success 0 is returned, on
7899 * a failure a netagive errno code is returned.
7901 * Callers must hold the rtnl semaphore.
7904 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7910 /* Don't allow namespace local devices to be moved. */
7912 if (dev->features & NETIF_F_NETNS_LOCAL)
7915 /* Ensure the device has been registrered */
7916 if (dev->reg_state != NETREG_REGISTERED)
7919 /* Get out if there is nothing todo */
7921 if (net_eq(dev_net(dev), net))
7924 /* Pick the destination device name, and ensure
7925 * we can use it in the destination network namespace.
7928 if (__dev_get_by_name(net, dev->name)) {
7929 /* We get here if we can't use the current device name */
7932 if (dev_get_valid_name(net, dev, pat) < 0)
7937 * And now a mini version of register_netdevice unregister_netdevice.
7940 /* If device is running close it first. */
7943 /* And unlink it from device chain */
7945 unlist_netdevice(dev);
7949 /* Shutdown queueing discipline. */
7952 /* Notify protocols, that we are about to destroy
7953 this device. They should clean all the things.
7955 Note that dev->reg_state stays at NETREG_REGISTERED.
7956 This is wanted because this way 8021q and macvlan know
7957 the device is just moving and can keep their slaves up.
7959 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7961 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7962 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7965 * Flush the unicast and multicast chains
7970 /* Send a netdev-removed uevent to the old namespace */
7971 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7972 netdev_adjacent_del_links(dev);
7974 /* Actually switch the network namespace */
7975 dev_net_set(dev, net);
7977 /* If there is an ifindex conflict assign a new one */
7978 if (__dev_get_by_index(net, dev->ifindex))
7979 dev->ifindex = dev_new_index(net);
7981 /* Send a netdev-add uevent to the new namespace */
7982 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7983 netdev_adjacent_add_links(dev);
7985 /* Fixup kobjects */
7986 err = device_rename(&dev->dev, dev->name);
7989 /* Add the device back in the hashes */
7990 list_netdevice(dev);
7992 /* Notify protocols, that a new device appeared. */
7993 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7996 * Prevent userspace races by waiting until the network
7997 * device is fully setup before sending notifications.
7999 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
8006 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
8008 static int dev_cpu_dead(unsigned int oldcpu)
8010 struct sk_buff **list_skb;
8011 struct sk_buff *skb;
8013 struct softnet_data *sd, *oldsd;
8015 local_irq_disable();
8016 cpu = smp_processor_id();
8017 sd = &per_cpu(softnet_data, cpu);
8018 oldsd = &per_cpu(softnet_data, oldcpu);
8020 /* Find end of our completion_queue. */
8021 list_skb = &sd->completion_queue;
8023 list_skb = &(*list_skb)->next;
8024 /* Append completion queue from offline CPU. */
8025 *list_skb = oldsd->completion_queue;
8026 oldsd->completion_queue = NULL;
8028 /* Append output queue from offline CPU. */
8029 if (oldsd->output_queue) {
8030 *sd->output_queue_tailp = oldsd->output_queue;
8031 sd->output_queue_tailp = oldsd->output_queue_tailp;
8032 oldsd->output_queue = NULL;
8033 oldsd->output_queue_tailp = &oldsd->output_queue;
8035 /* Append NAPI poll list from offline CPU, with one exception :
8036 * process_backlog() must be called by cpu owning percpu backlog.
8037 * We properly handle process_queue & input_pkt_queue later.
8039 while (!list_empty(&oldsd->poll_list)) {
8040 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
8044 list_del_init(&napi->poll_list);
8045 if (napi->poll == process_backlog)
8048 ____napi_schedule(sd, napi);
8051 raise_softirq_irqoff(NET_TX_SOFTIRQ);
8054 /* Process offline CPU's input_pkt_queue */
8055 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
8057 input_queue_head_incr(oldsd);
8059 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
8061 input_queue_head_incr(oldsd);
8068 * netdev_increment_features - increment feature set by one
8069 * @all: current feature set
8070 * @one: new feature set
8071 * @mask: mask feature set
8073 * Computes a new feature set after adding a device with feature set
8074 * @one to the master device with current feature set @all. Will not
8075 * enable anything that is off in @mask. Returns the new feature set.
