2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/dst_metadata.h>
103 #include <net/pkt_sched.h>
104 #include <net/checksum.h>
105 #include <net/xfrm.h>
106 #include <linux/highmem.h>
107 #include <linux/init.h>
108 #include <linux/module.h>
109 #include <linux/netpoll.h>
110 #include <linux/rcupdate.h>
111 #include <linux/delay.h>
112 #include <net/iw_handler.h>
113 #include <asm/current.h>
114 #include <linux/audit.h>
115 #include <linux/dmaengine.h>
116 #include <linux/err.h>
117 #include <linux/ctype.h>
118 #include <linux/if_arp.h>
119 #include <linux/if_vlan.h>
120 #include <linux/ip.h>
122 #include <net/mpls.h>
123 #include <linux/ipv6.h>
124 #include <linux/in.h>
125 #include <linux/jhash.h>
126 #include <linux/random.h>
127 #include <trace/events/napi.h>
128 #include <trace/events/net.h>
129 #include <trace/events/skb.h>
130 #include <linux/pci.h>
131 #include <linux/inetdevice.h>
132 #include <linux/cpu_rmap.h>
133 #include <linux/static_key.h>
134 #include <linux/hashtable.h>
135 #include <linux/vmalloc.h>
136 #include <linux/if_macvlan.h>
137 #include <linux/errqueue.h>
138 #include <linux/hrtimer.h>
139 #include <linux/netfilter_ingress.h>
141 #include "net-sysfs.h"
143 /* Instead of increasing this, you should create a hash table. */
144 #define MAX_GRO_SKBS 8
146 /* This should be increased if a protocol with a bigger head is added. */
147 #define GRO_MAX_HEAD (MAX_HEADER + 128)
149 static DEFINE_SPINLOCK(ptype_lock);
150 static DEFINE_SPINLOCK(offload_lock);
151 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
152 struct list_head ptype_all __read_mostly; /* Taps */
153 static struct list_head offload_base __read_mostly;
155 static int netif_rx_internal(struct sk_buff *skb);
156 static int call_netdevice_notifiers_info(unsigned long val,
157 struct net_device *dev,
158 struct netdev_notifier_info *info);
161 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
164 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
166 * Writers must hold the rtnl semaphore while they loop through the
167 * dev_base_head list, and hold dev_base_lock for writing when they do the
168 * actual updates. This allows pure readers to access the list even
169 * while a writer is preparing to update it.
171 * To put it another way, dev_base_lock is held for writing only to
172 * protect against pure readers; the rtnl semaphore provides the
173 * protection against other writers.
175 * See, for example usages, register_netdevice() and
176 * unregister_netdevice(), which must be called with the rtnl
179 DEFINE_RWLOCK(dev_base_lock);
180 EXPORT_SYMBOL(dev_base_lock);
182 /* protects napi_hash addition/deletion and napi_gen_id */
183 static DEFINE_SPINLOCK(napi_hash_lock);
185 static unsigned int napi_gen_id;
186 static DEFINE_HASHTABLE(napi_hash, 8);
188 static seqcount_t devnet_rename_seq;
190 static inline void dev_base_seq_inc(struct net *net)
192 while (++net->dev_base_seq == 0);
195 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
197 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
199 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
202 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
204 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
207 static inline void rps_lock(struct softnet_data *sd)
210 spin_lock(&sd->input_pkt_queue.lock);
214 static inline void rps_unlock(struct softnet_data *sd)
217 spin_unlock(&sd->input_pkt_queue.lock);
221 /* Device list insertion */
222 static void list_netdevice(struct net_device *dev)
224 struct net *net = dev_net(dev);
228 write_lock_bh(&dev_base_lock);
229 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
230 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
231 hlist_add_head_rcu(&dev->index_hlist,
232 dev_index_hash(net, dev->ifindex));
233 write_unlock_bh(&dev_base_lock);
235 dev_base_seq_inc(net);
238 /* Device list removal
239 * caller must respect a RCU grace period before freeing/reusing dev
241 static void unlist_netdevice(struct net_device *dev)
245 /* Unlink dev from the device chain */
246 write_lock_bh(&dev_base_lock);
247 list_del_rcu(&dev->dev_list);
248 hlist_del_rcu(&dev->name_hlist);
249 hlist_del_rcu(&dev->index_hlist);
250 write_unlock_bh(&dev_base_lock);
252 dev_base_seq_inc(dev_net(dev));
259 static RAW_NOTIFIER_HEAD(netdev_chain);
262 * Device drivers call our routines to queue packets here. We empty the
263 * queue in the local softnet handler.
266 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
267 EXPORT_PER_CPU_SYMBOL(softnet_data);
269 #ifdef CONFIG_LOCKDEP
271 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
272 * according to dev->type
274 static const unsigned short netdev_lock_type[] =
275 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
276 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
277 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
278 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
279 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
280 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
281 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
282 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
283 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
284 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
285 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
286 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
287 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
288 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
289 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
291 static const char *const netdev_lock_name[] =
292 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
293 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
294 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
295 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
296 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
297 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
298 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
299 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
300 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
301 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
302 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
303 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
304 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
305 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
306 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
308 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
309 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
311 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
315 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
316 if (netdev_lock_type[i] == dev_type)
318 /* the last key is used by default */
319 return ARRAY_SIZE(netdev_lock_type) - 1;
322 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
323 unsigned short dev_type)
327 i = netdev_lock_pos(dev_type);
328 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
329 netdev_lock_name[i]);
332 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
336 i = netdev_lock_pos(dev->type);
337 lockdep_set_class_and_name(&dev->addr_list_lock,
338 &netdev_addr_lock_key[i],
339 netdev_lock_name[i]);
342 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
343 unsigned short dev_type)
346 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
351 /*******************************************************************************
353 Protocol management and registration routines
355 *******************************************************************************/
358 * Add a protocol ID to the list. Now that the input handler is
359 * smarter we can dispense with all the messy stuff that used to be
362 * BEWARE!!! Protocol handlers, mangling input packets,
363 * MUST BE last in hash buckets and checking protocol handlers
364 * MUST start from promiscuous ptype_all chain in net_bh.
365 * It is true now, do not change it.
366 * Explanation follows: if protocol handler, mangling packet, will
367 * be the first on list, it is not able to sense, that packet
368 * is cloned and should be copied-on-write, so that it will
369 * change it and subsequent readers will get broken packet.
373 static inline struct list_head *ptype_head(const struct packet_type *pt)
375 if (pt->type == htons(ETH_P_ALL))
376 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
378 return pt->dev ? &pt->dev->ptype_specific :
379 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
383 * dev_add_pack - add packet handler
384 * @pt: packet type declaration
386 * Add a protocol handler to the networking stack. The passed &packet_type
387 * is linked into kernel lists and may not be freed until it has been
388 * removed from the kernel lists.
390 * This call does not sleep therefore it can not
391 * guarantee all CPU's that are in middle of receiving packets
392 * will see the new packet type (until the next received packet).
395 void dev_add_pack(struct packet_type *pt)
397 struct list_head *head = ptype_head(pt);
399 spin_lock(&ptype_lock);
400 list_add_rcu(&pt->list, head);
401 spin_unlock(&ptype_lock);
403 EXPORT_SYMBOL(dev_add_pack);
406 * __dev_remove_pack - remove packet handler
407 * @pt: packet type declaration
409 * Remove a protocol handler that was previously added to the kernel
410 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
411 * from the kernel lists and can be freed or reused once this function
414 * The packet type might still be in use by receivers
415 * and must not be freed until after all the CPU's have gone
416 * through a quiescent state.
418 void __dev_remove_pack(struct packet_type *pt)
420 struct list_head *head = ptype_head(pt);
421 struct packet_type *pt1;
423 spin_lock(&ptype_lock);
425 list_for_each_entry(pt1, head, list) {
427 list_del_rcu(&pt->list);
432 pr_warn("dev_remove_pack: %p not found\n", pt);
434 spin_unlock(&ptype_lock);
436 EXPORT_SYMBOL(__dev_remove_pack);
439 * dev_remove_pack - remove packet handler
440 * @pt: packet type declaration
442 * Remove a protocol handler that was previously added to the kernel
443 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
444 * from the kernel lists and can be freed or reused once this function
447 * This call sleeps to guarantee that no CPU is looking at the packet
450 void dev_remove_pack(struct packet_type *pt)
452 __dev_remove_pack(pt);
456 EXPORT_SYMBOL(dev_remove_pack);
460 * dev_add_offload - register offload handlers
461 * @po: protocol offload declaration
463 * Add protocol offload handlers to the networking stack. The passed
464 * &proto_offload is linked into kernel lists and may not be freed until
465 * it has been removed from the kernel lists.
467 * This call does not sleep therefore it can not
468 * guarantee all CPU's that are in middle of receiving packets
469 * will see the new offload handlers (until the next received packet).
471 void dev_add_offload(struct packet_offload *po)
473 struct packet_offload *elem;
475 spin_lock(&offload_lock);
476 list_for_each_entry(elem, &offload_base, list) {
477 if (po->priority < elem->priority)
480 list_add_rcu(&po->list, elem->list.prev);
481 spin_unlock(&offload_lock);
483 EXPORT_SYMBOL(dev_add_offload);
486 * __dev_remove_offload - remove offload handler
487 * @po: packet offload declaration
489 * Remove a protocol offload handler that was previously added to the
490 * kernel offload handlers by dev_add_offload(). The passed &offload_type
491 * is removed from the kernel lists and can be freed or reused once this
494 * The packet type might still be in use by receivers
495 * and must not be freed until after all the CPU's have gone
496 * through a quiescent state.
498 static void __dev_remove_offload(struct packet_offload *po)
500 struct list_head *head = &offload_base;
501 struct packet_offload *po1;
503 spin_lock(&offload_lock);
505 list_for_each_entry(po1, head, list) {
507 list_del_rcu(&po->list);
512 pr_warn("dev_remove_offload: %p not found\n", po);
514 spin_unlock(&offload_lock);
518 * dev_remove_offload - remove packet offload handler
519 * @po: packet offload declaration
521 * Remove a packet offload handler that was previously added to the kernel
522 * offload handlers by dev_add_offload(). The passed &offload_type is
523 * removed from the kernel lists and can be freed or reused once this
526 * This call sleeps to guarantee that no CPU is looking at the packet
529 void dev_remove_offload(struct packet_offload *po)
531 __dev_remove_offload(po);
535 EXPORT_SYMBOL(dev_remove_offload);
537 /******************************************************************************
539 Device Boot-time Settings Routines
541 *******************************************************************************/
543 /* Boot time configuration table */
544 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
547 * netdev_boot_setup_add - add new setup entry
548 * @name: name of the device
549 * @map: configured settings for the device
551 * Adds new setup entry to the dev_boot_setup list. The function
552 * returns 0 on error and 1 on success. This is a generic routine to
555 static int netdev_boot_setup_add(char *name, struct ifmap *map)
557 struct netdev_boot_setup *s;
561 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
562 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
563 memset(s[i].name, 0, sizeof(s[i].name));
564 strlcpy(s[i].name, name, IFNAMSIZ);
565 memcpy(&s[i].map, map, sizeof(s[i].map));
570 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
574 * netdev_boot_setup_check - check boot time settings
575 * @dev: the netdevice
577 * Check boot time settings for the device.
578 * The found settings are set for the device to be used
579 * later in the device probing.
580 * Returns 0 if no settings found, 1 if they are.
582 int netdev_boot_setup_check(struct net_device *dev)
584 struct netdev_boot_setup *s = dev_boot_setup;
587 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
588 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
589 !strcmp(dev->name, s[i].name)) {
590 dev->irq = s[i].map.irq;
591 dev->base_addr = s[i].map.base_addr;
592 dev->mem_start = s[i].map.mem_start;
593 dev->mem_end = s[i].map.mem_end;
599 EXPORT_SYMBOL(netdev_boot_setup_check);
603 * netdev_boot_base - get address from boot time settings
604 * @prefix: prefix for network device
605 * @unit: id for network device
607 * Check boot time settings for the base address of device.
608 * The found settings are set for the device to be used
609 * later in the device probing.
610 * Returns 0 if no settings found.
612 unsigned long netdev_boot_base(const char *prefix, int unit)
614 const struct netdev_boot_setup *s = dev_boot_setup;
618 sprintf(name, "%s%d", prefix, unit);
621 * If device already registered then return base of 1
622 * to indicate not to probe for this interface
624 if (__dev_get_by_name(&init_net, name))
627 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
628 if (!strcmp(name, s[i].name))
629 return s[i].map.base_addr;
634 * Saves at boot time configured settings for any netdevice.
636 int __init netdev_boot_setup(char *str)
641 str = get_options(str, ARRAY_SIZE(ints), ints);
646 memset(&map, 0, sizeof(map));
650 map.base_addr = ints[2];
652 map.mem_start = ints[3];
654 map.mem_end = ints[4];
656 /* Add new entry to the list */
657 return netdev_boot_setup_add(str, &map);
660 __setup("netdev=", netdev_boot_setup);
662 /*******************************************************************************
664 Device Interface Subroutines
666 *******************************************************************************/
669 * dev_get_iflink - get 'iflink' value of a interface
670 * @dev: targeted interface
672 * Indicates the ifindex the interface is linked to.
673 * Physical interfaces have the same 'ifindex' and 'iflink' values.
676 int dev_get_iflink(const struct net_device *dev)
678 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
679 return dev->netdev_ops->ndo_get_iflink(dev);
683 EXPORT_SYMBOL(dev_get_iflink);
686 * dev_fill_metadata_dst - Retrieve tunnel egress information.
687 * @dev: targeted interface
690 * For better visibility of tunnel traffic OVS needs to retrieve
691 * egress tunnel information for a packet. Following API allows
692 * user to get this info.
694 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
696 struct ip_tunnel_info *info;
698 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
701 info = skb_tunnel_info_unclone(skb);
704 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
707 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
709 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
712 * __dev_get_by_name - find a device by its name
713 * @net: the applicable net namespace
714 * @name: name to find
716 * Find an interface by name. Must be called under RTNL semaphore
717 * or @dev_base_lock. If the name is found a pointer to the device
718 * is returned. If the name is not found then %NULL is returned. The
719 * reference counters are not incremented so the caller must be
720 * careful with locks.
723 struct net_device *__dev_get_by_name(struct net *net, const char *name)
725 struct net_device *dev;
726 struct hlist_head *head = dev_name_hash(net, name);
728 hlist_for_each_entry(dev, head, name_hlist)
729 if (!strncmp(dev->name, name, IFNAMSIZ))
734 EXPORT_SYMBOL(__dev_get_by_name);
737 * dev_get_by_name_rcu - find a device by its name
738 * @net: the applicable net namespace
739 * @name: name to find
741 * Find an interface by name.
742 * If the name is found a pointer to the device is returned.
743 * If the name is not found then %NULL is returned.
744 * The reference counters are not incremented so the caller must be
745 * careful with locks. The caller must hold RCU lock.
748 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
750 struct net_device *dev;
751 struct hlist_head *head = dev_name_hash(net, name);
753 hlist_for_each_entry_rcu(dev, head, name_hlist)
754 if (!strncmp(dev->name, name, IFNAMSIZ))
759 EXPORT_SYMBOL(dev_get_by_name_rcu);
762 * dev_get_by_name - find a device by its name
763 * @net: the applicable net namespace
764 * @name: name to find
766 * Find an interface by name. This can be called from any
767 * context and does its own locking. The returned handle has
768 * the usage count incremented and the caller must use dev_put() to
769 * release it when it is no longer needed. %NULL is returned if no
770 * matching device is found.
773 struct net_device *dev_get_by_name(struct net *net, const char *name)
775 struct net_device *dev;
778 dev = dev_get_by_name_rcu(net, name);
784 EXPORT_SYMBOL(dev_get_by_name);
787 * __dev_get_by_index - find a device by its ifindex
788 * @net: the applicable net namespace
789 * @ifindex: index of device
791 * Search for an interface by index. Returns %NULL if the device
792 * is not found or a pointer to the device. The device has not
793 * had its reference counter increased so the caller must be careful
794 * about locking. The caller must hold either the RTNL semaphore
798 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
800 struct net_device *dev;
801 struct hlist_head *head = dev_index_hash(net, ifindex);
803 hlist_for_each_entry(dev, head, index_hlist)
804 if (dev->ifindex == ifindex)
809 EXPORT_SYMBOL(__dev_get_by_index);
812 * dev_get_by_index_rcu - find a device by its ifindex
813 * @net: the applicable net namespace
814 * @ifindex: index of device
816 * Search for an interface by index. Returns %NULL if the device
817 * is not found or a pointer to the device. The device has not
818 * had its reference counter increased so the caller must be careful
819 * about locking. The caller must hold RCU lock.
822 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
824 struct net_device *dev;
825 struct hlist_head *head = dev_index_hash(net, ifindex);
827 hlist_for_each_entry_rcu(dev, head, index_hlist)
828 if (dev->ifindex == ifindex)
833 EXPORT_SYMBOL(dev_get_by_index_rcu);
837 * dev_get_by_index - find a device by its ifindex
838 * @net: the applicable net namespace
839 * @ifindex: index of device
841 * Search for an interface by index. Returns NULL if the device
842 * is not found or a pointer to the device. The device returned has
843 * had a reference added and the pointer is safe until the user calls
844 * dev_put to indicate they have finished with it.
847 struct net_device *dev_get_by_index(struct net *net, int ifindex)
849 struct net_device *dev;
852 dev = dev_get_by_index_rcu(net, ifindex);
858 EXPORT_SYMBOL(dev_get_by_index);
861 * netdev_get_name - get a netdevice name, knowing its ifindex.
862 * @net: network namespace
863 * @name: a pointer to the buffer where the name will be stored.
864 * @ifindex: the ifindex of the interface to get the name from.
866 * The use of raw_seqcount_begin() and cond_resched() before
867 * retrying is required as we want to give the writers a chance
868 * to complete when CONFIG_PREEMPT is not set.
870 int netdev_get_name(struct net *net, char *name, int ifindex)
872 struct net_device *dev;
876 seq = raw_seqcount_begin(&devnet_rename_seq);
878 dev = dev_get_by_index_rcu(net, ifindex);
884 strcpy(name, dev->name);
886 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
895 * dev_getbyhwaddr_rcu - find a device by its hardware address
896 * @net: the applicable net namespace
897 * @type: media type of device
898 * @ha: hardware address
900 * Search for an interface by MAC address. Returns NULL if the device
901 * is not found or a pointer to the device.
