2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/rwsem.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <linux/bpf.h>
99 #include <net/net_namespace.h>
100 #include <net/sock.h>
101 #include <net/busy_poll.h>
102 #include <linux/rtnetlink.h>
103 #include <linux/stat.h>
105 #include <net/dst_metadata.h>
106 #include <net/pkt_sched.h>
107 #include <net/checksum.h>
108 #include <net/xfrm.h>
109 #include <linux/highmem.h>
110 #include <linux/init.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
125 #include <net/mpls.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
131 #include <trace/events/net.h>
132 #include <trace/events/skb.h>
133 #include <linux/pci.h>
134 #include <linux/inetdevice.h>
135 #include <linux/cpu_rmap.h>
136 #include <linux/static_key.h>
137 #include <linux/hashtable.h>
138 #include <linux/vmalloc.h>
139 #include <linux/if_macvlan.h>
140 #include <linux/errqueue.h>
141 #include <linux/hrtimer.h>
142 #include <linux/netfilter_ingress.h>
143 #include <linux/sctp.h>
144 #include <linux/crash_dump.h>
146 #include "net-sysfs.h"
148 /* Instead of increasing this, you should create a hash table. */
149 #define MAX_GRO_SKBS 8
151 /* This should be increased if a protocol with a bigger head is added. */
152 #define GRO_MAX_HEAD (MAX_HEADER + 128)
154 static DEFINE_SPINLOCK(ptype_lock);
155 static DEFINE_SPINLOCK(offload_lock);
156 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
157 struct list_head ptype_all __read_mostly; /* Taps */
158 static struct list_head offload_base __read_mostly;
160 static int netif_rx_internal(struct sk_buff *skb);
161 static int call_netdevice_notifiers_info(unsigned long val,
162 struct net_device *dev,
163 struct netdev_notifier_info *info);
166 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
169 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
171 * Writers must hold the rtnl semaphore while they loop through the
172 * dev_base_head list, and hold dev_base_lock for writing when they do the
173 * actual updates. This allows pure readers to access the list even
174 * while a writer is preparing to update it.
176 * To put it another way, dev_base_lock is held for writing only to
177 * protect against pure readers; the rtnl semaphore provides the
178 * protection against other writers.
180 * See, for example usages, register_netdevice() and
181 * unregister_netdevice(), which must be called with the rtnl
184 DEFINE_RWLOCK(dev_base_lock);
185 EXPORT_SYMBOL(dev_base_lock);
187 /* protects napi_hash addition/deletion and napi_gen_id */
188 static DEFINE_SPINLOCK(napi_hash_lock);
190 static unsigned int napi_gen_id = NR_CPUS;
191 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
193 static DECLARE_RWSEM(devnet_rename_sem);
195 static inline void dev_base_seq_inc(struct net *net)
197 while (++net->dev_base_seq == 0);
200 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
202 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
204 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
207 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
209 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
212 static inline void rps_lock(struct softnet_data *sd)
215 spin_lock(&sd->input_pkt_queue.lock);
219 static inline void rps_unlock(struct softnet_data *sd)
222 spin_unlock(&sd->input_pkt_queue.lock);
226 /* Device list insertion */
227 static void list_netdevice(struct net_device *dev)
229 struct net *net = dev_net(dev);
233 write_lock_bh(&dev_base_lock);
234 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
235 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
236 hlist_add_head_rcu(&dev->index_hlist,
237 dev_index_hash(net, dev->ifindex));
238 write_unlock_bh(&dev_base_lock);
240 dev_base_seq_inc(net);
243 /* Device list removal
244 * caller must respect a RCU grace period before freeing/reusing dev
246 static void unlist_netdevice(struct net_device *dev)
250 /* Unlink dev from the device chain */
251 write_lock_bh(&dev_base_lock);
252 list_del_rcu(&dev->dev_list);
253 hlist_del_rcu(&dev->name_hlist);
254 hlist_del_rcu(&dev->index_hlist);
255 write_unlock_bh(&dev_base_lock);
257 dev_base_seq_inc(dev_net(dev));
264 static RAW_NOTIFIER_HEAD(netdev_chain);
267 * Device drivers call our routines to queue packets here. We empty the
268 * queue in the local softnet handler.
271 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
272 EXPORT_PER_CPU_SYMBOL(softnet_data);
274 #ifdef CONFIG_LOCKDEP
276 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
277 * according to dev->type
279 static const unsigned short netdev_lock_type[] =
280 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
281 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
282 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
283 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
284 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
285 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
286 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
287 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
288 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
289 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
290 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
291 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
292 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
293 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
294 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
296 static const char *const netdev_lock_name[] =
297 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
298 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
299 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
300 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
301 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
302 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
303 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
304 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
305 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
306 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
307 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
308 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
309 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
310 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
311 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
313 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
316 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
320 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
321 if (netdev_lock_type[i] == dev_type)
323 /* the last key is used by default */
324 return ARRAY_SIZE(netdev_lock_type) - 1;
327 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
328 unsigned short dev_type)
332 i = netdev_lock_pos(dev_type);
333 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
334 netdev_lock_name[i]);
337 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
341 i = netdev_lock_pos(dev->type);
342 lockdep_set_class_and_name(&dev->addr_list_lock,
343 &netdev_addr_lock_key[i],
344 netdev_lock_name[i]);
347 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
348 unsigned short dev_type)
351 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
356 /*******************************************************************************
358 Protocol management and registration routines
360 *******************************************************************************/
363 * Add a protocol ID to the list. Now that the input handler is
364 * smarter we can dispense with all the messy stuff that used to be
367 * BEWARE!!! Protocol handlers, mangling input packets,
368 * MUST BE last in hash buckets and checking protocol handlers
369 * MUST start from promiscuous ptype_all chain in net_bh.
370 * It is true now, do not change it.
371 * Explanation follows: if protocol handler, mangling packet, will
372 * be the first on list, it is not able to sense, that packet
373 * is cloned and should be copied-on-write, so that it will
374 * change it and subsequent readers will get broken packet.
378 static inline struct list_head *ptype_head(const struct packet_type *pt)
380 if (pt->type == htons(ETH_P_ALL))
381 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
383 return pt->dev ? &pt->dev->ptype_specific :
384 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
388 * dev_add_pack - add packet handler
389 * @pt: packet type declaration
391 * Add a protocol handler to the networking stack. The passed &packet_type
392 * is linked into kernel lists and may not be freed until it has been
393 * removed from the kernel lists.
395 * This call does not sleep therefore it can not
396 * guarantee all CPU's that are in middle of receiving packets
397 * will see the new packet type (until the next received packet).
400 void dev_add_pack(struct packet_type *pt)
402 struct list_head *head = ptype_head(pt);
404 spin_lock(&ptype_lock);
405 list_add_rcu(&pt->list, head);
406 spin_unlock(&ptype_lock);
408 EXPORT_SYMBOL(dev_add_pack);
411 * __dev_remove_pack - remove packet handler
412 * @pt: packet type declaration
414 * Remove a protocol handler that was previously added to the kernel
415 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
416 * from the kernel lists and can be freed or reused once this function
419 * The packet type might still be in use by receivers
420 * and must not be freed until after all the CPU's have gone
421 * through a quiescent state.
423 void __dev_remove_pack(struct packet_type *pt)
425 struct list_head *head = ptype_head(pt);
426 struct packet_type *pt1;
428 spin_lock(&ptype_lock);
430 list_for_each_entry(pt1, head, list) {
432 list_del_rcu(&pt->list);
437 pr_warn("dev_remove_pack: %p not found\n", pt);
439 spin_unlock(&ptype_lock);
441 EXPORT_SYMBOL(__dev_remove_pack);
444 * dev_remove_pack - remove packet handler
445 * @pt: packet type declaration
447 * Remove a protocol handler that was previously added to the kernel
448 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
449 * from the kernel lists and can be freed or reused once this function
452 * This call sleeps to guarantee that no CPU is looking at the packet
455 void dev_remove_pack(struct packet_type *pt)
457 __dev_remove_pack(pt);
461 EXPORT_SYMBOL(dev_remove_pack);
465 * dev_add_offload - register offload handlers
466 * @po: protocol offload declaration
468 * Add protocol offload handlers to the networking stack. The passed
469 * &proto_offload is linked into kernel lists and may not be freed until
470 * it has been removed from the kernel lists.
472 * This call does not sleep therefore it can not
473 * guarantee all CPU's that are in middle of receiving packets
474 * will see the new offload handlers (until the next received packet).
476 void dev_add_offload(struct packet_offload *po)
478 struct packet_offload *elem;
480 spin_lock(&offload_lock);
481 list_for_each_entry(elem, &offload_base, list) {
482 if (po->priority < elem->priority)
485 list_add_rcu(&po->list, elem->list.prev);
486 spin_unlock(&offload_lock);
488 EXPORT_SYMBOL(dev_add_offload);
491 * __dev_remove_offload - remove offload handler
492 * @po: packet offload declaration
494 * Remove a protocol offload handler that was previously added to the
495 * kernel offload handlers by dev_add_offload(). The passed &offload_type
496 * is removed from the kernel lists and can be freed or reused once this
499 * The packet type might still be in use by receivers
500 * and must not be freed until after all the CPU's have gone
501 * through a quiescent state.
503 static void __dev_remove_offload(struct packet_offload *po)
505 struct list_head *head = &offload_base;
506 struct packet_offload *po1;
508 spin_lock(&offload_lock);
510 list_for_each_entry(po1, head, list) {
512 list_del_rcu(&po->list);
517 pr_warn("dev_remove_offload: %p not found\n", po);
519 spin_unlock(&offload_lock);
523 * dev_remove_offload - remove packet offload handler
524 * @po: packet offload declaration
526 * Remove a packet offload handler that was previously added to the kernel
527 * offload handlers by dev_add_offload(). The passed &offload_type is
528 * removed from the kernel lists and can be freed or reused once this
531 * This call sleeps to guarantee that no CPU is looking at the packet
534 void dev_remove_offload(struct packet_offload *po)
536 __dev_remove_offload(po);
540 EXPORT_SYMBOL(dev_remove_offload);
542 /******************************************************************************
544 Device Boot-time Settings Routines
546 *******************************************************************************/
548 /* Boot time configuration table */
549 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
552 * netdev_boot_setup_add - add new setup entry
553 * @name: name of the device
554 * @map: configured settings for the device
556 * Adds new setup entry to the dev_boot_setup list. The function
557 * returns 0 on error and 1 on success. This is a generic routine to
560 static int netdev_boot_setup_add(char *name, struct ifmap *map)
562 struct netdev_boot_setup *s;
566 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
567 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
568 memset(s[i].name, 0, sizeof(s[i].name));
569 strlcpy(s[i].name, name, IFNAMSIZ);
570 memcpy(&s[i].map, map, sizeof(s[i].map));
575 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
579 * netdev_boot_setup_check - check boot time settings
580 * @dev: the netdevice
582 * Check boot time settings for the device.
583 * The found settings are set for the device to be used
584 * later in the device probing.
585 * Returns 0 if no settings found, 1 if they are.
587 int netdev_boot_setup_check(struct net_device *dev)
589 struct netdev_boot_setup *s = dev_boot_setup;
592 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
593 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
594 !strcmp(dev->name, s[i].name)) {
595 dev->irq = s[i].map.irq;
596 dev->base_addr = s[i].map.base_addr;
597 dev->mem_start = s[i].map.mem_start;
598 dev->mem_end = s[i].map.mem_end;
604 EXPORT_SYMBOL(netdev_boot_setup_check);
608 * netdev_boot_base - get address from boot time settings
609 * @prefix: prefix for network device
610 * @unit: id for network device
612 * Check boot time settings for the base address of device.
613 * The found settings are set for the device to be used
614 * later in the device probing.
615 * Returns 0 if no settings found.
617 unsigned long netdev_boot_base(const char *prefix, int unit)
619 const struct netdev_boot_setup *s = dev_boot_setup;
623 sprintf(name, "%s%d", prefix, unit);
626 * If device already registered then return base of 1
627 * to indicate not to probe for this interface
629 if (__dev_get_by_name(&init_net, name))
632 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
633 if (!strcmp(name, s[i].name))
634 return s[i].map.base_addr;
639 * Saves at boot time configured settings for any netdevice.
641 int __init netdev_boot_setup(char *str)
646 str = get_options(str, ARRAY_SIZE(ints), ints);
651 memset(&map, 0, sizeof(map));
655 map.base_addr = ints[2];
657 map.mem_start = ints[3];
659 map.mem_end = ints[4];
661 /* Add new entry to the list */
662 return netdev_boot_setup_add(str, &map);
665 __setup("netdev=", netdev_boot_setup);
667 /*******************************************************************************
669 Device Interface Subroutines
671 *******************************************************************************/
674 * dev_get_iflink - get 'iflink' value of a interface
675 * @dev: targeted interface
677 * Indicates the ifindex the interface is linked to.
678 * Physical interfaces have the same 'ifindex' and 'iflink' values.
681 int dev_get_iflink(const struct net_device *dev)
683 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
684 return dev->netdev_ops->ndo_get_iflink(dev);
688 EXPORT_SYMBOL(dev_get_iflink);
691 * dev_fill_metadata_dst - Retrieve tunnel egress information.
692 * @dev: targeted interface
695 * For better visibility of tunnel traffic OVS needs to retrieve
696 * egress tunnel information for a packet. Following API allows
697 * user to get this info.
699 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
701 struct ip_tunnel_info *info;
703 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
706 info = skb_tunnel_info_unclone(skb);
709 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
712 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
714 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
717 * __dev_get_by_name - find a device by its name
718 * @net: the applicable net namespace
719 * @name: name to find
721 * Find an interface by name. Must be called under RTNL semaphore
722 * or @dev_base_lock. If the name is found a pointer to the device
723 * is returned. If the name is not found then %NULL is returned. The
724 * reference counters are not incremented so the caller must be
725 * careful with locks.
728 struct net_device *__dev_get_by_name(struct net *net, const char *name)
730 struct net_device *dev;
731 struct hlist_head *head = dev_name_hash(net, name);
733 hlist_for_each_entry(dev, head, name_hlist)
734 if (!strncmp(dev->name, name, IFNAMSIZ))
739 EXPORT_SYMBOL(__dev_get_by_name);
742 * dev_get_by_name_rcu - find a device by its name
743 * @net: the applicable net namespace
744 * @name: name to find
746 * Find an interface by name.
747 * If the name is found a pointer to the device is returned.
748 * If the name is not found then %NULL is returned.
749 * The reference counters are not incremented so the caller must be
750 * careful with locks. The caller must hold RCU lock.
753 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
755 struct net_device *dev;
756 struct hlist_head *head = dev_name_hash(net, name);
758 hlist_for_each_entry_rcu(dev, head, name_hlist)
759 if (!strncmp(dev->name, name, IFNAMSIZ))
764 EXPORT_SYMBOL(dev_get_by_name_rcu);
767 * dev_get_by_name - find a device by its name
768 * @net: the applicable net namespace
769 * @name: name to find
771 * Find an interface by name. This can be called from any
772 * context and does its own locking. The returned handle has
773 * the usage count incremented and the caller must use dev_put() to
774 * release it when it is no longer needed. %NULL is returned if no
775 * matching device is found.
778 struct net_device *dev_get_by_name(struct net *net, const char *name)
780 struct net_device *dev;
783 dev = dev_get_by_name_rcu(net, name);
789 EXPORT_SYMBOL(dev_get_by_name);
792 * __dev_get_by_index - find a device by its ifindex
793 * @net: the applicable net namespace
794 * @ifindex: index of device
796 * Search for an interface by index. Returns %NULL if the device
797 * is not found or a pointer to the device. The device has not
798 * had its reference counter increased so the caller must be careful
799 * about locking. The caller must hold either the RTNL semaphore
803 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
805 struct net_device *dev;
806 struct hlist_head *head = dev_index_hash(net, ifindex);
808 hlist_for_each_entry(dev, head, index_hlist)
809 if (dev->ifindex == ifindex)
814 EXPORT_SYMBOL(__dev_get_by_index);
817 * dev_get_by_index_rcu - find a device by its ifindex
818 * @net: the applicable net namespace
819 * @ifindex: index of device
821 * Search for an interface by index. Returns %NULL if the device
822 * is not found or a pointer to the device. The device has not
823 * had its reference counter increased so the caller must be careful
824 * about locking. The caller must hold RCU lock.
827 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
829 struct net_device *dev;
830 struct hlist_head *head = dev_index_hash(net, ifindex);
832 hlist_for_each_entry_rcu(dev, head, index_hlist)
833 if (dev->ifindex == ifindex)
838 EXPORT_SYMBOL(dev_get_by_index_rcu);
842 * dev_get_by_index - find a device by its ifindex
843 * @net: the applicable net namespace
844 * @ifindex: index of device
846 * Search for an interface by index. Returns NULL if the device
847 * is not found or a pointer to the device. The device returned has
848 * had a reference added and the pointer is safe until the user calls
849 * dev_put to indicate they have finished with it.
852 struct net_device *dev_get_by_index(struct net *net, int ifindex)
854 struct net_device *dev;
857 dev = dev_get_by_index_rcu(net, ifindex);
863 EXPORT_SYMBOL(dev_get_by_index);
866 * netdev_get_name - get a netdevice name, knowing its ifindex.
867 * @net: network namespace
868 * @name: a pointer to the buffer where the name will be stored.
869 * @ifindex: the ifindex of the interface to get the name from.
871 int netdev_get_name(struct net *net, char *name, int ifindex)
873 struct net_device *dev;
876 down_read(&devnet_rename_sem);
879 dev = dev_get_by_index_rcu(net, ifindex);
885 strcpy(name, dev->name);
890 up_read(&devnet_rename_sem);
895 * dev_getbyhwaddr_rcu - find a device by its hardware address
896 * @net: the applicable net namespace
897 * @type: media type of device
898 * @ha: hardware address
900 * Search for an interface by MAC address. Returns NULL if the device
901 * is not found or a pointer to the device.
902 * The caller must hold RCU or RTNL.
903 * The returned device has not had its ref count increased
904 * and the caller must therefore be careful about locking
908 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
911 struct net_device *dev;
913 for_each_netdev_rcu(net, dev)
914 if (dev->type == type &&
915 !memcmp(dev->dev_addr, ha, dev->addr_len))
920 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
922 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
924 struct net_device *dev;
927 for_each_netdev(net, dev)
928 if (dev->type == type)
933 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
935 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
937 struct net_device *dev, *ret = NULL;
940 for_each_netdev_rcu(net, dev)
941 if (dev->type == type) {
949 EXPORT_SYMBOL(dev_getfirstbyhwtype);
952 * __dev_get_by_flags - find any device with given flags
953 * @net: the applicable net namespace
954 * @if_flags: IFF_* values
955 * @mask: bitmask of bits in if_flags to check
957 * Search for any interface with the given flags. Returns NULL if a device
958 * is not found or a pointer to the device. Must be called inside
959 * rtnl_lock(), and result refcount is unchanged.
962 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
965 struct net_device *dev, *ret;
970 for_each_netdev(net, dev) {
971 if (((dev->flags ^ if_flags) & mask) == 0) {
978 EXPORT_SYMBOL(__dev_get_by_flags);
981 * dev_valid_name - check if name is okay for network device
984 * Network device names need to be valid file names to
985 * to allow sysfs to work. We also disallow any kind of
988 bool dev_valid_name(const char *name)
992 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
994 if (!strcmp(name, ".") || !strcmp(name, ".."))
998 if (*name == '/' || *name == ':' || isspace(*name))
1004 EXPORT_SYMBOL(dev_valid_name);
1007 * __dev_alloc_name - allocate a name for a device
1008 * @net: network namespace to allocate the device name in
1009 * @name: name format string
1010 * @buf: scratch buffer and result name string
1012 * Passed a format string - eg "lt%d" it will try and find a suitable
1013 * id. It scans list of devices to build up a free map, then chooses
1014 * the first empty slot. The caller must hold the dev_base or rtnl lock
1015 * while allocating the name and adding the device in order to avoid
1017 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1018 * Returns the number of the unit assigned or a negative errno code.
1021 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1025 const int max_netdevices = 8*PAGE_SIZE;
1026 unsigned long *inuse;
1027 struct net_device *d;
1029 p = strnchr(name, IFNAMSIZ-1, '%');
1032 * Verify the string as this thing may have come from
1033 * the user. There must be either one "%d" and no other "%"
1036 if (p[1] != 'd' || strchr(p + 2, '%'))
1039 /* Use one page as a bit array of possible slots */
1040 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1044 for_each_netdev(net, d) {
1045 if (!sscanf(d->name, name, &i))
1047 if (i < 0 || i >= max_netdevices)
1050 /* avoid cases where sscanf is not exact inverse of printf */
1051 snprintf(buf, IFNAMSIZ, name, i);
1052 if (!strncmp(buf, d->name, IFNAMSIZ))
1056 i = find_first_zero_bit(inuse, max_netdevices);
1057 free_page((unsigned long) inuse);
1061 snprintf(buf, IFNAMSIZ, name, i);
1062 if (!__dev_get_by_name(net, buf))
1065 /* It is possible to run out of possible slots
1066 * when the name is long and there isn't enough space left
1067 * for the digits, or if all bits are used.
1073 * dev_alloc_name - allocate a name for a device
1075 * @name: name format string
1077 * Passed a format string - eg "lt%d" it will try and find a suitable
1078 * id. It scans list of devices to build up a free map, then chooses
1079 * the first empty slot. The caller must hold the dev_base or rtnl lock
1080 * while allocating the name and adding the device in order to avoid
1082 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1083 * Returns the number of the unit assigned or a negative errno code.
1086 int dev_alloc_name(struct net_device *dev, const char *name)
1092 BUG_ON(!dev_net(dev));
1094 ret = __dev_alloc_name(net, name, buf);
1096 strlcpy(dev->name, buf, IFNAMSIZ);
1099 EXPORT_SYMBOL(dev_alloc_name);
1101 static int dev_alloc_name_ns(struct net *net,
1102 struct net_device *dev,
1108 ret = __dev_alloc_name(net, name, buf);
1110 strlcpy(dev->name, buf, IFNAMSIZ);
1114 int dev_get_valid_name(struct net *net, struct net_device *dev,
1119 if (!dev_valid_name(name))
1122 if (strchr(name, '%'))
1123 return dev_alloc_name_ns(net, dev, name);
1124 else if (__dev_get_by_name(net, name))
1126 else if (dev->name != name)
1127 strlcpy(dev->name, name, IFNAMSIZ);
1131 EXPORT_SYMBOL(dev_get_valid_name);
1134 * dev_change_name - change name of a device
1136 * @newname: name (or format string) must be at least IFNAMSIZ
1138 * Change name of a device, can pass format strings "eth%d".