8077 netdev_features_t netdev_increment_features(netdev_features_t all,
8078 netdev_features_t one, netdev_features_t mask)
8080 if (mask & NETIF_F_HW_CSUM)
8081 mask |= NETIF_F_CSUM_MASK;
8082 mask |= NETIF_F_VLAN_CHALLENGED;
8084 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
8085 all &= one | ~NETIF_F_ALL_FOR_ALL;
8087 /* If one device supports hw checksumming, set for all. */
8088 if (all & NETIF_F_HW_CSUM)
8089 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
8093 EXPORT_SYMBOL(netdev_increment_features);
8095 static struct hlist_head * __net_init netdev_create_hash(void)
8098 struct hlist_head *hash;
8100 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
8102 for (i = 0; i < NETDEV_HASHENTRIES; i++)
8103 INIT_HLIST_HEAD(&hash[i]);
8108 /* Initialize per network namespace state */
8109 static int __net_init netdev_init(struct net *net)
8111 if (net != &init_net)
8112 INIT_LIST_HEAD(&net->dev_base_head);
8114 net->dev_name_head = netdev_create_hash();
8115 if (net->dev_name_head == NULL)
8118 net->dev_index_head = netdev_create_hash();
8119 if (net->dev_index_head == NULL)
8125 kfree(net->dev_name_head);
8131 * netdev_drivername - network driver for the device
8132 * @dev: network device
8134 * Determine network driver for device.
8136 const char *netdev_drivername(const struct net_device *dev)
8138 const struct device_driver *driver;
8139 const struct device *parent;
8140 const char *empty = "";
8142 parent = dev->dev.parent;
8146 driver = parent->driver;
8147 if (driver && driver->name)
8148 return driver->name;
8152 static void __netdev_printk(const char *level, const struct net_device *dev,
8153 struct va_format *vaf)
8155 if (dev && dev->dev.parent) {
8156 dev_printk_emit(level[1] - '0',
8159 dev_driver_string(dev->dev.parent),
8160 dev_name(dev->dev.parent),
8161 netdev_name(dev), netdev_reg_state(dev),
8164 printk("%s%s%s: %pV",
8165 level, netdev_name(dev), netdev_reg_state(dev), vaf);
8167 printk("%s(NULL net_device): %pV", level, vaf);
8171 void netdev_printk(const char *level, const struct net_device *dev,
8172 const char *format, ...)
8174 struct va_format vaf;
8177 va_start(args, format);
8182 __netdev_printk(level, dev, &vaf);
8186 EXPORT_SYMBOL(netdev_printk);
8188 #define define_netdev_printk_level(func, level) \
8189 void func(const struct net_device *dev, const char *fmt, ...) \
8191 struct va_format vaf; \
8194 va_start(args, fmt); \
8199 __netdev_printk(level, dev, &vaf); \
8203 EXPORT_SYMBOL(func);
8205 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
8206 define_netdev_printk_level(netdev_alert, KERN_ALERT);
8207 define_netdev_printk_level(netdev_crit, KERN_CRIT);
8208 define_netdev_printk_level(netdev_err, KERN_ERR);
8209 define_netdev_printk_level(netdev_warn, KERN_WARNING);
8210 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
8211 define_netdev_printk_level(netdev_info, KERN_INFO);
8213 static void __net_exit netdev_exit(struct net *net)
8215 kfree(net->dev_name_head);
8216 kfree(net->dev_index_head);
8219 static struct pernet_operations __net_initdata netdev_net_ops = {
8220 .init = netdev_init,
8221 .exit = netdev_exit,
8224 static void __net_exit default_device_exit(struct net *net)
8226 struct net_device *dev, *aux;
8228 * Push all migratable network devices back to the
8229 * initial network namespace
8232 for_each_netdev_safe(net, dev, aux) {
8234 char fb_name[IFNAMSIZ];
8236 /* Ignore unmoveable devices (i.e. loopback) */
8237 if (dev->features & NETIF_F_NETNS_LOCAL)
8240 /* Leave virtual devices for the generic cleanup */
8241 if (dev->rtnl_link_ops)
8244 /* Push remaining network devices to init_net */
8245 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
8246 err = dev_change_net_namespace(dev, &init_net, fb_name);
8248 pr_emerg("%s: failed to move %s to init_net: %d\n",
8249 __func__, dev->name, err);
8256 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
8258 /* Return with the rtnl_lock held when there are no network
8259 * devices unregistering in any network namespace in net_list.