902 * The caller must hold RCU or RTNL.
903 * The returned device has not had its ref count increased
904 * and the caller must therefore be careful about locking
908 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
911 struct net_device *dev;
913 for_each_netdev_rcu(net, dev)
914 if (dev->type == type &&
915 !memcmp(dev->dev_addr, ha, dev->addr_len))
920 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
922 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
924 struct net_device *dev;
927 for_each_netdev(net, dev)
928 if (dev->type == type)
933 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
935 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
937 struct net_device *dev, *ret = NULL;
940 for_each_netdev_rcu(net, dev)
941 if (dev->type == type) {
949 EXPORT_SYMBOL(dev_getfirstbyhwtype);
952 * __dev_get_by_flags - find any device with given flags
953 * @net: the applicable net namespace
954 * @if_flags: IFF_* values
955 * @mask: bitmask of bits in if_flags to check
957 * Search for any interface with the given flags. Returns NULL if a device
958 * is not found or a pointer to the device. Must be called inside
959 * rtnl_lock(), and result refcount is unchanged.
962 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
965 struct net_device *dev, *ret;
970 for_each_netdev(net, dev) {
971 if (((dev->flags ^ if_flags) & mask) == 0) {
978 EXPORT_SYMBOL(__dev_get_by_flags);
981 * dev_valid_name - check if name is okay for network device
984 * Network device names need to be valid file names to
985 * to allow sysfs to work. We also disallow any kind of
988 bool dev_valid_name(const char *name)
992 if (strlen(name) >= IFNAMSIZ)
994 if (!strcmp(name, ".") || !strcmp(name, ".."))
998 if (*name == '/' || *name == ':' || isspace(*name))
1004 EXPORT_SYMBOL(dev_valid_name);
1007 * __dev_alloc_name - allocate a name for a device
1008 * @net: network namespace to allocate the device name in
1009 * @name: name format string
1010 * @buf: scratch buffer and result name string
1012 * Passed a format string - eg "lt%d" it will try and find a suitable
1013 * id. It scans list of devices to build up a free map, then chooses
1014 * the first empty slot. The caller must hold the dev_base or rtnl lock
1015 * while allocating the name and adding the device in order to avoid
1017 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1018 * Returns the number of the unit assigned or a negative errno code.
1021 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1025 const int max_netdevices = 8*PAGE_SIZE;
1026 unsigned long *inuse;
1027 struct net_device *d;
1029 p = strnchr(name, IFNAMSIZ-1, '%');
1032 * Verify the string as this thing may have come from
1033 * the user. There must be either one "%d" and no other "%"
1036 if (p[1] != 'd' || strchr(p + 2, '%'))
1039 /* Use one page as a bit array of possible slots */
1040 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1044 for_each_netdev(net, d) {
1045 if (!sscanf(d->name, name, &i))
1047 if (i < 0 || i >= max_netdevices)
1050 /* avoid cases where sscanf is not exact inverse of printf */
1051 snprintf(buf, IFNAMSIZ, name, i);
1052 if (!strncmp(buf, d->name, IFNAMSIZ))
1056 i = find_first_zero_bit(inuse, max_netdevices);
1057 free_page((unsigned long) inuse);
1061 snprintf(buf, IFNAMSIZ, name, i);
1062 if (!__dev_get_by_name(net, buf))
1065 /* It is possible to run out of possible slots
1066 * when the name is long and there isn't enough space left
1067 * for the digits, or if all bits are used.
1073 * dev_alloc_name - allocate a name for a device
1075 * @name: name format string
1077 * Passed a format string - eg "lt%d" it will try and find a suitable
1078 * id. It scans list of devices to build up a free map, then chooses
1079 * the first empty slot. The caller must hold the dev_base or rtnl lock
1080 * while allocating the name and adding the device in order to avoid
1082 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1083 * Returns the number of the unit assigned or a negative errno code.
1086 int dev_alloc_name(struct net_device *dev, const char *name)
1092 BUG_ON(!dev_net(dev));
1094 ret = __dev_alloc_name(net, name, buf);
1096 strlcpy(dev->name, buf, IFNAMSIZ);
1099 EXPORT_SYMBOL(dev_alloc_name);
1101 static int dev_alloc_name_ns(struct net *net,
1102 struct net_device *dev,
1108 ret = __dev_alloc_name(net, name, buf);
1110 strlcpy(dev->name, buf, IFNAMSIZ);
1114 static int dev_get_valid_name(struct net *net,
1115 struct net_device *dev,
1120 if (!dev_valid_name(name))
1123 if (strchr(name, '%'))
1124 return dev_alloc_name_ns(net, dev, name);
1125 else if (__dev_get_by_name(net, name))
1127 else if (dev->name != name)
1128 strlcpy(dev->name, name, IFNAMSIZ);
1134 * dev_change_name - change name of a device
1136 * @newname: name (or format string) must be at least IFNAMSIZ
1138 * Change name of a device, can pass format strings "eth%d".
1141 int dev_change_name(struct net_device *dev, const char *newname)
1143 unsigned char old_assign_type;
1144 char oldname[IFNAMSIZ];
1150 BUG_ON(!dev_net(dev));
1153 if (dev->flags & IFF_UP)
1156 write_seqcount_begin(&devnet_rename_seq);
1158 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1159 write_seqcount_end(&devnet_rename_seq);
1163 memcpy(oldname, dev->name, IFNAMSIZ);
1165 err = dev_get_valid_name(net, dev, newname);
1167 write_seqcount_end(&devnet_rename_seq);
1171 if (oldname[0] && !strchr(oldname, '%'))
1172 netdev_info(dev, "renamed from %s\n", oldname);
1174 old_assign_type = dev->name_assign_type;
1175 dev->name_assign_type = NET_NAME_RENAMED;
1178 ret = device_rename(&dev->dev, dev->name);
1180 memcpy(dev->name, oldname, IFNAMSIZ);
1181 dev->name_assign_type = old_assign_type;
1182 write_seqcount_end(&devnet_rename_seq);
1186 write_seqcount_end(&devnet_rename_seq);
1188 netdev_adjacent_rename_links(dev, oldname);
1190 write_lock_bh(&dev_base_lock);
1191 hlist_del_rcu(&dev->name_hlist);
1192 write_unlock_bh(&dev_base_lock);
1196 write_lock_bh(&dev_base_lock);
1197 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1198 write_unlock_bh(&dev_base_lock);
1200 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1201 ret = notifier_to_errno(ret);
1204 /* err >= 0 after dev_alloc_name() or stores the first errno */
1207 write_seqcount_begin(&devnet_rename_seq);
1208 memcpy(dev->name, oldname, IFNAMSIZ);
1209 memcpy(oldname, newname, IFNAMSIZ);
1210 dev->name_assign_type = old_assign_type;
1211 old_assign_type = NET_NAME_RENAMED;
1214 pr_err("%s: name change rollback failed: %d\n",
1223 * dev_set_alias - change ifalias of a device
1225 * @alias: name up to IFALIASZ
1226 * @len: limit of bytes to copy from info
1228 * Set ifalias for a device,
1230 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1236 if (len >= IFALIASZ)
1240 kfree(dev->ifalias);
1241 dev->ifalias = NULL;
1245 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1248 dev->ifalias = new_ifalias;
1250 strlcpy(dev->ifalias, alias, len+1);
1256 * netdev_features_change - device changes features
1257 * @dev: device to cause notification
1259 * Called to indicate a device has changed features.
1261 void netdev_features_change(struct net_device *dev)
1263 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1265 EXPORT_SYMBOL(netdev_features_change);
1268 * netdev_state_change - device changes state
1269 * @dev: device to cause notification
1271 * Called to indicate a device has changed state. This function calls
1272 * the notifier chains for netdev_chain and sends a NEWLINK message
1273 * to the routing socket.
1275 void netdev_state_change(struct net_device *dev)
1277 if (dev->flags & IFF_UP) {
1278 struct netdev_notifier_change_info change_info;
1280 change_info.flags_changed = 0;
1281 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1283 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1286 EXPORT_SYMBOL(netdev_state_change);
1289 * netdev_notify_peers - notify network peers about existence of @dev
1290 * @dev: network device
1292 * Generate traffic such that interested network peers are aware of
1293 * @dev, such as by generating a gratuitous ARP. This may be used when
1294 * a device wants to inform the rest of the network about some sort of
1295 * reconfiguration such as a failover event or virtual machine
1298 void netdev_notify_peers(struct net_device *dev)
1301 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1304 EXPORT_SYMBOL(netdev_notify_peers);
1306 static int __dev_open(struct net_device *dev)
1308 const struct net_device_ops *ops = dev->netdev_ops;
1313 if (!netif_device_present(dev))
1316 /* Block netpoll from trying to do any rx path servicing.
1317 * If we don't do this there is a chance ndo_poll_controller
1318 * or ndo_poll may be running while we open the device
1320 netpoll_poll_disable(dev);
1322 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1323 ret = notifier_to_errno(ret);
1327 set_bit(__LINK_STATE_START, &dev->state);
1329 if (ops->ndo_validate_addr)
1330 ret = ops->ndo_validate_addr(dev);
1332 if (!ret && ops->ndo_open)
1333 ret = ops->ndo_open(dev);
1335 netpoll_poll_enable(dev);
1338 clear_bit(__LINK_STATE_START, &dev->state);
1340 dev->flags |= IFF_UP;
1341 dev_set_rx_mode(dev);
1343 add_device_randomness(dev->dev_addr, dev->addr_len);
1350 * dev_open - prepare an interface for use.
1351 * @dev: device to open
1353 * Takes a device from down to up state. The device's private open
1354 * function is invoked and then the multicast lists are loaded. Finally
1355 * the device is moved into the up state and a %NETDEV_UP message is
1356 * sent to the netdev notifier chain.
1358 * Calling this function on an active interface is a nop. On a failure
1359 * a negative errno code is returned.
1361 int dev_open(struct net_device *dev)
1365 if (dev->flags & IFF_UP)
1368 ret = __dev_open(dev);
1372 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1373 call_netdevice_notifiers(NETDEV_UP, dev);
1377 EXPORT_SYMBOL(dev_open);
1379 static int __dev_close_many(struct list_head *head)
1381 struct net_device *dev;
1386 list_for_each_entry(dev, head, close_list) {
1387 /* Temporarily disable netpoll until the interface is down */
1388 netpoll_poll_disable(dev);
1390 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1392 clear_bit(__LINK_STATE_START, &dev->state);
1394 /* Synchronize to scheduled poll. We cannot touch poll list, it
1395 * can be even on different cpu. So just clear netif_running().
1397 * dev->stop() will invoke napi_disable() on all of it's
1398 * napi_struct instances on this device.
1400 smp_mb__after_atomic(); /* Commit netif_running(). */
1403 dev_deactivate_many(head);
1405 list_for_each_entry(dev, head, close_list) {
1406 const struct net_device_ops *ops = dev->netdev_ops;
1409 * Call the device specific close. This cannot fail.
1410 * Only if device is UP
1412 * We allow it to be called even after a DETACH hot-plug
1418 dev->flags &= ~IFF_UP;
1419 netpoll_poll_enable(dev);
1425 static int __dev_close(struct net_device *dev)
1430 list_add(&dev->close_list, &single);
1431 retval = __dev_close_many(&single);
1437 int dev_close_many(struct list_head *head, bool unlink)
1439 struct net_device *dev, *tmp;
1441 /* Remove the devices that don't need to be closed */
1442 list_for_each_entry_safe(dev, tmp, head, close_list)
1443 if (!(dev->flags & IFF_UP))
1444 list_del_init(&dev->close_list);
1446 __dev_close_many(head);
1448 list_for_each_entry_safe(dev, tmp, head, close_list) {
1449 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1450 call_netdevice_notifiers(NETDEV_DOWN, dev);
1452 list_del_init(&dev->close_list);
1457 EXPORT_SYMBOL(dev_close_many);
1460 * dev_close - shutdown an interface.
1461 * @dev: device to shutdown
1463 * This function moves an active device into down state. A
1464 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1465 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1468 int dev_close(struct net_device *dev)
1470 if (dev->flags & IFF_UP) {
1473 list_add(&dev->close_list, &single);
1474 dev_close_many(&single, true);
1479 EXPORT_SYMBOL(dev_close);
1483 * dev_disable_lro - disable Large Receive Offload on a device
1486 * Disable Large Receive Offload (LRO) on a net device. Must be
1487 * called under RTNL. This is needed if received packets may be
1488 * forwarded to another interface.
1490 void dev_disable_lro(struct net_device *dev)
1492 struct net_device *lower_dev;
1493 struct list_head *iter;
1495 dev->wanted_features &= ~NETIF_F_LRO;
1496 netdev_update_features(dev);
1498 if (unlikely(dev->features & NETIF_F_LRO))
1499 netdev_WARN(dev, "failed to disable LRO!\n");
1501 netdev_for_each_lower_dev(dev, lower_dev, iter)
1502 dev_disable_lro(lower_dev);
1504 EXPORT_SYMBOL(dev_disable_lro);
1506 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1507 struct net_device *dev)
1509 struct netdev_notifier_info info;
1511 netdev_notifier_info_init(&info, dev);
1512 return nb->notifier_call(nb, val, &info);
1515 static int dev_boot_phase = 1;
1518 * register_netdevice_notifier - register a network notifier block
1521 * Register a notifier to be called when network device events occur.
1522 * The notifier passed is linked into the kernel structures and must
1523 * not be reused until it has been unregistered. A negative errno code
1524 * is returned on a failure.
1526 * When registered all registration and up events are replayed
1527 * to the new notifier to allow device to have a race free
1528 * view of the network device list.
1531 int register_netdevice_notifier(struct notifier_block *nb)
1533 struct net_device *dev;
1534 struct net_device *last;
1539 err = raw_notifier_chain_register(&netdev_chain, nb);
1545 for_each_netdev(net, dev) {
1546 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1547 err = notifier_to_errno(err);
1551 if (!(dev->flags & IFF_UP))
1554 call_netdevice_notifier(nb, NETDEV_UP, dev);
1565 for_each_netdev(net, dev) {
1569 if (dev->flags & IFF_UP) {
1570 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1572 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1574 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1579 raw_notifier_chain_unregister(&netdev_chain, nb);
1582 EXPORT_SYMBOL(register_netdevice_notifier);
1585 * unregister_netdevice_notifier - unregister a network notifier block
1588 * Unregister a notifier previously registered by
1589 * register_netdevice_notifier(). The notifier is unlinked into the
1590 * kernel structures and may then be reused. A negative errno code
1591 * is returned on a failure.
1593 * After unregistering unregister and down device events are synthesized
1594 * for all devices on the device list to the removed notifier to remove
1595 * the need for special case cleanup code.
1598 int unregister_netdevice_notifier(struct notifier_block *nb)
1600 struct net_device *dev;
1605 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1610 for_each_netdev(net, dev) {
1611 if (dev->flags & IFF_UP) {
1612 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1614 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1616 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1623 EXPORT_SYMBOL(unregister_netdevice_notifier);
1626 * call_netdevice_notifiers_info - call all network notifier blocks
1627 * @val: value passed unmodified to notifier function
1628 * @dev: net_device pointer passed unmodified to notifier function
1629 * @info: notifier information data
1631 * Call all network notifier blocks. Parameters and return value
1632 * are as for raw_notifier_call_chain().
1635 static int call_netdevice_notifiers_info(unsigned long val,
1636 struct net_device *dev,
1637 struct netdev_notifier_info *info)
1640 netdev_notifier_info_init(info, dev);
1641 return raw_notifier_call_chain(&netdev_chain, val, info);
1645 * call_netdevice_notifiers - call all network notifier blocks
1646 * @val: value passed unmodified to notifier function
1647 * @dev: net_device pointer passed unmodified to notifier function
1649 * Call all network notifier blocks. Parameters and return value
1650 * are as for raw_notifier_call_chain().
1653 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1655 struct netdev_notifier_info info;
1657 return call_netdevice_notifiers_info(val, dev, &info);
1659 EXPORT_SYMBOL(call_netdevice_notifiers);
1661 #ifdef CONFIG_NET_INGRESS
1662 static struct static_key ingress_needed __read_mostly;
1664 void net_inc_ingress_queue(void)
1666 static_key_slow_inc(&ingress_needed);
1668 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1670 void net_dec_ingress_queue(void)
1672 static_key_slow_dec(&ingress_needed);
1674 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1677 static struct static_key netstamp_needed __read_mostly;
1678 #ifdef HAVE_JUMP_LABEL
1679 /* We are not allowed to call static_key_slow_dec() from irq context
1680 * If net_disable_timestamp() is called from irq context, defer the
1681 * static_key_slow_dec() calls.
1683 static atomic_t netstamp_needed_deferred;
1686 void net_enable_timestamp(void)
1688 #ifdef HAVE_JUMP_LABEL
1689 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1693 static_key_slow_dec(&netstamp_needed);
1697 static_key_slow_inc(&netstamp_needed);
1699 EXPORT_SYMBOL(net_enable_timestamp);
1701 void net_disable_timestamp(void)
1703 #ifdef HAVE_JUMP_LABEL
1704 if (in_interrupt()) {
1705 atomic_inc(&netstamp_needed_deferred);
1709 static_key_slow_dec(&netstamp_needed);
1711 EXPORT_SYMBOL(net_disable_timestamp);
1713 static inline void net_timestamp_set(struct sk_buff *skb)
1715 skb->tstamp.tv64 = 0;
1716 if (static_key_false(&netstamp_needed))
1717 __net_timestamp(skb);
1720 #define net_timestamp_check(COND, SKB) \
1721 if (static_key_false(&netstamp_needed)) { \
1722 if ((COND) && !(SKB)->tstamp.tv64) \
1723 __net_timestamp(SKB); \
1726 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1730 if (!(dev->flags & IFF_UP))
1733 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1734 if (skb->len <= len)
1737 /* if TSO is enabled, we don't care about the length as the packet
1738 * could be forwarded without being segmented before
1740 if (skb_is_gso(skb))
1745 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1747 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1749 if (skb_orphan_frags(skb, GFP_ATOMIC) ||
1750 unlikely(!is_skb_forwardable(dev, skb))) {
1751 atomic_long_inc(&dev->rx_dropped);
1756 skb_scrub_packet(skb, true);
1758 skb->protocol = eth_type_trans(skb, dev);
1759 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1763 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1766 * dev_forward_skb - loopback an skb to another netif
1768 * @dev: destination network device
1769 * @skb: buffer to forward
1772 * NET_RX_SUCCESS (no congestion)
1773 * NET_RX_DROP (packet was dropped, but freed)
1775 * dev_forward_skb can be used for injecting an skb from the
1776 * start_xmit function of one device into the receive queue
1777 * of another device.