1141 int dev_change_name(struct net_device *dev, const char *newname)
1143 unsigned char old_assign_type;
1144 char oldname[IFNAMSIZ];
1150 BUG_ON(!dev_net(dev));
1153 if (dev->flags & IFF_UP)
1156 down_write(&devnet_rename_sem);
1158 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1159 up_write(&devnet_rename_sem);
1163 memcpy(oldname, dev->name, IFNAMSIZ);
1165 err = dev_get_valid_name(net, dev, newname);
1167 up_write(&devnet_rename_sem);
1171 if (oldname[0] && !strchr(oldname, '%'))
1172 netdev_info(dev, "renamed from %s\n", oldname);
1174 old_assign_type = dev->name_assign_type;
1175 dev->name_assign_type = NET_NAME_RENAMED;
1178 ret = device_rename(&dev->dev, dev->name);
1180 memcpy(dev->name, oldname, IFNAMSIZ);
1181 dev->name_assign_type = old_assign_type;
1182 up_write(&devnet_rename_sem);
1186 up_write(&devnet_rename_sem);
1188 netdev_adjacent_rename_links(dev, oldname);
1190 write_lock_bh(&dev_base_lock);
1191 hlist_del_rcu(&dev->name_hlist);
1192 write_unlock_bh(&dev_base_lock);
1196 write_lock_bh(&dev_base_lock);
1197 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1198 write_unlock_bh(&dev_base_lock);
1200 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1201 ret = notifier_to_errno(ret);
1204 /* err >= 0 after dev_alloc_name() or stores the first errno */
1207 down_write(&devnet_rename_sem);
1208 memcpy(dev->name, oldname, IFNAMSIZ);
1209 memcpy(oldname, newname, IFNAMSIZ);
1210 dev->name_assign_type = old_assign_type;
1211 old_assign_type = NET_NAME_RENAMED;
1214 pr_err("%s: name change rollback failed: %d\n",
1223 * dev_set_alias - change ifalias of a device
1225 * @alias: name up to IFALIASZ
1226 * @len: limit of bytes to copy from info
1228 * Set ifalias for a device,
1230 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1236 if (len >= IFALIASZ)
1240 kfree(dev->ifalias);
1241 dev->ifalias = NULL;
1245 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1248 dev->ifalias = new_ifalias;
1249 memcpy(dev->ifalias, alias, len);
1250 dev->ifalias[len] = 0;
1257 * netdev_features_change - device changes features
1258 * @dev: device to cause notification
1260 * Called to indicate a device has changed features.
1262 void netdev_features_change(struct net_device *dev)
1264 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1266 EXPORT_SYMBOL(netdev_features_change);
1269 * netdev_state_change - device changes state
1270 * @dev: device to cause notification
1272 * Called to indicate a device has changed state. This function calls
1273 * the notifier chains for netdev_chain and sends a NEWLINK message
1274 * to the routing socket.
1276 void netdev_state_change(struct net_device *dev)
1278 if (dev->flags & IFF_UP) {
1279 struct netdev_notifier_change_info change_info;
1281 change_info.flags_changed = 0;
1282 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1284 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1287 EXPORT_SYMBOL(netdev_state_change);
1290 * netdev_notify_peers - notify network peers about existence of @dev
1291 * @dev: network device
1293 * Generate traffic such that interested network peers are aware of
1294 * @dev, such as by generating a gratuitous ARP. This may be used when
1295 * a device wants to inform the rest of the network about some sort of
1296 * reconfiguration such as a failover event or virtual machine
1299 void netdev_notify_peers(struct net_device *dev)
1302 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1303 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1306 EXPORT_SYMBOL(netdev_notify_peers);
1308 static int __dev_open(struct net_device *dev)
1310 const struct net_device_ops *ops = dev->netdev_ops;
1315 if (!netif_device_present(dev))
1318 /* Block netpoll from trying to do any rx path servicing.
1319 * If we don't do this there is a chance ndo_poll_controller
1320 * or ndo_poll may be running while we open the device
1322 netpoll_poll_disable(dev);
1324 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1325 ret = notifier_to_errno(ret);
1329 set_bit(__LINK_STATE_START, &dev->state);
1331 if (ops->ndo_validate_addr)
1332 ret = ops->ndo_validate_addr(dev);
1334 if (!ret && ops->ndo_open)
1335 ret = ops->ndo_open(dev);
1337 netpoll_poll_enable(dev);
1340 clear_bit(__LINK_STATE_START, &dev->state);
1342 dev->flags |= IFF_UP;
1343 dev_set_rx_mode(dev);
1345 add_device_randomness(dev->dev_addr, dev->addr_len);
1352 * dev_open - prepare an interface for use.
1353 * @dev: device to open
1355 * Takes a device from down to up state. The device's private open
1356 * function is invoked and then the multicast lists are loaded. Finally
1357 * the device is moved into the up state and a %NETDEV_UP message is
1358 * sent to the netdev notifier chain.
1360 * Calling this function on an active interface is a nop. On a failure
1361 * a negative errno code is returned.
1363 int dev_open(struct net_device *dev)
1367 if (dev->flags & IFF_UP)
1370 ret = __dev_open(dev);
1374 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1375 call_netdevice_notifiers(NETDEV_UP, dev);
1379 EXPORT_SYMBOL(dev_open);
1381 static int __dev_close_many(struct list_head *head)
1383 struct net_device *dev;
1388 list_for_each_entry(dev, head, close_list) {
1389 /* Temporarily disable netpoll until the interface is down */
1390 netpoll_poll_disable(dev);
1392 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1394 clear_bit(__LINK_STATE_START, &dev->state);
1396 /* Synchronize to scheduled poll. We cannot touch poll list, it
1397 * can be even on different cpu. So just clear netif_running().
1399 * dev->stop() will invoke napi_disable() on all of it's
1400 * napi_struct instances on this device.
1402 smp_mb__after_atomic(); /* Commit netif_running(). */
1405 dev_deactivate_many(head);
1407 list_for_each_entry(dev, head, close_list) {
1408 const struct net_device_ops *ops = dev->netdev_ops;
1411 * Call the device specific close. This cannot fail.
1412 * Only if device is UP
1414 * We allow it to be called even after a DETACH hot-plug
1420 dev->flags &= ~IFF_UP;
1421 netpoll_poll_enable(dev);
1427 static int __dev_close(struct net_device *dev)
1432 list_add(&dev->close_list, &single);
1433 retval = __dev_close_many(&single);
1439 int dev_close_many(struct list_head *head, bool unlink)
1441 struct net_device *dev, *tmp;
1443 /* Remove the devices that don't need to be closed */
1444 list_for_each_entry_safe(dev, tmp, head, close_list)
1445 if (!(dev->flags & IFF_UP))
1446 list_del_init(&dev->close_list);
1448 __dev_close_many(head);
1450 list_for_each_entry_safe(dev, tmp, head, close_list) {
1451 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1452 call_netdevice_notifiers(NETDEV_DOWN, dev);
1454 list_del_init(&dev->close_list);
1459 EXPORT_SYMBOL(dev_close_many);
1462 * dev_close - shutdown an interface.
1463 * @dev: device to shutdown
1465 * This function moves an active device into down state. A
1466 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1467 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1470 int dev_close(struct net_device *dev)
1472 if (dev->flags & IFF_UP) {
1475 list_add(&dev->close_list, &single);
1476 dev_close_many(&single, true);
1481 EXPORT_SYMBOL(dev_close);
1485 * dev_disable_lro - disable Large Receive Offload on a device
1488 * Disable Large Receive Offload (LRO) on a net device. Must be
1489 * called under RTNL. This is needed if received packets may be
1490 * forwarded to another interface.
1492 void dev_disable_lro(struct net_device *dev)
1494 struct net_device *lower_dev;
1495 struct list_head *iter;
1497 dev->wanted_features &= ~NETIF_F_LRO;
1498 netdev_update_features(dev);
1500 if (unlikely(dev->features & NETIF_F_LRO))
1501 netdev_WARN(dev, "failed to disable LRO!\n");
1503 netdev_for_each_lower_dev(dev, lower_dev, iter)
1504 dev_disable_lro(lower_dev);
1506 EXPORT_SYMBOL(dev_disable_lro);
1508 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1509 struct net_device *dev)
1511 struct netdev_notifier_info info;
1513 netdev_notifier_info_init(&info, dev);
1514 return nb->notifier_call(nb, val, &info);
1517 static int dev_boot_phase = 1;
1520 * register_netdevice_notifier - register a network notifier block
1523 * Register a notifier to be called when network device events occur.
1524 * The notifier passed is linked into the kernel structures and must
1525 * not be reused until it has been unregistered. A negative errno code
1526 * is returned on a failure.
1528 * When registered all registration and up events are replayed
1529 * to the new notifier to allow device to have a race free
1530 * view of the network device list.
1533 int register_netdevice_notifier(struct notifier_block *nb)
1535 struct net_device *dev;
1536 struct net_device *last;
1541 err = raw_notifier_chain_register(&netdev_chain, nb);
1547 for_each_netdev(net, dev) {
1548 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1549 err = notifier_to_errno(err);
1553 if (!(dev->flags & IFF_UP))
1556 call_netdevice_notifier(nb, NETDEV_UP, dev);
1567 for_each_netdev(net, dev) {
1571 if (dev->flags & IFF_UP) {
1572 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1574 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1576 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1581 raw_notifier_chain_unregister(&netdev_chain, nb);
1584 EXPORT_SYMBOL(register_netdevice_notifier);
1587 * unregister_netdevice_notifier - unregister a network notifier block
1590 * Unregister a notifier previously registered by
1591 * register_netdevice_notifier(). The notifier is unlinked into the
1592 * kernel structures and may then be reused. A negative errno code
1593 * is returned on a failure.
1595 * After unregistering unregister and down device events are synthesized
1596 * for all devices on the device list to the removed notifier to remove
1597 * the need for special case cleanup code.
1600 int unregister_netdevice_notifier(struct notifier_block *nb)
1602 struct net_device *dev;
1607 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1612 for_each_netdev(net, dev) {
1613 if (dev->flags & IFF_UP) {
1614 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1616 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1618 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1625 EXPORT_SYMBOL(unregister_netdevice_notifier);
1628 * call_netdevice_notifiers_info - call all network notifier blocks
1629 * @val: value passed unmodified to notifier function
1630 * @dev: net_device pointer passed unmodified to notifier function
1631 * @info: notifier information data
1633 * Call all network notifier blocks. Parameters and return value
1634 * are as for raw_notifier_call_chain().
1637 static int call_netdevice_notifiers_info(unsigned long val,
1638 struct net_device *dev,
1639 struct netdev_notifier_info *info)
1642 netdev_notifier_info_init(info, dev);
1643 return raw_notifier_call_chain(&netdev_chain, val, info);
1647 * call_netdevice_notifiers - call all network notifier blocks
1648 * @val: value passed unmodified to notifier function
1649 * @dev: net_device pointer passed unmodified to notifier function
1651 * Call all network notifier blocks. Parameters and return value
1652 * are as for raw_notifier_call_chain().
1655 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1657 struct netdev_notifier_info info;
1659 return call_netdevice_notifiers_info(val, dev, &info);
1661 EXPORT_SYMBOL(call_netdevice_notifiers);
1664 * call_netdevice_notifiers_mtu - call all network notifier blocks
1665 * @val: value passed unmodified to notifier function
1666 * @dev: net_device pointer passed unmodified to notifier function
1667 * @arg: additional u32 argument passed to the notifier function
1669 * Call all network notifier blocks. Parameters and return value
1670 * are as for raw_notifier_call_chain().
1672 static int call_netdevice_notifiers_mtu(unsigned long val,
1673 struct net_device *dev, u32 arg)
1675 struct netdev_notifier_info_ext info = {
1680 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
1682 return call_netdevice_notifiers_info(val, dev, &info.info);
1685 #ifdef CONFIG_NET_INGRESS
1686 static struct static_key ingress_needed __read_mostly;
1688 void net_inc_ingress_queue(void)
1690 static_key_slow_inc(&ingress_needed);
1692 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1694 void net_dec_ingress_queue(void)
1696 static_key_slow_dec(&ingress_needed);
1698 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1701 #ifdef CONFIG_NET_EGRESS
1702 static struct static_key egress_needed __read_mostly;
1704 void net_inc_egress_queue(void)
1706 static_key_slow_inc(&egress_needed);
1708 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1710 void net_dec_egress_queue(void)
1712 static_key_slow_dec(&egress_needed);
1714 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1717 static struct static_key netstamp_needed __read_mostly;
1718 #ifdef HAVE_JUMP_LABEL
1719 static atomic_t netstamp_needed_deferred;
1720 static atomic_t netstamp_wanted;
1721 static void netstamp_clear(struct work_struct *work)
1723 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1726 wanted = atomic_add_return(deferred, &netstamp_wanted);
1728 static_key_enable(&netstamp_needed);
1730 static_key_disable(&netstamp_needed);
1732 static DECLARE_WORK(netstamp_work, netstamp_clear);
1735 void net_enable_timestamp(void)
1737 #ifdef HAVE_JUMP_LABEL
1741 wanted = atomic_read(&netstamp_wanted);
1744 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1747 atomic_inc(&netstamp_needed_deferred);
1748 schedule_work(&netstamp_work);
1750 static_key_slow_inc(&netstamp_needed);
1753 EXPORT_SYMBOL(net_enable_timestamp);
1755 void net_disable_timestamp(void)
1757 #ifdef HAVE_JUMP_LABEL
1761 wanted = atomic_read(&netstamp_wanted);
1764 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
1767 atomic_dec(&netstamp_needed_deferred);
1768 schedule_work(&netstamp_work);
1770 static_key_slow_dec(&netstamp_needed);
1773 EXPORT_SYMBOL(net_disable_timestamp);
1775 static inline void net_timestamp_set(struct sk_buff *skb)
1777 skb->tstamp.tv64 = 0;
1778 if (static_key_false(&netstamp_needed))
1779 __net_timestamp(skb);
1782 #define net_timestamp_check(COND, SKB) \
1783 if (static_key_false(&netstamp_needed)) { \
1784 if ((COND) && !(SKB)->tstamp.tv64) \
1785 __net_timestamp(SKB); \
1788 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1792 if (!(dev->flags & IFF_UP))
1795 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1796 if (skb->len <= len)
1799 /* if TSO is enabled, we don't care about the length as the packet
1800 * could be forwarded without being segmented before
1802 if (skb_is_gso(skb))
1807 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1809 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1811 int ret = ____dev_forward_skb(dev, skb);
1814 skb->protocol = eth_type_trans(skb, dev);
1815 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1820 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1823 * dev_forward_skb - loopback an skb to another netif
1825 * @dev: destination network device
1826 * @skb: buffer to forward
1829 * NET_RX_SUCCESS (no congestion)
1830 * NET_RX_DROP (packet was dropped, but freed)
1832 * dev_forward_skb can be used for injecting an skb from the
1833 * start_xmit function of one device into the receive queue
1834 * of another device.
1836 * The receiving device may be in another namespace, so
1837 * we have to clear all information in the skb that could
1838 * impact namespace isolation.
1840 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1842 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1844 EXPORT_SYMBOL_GPL(dev_forward_skb);
1846 static inline int deliver_skb(struct sk_buff *skb,
1847 struct packet_type *pt_prev,
1848 struct net_device *orig_dev)
1850 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1852 atomic_inc(&skb->users);
1853 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1856 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1857 struct packet_type **pt,
1858 struct net_device *orig_dev,
1860 struct list_head *ptype_list)
1862 struct packet_type *ptype, *pt_prev = *pt;
1864 list_for_each_entry_rcu(ptype, ptype_list, list) {
1865 if (ptype->type != type)
1868 deliver_skb(skb, pt_prev, orig_dev);
1874 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1876 if (!ptype->af_packet_priv || !skb->sk)
1879 if (ptype->id_match)
1880 return ptype->id_match(ptype, skb->sk);
1881 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1888 * Support routine. Sends outgoing frames to any network
1889 * taps currently in use.
1892 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1894 struct packet_type *ptype;
1895 struct sk_buff *skb2 = NULL;
1896 struct packet_type *pt_prev = NULL;
1897 struct list_head *ptype_list = &ptype_all;
1901 list_for_each_entry_rcu(ptype, ptype_list, list) {
1902 /* Never send packets back to the socket
1903 * they originated from - MvS (miquels@drinkel.ow.org)
1905 if (skb_loop_sk(ptype, skb))
1909 deliver_skb(skb2, pt_prev, skb->dev);
1914 /* need to clone skb, done only once */
1915 skb2 = skb_clone(skb, GFP_ATOMIC);
1919 net_timestamp_set(skb2);
1921 /* skb->nh should be correctly
1922 * set by sender, so that the second statement is
1923 * just protection against buggy protocols.
1925 skb_reset_mac_header(skb2);
1927 if (skb_network_header(skb2) < skb2->data ||
1928 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1929 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1930 ntohs(skb2->protocol),
1932 skb_reset_network_header(skb2);
1935 skb2->transport_header = skb2->network_header;
1936 skb2->pkt_type = PACKET_OUTGOING;
1940 if (ptype_list == &ptype_all) {
1941 ptype_list = &dev->ptype_all;
1946 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1949 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
1952 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1953 * @dev: Network device
1954 * @txq: number of queues available
1956 * If real_num_tx_queues is changed the tc mappings may no longer be
1957 * valid. To resolve this verify the tc mapping remains valid and if
1958 * not NULL the mapping. With no priorities mapping to this
1959 * offset/count pair it will no longer be used. In the worst case TC0
1960 * is invalid nothing can be done so disable priority mappings. If is
1961 * expected that drivers will fix this mapping if they can before
1962 * calling netif_set_real_num_tx_queues.
1964 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1967 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1969 /* If TC0 is invalidated disable TC mapping */
1970 if (tc->offset + tc->count > txq) {
1971 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1976 /* Invalidated prio to tc mappings set to TC0 */
1977 for (i = 1; i < TC_BITMASK + 1; i++) {
1978 int q = netdev_get_prio_tc_map(dev, i);
1980 tc = &dev->tc_to_txq[q];
1981 if (tc->offset + tc->count > txq) {
1982 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1984 netdev_set_prio_tc_map(dev, i, 0);
1990 static DEFINE_MUTEX(xps_map_mutex);
1991 #define xmap_dereference(P) \
1992 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1994 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1997 struct xps_map *map = NULL;
2001 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2003 for (pos = 0; map && pos < map->len; pos++) {
2004 if (map->queues[pos] == index) {
2006 map->queues[pos] = map->queues[--map->len];
2008 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
2009 kfree_rcu(map, rcu);
2019 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2021 struct xps_dev_maps *dev_maps;
2023 bool active = false;
2025 mutex_lock(&xps_map_mutex);
2026 dev_maps = xmap_dereference(dev->xps_maps);
2031 for_each_possible_cpu(cpu) {
2032 for (i = index; i < dev->num_tx_queues; i++) {
2033 if (!remove_xps_queue(dev_maps, cpu, i))
2036 if (i == dev->num_tx_queues)
2041 RCU_INIT_POINTER(dev->xps_maps, NULL);
2042 kfree_rcu(dev_maps, rcu);
2045 for (i = index; i < dev->num_tx_queues; i++)
2046 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2050 mutex_unlock(&xps_map_mutex);
2053 static struct xps_map *expand_xps_map(struct xps_map *map,
2056 struct xps_map *new_map;
2057 int alloc_len = XPS_MIN_MAP_ALLOC;
2060 for (pos = 0; map && pos < map->len; pos++) {
2061 if (map->queues[pos] != index)
2066 /* Need to add queue to this CPU's existing map */
2068 if (pos < map->alloc_len)
2071 alloc_len = map->alloc_len * 2;
2074 /* Need to allocate new map to store queue on this CPU's map */
2075 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2080 for (i = 0; i < pos; i++)
2081 new_map->queues[i] = map->queues[i];
2082 new_map->alloc_len = alloc_len;
2088 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2091 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2092 struct xps_map *map, *new_map;
2093 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2094 int cpu, numa_node_id = -2;
2095 bool active = false;
2097 mutex_lock(&xps_map_mutex);
2099 dev_maps = xmap_dereference(dev->xps_maps);
2101 /* allocate memory for queue storage */
2102 for_each_online_cpu(cpu) {
2103 if (!cpumask_test_cpu(cpu, mask))
2107 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2108 if (!new_dev_maps) {
2109 mutex_unlock(&xps_map_mutex);
2113 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2116 map = expand_xps_map(map, cpu, index);
2120 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2124 goto out_no_new_maps;
2126 for_each_possible_cpu(cpu) {
2127 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2128 /* add queue to CPU maps */
2131 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2132 while ((pos < map->len) && (map->queues[pos] != index))
2135 if (pos == map->len)
2136 map->queues[map->len++] = index;
2138 if (numa_node_id == -2)
2139 numa_node_id = cpu_to_node(cpu);
2140 else if (numa_node_id != cpu_to_node(cpu))
2143 } else if (dev_maps) {
2144 /* fill in the new device map from the old device map */
2145 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2146 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2151 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2153 /* Cleanup old maps */
2155 for_each_possible_cpu(cpu) {
2156 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2157 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2158 if (map && map != new_map)
2159 kfree_rcu(map, rcu);
2162 kfree_rcu(dev_maps, rcu);
2165 dev_maps = new_dev_maps;
2169 /* update Tx queue numa node */
2170 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2171 (numa_node_id >= 0) ? numa_node_id :
2177 /* removes queue from unused CPUs */
2178 for_each_possible_cpu(cpu) {
2179 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2182 if (remove_xps_queue(dev_maps, cpu, index))
2186 /* free map if not active */
2188 RCU_INIT_POINTER(dev->xps_maps, NULL);
2189 kfree_rcu(dev_maps, rcu);
2193 mutex_unlock(&xps_map_mutex);
2197 /* remove any maps that we added */
2198 for_each_possible_cpu(cpu) {
2199 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2200 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2202 if (new_map && new_map != map)
2206 mutex_unlock(&xps_map_mutex);
2208 kfree(new_dev_maps);
2211 EXPORT_SYMBOL(netif_set_xps_queue);
2215 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2216 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2218 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2223 disabling = txq < dev->real_num_tx_queues;
2225 if (txq < 1 || txq > dev->num_tx_queues)
2228 if (dev->reg_state == NETREG_REGISTERED ||
2229 dev->reg_state == NETREG_UNREGISTERING) {
2232 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2238 netif_setup_tc(dev, txq);
2240 dev->real_num_tx_queues = txq;
2244 qdisc_reset_all_tx_gt(dev, txq);
2246 netif_reset_xps_queues_gt(dev, txq);
2250 dev->real_num_tx_queues = txq;
2255 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2259 * netif_set_real_num_rx_queues - set actual number of RX queues used
2260 * @dev: Network device
2261 * @rxq: Actual number of RX queues
2263 * This must be called either with the rtnl_lock held or before
2264 * registration of the net device. Returns 0 on success, or a
2265 * negative error code. If called before registration, it always
2268 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2272 if (rxq < 1 || rxq > dev->num_rx_queues)
2275 if (dev->reg_state == NETREG_REGISTERED) {
2278 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2284 dev->real_num_rx_queues = rxq;
2287 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2291 * netif_get_num_default_rss_queues - default number of RSS queues
2293 * This routine should set an upper limit on the number of RSS queues
2294 * used by default by multiqueue devices.
2296 int netif_get_num_default_rss_queues(void)
2298 return is_kdump_kernel() ?