8263 DEFINE_WAIT_FUNC(wait, woken_wake_function);
8265 add_wait_queue(&netdev_unregistering_wq, &wait);
8267 unregistering = false;
8269 list_for_each_entry(net, net_list, exit_list) {
8270 if (net->dev_unreg_count > 0) {
8271 unregistering = true;
8279 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
8281 remove_wait_queue(&netdev_unregistering_wq, &wait);
8284 static void __net_exit default_device_exit_batch(struct list_head *net_list)
8286 /* At exit all network devices most be removed from a network
8287 * namespace. Do this in the reverse order of registration.
8288 * Do this across as many network namespaces as possible to
8289 * improve batching efficiency.
8291 struct net_device *dev;
8293 LIST_HEAD(dev_kill_list);
8295 /* To prevent network device cleanup code from dereferencing
8296 * loopback devices or network devices that have been freed
8297 * wait here for all pending unregistrations to complete,
8298 * before unregistring the loopback device and allowing the
8299 * network namespace be freed.
8301 * The netdev todo list containing all network devices
8302 * unregistrations that happen in default_device_exit_batch
8303 * will run in the rtnl_unlock() at the end of
8304 * default_device_exit_batch.
8306 rtnl_lock_unregistering(net_list);
8307 list_for_each_entry(net, net_list, exit_list) {
8308 for_each_netdev_reverse(net, dev) {
8309 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
8310 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
8312 unregister_netdevice_queue(dev, &dev_kill_list);
8315 unregister_netdevice_many(&dev_kill_list);
8319 static struct pernet_operations __net_initdata default_device_ops = {
8320 .exit = default_device_exit,
8321 .exit_batch = default_device_exit_batch,
8325 * Initialize the DEV module. At boot time this walks the device list and
8326 * unhooks any devices that fail to initialise (normally hardware not
8327 * present) and leaves us with a valid list of present and active devices.
8332 * This is called single threaded during boot, so no need
8333 * to take the rtnl semaphore.
8335 static int __init net_dev_init(void)
8337 int i, rc = -ENOMEM;
8339 BUG_ON(!dev_boot_phase);
8341 if (dev_proc_init())
8344 if (netdev_kobject_init())
8347 INIT_LIST_HEAD(&ptype_all);
8348 for (i = 0; i < PTYPE_HASH_SIZE; i++)
8349 INIT_LIST_HEAD(&ptype_base[i]);
8351 INIT_LIST_HEAD(&offload_base);
8353 if (register_pernet_subsys(&netdev_net_ops))
8357 * Initialise the packet receive queues.
8360 for_each_possible_cpu(i) {
8361 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
8362 struct softnet_data *sd = &per_cpu(softnet_data, i);
8364 INIT_WORK(flush, flush_backlog);
8366 skb_queue_head_init(&sd->input_pkt_queue);
8367 skb_queue_head_init(&sd->process_queue);
8368 INIT_LIST_HEAD(&sd->poll_list);
8369 sd->output_queue_tailp = &sd->output_queue;
8371 sd->csd.func = rps_trigger_softirq;
8376 sd->backlog.poll = process_backlog;
8377 sd->backlog.weight = weight_p;
8382 /* The loopback device is special if any other network devices
8383 * is present in a network namespace the loopback device must
8384 * be present. Since we now dynamically allocate and free the
8385 * loopback device ensure this invariant is maintained by
8386 * keeping the loopback device as the first device on the
8387 * list of network devices. Ensuring the loopback devices
8388 * is the first device that appears and the last network device
8391 if (register_pernet_device(&loopback_net_ops))
8394 if (register_pernet_device(&default_device_ops))
8397 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
8398 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
8400 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
8401 NULL, dev_cpu_dead);
8409 subsys_initcall(net_dev_init);