1779 * The receiving device may be in another namespace, so
1780 * we have to clear all information in the skb that could
1781 * impact namespace isolation.
1783 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1785 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1787 EXPORT_SYMBOL_GPL(dev_forward_skb);
1789 static inline int deliver_skb(struct sk_buff *skb,
1790 struct packet_type *pt_prev,
1791 struct net_device *orig_dev)
1793 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1795 atomic_inc(&skb->users);
1796 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1799 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1800 struct packet_type **pt,
1801 struct net_device *orig_dev,
1803 struct list_head *ptype_list)
1805 struct packet_type *ptype, *pt_prev = *pt;
1807 list_for_each_entry_rcu(ptype, ptype_list, list) {
1808 if (ptype->type != type)
1811 deliver_skb(skb, pt_prev, orig_dev);
1817 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1819 if (!ptype->af_packet_priv || !skb->sk)
1822 if (ptype->id_match)
1823 return ptype->id_match(ptype, skb->sk);
1824 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1831 * Support routine. Sends outgoing frames to any network
1832 * taps currently in use.
1835 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1837 struct packet_type *ptype;
1838 struct sk_buff *skb2 = NULL;
1839 struct packet_type *pt_prev = NULL;
1840 struct list_head *ptype_list = &ptype_all;
1844 list_for_each_entry_rcu(ptype, ptype_list, list) {
1845 /* Never send packets back to the socket
1846 * they originated from - MvS (miquels@drinkel.ow.org)
1848 if (skb_loop_sk(ptype, skb))
1852 deliver_skb(skb2, pt_prev, skb->dev);
1857 /* need to clone skb, done only once */
1858 skb2 = skb_clone(skb, GFP_ATOMIC);
1862 net_timestamp_set(skb2);
1864 /* skb->nh should be correctly
1865 * set by sender, so that the second statement is
1866 * just protection against buggy protocols.
1868 skb_reset_mac_header(skb2);
1870 if (skb_network_header(skb2) < skb2->data ||
1871 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1872 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1873 ntohs(skb2->protocol),
1875 skb_reset_network_header(skb2);
1878 skb2->transport_header = skb2->network_header;
1879 skb2->pkt_type = PACKET_OUTGOING;
1883 if (ptype_list == &ptype_all) {
1884 ptype_list = &dev->ptype_all;
1889 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1894 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1895 * @dev: Network device
1896 * @txq: number of queues available
1898 * If real_num_tx_queues is changed the tc mappings may no longer be
1899 * valid. To resolve this verify the tc mapping remains valid and if
1900 * not NULL the mapping. With no priorities mapping to this
1901 * offset/count pair it will no longer be used. In the worst case TC0
1902 * is invalid nothing can be done so disable priority mappings. If is
1903 * expected that drivers will fix this mapping if they can before
1904 * calling netif_set_real_num_tx_queues.
1906 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1909 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1911 /* If TC0 is invalidated disable TC mapping */
1912 if (tc->offset + tc->count > txq) {
1913 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1918 /* Invalidated prio to tc mappings set to TC0 */
1919 for (i = 1; i < TC_BITMASK + 1; i++) {
1920 int q = netdev_get_prio_tc_map(dev, i);
1922 tc = &dev->tc_to_txq[q];
1923 if (tc->offset + tc->count > txq) {
1924 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1926 netdev_set_prio_tc_map(dev, i, 0);
1932 static DEFINE_MUTEX(xps_map_mutex);
1933 #define xmap_dereference(P) \
1934 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1936 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1939 struct xps_map *map = NULL;
1943 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1945 for (pos = 0; map && pos < map->len; pos++) {
1946 if (map->queues[pos] == index) {
1948 map->queues[pos] = map->queues[--map->len];
1950 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1951 kfree_rcu(map, rcu);
1961 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1963 struct xps_dev_maps *dev_maps;
1965 bool active = false;
1967 mutex_lock(&xps_map_mutex);
1968 dev_maps = xmap_dereference(dev->xps_maps);
1973 for_each_possible_cpu(cpu) {
1974 for (i = index; i < dev->num_tx_queues; i++) {
1975 if (!remove_xps_queue(dev_maps, cpu, i))
1978 if (i == dev->num_tx_queues)
1983 RCU_INIT_POINTER(dev->xps_maps, NULL);
1984 kfree_rcu(dev_maps, rcu);
1987 for (i = index; i < dev->num_tx_queues; i++)
1988 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1992 mutex_unlock(&xps_map_mutex);
1995 static struct xps_map *expand_xps_map(struct xps_map *map,
1998 struct xps_map *new_map;
1999 int alloc_len = XPS_MIN_MAP_ALLOC;
2002 for (pos = 0; map && pos < map->len; pos++) {
2003 if (map->queues[pos] != index)
2008 /* Need to add queue to this CPU's existing map */
2010 if (pos < map->alloc_len)
2013 alloc_len = map->alloc_len * 2;
2016 /* Need to allocate new map to store queue on this CPU's map */
2017 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2022 for (i = 0; i < pos; i++)
2023 new_map->queues[i] = map->queues[i];
2024 new_map->alloc_len = alloc_len;
2030 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2033 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2034 struct xps_map *map, *new_map;
2035 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2036 int cpu, numa_node_id = -2;
2037 bool active = false;
2039 mutex_lock(&xps_map_mutex);
2041 dev_maps = xmap_dereference(dev->xps_maps);
2043 /* allocate memory for queue storage */
2044 for_each_online_cpu(cpu) {
2045 if (!cpumask_test_cpu(cpu, mask))
2049 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2050 if (!new_dev_maps) {
2051 mutex_unlock(&xps_map_mutex);
2055 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2058 map = expand_xps_map(map, cpu, index);
2062 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2066 goto out_no_new_maps;
2068 for_each_possible_cpu(cpu) {
2069 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2070 /* add queue to CPU maps */
2073 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2074 while ((pos < map->len) && (map->queues[pos] != index))
2077 if (pos == map->len)
2078 map->queues[map->len++] = index;
2080 if (numa_node_id == -2)
2081 numa_node_id = cpu_to_node(cpu);
2082 else if (numa_node_id != cpu_to_node(cpu))
2085 } else if (dev_maps) {
2086 /* fill in the new device map from the old device map */
2087 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2088 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2093 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2095 /* Cleanup old maps */
2097 for_each_possible_cpu(cpu) {
2098 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2099 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2100 if (map && map != new_map)
2101 kfree_rcu(map, rcu);
2104 kfree_rcu(dev_maps, rcu);
2107 dev_maps = new_dev_maps;
2111 /* update Tx queue numa node */
2112 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2113 (numa_node_id >= 0) ? numa_node_id :
2119 /* removes queue from unused CPUs */
2120 for_each_possible_cpu(cpu) {
2121 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2124 if (remove_xps_queue(dev_maps, cpu, index))
2128 /* free map if not active */
2130 RCU_INIT_POINTER(dev->xps_maps, NULL);
2131 kfree_rcu(dev_maps, rcu);
2135 mutex_unlock(&xps_map_mutex);
2139 /* remove any maps that we added */
2140 for_each_possible_cpu(cpu) {
2141 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2142 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2144 if (new_map && new_map != map)
2148 mutex_unlock(&xps_map_mutex);
2150 kfree(new_dev_maps);
2153 EXPORT_SYMBOL(netif_set_xps_queue);
2157 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2158 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2160 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2164 if (txq < 1 || txq > dev->num_tx_queues)
2167 if (dev->reg_state == NETREG_REGISTERED ||
2168 dev->reg_state == NETREG_UNREGISTERING) {
2171 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2177 netif_setup_tc(dev, txq);
2179 if (txq < dev->real_num_tx_queues) {
2180 qdisc_reset_all_tx_gt(dev, txq);
2182 netif_reset_xps_queues_gt(dev, txq);
2187 dev->real_num_tx_queues = txq;
2190 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2194 * netif_set_real_num_rx_queues - set actual number of RX queues used
2195 * @dev: Network device
2196 * @rxq: Actual number of RX queues
2198 * This must be called either with the rtnl_lock held or before
2199 * registration of the net device. Returns 0 on success, or a
2200 * negative error code. If called before registration, it always
2203 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2207 if (rxq < 1 || rxq > dev->num_rx_queues)
2210 if (dev->reg_state == NETREG_REGISTERED) {
2213 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2219 dev->real_num_rx_queues = rxq;
2222 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2226 * netif_get_num_default_rss_queues - default number of RSS queues
2228 * This routine should set an upper limit on the number of RSS queues
2229 * used by default by multiqueue devices.
2231 int netif_get_num_default_rss_queues(void)
2233 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2235 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2237 static inline void __netif_reschedule(struct Qdisc *q)
2239 struct softnet_data *sd;
2240 unsigned long flags;
2242 local_irq_save(flags);
2243 sd = this_cpu_ptr(&softnet_data);
2244 q->next_sched = NULL;
2245 *sd->output_queue_tailp = q;
2246 sd->output_queue_tailp = &q->next_sched;
2247 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2248 local_irq_restore(flags);
2251 void __netif_schedule(struct Qdisc *q)
2253 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2254 __netif_reschedule(q);
2256 EXPORT_SYMBOL(__netif_schedule);
2258 struct dev_kfree_skb_cb {
2259 enum skb_free_reason reason;
2262 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2264 return (struct dev_kfree_skb_cb *)skb->cb;
2267 void netif_schedule_queue(struct netdev_queue *txq)
2270 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2271 struct Qdisc *q = rcu_dereference(txq->qdisc);
2273 __netif_schedule(q);
2277 EXPORT_SYMBOL(netif_schedule_queue);
2280 * netif_wake_subqueue - allow sending packets on subqueue
2281 * @dev: network device
2282 * @queue_index: sub queue index
2284 * Resume individual transmit queue of a device with multiple transmit queues.
2286 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2288 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2290 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2294 q = rcu_dereference(txq->qdisc);
2295 __netif_schedule(q);
2299 EXPORT_SYMBOL(netif_wake_subqueue);
2301 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2303 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2307 q = rcu_dereference(dev_queue->qdisc);
2308 __netif_schedule(q);
2312 EXPORT_SYMBOL(netif_tx_wake_queue);
2314 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2316 unsigned long flags;
2318 if (likely(atomic_read(&skb->users) == 1)) {
2320 atomic_set(&skb->users, 0);
2321 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2324 get_kfree_skb_cb(skb)->reason = reason;
2325 local_irq_save(flags);
2326 skb->next = __this_cpu_read(softnet_data.completion_queue);
2327 __this_cpu_write(softnet_data.completion_queue, skb);
2328 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2329 local_irq_restore(flags);
2331 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2333 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2335 if (in_irq() || irqs_disabled())
2336 __dev_kfree_skb_irq(skb, reason);
2340 EXPORT_SYMBOL(__dev_kfree_skb_any);
2344 * netif_device_detach - mark device as removed
2345 * @dev: network device
2347 * Mark device as removed from system and therefore no longer available.
2349 void netif_device_detach(struct net_device *dev)
2351 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2352 netif_running(dev)) {
2353 netif_tx_stop_all_queues(dev);
2356 EXPORT_SYMBOL(netif_device_detach);
2359 * netif_device_attach - mark device as attached
2360 * @dev: network device
2362 * Mark device as attached from system and restart if needed.
2364 void netif_device_attach(struct net_device *dev)
2366 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2367 netif_running(dev)) {
2368 netif_tx_wake_all_queues(dev);
2369 __netdev_watchdog_up(dev);
2372 EXPORT_SYMBOL(netif_device_attach);
2375 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2376 * to be used as a distribution range.
2378 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2379 unsigned int num_tx_queues)
2383 u16 qcount = num_tx_queues;
2385 if (skb_rx_queue_recorded(skb)) {
2386 hash = skb_get_rx_queue(skb);
2387 while (unlikely(hash >= num_tx_queues))
2388 hash -= num_tx_queues;
2393 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2394 qoffset = dev->tc_to_txq[tc].offset;
2395 qcount = dev->tc_to_txq[tc].count;
2398 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2400 EXPORT_SYMBOL(__skb_tx_hash);
2402 static void skb_warn_bad_offload(const struct sk_buff *skb)
2404 static const netdev_features_t null_features = 0;
2405 struct net_device *dev = skb->dev;
2406 const char *driver = "";
2408 if (!net_ratelimit())
2411 if (dev && dev->dev.parent)
2412 driver = dev_driver_string(dev->dev.parent);
2414 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2415 "gso_type=%d ip_summed=%d\n",
2416 driver, dev ? &dev->features : &null_features,
2417 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2418 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2419 skb_shinfo(skb)->gso_type, skb->ip_summed);
2423 * Invalidate hardware checksum when packet is to be mangled, and
2424 * complete checksum manually on outgoing path.
2426 int skb_checksum_help(struct sk_buff *skb)
2429 int ret = 0, offset;
2431 if (skb->ip_summed == CHECKSUM_COMPLETE)
2432 goto out_set_summed;
2434 if (unlikely(skb_shinfo(skb)->gso_size)) {
2435 skb_warn_bad_offload(skb);
2439 /* Before computing a checksum, we should make sure no frag could
2440 * be modified by an external entity : checksum could be wrong.
2442 if (skb_has_shared_frag(skb)) {
2443 ret = __skb_linearize(skb);
2448 offset = skb_checksum_start_offset(skb);
2449 BUG_ON(offset >= skb_headlen(skb));
2450 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2452 offset += skb->csum_offset;
2453 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2455 if (skb_cloned(skb) &&
2456 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2457 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2462 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2464 skb->ip_summed = CHECKSUM_NONE;
2468 EXPORT_SYMBOL(skb_checksum_help);
2470 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2472 __be16 type = skb->protocol;
2474 /* Tunnel gso handlers can set protocol to ethernet. */
2475 if (type == htons(ETH_P_TEB)) {
2478 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2481 eth = (struct ethhdr *)skb_mac_header(skb);
2482 type = eth->h_proto;
2485 return __vlan_get_protocol(skb, type, depth);
2489 * skb_mac_gso_segment - mac layer segmentation handler.
2490 * @skb: buffer to segment
2491 * @features: features for the output path (see dev->features)
2493 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2494 netdev_features_t features)
2496 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2497 struct packet_offload *ptype;
2498 int vlan_depth = skb->mac_len;
2499 __be16 type = skb_network_protocol(skb, &vlan_depth);
2501 if (unlikely(!type))
2502 return ERR_PTR(-EINVAL);
2504 __skb_pull(skb, vlan_depth);
2507 list_for_each_entry_rcu(ptype, &offload_base, list) {
2508 if (ptype->type == type && ptype->callbacks.gso_segment) {
2509 segs = ptype->callbacks.gso_segment(skb, features);
2515 __skb_push(skb, skb->data - skb_mac_header(skb));
2519 EXPORT_SYMBOL(skb_mac_gso_segment);
2522 /* openvswitch calls this on rx path, so we need a different check.
2524 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2527 return skb->ip_summed != CHECKSUM_PARTIAL;
2529 return skb->ip_summed == CHECKSUM_NONE;
2533 * __skb_gso_segment - Perform segmentation on skb.
2534 * @skb: buffer to segment
2535 * @features: features for the output path (see dev->features)
2536 * @tx_path: whether it is called in TX path
2538 * This function segments the given skb and returns a list of segments.
2540 * It may return NULL if the skb requires no segmentation. This is
2541 * only possible when GSO is used for verifying header integrity.
2543 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2544 netdev_features_t features, bool tx_path)
2546 if (unlikely(skb_needs_check(skb, tx_path))) {
2549 skb_warn_bad_offload(skb);
2551 err = skb_cow_head(skb, 0);
2553 return ERR_PTR(err);
2556 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2557 SKB_GSO_CB(skb)->encap_level = 0;
2559 skb_reset_mac_header(skb);
2560 skb_reset_mac_len(skb);
2562 return skb_mac_gso_segment(skb, features);
2564 EXPORT_SYMBOL(__skb_gso_segment);
2566 /* Take action when hardware reception checksum errors are detected. */
2568 void netdev_rx_csum_fault(struct net_device *dev)
2570 if (net_ratelimit()) {
2571 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2575 EXPORT_SYMBOL(netdev_rx_csum_fault);
2578 /* Actually, we should eliminate this check as soon as we know, that:
2579 * 1. IOMMU is present and allows to map all the memory.
2580 * 2. No high memory really exists on this machine.
2583 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2585 #ifdef CONFIG_HIGHMEM
2587 if (!(dev->features & NETIF_F_HIGHDMA)) {
2588 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2589 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2590 if (PageHighMem(skb_frag_page(frag)))
2595 if (PCI_DMA_BUS_IS_PHYS) {
2596 struct device *pdev = dev->dev.parent;
2600 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2601 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2602 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2603 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2611 /* If MPLS offload request, verify we are testing hardware MPLS features
2612 * instead of standard features for the netdev.