2299 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2301 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2303 static void __netif_reschedule(struct Qdisc *q)
2305 struct softnet_data *sd;
2306 unsigned long flags;
2308 local_irq_save(flags);
2309 sd = this_cpu_ptr(&softnet_data);
2310 q->next_sched = NULL;
2311 *sd->output_queue_tailp = q;
2312 sd->output_queue_tailp = &q->next_sched;
2313 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2314 local_irq_restore(flags);
2317 void __netif_schedule(struct Qdisc *q)
2319 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2320 __netif_reschedule(q);
2322 EXPORT_SYMBOL(__netif_schedule);
2324 struct dev_kfree_skb_cb {
2325 enum skb_free_reason reason;
2328 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2330 return (struct dev_kfree_skb_cb *)skb->cb;
2333 void netif_schedule_queue(struct netdev_queue *txq)
2336 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2337 struct Qdisc *q = rcu_dereference(txq->qdisc);
2339 __netif_schedule(q);
2343 EXPORT_SYMBOL(netif_schedule_queue);
2346 * netif_wake_subqueue - allow sending packets on subqueue
2347 * @dev: network device
2348 * @queue_index: sub queue index
2350 * Resume individual transmit queue of a device with multiple transmit queues.
2352 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2354 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2356 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2360 q = rcu_dereference(txq->qdisc);
2361 __netif_schedule(q);
2365 EXPORT_SYMBOL(netif_wake_subqueue);
2367 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2369 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2373 q = rcu_dereference(dev_queue->qdisc);
2374 __netif_schedule(q);
2378 EXPORT_SYMBOL(netif_tx_wake_queue);
2380 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2382 unsigned long flags;
2387 if (likely(atomic_read(&skb->users) == 1)) {
2389 atomic_set(&skb->users, 0);
2390 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2393 get_kfree_skb_cb(skb)->reason = reason;
2394 local_irq_save(flags);
2395 skb->next = __this_cpu_read(softnet_data.completion_queue);
2396 __this_cpu_write(softnet_data.completion_queue, skb);
2397 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2398 local_irq_restore(flags);
2400 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2402 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2404 if (in_irq() || irqs_disabled())
2405 __dev_kfree_skb_irq(skb, reason);
2409 EXPORT_SYMBOL(__dev_kfree_skb_any);
2413 * netif_device_detach - mark device as removed
2414 * @dev: network device
2416 * Mark device as removed from system and therefore no longer available.
2418 void netif_device_detach(struct net_device *dev)
2420 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2421 netif_running(dev)) {
2422 netif_tx_stop_all_queues(dev);
2425 EXPORT_SYMBOL(netif_device_detach);
2428 * netif_device_attach - mark device as attached
2429 * @dev: network device
2431 * Mark device as attached from system and restart if needed.
2433 void netif_device_attach(struct net_device *dev)
2435 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2436 netif_running(dev)) {
2437 netif_tx_wake_all_queues(dev);
2438 __netdev_watchdog_up(dev);
2441 EXPORT_SYMBOL(netif_device_attach);
2444 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2445 * to be used as a distribution range.
2447 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2448 unsigned int num_tx_queues)
2452 u16 qcount = num_tx_queues;
2454 if (skb_rx_queue_recorded(skb)) {
2455 hash = skb_get_rx_queue(skb);
2456 while (unlikely(hash >= num_tx_queues))
2457 hash -= num_tx_queues;
2462 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2463 qoffset = dev->tc_to_txq[tc].offset;
2464 qcount = dev->tc_to_txq[tc].count;
2467 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2469 EXPORT_SYMBOL(__skb_tx_hash);
2471 static void skb_warn_bad_offload(const struct sk_buff *skb)
2473 static const netdev_features_t null_features;
2474 struct net_device *dev = skb->dev;
2475 const char *name = "";
2477 if (!net_ratelimit())
2481 if (dev->dev.parent)
2482 name = dev_driver_string(dev->dev.parent);
2484 name = netdev_name(dev);
2486 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2487 "gso_type=%d ip_summed=%d\n",
2488 name, dev ? &dev->features : &null_features,
2489 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2490 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2491 skb_shinfo(skb)->gso_type, skb->ip_summed);
2495 * Invalidate hardware checksum when packet is to be mangled, and
2496 * complete checksum manually on outgoing path.
2498 int skb_checksum_help(struct sk_buff *skb)
2501 int ret = 0, offset;
2503 if (skb->ip_summed == CHECKSUM_COMPLETE)
2504 goto out_set_summed;
2506 if (unlikely(skb_shinfo(skb)->gso_size)) {
2507 skb_warn_bad_offload(skb);
2511 /* Before computing a checksum, we should make sure no frag could
2512 * be modified by an external entity : checksum could be wrong.
2514 if (skb_has_shared_frag(skb)) {
2515 ret = __skb_linearize(skb);
2520 offset = skb_checksum_start_offset(skb);
2521 BUG_ON(offset >= skb_headlen(skb));
2522 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2524 offset += skb->csum_offset;
2525 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2527 if (skb_cloned(skb) &&
2528 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2529 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2534 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2536 skb->ip_summed = CHECKSUM_NONE;
2540 EXPORT_SYMBOL(skb_checksum_help);
2542 /* skb_csum_offload_check - Driver helper function to determine if a device
2543 * with limited checksum offload capabilities is able to offload the checksum
2544 * for a given packet.
2547 * skb - sk_buff for the packet in question
2548 * spec - contains the description of what device can offload
2549 * csum_encapped - returns true if the checksum being offloaded is
2550 * encpasulated. That is it is checksum for the transport header
2551 * in the inner headers.
2552 * checksum_help - when set indicates that helper function should
2553 * call skb_checksum_help if offload checks fail
2556 * true: Packet has passed the checksum checks and should be offloadable to
2557 * the device (a driver may still need to check for additional
2558 * restrictions of its device)
2559 * false: Checksum is not offloadable. If checksum_help was set then
2560 * skb_checksum_help was called to resolve checksum for non-GSO
2561 * packets and when IP protocol is not SCTP
2563 bool __skb_csum_offload_chk(struct sk_buff *skb,
2564 const struct skb_csum_offl_spec *spec,
2565 bool *csum_encapped,
2569 struct ipv6hdr *ipv6;
2574 if (skb->protocol == htons(ETH_P_8021Q) ||
2575 skb->protocol == htons(ETH_P_8021AD)) {
2576 if (!spec->vlan_okay)
2580 /* We check whether the checksum refers to a transport layer checksum in
2581 * the outermost header or an encapsulated transport layer checksum that
2582 * corresponds to the inner headers of the skb. If the checksum is for
2583 * something else in the packet we need help.
2585 if (skb_checksum_start_offset(skb) == skb_transport_offset(skb)) {
2586 /* Non-encapsulated checksum */
2587 protocol = eproto_to_ipproto(vlan_get_protocol(skb));
2588 nhdr = skb_network_header(skb);
2589 *csum_encapped = false;
2590 if (spec->no_not_encapped)
2592 } else if (skb->encapsulation && spec->encap_okay &&
2593 skb_checksum_start_offset(skb) ==
2594 skb_inner_transport_offset(skb)) {
2595 /* Encapsulated checksum */
2596 *csum_encapped = true;
2597 switch (skb->inner_protocol_type) {
2598 case ENCAP_TYPE_ETHER:
2599 protocol = eproto_to_ipproto(skb->inner_protocol);
2601 case ENCAP_TYPE_IPPROTO:
2602 protocol = skb->inner_protocol;
2605 nhdr = skb_inner_network_header(skb);
2612 if (!spec->ipv4_okay)
2615 ip_proto = iph->protocol;
2616 if (iph->ihl != 5 && !spec->ip_options_okay)
2620 if (!spec->ipv6_okay)
2622 if (spec->no_encapped_ipv6 && *csum_encapped)
2625 nhdr += sizeof(*ipv6);
2626 ip_proto = ipv6->nexthdr;
2635 if (!spec->tcp_okay ||
2636 skb->csum_offset != offsetof(struct tcphdr, check))
2640 if (!spec->udp_okay ||
2641 skb->csum_offset != offsetof(struct udphdr, check))
2645 if (!spec->sctp_okay ||
2646 skb->csum_offset != offsetof(struct sctphdr, checksum))
2650 case NEXTHDR_ROUTING:
2651 case NEXTHDR_DEST: {
2654 if (protocol != IPPROTO_IPV6 || !spec->ext_hdrs_okay)
2657 ip_proto = opthdr[0];
2658 nhdr += (opthdr[1] + 1) << 3;
2660 goto ip_proto_again;
2666 /* Passed the tests for offloading checksum */
2670 if (csum_help && !skb_shinfo(skb)->gso_size)
2671 skb_checksum_help(skb);
2675 EXPORT_SYMBOL(__skb_csum_offload_chk);
2677 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2679 __be16 type = skb->protocol;
2681 /* Tunnel gso handlers can set protocol to ethernet. */
2682 if (type == htons(ETH_P_TEB)) {
2685 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2688 eth = (struct ethhdr *)skb->data;
2689 type = eth->h_proto;
2692 return __vlan_get_protocol(skb, type, depth);
2696 * skb_mac_gso_segment - mac layer segmentation handler.
2697 * @skb: buffer to segment
2698 * @features: features for the output path (see dev->features)
2700 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2701 netdev_features_t features)
2703 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2704 struct packet_offload *ptype;
2705 int vlan_depth = skb->mac_len;
2706 __be16 type = skb_network_protocol(skb, &vlan_depth);
2708 if (unlikely(!type))
2709 return ERR_PTR(-EINVAL);
2711 __skb_pull(skb, vlan_depth);
2714 list_for_each_entry_rcu(ptype, &offload_base, list) {
2715 if (ptype->type == type && ptype->callbacks.gso_segment) {
2716 segs = ptype->callbacks.gso_segment(skb, features);
2722 __skb_push(skb, skb->data - skb_mac_header(skb));
2726 EXPORT_SYMBOL(skb_mac_gso_segment);
2729 /* openvswitch calls this on rx path, so we need a different check.
2731 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2734 return skb->ip_summed != CHECKSUM_PARTIAL &&
2735 skb->ip_summed != CHECKSUM_UNNECESSARY;
2737 return skb->ip_summed == CHECKSUM_NONE;
2741 * __skb_gso_segment - Perform segmentation on skb.
2742 * @skb: buffer to segment
2743 * @features: features for the output path (see dev->features)
2744 * @tx_path: whether it is called in TX path
2746 * This function segments the given skb and returns a list of segments.
2748 * It may return NULL if the skb requires no segmentation. This is
2749 * only possible when GSO is used for verifying header integrity.
2751 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2753 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2754 netdev_features_t features, bool tx_path)
2756 struct sk_buff *segs;
2758 if (unlikely(skb_needs_check(skb, tx_path))) {
2761 /* We're going to init ->check field in TCP or UDP header */
2762 err = skb_cow_head(skb, 0);
2764 return ERR_PTR(err);
2767 /* Only report GSO partial support if it will enable us to
2768 * support segmentation on this frame without needing additional
2771 if (features & NETIF_F_GSO_PARTIAL) {
2772 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
2773 struct net_device *dev = skb->dev;
2775 partial_features |= dev->features & dev->gso_partial_features;
2776 if (!skb_gso_ok(skb, features | partial_features))
2777 features &= ~NETIF_F_GSO_PARTIAL;
2780 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2781 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2783 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2784 SKB_GSO_CB(skb)->encap_level = 0;
2786 skb_reset_mac_header(skb);
2787 skb_reset_mac_len(skb);
2789 segs = skb_mac_gso_segment(skb, features);
2791 if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
2792 skb_warn_bad_offload(skb);
2796 EXPORT_SYMBOL(__skb_gso_segment);
2798 /* Take action when hardware reception checksum errors are detected. */
2800 void netdev_rx_csum_fault(struct net_device *dev)
2802 if (net_ratelimit()) {
2803 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2807 EXPORT_SYMBOL(netdev_rx_csum_fault);
2810 /* Actually, we should eliminate this check as soon as we know, that:
2811 * 1. IOMMU is present and allows to map all the memory.
2812 * 2. No high memory really exists on this machine.
2815 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2817 #ifdef CONFIG_HIGHMEM
2819 if (!(dev->features & NETIF_F_HIGHDMA)) {
2820 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2821 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2822 if (PageHighMem(skb_frag_page(frag)))
2827 if (PCI_DMA_BUS_IS_PHYS) {
2828 struct device *pdev = dev->dev.parent;
2832 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2833 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2834 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2835 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2843 /* If MPLS offload request, verify we are testing hardware MPLS features
2844 * instead of standard features for the netdev.
2846 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2847 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2848 netdev_features_t features,
2851 if (eth_p_mpls(type))
2852 features &= skb->dev->mpls_features;
2857 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2858 netdev_features_t features,
2865 static netdev_features_t harmonize_features(struct sk_buff *skb,
2866 netdev_features_t features)
2871 type = skb_network_protocol(skb, &tmp);
2872 features = net_mpls_features(skb, features, type);
2874 if (skb->ip_summed != CHECKSUM_NONE &&
2875 !can_checksum_protocol(features, type)) {
2876 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2878 if (illegal_highdma(skb->dev, skb))
2879 features &= ~NETIF_F_SG;
2884 netdev_features_t passthru_features_check(struct sk_buff *skb,
2885 struct net_device *dev,
2886 netdev_features_t features)
2890 EXPORT_SYMBOL(passthru_features_check);
2892 static netdev_features_t dflt_features_check(struct sk_buff *skb,
2893 struct net_device *dev,
2894 netdev_features_t features)
2896 return vlan_features_check(skb, features);
2899 static netdev_features_t gso_features_check(const struct sk_buff *skb,
2900 struct net_device *dev,
2901 netdev_features_t features)
2903 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2905 if (gso_segs > dev->gso_max_segs)
2906 return features & ~NETIF_F_GSO_MASK;
2908 /* Support for GSO partial features requires software
2909 * intervention before we can actually process the packets
2910 * so we need to strip support for any partial features now
2911 * and we can pull them back in after we have partially
2912 * segmented the frame.
2914 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
2915 features &= ~dev->gso_partial_features;
2917 /* Make sure to clear the IPv4 ID mangling feature if the
2918 * IPv4 header has the potential to be fragmented.
2920 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2921 struct iphdr *iph = skb->encapsulation ?
2922 inner_ip_hdr(skb) : ip_hdr(skb);
2924 if (!(iph->frag_off & htons(IP_DF)))
2925 features &= ~NETIF_F_TSO_MANGLEID;
2931 netdev_features_t netif_skb_features(struct sk_buff *skb)
2933 struct net_device *dev = skb->dev;
2934 netdev_features_t features = dev->features;
2936 if (skb_is_gso(skb))
2937 features = gso_features_check(skb, dev, features);
2939 /* If encapsulation offload request, verify we are testing
2940 * hardware encapsulation features instead of standard
2941 * features for the netdev
2943 if (skb->encapsulation)
2944 features &= dev->hw_enc_features;
2946 if (skb_vlan_tagged(skb))
2947 features = netdev_intersect_features(features,
2948 dev->vlan_features |
2949 NETIF_F_HW_VLAN_CTAG_TX |
2950 NETIF_F_HW_VLAN_STAG_TX);
2952 if (dev->netdev_ops->ndo_features_check)
2953 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2956 features &= dflt_features_check(skb, dev, features);
2958 return harmonize_features(skb, features);
2960 EXPORT_SYMBOL(netif_skb_features);
2962 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2963 struct netdev_queue *txq, bool more)
2968 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2969 dev_queue_xmit_nit(skb, dev);
2972 trace_net_dev_start_xmit(skb, dev);
2973 rc = netdev_start_xmit(skb, dev, txq, more);
2974 trace_net_dev_xmit(skb, rc, dev, len);
2979 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2980 struct netdev_queue *txq, int *ret)
2982 struct sk_buff *skb = first;
2983 int rc = NETDEV_TX_OK;
2986 struct sk_buff *next = skb->next;
2989 rc = xmit_one(skb, dev, txq, next != NULL);
2990 if (unlikely(!dev_xmit_complete(rc))) {
2996 if (netif_tx_queue_stopped(txq) && skb) {
2997 rc = NETDEV_TX_BUSY;
3007 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3008 netdev_features_t features)
3010 if (skb_vlan_tag_present(skb) &&
3011 !vlan_hw_offload_capable(features, skb->vlan_proto))
3012 skb = __vlan_hwaccel_push_inside(skb);
3016 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
3018 netdev_features_t features;
3020 features = netif_skb_features(skb);
3021 skb = validate_xmit_vlan(skb, features);
3025 if (netif_needs_gso(skb, features)) {
3026 struct sk_buff *segs;
3028 segs = skb_gso_segment(skb, features);
3036 if (skb_needs_linearize(skb, features) &&
3037 __skb_linearize(skb))
3040 /* If packet is not checksummed and device does not
3041 * support checksumming for this protocol, complete
3042 * checksumming here.
3044 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3045 if (skb->encapsulation)
3046 skb_set_inner_transport_header(skb,
3047 skb_checksum_start_offset(skb));
3049 skb_set_transport_header(skb,
3050 skb_checksum_start_offset(skb));
3051 if (!(features & NETIF_F_CSUM_MASK) &&
3052 skb_checksum_help(skb))
3062 atomic_long_inc(&dev->tx_dropped);
3066 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
3068 struct sk_buff *next, *head = NULL, *tail;
3070 for (; skb != NULL; skb = next) {
3074 /* in case skb wont be segmented, point to itself */
3077 skb = validate_xmit_skb(skb, dev);
3085 /* If skb was segmented, skb->prev points to
3086 * the last segment. If not, it still contains skb.
3092 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3094 static void qdisc_pkt_len_init(struct sk_buff *skb)
3096 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3098 qdisc_skb_cb(skb)->pkt_len = skb->len;
3100 /* To get more precise estimation of bytes sent on wire,
3101 * we add to pkt_len the headers size of all segments
3103 if (shinfo->gso_size) {
3104 unsigned int hdr_len;
3105 u16 gso_segs = shinfo->gso_segs;
3107 /* mac layer + network layer */
3108 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3110 /* + transport layer */
3111 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3112 const struct tcphdr *th;
3113 struct tcphdr _tcphdr;
3115 th = skb_header_pointer(skb, skb_transport_offset(skb),
3116 sizeof(_tcphdr), &_tcphdr);
3118 hdr_len += __tcp_hdrlen(th);
3120 struct udphdr _udphdr;
3122 if (skb_header_pointer(skb, skb_transport_offset(skb),
3123 sizeof(_udphdr), &_udphdr))
3124 hdr_len += sizeof(struct udphdr);
3127 if (shinfo->gso_type & SKB_GSO_DODGY)
3128 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3131 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3135 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3136 struct net_device *dev,
3137 struct netdev_queue *txq)
3139 spinlock_t *root_lock = qdisc_lock(q);
3140 struct sk_buff *to_free = NULL;
3144 qdisc_calculate_pkt_len(skb, q);
3146 * Heuristic to force contended enqueues to serialize on a
3147 * separate lock before trying to get qdisc main lock.
3148 * This permits qdisc->running owner to get the lock more
3149 * often and dequeue packets faster.
3151 contended = qdisc_is_running(q);
3152 if (unlikely(contended))
3153 spin_lock(&q->busylock);
3155 spin_lock(root_lock);
3156 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3157 __qdisc_drop(skb, &to_free);
3159 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3160 qdisc_run_begin(q)) {
3162 * This is a work-conserving queue; there are no old skbs
3163 * waiting to be sent out; and the qdisc is not running -
3164 * xmit the skb directly.
3167 qdisc_bstats_update(q, skb);
3169 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3170 if (unlikely(contended)) {
3171 spin_unlock(&q->busylock);
3178 rc = NET_XMIT_SUCCESS;
3180 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3181 if (qdisc_run_begin(q)) {
3182 if (unlikely(contended)) {
3183 spin_unlock(&q->busylock);
3189 spin_unlock(root_lock);
3190 if (unlikely(to_free))
3191 kfree_skb_list(to_free);
3192 if (unlikely(contended))
3193 spin_unlock(&q->busylock);
3197 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3198 static void skb_update_prio(struct sk_buff *skb)
3200 const struct netprio_map *map;
3201 const struct sock *sk;
3202 unsigned int prioidx;
3206 map = rcu_dereference_bh(skb->dev->priomap);
3209 sk = skb_to_full_sk(skb);
3213 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3215 if (prioidx < map->priomap_len)
3216 skb->priority = map->priomap[prioidx];
3219 #define skb_update_prio(skb)
3222 DEFINE_PER_CPU(int, xmit_recursion);
3223 EXPORT_SYMBOL(xmit_recursion);
3226 * dev_loopback_xmit - loop back @skb
3227 * @net: network namespace this loopback is happening in
3228 * @sk: sk needed to be a netfilter okfn
3229 * @skb: buffer to transmit
3231 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3233 skb_reset_mac_header(skb);
3234 __skb_pull(skb, skb_network_offset(skb));
3235 skb->pkt_type = PACKET_LOOPBACK;
3236 skb->ip_summed = CHECKSUM_UNNECESSARY;
3237 WARN_ON(!skb_dst(skb));
3242 EXPORT_SYMBOL(dev_loopback_xmit);
3244 #ifdef CONFIG_NET_EGRESS
3245 static struct sk_buff *
3246 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3248 struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list);
3249 struct tcf_result cl_res;
3254 /* skb->tc_verd and qdisc_skb_cb(skb)->pkt_len were already set
3255 * earlier by the caller.
3257 qdisc_bstats_cpu_update(cl->q, skb);
3259 switch (tc_classify(skb, cl, &cl_res, false)) {
3261 case TC_ACT_RECLASSIFY:
3262 skb->tc_index = TC_H_MIN(cl_res.classid);
3265 qdisc_qstats_cpu_drop(cl->q);
3266 *ret = NET_XMIT_DROP;
3271 *ret = NET_XMIT_SUCCESS;
3274 case TC_ACT_REDIRECT:
3275 /* No need to push/pop skb's mac_header here on egress! */
3276 skb_do_redirect(skb);
3277 *ret = NET_XMIT_SUCCESS;
3285 #endif /* CONFIG_NET_EGRESS */
3287 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3290 struct xps_dev_maps *dev_maps;
3291 struct xps_map *map;
3292 int queue_index = -1;
3295 dev_maps = rcu_dereference(dev->xps_maps);
3297 map = rcu_dereference(
3298 dev_maps->cpu_map[skb->sender_cpu - 1]);
3301 queue_index = map->queues[0];
3303 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3305 if (unlikely(queue_index >= dev->real_num_tx_queues))
3317 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3319 struct sock *sk = skb->sk;
3320 int queue_index = sk_tx_queue_get(sk);
3322 if (queue_index < 0 || skb->ooo_okay ||
3323 queue_index >= dev->real_num_tx_queues) {
3324 int new_index = get_xps_queue(dev, skb);
3326 new_index = skb_tx_hash(dev, skb);
3328 if (queue_index != new_index && sk &&
3330 rcu_access_pointer(sk->sk_dst_cache))
3331 sk_tx_queue_set(sk, new_index);
3333 queue_index = new_index;
3339 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3340 struct sk_buff *skb,
3343 int queue_index = 0;
3346 u32 sender_cpu = skb->sender_cpu - 1;
3348 if (sender_cpu >= (u32)NR_CPUS)
3349 skb->sender_cpu = raw_smp_processor_id() + 1;
3352 if (dev->real_num_tx_queues != 1) {
3353 const struct net_device_ops *ops = dev->netdev_ops;
3354 if (ops->ndo_select_queue)
3355 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3358 queue_index = __netdev_pick_tx(dev, skb);
3361 queue_index = netdev_cap_txqueue(dev, queue_index);
3364 skb_set_queue_mapping(skb, queue_index);
3365 return netdev_get_tx_queue(dev, queue_index);
3369 * __dev_queue_xmit - transmit a buffer
3370 * @skb: buffer to transmit
3371 * @accel_priv: private data used for L2 forwarding offload
3373 * Queue a buffer for transmission to a network device. The caller must
3374 * have set the device and priority and built the buffer before calling
3375 * this function. The function can be called from an interrupt.