2614 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2615 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2616 netdev_features_t features,
2619 if (eth_p_mpls(type))
2620 features &= skb->dev->mpls_features;
2625 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2626 netdev_features_t features,
2633 static netdev_features_t harmonize_features(struct sk_buff *skb,
2634 netdev_features_t features)
2639 type = skb_network_protocol(skb, &tmp);
2640 features = net_mpls_features(skb, features, type);
2642 if (skb->ip_summed != CHECKSUM_NONE &&
2643 !can_checksum_protocol(features, type)) {
2644 features &= ~NETIF_F_ALL_CSUM;
2645 } else if (illegal_highdma(skb->dev, skb)) {
2646 features &= ~NETIF_F_SG;
2652 netdev_features_t passthru_features_check(struct sk_buff *skb,
2653 struct net_device *dev,
2654 netdev_features_t features)
2658 EXPORT_SYMBOL(passthru_features_check);
2660 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2661 struct net_device *dev,
2662 netdev_features_t features)
2664 return vlan_features_check(skb, features);
2667 netdev_features_t netif_skb_features(struct sk_buff *skb)
2669 struct net_device *dev = skb->dev;
2670 netdev_features_t features = dev->features;
2671 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2673 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2674 features &= ~NETIF_F_GSO_MASK;
2676 /* If encapsulation offload request, verify we are testing
2677 * hardware encapsulation features instead of standard
2678 * features for the netdev
2680 if (skb->encapsulation)
2681 features &= dev->hw_enc_features;
2683 if (skb_vlan_tagged(skb))
2684 features = netdev_intersect_features(features,
2685 dev->vlan_features |
2686 NETIF_F_HW_VLAN_CTAG_TX |
2687 NETIF_F_HW_VLAN_STAG_TX);
2689 if (dev->netdev_ops->ndo_features_check)
2690 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2693 features &= dflt_features_check(skb, dev, features);
2695 return harmonize_features(skb, features);
2697 EXPORT_SYMBOL(netif_skb_features);
2699 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2700 struct netdev_queue *txq, bool more)
2705 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2706 dev_queue_xmit_nit(skb, dev);
2709 trace_net_dev_start_xmit(skb, dev);
2710 rc = netdev_start_xmit(skb, dev, txq, more);
2711 trace_net_dev_xmit(skb, rc, dev, len);
2716 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2717 struct netdev_queue *txq, int *ret)
2719 struct sk_buff *skb = first;
2720 int rc = NETDEV_TX_OK;
2723 struct sk_buff *next = skb->next;
2726 rc = xmit_one(skb, dev, txq, next != NULL);
2727 if (unlikely(!dev_xmit_complete(rc))) {
2733 if (netif_xmit_stopped(txq) && skb) {
2734 rc = NETDEV_TX_BUSY;
2744 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2745 netdev_features_t features)
2747 if (skb_vlan_tag_present(skb) &&
2748 !vlan_hw_offload_capable(features, skb->vlan_proto))
2749 skb = __vlan_hwaccel_push_inside(skb);
2753 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2755 netdev_features_t features;
2760 features = netif_skb_features(skb);
2761 skb = validate_xmit_vlan(skb, features);
2765 if (netif_needs_gso(skb, features)) {
2766 struct sk_buff *segs;
2768 segs = skb_gso_segment(skb, features);
2776 if (skb_needs_linearize(skb, features) &&
2777 __skb_linearize(skb))
2780 /* If packet is not checksummed and device does not
2781 * support checksumming for this protocol, complete
2782 * checksumming here.
2784 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2785 if (skb->encapsulation)
2786 skb_set_inner_transport_header(skb,
2787 skb_checksum_start_offset(skb));
2789 skb_set_transport_header(skb,
2790 skb_checksum_start_offset(skb));
2791 if (!(features & NETIF_F_ALL_CSUM) &&
2792 skb_checksum_help(skb))
2805 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2807 struct sk_buff *next, *head = NULL, *tail;
2809 for (; skb != NULL; skb = next) {
2813 /* in case skb wont be segmented, point to itself */
2816 skb = validate_xmit_skb(skb, dev);
2824 /* If skb was segmented, skb->prev points to
2825 * the last segment. If not, it still contains skb.
2832 static void qdisc_pkt_len_init(struct sk_buff *skb)
2834 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2836 qdisc_skb_cb(skb)->pkt_len = skb->len;
2838 /* To get more precise estimation of bytes sent on wire,
2839 * we add to pkt_len the headers size of all segments
2841 if (shinfo->gso_size) {
2842 unsigned int hdr_len;
2843 u16 gso_segs = shinfo->gso_segs;
2845 /* mac layer + network layer */
2846 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2848 /* + transport layer */
2849 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2850 hdr_len += tcp_hdrlen(skb);
2852 hdr_len += sizeof(struct udphdr);
2854 if (shinfo->gso_type & SKB_GSO_DODGY)
2855 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2858 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2862 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2863 struct net_device *dev,
2864 struct netdev_queue *txq)
2866 spinlock_t *root_lock = qdisc_lock(q);
2870 qdisc_pkt_len_init(skb);
2871 qdisc_calculate_pkt_len(skb, q);
2873 * Heuristic to force contended enqueues to serialize on a
2874 * separate lock before trying to get qdisc main lock.
2875 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2876 * often and dequeue packets faster.
2878 contended = qdisc_is_running(q);
2879 if (unlikely(contended))
2880 spin_lock(&q->busylock);
2882 spin_lock(root_lock);
2883 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2886 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2887 qdisc_run_begin(q)) {
2889 * This is a work-conserving queue; there are no old skbs
2890 * waiting to be sent out; and the qdisc is not running -
2891 * xmit the skb directly.
2894 qdisc_bstats_update(q, skb);
2896 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2897 if (unlikely(contended)) {
2898 spin_unlock(&q->busylock);
2905 rc = NET_XMIT_SUCCESS;
2907 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2908 if (qdisc_run_begin(q)) {
2909 if (unlikely(contended)) {
2910 spin_unlock(&q->busylock);
2916 spin_unlock(root_lock);
2917 if (unlikely(contended))
2918 spin_unlock(&q->busylock);
2922 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2923 static void skb_update_prio(struct sk_buff *skb)
2925 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2927 if (!skb->priority && skb->sk && map) {
2928 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2930 if (prioidx < map->priomap_len)
2931 skb->priority = map->priomap[prioidx];
2935 #define skb_update_prio(skb)
2938 DEFINE_PER_CPU(int, xmit_recursion);
2939 EXPORT_SYMBOL(xmit_recursion);
2941 #define RECURSION_LIMIT 10
2944 * dev_loopback_xmit - loop back @skb
2945 * @net: network namespace this loopback is happening in
2946 * @sk: sk needed to be a netfilter okfn
2947 * @skb: buffer to transmit
2949 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
2951 skb_reset_mac_header(skb);
2952 __skb_pull(skb, skb_network_offset(skb));
2953 skb->pkt_type = PACKET_LOOPBACK;
2954 skb->ip_summed = CHECKSUM_UNNECESSARY;
2955 WARN_ON(!skb_dst(skb));
2960 EXPORT_SYMBOL(dev_loopback_xmit);
2962 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2965 struct xps_dev_maps *dev_maps;
2966 struct xps_map *map;
2967 int queue_index = -1;
2970 dev_maps = rcu_dereference(dev->xps_maps);
2972 map = rcu_dereference(
2973 dev_maps->cpu_map[skb->sender_cpu - 1]);
2976 queue_index = map->queues[0];
2978 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
2980 if (unlikely(queue_index >= dev->real_num_tx_queues))
2992 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
2994 struct sock *sk = skb->sk;
2995 int queue_index = sk_tx_queue_get(sk);
2997 if (queue_index < 0 || skb->ooo_okay ||
2998 queue_index >= dev->real_num_tx_queues) {
2999 int new_index = get_xps_queue(dev, skb);
3001 new_index = skb_tx_hash(dev, skb);
3003 if (queue_index != new_index && sk &&
3005 rcu_access_pointer(sk->sk_dst_cache))
3006 sk_tx_queue_set(sk, new_index);
3008 queue_index = new_index;
3014 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3015 struct sk_buff *skb,
3018 int queue_index = 0;
3021 if (skb->sender_cpu == 0)
3022 skb->sender_cpu = raw_smp_processor_id() + 1;
3025 if (dev->real_num_tx_queues != 1) {
3026 const struct net_device_ops *ops = dev->netdev_ops;
3027 if (ops->ndo_select_queue)
3028 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3031 queue_index = __netdev_pick_tx(dev, skb);
3034 queue_index = netdev_cap_txqueue(dev, queue_index);
3037 skb_set_queue_mapping(skb, queue_index);
3038 return netdev_get_tx_queue(dev, queue_index);
3042 * __dev_queue_xmit - transmit a buffer
3043 * @skb: buffer to transmit
3044 * @accel_priv: private data used for L2 forwarding offload
3046 * Queue a buffer for transmission to a network device. The caller must
3047 * have set the device and priority and built the buffer before calling
3048 * this function. The function can be called from an interrupt.
3050 * A negative errno code is returned on a failure. A success does not
3051 * guarantee the frame will be transmitted as it may be dropped due
3052 * to congestion or traffic shaping.
3054 * -----------------------------------------------------------------------------------
3055 * I notice this method can also return errors from the queue disciplines,
3056 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3059 * Regardless of the return value, the skb is consumed, so it is currently
3060 * difficult to retry a send to this method. (You can bump the ref count
3061 * before sending to hold a reference for retry if you are careful.)
3063 * When calling this method, interrupts MUST be enabled. This is because
3064 * the BH enable code must have IRQs enabled so that it will not deadlock.
3067 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3069 struct net_device *dev = skb->dev;
3070 struct netdev_queue *txq;
3074 skb_reset_mac_header(skb);
3076 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3077 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3079 /* Disable soft irqs for various locks below. Also
3080 * stops preemption for RCU.
3084 skb_update_prio(skb);
3086 /* If device/qdisc don't need skb->dst, release it right now while
3087 * its hot in this cpu cache.
3089 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3094 #ifdef CONFIG_NET_SWITCHDEV
3095 /* Don't forward if offload device already forwarded */
3096 if (skb->offload_fwd_mark &&
3097 skb->offload_fwd_mark == dev->offload_fwd_mark) {
3099 rc = NET_XMIT_SUCCESS;
3104 txq = netdev_pick_tx(dev, skb, accel_priv);
3105 q = rcu_dereference_bh(txq->qdisc);
3107 #ifdef CONFIG_NET_CLS_ACT
3108 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3110 trace_net_dev_queue(skb);
3112 rc = __dev_xmit_skb(skb, q, dev, txq);
3116 /* The device has no queue. Common case for software devices:
3117 loopback, all the sorts of tunnels...
3119 Really, it is unlikely that netif_tx_lock protection is necessary
3120 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3122 However, it is possible, that they rely on protection
3125 Check this and shot the lock. It is not prone from deadlocks.
3126 Either shot noqueue qdisc, it is even simpler 8)
3128 if (dev->flags & IFF_UP) {
3129 int cpu = smp_processor_id(); /* ok because BHs are off */
3131 if (txq->xmit_lock_owner != cpu) {
3133 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
3134 goto recursion_alert;
3136 skb = validate_xmit_skb(skb, dev);
3140 HARD_TX_LOCK(dev, txq, cpu);
3142 if (!netif_xmit_stopped(txq)) {
3143 __this_cpu_inc(xmit_recursion);
3144 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3145 __this_cpu_dec(xmit_recursion);
3146 if (dev_xmit_complete(rc)) {
3147 HARD_TX_UNLOCK(dev, txq);
3151 HARD_TX_UNLOCK(dev, txq);
3152 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3155 /* Recursion is detected! It is possible,
3159 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3166 rcu_read_unlock_bh();
3168 atomic_long_inc(&dev->tx_dropped);
3169 kfree_skb_list(skb);
3172 rcu_read_unlock_bh();
3176 int dev_queue_xmit(struct sk_buff *skb)
3178 return __dev_queue_xmit(skb, NULL);
3180 EXPORT_SYMBOL(dev_queue_xmit);
3182 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3184 return __dev_queue_xmit(skb, accel_priv);
3186 EXPORT_SYMBOL(dev_queue_xmit_accel);
3189 /*=======================================================================
3191 =======================================================================*/
3193 int netdev_max_backlog __read_mostly = 1000;
3194 EXPORT_SYMBOL(netdev_max_backlog);
3196 int netdev_tstamp_prequeue __read_mostly = 1;
3197 int netdev_budget __read_mostly = 300;
3198 int weight_p __read_mostly = 64; /* old backlog weight */
3200 /* Called with irq disabled */
3201 static inline void ____napi_schedule(struct softnet_data *sd,
3202 struct napi_struct *napi)
3204 list_add_tail(&napi->poll_list, &sd->poll_list);
3205 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3210 /* One global table that all flow-based protocols share. */
3211 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3212 EXPORT_SYMBOL(rps_sock_flow_table);
3213 u32 rps_cpu_mask __read_mostly;
3214 EXPORT_SYMBOL(rps_cpu_mask);
3216 struct static_key rps_needed __read_mostly;
3218 static struct rps_dev_flow *
3219 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3220 struct rps_dev_flow *rflow, u16 next_cpu)
3222 if (next_cpu < nr_cpu_ids) {
3223 #ifdef CONFIG_RFS_ACCEL
3224 struct netdev_rx_queue *rxqueue;
3225 struct rps_dev_flow_table *flow_table;
3226 struct rps_dev_flow *old_rflow;
3231 /* Should we steer this flow to a different hardware queue? */
3232 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3233 !(dev->features & NETIF_F_NTUPLE))
3235 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3236 if (rxq_index == skb_get_rx_queue(skb))
3239 rxqueue = dev->_rx + rxq_index;
3240 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3243 flow_id = skb_get_hash(skb) & flow_table->mask;
3244 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3245 rxq_index, flow_id);
3249 rflow = &flow_table->flows[flow_id];
3251 if (old_rflow->filter == rflow->filter)
3252 old_rflow->filter = RPS_NO_FILTER;
3256 per_cpu(softnet_data, next_cpu).input_queue_head;
3259 rflow->cpu = next_cpu;
3264 * get_rps_cpu is called from netif_receive_skb and returns the target
3265 * CPU from the RPS map of the receiving queue for a given skb.
3266 * rcu_read_lock must be held on entry.
3268 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3269 struct rps_dev_flow **rflowp)
3271 const struct rps_sock_flow_table *sock_flow_table;
3272 struct netdev_rx_queue *rxqueue = dev->_rx;
3273 struct rps_dev_flow_table *flow_table;
3274 struct rps_map *map;
3279 if (skb_rx_queue_recorded(skb)) {
3280 u16 index = skb_get_rx_queue(skb);
3282 if (unlikely(index >= dev->real_num_rx_queues)) {
3283 WARN_ONCE(dev->real_num_rx_queues > 1,
3284 "%s received packet on queue %u, but number "
3285 "of RX queues is %u\n",
3286 dev->name, index, dev->real_num_rx_queues);
3292 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3294 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3295 map = rcu_dereference(rxqueue->rps_map);
3296 if (!flow_table && !map)
3299 skb_reset_network_header(skb);
3300 hash = skb_get_hash(skb);
3304 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3305 if (flow_table && sock_flow_table) {
3306 struct rps_dev_flow *rflow;
3310 /* First check into global flow table if there is a match */
3311 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3312 if ((ident ^ hash) & ~rps_cpu_mask)
3315 next_cpu = ident & rps_cpu_mask;
3317 /* OK, now we know there is a match,
3318 * we can look at the local (per receive queue) flow table
3320 rflow = &flow_table->flows[hash & flow_table->mask];
3324 * If the desired CPU (where last recvmsg was done) is
3325 * different from current CPU (one in the rx-queue flow
3326 * table entry), switch if one of the following holds:
3327 * - Current CPU is unset (>= nr_cpu_ids).
3328 * - Current CPU is offline.
3329 * - The current CPU's queue tail has advanced beyond the
3330 * last packet that was enqueued using this table entry.
3331 * This guarantees that all previous packets for the flow
3332 * have been dequeued, thus preserving in order delivery.
3334 if (unlikely(tcpu != next_cpu) &&
3335 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3336 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3337 rflow->last_qtail)) >= 0)) {
3339 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3342 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3352 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3353 if (cpu_online(tcpu)) {
3363 #ifdef CONFIG_RFS_ACCEL
3366 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3367 * @dev: Device on which the filter was set
3368 * @rxq_index: RX queue index
3369 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3370 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3372 * Drivers that implement ndo_rx_flow_steer() should periodically call
3373 * this function for each installed filter and remove the filters for
3374 * which it returns %true.
3376 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3377 u32 flow_id, u16 filter_id)
3379 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3380 struct rps_dev_flow_table *flow_table;
3381 struct rps_dev_flow *rflow;
3386 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3387 if (flow_table && flow_id <= flow_table->mask) {
3388 rflow = &flow_table->flows[flow_id];
3389 cpu = ACCESS_ONCE(rflow->cpu);
3390 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3391 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3392 rflow->last_qtail) <
3393 (int)(10 * flow_table->mask)))
3399 EXPORT_SYMBOL(rps_may_expire_flow);
3401 #endif /* CONFIG_RFS_ACCEL */
3403 /* Called from hardirq (IPI) context */
3404 static void rps_trigger_softirq(void *data)
3406 struct softnet_data *sd = data;
3408 ____napi_schedule(sd, &sd->backlog);
3412 #endif /* CONFIG_RPS */
3415 * Check if this softnet_data structure is another cpu one
3416 * If yes, queue it to our IPI list and return 1
3419 static int rps_ipi_queued(struct softnet_data *sd)
3422 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3425 sd->rps_ipi_next = mysd->rps_ipi_list;
3426 mysd->rps_ipi_list = sd;
3428 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3431 #endif /* CONFIG_RPS */
3435 #ifdef CONFIG_NET_FLOW_LIMIT
3436 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3439 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3441 #ifdef CONFIG_NET_FLOW_LIMIT
3442 struct sd_flow_limit *fl;
3443 struct softnet_data *sd;
3444 unsigned int old_flow, new_flow;
3446 if (qlen < (netdev_max_backlog >> 1))
3449 sd = this_cpu_ptr(&softnet_data);
3452 fl = rcu_dereference(sd->flow_limit);
3454 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3455 old_flow = fl->history[fl->history_head];
3456 fl->history[fl->history_head] = new_flow;
3459 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3461 if (likely(fl->buckets[old_flow]))
3462 fl->buckets[old_flow]--;
3464 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3476 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3477 * queue (may be a remote CPU queue).