3377 * A negative errno code is returned on a failure. A success does not
3378 * guarantee the frame will be transmitted as it may be dropped due
3379 * to congestion or traffic shaping.
3381 * -----------------------------------------------------------------------------------
3382 * I notice this method can also return errors from the queue disciplines,
3383 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3386 * Regardless of the return value, the skb is consumed, so it is currently
3387 * difficult to retry a send to this method. (You can bump the ref count
3388 * before sending to hold a reference for retry if you are careful.)
3390 * When calling this method, interrupts MUST be enabled. This is because
3391 * the BH enable code must have IRQs enabled so that it will not deadlock.
3394 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3396 struct net_device *dev = skb->dev;
3397 struct netdev_queue *txq;
3401 skb_reset_mac_header(skb);
3403 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3404 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3406 /* Disable soft irqs for various locks below. Also
3407 * stops preemption for RCU.
3411 skb_update_prio(skb);
3413 qdisc_pkt_len_init(skb);
3414 #ifdef CONFIG_NET_CLS_ACT
3415 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3416 # ifdef CONFIG_NET_EGRESS
3417 if (static_key_false(&egress_needed)) {
3418 skb = sch_handle_egress(skb, &rc, dev);
3424 /* If device/qdisc don't need skb->dst, release it right now while
3425 * its hot in this cpu cache.
3427 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3432 txq = netdev_pick_tx(dev, skb, accel_priv);
3433 q = rcu_dereference_bh(txq->qdisc);
3435 trace_net_dev_queue(skb);
3437 rc = __dev_xmit_skb(skb, q, dev, txq);
3441 /* The device has no queue. Common case for software devices:
3442 loopback, all the sorts of tunnels...
3444 Really, it is unlikely that netif_tx_lock protection is necessary
3445 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3447 However, it is possible, that they rely on protection
3450 Check this and shot the lock. It is not prone from deadlocks.
3451 Either shot noqueue qdisc, it is even simpler 8)
3453 if (dev->flags & IFF_UP) {
3454 int cpu = smp_processor_id(); /* ok because BHs are off */
3456 if (txq->xmit_lock_owner != cpu) {
3457 if (unlikely(__this_cpu_read(xmit_recursion) >
3458 XMIT_RECURSION_LIMIT))
3459 goto recursion_alert;
3461 skb = validate_xmit_skb(skb, dev);
3465 HARD_TX_LOCK(dev, txq, cpu);
3467 if (!netif_xmit_stopped(txq)) {
3468 __this_cpu_inc(xmit_recursion);
3469 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3470 __this_cpu_dec(xmit_recursion);
3471 if (dev_xmit_complete(rc)) {
3472 HARD_TX_UNLOCK(dev, txq);
3476 HARD_TX_UNLOCK(dev, txq);
3477 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3480 /* Recursion is detected! It is possible,
3484 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3490 rcu_read_unlock_bh();
3492 atomic_long_inc(&dev->tx_dropped);
3493 kfree_skb_list(skb);
3496 rcu_read_unlock_bh();
3500 int dev_queue_xmit(struct sk_buff *skb)
3502 return __dev_queue_xmit(skb, NULL);
3504 EXPORT_SYMBOL(dev_queue_xmit);
3506 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3508 return __dev_queue_xmit(skb, accel_priv);
3510 EXPORT_SYMBOL(dev_queue_xmit_accel);
3513 /*=======================================================================
3515 =======================================================================*/
3517 int netdev_max_backlog __read_mostly = 1000;
3518 EXPORT_SYMBOL(netdev_max_backlog);
3520 int netdev_tstamp_prequeue __read_mostly = 1;
3521 int netdev_budget __read_mostly = 300;
3522 int weight_p __read_mostly = 64; /* old backlog weight */
3524 /* Called with irq disabled */
3525 static inline void ____napi_schedule(struct softnet_data *sd,
3526 struct napi_struct *napi)
3528 list_add_tail(&napi->poll_list, &sd->poll_list);
3529 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3534 /* One global table that all flow-based protocols share. */
3535 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3536 EXPORT_SYMBOL(rps_sock_flow_table);
3537 u32 rps_cpu_mask __read_mostly;
3538 EXPORT_SYMBOL(rps_cpu_mask);
3540 struct static_key rps_needed __read_mostly;
3541 EXPORT_SYMBOL(rps_needed);
3543 static struct rps_dev_flow *
3544 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3545 struct rps_dev_flow *rflow, u16 next_cpu)
3547 if (next_cpu < nr_cpu_ids) {
3548 #ifdef CONFIG_RFS_ACCEL
3549 struct netdev_rx_queue *rxqueue;
3550 struct rps_dev_flow_table *flow_table;
3551 struct rps_dev_flow *old_rflow;
3556 /* Should we steer this flow to a different hardware queue? */
3557 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3558 !(dev->features & NETIF_F_NTUPLE))
3560 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3561 if (rxq_index == skb_get_rx_queue(skb))
3564 rxqueue = dev->_rx + rxq_index;
3565 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3568 flow_id = skb_get_hash(skb) & flow_table->mask;
3569 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3570 rxq_index, flow_id);
3574 rflow = &flow_table->flows[flow_id];
3576 if (old_rflow->filter == rflow->filter)
3577 old_rflow->filter = RPS_NO_FILTER;
3581 per_cpu(softnet_data, next_cpu).input_queue_head;
3584 rflow->cpu = next_cpu;
3589 * get_rps_cpu is called from netif_receive_skb and returns the target
3590 * CPU from the RPS map of the receiving queue for a given skb.
3591 * rcu_read_lock must be held on entry.
3593 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3594 struct rps_dev_flow **rflowp)
3596 const struct rps_sock_flow_table *sock_flow_table;
3597 struct netdev_rx_queue *rxqueue = dev->_rx;
3598 struct rps_dev_flow_table *flow_table;
3599 struct rps_map *map;
3604 if (skb_rx_queue_recorded(skb)) {
3605 u16 index = skb_get_rx_queue(skb);
3607 if (unlikely(index >= dev->real_num_rx_queues)) {
3608 WARN_ONCE(dev->real_num_rx_queues > 1,
3609 "%s received packet on queue %u, but number "
3610 "of RX queues is %u\n",
3611 dev->name, index, dev->real_num_rx_queues);
3617 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3619 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3620 map = rcu_dereference(rxqueue->rps_map);
3621 if (!flow_table && !map)
3624 skb_reset_network_header(skb);
3625 hash = skb_get_hash(skb);
3629 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3630 if (flow_table && sock_flow_table) {
3631 struct rps_dev_flow *rflow;
3635 /* First check into global flow table if there is a match */
3636 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3637 if ((ident ^ hash) & ~rps_cpu_mask)
3640 next_cpu = ident & rps_cpu_mask;
3642 /* OK, now we know there is a match,
3643 * we can look at the local (per receive queue) flow table
3645 rflow = &flow_table->flows[hash & flow_table->mask];
3649 * If the desired CPU (where last recvmsg was done) is
3650 * different from current CPU (one in the rx-queue flow
3651 * table entry), switch if one of the following holds:
3652 * - Current CPU is unset (>= nr_cpu_ids).
3653 * - Current CPU is offline.
3654 * - The current CPU's queue tail has advanced beyond the
3655 * last packet that was enqueued using this table entry.
3656 * This guarantees that all previous packets for the flow
3657 * have been dequeued, thus preserving in order delivery.
3659 if (unlikely(tcpu != next_cpu) &&
3660 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3661 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3662 rflow->last_qtail)) >= 0)) {
3664 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3667 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3677 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3678 if (cpu_online(tcpu)) {
3688 #ifdef CONFIG_RFS_ACCEL
3691 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3692 * @dev: Device on which the filter was set
3693 * @rxq_index: RX queue index
3694 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3695 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3697 * Drivers that implement ndo_rx_flow_steer() should periodically call
3698 * this function for each installed filter and remove the filters for
3699 * which it returns %true.
3701 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3702 u32 flow_id, u16 filter_id)
3704 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3705 struct rps_dev_flow_table *flow_table;
3706 struct rps_dev_flow *rflow;
3711 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3712 if (flow_table && flow_id <= flow_table->mask) {
3713 rflow = &flow_table->flows[flow_id];
3714 cpu = ACCESS_ONCE(rflow->cpu);
3715 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3716 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3717 rflow->last_qtail) <
3718 (int)(10 * flow_table->mask)))
3724 EXPORT_SYMBOL(rps_may_expire_flow);
3726 #endif /* CONFIG_RFS_ACCEL */
3728 /* Called from hardirq (IPI) context */
3729 static void rps_trigger_softirq(void *data)
3731 struct softnet_data *sd = data;
3733 ____napi_schedule(sd, &sd->backlog);
3737 #endif /* CONFIG_RPS */
3740 * Check if this softnet_data structure is another cpu one
3741 * If yes, queue it to our IPI list and return 1
3744 static int rps_ipi_queued(struct softnet_data *sd)
3747 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3750 sd->rps_ipi_next = mysd->rps_ipi_list;
3751 mysd->rps_ipi_list = sd;
3753 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3756 #endif /* CONFIG_RPS */
3760 #ifdef CONFIG_NET_FLOW_LIMIT
3761 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3764 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3766 #ifdef CONFIG_NET_FLOW_LIMIT
3767 struct sd_flow_limit *fl;
3768 struct softnet_data *sd;
3769 unsigned int old_flow, new_flow;
3771 if (qlen < (netdev_max_backlog >> 1))
3774 sd = this_cpu_ptr(&softnet_data);
3777 fl = rcu_dereference(sd->flow_limit);
3779 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3780 old_flow = fl->history[fl->history_head];
3781 fl->history[fl->history_head] = new_flow;
3784 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3786 if (likely(fl->buckets[old_flow]))
3787 fl->buckets[old_flow]--;
3789 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3801 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3802 * queue (may be a remote CPU queue).
3804 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3805 unsigned int *qtail)
3807 struct softnet_data *sd;
3808 unsigned long flags;
3811 sd = &per_cpu(softnet_data, cpu);
3813 local_irq_save(flags);
3816 if (!netif_running(skb->dev))
3818 qlen = skb_queue_len(&sd->input_pkt_queue);
3819 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3822 __skb_queue_tail(&sd->input_pkt_queue, skb);
3823 input_queue_tail_incr_save(sd, qtail);
3825 local_irq_restore(flags);
3826 return NET_RX_SUCCESS;
3829 /* Schedule NAPI for backlog device
3830 * We can use non atomic operation since we own the queue lock
3832 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3833 if (!rps_ipi_queued(sd))
3834 ____napi_schedule(sd, &sd->backlog);
3843 local_irq_restore(flags);
3845 atomic_long_inc(&skb->dev->rx_dropped);
3850 static int netif_rx_internal(struct sk_buff *skb)
3854 net_timestamp_check(netdev_tstamp_prequeue, skb);
3856 trace_netif_rx(skb);
3858 if (static_key_false(&rps_needed)) {
3859 struct rps_dev_flow voidflow, *rflow = &voidflow;
3865 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3867 cpu = smp_processor_id();
3869 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3877 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3884 * netif_rx - post buffer to the network code
3885 * @skb: buffer to post
3887 * This function receives a packet from a device driver and queues it for
3888 * the upper (protocol) levels to process. It always succeeds. The buffer
3889 * may be dropped during processing for congestion control or by the
3893 * NET_RX_SUCCESS (no congestion)
3894 * NET_RX_DROP (packet was dropped)
3898 int netif_rx(struct sk_buff *skb)
3900 trace_netif_rx_entry(skb);
3902 return netif_rx_internal(skb);
3904 EXPORT_SYMBOL(netif_rx);
3906 int netif_rx_ni(struct sk_buff *skb)
3910 trace_netif_rx_ni_entry(skb);
3913 err = netif_rx_internal(skb);
3914 if (local_softirq_pending())
3920 EXPORT_SYMBOL(netif_rx_ni);
3922 static __latent_entropy void net_tx_action(struct softirq_action *h)
3924 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3926 if (sd->completion_queue) {
3927 struct sk_buff *clist;
3929 local_irq_disable();
3930 clist = sd->completion_queue;
3931 sd->completion_queue = NULL;
3935 struct sk_buff *skb = clist;
3936 clist = clist->next;
3938 WARN_ON(atomic_read(&skb->users));
3939 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3940 trace_consume_skb(skb);
3942 trace_kfree_skb(skb, net_tx_action);
3944 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
3947 __kfree_skb_defer(skb);
3950 __kfree_skb_flush();
3953 if (sd->output_queue) {
3956 local_irq_disable();
3957 head = sd->output_queue;
3958 sd->output_queue = NULL;
3959 sd->output_queue_tailp = &sd->output_queue;
3963 struct Qdisc *q = head;
3964 spinlock_t *root_lock;
3966 head = head->next_sched;
3968 root_lock = qdisc_lock(q);
3969 spin_lock(root_lock);
3970 /* We need to make sure head->next_sched is read
3971 * before clearing __QDISC_STATE_SCHED
3973 smp_mb__before_atomic();
3974 clear_bit(__QDISC_STATE_SCHED, &q->state);
3976 spin_unlock(root_lock);
3981 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
3982 /* This hook is defined here for ATM LANE */
3983 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3984 unsigned char *addr) __read_mostly;
3985 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3988 static inline struct sk_buff *
3989 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
3990 struct net_device *orig_dev)
3992 #ifdef CONFIG_NET_CLS_ACT
3993 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3994 struct tcf_result cl_res;
3996 /* If there's at least one ingress present somewhere (so
3997 * we get here via enabled static key), remaining devices
3998 * that are not configured with an ingress qdisc will bail
4004 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4008 qdisc_skb_cb(skb)->pkt_len = skb->len;
4009 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
4010 qdisc_bstats_cpu_update(cl->q, skb);
4012 switch (tc_classify(skb, cl, &cl_res, false)) {
4014 case TC_ACT_RECLASSIFY:
4015 skb->tc_index = TC_H_MIN(cl_res.classid);
4018 qdisc_qstats_cpu_drop(cl->q);
4025 case TC_ACT_REDIRECT:
4026 /* skb_mac_header check was done by cls/act_bpf, so
4027 * we can safely push the L2 header back before
4028 * redirecting to another netdev
4030 __skb_push(skb, skb->mac_len);
4031 skb_do_redirect(skb);
4036 #endif /* CONFIG_NET_CLS_ACT */
4041 * netdev_is_rx_handler_busy - check if receive handler is registered
4042 * @dev: device to check
4044 * Check if a receive handler is already registered for a given device.
4045 * Return true if there one.
4047 * The caller must hold the rtnl_mutex.
4049 bool netdev_is_rx_handler_busy(struct net_device *dev)
4052 return dev && rtnl_dereference(dev->rx_handler);
4054 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4057 * netdev_rx_handler_register - register receive handler
4058 * @dev: device to register a handler for
4059 * @rx_handler: receive handler to register
4060 * @rx_handler_data: data pointer that is used by rx handler
4062 * Register a receive handler for a device. This handler will then be
4063 * called from __netif_receive_skb. A negative errno code is returned
4066 * The caller must hold the rtnl_mutex.
4068 * For a general description of rx_handler, see enum rx_handler_result.
4070 int netdev_rx_handler_register(struct net_device *dev,
4071 rx_handler_func_t *rx_handler,
4072 void *rx_handler_data)
4076 if (dev->rx_handler)
4079 /* Note: rx_handler_data must be set before rx_handler */
4080 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4081 rcu_assign_pointer(dev->rx_handler, rx_handler);
4085 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4088 * netdev_rx_handler_unregister - unregister receive handler
4089 * @dev: device to unregister a handler from
4091 * Unregister a receive handler from a device.
4093 * The caller must hold the rtnl_mutex.
4095 void netdev_rx_handler_unregister(struct net_device *dev)
4099 RCU_INIT_POINTER(dev->rx_handler, NULL);
4100 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4101 * section has a guarantee to see a non NULL rx_handler_data
4105 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4107 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4110 * Limit the use of PFMEMALLOC reserves to those protocols that implement
4111 * the special handling of PFMEMALLOC skbs.
4113 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4115 switch (skb->protocol) {
4116 case htons(ETH_P_ARP):
4117 case htons(ETH_P_IP):
4118 case htons(ETH_P_IPV6):
4119 case htons(ETH_P_8021Q):
4120 case htons(ETH_P_8021AD):
4127 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4128 int *ret, struct net_device *orig_dev)
4130 #ifdef CONFIG_NETFILTER_INGRESS
4131 if (nf_hook_ingress_active(skb)) {
4135 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4140 ingress_retval = nf_hook_ingress(skb);
4142 return ingress_retval;
4144 #endif /* CONFIG_NETFILTER_INGRESS */
4148 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
4150 struct packet_type *ptype, *pt_prev;
4151 rx_handler_func_t *rx_handler;
4152 struct net_device *orig_dev;
4153 bool deliver_exact = false;
4154 int ret = NET_RX_DROP;
4157 net_timestamp_check(!netdev_tstamp_prequeue, skb);
4159 trace_netif_receive_skb(skb);
4161 orig_dev = skb->dev;
4163 skb_reset_network_header(skb);
4164 if (!skb_transport_header_was_set(skb))
4165 skb_reset_transport_header(skb);
4166 skb_reset_mac_len(skb);
4171 skb->skb_iif = skb->dev->ifindex;
4173 __this_cpu_inc(softnet_data.processed);
4175 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4176 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4177 skb = skb_vlan_untag(skb);
4182 #ifdef CONFIG_NET_CLS_ACT
4183 if (skb->tc_verd & TC_NCLS) {
4184 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
4192 list_for_each_entry_rcu(ptype, &ptype_all, list) {
4194 ret = deliver_skb(skb, pt_prev, orig_dev);
4198 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4200 ret = deliver_skb(skb, pt_prev, orig_dev);
4205 #ifdef CONFIG_NET_INGRESS
4206 if (static_key_false(&ingress_needed)) {
4207 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4211 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4215 #ifdef CONFIG_NET_CLS_ACT
4219 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4222 if (skb_vlan_tag_present(skb)) {
4224 ret = deliver_skb(skb, pt_prev, orig_dev);
4227 if (vlan_do_receive(&skb))
4229 else if (unlikely(!skb))
4233 rx_handler = rcu_dereference(skb->dev->rx_handler);
4236 ret = deliver_skb(skb, pt_prev, orig_dev);
4239 switch (rx_handler(&skb)) {
4240 case RX_HANDLER_CONSUMED:
4241 ret = NET_RX_SUCCESS;
4243 case RX_HANDLER_ANOTHER:
4245 case RX_HANDLER_EXACT:
4246 deliver_exact = true;
4247 case RX_HANDLER_PASS:
4254 if (unlikely(skb_vlan_tag_present(skb))) {
4255 if (skb_vlan_tag_get_id(skb))
4256 skb->pkt_type = PACKET_OTHERHOST;
4257 /* Note: we might in the future use prio bits
4258 * and set skb->priority like in vlan_do_receive()
4259 * For the time being, just ignore Priority Code Point
4264 type = skb->protocol;
4266 /* deliver only exact match when indicated */
4267 if (likely(!deliver_exact)) {
4268 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4269 &ptype_base[ntohs(type) &
4273 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4274 &orig_dev->ptype_specific);
4276 if (unlikely(skb->dev != orig_dev)) {
4277 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4278 &skb->dev->ptype_specific);
4282 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
4285 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4289 atomic_long_inc(&skb->dev->rx_dropped);
4291 atomic_long_inc(&skb->dev->rx_nohandler);
4293 /* Jamal, now you will not able to escape explaining
4294 * me how you were going to use this. :-)
4303 static int __netif_receive_skb(struct sk_buff *skb)
4307 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4308 unsigned long pflags = current->flags;
4311 * PFMEMALLOC skbs are special, they should
4312 * - be delivered to SOCK_MEMALLOC sockets only
4313 * - stay away from userspace
4314 * - have bounded memory usage
4316 * Use PF_MEMALLOC as this saves us from propagating the allocation
4317 * context down to all allocation sites.
4319 current->flags |= PF_MEMALLOC;
4320 ret = __netif_receive_skb_core(skb, true);
4321 tsk_restore_flags(current, pflags, PF_MEMALLOC);
4323 ret = __netif_receive_skb_core(skb, false);
4328 static int netif_receive_skb_internal(struct sk_buff *skb)
4332 net_timestamp_check(netdev_tstamp_prequeue, skb);
4334 if (skb_defer_rx_timestamp(skb))
4335 return NET_RX_SUCCESS;
4340 if (static_key_false(&rps_needed)) {
4341 struct rps_dev_flow voidflow, *rflow = &voidflow;
4342 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4345 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4351 ret = __netif_receive_skb(skb);
4357 * netif_receive_skb - process receive buffer from network
4358 * @skb: buffer to process
4360 * netif_receive_skb() is the main receive data processing function.
4361 * It always succeeds. The buffer may be dropped during processing
4362 * for congestion control or by the protocol layers.
4364 * This function may only be called from softirq context and interrupts
4365 * should be enabled.
4367 * Return values (usually ignored):
4368 * NET_RX_SUCCESS: no congestion
4369 * NET_RX_DROP: packet was dropped
4371 int netif_receive_skb(struct sk_buff *skb)
4373 trace_netif_receive_skb_entry(skb);
4375 return netif_receive_skb_internal(skb);
4377 EXPORT_SYMBOL(netif_receive_skb);
4379 DEFINE_PER_CPU(struct work_struct, flush_works);
4381 /* Network device is going away, flush any packets still pending */
4382 static void flush_backlog(struct work_struct *work)
4384 struct sk_buff *skb, *tmp;
4385 struct softnet_data *sd;
4388 sd = this_cpu_ptr(&softnet_data);
4390 local_irq_disable();
4392 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4393 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4394 __skb_unlink(skb, &sd->input_pkt_queue);
4395 dev_kfree_skb_irq(skb);
4396 input_queue_head_incr(sd);
4402 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4403 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4404 __skb_unlink(skb, &sd->process_queue);
4406 input_queue_head_incr(sd);
4412 static void flush_all_backlogs(void)
4418 for_each_online_cpu(cpu)
4419 queue_work_on(cpu, system_highpri_wq,
4420 per_cpu_ptr(&flush_works, cpu));
4422 for_each_online_cpu(cpu)
4423 flush_work(per_cpu_ptr(&flush_works, cpu));
4428 static int napi_gro_complete(struct sk_buff *skb)
4430 struct packet_offload *ptype;
4431 __be16 type = skb->protocol;
4432 struct list_head *head = &offload_base;
4435 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4437 if (NAPI_GRO_CB(skb)->count == 1) {
4438 skb_shinfo(skb)->gso_size = 0;
4443 list_for_each_entry_rcu(ptype, head, list) {
4444 if (ptype->type != type || !ptype->callbacks.gro_complete)
4447 err = ptype->callbacks.gro_complete(skb, 0);
4453 WARN_ON(&ptype->list == head);
4455 return NET_RX_SUCCESS;
4459 return netif_receive_skb_internal(skb);
4462 /* napi->gro_list contains packets ordered by age.