3479 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3480 unsigned int *qtail)
3482 struct softnet_data *sd;
3483 unsigned long flags;
3486 sd = &per_cpu(softnet_data, cpu);
3488 local_irq_save(flags);
3491 if (!netif_running(skb->dev))
3493 qlen = skb_queue_len(&sd->input_pkt_queue);
3494 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3497 __skb_queue_tail(&sd->input_pkt_queue, skb);
3498 input_queue_tail_incr_save(sd, qtail);
3500 local_irq_restore(flags);
3501 return NET_RX_SUCCESS;
3504 /* Schedule NAPI for backlog device
3505 * We can use non atomic operation since we own the queue lock
3507 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3508 if (!rps_ipi_queued(sd))
3509 ____napi_schedule(sd, &sd->backlog);
3518 local_irq_restore(flags);
3520 atomic_long_inc(&skb->dev->rx_dropped);
3525 static int netif_rx_internal(struct sk_buff *skb)
3529 net_timestamp_check(netdev_tstamp_prequeue, skb);
3531 trace_netif_rx(skb);
3533 if (static_key_false(&rps_needed)) {
3534 struct rps_dev_flow voidflow, *rflow = &voidflow;
3540 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3542 cpu = smp_processor_id();
3544 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3552 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3559 * netif_rx - post buffer to the network code
3560 * @skb: buffer to post
3562 * This function receives a packet from a device driver and queues it for
3563 * the upper (protocol) levels to process. It always succeeds. The buffer
3564 * may be dropped during processing for congestion control or by the
3568 * NET_RX_SUCCESS (no congestion)
3569 * NET_RX_DROP (packet was dropped)
3573 int netif_rx(struct sk_buff *skb)
3575 trace_netif_rx_entry(skb);
3577 return netif_rx_internal(skb);
3579 EXPORT_SYMBOL(netif_rx);
3581 int netif_rx_ni(struct sk_buff *skb)
3585 trace_netif_rx_ni_entry(skb);
3588 err = netif_rx_internal(skb);
3589 if (local_softirq_pending())
3595 EXPORT_SYMBOL(netif_rx_ni);
3597 static void net_tx_action(struct softirq_action *h)
3599 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3601 if (sd->completion_queue) {
3602 struct sk_buff *clist;
3604 local_irq_disable();
3605 clist = sd->completion_queue;
3606 sd->completion_queue = NULL;
3610 struct sk_buff *skb = clist;
3611 clist = clist->next;
3613 WARN_ON(atomic_read(&skb->users));
3614 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3615 trace_consume_skb(skb);
3617 trace_kfree_skb(skb, net_tx_action);
3622 if (sd->output_queue) {
3625 local_irq_disable();
3626 head = sd->output_queue;
3627 sd->output_queue = NULL;
3628 sd->output_queue_tailp = &sd->output_queue;
3632 struct Qdisc *q = head;
3633 spinlock_t *root_lock;
3635 head = head->next_sched;
3637 root_lock = qdisc_lock(q);
3638 if (spin_trylock(root_lock)) {
3639 smp_mb__before_atomic();
3640 clear_bit(__QDISC_STATE_SCHED,
3643 spin_unlock(root_lock);
3645 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3647 __netif_reschedule(q);
3649 smp_mb__before_atomic();
3650 clear_bit(__QDISC_STATE_SCHED,
3658 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3659 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3660 /* This hook is defined here for ATM LANE */
3661 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3662 unsigned char *addr) __read_mostly;
3663 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3666 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3667 struct packet_type **pt_prev,
3668 int *ret, struct net_device *orig_dev)
3670 #ifdef CONFIG_NET_CLS_ACT
3671 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3672 struct tcf_result cl_res;
3674 /* If there's at least one ingress present somewhere (so
3675 * we get here via enabled static key), remaining devices
3676 * that are not configured with an ingress qdisc will bail
3682 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3686 qdisc_skb_cb(skb)->pkt_len = skb->len;
3687 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3688 qdisc_bstats_cpu_update(cl->q, skb);
3690 switch (tc_classify(skb, cl, &cl_res, false)) {
3692 case TC_ACT_RECLASSIFY:
3693 skb->tc_index = TC_H_MIN(cl_res.classid);
3696 qdisc_qstats_cpu_drop(cl->q);
3701 case TC_ACT_REDIRECT:
3702 /* skb_mac_header check was done by cls/act_bpf, so
3703 * we can safely push the L2 header back before
3704 * redirecting to another netdev
3706 __skb_push(skb, skb->mac_len);
3707 skb_do_redirect(skb);
3712 #endif /* CONFIG_NET_CLS_ACT */
3717 * netdev_rx_handler_register - register receive handler
3718 * @dev: device to register a handler for
3719 * @rx_handler: receive handler to register
3720 * @rx_handler_data: data pointer that is used by rx handler
3722 * Register a receive handler for a device. This handler will then be
3723 * called from __netif_receive_skb. A negative errno code is returned
3726 * The caller must hold the rtnl_mutex.
3728 * For a general description of rx_handler, see enum rx_handler_result.
3730 int netdev_rx_handler_register(struct net_device *dev,
3731 rx_handler_func_t *rx_handler,
3732 void *rx_handler_data)
3736 if (dev->rx_handler)
3739 /* Note: rx_handler_data must be set before rx_handler */
3740 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3741 rcu_assign_pointer(dev->rx_handler, rx_handler);
3745 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3748 * netdev_rx_handler_unregister - unregister receive handler
3749 * @dev: device to unregister a handler from
3751 * Unregister a receive handler from a device.
3753 * The caller must hold the rtnl_mutex.
3755 void netdev_rx_handler_unregister(struct net_device *dev)
3759 RCU_INIT_POINTER(dev->rx_handler, NULL);
3760 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3761 * section has a guarantee to see a non NULL rx_handler_data
3765 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3767 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3770 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3771 * the special handling of PFMEMALLOC skbs.
3773 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3775 switch (skb->protocol) {
3776 case htons(ETH_P_ARP):
3777 case htons(ETH_P_IP):
3778 case htons(ETH_P_IPV6):
3779 case htons(ETH_P_8021Q):
3780 case htons(ETH_P_8021AD):
3787 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
3788 int *ret, struct net_device *orig_dev)
3790 #ifdef CONFIG_NETFILTER_INGRESS
3791 if (nf_hook_ingress_active(skb)) {
3793 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3797 return nf_hook_ingress(skb);
3799 #endif /* CONFIG_NETFILTER_INGRESS */
3803 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3805 struct packet_type *ptype, *pt_prev;
3806 rx_handler_func_t *rx_handler;
3807 struct net_device *orig_dev;
3808 bool deliver_exact = false;
3809 int ret = NET_RX_DROP;
3812 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3814 trace_netif_receive_skb(skb);
3816 orig_dev = skb->dev;
3818 skb_reset_network_header(skb);
3819 if (!skb_transport_header_was_set(skb))
3820 skb_reset_transport_header(skb);
3821 skb_reset_mac_len(skb);
3826 skb->skb_iif = skb->dev->ifindex;
3828 __this_cpu_inc(softnet_data.processed);
3830 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3831 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3832 skb = skb_vlan_untag(skb);
3837 #ifdef CONFIG_NET_CLS_ACT
3838 if (skb->tc_verd & TC_NCLS) {
3839 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3847 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3849 ret = deliver_skb(skb, pt_prev, orig_dev);
3853 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3855 ret = deliver_skb(skb, pt_prev, orig_dev);
3860 #ifdef CONFIG_NET_INGRESS
3861 if (static_key_false(&ingress_needed)) {
3862 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3866 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
3870 #ifdef CONFIG_NET_CLS_ACT
3874 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3877 if (skb_vlan_tag_present(skb)) {
3879 ret = deliver_skb(skb, pt_prev, orig_dev);
3882 if (vlan_do_receive(&skb))
3884 else if (unlikely(!skb))
3888 rx_handler = rcu_dereference(skb->dev->rx_handler);
3891 ret = deliver_skb(skb, pt_prev, orig_dev);
3894 switch (rx_handler(&skb)) {
3895 case RX_HANDLER_CONSUMED:
3896 ret = NET_RX_SUCCESS;
3898 case RX_HANDLER_ANOTHER:
3900 case RX_HANDLER_EXACT:
3901 deliver_exact = true;
3902 case RX_HANDLER_PASS:
3909 if (unlikely(skb_vlan_tag_present(skb))) {
3910 if (skb_vlan_tag_get_id(skb))
3911 skb->pkt_type = PACKET_OTHERHOST;
3912 /* Note: we might in the future use prio bits
3913 * and set skb->priority like in vlan_do_receive()
3914 * For the time being, just ignore Priority Code Point
3919 type = skb->protocol;
3921 /* deliver only exact match when indicated */
3922 if (likely(!deliver_exact)) {
3923 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3924 &ptype_base[ntohs(type) &
3928 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3929 &orig_dev->ptype_specific);
3931 if (unlikely(skb->dev != orig_dev)) {
3932 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3933 &skb->dev->ptype_specific);
3937 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3940 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3943 atomic_long_inc(&skb->dev->rx_dropped);
3945 /* Jamal, now you will not able to escape explaining
3946 * me how you were going to use this. :-)
3955 static int __netif_receive_skb(struct sk_buff *skb)
3959 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3960 unsigned long pflags = current->flags;
3963 * PFMEMALLOC skbs are special, they should
3964 * - be delivered to SOCK_MEMALLOC sockets only
3965 * - stay away from userspace
3966 * - have bounded memory usage
3968 * Use PF_MEMALLOC as this saves us from propagating the allocation
3969 * context down to all allocation sites.
3971 current->flags |= PF_MEMALLOC;
3972 ret = __netif_receive_skb_core(skb, true);
3973 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3975 ret = __netif_receive_skb_core(skb, false);
3980 static int netif_receive_skb_internal(struct sk_buff *skb)
3984 net_timestamp_check(netdev_tstamp_prequeue, skb);
3986 if (skb_defer_rx_timestamp(skb))
3987 return NET_RX_SUCCESS;
3992 if (static_key_false(&rps_needed)) {
3993 struct rps_dev_flow voidflow, *rflow = &voidflow;
3994 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
3997 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4003 ret = __netif_receive_skb(skb);
4009 * netif_receive_skb - process receive buffer from network
4010 * @skb: buffer to process
4012 * netif_receive_skb() is the main receive data processing function.
4013 * It always succeeds. The buffer may be dropped during processing
4014 * for congestion control or by the protocol layers.
4016 * This function may only be called from softirq context and interrupts
4017 * should be enabled.
4019 * Return values (usually ignored):
4020 * NET_RX_SUCCESS: no congestion
4021 * NET_RX_DROP: packet was dropped
4023 int netif_receive_skb(struct sk_buff *skb)
4025 trace_netif_receive_skb_entry(skb);
4027 return netif_receive_skb_internal(skb);
4029 EXPORT_SYMBOL(netif_receive_skb);
4031 /* Network device is going away, flush any packets still pending
4032 * Called with irqs disabled.
4034 static void flush_backlog(void *arg)
4036 struct net_device *dev = arg;
4037 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4038 struct sk_buff *skb, *tmp;
4041 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4042 if (skb->dev == dev) {
4043 __skb_unlink(skb, &sd->input_pkt_queue);
4045 input_queue_head_incr(sd);
4050 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4051 if (skb->dev == dev) {
4052 __skb_unlink(skb, &sd->process_queue);
4054 input_queue_head_incr(sd);
4059 static int napi_gro_complete(struct sk_buff *skb)
4061 struct packet_offload *ptype;
4062 __be16 type = skb->protocol;
4063 struct list_head *head = &offload_base;
4066 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4068 if (NAPI_GRO_CB(skb)->count == 1) {
4069 skb_shinfo(skb)->gso_size = 0;
4074 list_for_each_entry_rcu(ptype, head, list) {
4075 if (ptype->type != type || !ptype->callbacks.gro_complete)
4078 err = ptype->callbacks.gro_complete(skb, 0);
4084 WARN_ON(&ptype->list == head);
4086 return NET_RX_SUCCESS;
4090 return netif_receive_skb_internal(skb);
4093 /* napi->gro_list contains packets ordered by age.
4094 * youngest packets at the head of it.
4095 * Complete skbs in reverse order to reduce latencies.
4097 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4099 struct sk_buff *skb, *prev = NULL;
4101 /* scan list and build reverse chain */
4102 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4107 for (skb = prev; skb; skb = prev) {
4110 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4114 napi_gro_complete(skb);
4118 napi->gro_list = NULL;
4120 EXPORT_SYMBOL(napi_gro_flush);
4122 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4125 unsigned int maclen = skb->dev->hard_header_len;
4126 u32 hash = skb_get_hash_raw(skb);
4128 for (p = napi->gro_list; p; p = p->next) {
4129 unsigned long diffs;
4131 NAPI_GRO_CB(p)->flush = 0;
4133 if (hash != skb_get_hash_raw(p)) {
4134 NAPI_GRO_CB(p)->same_flow = 0;
4138 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4139 diffs |= p->vlan_tci ^ skb->vlan_tci;
4140 if (maclen == ETH_HLEN)
4141 diffs |= compare_ether_header(skb_mac_header(p),
4142 skb_mac_header(skb));
4144 diffs = memcmp(skb_mac_header(p),
4145 skb_mac_header(skb),
4147 NAPI_GRO_CB(p)->same_flow = !diffs;
4151 static void skb_gro_reset_offset(struct sk_buff *skb)
4153 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4154 const skb_frag_t *frag0 = &pinfo->frags[0];
4156 NAPI_GRO_CB(skb)->data_offset = 0;
4157 NAPI_GRO_CB(skb)->frag0 = NULL;
4158 NAPI_GRO_CB(skb)->frag0_len = 0;
4160 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4162 !PageHighMem(skb_frag_page(frag0))) {
4163 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4164 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
4168 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4170 struct skb_shared_info *pinfo = skb_shinfo(skb);
4172 BUG_ON(skb->end - skb->tail < grow);
4174 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4176 skb->data_len -= grow;
4179 pinfo->frags[0].page_offset += grow;
4180 skb_frag_size_sub(&pinfo->frags[0], grow);
4182 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4183 skb_frag_unref(skb, 0);
4184 memmove(pinfo->frags, pinfo->frags + 1,
4185 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4189 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4191 struct sk_buff **pp = NULL;
4192 struct packet_offload *ptype;
4193 __be16 type = skb->protocol;
4194 struct list_head *head = &offload_base;
4196 enum gro_result ret;
4199 if (!(skb->dev->features & NETIF_F_GRO))
4202 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4205 gro_list_prepare(napi, skb);
4208 list_for_each_entry_rcu(ptype, head, list) {
4209 if (ptype->type != type || !ptype->callbacks.gro_receive)
4212 skb_set_network_header(skb, skb_gro_offset(skb));
4213 skb_reset_mac_len(skb);
4214 NAPI_GRO_CB(skb)->same_flow = 0;
4215 NAPI_GRO_CB(skb)->flush = 0;
4216 NAPI_GRO_CB(skb)->free = 0;
4217 NAPI_GRO_CB(skb)->udp_mark = 0;
4218 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4220 /* Setup for GRO checksum validation */
4221 switch (skb->ip_summed) {
4222 case CHECKSUM_COMPLETE:
4223 NAPI_GRO_CB(skb)->csum = skb->csum;
4224 NAPI_GRO_CB(skb)->csum_valid = 1;
4225 NAPI_GRO_CB(skb)->csum_cnt = 0;
4227 case CHECKSUM_UNNECESSARY:
4228 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4229 NAPI_GRO_CB(skb)->csum_valid = 0;
4232 NAPI_GRO_CB(skb)->csum_cnt = 0;
4233 NAPI_GRO_CB(skb)->csum_valid = 0;
4236 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4241 if (&ptype->list == head)
4244 same_flow = NAPI_GRO_CB(skb)->same_flow;
4245 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4248 struct sk_buff *nskb = *pp;
4252 napi_gro_complete(nskb);
4259 if (NAPI_GRO_CB(skb)->flush)
4262 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4263 struct sk_buff *nskb = napi->gro_list;
4265 /* locate the end of the list to select the 'oldest' flow */
4266 while (nskb->next) {
4272 napi_gro_complete(nskb);
4276 NAPI_GRO_CB(skb)->count = 1;
4277 NAPI_GRO_CB(skb)->age = jiffies;
4278 NAPI_GRO_CB(skb)->last = skb;
4279 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4280 skb->next = napi->gro_list;
4281 napi->gro_list = skb;
4285 grow = skb_gro_offset(skb) - skb_headlen(skb);
4287 gro_pull_from_frag0(skb, grow);
4296 struct packet_offload *gro_find_receive_by_type(__be16 type)
4298 struct list_head *offload_head = &offload_base;
4299 struct packet_offload *ptype;
4301 list_for_each_entry_rcu(ptype, offload_head, list) {
4302 if (ptype->type != type || !ptype->callbacks.gro_receive)
4308 EXPORT_SYMBOL(gro_find_receive_by_type);
4310 struct packet_offload *gro_find_complete_by_type(__be16 type)
4312 struct list_head *offload_head = &offload_base;
4313 struct packet_offload *ptype;
4315 list_for_each_entry_rcu(ptype, offload_head, list) {
4316 if (ptype->type != type || !ptype->callbacks.gro_complete)
4322 EXPORT_SYMBOL(gro_find_complete_by_type);
4324 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4328 if (netif_receive_skb_internal(skb))
4336 case GRO_MERGED_FREE:
4337 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4338 kmem_cache_free(skbuff_head_cache, skb);
4351 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4353 trace_napi_gro_receive_entry(skb);
4355 skb_gro_reset_offset(skb);
4357 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4359 EXPORT_SYMBOL(napi_gro_receive);
4361 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4363 if (unlikely(skb->pfmemalloc)) {
4367 __skb_pull(skb, skb_headlen(skb));
4368 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4369 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4371 skb->dev = napi->dev;
4373 skb->encapsulation = 0;
4374 skb_shinfo(skb)->gso_type = 0;
4375 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4380 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4382 struct sk_buff *skb = napi->skb;
4385 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4390 EXPORT_SYMBOL(napi_get_frags);
4392 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4393 struct sk_buff *skb,
4399 __skb_push(skb, ETH_HLEN);
4400 skb->protocol = eth_type_trans(skb, skb->dev);
4401 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4406 case GRO_MERGED_FREE:
4407 napi_reuse_skb(napi, skb);
4417 /* Upper GRO stack assumes network header starts at gro_offset=0
4418 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4419 * We copy ethernet header into skb->data to have a common layout.
4421 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4423 struct sk_buff *skb = napi->skb;
4424 const struct ethhdr *eth;
4425 unsigned int hlen = sizeof(*eth);
4429 skb_reset_mac_header(skb);
4430 skb_gro_reset_offset(skb);
4432 eth = skb_gro_header_fast(skb, 0);
4433 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4434 eth = skb_gro_header_slow(skb, hlen, 0);
4435 if (unlikely(!eth)) {
4436 napi_reuse_skb(napi, skb);
4440 gro_pull_from_frag0(skb, hlen);
4441 NAPI_GRO_CB(skb)->frag0 += hlen;
4442 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4444 __skb_pull(skb, hlen);
4447 * This works because the only protocols we care about don't require
4449 * We'll fix it up properly in napi_frags_finish()
4451 skb->protocol = eth->h_proto;
4456 gro_result_t napi_gro_frags(struct napi_struct *napi)
4458 struct sk_buff *skb = napi_frags_skb(napi);
4463 trace_napi_gro_frags_entry(skb);
4465 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4467 EXPORT_SYMBOL(napi_gro_frags);
4469 /* Compute the checksum from gro_offset and return the folded value
4470 * after adding in any pseudo checksum.
4472 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4477 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4479 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4480 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4482 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4483 !skb->csum_complete_sw)
4484 netdev_rx_csum_fault(skb->dev);
4487 NAPI_GRO_CB(skb)->csum = wsum;
4488 NAPI_GRO_CB(skb)->csum_valid = 1;
4492 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4495 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4496 * Note: called with local irq disabled, but exits with local irq enabled.
4498 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4501 struct softnet_data *remsd = sd->rps_ipi_list;
4504 sd->rps_ipi_list = NULL;
4508 /* Send pending IPI's to kick RPS processing on remote cpus. */
4510 struct softnet_data *next = remsd->rps_ipi_next;
4512 if (cpu_online(remsd->cpu))
4513 smp_call_function_single_async(remsd->cpu,
4522 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4525 return sd->rps_ipi_list != NULL;
4531 static int process_backlog(struct napi_struct *napi, int quota)
4534 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4536 /* Check if we have pending ipi, its better to send them now,
4537 * not waiting net_rx_action() end.
4539 if (sd_has_rps_ipi_waiting(sd)) {
4540 local_irq_disable();
4541 net_rps_action_and_irq_enable(sd);
4544 napi->weight = weight_p;
4545 local_irq_disable();
4547 struct sk_buff *skb;
4549 while ((skb = __skb_dequeue(&sd->process_queue))) {
4552 __netif_receive_skb(skb);
4554 local_irq_disable();
4555 input_queue_head_incr(sd);
4556 if (++work >= quota) {
4563 if (skb_queue_empty(&sd->input_pkt_queue)) {
4565 * Inline a custom version of __napi_complete().
4566 * only current cpu owns and manipulates this napi,
4567 * and NAPI_STATE_SCHED is the only possible flag set
4569 * We can use a plain write instead of clear_bit(),
4570 * and we dont need an smp_mb() memory barrier.
4578 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4579 &sd->process_queue);
4588 * __napi_schedule - schedule for receive
4589 * @n: entry to schedule
4591 * The entry's receive function will be scheduled to run.
4592 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4594 void __napi_schedule(struct napi_struct *n)
4596 unsigned long flags;
4598 local_irq_save(flags);
4599 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4600 local_irq_restore(flags);
4602 EXPORT_SYMBOL(__napi_schedule);
4605 * __napi_schedule_irqoff - schedule for receive
4606 * @n: entry to schedule
4608 * Variant of __napi_schedule() assuming hard irqs are masked
4610 void __napi_schedule_irqoff(struct napi_struct *n)
4612 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4614 EXPORT_SYMBOL(__napi_schedule_irqoff);
4616 void __napi_complete(struct napi_struct *n)
4618 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4620 list_del_init(&n->poll_list);
4621 smp_mb__before_atomic();
4622 clear_bit(NAPI_STATE_SCHED, &n->state);
4624 EXPORT_SYMBOL(__napi_complete);
4626 void napi_complete_done(struct napi_struct *n, int work_done)
4628 unsigned long flags;
4631 * don't let napi dequeue from the cpu poll list
4632 * just in case its running on a different cpu
4634 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4638 unsigned long timeout = 0;
4641 timeout = n->dev->gro_flush_timeout;
4644 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4645 HRTIMER_MODE_REL_PINNED);
4647 napi_gro_flush(n, false);
4649 if (likely(list_empty(&n->poll_list))) {
4650 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4652 /* If n->poll_list is not empty, we need to mask irqs */
4653 local_irq_save(flags);
4655 local_irq_restore(flags);
4658 EXPORT_SYMBOL(napi_complete_done);
4660 /* must be called under rcu_read_lock(), as we dont take a reference */
4661 struct napi_struct *napi_by_id(unsigned int napi_id)
4663 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4664 struct napi_struct *napi;
4666 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4667 if (napi->napi_id == napi_id)
4672 EXPORT_SYMBOL_GPL(napi_by_id);
4674 void napi_hash_add(struct napi_struct *napi)
4676 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4678 spin_lock(&napi_hash_lock);
4680 /* 0 is not a valid id, we also skip an id that is taken
4681 * we expect both events to be extremely rare
4684 while (!napi->napi_id) {
4685 napi->napi_id = ++napi_gen_id;
4686 if (napi_by_id(napi->napi_id))
4690 hlist_add_head_rcu(&napi->napi_hash_node,
4691 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4693 spin_unlock(&napi_hash_lock);
4696 EXPORT_SYMBOL_GPL(napi_hash_add);
4698 /* Warning : caller is responsible to make sure rcu grace period
4699 * is respected before freeing memory containing @napi
4701 void napi_hash_del(struct napi_struct *napi)
4703 spin_lock(&napi_hash_lock);
4705 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4706 hlist_del_rcu(&napi->napi_hash_node);
4708 spin_unlock(&napi_hash_lock);
4710 EXPORT_SYMBOL_GPL(napi_hash_del);
4712 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4714 struct napi_struct *napi;
4716 napi = container_of(timer, struct napi_struct, timer);
4718 napi_schedule(napi);
4720 return HRTIMER_NORESTART;
4723 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4724 int (*poll)(struct napi_struct *, int), int weight)
4726 INIT_LIST_HEAD(&napi->poll_list);
4727 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4728 napi->timer.function = napi_watchdog;
4729 napi->gro_count = 0;
4730 napi->gro_list = NULL;
4733 if (weight > NAPI_POLL_WEIGHT)
4734 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4736 napi->weight = weight;
4737 list_add(&napi->dev_list, &dev->napi_list);
4739 #ifdef CONFIG_NETPOLL
4740 spin_lock_init(&napi->poll_lock);
4741 napi->poll_owner = -1;
4743 set_bit(NAPI_STATE_SCHED, &napi->state);
4745 EXPORT_SYMBOL(netif_napi_add);
4747 void napi_disable(struct napi_struct *n)
4750 set_bit(NAPI_STATE_DISABLE, &n->state);
4752 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4754 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
4757 hrtimer_cancel(&n->timer);
4759 clear_bit(NAPI_STATE_DISABLE, &n->state);
4761 EXPORT_SYMBOL(napi_disable);
4763 void netif_napi_del(struct napi_struct *napi)
4765 list_del_init(&napi->dev_list);
4766 napi_free_frags(napi);
4768 kfree_skb_list(napi->gro_list);
4769 napi->gro_list = NULL;
4770 napi->gro_count = 0;
4772 EXPORT_SYMBOL(netif_napi_del);
4774 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4779 list_del_init(&n->poll_list);
4781 have = netpoll_poll_lock(n);
4785 /* This NAPI_STATE_SCHED test is for avoiding a race
4786 * with netpoll's poll_napi(). Only the entity which
4787 * obtains the lock and sees NAPI_STATE_SCHED set will
4788 * actually make the ->poll() call. Therefore we avoid
4789 * accidentally calling ->poll() when NAPI is not scheduled.
4792 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4793 work = n->poll(n, weight);
4797 WARN_ON_ONCE(work > weight);
4799 if (likely(work < weight))
4802 /* Drivers must not modify the NAPI state if they
4803 * consume the entire weight. In such cases this code
4804 * still "owns" the NAPI instance and therefore can
4805 * move the instance around on the list at-will.
4807 if (unlikely(napi_disable_pending(n))) {
4813 /* flush too old packets
4814 * If HZ < 1000, flush all packets.
4816 napi_gro_flush(n, HZ >= 1000);
4819 /* Some drivers may have called napi_schedule
4820 * prior to exhausting their budget.
4822 if (unlikely(!list_empty(&n->poll_list))) {
4823 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4824 n->dev ? n->dev->name : "backlog");
4828 list_add_tail(&n->poll_list, repoll);
4831 netpoll_poll_unlock(have);
4836 static void net_rx_action(struct softirq_action *h)
4838 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4839 unsigned long time_limit = jiffies + 2;
4840 int budget = netdev_budget;
4844 local_irq_disable();
4845 list_splice_init(&sd->poll_list, &list);
4849 struct napi_struct *n;
4851 if (list_empty(&list)) {
4852 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4857 n = list_first_entry(&list, struct napi_struct, poll_list);
4858 budget -= napi_poll(n, &repoll);
4860 /* If softirq window is exhausted then punt.
4861 * Allow this to run for 2 jiffies since which will allow
4862 * an average latency of 1.5/HZ.
4864 if (unlikely(budget <= 0 ||
4865 time_after_eq(jiffies, time_limit))) {
4871 local_irq_disable();
4873 list_splice_tail_init(&sd->poll_list, &list);
4874 list_splice_tail(&repoll, &list);
4875 list_splice(&list, &sd->poll_list);
4876 if (!list_empty(&sd->poll_list))
4877 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4879 net_rps_action_and_irq_enable(sd);
4882 struct netdev_adjacent {
4883 struct net_device *dev;
4885 /* upper master flag, there can only be one master device per list */
4888 /* counter for the number of times this device was added to us */
4891 /* private field for the users */
4894 struct list_head list;
4895 struct rcu_head rcu;
4898 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
4899 struct list_head *adj_list)
4901 struct netdev_adjacent *adj;
4903 list_for_each_entry(adj, adj_list, list) {
4904 if (adj->dev == adj_dev)
4911 * netdev_has_upper_dev - Check if device is linked to an upper device
4913 * @upper_dev: upper device to check
4915 * Find out if a device is linked to specified upper device and return true
4916 * in case it is. Note that this checks only immediate upper device,
4917 * not through a complete stack of devices. The caller must hold the RTNL lock.
4919 bool netdev_has_upper_dev(struct net_device *dev,
4920 struct net_device *upper_dev)
4924 return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper);
4926 EXPORT_SYMBOL(netdev_has_upper_dev);
4929 * netdev_has_any_upper_dev - Check if device is linked to some device
4932 * Find out if a device is linked to an upper device and return true in case
4933 * it is. The caller must hold the RTNL lock.
4935 static bool netdev_has_any_upper_dev(struct net_device *dev)
4939 return !list_empty(&dev->all_adj_list.upper);
4943 * netdev_master_upper_dev_get - Get master upper device
4946 * Find a master upper device and return pointer to it or NULL in case
4947 * it's not there. The caller must hold the RTNL lock.
4949 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4951 struct netdev_adjacent *upper;
4955 if (list_empty(&dev->adj_list.upper))
4958 upper = list_first_entry(&dev->adj_list.upper,
4959 struct netdev_adjacent, list);
4960 if (likely(upper->master))
4964 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4966 void *netdev_adjacent_get_private(struct list_head *adj_list)
4968 struct netdev_adjacent *adj;
4970 adj = list_entry(adj_list, struct netdev_adjacent, list);
4972 return adj->private;
4974 EXPORT_SYMBOL(netdev_adjacent_get_private);
4977 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4979 * @iter: list_head ** of the current position
4981 * Gets the next device from the dev's upper list, starting from iter
4982 * position. The caller must hold RCU read lock.
4984 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4985 struct list_head **iter)
4987 struct netdev_adjacent *upper;
4989 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4991 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4993 if (&upper->list == &dev->adj_list.upper)
4996 *iter = &upper->list;
5000 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5003 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
5005 * @iter: list_head ** of the current position
5007 * Gets the next device from the dev's upper list, starting from iter
5008 * position. The caller must hold RCU read lock.
5010 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
5011 struct list_head **iter)
5013 struct netdev_adjacent *upper;
5015 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5017 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5019 if (&upper->list == &dev->all_adj_list.upper)
5022 *iter = &upper->list;
5026 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
5029 * netdev_lower_get_next_private - Get the next ->private from the
5030 * lower neighbour list
5032 * @iter: list_head ** of the current position
5034 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5035 * list, starting from iter position. The caller must hold either hold the
5036 * RTNL lock or its own locking that guarantees that the neighbour lower
5037 * list will remain unchanged.
5039 void *netdev_lower_get_next_private(struct net_device *dev,
5040 struct list_head **iter)
5042 struct netdev_adjacent *lower;
5044 lower = list_entry(*iter, struct netdev_adjacent, list);
5046 if (&lower->list == &dev->adj_list.lower)
5049 *iter = lower->list.next;
5051 return lower->private;
5053 EXPORT_SYMBOL(netdev_lower_get_next_private);
5056 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5057 * lower neighbour list, RCU
5060 * @iter: list_head ** of the current position
5062 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5063 * list, starting from iter position. The caller must hold RCU read lock.
5065 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5066 struct list_head **iter)
5068 struct netdev_adjacent *lower;
5070 WARN_ON_ONCE(!rcu_read_lock_held());
5072 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5074 if (&lower->list == &dev->adj_list.lower)
5077 *iter = &lower->list;
5079 return lower->private;
5081 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5084 * netdev_lower_get_next - Get the next device from the lower neighbour
5087 * @iter: list_head ** of the current position
5089 * Gets the next netdev_adjacent from the dev's lower neighbour
5090 * list, starting from iter position. The caller must hold RTNL lock or
5091 * its own locking that guarantees that the neighbour lower
5092 * list will remain unchanged.
5094 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5096 struct netdev_adjacent *lower;
5098 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5100 if (&lower->list == &dev->adj_list.lower)
5103 *iter = &lower->list;
5107 EXPORT_SYMBOL(netdev_lower_get_next);
5110 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5111 * lower neighbour list, RCU
5115 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5116 * list. The caller must hold RCU read lock.
5118 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5120 struct netdev_adjacent *lower;
5122 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5123 struct netdev_adjacent, list);
5125 return lower->private;
5128 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5131 * netdev_master_upper_dev_get_rcu - Get master upper device
5134 * Find a master upper device and return pointer to it or NULL in case
5135 * it's not there. The caller must hold the RCU read lock.
5137 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5139 struct netdev_adjacent *upper;
5141 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5142 struct netdev_adjacent, list);
5143 if (upper && likely(upper->master))
5147 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5149 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5150 struct net_device *adj_dev,
5151 struct list_head *dev_list)
5153 char linkname[IFNAMSIZ+7];
5154 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5155 "upper_%s" : "lower_%s", adj_dev->name);
5156 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5159 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5161 struct list_head *dev_list)
5163 char linkname[IFNAMSIZ+7];
5164 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5165 "upper_%s" : "lower_%s", name);
5166 sysfs_remove_link(&(dev->dev.kobj), linkname);
5169 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5170 struct net_device *adj_dev,
5171 struct list_head *dev_list)
5173 return (dev_list == &dev->adj_list.upper ||
5174 dev_list == &dev->adj_list.lower) &&
5175 net_eq(dev_net(dev), dev_net(adj_dev));
5178 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5179 struct net_device *adj_dev,
5180 struct list_head *dev_list,
5181 void *private, bool master)
5183 struct netdev_adjacent *adj;
5186 adj = __netdev_find_adj(adj_dev, dev_list);
5193 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5198 adj->master = master;
5200 adj->private = private;
5203 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5204 adj_dev->name, dev->name, adj_dev->name);
5206 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5207 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5212 /* Ensure that master link is always the first item in list. */
5214 ret = sysfs_create_link(&(dev->dev.kobj),
5215 &(adj_dev->dev.kobj), "master");
5217 goto remove_symlinks;
5219 list_add_rcu(&adj->list, dev_list);
5221 list_add_tail_rcu(&adj->list, dev_list);
5227 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5228 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5236 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5237 struct net_device *adj_dev,
5238 struct list_head *dev_list)
5240 struct netdev_adjacent *adj;
5242 adj = __netdev_find_adj(adj_dev, dev_list);
5245 pr_err("tried to remove device %s from %s\n",
5246 dev->name, adj_dev->name);
5250 if (adj->ref_nr > 1) {
5251 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
5258 sysfs_remove_link(&(dev->dev.kobj), "master");
5260 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5261 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5263 list_del_rcu(&adj->list);
5264 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5265 adj_dev->name, dev->name, adj_dev->name);
5267 kfree_rcu(adj, rcu);
5270 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5271 struct net_device *upper_dev,
5272 struct list_head *up_list,
5273 struct list_head *down_list,
5274 void *private, bool master)
5278 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5283 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5286 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5293 static int __netdev_adjacent_dev_link(struct net_device *dev,
5294 struct net_device *upper_dev)
5296 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5297 &dev->all_adj_list.upper,
5298 &upper_dev->all_adj_list.lower,
5302 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5303 struct net_device *upper_dev,
5304 struct list_head *up_list,
5305 struct list_head *down_list)
5307 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5308 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5311 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5312 struct net_device *upper_dev)
5314 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5315 &dev->all_adj_list.upper,
5316 &upper_dev->all_adj_list.lower);
5319 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5320 struct net_device *upper_dev,
5321 void *private, bool master)
5323 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5328 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5329 &dev->adj_list.upper,
5330 &upper_dev->adj_list.lower,
5333 __netdev_adjacent_dev_unlink(dev, upper_dev);
5340 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5341 struct net_device *upper_dev)
5343 __netdev_adjacent_dev_unlink(dev, upper_dev);
5344 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5345 &dev->adj_list.upper,
5346 &upper_dev->adj_list.lower);
5349 static int __netdev_upper_dev_link(struct net_device *dev,
5350 struct net_device *upper_dev, bool master,
5353 struct netdev_notifier_changeupper_info changeupper_info;
5354 struct netdev_adjacent *i, *j, *to_i, *to_j;
5359 if (dev == upper_dev)
5362 /* To prevent loops, check if dev is not upper device to upper_dev. */
5363 if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper))
5366 if (__netdev_find_adj(upper_dev, &dev->adj_list.upper))
5369 if (master && netdev_master_upper_dev_get(dev))
5372 changeupper_info.upper_dev = upper_dev;
5373 changeupper_info.master = master;
5374 changeupper_info.linking = true;
5376 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5377 &changeupper_info.info);
5378 ret = notifier_to_errno(ret);
5382 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5387 /* Now that we linked these devs, make all the upper_dev's
5388 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5389 * versa, and don't forget the devices itself. All of these
5390 * links are non-neighbours.