4463 * youngest packets at the head of it.
4464 * Complete skbs in reverse order to reduce latencies.
4466 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4468 struct sk_buff *skb, *prev = NULL;
4470 /* scan list and build reverse chain */
4471 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4476 for (skb = prev; skb; skb = prev) {
4479 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4483 napi_gro_complete(skb);
4487 napi->gro_list = NULL;
4489 EXPORT_SYMBOL(napi_gro_flush);
4491 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4494 unsigned int maclen = skb->dev->hard_header_len;
4495 u32 hash = skb_get_hash_raw(skb);
4497 for (p = napi->gro_list; p; p = p->next) {
4498 unsigned long diffs;
4500 NAPI_GRO_CB(p)->flush = 0;
4502 if (hash != skb_get_hash_raw(p)) {
4503 NAPI_GRO_CB(p)->same_flow = 0;
4507 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4508 diffs |= p->vlan_tci ^ skb->vlan_tci;
4509 diffs |= skb_metadata_dst_cmp(p, skb);
4510 if (maclen == ETH_HLEN)
4511 diffs |= compare_ether_header(skb_mac_header(p),
4512 skb_mac_header(skb));
4514 diffs = memcmp(skb_mac_header(p),
4515 skb_mac_header(skb),
4517 NAPI_GRO_CB(p)->same_flow = !diffs;
4521 static void skb_gro_reset_offset(struct sk_buff *skb)
4523 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4524 const skb_frag_t *frag0 = &pinfo->frags[0];
4526 NAPI_GRO_CB(skb)->data_offset = 0;
4527 NAPI_GRO_CB(skb)->frag0 = NULL;
4528 NAPI_GRO_CB(skb)->frag0_len = 0;
4530 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4532 !PageHighMem(skb_frag_page(frag0))) {
4533 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4534 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
4535 skb_frag_size(frag0),
4536 skb->end - skb->tail);
4540 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4542 struct skb_shared_info *pinfo = skb_shinfo(skb);
4544 BUG_ON(skb->end - skb->tail < grow);
4546 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4548 skb->data_len -= grow;
4551 pinfo->frags[0].page_offset += grow;
4552 skb_frag_size_sub(&pinfo->frags[0], grow);
4554 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4555 skb_frag_unref(skb, 0);
4556 memmove(pinfo->frags, pinfo->frags + 1,
4557 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4561 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4563 struct sk_buff **pp = NULL;
4564 struct packet_offload *ptype;
4565 __be16 type = skb->protocol;
4566 struct list_head *head = &offload_base;
4568 enum gro_result ret;
4571 if (!(skb->dev->features & NETIF_F_GRO))
4574 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4577 gro_list_prepare(napi, skb);
4580 list_for_each_entry_rcu(ptype, head, list) {
4581 if (ptype->type != type || !ptype->callbacks.gro_receive)
4584 skb_set_network_header(skb, skb_gro_offset(skb));
4585 skb_reset_mac_len(skb);
4586 NAPI_GRO_CB(skb)->same_flow = 0;
4587 NAPI_GRO_CB(skb)->flush = 0;
4588 NAPI_GRO_CB(skb)->free = 0;
4589 NAPI_GRO_CB(skb)->encap_mark = 0;
4590 NAPI_GRO_CB(skb)->recursion_counter = 0;
4591 NAPI_GRO_CB(skb)->is_fou = 0;
4592 NAPI_GRO_CB(skb)->is_atomic = 1;
4593 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4595 /* Setup for GRO checksum validation */
4596 switch (skb->ip_summed) {
4597 case CHECKSUM_COMPLETE:
4598 NAPI_GRO_CB(skb)->csum = skb->csum;
4599 NAPI_GRO_CB(skb)->csum_valid = 1;
4600 NAPI_GRO_CB(skb)->csum_cnt = 0;
4602 case CHECKSUM_UNNECESSARY:
4603 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4604 NAPI_GRO_CB(skb)->csum_valid = 0;
4607 NAPI_GRO_CB(skb)->csum_cnt = 0;
4608 NAPI_GRO_CB(skb)->csum_valid = 0;
4611 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4616 if (&ptype->list == head)
4619 same_flow = NAPI_GRO_CB(skb)->same_flow;
4620 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4623 struct sk_buff *nskb = *pp;
4627 napi_gro_complete(nskb);
4634 if (NAPI_GRO_CB(skb)->flush)
4637 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4638 struct sk_buff *nskb = napi->gro_list;
4640 /* locate the end of the list to select the 'oldest' flow */
4641 while (nskb->next) {
4647 napi_gro_complete(nskb);
4651 NAPI_GRO_CB(skb)->count = 1;
4652 NAPI_GRO_CB(skb)->age = jiffies;
4653 NAPI_GRO_CB(skb)->last = skb;
4654 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4655 skb->next = napi->gro_list;
4656 napi->gro_list = skb;
4660 grow = skb_gro_offset(skb) - skb_headlen(skb);
4662 gro_pull_from_frag0(skb, grow);
4671 struct packet_offload *gro_find_receive_by_type(__be16 type)
4673 struct list_head *offload_head = &offload_base;
4674 struct packet_offload *ptype;
4676 list_for_each_entry_rcu(ptype, offload_head, list) {
4677 if (ptype->type != type || !ptype->callbacks.gro_receive)
4683 EXPORT_SYMBOL(gro_find_receive_by_type);
4685 struct packet_offload *gro_find_complete_by_type(__be16 type)
4687 struct list_head *offload_head = &offload_base;
4688 struct packet_offload *ptype;
4690 list_for_each_entry_rcu(ptype, offload_head, list) {
4691 if (ptype->type != type || !ptype->callbacks.gro_complete)
4697 EXPORT_SYMBOL(gro_find_complete_by_type);
4699 static void napi_skb_free_stolen_head(struct sk_buff *skb)
4702 kmem_cache_free(skbuff_head_cache, skb);
4705 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4709 if (netif_receive_skb_internal(skb))
4717 case GRO_MERGED_FREE:
4718 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4719 napi_skb_free_stolen_head(skb);
4732 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4734 skb_mark_napi_id(skb, napi);
4735 trace_napi_gro_receive_entry(skb);
4737 skb_gro_reset_offset(skb);
4739 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4741 EXPORT_SYMBOL(napi_gro_receive);
4743 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4745 if (unlikely(skb->pfmemalloc)) {
4749 __skb_pull(skb, skb_headlen(skb));
4750 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4751 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4753 skb->dev = napi->dev;
4756 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
4757 skb->pkt_type = PACKET_HOST;
4759 skb->encapsulation = 0;
4760 skb_shinfo(skb)->gso_type = 0;
4761 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4766 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4768 struct sk_buff *skb = napi->skb;
4771 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4774 skb_mark_napi_id(skb, napi);
4779 EXPORT_SYMBOL(napi_get_frags);
4781 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4782 struct sk_buff *skb,
4788 __skb_push(skb, ETH_HLEN);
4789 skb->protocol = eth_type_trans(skb, skb->dev);
4790 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4795 napi_reuse_skb(napi, skb);
4798 case GRO_MERGED_FREE:
4799 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4800 napi_skb_free_stolen_head(skb);
4802 napi_reuse_skb(napi, skb);
4812 /* Upper GRO stack assumes network header starts at gro_offset=0
4813 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4814 * We copy ethernet header into skb->data to have a common layout.
4816 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4818 struct sk_buff *skb = napi->skb;
4819 const struct ethhdr *eth;
4820 unsigned int hlen = sizeof(*eth);
4824 skb_reset_mac_header(skb);
4825 skb_gro_reset_offset(skb);
4827 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4828 eth = skb_gro_header_slow(skb, hlen, 0);
4829 if (unlikely(!eth)) {
4830 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
4831 __func__, napi->dev->name);
4832 napi_reuse_skb(napi, skb);
4836 eth = (const struct ethhdr *)skb->data;
4837 gro_pull_from_frag0(skb, hlen);
4838 NAPI_GRO_CB(skb)->frag0 += hlen;
4839 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4841 __skb_pull(skb, hlen);
4844 * This works because the only protocols we care about don't require
4846 * We'll fix it up properly in napi_frags_finish()
4848 skb->protocol = eth->h_proto;
4853 gro_result_t napi_gro_frags(struct napi_struct *napi)
4855 struct sk_buff *skb = napi_frags_skb(napi);
4860 trace_napi_gro_frags_entry(skb);
4862 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4864 EXPORT_SYMBOL(napi_gro_frags);
4866 /* Compute the checksum from gro_offset and return the folded value
4867 * after adding in any pseudo checksum.
4869 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4874 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4876 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4877 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4879 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4880 !skb->csum_complete_sw)
4881 netdev_rx_csum_fault(skb->dev);
4884 NAPI_GRO_CB(skb)->csum = wsum;
4885 NAPI_GRO_CB(skb)->csum_valid = 1;
4889 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4892 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4893 * Note: called with local irq disabled, but exits with local irq enabled.
4895 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4898 struct softnet_data *remsd = sd->rps_ipi_list;
4901 sd->rps_ipi_list = NULL;
4905 /* Send pending IPI's to kick RPS processing on remote cpus. */
4907 struct softnet_data *next = remsd->rps_ipi_next;
4909 if (cpu_online(remsd->cpu))
4910 smp_call_function_single_async(remsd->cpu,
4919 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4922 return sd->rps_ipi_list != NULL;
4928 static int process_backlog(struct napi_struct *napi, int quota)
4930 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4934 /* Check if we have pending ipi, its better to send them now,
4935 * not waiting net_rx_action() end.
4937 if (sd_has_rps_ipi_waiting(sd)) {
4938 local_irq_disable();
4939 net_rps_action_and_irq_enable(sd);
4942 napi->weight = weight_p;
4944 struct sk_buff *skb;
4946 while ((skb = __skb_dequeue(&sd->process_queue))) {
4948 __netif_receive_skb(skb);
4950 input_queue_head_incr(sd);
4951 if (++work >= quota)
4956 local_irq_disable();
4958 if (skb_queue_empty(&sd->input_pkt_queue)) {
4960 * Inline a custom version of __napi_complete().
4961 * only current cpu owns and manipulates this napi,
4962 * and NAPI_STATE_SCHED is the only possible flag set
4964 * We can use a plain write instead of clear_bit(),
4965 * and we dont need an smp_mb() memory barrier.
4970 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4971 &sd->process_queue);
4981 * __napi_schedule - schedule for receive
4982 * @n: entry to schedule
4984 * The entry's receive function will be scheduled to run.
4985 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4987 void __napi_schedule(struct napi_struct *n)
4989 unsigned long flags;
4991 local_irq_save(flags);
4992 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4993 local_irq_restore(flags);
4995 EXPORT_SYMBOL(__napi_schedule);
4998 * __napi_schedule_irqoff - schedule for receive
4999 * @n: entry to schedule
5001 * Variant of __napi_schedule() assuming hard irqs are masked.
5003 * On PREEMPT_RT enabled kernels this maps to __napi_schedule()
5004 * because the interrupt disabled assumption might not be true
5005 * due to force-threaded interrupts and spinlock substitution.
5007 void __napi_schedule_irqoff(struct napi_struct *n)
5009 if (!IS_ENABLED(CONFIG_PREEMPT_RT))
5010 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5014 EXPORT_SYMBOL(__napi_schedule_irqoff);
5016 void __napi_complete(struct napi_struct *n)
5018 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
5020 list_del_init(&n->poll_list);
5021 smp_mb__before_atomic();
5022 clear_bit(NAPI_STATE_SCHED, &n->state);
5024 EXPORT_SYMBOL(__napi_complete);
5026 void napi_complete_done(struct napi_struct *n, int work_done)
5028 unsigned long flags;
5031 * don't let napi dequeue from the cpu poll list
5032 * just in case its running on a different cpu
5034 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
5038 unsigned long timeout = 0;
5041 timeout = n->dev->gro_flush_timeout;
5044 hrtimer_start(&n->timer, ns_to_ktime(timeout),
5045 HRTIMER_MODE_REL_PINNED);
5047 napi_gro_flush(n, false);
5049 if (likely(list_empty(&n->poll_list))) {
5050 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
5052 /* If n->poll_list is not empty, we need to mask irqs */
5053 local_irq_save(flags);
5055 local_irq_restore(flags);
5058 EXPORT_SYMBOL(napi_complete_done);
5060 /* must be called under rcu_read_lock(), as we dont take a reference */
5061 static struct napi_struct *napi_by_id(unsigned int napi_id)
5063 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
5064 struct napi_struct *napi;
5066 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
5067 if (napi->napi_id == napi_id)
5073 #if defined(CONFIG_NET_RX_BUSY_POLL)
5074 #define BUSY_POLL_BUDGET 8
5075 bool sk_busy_loop(struct sock *sk, int nonblock)
5077 unsigned long end_time = !nonblock ? sk_busy_loop_end_time(sk) : 0;
5078 int (*busy_poll)(struct napi_struct *dev);
5079 struct napi_struct *napi;
5084 napi = napi_by_id(sk->sk_napi_id);
5088 /* Note: ndo_busy_poll method is optional in linux-4.5 */
5089 if (napi->dev->netdev_ops)
5090 busy_poll = napi->dev->netdev_ops->ndo_busy_poll;
5098 rc = busy_poll(napi);
5099 } else if (napi_schedule_prep(napi)) {
5100 void *have = netpoll_poll_lock(napi);
5102 if (test_bit(NAPI_STATE_SCHED, &napi->state)) {
5103 rc = napi->poll(napi, BUSY_POLL_BUDGET);
5104 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
5105 if (rc == BUSY_POLL_BUDGET) {
5106 napi_complete_done(napi, rc);
5107 napi_schedule(napi);
5110 netpoll_poll_unlock(have);
5113 __NET_ADD_STATS(sock_net(sk),
5114 LINUX_MIB_BUSYPOLLRXPACKETS, rc);
5117 if (rc == LL_FLUSH_FAILED)
5118 break; /* permanent failure */
5121 } while (!nonblock && skb_queue_empty(&sk->sk_receive_queue) &&
5122 !need_resched() && !busy_loop_timeout(end_time));
5124 rc = !skb_queue_empty(&sk->sk_receive_queue);
5129 EXPORT_SYMBOL(sk_busy_loop);
5131 #endif /* CONFIG_NET_RX_BUSY_POLL */
5133 void napi_hash_add(struct napi_struct *napi)
5135 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
5136 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
5139 spin_lock(&napi_hash_lock);
5141 /* 0..NR_CPUS+1 range is reserved for sender_cpu use */
5143 if (unlikely(++napi_gen_id < NR_CPUS + 1))
5144 napi_gen_id = NR_CPUS + 1;
5145 } while (napi_by_id(napi_gen_id));
5146 napi->napi_id = napi_gen_id;
5148 hlist_add_head_rcu(&napi->napi_hash_node,
5149 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
5151 spin_unlock(&napi_hash_lock);
5153 EXPORT_SYMBOL_GPL(napi_hash_add);
5155 /* Warning : caller is responsible to make sure rcu grace period
5156 * is respected before freeing memory containing @napi
5158 bool napi_hash_del(struct napi_struct *napi)
5160 bool rcu_sync_needed = false;
5162 spin_lock(&napi_hash_lock);
5164 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
5165 rcu_sync_needed = true;
5166 hlist_del_rcu(&napi->napi_hash_node);
5168 spin_unlock(&napi_hash_lock);
5169 return rcu_sync_needed;
5171 EXPORT_SYMBOL_GPL(napi_hash_del);
5173 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
5175 struct napi_struct *napi;
5177 napi = container_of(timer, struct napi_struct, timer);
5179 napi_schedule(napi);
5181 return HRTIMER_NORESTART;
5184 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
5185 int (*poll)(struct napi_struct *, int), int weight)
5187 INIT_LIST_HEAD(&napi->poll_list);
5188 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
5189 napi->timer.function = napi_watchdog;
5190 napi->gro_count = 0;
5191 napi->gro_list = NULL;
5194 if (weight > NAPI_POLL_WEIGHT)
5195 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5197 napi->weight = weight;
5199 #ifdef CONFIG_NETPOLL
5200 spin_lock_init(&napi->poll_lock);
5201 napi->poll_owner = -1;
5203 set_bit(NAPI_STATE_SCHED, &napi->state);
5204 set_bit(NAPI_STATE_NPSVC, &napi->state);
5205 list_add_rcu(&napi->dev_list, &dev->napi_list);
5206 napi_hash_add(napi);
5208 EXPORT_SYMBOL(netif_napi_add);
5210 void napi_disable(struct napi_struct *n)
5213 set_bit(NAPI_STATE_DISABLE, &n->state);
5215 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
5217 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
5220 hrtimer_cancel(&n->timer);
5222 clear_bit(NAPI_STATE_DISABLE, &n->state);
5224 EXPORT_SYMBOL(napi_disable);
5226 /* Must be called in process context */
5227 void netif_napi_del(struct napi_struct *napi)
5230 if (napi_hash_del(napi))
5232 list_del_init(&napi->dev_list);
5233 napi_free_frags(napi);
5235 kfree_skb_list(napi->gro_list);
5236 napi->gro_list = NULL;
5237 napi->gro_count = 0;
5239 EXPORT_SYMBOL(netif_napi_del);
5241 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
5246 list_del_init(&n->poll_list);
5248 have = netpoll_poll_lock(n);
5252 /* This NAPI_STATE_SCHED test is for avoiding a race
5253 * with netpoll's poll_napi(). Only the entity which
5254 * obtains the lock and sees NAPI_STATE_SCHED set will
5255 * actually make the ->poll() call. Therefore we avoid
5256 * accidentally calling ->poll() when NAPI is not scheduled.
5259 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
5260 work = n->poll(n, weight);
5261 trace_napi_poll(n, work, weight);
5264 WARN_ON_ONCE(work > weight);
5266 if (likely(work < weight))
5269 /* Drivers must not modify the NAPI state if they
5270 * consume the entire weight. In such cases this code
5271 * still "owns" the NAPI instance and therefore can
5272 * move the instance around on the list at-will.
5274 if (unlikely(napi_disable_pending(n))) {
5280 /* flush too old packets
5281 * If HZ < 1000, flush all packets.
5283 napi_gro_flush(n, HZ >= 1000);
5286 /* Some drivers may have called napi_schedule
5287 * prior to exhausting their budget.
5289 if (unlikely(!list_empty(&n->poll_list))) {
5290 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5291 n->dev ? n->dev->name : "backlog");
5295 list_add_tail(&n->poll_list, repoll);
5298 netpoll_poll_unlock(have);
5303 static __latent_entropy void net_rx_action(struct softirq_action *h)
5305 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5306 unsigned long time_limit = jiffies + 2;
5307 int budget = netdev_budget;
5311 local_irq_disable();
5312 list_splice_init(&sd->poll_list, &list);
5316 struct napi_struct *n;
5318 if (list_empty(&list)) {
5319 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
5324 n = list_first_entry(&list, struct napi_struct, poll_list);
5325 budget -= napi_poll(n, &repoll);
5327 /* If softirq window is exhausted then punt.
5328 * Allow this to run for 2 jiffies since which will allow
5329 * an average latency of 1.5/HZ.
5331 if (unlikely(budget <= 0 ||
5332 time_after_eq(jiffies, time_limit))) {
5338 __kfree_skb_flush();
5339 local_irq_disable();
5341 list_splice_tail_init(&sd->poll_list, &list);
5342 list_splice_tail(&repoll, &list);
5343 list_splice(&list, &sd->poll_list);
5344 if (!list_empty(&sd->poll_list))
5345 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5347 net_rps_action_and_irq_enable(sd);
5350 struct netdev_adjacent {
5351 struct net_device *dev;
5353 /* upper master flag, there can only be one master device per list */
5356 /* counter for the number of times this device was added to us */
5359 /* private field for the users */
5362 struct list_head list;
5363 struct rcu_head rcu;
5366 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
5367 struct list_head *adj_list)
5369 struct netdev_adjacent *adj;
5371 list_for_each_entry(adj, adj_list, list) {
5372 if (adj->dev == adj_dev)
5379 * netdev_has_upper_dev - Check if device is linked to an upper device
5381 * @upper_dev: upper device to check
5383 * Find out if a device is linked to specified upper device and return true
5384 * in case it is. Note that this checks only immediate upper device,
5385 * not through a complete stack of devices. The caller must hold the RTNL lock.
5387 bool netdev_has_upper_dev(struct net_device *dev,
5388 struct net_device *upper_dev)
5392 return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper);
5394 EXPORT_SYMBOL(netdev_has_upper_dev);
5397 * netdev_has_any_upper_dev - Check if device is linked to some device
5400 * Find out if a device is linked to an upper device and return true in case
5401 * it is. The caller must hold the RTNL lock.
5403 bool netdev_has_any_upper_dev(struct net_device *dev)
5407 return !list_empty(&dev->all_adj_list.upper);
5409 EXPORT_SYMBOL(netdev_has_any_upper_dev);
5412 * netdev_master_upper_dev_get - Get master upper device
5415 * Find a master upper device and return pointer to it or NULL in case
5416 * it's not there. The caller must hold the RTNL lock.
5418 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5420 struct netdev_adjacent *upper;
5424 if (list_empty(&dev->adj_list.upper))
5427 upper = list_first_entry(&dev->adj_list.upper,
5428 struct netdev_adjacent, list);
5429 if (likely(upper->master))
5433 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5435 void *netdev_adjacent_get_private(struct list_head *adj_list)
5437 struct netdev_adjacent *adj;
5439 adj = list_entry(adj_list, struct netdev_adjacent, list);
5441 return adj->private;
5443 EXPORT_SYMBOL(netdev_adjacent_get_private);
5446 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5448 * @iter: list_head ** of the current position
5450 * Gets the next device from the dev's upper list, starting from iter
5451 * position. The caller must hold RCU read lock.
5453 struct net_device *netdev_upper_get_next_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;
5469 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5472 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
5474 * @iter: list_head ** of the current position
5476 * Gets the next device from the dev's upper list, starting from iter
5477 * position. The caller must hold RCU read lock.
5479 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
5480 struct list_head **iter)
5482 struct netdev_adjacent *upper;
5484 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5486 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5488 if (&upper->list == &dev->all_adj_list.upper)
5491 *iter = &upper->list;
5495 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
5498 * netdev_lower_get_next_private - Get the next ->private from the
5499 * lower neighbour list
5501 * @iter: list_head ** of the current position
5503 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5504 * list, starting from iter position. The caller must hold either hold the
5505 * RTNL lock or its own locking that guarantees that the neighbour lower
5506 * list will remain unchanged.