5392 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5393 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5394 pr_debug("Interlinking %s with %s, non-neighbour\n",
5395 i->dev->name, j->dev->name);
5396 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5402 /* add dev to every upper_dev's upper device */
5403 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5404 pr_debug("linking %s's upper device %s with %s\n",
5405 upper_dev->name, i->dev->name, dev->name);
5406 ret = __netdev_adjacent_dev_link(dev, i->dev);
5408 goto rollback_upper_mesh;
5411 /* add upper_dev to every dev's lower device */
5412 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5413 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5414 i->dev->name, upper_dev->name);
5415 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5417 goto rollback_lower_mesh;
5420 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5421 &changeupper_info.info);
5424 rollback_lower_mesh:
5426 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5429 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5434 rollback_upper_mesh:
5436 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5439 __netdev_adjacent_dev_unlink(dev, i->dev);
5447 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5448 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5449 if (i == to_i && j == to_j)
5451 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5457 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5463 * netdev_upper_dev_link - Add a link to the upper device
5465 * @upper_dev: new upper device
5467 * Adds a link to device which is upper to this one. The caller must hold
5468 * the RTNL lock. On a failure a negative errno code is returned.
5469 * On success the reference counts are adjusted and the function
5472 int netdev_upper_dev_link(struct net_device *dev,
5473 struct net_device *upper_dev)
5475 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5477 EXPORT_SYMBOL(netdev_upper_dev_link);
5480 * netdev_master_upper_dev_link - Add a master link to the upper device
5482 * @upper_dev: new upper device
5484 * Adds a link to device which is upper to this one. In this case, only
5485 * one master upper device can be linked, although other non-master devices
5486 * might be linked as well. The caller must hold the RTNL lock.
5487 * On a failure a negative errno code is returned. On success the reference
5488 * counts are adjusted and the function returns zero.
5490 int netdev_master_upper_dev_link(struct net_device *dev,
5491 struct net_device *upper_dev)
5493 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5495 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5497 int netdev_master_upper_dev_link_private(struct net_device *dev,
5498 struct net_device *upper_dev,
5501 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5503 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5506 * netdev_upper_dev_unlink - Removes a link to upper device
5508 * @upper_dev: new upper device
5510 * Removes a link to device which is upper to this one. The caller must hold
5513 void netdev_upper_dev_unlink(struct net_device *dev,
5514 struct net_device *upper_dev)
5516 struct netdev_notifier_changeupper_info changeupper_info;
5517 struct netdev_adjacent *i, *j;
5520 changeupper_info.upper_dev = upper_dev;
5521 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
5522 changeupper_info.linking = false;
5524 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5525 &changeupper_info.info);
5527 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5529 /* Here is the tricky part. We must remove all dev's lower
5530 * devices from all upper_dev's upper devices and vice
5531 * versa, to maintain the graph relationship.
5533 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5534 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5535 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5537 /* remove also the devices itself from lower/upper device
5540 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5541 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5543 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5544 __netdev_adjacent_dev_unlink(dev, i->dev);
5546 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5547 &changeupper_info.info);
5549 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5552 * netdev_bonding_info_change - Dispatch event about slave change
5554 * @bonding_info: info to dispatch
5556 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5557 * The caller must hold the RTNL lock.
5559 void netdev_bonding_info_change(struct net_device *dev,
5560 struct netdev_bonding_info *bonding_info)
5562 struct netdev_notifier_bonding_info info;
5564 memcpy(&info.bonding_info, bonding_info,
5565 sizeof(struct netdev_bonding_info));
5566 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5569 EXPORT_SYMBOL(netdev_bonding_info_change);
5571 static void netdev_adjacent_add_links(struct net_device *dev)
5573 struct netdev_adjacent *iter;
5575 struct net *net = dev_net(dev);
5577 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5578 if (!net_eq(net,dev_net(iter->dev)))
5580 netdev_adjacent_sysfs_add(iter->dev, dev,
5581 &iter->dev->adj_list.lower);
5582 netdev_adjacent_sysfs_add(dev, iter->dev,
5583 &dev->adj_list.upper);
5586 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5587 if (!net_eq(net,dev_net(iter->dev)))
5589 netdev_adjacent_sysfs_add(iter->dev, dev,
5590 &iter->dev->adj_list.upper);
5591 netdev_adjacent_sysfs_add(dev, iter->dev,
5592 &dev->adj_list.lower);
5596 static void netdev_adjacent_del_links(struct net_device *dev)
5598 struct netdev_adjacent *iter;
5600 struct net *net = dev_net(dev);
5602 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5603 if (!net_eq(net,dev_net(iter->dev)))
5605 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5606 &iter->dev->adj_list.lower);
5607 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5608 &dev->adj_list.upper);
5611 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5612 if (!net_eq(net,dev_net(iter->dev)))
5614 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5615 &iter->dev->adj_list.upper);
5616 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5617 &dev->adj_list.lower);
5621 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5623 struct netdev_adjacent *iter;
5625 struct net *net = dev_net(dev);
5627 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5628 if (!net_eq(net,dev_net(iter->dev)))
5630 netdev_adjacent_sysfs_del(iter->dev, oldname,
5631 &iter->dev->adj_list.lower);
5632 netdev_adjacent_sysfs_add(iter->dev, dev,
5633 &iter->dev->adj_list.lower);
5636 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5637 if (!net_eq(net,dev_net(iter->dev)))
5639 netdev_adjacent_sysfs_del(iter->dev, oldname,
5640 &iter->dev->adj_list.upper);
5641 netdev_adjacent_sysfs_add(iter->dev, dev,
5642 &iter->dev->adj_list.upper);
5646 void *netdev_lower_dev_get_private(struct net_device *dev,
5647 struct net_device *lower_dev)
5649 struct netdev_adjacent *lower;
5653 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
5657 return lower->private;
5659 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5662 int dev_get_nest_level(struct net_device *dev,
5663 bool (*type_check)(struct net_device *dev))
5665 struct net_device *lower = NULL;
5666 struct list_head *iter;
5672 netdev_for_each_lower_dev(dev, lower, iter) {
5673 nest = dev_get_nest_level(lower, type_check);
5674 if (max_nest < nest)
5678 if (type_check(dev))
5683 EXPORT_SYMBOL(dev_get_nest_level);
5685 static void dev_change_rx_flags(struct net_device *dev, int flags)
5687 const struct net_device_ops *ops = dev->netdev_ops;
5689 if (ops->ndo_change_rx_flags)
5690 ops->ndo_change_rx_flags(dev, flags);
5693 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5695 unsigned int old_flags = dev->flags;
5701 dev->flags |= IFF_PROMISC;
5702 dev->promiscuity += inc;
5703 if (dev->promiscuity == 0) {
5706 * If inc causes overflow, untouch promisc and return error.
5709 dev->flags &= ~IFF_PROMISC;
5711 dev->promiscuity -= inc;
5712 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5717 if (dev->flags != old_flags) {
5718 pr_info("device %s %s promiscuous mode\n",
5720 dev->flags & IFF_PROMISC ? "entered" : "left");
5721 if (audit_enabled) {
5722 current_uid_gid(&uid, &gid);
5723 audit_log(current->audit_context, GFP_ATOMIC,
5724 AUDIT_ANOM_PROMISCUOUS,
5725 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5726 dev->name, (dev->flags & IFF_PROMISC),
5727 (old_flags & IFF_PROMISC),
5728 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5729 from_kuid(&init_user_ns, uid),
5730 from_kgid(&init_user_ns, gid),
5731 audit_get_sessionid(current));
5734 dev_change_rx_flags(dev, IFF_PROMISC);
5737 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5742 * dev_set_promiscuity - update promiscuity count on a device
5746 * Add or remove promiscuity from a device. While the count in the device
5747 * remains above zero the interface remains promiscuous. Once it hits zero
5748 * the device reverts back to normal filtering operation. A negative inc
5749 * value is used to drop promiscuity on the device.
5750 * Return 0 if successful or a negative errno code on error.
5752 int dev_set_promiscuity(struct net_device *dev, int inc)
5754 unsigned int old_flags = dev->flags;
5757 err = __dev_set_promiscuity(dev, inc, true);
5760 if (dev->flags != old_flags)
5761 dev_set_rx_mode(dev);
5764 EXPORT_SYMBOL(dev_set_promiscuity);
5766 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5768 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5772 dev->flags |= IFF_ALLMULTI;
5773 dev->allmulti += inc;
5774 if (dev->allmulti == 0) {
5777 * If inc causes overflow, untouch allmulti and return error.
5780 dev->flags &= ~IFF_ALLMULTI;
5782 dev->allmulti -= inc;
5783 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5788 if (dev->flags ^ old_flags) {
5789 dev_change_rx_flags(dev, IFF_ALLMULTI);
5790 dev_set_rx_mode(dev);
5792 __dev_notify_flags(dev, old_flags,
5793 dev->gflags ^ old_gflags);
5799 * dev_set_allmulti - update allmulti count on a device
5803 * Add or remove reception of all multicast frames to a device. While the
5804 * count in the device remains above zero the interface remains listening
5805 * to all interfaces. Once it hits zero the device reverts back to normal
5806 * filtering operation. A negative @inc value is used to drop the counter
5807 * when releasing a resource needing all multicasts.
5808 * Return 0 if successful or a negative errno code on error.
5811 int dev_set_allmulti(struct net_device *dev, int inc)
5813 return __dev_set_allmulti(dev, inc, true);
5815 EXPORT_SYMBOL(dev_set_allmulti);
5818 * Upload unicast and multicast address lists to device and
5819 * configure RX filtering. When the device doesn't support unicast
5820 * filtering it is put in promiscuous mode while unicast addresses
5823 void __dev_set_rx_mode(struct net_device *dev)
5825 const struct net_device_ops *ops = dev->netdev_ops;
5827 /* dev_open will call this function so the list will stay sane. */
5828 if (!(dev->flags&IFF_UP))
5831 if (!netif_device_present(dev))
5834 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5835 /* Unicast addresses changes may only happen under the rtnl,
5836 * therefore calling __dev_set_promiscuity here is safe.
5838 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5839 __dev_set_promiscuity(dev, 1, false);
5840 dev->uc_promisc = true;
5841 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5842 __dev_set_promiscuity(dev, -1, false);
5843 dev->uc_promisc = false;
5847 if (ops->ndo_set_rx_mode)
5848 ops->ndo_set_rx_mode(dev);
5851 void dev_set_rx_mode(struct net_device *dev)
5853 netif_addr_lock_bh(dev);
5854 __dev_set_rx_mode(dev);
5855 netif_addr_unlock_bh(dev);
5859 * dev_get_flags - get flags reported to userspace
5862 * Get the combination of flag bits exported through APIs to userspace.
5864 unsigned int dev_get_flags(const struct net_device *dev)
5868 flags = (dev->flags & ~(IFF_PROMISC |
5873 (dev->gflags & (IFF_PROMISC |
5876 if (netif_running(dev)) {
5877 if (netif_oper_up(dev))
5878 flags |= IFF_RUNNING;
5879 if (netif_carrier_ok(dev))
5880 flags |= IFF_LOWER_UP;
5881 if (netif_dormant(dev))
5882 flags |= IFF_DORMANT;
5887 EXPORT_SYMBOL(dev_get_flags);
5889 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5891 unsigned int old_flags = dev->flags;
5897 * Set the flags on our device.
5900 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5901 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5903 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5907 * Load in the correct multicast list now the flags have changed.
5910 if ((old_flags ^ flags) & IFF_MULTICAST)
5911 dev_change_rx_flags(dev, IFF_MULTICAST);
5913 dev_set_rx_mode(dev);
5916 * Have we downed the interface. We handle IFF_UP ourselves
5917 * according to user attempts to set it, rather than blindly
5922 if ((old_flags ^ flags) & IFF_UP)
5923 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5925 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5926 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5927 unsigned int old_flags = dev->flags;
5929 dev->gflags ^= IFF_PROMISC;
5931 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5932 if (dev->flags != old_flags)
5933 dev_set_rx_mode(dev);
5936 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5937 is important. Some (broken) drivers set IFF_PROMISC, when
5938 IFF_ALLMULTI is requested not asking us and not reporting.
5940 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5941 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5943 dev->gflags ^= IFF_ALLMULTI;
5944 __dev_set_allmulti(dev, inc, false);
5950 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5951 unsigned int gchanges)
5953 unsigned int changes = dev->flags ^ old_flags;
5956 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5958 if (changes & IFF_UP) {
5959 if (dev->flags & IFF_UP)
5960 call_netdevice_notifiers(NETDEV_UP, dev);
5962 call_netdevice_notifiers(NETDEV_DOWN, dev);
5965 if (dev->flags & IFF_UP &&
5966 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5967 struct netdev_notifier_change_info change_info;
5969 change_info.flags_changed = changes;
5970 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5976 * dev_change_flags - change device settings
5978 * @flags: device state flags
5980 * Change settings on device based state flags. The flags are
5981 * in the userspace exported format.
5983 int dev_change_flags(struct net_device *dev, unsigned int flags)
5986 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5988 ret = __dev_change_flags(dev, flags);
5992 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5993 __dev_notify_flags(dev, old_flags, changes);
5996 EXPORT_SYMBOL(dev_change_flags);
5998 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6000 const struct net_device_ops *ops = dev->netdev_ops;
6002 if (ops->ndo_change_mtu)
6003 return ops->ndo_change_mtu(dev, new_mtu);
6010 * dev_set_mtu - Change maximum transfer unit
6012 * @new_mtu: new transfer unit
6014 * Change the maximum transfer size of the network device.
6016 int dev_set_mtu(struct net_device *dev, int new_mtu)
6020 if (new_mtu == dev->mtu)
6023 /* MTU must be positive. */
6027 if (!netif_device_present(dev))
6030 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6031 err = notifier_to_errno(err);
6035 orig_mtu = dev->mtu;
6036 err = __dev_set_mtu(dev, new_mtu);
6039 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6040 err = notifier_to_errno(err);
6042 /* setting mtu back and notifying everyone again,
6043 * so that they have a chance to revert changes.
6045 __dev_set_mtu(dev, orig_mtu);
6046 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6051 EXPORT_SYMBOL(dev_set_mtu);
6054 * dev_set_group - Change group this device belongs to
6056 * @new_group: group this device should belong to
6058 void dev_set_group(struct net_device *dev, int new_group)
6060 dev->group = new_group;
6062 EXPORT_SYMBOL(dev_set_group);
6065 * dev_set_mac_address - Change Media Access Control Address
6069 * Change the hardware (MAC) address of the device
6071 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6073 const struct net_device_ops *ops = dev->netdev_ops;
6076 if (!ops->ndo_set_mac_address)
6078 if (sa->sa_family != dev->type)
6080 if (!netif_device_present(dev))
6082 err = ops->ndo_set_mac_address(dev, sa);
6085 dev->addr_assign_type = NET_ADDR_SET;
6086 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6087 add_device_randomness(dev->dev_addr, dev->addr_len);
6090 EXPORT_SYMBOL(dev_set_mac_address);
6093 * dev_change_carrier - Change device carrier
6095 * @new_carrier: new value
6097 * Change device carrier
6099 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6101 const struct net_device_ops *ops = dev->netdev_ops;
6103 if (!ops->ndo_change_carrier)
6105 if (!netif_device_present(dev))
6107 return ops->ndo_change_carrier(dev, new_carrier);
6109 EXPORT_SYMBOL(dev_change_carrier);
6112 * dev_get_phys_port_id - Get device physical port ID
6116 * Get device physical port ID
6118 int dev_get_phys_port_id(struct net_device *dev,
6119 struct netdev_phys_item_id *ppid)
6121 const struct net_device_ops *ops = dev->netdev_ops;
6123 if (!ops->ndo_get_phys_port_id)
6125 return ops->ndo_get_phys_port_id(dev, ppid);
6127 EXPORT_SYMBOL(dev_get_phys_port_id);
6130 * dev_get_phys_port_name - Get device physical port name
6134 * Get device physical port name
6136 int dev_get_phys_port_name(struct net_device *dev,
6137 char *name, size_t len)
6139 const struct net_device_ops *ops = dev->netdev_ops;
6141 if (!ops->ndo_get_phys_port_name)
6143 return ops->ndo_get_phys_port_name(dev, name, len);
6145 EXPORT_SYMBOL(dev_get_phys_port_name);
6148 * dev_change_proto_down - update protocol port state information
6150 * @proto_down: new value
6152 * This info can be used by switch drivers to set the phys state of the
6155 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6157 const struct net_device_ops *ops = dev->netdev_ops;
6159 if (!ops->ndo_change_proto_down)
6161 if (!netif_device_present(dev))
6163 return ops->ndo_change_proto_down(dev, proto_down);
6165 EXPORT_SYMBOL(dev_change_proto_down);
6168 * dev_new_index - allocate an ifindex
6169 * @net: the applicable net namespace
6171 * Returns a suitable unique value for a new device interface
6172 * number. The caller must hold the rtnl semaphore or the
6173 * dev_base_lock to be sure it remains unique.
6175 static int dev_new_index(struct net *net)
6177 int ifindex = net->ifindex;
6181 if (!__dev_get_by_index(net, ifindex))
6182 return net->ifindex = ifindex;
6186 /* Delayed registration/unregisteration */
6187 static LIST_HEAD(net_todo_list);
6188 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6190 static void net_set_todo(struct net_device *dev)
6192 list_add_tail(&dev->todo_list, &net_todo_list);
6193 dev_net(dev)->dev_unreg_count++;
6196 static void rollback_registered_many(struct list_head *head)
6198 struct net_device *dev, *tmp;
6199 LIST_HEAD(close_head);
6201 BUG_ON(dev_boot_phase);
6204 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6205 /* Some devices call without registering
6206 * for initialization unwind. Remove those
6207 * devices and proceed with the remaining.
6209 if (dev->reg_state == NETREG_UNINITIALIZED) {
6210 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6214 list_del(&dev->unreg_list);
6217 dev->dismantle = true;
6218 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6221 /* If device is running, close it first. */
6222 list_for_each_entry(dev, head, unreg_list)
6223 list_add_tail(&dev->close_list, &close_head);
6224 dev_close_many(&close_head, true);
6226 list_for_each_entry(dev, head, unreg_list) {
6227 /* And unlink it from device chain. */
6228 unlist_netdevice(dev);
6230 dev->reg_state = NETREG_UNREGISTERING;
6231 on_each_cpu(flush_backlog, dev, 1);
6236 list_for_each_entry(dev, head, unreg_list) {
6237 struct sk_buff *skb = NULL;
6239 /* Shutdown queueing discipline. */
6243 /* Notify protocols, that we are about to destroy
6244 this device. They should clean all the things.