5508 void *netdev_lower_get_next_private(struct net_device *dev,
5509 struct list_head **iter)
5511 struct netdev_adjacent *lower;
5513 lower = list_entry(*iter, struct netdev_adjacent, list);
5515 if (&lower->list == &dev->adj_list.lower)
5518 *iter = lower->list.next;
5520 return lower->private;
5522 EXPORT_SYMBOL(netdev_lower_get_next_private);
5525 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5526 * lower neighbour list, RCU
5529 * @iter: list_head ** of the current position
5531 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5532 * list, starting from iter position. The caller must hold RCU read lock.
5534 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5535 struct list_head **iter)
5537 struct netdev_adjacent *lower;
5539 WARN_ON_ONCE(!rcu_read_lock_held());
5541 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5543 if (&lower->list == &dev->adj_list.lower)
5546 *iter = &lower->list;
5548 return lower->private;
5550 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5553 * netdev_lower_get_next - Get the next device from the lower neighbour
5556 * @iter: list_head ** of the current position
5558 * Gets the next netdev_adjacent from the dev's lower neighbour
5559 * list, starting from iter position. The caller must hold RTNL lock or
5560 * its own locking that guarantees that the neighbour lower
5561 * list will remain unchanged.
5563 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5565 struct netdev_adjacent *lower;
5567 lower = list_entry(*iter, struct netdev_adjacent, list);
5569 if (&lower->list == &dev->adj_list.lower)
5572 *iter = lower->list.next;
5576 EXPORT_SYMBOL(netdev_lower_get_next);
5579 * netdev_all_lower_get_next - Get the next device from all lower neighbour list
5581 * @iter: list_head ** of the current position
5583 * Gets the next netdev_adjacent from the dev's all lower neighbour
5584 * list, starting from iter position. The caller must hold RTNL lock or
5585 * its own locking that guarantees that the neighbour all lower
5586 * list will remain unchanged.
5588 struct net_device *netdev_all_lower_get_next(struct net_device *dev, struct list_head **iter)
5590 struct netdev_adjacent *lower;
5592 lower = list_entry(*iter, struct netdev_adjacent, list);
5594 if (&lower->list == &dev->all_adj_list.lower)
5597 *iter = lower->list.next;
5601 EXPORT_SYMBOL(netdev_all_lower_get_next);
5604 * netdev_all_lower_get_next_rcu - Get the next device from all
5605 * lower neighbour list, RCU variant
5607 * @iter: list_head ** of the current position
5609 * Gets the next netdev_adjacent from the dev's all lower neighbour
5610 * list, starting from iter position. The caller must hold RCU read lock.
5612 struct net_device *netdev_all_lower_get_next_rcu(struct net_device *dev,
5613 struct list_head **iter)
5615 struct netdev_adjacent *lower;
5617 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5619 if (&lower->list == &dev->all_adj_list.lower)
5622 *iter = &lower->list;
5626 EXPORT_SYMBOL(netdev_all_lower_get_next_rcu);
5629 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5630 * lower neighbour list, RCU
5634 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5635 * list. The caller must hold RCU read lock.
5637 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5639 struct netdev_adjacent *lower;
5641 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5642 struct netdev_adjacent, list);
5644 return lower->private;
5647 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5650 * netdev_master_upper_dev_get_rcu - Get master upper device
5653 * Find a master upper device and return pointer to it or NULL in case
5654 * it's not there. The caller must hold the RCU read lock.
5656 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5658 struct netdev_adjacent *upper;
5660 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5661 struct netdev_adjacent, list);
5662 if (upper && likely(upper->master))
5666 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5668 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5669 struct net_device *adj_dev,
5670 struct list_head *dev_list)
5672 char linkname[IFNAMSIZ+7];
5673 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5674 "upper_%s" : "lower_%s", adj_dev->name);
5675 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5678 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5680 struct list_head *dev_list)
5682 char linkname[IFNAMSIZ+7];
5683 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5684 "upper_%s" : "lower_%s", name);
5685 sysfs_remove_link(&(dev->dev.kobj), linkname);
5688 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5689 struct net_device *adj_dev,
5690 struct list_head *dev_list)
5692 return (dev_list == &dev->adj_list.upper ||
5693 dev_list == &dev->adj_list.lower) &&
5694 net_eq(dev_net(dev), dev_net(adj_dev));
5697 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5698 struct net_device *adj_dev,
5700 struct list_head *dev_list,
5701 void *private, bool master)
5703 struct netdev_adjacent *adj;
5706 adj = __netdev_find_adj(adj_dev, dev_list);
5709 adj->ref_nr += ref_nr;
5713 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5718 adj->master = master;
5719 adj->ref_nr = ref_nr;
5720 adj->private = private;
5723 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5724 adj_dev->name, dev->name, adj_dev->name);
5726 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5727 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5732 /* Ensure that master link is always the first item in list. */
5734 ret = sysfs_create_link(&(dev->dev.kobj),
5735 &(adj_dev->dev.kobj), "master");
5737 goto remove_symlinks;
5739 list_add_rcu(&adj->list, dev_list);
5741 list_add_tail_rcu(&adj->list, dev_list);
5747 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5748 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5756 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5757 struct net_device *adj_dev,
5759 struct list_head *dev_list)
5761 struct netdev_adjacent *adj;
5763 adj = __netdev_find_adj(adj_dev, dev_list);
5766 pr_err("tried to remove device %s from %s\n",
5767 dev->name, adj_dev->name);
5771 if (adj->ref_nr > ref_nr) {
5772 pr_debug("%s to %s ref_nr-%d = %d\n", dev->name, adj_dev->name,
5773 ref_nr, adj->ref_nr-ref_nr);
5774 adj->ref_nr -= ref_nr;
5779 sysfs_remove_link(&(dev->dev.kobj), "master");
5781 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5782 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5784 list_del_rcu(&adj->list);
5785 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5786 adj_dev->name, dev->name, adj_dev->name);
5788 kfree_rcu(adj, rcu);
5791 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5792 struct net_device *upper_dev,
5794 struct list_head *up_list,
5795 struct list_head *down_list,
5796 void *private, bool master)
5800 ret = __netdev_adjacent_dev_insert(dev, upper_dev, ref_nr, up_list,
5805 ret = __netdev_adjacent_dev_insert(upper_dev, dev, ref_nr, down_list,
5808 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
5815 static int __netdev_adjacent_dev_link(struct net_device *dev,
5816 struct net_device *upper_dev,
5819 return __netdev_adjacent_dev_link_lists(dev, upper_dev, ref_nr,
5820 &dev->all_adj_list.upper,
5821 &upper_dev->all_adj_list.lower,
5825 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5826 struct net_device *upper_dev,
5828 struct list_head *up_list,
5829 struct list_head *down_list)
5831 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
5832 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
5835 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5836 struct net_device *upper_dev,
5839 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, ref_nr,
5840 &dev->all_adj_list.upper,
5841 &upper_dev->all_adj_list.lower);
5844 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5845 struct net_device *upper_dev,
5846 void *private, bool master)
5848 int ret = __netdev_adjacent_dev_link(dev, upper_dev, 1);
5853 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, 1,
5854 &dev->adj_list.upper,
5855 &upper_dev->adj_list.lower,
5858 __netdev_adjacent_dev_unlink(dev, upper_dev, 1);
5865 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5866 struct net_device *upper_dev)
5868 __netdev_adjacent_dev_unlink(dev, upper_dev, 1);
5869 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
5870 &dev->adj_list.upper,
5871 &upper_dev->adj_list.lower);
5874 static int __netdev_upper_dev_link(struct net_device *dev,
5875 struct net_device *upper_dev, bool master,
5876 void *upper_priv, void *upper_info)
5878 struct netdev_notifier_changeupper_info changeupper_info;
5879 struct netdev_adjacent *i, *j, *to_i, *to_j;
5884 if (dev == upper_dev)
5887 /* To prevent loops, check if dev is not upper device to upper_dev. */
5888 if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper))
5891 if (__netdev_find_adj(upper_dev, &dev->adj_list.upper))
5894 if (master && netdev_master_upper_dev_get(dev))
5897 changeupper_info.upper_dev = upper_dev;
5898 changeupper_info.master = master;
5899 changeupper_info.linking = true;
5900 changeupper_info.upper_info = upper_info;
5902 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5903 &changeupper_info.info);
5904 ret = notifier_to_errno(ret);
5908 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
5913 /* Now that we linked these devs, make all the upper_dev's
5914 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5915 * versa, and don't forget the devices itself. All of these
5916 * links are non-neighbours.
5918 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5919 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5920 pr_debug("Interlinking %s with %s, non-neighbour\n",
5921 i->dev->name, j->dev->name);
5922 ret = __netdev_adjacent_dev_link(i->dev, j->dev, i->ref_nr);
5928 /* add dev to every upper_dev's upper device */
5929 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5930 pr_debug("linking %s's upper device %s with %s\n",
5931 upper_dev->name, i->dev->name, dev->name);
5932 ret = __netdev_adjacent_dev_link(dev, i->dev, i->ref_nr);
5934 goto rollback_upper_mesh;
5937 /* add upper_dev to every dev's lower device */
5938 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5939 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5940 i->dev->name, upper_dev->name);
5941 ret = __netdev_adjacent_dev_link(i->dev, upper_dev, i->ref_nr);
5943 goto rollback_lower_mesh;
5946 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5947 &changeupper_info.info);
5948 ret = notifier_to_errno(ret);
5950 goto rollback_lower_mesh;
5954 rollback_lower_mesh:
5956 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5959 __netdev_adjacent_dev_unlink(i->dev, upper_dev, i->ref_nr);
5964 rollback_upper_mesh:
5966 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5969 __netdev_adjacent_dev_unlink(dev, i->dev, i->ref_nr);
5977 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5978 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5979 if (i == to_i && j == to_j)
5981 __netdev_adjacent_dev_unlink(i->dev, j->dev, i->ref_nr);
5987 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5993 * netdev_upper_dev_link - Add a link to the upper device
5995 * @upper_dev: new upper device
5997 * Adds a link to device which is upper to this one. The caller must hold
5998 * the RTNL lock. On a failure a negative errno code is returned.
5999 * On success the reference counts are adjusted and the function
6002 int netdev_upper_dev_link(struct net_device *dev,
6003 struct net_device *upper_dev)
6005 return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL);
6007 EXPORT_SYMBOL(netdev_upper_dev_link);
6010 * netdev_master_upper_dev_link - Add a master link to the upper device
6012 * @upper_dev: new upper device
6013 * @upper_priv: upper device private
6014 * @upper_info: upper info to be passed down via notifier
6016 * Adds a link to device which is upper to this one. In this case, only
6017 * one master upper device can be linked, although other non-master devices
6018 * might be linked as well. The caller must hold the RTNL lock.
6019 * On a failure a negative errno code is returned. On success the reference
6020 * counts are adjusted and the function returns zero.
6022 int netdev_master_upper_dev_link(struct net_device *dev,
6023 struct net_device *upper_dev,
6024 void *upper_priv, void *upper_info)
6026 return __netdev_upper_dev_link(dev, upper_dev, true,
6027 upper_priv, upper_info);
6029 EXPORT_SYMBOL(netdev_master_upper_dev_link);
6032 * netdev_upper_dev_unlink - Removes a link to upper device
6034 * @upper_dev: new upper device
6036 * Removes a link to device which is upper to this one. The caller must hold
6039 void netdev_upper_dev_unlink(struct net_device *dev,
6040 struct net_device *upper_dev)
6042 struct netdev_notifier_changeupper_info changeupper_info;
6043 struct netdev_adjacent *i, *j;
6046 changeupper_info.upper_dev = upper_dev;
6047 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
6048 changeupper_info.linking = false;
6050 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
6051 &changeupper_info.info);
6053 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
6055 /* Here is the tricky part. We must remove all dev's lower
6056 * devices from all upper_dev's upper devices and vice
6057 * versa, to maintain the graph relationship.
6059 list_for_each_entry(i, &dev->all_adj_list.lower, list)
6060 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
6061 __netdev_adjacent_dev_unlink(i->dev, j->dev, i->ref_nr);
6063 /* remove also the devices itself from lower/upper device
6066 list_for_each_entry(i, &dev->all_adj_list.lower, list)
6067 __netdev_adjacent_dev_unlink(i->dev, upper_dev, i->ref_nr);
6069 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
6070 __netdev_adjacent_dev_unlink(dev, i->dev, i->ref_nr);
6072 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
6073 &changeupper_info.info);
6075 EXPORT_SYMBOL(netdev_upper_dev_unlink);
6078 * netdev_bonding_info_change - Dispatch event about slave change
6080 * @bonding_info: info to dispatch
6082 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
6083 * The caller must hold the RTNL lock.
6085 void netdev_bonding_info_change(struct net_device *dev,
6086 struct netdev_bonding_info *bonding_info)
6088 struct netdev_notifier_bonding_info info;
6090 memcpy(&info.bonding_info, bonding_info,
6091 sizeof(struct netdev_bonding_info));
6092 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
6095 EXPORT_SYMBOL(netdev_bonding_info_change);
6097 static void netdev_adjacent_add_links(struct net_device *dev)
6099 struct netdev_adjacent *iter;
6101 struct net *net = dev_net(dev);
6103 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6104 if (!net_eq(net, dev_net(iter->dev)))
6106 netdev_adjacent_sysfs_add(iter->dev, dev,
6107 &iter->dev->adj_list.lower);
6108 netdev_adjacent_sysfs_add(dev, iter->dev,
6109 &dev->adj_list.upper);
6112 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6113 if (!net_eq(net, dev_net(iter->dev)))
6115 netdev_adjacent_sysfs_add(iter->dev, dev,
6116 &iter->dev->adj_list.upper);
6117 netdev_adjacent_sysfs_add(dev, iter->dev,
6118 &dev->adj_list.lower);
6122 static void netdev_adjacent_del_links(struct net_device *dev)
6124 struct netdev_adjacent *iter;
6126 struct net *net = dev_net(dev);
6128 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6129 if (!net_eq(net, dev_net(iter->dev)))
6131 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6132 &iter->dev->adj_list.lower);
6133 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6134 &dev->adj_list.upper);
6137 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6138 if (!net_eq(net, dev_net(iter->dev)))
6140 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6141 &iter->dev->adj_list.upper);
6142 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6143 &dev->adj_list.lower);
6147 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
6149 struct netdev_adjacent *iter;
6151 struct net *net = dev_net(dev);
6153 list_for_each_entry(iter, &dev->adj_list.upper, list) {
6154 if (!net_eq(net, dev_net(iter->dev)))
6156 netdev_adjacent_sysfs_del(iter->dev, oldname,
6157 &iter->dev->adj_list.lower);
6158 netdev_adjacent_sysfs_add(iter->dev, dev,
6159 &iter->dev->adj_list.lower);
6162 list_for_each_entry(iter, &dev->adj_list.lower, list) {
6163 if (!net_eq(net, dev_net(iter->dev)))
6165 netdev_adjacent_sysfs_del(iter->dev, oldname,
6166 &iter->dev->adj_list.upper);
6167 netdev_adjacent_sysfs_add(iter->dev, dev,
6168 &iter->dev->adj_list.upper);
6172 void *netdev_lower_dev_get_private(struct net_device *dev,
6173 struct net_device *lower_dev)
6175 struct netdev_adjacent *lower;
6179 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
6183 return lower->private;
6185 EXPORT_SYMBOL(netdev_lower_dev_get_private);
6188 int dev_get_nest_level(struct net_device *dev)
6190 struct net_device *lower = NULL;
6191 struct list_head *iter;
6197 netdev_for_each_lower_dev(dev, lower, iter) {
6198 nest = dev_get_nest_level(lower);
6199 if (max_nest < nest)
6203 return max_nest + 1;
6205 EXPORT_SYMBOL(dev_get_nest_level);
6208 * netdev_lower_change - Dispatch event about lower device state change
6209 * @lower_dev: device
6210 * @lower_state_info: state to dispatch
6212 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
6213 * The caller must hold the RTNL lock.
6215 void netdev_lower_state_changed(struct net_device *lower_dev,
6216 void *lower_state_info)
6218 struct netdev_notifier_changelowerstate_info changelowerstate_info;
6221 changelowerstate_info.lower_state_info = lower_state_info;
6222 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev,
6223 &changelowerstate_info.info);
6225 EXPORT_SYMBOL(netdev_lower_state_changed);
6227 int netdev_default_l2upper_neigh_construct(struct net_device *dev,
6228 struct neighbour *n)
6230 struct net_device *lower_dev, *stop_dev;
6231 struct list_head *iter;
6234 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6235 if (!lower_dev->netdev_ops->ndo_neigh_construct)
6237 err = lower_dev->netdev_ops->ndo_neigh_construct(lower_dev, n);
6239 stop_dev = lower_dev;
6246 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6247 if (lower_dev == stop_dev)
6249 if (!lower_dev->netdev_ops->ndo_neigh_destroy)
6251 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n);
6255 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_construct);
6257 void netdev_default_l2upper_neigh_destroy(struct net_device *dev,
6258 struct neighbour *n)
6260 struct net_device *lower_dev;
6261 struct list_head *iter;
6263 netdev_for_each_lower_dev(dev, lower_dev, iter) {
6264 if (!lower_dev->netdev_ops->ndo_neigh_destroy)
6266 lower_dev->netdev_ops->ndo_neigh_destroy(lower_dev, n);
6269 EXPORT_SYMBOL_GPL(netdev_default_l2upper_neigh_destroy);
6271 static void dev_change_rx_flags(struct net_device *dev, int flags)
6273 const struct net_device_ops *ops = dev->netdev_ops;
6275 if (ops->ndo_change_rx_flags)
6276 ops->ndo_change_rx_flags(dev, flags);
6279 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
6281 unsigned int old_flags = dev->flags;
6287 dev->flags |= IFF_PROMISC;
6288 dev->promiscuity += inc;
6289 if (dev->promiscuity == 0) {
6292 * If inc causes overflow, untouch promisc and return error.
6295 dev->flags &= ~IFF_PROMISC;
6297 dev->promiscuity -= inc;
6298 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6303 if (dev->flags != old_flags) {
6304 pr_info("device %s %s promiscuous mode\n",
6306 dev->flags & IFF_PROMISC ? "entered" : "left");
6307 if (audit_enabled) {
6308 current_uid_gid(&uid, &gid);
6309 audit_log(current->audit_context, GFP_ATOMIC,
6310 AUDIT_ANOM_PROMISCUOUS,
6311 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6312 dev->name, (dev->flags & IFF_PROMISC),
6313 (old_flags & IFF_PROMISC),
6314 from_kuid(&init_user_ns, audit_get_loginuid(current)),
6315 from_kuid(&init_user_ns, uid),
6316 from_kgid(&init_user_ns, gid),
6317 audit_get_sessionid(current));
6320 dev_change_rx_flags(dev, IFF_PROMISC);
6323 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
6328 * dev_set_promiscuity - update promiscuity count on a device
6332 * Add or remove promiscuity from a device. While the count in the device
6333 * remains above zero the interface remains promiscuous. Once it hits zero
6334 * the device reverts back to normal filtering operation. A negative inc
6335 * value is used to drop promiscuity on the device.
6336 * Return 0 if successful or a negative errno code on error.
6338 int dev_set_promiscuity(struct net_device *dev, int inc)
6340 unsigned int old_flags = dev->flags;
6343 err = __dev_set_promiscuity(dev, inc, true);
6346 if (dev->flags != old_flags)
6347 dev_set_rx_mode(dev);
6350 EXPORT_SYMBOL(dev_set_promiscuity);
6352 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
6354 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
6358 dev->flags |= IFF_ALLMULTI;
6359 dev->allmulti += inc;
6360 if (dev->allmulti == 0) {
6363 * If inc causes overflow, untouch allmulti and return error.
6366 dev->flags &= ~IFF_ALLMULTI;
6368 dev->allmulti -= inc;
6369 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6374 if (dev->flags ^ old_flags) {
6375 dev_change_rx_flags(dev, IFF_ALLMULTI);
6376 dev_set_rx_mode(dev);
6378 __dev_notify_flags(dev, old_flags,
6379 dev->gflags ^ old_gflags);
6385 * dev_set_allmulti - update allmulti count on a device
6389 * Add or remove reception of all multicast frames to a device. While the
6390 * count in the device remains above zero the interface remains listening
6391 * to all interfaces. Once it hits zero the device reverts back to normal
6392 * filtering operation. A negative @inc value is used to drop the counter
6393 * when releasing a resource needing all multicasts.
6394 * Return 0 if successful or a negative errno code on error.
6397 int dev_set_allmulti(struct net_device *dev, int inc)
6399 return __dev_set_allmulti(dev, inc, true);
6401 EXPORT_SYMBOL(dev_set_allmulti);
6404 * Upload unicast and multicast address lists to device and
6405 * configure RX filtering. When the device doesn't support unicast
6406 * filtering it is put in promiscuous mode while unicast addresses
6409 void __dev_set_rx_mode(struct net_device *dev)
6411 const struct net_device_ops *ops = dev->netdev_ops;
6413 /* dev_open will call this function so the list will stay sane. */
6414 if (!(dev->flags&IFF_UP))
6417 if (!netif_device_present(dev))
6420 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
6421 /* Unicast addresses changes may only happen under the rtnl,
6422 * therefore calling __dev_set_promiscuity here is safe.
6424 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
6425 __dev_set_promiscuity(dev, 1, false);
6426 dev->uc_promisc = true;
6427 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
6428 __dev_set_promiscuity(dev, -1, false);
6429 dev->uc_promisc = false;
6433 if (ops->ndo_set_rx_mode)
6434 ops->ndo_set_rx_mode(dev);
6437 void dev_set_rx_mode(struct net_device *dev)
6439 netif_addr_lock_bh(dev);
6440 __dev_set_rx_mode(dev);
6441 netif_addr_unlock_bh(dev);
6445 * dev_get_flags - get flags reported to userspace
6448 * Get the combination of flag bits exported through APIs to userspace.
6450 unsigned int dev_get_flags(const struct net_device *dev)
6454 flags = (dev->flags & ~(IFF_PROMISC |
6459 (dev->gflags & (IFF_PROMISC |
6462 if (netif_running(dev)) {
6463 if (netif_oper_up(dev))
6464 flags |= IFF_RUNNING;
6465 if (netif_carrier_ok(dev))
6466 flags |= IFF_LOWER_UP;
6467 if (netif_dormant(dev))
6468 flags |= IFF_DORMANT;
6473 EXPORT_SYMBOL(dev_get_flags);
6475 int __dev_change_flags(struct net_device *dev, unsigned int flags)
6477 unsigned int old_flags = dev->flags;
6483 * Set the flags on our device.
6486 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
6487 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
6489 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6493 * Load in the correct multicast list now the flags have changed.
6496 if ((old_flags ^ flags) & IFF_MULTICAST)
6497 dev_change_rx_flags(dev, IFF_MULTICAST);
6499 dev_set_rx_mode(dev);
6502 * Have we downed the interface. We handle IFF_UP ourselves
6503 * according to user attempts to set it, rather than blindly
6508 if ((old_flags ^ flags) & IFF_UP)
6509 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
6511 if ((flags ^ dev->gflags) & IFF_PROMISC) {
6512 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6513 unsigned int old_flags = dev->flags;
6515 dev->gflags ^= IFF_PROMISC;
6517 if (__dev_set_promiscuity(dev, inc, false) >= 0)
6518 if (dev->flags != old_flags)
6519 dev_set_rx_mode(dev);
6522 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
6523 is important. Some (broken) drivers set IFF_PROMISC, when
6524 IFF_ALLMULTI is requested not asking us and not reporting.
6526 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6527 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6529 dev->gflags ^= IFF_ALLMULTI;
6530 __dev_set_allmulti(dev, inc, false);
6536 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6537 unsigned int gchanges)
6539 unsigned int changes = dev->flags ^ old_flags;
6542 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6544 if (changes & IFF_UP) {
6545 if (dev->flags & IFF_UP)
6546 call_netdevice_notifiers(NETDEV_UP, dev);
6548 call_netdevice_notifiers(NETDEV_DOWN, dev);
6551 if (dev->flags & IFF_UP &&
6552 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6553 struct netdev_notifier_change_info change_info;
6555 change_info.flags_changed = changes;
6556 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6562 * dev_change_flags - change device settings
6564 * @flags: device state flags
6566 * Change settings on device based state flags. The flags are
6567 * in the userspace exported format.
6569 int dev_change_flags(struct net_device *dev, unsigned int flags)
6572 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6574 ret = __dev_change_flags(dev, flags);
6578 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6579 __dev_notify_flags(dev, old_flags, changes);
6582 EXPORT_SYMBOL(dev_change_flags);
6584 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6586 const struct net_device_ops *ops = dev->netdev_ops;
6588 if (ops->ndo_change_mtu)
6589 return ops->ndo_change_mtu(dev, new_mtu);
6591 /* Pairs with all the lockless reads of dev->mtu in the stack */
6592 WRITE_ONCE(dev->mtu, new_mtu);
6597 * dev_set_mtu - Change maximum transfer unit
6599 * @new_mtu: new transfer unit
6601 * Change the maximum transfer size of the network device.
6603 int dev_set_mtu(struct net_device *dev, int new_mtu)
6607 if (new_mtu == dev->mtu)
6610 /* MTU must be positive. */
6614 if (!netif_device_present(dev))
6617 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6618 err = notifier_to_errno(err);
6622 orig_mtu = dev->mtu;
6623 err = __dev_set_mtu(dev, new_mtu);
6626 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
6628 err = notifier_to_errno(err);
6630 /* setting mtu back and notifying everyone again,
6631 * so that they have a chance to revert changes.
6633 __dev_set_mtu(dev, orig_mtu);
6634 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
6640 EXPORT_SYMBOL(dev_set_mtu);
6643 * dev_set_group - Change group this device belongs to
6645 * @new_group: group this device should belong to
6647 void dev_set_group(struct net_device *dev, int new_group)
6649 dev->group = new_group;
6651 EXPORT_SYMBOL(dev_set_group);
6654 * dev_set_mac_address - Change Media Access Control Address
6658 * Change the hardware (MAC) address of the device
6660 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6662 const struct net_device_ops *ops = dev->netdev_ops;
6665 if (!ops->ndo_set_mac_address)
6667 if (sa->sa_family != dev->type)
6669 if (!netif_device_present(dev))
6671 err = ops->ndo_set_mac_address(dev, sa);
6674 dev->addr_assign_type = NET_ADDR_SET;
6675 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6676 add_device_randomness(dev->dev_addr, dev->addr_len);
6679 EXPORT_SYMBOL(dev_set_mac_address);
6682 * dev_change_carrier - Change device carrier
6684 * @new_carrier: new value
6686 * Change device carrier
6688 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6690 const struct net_device_ops *ops = dev->netdev_ops;
6692 if (!ops->ndo_change_carrier)
6694 if (!netif_device_present(dev))
6696 return ops->ndo_change_carrier(dev, new_carrier);
6698 EXPORT_SYMBOL(dev_change_carrier);
6701 * dev_get_phys_port_id - Get device physical port ID
6705 * Get device physical port ID
6707 int dev_get_phys_port_id(struct net_device *dev,
6708 struct netdev_phys_item_id *ppid)
6710 const struct net_device_ops *ops = dev->netdev_ops;
6712 if (!ops->ndo_get_phys_port_id)
6714 return ops->ndo_get_phys_port_id(dev, ppid);
6716 EXPORT_SYMBOL(dev_get_phys_port_id);
6719 * dev_get_phys_port_name - Get device physical port name
6722 * @len: limit of bytes to copy to name
6724 * Get device physical port name
6726 int dev_get_phys_port_name(struct net_device *dev,
6727 char *name, size_t len)
6729 const struct net_device_ops *ops = dev->netdev_ops;
6731 if (!ops->ndo_get_phys_port_name)
6733 return ops->ndo_get_phys_port_name(dev, name, len);
6735 EXPORT_SYMBOL(dev_get_phys_port_name);
6738 * dev_change_proto_down - update protocol port state information
6740 * @proto_down: new value
6742 * This info can be used by switch drivers to set the phys state of the
6745 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6747 const struct net_device_ops *ops = dev->netdev_ops;
6749 if (!ops->ndo_change_proto_down)
6751 if (!netif_device_present(dev))
6753 return ops->ndo_change_proto_down(dev, proto_down);
6755 EXPORT_SYMBOL(dev_change_proto_down);
6758 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
6760 * @fd: new program fd or negative value to clear
6762 * Set or clear a bpf program for a device
6764 int dev_change_xdp_fd(struct net_device *dev, int fd)
6766 const struct net_device_ops *ops = dev->netdev_ops;
6767 struct bpf_prog *prog = NULL;
6768 struct netdev_xdp xdp = {};
6774 prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP);
6776 return PTR_ERR(prog);
6779 xdp.command = XDP_SETUP_PROG;
6781 err = ops->ndo_xdp(dev, &xdp);
6782 if (err < 0 && prog)
6787 EXPORT_SYMBOL(dev_change_xdp_fd);
6790 * dev_new_index - allocate an ifindex
6791 * @net: the applicable net namespace
6793 * Returns a suitable unique value for a new device interface
6794 * number. The caller must hold the rtnl semaphore or the
6795 * dev_base_lock to be sure it remains unique.
6797 static int dev_new_index(struct net *net)
6799 int ifindex = net->ifindex;
6803 if (!__dev_get_by_index(net, ifindex))
6804 return net->ifindex = ifindex;
6808 /* Delayed registration/unregisteration */
6809 static LIST_HEAD(net_todo_list);
6810 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6812 static void net_set_todo(struct net_device *dev)
6814 list_add_tail(&dev->todo_list, &net_todo_list);
6815 dev_net(dev)->dev_unreg_count++;
6818 static void rollback_registered_many(struct list_head *head)
6820 struct net_device *dev, *tmp;
6821 LIST_HEAD(close_head);
6823 BUG_ON(dev_boot_phase);
6826 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6827 /* Some devices call without registering
6828 * for initialization unwind. Remove those
6829 * devices and proceed with the remaining.
6831 if (dev->reg_state == NETREG_UNINITIALIZED) {
6832 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6836 list_del(&dev->unreg_list);
6839 dev->dismantle = true;
6840 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6843 /* If device is running, close it first. */
6844 list_for_each_entry(dev, head, unreg_list)
6845 list_add_tail(&dev->close_list, &close_head);
6846 dev_close_many(&close_head, true);
6848 list_for_each_entry(dev, head, unreg_list) {
6849 /* And unlink it from device chain. */
6850 unlist_netdevice(dev);
6852 dev->reg_state = NETREG_UNREGISTERING;
6854 flush_all_backlogs();
6858 list_for_each_entry(dev, head, unreg_list) {
6859 struct sk_buff *skb = NULL;
6861 /* Shutdown queueing discipline. */
6865 /* Notify protocols, that we are about to destroy
6866 this device. They should clean all the things.
6868 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6870 if (!dev->rtnl_link_ops ||
6871 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6872 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6876 * Flush the unicast and multicast chains
6881 if (dev->netdev_ops->ndo_uninit)
6882 dev->netdev_ops->ndo_uninit(dev);
6885 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6887 /* Notifier chain MUST detach us all upper devices. */
6888 WARN_ON(netdev_has_any_upper_dev(dev));
6890 /* Remove entries from kobject tree */
6891 netdev_unregister_kobject(dev);
6893 /* Remove XPS queueing entries */
6894 netif_reset_xps_queues_gt(dev, 0);
6900 list_for_each_entry(dev, head, unreg_list)
6904 static void rollback_registered(struct net_device *dev)
6908 list_add(&dev->unreg_list, &single);
6909 rollback_registered_many(&single);
6913 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6914 struct net_device *upper, netdev_features_t features)
6916 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6917 netdev_features_t feature;
6920 for_each_netdev_feature(upper_disables, feature_bit) {
6921 feature = __NETIF_F_BIT(feature_bit);
6922 if (!(upper->wanted_features & feature)
6923 && (features & feature)) {
6924 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6925 &feature, upper->name);
6926 features &= ~feature;
6933 static void netdev_sync_lower_features(struct net_device *upper,
6934 struct net_device *lower, netdev_features_t features)
6936 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6937 netdev_features_t feature;
6940 for_each_netdev_feature(upper_disables, feature_bit) {
6941 feature = __NETIF_F_BIT(feature_bit);
6942 if (!(features & feature) && (lower->features & feature)) {
6943 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6944 &feature, lower->name);
6945 lower->wanted_features &= ~feature;
6946 __netdev_update_features(lower);
6948 if (unlikely(lower->features & feature))
6949 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6950 &feature, lower->name);
6952 netdev_features_change(lower);
6957 static netdev_features_t netdev_fix_features(struct net_device *dev,
6958 netdev_features_t features)
6960 /* Fix illegal checksum combinations */
6961 if ((features & NETIF_F_HW_CSUM) &&
6962 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6963 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6964 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6967 /* TSO requires that SG is present as well. */
6968 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6969 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6970 features &= ~NETIF_F_ALL_TSO;
6973 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6974 !(features & NETIF_F_IP_CSUM)) {
6975 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6976 features &= ~NETIF_F_TSO;
6977 features &= ~NETIF_F_TSO_ECN;
6980 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6981 !(features & NETIF_F_IPV6_CSUM)) {
6982 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6983 features &= ~NETIF_F_TSO6;
6986 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
6987 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
6988 features &= ~NETIF_F_TSO_MANGLEID;
6990 /* TSO ECN requires that TSO is present as well. */
6991 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6992 features &= ~NETIF_F_TSO_ECN;
6994 /* Software GSO depends on SG. */
6995 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6996 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6997 features &= ~NETIF_F_GSO;
7000 /* UFO needs SG and checksumming */
7001 if (features & NETIF_F_UFO) {
7002 /* maybe split UFO into V4 and V6? */
7003 if (!(features & NETIF_F_HW_CSUM) &&
7004 ((features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) !=
7005 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))) {
7007 "Dropping NETIF_F_UFO since no checksum offload features.\n");
7008 features &= ~NETIF_F_UFO;
7011 if (!(features & NETIF_F_SG)) {
7013 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
7014 features &= ~NETIF_F_UFO;
7018 /* GSO partial features require GSO partial be set */
7019 if ((features & dev->gso_partial_features) &&
7020 !(features & NETIF_F_GSO_PARTIAL)) {
7022 "Dropping partially supported GSO features since no GSO partial.\n");
7023 features &= ~dev->gso_partial_features;
7026 #ifdef CONFIG_NET_RX_BUSY_POLL
7027 if (dev->netdev_ops->ndo_busy_poll)
7028 features |= NETIF_F_BUSY_POLL;
7031 features &= ~NETIF_F_BUSY_POLL;
7036 int __netdev_update_features(struct net_device *dev)
7038 struct net_device *upper, *lower;
7039 netdev_features_t features;
7040 struct list_head *iter;
7045 features = netdev_get_wanted_features(dev);
7047 if (dev->netdev_ops->ndo_fix_features)
7048 features = dev->netdev_ops->ndo_fix_features(dev, features);
7050 /* driver might be less strict about feature dependencies */
7051 features = netdev_fix_features(dev, features);
7053 /* some features can't be enabled if they're off an an upper device */
7054 netdev_for_each_upper_dev_rcu(dev, upper, iter)
7055 features = netdev_sync_upper_features(dev, upper, features);
7057 if (dev->features == features)
7060 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
7061 &dev->features, &features);
7063 if (dev->netdev_ops->ndo_set_features)
7064 err = dev->netdev_ops->ndo_set_features(dev, features);
7068 if (unlikely(err < 0)) {
7070 "set_features() failed (%d); wanted %pNF, left %pNF\n",
7071 err, &features, &dev->features);
7072 /* return non-0 since some features might have changed and
7073 * it's better to fire a spurious notification than miss it
7079 /* some features must be disabled on lower devices when disabled
7080 * on an upper device (think: bonding master or bridge)
7082 netdev_for_each_lower_dev(dev, lower, iter)
7083 netdev_sync_lower_features(dev, lower, features);
7086 dev->features = features;
7088 return err < 0 ? 0 : 1;
7092 * netdev_update_features - recalculate device features
7093 * @dev: the device to check
7095 * Recalculate dev->features set and send notifications if it
7096 * has changed. Should be called after driver or hardware dependent
7097 * conditions might have changed that influence the features.
7099 void netdev_update_features(struct net_device *dev)
7101 if (__netdev_update_features(dev))
7102 netdev_features_change(dev);
7104 EXPORT_SYMBOL(netdev_update_features);
7107 * netdev_change_features - recalculate device features
7108 * @dev: the device to check
7110 * Recalculate dev->features set and send notifications even
7111 * if they have not changed. Should be called instead of
7112 * netdev_update_features() if also dev->vlan_features might
7113 * have changed to allow the changes to be propagated to stacked
7116 void netdev_change_features(struct net_device *dev)
7118 __netdev_update_features(dev);
7119 netdev_features_change(dev);
7121 EXPORT_SYMBOL(netdev_change_features);
7124 * netif_stacked_transfer_operstate - transfer operstate
7125 * @rootdev: the root or lower level device to transfer state from
7126 * @dev: the device to transfer operstate to
7128 * Transfer operational state from root to device. This is normally
7129 * called when a stacking relationship exists between the root
7130 * device and the device(a leaf device).
7132 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
7133 struct net_device *dev)
7135 if (rootdev->operstate == IF_OPER_DORMANT)
7136 netif_dormant_on(dev);
7138 netif_dormant_off(dev);
7140 if (netif_carrier_ok(rootdev)) {
7141 if (!netif_carrier_ok(dev))
7142 netif_carrier_on(dev);
7144 if (netif_carrier_ok(dev))
7145 netif_carrier_off(dev);
7148 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
7151 static int netif_alloc_rx_queues(struct net_device *dev)
7153 unsigned int i, count = dev->num_rx_queues;
7154 struct netdev_rx_queue *rx;
7155 size_t sz = count * sizeof(*rx);
7159 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7167 for (i = 0; i < count; i++)
7173 static void netdev_init_one_queue(struct net_device *dev,
7174 struct netdev_queue *queue, void *_unused)
7176 /* Initialize queue lock */
7177 spin_lock_init(&queue->_xmit_lock);
7178 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
7179 queue->xmit_lock_owner = -1;
7180 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
7183 dql_init(&queue->dql, HZ);
7187 static void netif_free_tx_queues(struct net_device *dev)
7192 static int netif_alloc_netdev_queues(struct net_device *dev)
7194 unsigned int count = dev->num_tx_queues;
7195 struct netdev_queue *tx;
7196 size_t sz = count * sizeof(*tx);
7198 if (count < 1 || count > 0xffff)
7201 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7209 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
7210 spin_lock_init(&dev->tx_global_lock);
7215 void netif_tx_stop_all_queues(struct net_device *dev)
7219 for (i = 0; i < dev->num_tx_queues; i++) {
7220 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
7221 netif_tx_stop_queue(txq);
7224 EXPORT_SYMBOL(netif_tx_stop_all_queues);
7227 * register_netdevice - register a network device
7228 * @dev: device to register
7230 * Take a completed network device structure and add it to the kernel
7231 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7232 * chain. 0 is returned on success. A negative errno code is returned
7233 * on a failure to set up the device, or if the name is a duplicate.
7235 * Callers must hold the rtnl semaphore. You may want
7236 * register_netdev() instead of this.
7239 * The locking appears insufficient to guarantee two parallel registers
7240 * will not get the same name.
7243 int register_netdevice(struct net_device *dev)
7246 struct net *net = dev_net(dev);
7248 BUG_ON(dev_boot_phase);
7253 /* When net_device's are persistent, this will be fatal. */
7254 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
7257 spin_lock_init(&dev->addr_list_lock);
7258 netdev_set_addr_lockdep_class(dev);
7260 ret = dev_get_valid_name(net, dev, dev->name);
7264 /* Init, if this function is available */
7265 if (dev->netdev_ops->ndo_init) {
7266 ret = dev->netdev_ops->ndo_init(dev);
7274 if (((dev->hw_features | dev->features) &
7275 NETIF_F_HW_VLAN_CTAG_FILTER) &&
7276 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
7277 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
7278 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
7285 dev->ifindex = dev_new_index(net);
7286 else if (__dev_get_by_index(net, dev->ifindex))
7289 /* Transfer changeable features to wanted_features and enable
7290 * software offloads (GSO and GRO).
7292 dev->hw_features |= NETIF_F_SOFT_FEATURES;
7293 dev->features |= NETIF_F_SOFT_FEATURES;
7294 dev->wanted_features = dev->features & dev->hw_features;
7296 if (!(dev->flags & IFF_LOOPBACK))
7297 dev->hw_features |= NETIF_F_NOCACHE_COPY;
7299 /* If IPv4 TCP segmentation offload is supported we should also
7300 * allow the device to enable segmenting the frame with the option
7301 * of ignoring a static IP ID value. This doesn't enable the
7302 * feature itself but allows the user to enable it later.
7304 if (dev->hw_features & NETIF_F_TSO)
7305 dev->hw_features |= NETIF_F_TSO_MANGLEID;
7306 if (dev->vlan_features & NETIF_F_TSO)
7307 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
7308 if (dev->mpls_features & NETIF_F_TSO)
7309 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
7310 if (dev->hw_enc_features & NETIF_F_TSO)
7311 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
7313 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
7315 dev->vlan_features |= NETIF_F_HIGHDMA;
7317 /* Make NETIF_F_SG inheritable to tunnel devices.
7319 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
7321 /* Make NETIF_F_SG inheritable to MPLS.
7323 dev->mpls_features |= NETIF_F_SG;
7325 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
7326 ret = notifier_to_errno(ret);
7330 ret = netdev_register_kobject(dev);
7333 dev->reg_state = NETREG_REGISTERED;
7335 __netdev_update_features(dev);
7338 * Default initial state at registry is that the
7339 * device is present.
7342 set_bit(__LINK_STATE_PRESENT, &dev->state);
7344 linkwatch_init_dev(dev);
7346 dev_init_scheduler(dev);
7348 list_netdevice(dev);
7349 add_device_randomness(dev->dev_addr, dev->addr_len);
7351 /* If the device has permanent device address, driver should
7352 * set dev_addr and also addr_assign_type should be set to
7353 * NET_ADDR_PERM (default value).
7355 if (dev->addr_assign_type == NET_ADDR_PERM)
7356 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
7358 /* Notify protocols, that a new device appeared. */
7359 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
7360 ret = notifier_to_errno(ret);
7362 rollback_registered(dev);
7365 dev->reg_state = NETREG_UNREGISTERED;
7366 /* We should put the kobject that hold in
7367 * netdev_unregister_kobject(), otherwise
7368 * the net device cannot be freed when
7369 * driver calls free_netdev(), because the
7370 * kobject is being hold.
7372 kobject_put(&dev->dev.kobj);
7375 * Prevent userspace races by waiting until the network
7376 * device is fully setup before sending notifications.
7378 if (!dev->rtnl_link_ops ||
7379 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7380 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7386 if (dev->netdev_ops->ndo_uninit)
7387 dev->netdev_ops->ndo_uninit(dev);
7390 EXPORT_SYMBOL(register_netdevice);
7393 * init_dummy_netdev - init a dummy network device for NAPI
7394 * @dev: device to init
7396 * This takes a network device structure and initialize the minimum
7397 * amount of fields so it can be used to schedule NAPI polls without
7398 * registering a full blown interface. This is to be used by drivers
7399 * that need to tie several hardware interfaces to a single NAPI
7400 * poll scheduler due to HW limitations.
7402 int init_dummy_netdev(struct net_device *dev)
7404 /* Clear everything. Note we don't initialize spinlocks
7405 * are they aren't supposed to be taken by any of the
7406 * NAPI code and this dummy netdev is supposed to be
7407 * only ever used for NAPI polls
7409 memset(dev, 0, sizeof(struct net_device));
7411 /* make sure we BUG if trying to hit standard
7412 * register/unregister code path
7414 dev->reg_state = NETREG_DUMMY;
7416 /* NAPI wants this */
7417 INIT_LIST_HEAD(&dev->napi_list);
7419 /* a dummy interface is started by default */
7420 set_bit(__LINK_STATE_PRESENT, &dev->state);
7421 set_bit(__LINK_STATE_START, &dev->state);
7423 /* Note : We dont allocate pcpu_refcnt for dummy devices,
7424 * because users of this 'device' dont need to change
7430 EXPORT_SYMBOL_GPL(init_dummy_netdev);
7434 * register_netdev - register a network device
7435 * @dev: device to register
7437 * Take a completed network device structure and add it to the kernel
7438 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7439 * chain. 0 is returned on success. A negative errno code is returned
7440 * on a failure to set up the device, or if the name is a duplicate.
7442 * This is a wrapper around register_netdevice that takes the rtnl semaphore
7443 * and expands the device name if you passed a format string to
7446 int register_netdev(struct net_device *dev)
7451 err = register_netdevice(dev);
7455 EXPORT_SYMBOL(register_netdev);
7457 int netdev_refcnt_read(const struct net_device *dev)
7461 for_each_possible_cpu(i)
7462 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
7465 EXPORT_SYMBOL(netdev_refcnt_read);
7468 * netdev_wait_allrefs - wait until all references are gone.
7469 * @dev: target net_device
7471 * This is called when unregistering network devices.
7473 * Any protocol or device that holds a reference should register
7474 * for netdevice notification, and cleanup and put back the
7475 * reference if they receive an UNREGISTER event.
7476 * We can get stuck here if buggy protocols don't correctly
7479 static void netdev_wait_allrefs(struct net_device *dev)
7481 unsigned long rebroadcast_time, warning_time;
7484 linkwatch_forget_dev(dev);
7486 rebroadcast_time = warning_time = jiffies;
7487 refcnt = netdev_refcnt_read(dev);
7489 while (refcnt != 0) {
7490 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
7493 /* Rebroadcast unregister notification */
7494 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7500 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7501 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
7503 /* We must not have linkwatch events
7504 * pending on unregister. If this
7505 * happens, we simply run the queue
7506 * unscheduled, resulting in a noop
7509 linkwatch_run_queue();
7514 rebroadcast_time = jiffies;
7519 refcnt = netdev_refcnt_read(dev);
7521 if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) {
7522 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
7524 warning_time = jiffies;
7533 * register_netdevice(x1);
7534 * register_netdevice(x2);
7536 * unregister_netdevice(y1);
7537 * unregister_netdevice(y2);
7543 * We are invoked by rtnl_unlock().
7544 * This allows us to deal with problems:
7545 * 1) We can delete sysfs objects which invoke hotplug
7546 * without deadlocking with linkwatch via keventd.
7547 * 2) Since we run with the RTNL semaphore not held, we can sleep
7548 * safely in order to wait for the netdev refcnt to drop to zero.
7550 * We must not return until all unregister events added during
7551 * the interval the lock was held have been completed.
7553 void netdev_run_todo(void)
7555 struct list_head list;
7557 /* Snapshot list, allow later requests */
7558 list_replace_init(&net_todo_list, &list);
7563 /* Wait for rcu callbacks to finish before next phase */
7564 if (!list_empty(&list))
7567 while (!list_empty(&list)) {
7568 struct net_device *dev
7569 = list_first_entry(&list, struct net_device, todo_list);
7570 list_del(&dev->todo_list);
7573 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7576 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
7577 pr_err("network todo '%s' but state %d\n",
7578 dev->name, dev->reg_state);
7583 dev->reg_state = NETREG_UNREGISTERED;
7585 netdev_wait_allrefs(dev);
7588 BUG_ON(netdev_refcnt_read(dev));
7589 BUG_ON(!list_empty(&dev->ptype_all));
7590 BUG_ON(!list_empty(&dev->ptype_specific));
7591 WARN_ON(rcu_access_pointer(dev->ip_ptr));
7592 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
7593 WARN_ON(dev->dn_ptr);
7595 if (dev->destructor)
7596 dev->destructor(dev);
7598 /* Report a network device has been unregistered */
7600 dev_net(dev)->dev_unreg_count--;
7602 wake_up(&netdev_unregistering_wq);
7604 /* Free network device */
7605 kobject_put(&dev->dev.kobj);
7609 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
7610 * all the same fields in the same order as net_device_stats, with only
7611 * the type differing, but rtnl_link_stats64 may have additional fields
7612 * at the end for newer counters.
7614 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7615 const struct net_device_stats *netdev_stats)
7617 #if BITS_PER_LONG == 64
7618 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
7619 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
7620 /* zero out counters that only exist in rtnl_link_stats64 */
7621 memset((char *)stats64 + sizeof(*netdev_stats), 0,
7622 sizeof(*stats64) - sizeof(*netdev_stats));
7624 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
7625 const unsigned long *src = (const unsigned long *)netdev_stats;
7626 u64 *dst = (u64 *)stats64;
7628 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
7629 for (i = 0; i < n; i++)
7631 /* zero out counters that only exist in rtnl_link_stats64 */
7632 memset((char *)stats64 + n * sizeof(u64), 0,
7633 sizeof(*stats64) - n * sizeof(u64));
7636 EXPORT_SYMBOL(netdev_stats_to_stats64);
7639 * dev_get_stats - get network device statistics
7640 * @dev: device to get statistics from
7641 * @storage: place to store stats
7643 * Get network statistics from device. Return @storage.
7644 * The device driver may provide its own method by setting
7645 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7646 * otherwise the internal statistics structure is used.
7648 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
7649 struct rtnl_link_stats64 *storage)
7651 const struct net_device_ops *ops = dev->netdev_ops;
7653 if (ops->ndo_get_stats64) {
7654 memset(storage, 0, sizeof(*storage));
7655 ops->ndo_get_stats64(dev, storage);
7656 } else if (ops->ndo_get_stats) {
7657 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7659 netdev_stats_to_stats64(storage, &dev->stats);
7661 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
7662 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
7663 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
7666 EXPORT_SYMBOL(dev_get_stats);
7668 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7670 struct netdev_queue *queue = dev_ingress_queue(dev);
7672 #ifdef CONFIG_NET_CLS_ACT
7675 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7678 netdev_init_one_queue(dev, queue, NULL);
7679 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7680 queue->qdisc_sleeping = &noop_qdisc;
7681 rcu_assign_pointer(dev->ingress_queue, queue);
7686 static const struct ethtool_ops default_ethtool_ops;
7688 void netdev_set_default_ethtool_ops(struct net_device *dev,
7689 const struct ethtool_ops *ops)
7691 if (dev->ethtool_ops == &default_ethtool_ops)
7692 dev->ethtool_ops = ops;
7694 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7696 void netdev_freemem(struct net_device *dev)
7698 char *addr = (char *)dev - dev->padded;
7704 * alloc_netdev_mqs - allocate network device
7705 * @sizeof_priv: size of private data to allocate space for
7706 * @name: device name format string
7707 * @name_assign_type: origin of device name
7708 * @setup: callback to initialize device
7709 * @txqs: the number of TX subqueues to allocate
7710 * @rxqs: the number of RX subqueues to allocate
7712 * Allocates a struct net_device with private data area for driver use
7713 * and performs basic initialization. Also allocates subqueue structs
7714 * for each queue on the device.
7716 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7717 unsigned char name_assign_type,
7718 void (*setup)(struct net_device *),
7719 unsigned int txqs, unsigned int rxqs)
7721 struct net_device *dev;
7723 struct net_device *p;
7725 BUG_ON(strlen(name) >= sizeof(dev->name));
7728 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7734 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7739 alloc_size = sizeof(struct net_device);
7741 /* ensure 32-byte alignment of private area */
7742 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7743 alloc_size += sizeof_priv;
7745 /* ensure 32-byte alignment of whole construct */
7746 alloc_size += NETDEV_ALIGN - 1;
7748 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7750 p = vzalloc(alloc_size);
7754 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7755 dev->padded = (char *)dev - (char *)p;
7757 dev->pcpu_refcnt = alloc_percpu(int);
7758 if (!dev->pcpu_refcnt)
7761 if (dev_addr_init(dev))
7767 dev_net_set(dev, &init_net);
7769 dev->gso_max_size = GSO_MAX_SIZE;
7770 dev->gso_max_segs = GSO_MAX_SEGS;
7772 INIT_LIST_HEAD(&dev->napi_list);
7773 INIT_LIST_HEAD(&dev->unreg_list);
7774 INIT_LIST_HEAD(&dev->close_list);
7775 INIT_LIST_HEAD(&dev->link_watch_list);
7776 INIT_LIST_HEAD(&dev->adj_list.upper);
7777 INIT_LIST_HEAD(&dev->adj_list.lower);
7778 INIT_LIST_HEAD(&dev->all_adj_list.upper);
7779 INIT_LIST_HEAD(&dev->all_adj_list.lower);
7780 INIT_LIST_HEAD(&dev->ptype_all);
7781 INIT_LIST_HEAD(&dev->ptype_specific);
7782 #ifdef CONFIG_NET_SCHED
7783 hash_init(dev->qdisc_hash);
7785 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7788 if (!dev->tx_queue_len) {
7789 dev->priv_flags |= IFF_NO_QUEUE;
7790 dev->tx_queue_len = 1;
7793 dev->num_tx_queues = txqs;
7794 dev->real_num_tx_queues = txqs;
7795 if (netif_alloc_netdev_queues(dev))
7799 dev->num_rx_queues = rxqs;
7800 dev->real_num_rx_queues = rxqs;
7801 if (netif_alloc_rx_queues(dev))
7805 strcpy(dev->name, name);
7806 dev->name_assign_type = name_assign_type;
7807 dev->group = INIT_NETDEV_GROUP;
7808 if (!dev->ethtool_ops)
7809 dev->ethtool_ops = &default_ethtool_ops;
7811 nf_hook_ingress_init(dev);
7820 free_percpu(dev->pcpu_refcnt);
7822 netdev_freemem(dev);
7825 EXPORT_SYMBOL(alloc_netdev_mqs);
7828 * free_netdev - free network device
7831 * This function does the last stage of destroying an allocated device
7832 * interface. The reference to the device object is released.
7833 * If this is the last reference then it will be freed.
7834 * Must be called in process context.
7836 void free_netdev(struct net_device *dev)
7838 struct napi_struct *p, *n;
7841 netif_free_tx_queues(dev);
7846 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7848 /* Flush device addresses */
7849 dev_addr_flush(dev);
7851 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7854 free_percpu(dev->pcpu_refcnt);
7855 dev->pcpu_refcnt = NULL;
7857 /* Compatibility with error handling in drivers */
7858 if (dev->reg_state == NETREG_UNINITIALIZED) {
7859 netdev_freemem(dev);
7863 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7864 dev->reg_state = NETREG_RELEASED;
7866 /* will free via device release */
7867 put_device(&dev->dev);
7869 EXPORT_SYMBOL(free_netdev);
7872 * synchronize_net - Synchronize with packet receive processing
7874 * Wait for packets currently being received to be done.
7875 * Does not block later packets from starting.
7877 void synchronize_net(void)
7880 if (rtnl_is_locked())
7881 synchronize_rcu_expedited();
7885 EXPORT_SYMBOL(synchronize_net);
7888 * unregister_netdevice_queue - remove device from the kernel
7892 * This function shuts down a device interface and removes it
7893 * from the kernel tables.
7894 * If head not NULL, device is queued to be unregistered later.
7896 * Callers must hold the rtnl semaphore. You may want
7897 * unregister_netdev() instead of this.
7900 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7905 list_move_tail(&dev->unreg_list, head);
7907 rollback_registered(dev);
7908 /* Finish processing unregister after unlock */
7912 EXPORT_SYMBOL(unregister_netdevice_queue);
7915 * unregister_netdevice_many - unregister many devices
7916 * @head: list of devices
7918 * Note: As most callers use a stack allocated list_head,
7919 * we force a list_del() to make sure stack wont be corrupted later.
7921 void unregister_netdevice_many(struct list_head *head)
7923 struct net_device *dev;
7925 if (!list_empty(head)) {
7926 rollback_registered_many(head);
7927 list_for_each_entry(dev, head, unreg_list)
7932 EXPORT_SYMBOL(unregister_netdevice_many);
7935 * unregister_netdev - remove device from the kernel
7938 * This function shuts down a device interface and removes it
7939 * from the kernel tables.
7941 * This is just a wrapper for unregister_netdevice that takes
7942 * the rtnl semaphore. In general you want to use this and not
7943 * unregister_netdevice.
7945 void unregister_netdev(struct net_device *dev)
7948 unregister_netdevice(dev);
7951 EXPORT_SYMBOL(unregister_netdev);
7954 * dev_change_net_namespace - move device to different nethost namespace
7956 * @net: network namespace
7957 * @pat: If not NULL name pattern to try if the current device name
7958 * is already taken in the destination network namespace.
7960 * This function shuts down a device interface and moves it
7961 * to a new network namespace. On success 0 is returned, on
7962 * a failure a netagive errno code is returned.
7964 * Callers must hold the rtnl semaphore.
7967 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7973 /* Don't allow namespace local devices to be moved. */
7975 if (dev->features & NETIF_F_NETNS_LOCAL)
7978 /* Ensure the device has been registrered */
7979 if (dev->reg_state != NETREG_REGISTERED)
7982 /* Get out if there is nothing todo */
7984 if (net_eq(dev_net(dev), net))
7987 /* Pick the destination device name, and ensure
7988 * we can use it in the destination network namespace.
7991 if (__dev_get_by_name(net, dev->name)) {
7992 /* We get here if we can't use the current device name */
7995 err = dev_get_valid_name(net, dev, pat);
8001 * And now a mini version of register_netdevice unregister_netdevice.
8004 /* If device is running close it first. */
8007 /* And unlink it from device chain */
8008 unlist_netdevice(dev);
8012 /* Shutdown queueing discipline. */
8015 /* Notify protocols, that we are about to destroy
8016 this device. They should clean all the things.
8018 Note that dev->reg_state stays at NETREG_REGISTERED.
8019 This is wanted because this way 8021q and macvlan know
8020 the device is just moving and can keep their slaves up.
8022 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8024 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
8025 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
8028 * Flush the unicast and multicast chains
8033 /* Send a netdev-removed uevent to the old namespace */
8034 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
8035 netdev_adjacent_del_links(dev);
8037 /* Actually switch the network namespace */
8038 dev_net_set(dev, net);
8040 /* If there is an ifindex conflict assign a new one */
8041 if (__dev_get_by_index(net, dev->ifindex))
8042 dev->ifindex = dev_new_index(net);
8044 /* Send a netdev-add uevent to the new namespace */
8045 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
8046 netdev_adjacent_add_links(dev);
8048 /* Fixup kobjects */
8049 err = device_rename(&dev->dev, dev->name);
8052 /* Add the device back in the hashes */
8053 list_netdevice(dev);
8055 /* Notify protocols, that a new device appeared. */
8056 call_netdevice_notifiers(NETDEV_REGISTER, dev);
8059 * Prevent userspace races by waiting until the network
8060 * device is fully setup before sending notifications.
8062 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
8069 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
8071 static int dev_cpu_callback(struct notifier_block *nfb,
8072 unsigned long action,
8075 struct sk_buff **list_skb;
8076 struct sk_buff *skb;
8077 unsigned int cpu, oldcpu = (unsigned long)ocpu;
8078 struct softnet_data *sd, *oldsd;
8080 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
8083 local_irq_disable();
8084 cpu = smp_processor_id();
8085 sd = &per_cpu(softnet_data, cpu);
8086 oldsd = &per_cpu(softnet_data, oldcpu);
8088 /* Find end of our completion_queue. */
8089 list_skb = &sd->completion_queue;
8091 list_skb = &(*list_skb)->next;
8092 /* Append completion queue from offline CPU. */
8093 *list_skb = oldsd->completion_queue;
8094 oldsd->completion_queue = NULL;
8096 /* Append output queue from offline CPU. */
8097 if (oldsd->output_queue) {
8098 *sd->output_queue_tailp = oldsd->output_queue;
8099 sd->output_queue_tailp = oldsd->output_queue_tailp;
8100 oldsd->output_queue = NULL;
8101 oldsd->output_queue_tailp = &oldsd->output_queue;
8103 /* Append NAPI poll list from offline CPU, with one exception :
8104 * process_backlog() must be called by cpu owning percpu backlog.
8105 * We properly handle process_queue & input_pkt_queue later.
8107 while (!list_empty(&oldsd->poll_list)) {
8108 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
8112 list_del_init(&napi->poll_list);
8113 if (napi->poll == process_backlog)
8116 ____napi_schedule(sd, napi);
8119 raise_softirq_irqoff(NET_TX_SOFTIRQ);
8122 /* Process offline CPU's input_pkt_queue */
8123 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
8125 input_queue_head_incr(oldsd);
8127 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
8129 input_queue_head_incr(oldsd);
8137 * netdev_increment_features - increment feature set by one
8138 * @all: current feature set
8139 * @one: new feature set
8140 * @mask: mask feature set
8142 * Computes a new feature set after adding a device with feature set
8143 * @one to the master device with current feature set @all. Will not
8144 * enable anything that is off in @mask. Returns the new feature set.
8146 netdev_features_t netdev_increment_features(netdev_features_t all,
8147 netdev_features_t one, netdev_features_t mask)
8149 if (mask & NETIF_F_HW_CSUM)
8150 mask |= NETIF_F_CSUM_MASK;
8151 mask |= NETIF_F_VLAN_CHALLENGED;
8153 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
8154 all &= one | ~NETIF_F_ALL_FOR_ALL;
8156 /* If one device supports hw checksumming, set for all. */
8157 if (all & NETIF_F_HW_CSUM)
8158 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
8162 EXPORT_SYMBOL(netdev_increment_features);
8164 static struct hlist_head * __net_init netdev_create_hash(void)
8167 struct hlist_head *hash;
8169 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
8171 for (i = 0; i < NETDEV_HASHENTRIES; i++)
8172 INIT_HLIST_HEAD(&hash[i]);
8177 /* Initialize per network namespace state */
8178 static int __net_init netdev_init(struct net *net)
8180 if (net != &init_net)
8181 INIT_LIST_HEAD(&net->dev_base_head);
8183 net->dev_name_head = netdev_create_hash();
8184 if (net->dev_name_head == NULL)
8187 net->dev_index_head = netdev_create_hash();
8188 if (net->dev_index_head == NULL)
8194 kfree(net->dev_name_head);
8200 * netdev_drivername - network driver for the device
8201 * @dev: network device
8203 * Determine network driver for device.
8205 const char *netdev_drivername(const struct net_device *dev)
8207 const struct device_driver *driver;
8208 const struct device *parent;
8209 const char *empty = "";
8211 parent = dev->dev.parent;
8215 driver = parent->driver;
8216 if (driver && driver->name)
8217 return driver->name;
8221 static void __netdev_printk(const char *level, const struct net_device *dev,
8222 struct va_format *vaf)
8224 if (dev && dev->dev.parent) {
8225 dev_printk_emit(level[1] - '0',
8228 dev_driver_string(dev->dev.parent),
8229 dev_name(dev->dev.parent),
8230 netdev_name(dev), netdev_reg_state(dev),
8233 printk("%s%s%s: %pV",
8234 level, netdev_name(dev), netdev_reg_state(dev), vaf);
8236 printk("%s(NULL net_device): %pV", level, vaf);
8240 void netdev_printk(const char *level, const struct net_device *dev,
8241 const char *format, ...)
8243 struct va_format vaf;
8246 va_start(args, format);
8251 __netdev_printk(level, dev, &vaf);
8255 EXPORT_SYMBOL(netdev_printk);
8257 #define define_netdev_printk_level(func, level) \
8258 void func(const struct net_device *dev, const char *fmt, ...) \
8260 struct va_format vaf; \
8263 va_start(args, fmt); \
8268 __netdev_printk(level, dev, &vaf); \
8272 EXPORT_SYMBOL(func);
8274 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
8275 define_netdev_printk_level(netdev_alert, KERN_ALERT);
8276 define_netdev_printk_level(netdev_crit, KERN_CRIT);
8277 define_netdev_printk_level(netdev_err, KERN_ERR);
8278 define_netdev_printk_level(netdev_warn, KERN_WARNING);
8279 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
8280 define_netdev_printk_level(netdev_info, KERN_INFO);
8282 static void __net_exit netdev_exit(struct net *net)
8284 kfree(net->dev_name_head);
8285 kfree(net->dev_index_head);
8288 static struct pernet_operations __net_initdata netdev_net_ops = {
8289 .init = netdev_init,
8290 .exit = netdev_exit,
8293 static void __net_exit default_device_exit(struct net *net)
8295 struct net_device *dev, *aux;
8297 * Push all migratable network devices back to the
8298 * initial network namespace
8301 for_each_netdev_safe(net, dev, aux) {
8303 char fb_name[IFNAMSIZ];
8305 /* Ignore unmoveable devices (i.e. loopback) */
8306 if (dev->features & NETIF_F_NETNS_LOCAL)
8309 /* Leave virtual devices for the generic cleanup */
8310 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
8313 /* Push remaining network devices to init_net */
8314 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
8315 if (__dev_get_by_name(&init_net, fb_name))
8316 snprintf(fb_name, IFNAMSIZ, "dev%%d");
8317 err = dev_change_net_namespace(dev, &init_net, fb_name);
8319 pr_emerg("%s: failed to move %s to init_net: %d\n",
8320 __func__, dev->name, err);
8327 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
8329 /* Return with the rtnl_lock held when there are no network
8330 * devices unregistering in any network namespace in net_list.
8334 DEFINE_WAIT_FUNC(wait, woken_wake_function);
8336 add_wait_queue(&netdev_unregistering_wq, &wait);
8338 unregistering = false;
8340 list_for_each_entry(net, net_list, exit_list) {
8341 if (net->dev_unreg_count > 0) {
8342 unregistering = true;
8350 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
8352 remove_wait_queue(&netdev_unregistering_wq, &wait);
8355 static void __net_exit default_device_exit_batch(struct list_head *net_list)
8357 /* At exit all network devices most be removed from a network
8358 * namespace. Do this in the reverse order of registration.
8359 * Do this across as many network namespaces as possible to
8360 * improve batching efficiency.
8362 struct net_device *dev;
8364 LIST_HEAD(dev_kill_list);
8366 /* To prevent network device cleanup code from dereferencing
8367 * loopback devices or network devices that have been freed
8368 * wait here for all pending unregistrations to complete,
8369 * before unregistring the loopback device and allowing the
8370 * network namespace be freed.
8372 * The netdev todo list containing all network devices
8373 * unregistrations that happen in default_device_exit_batch
8374 * will run in the rtnl_unlock() at the end of
8375 * default_device_exit_batch.
8377 rtnl_lock_unregistering(net_list);
8378 list_for_each_entry(net, net_list, exit_list) {
8379 for_each_netdev_reverse(net, dev) {
8380 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
8381 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
8383 unregister_netdevice_queue(dev, &dev_kill_list);
8386 unregister_netdevice_many(&dev_kill_list);
8390 static struct pernet_operations __net_initdata default_device_ops = {
8391 .exit = default_device_exit,
8392 .exit_batch = default_device_exit_batch,
8396 * Initialize the DEV module. At boot time this walks the device list and
8397 * unhooks any devices that fail to initialise (normally hardware not
8398 * present) and leaves us with a valid list of present and active devices.
8403 * This is called single threaded during boot, so no need
8404 * to take the rtnl semaphore.
8406 static int __init net_dev_init(void)
8408 int i, rc = -ENOMEM;
8410 BUG_ON(!dev_boot_phase);
8412 if (dev_proc_init())
8415 if (netdev_kobject_init())
8418 INIT_LIST_HEAD(&ptype_all);
8419 for (i = 0; i < PTYPE_HASH_SIZE; i++)
8420 INIT_LIST_HEAD(&ptype_base[i]);
8422 INIT_LIST_HEAD(&offload_base);
8424 if (register_pernet_subsys(&netdev_net_ops))
8428 * Initialise the packet receive queues.
8431 for_each_possible_cpu(i) {
8432 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
8433 struct softnet_data *sd = &per_cpu(softnet_data, i);
8435 INIT_WORK(flush, flush_backlog);
8437 skb_queue_head_init(&sd->input_pkt_queue);
8438 skb_queue_head_init(&sd->process_queue);
8439 INIT_LIST_HEAD(&sd->poll_list);
8440 sd->output_queue_tailp = &sd->output_queue;
8442 sd->csd.func = rps_trigger_softirq;
8447 sd->backlog.poll = process_backlog;
8448 sd->backlog.weight = weight_p;
8453 /* The loopback device is special if any other network devices
8454 * is present in a network namespace the loopback device must
8455 * be present. Since we now dynamically allocate and free the
8456 * loopback device ensure this invariant is maintained by
8457 * keeping the loopback device as the first device on the
8458 * list of network devices. Ensuring the loopback devices
8459 * is the first device that appears and the last network device
8462 if (register_pernet_device(&loopback_net_ops))
8465 if (register_pernet_device(&default_device_ops))
8468 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
8469 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
8471 hotcpu_notifier(dev_cpu_callback, 0);
8478 subsys_initcall(net_dev_init);