6246 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6248 if (!dev->rtnl_link_ops ||
6249 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6250 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6254 * Flush the unicast and multicast chains
6259 if (dev->netdev_ops->ndo_uninit)
6260 dev->netdev_ops->ndo_uninit(dev);
6263 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6265 /* Notifier chain MUST detach us all upper devices. */
6266 WARN_ON(netdev_has_any_upper_dev(dev));
6268 /* Remove entries from kobject tree */
6269 netdev_unregister_kobject(dev);
6271 /* Remove XPS queueing entries */
6272 netif_reset_xps_queues_gt(dev, 0);
6278 list_for_each_entry(dev, head, unreg_list)
6282 static void rollback_registered(struct net_device *dev)
6286 list_add(&dev->unreg_list, &single);
6287 rollback_registered_many(&single);
6291 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6292 struct net_device *upper, netdev_features_t features)
6294 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6295 netdev_features_t feature;
6298 for_each_netdev_feature(&upper_disables, feature_bit) {
6299 feature = __NETIF_F_BIT(feature_bit);
6300 if (!(upper->wanted_features & feature)
6301 && (features & feature)) {
6302 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6303 &feature, upper->name);
6304 features &= ~feature;
6311 static void netdev_sync_lower_features(struct net_device *upper,
6312 struct net_device *lower, netdev_features_t features)
6314 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6315 netdev_features_t feature;
6318 for_each_netdev_feature(&upper_disables, feature_bit) {
6319 feature = __NETIF_F_BIT(feature_bit);
6320 if (!(features & feature) && (lower->features & feature)) {
6321 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6322 &feature, lower->name);
6323 lower->wanted_features &= ~feature;
6324 netdev_update_features(lower);
6326 if (unlikely(lower->features & feature))
6327 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6328 &feature, lower->name);
6333 static netdev_features_t netdev_fix_features(struct net_device *dev,
6334 netdev_features_t features)
6336 /* Fix illegal checksum combinations */
6337 if ((features & NETIF_F_HW_CSUM) &&
6338 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6339 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6340 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6343 /* TSO requires that SG is present as well. */
6344 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6345 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6346 features &= ~NETIF_F_ALL_TSO;
6349 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6350 !(features & NETIF_F_IP_CSUM)) {
6351 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6352 features &= ~NETIF_F_TSO;
6353 features &= ~NETIF_F_TSO_ECN;
6356 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6357 !(features & NETIF_F_IPV6_CSUM)) {
6358 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6359 features &= ~NETIF_F_TSO6;
6362 /* TSO ECN requires that TSO is present as well. */
6363 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6364 features &= ~NETIF_F_TSO_ECN;
6366 /* Software GSO depends on SG. */
6367 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6368 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6369 features &= ~NETIF_F_GSO;
6372 /* UFO needs SG and checksumming */
6373 if (features & NETIF_F_UFO) {
6374 /* maybe split UFO into V4 and V6? */
6375 if (!((features & NETIF_F_GEN_CSUM) ||
6376 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6377 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6379 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6380 features &= ~NETIF_F_UFO;
6383 if (!(features & NETIF_F_SG)) {
6385 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6386 features &= ~NETIF_F_UFO;
6390 #ifdef CONFIG_NET_RX_BUSY_POLL
6391 if (dev->netdev_ops->ndo_busy_poll)
6392 features |= NETIF_F_BUSY_POLL;
6395 features &= ~NETIF_F_BUSY_POLL;
6400 int __netdev_update_features(struct net_device *dev)
6402 struct net_device *upper, *lower;
6403 netdev_features_t features;
6404 struct list_head *iter;
6409 features = netdev_get_wanted_features(dev);
6411 if (dev->netdev_ops->ndo_fix_features)
6412 features = dev->netdev_ops->ndo_fix_features(dev, features);
6414 /* driver might be less strict about feature dependencies */
6415 features = netdev_fix_features(dev, features);
6417 /* some features can't be enabled if they're off an an upper device */
6418 netdev_for_each_upper_dev_rcu(dev, upper, iter)
6419 features = netdev_sync_upper_features(dev, upper, features);
6421 if (dev->features == features)
6424 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6425 &dev->features, &features);
6427 if (dev->netdev_ops->ndo_set_features)
6428 err = dev->netdev_ops->ndo_set_features(dev, features);
6432 if (unlikely(err < 0)) {
6434 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6435 err, &features, &dev->features);
6440 /* some features must be disabled on lower devices when disabled
6441 * on an upper device (think: bonding master or bridge)
6443 netdev_for_each_lower_dev(dev, lower, iter)
6444 netdev_sync_lower_features(dev, lower, features);
6447 dev->features = features;
6449 return err < 0 ? 0 : 1;
6453 * netdev_update_features - recalculate device features
6454 * @dev: the device to check
6456 * Recalculate dev->features set and send notifications if it
6457 * has changed. Should be called after driver or hardware dependent
6458 * conditions might have changed that influence the features.
6460 void netdev_update_features(struct net_device *dev)
6462 if (__netdev_update_features(dev))
6463 netdev_features_change(dev);
6465 EXPORT_SYMBOL(netdev_update_features);
6468 * netdev_change_features - recalculate device features
6469 * @dev: the device to check
6471 * Recalculate dev->features set and send notifications even
6472 * if they have not changed. Should be called instead of
6473 * netdev_update_features() if also dev->vlan_features might
6474 * have changed to allow the changes to be propagated to stacked
6477 void netdev_change_features(struct net_device *dev)
6479 __netdev_update_features(dev);
6480 netdev_features_change(dev);
6482 EXPORT_SYMBOL(netdev_change_features);
6485 * netif_stacked_transfer_operstate - transfer operstate
6486 * @rootdev: the root or lower level device to transfer state from
6487 * @dev: the device to transfer operstate to
6489 * Transfer operational state from root to device. This is normally
6490 * called when a stacking relationship exists between the root
6491 * device and the device(a leaf device).
6493 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6494 struct net_device *dev)
6496 if (rootdev->operstate == IF_OPER_DORMANT)
6497 netif_dormant_on(dev);
6499 netif_dormant_off(dev);
6501 if (netif_carrier_ok(rootdev)) {
6502 if (!netif_carrier_ok(dev))
6503 netif_carrier_on(dev);
6505 if (netif_carrier_ok(dev))
6506 netif_carrier_off(dev);
6509 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6512 static int netif_alloc_rx_queues(struct net_device *dev)
6514 unsigned int i, count = dev->num_rx_queues;
6515 struct netdev_rx_queue *rx;
6516 size_t sz = count * sizeof(*rx);
6520 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6528 for (i = 0; i < count; i++)
6534 static void netdev_init_one_queue(struct net_device *dev,
6535 struct netdev_queue *queue, void *_unused)
6537 /* Initialize queue lock */
6538 spin_lock_init(&queue->_xmit_lock);
6539 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6540 queue->xmit_lock_owner = -1;
6541 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6544 dql_init(&queue->dql, HZ);
6548 static void netif_free_tx_queues(struct net_device *dev)
6553 static int netif_alloc_netdev_queues(struct net_device *dev)
6555 unsigned int count = dev->num_tx_queues;
6556 struct netdev_queue *tx;
6557 size_t sz = count * sizeof(*tx);
6559 if (count < 1 || count > 0xffff)
6562 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6570 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6571 spin_lock_init(&dev->tx_global_lock);
6576 void netif_tx_stop_all_queues(struct net_device *dev)
6580 for (i = 0; i < dev->num_tx_queues; i++) {
6581 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
6582 netif_tx_stop_queue(txq);
6585 EXPORT_SYMBOL(netif_tx_stop_all_queues);
6588 * register_netdevice - register a network device
6589 * @dev: device to register
6591 * Take a completed network device structure and add it to the kernel
6592 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6593 * chain. 0 is returned on success. A negative errno code is returned
6594 * on a failure to set up the device, or if the name is a duplicate.
6596 * Callers must hold the rtnl semaphore. You may want
6597 * register_netdev() instead of this.
6600 * The locking appears insufficient to guarantee two parallel registers
6601 * will not get the same name.
6604 int register_netdevice(struct net_device *dev)
6607 struct net *net = dev_net(dev);
6609 BUG_ON(dev_boot_phase);
6614 /* When net_device's are persistent, this will be fatal. */
6615 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6618 spin_lock_init(&dev->addr_list_lock);
6619 netdev_set_addr_lockdep_class(dev);
6621 ret = dev_get_valid_name(net, dev, dev->name);
6625 /* Init, if this function is available */
6626 if (dev->netdev_ops->ndo_init) {
6627 ret = dev->netdev_ops->ndo_init(dev);
6635 if (((dev->hw_features | dev->features) &
6636 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6637 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6638 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6639 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6646 dev->ifindex = dev_new_index(net);
6647 else if (__dev_get_by_index(net, dev->ifindex))
6650 /* Transfer changeable features to wanted_features and enable
6651 * software offloads (GSO and GRO).
6653 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6654 dev->features |= NETIF_F_SOFT_FEATURES;
6655 dev->wanted_features = dev->features & dev->hw_features;
6657 if (!(dev->flags & IFF_LOOPBACK)) {
6658 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6661 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6663 dev->vlan_features |= NETIF_F_HIGHDMA;
6665 /* Make NETIF_F_SG inheritable to tunnel devices.
6667 dev->hw_enc_features |= NETIF_F_SG;
6669 /* Make NETIF_F_SG inheritable to MPLS.
6671 dev->mpls_features |= NETIF_F_SG;
6673 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6674 ret = notifier_to_errno(ret);
6678 ret = netdev_register_kobject(dev);
6681 dev->reg_state = NETREG_REGISTERED;
6683 __netdev_update_features(dev);
6686 * Default initial state at registry is that the
6687 * device is present.
6690 set_bit(__LINK_STATE_PRESENT, &dev->state);
6692 linkwatch_init_dev(dev);
6694 dev_init_scheduler(dev);
6696 list_netdevice(dev);
6697 add_device_randomness(dev->dev_addr, dev->addr_len);
6699 /* If the device has permanent device address, driver should
6700 * set dev_addr and also addr_assign_type should be set to
6701 * NET_ADDR_PERM (default value).
6703 if (dev->addr_assign_type == NET_ADDR_PERM)
6704 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6706 /* Notify protocols, that a new device appeared. */
6707 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6708 ret = notifier_to_errno(ret);
6710 rollback_registered(dev);
6711 dev->reg_state = NETREG_UNREGISTERED;
6714 * Prevent userspace races by waiting until the network
6715 * device is fully setup before sending notifications.
6717 if (!dev->rtnl_link_ops ||
6718 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6719 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6725 if (dev->netdev_ops->ndo_uninit)
6726 dev->netdev_ops->ndo_uninit(dev);
6729 EXPORT_SYMBOL(register_netdevice);
6732 * init_dummy_netdev - init a dummy network device for NAPI
6733 * @dev: device to init
6735 * This takes a network device structure and initialize the minimum
6736 * amount of fields so it can be used to schedule NAPI polls without
6737 * registering a full blown interface. This is to be used by drivers
6738 * that need to tie several hardware interfaces to a single NAPI
6739 * poll scheduler due to HW limitations.
6741 int init_dummy_netdev(struct net_device *dev)
6743 /* Clear everything. Note we don't initialize spinlocks
6744 * are they aren't supposed to be taken by any of the
6745 * NAPI code and this dummy netdev is supposed to be
6746 * only ever used for NAPI polls
6748 memset(dev, 0, sizeof(struct net_device));
6750 /* make sure we BUG if trying to hit standard
6751 * register/unregister code path
6753 dev->reg_state = NETREG_DUMMY;
6755 /* NAPI wants this */
6756 INIT_LIST_HEAD(&dev->napi_list);
6758 /* a dummy interface is started by default */
6759 set_bit(__LINK_STATE_PRESENT, &dev->state);
6760 set_bit(__LINK_STATE_START, &dev->state);
6762 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6763 * because users of this 'device' dont need to change
6769 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6773 * register_netdev - register a network device
6774 * @dev: device to register
6776 * Take a completed network device structure and add it to the kernel
6777 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6778 * chain. 0 is returned on success. A negative errno code is returned
6779 * on a failure to set up the device, or if the name is a duplicate.
6781 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6782 * and expands the device name if you passed a format string to
6785 int register_netdev(struct net_device *dev)
6790 err = register_netdevice(dev);
6794 EXPORT_SYMBOL(register_netdev);
6796 int netdev_refcnt_read(const struct net_device *dev)
6800 for_each_possible_cpu(i)
6801 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6804 EXPORT_SYMBOL(netdev_refcnt_read);
6807 * netdev_wait_allrefs - wait until all references are gone.
6808 * @dev: target net_device
6810 * This is called when unregistering network devices.
6812 * Any protocol or device that holds a reference should register
6813 * for netdevice notification, and cleanup and put back the
6814 * reference if they receive an UNREGISTER event.
6815 * We can get stuck here if buggy protocols don't correctly
6818 static void netdev_wait_allrefs(struct net_device *dev)
6820 unsigned long rebroadcast_time, warning_time;
6823 linkwatch_forget_dev(dev);
6825 rebroadcast_time = warning_time = jiffies;
6826 refcnt = netdev_refcnt_read(dev);
6828 while (refcnt != 0) {
6829 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6832 /* Rebroadcast unregister notification */
6833 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6839 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6840 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6842 /* We must not have linkwatch events
6843 * pending on unregister. If this
6844 * happens, we simply run the queue
6845 * unscheduled, resulting in a noop
6848 linkwatch_run_queue();
6853 rebroadcast_time = jiffies;
6858 refcnt = netdev_refcnt_read(dev);
6860 if (time_after(jiffies, warning_time + 10 * HZ)) {
6861 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6863 warning_time = jiffies;
6872 * register_netdevice(x1);
6873 * register_netdevice(x2);
6875 * unregister_netdevice(y1);
6876 * unregister_netdevice(y2);
6882 * We are invoked by rtnl_unlock().
6883 * This allows us to deal with problems:
6884 * 1) We can delete sysfs objects which invoke hotplug
6885 * without deadlocking with linkwatch via keventd.
6886 * 2) Since we run with the RTNL semaphore not held, we can sleep
6887 * safely in order to wait for the netdev refcnt to drop to zero.
6889 * We must not return until all unregister events added during
6890 * the interval the lock was held have been completed.
6892 void netdev_run_todo(void)
6894 struct list_head list;
6896 /* Snapshot list, allow later requests */
6897 list_replace_init(&net_todo_list, &list);
6902 /* Wait for rcu callbacks to finish before next phase */
6903 if (!list_empty(&list))
6906 while (!list_empty(&list)) {
6907 struct net_device *dev
6908 = list_first_entry(&list, struct net_device, todo_list);
6909 list_del(&dev->todo_list);
6912 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6915 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6916 pr_err("network todo '%s' but state %d\n",
6917 dev->name, dev->reg_state);
6922 dev->reg_state = NETREG_UNREGISTERED;
6924 netdev_wait_allrefs(dev);
6927 BUG_ON(netdev_refcnt_read(dev));
6928 BUG_ON(!list_empty(&dev->ptype_all));
6929 BUG_ON(!list_empty(&dev->ptype_specific));
6930 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6931 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6932 WARN_ON(dev->dn_ptr);
6934 if (dev->destructor)
6935 dev->destructor(dev);
6937 /* Report a network device has been unregistered */
6939 dev_net(dev)->dev_unreg_count--;
6941 wake_up(&netdev_unregistering_wq);
6943 /* Free network device */
6944 kobject_put(&dev->dev.kobj);
6948 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6949 * fields in the same order, with only the type differing.
6951 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6952 const struct net_device_stats *netdev_stats)
6954 #if BITS_PER_LONG == 64
6955 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6956 memcpy(stats64, netdev_stats, sizeof(*stats64));
6958 size_t i, n = sizeof(*stats64) / sizeof(u64);
6959 const unsigned long *src = (const unsigned long *)netdev_stats;
6960 u64 *dst = (u64 *)stats64;
6962 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6963 sizeof(*stats64) / sizeof(u64));
6964 for (i = 0; i < n; i++)
6968 EXPORT_SYMBOL(netdev_stats_to_stats64);
6971 * dev_get_stats - get network device statistics
6972 * @dev: device to get statistics from
6973 * @storage: place to store stats
6975 * Get network statistics from device. Return @storage.
6976 * The device driver may provide its own method by setting
6977 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6978 * otherwise the internal statistics structure is used.
6980 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6981 struct rtnl_link_stats64 *storage)
6983 const struct net_device_ops *ops = dev->netdev_ops;
6985 if (ops->ndo_get_stats64) {
6986 memset(storage, 0, sizeof(*storage));
6987 ops->ndo_get_stats64(dev, storage);
6988 } else if (ops->ndo_get_stats) {
6989 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6991 netdev_stats_to_stats64(storage, &dev->stats);
6993 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6994 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6997 EXPORT_SYMBOL(dev_get_stats);
6999 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7001 struct netdev_queue *queue = dev_ingress_queue(dev);
7003 #ifdef CONFIG_NET_CLS_ACT
7006 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7009 netdev_init_one_queue(dev, queue, NULL);
7010 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7011 queue->qdisc_sleeping = &noop_qdisc;
7012 rcu_assign_pointer(dev->ingress_queue, queue);
7017 static const struct ethtool_ops default_ethtool_ops;
7019 void netdev_set_default_ethtool_ops(struct net_device *dev,
7020 const struct ethtool_ops *ops)
7022 if (dev->ethtool_ops == &default_ethtool_ops)
7023 dev->ethtool_ops = ops;
7025 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7027 void netdev_freemem(struct net_device *dev)
7029 char *addr = (char *)dev - dev->padded;
7035 * alloc_netdev_mqs - allocate network device
7036 * @sizeof_priv: size of private data to allocate space for
7037 * @name: device name format string
7038 * @name_assign_type: origin of device name
7039 * @setup: callback to initialize device
7040 * @txqs: the number of TX subqueues to allocate
7041 * @rxqs: the number of RX subqueues to allocate
7043 * Allocates a struct net_device with private data area for driver use
7044 * and performs basic initialization. Also allocates subqueue structs
7045 * for each queue on the device.
7047 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7048 unsigned char name_assign_type,
7049 void (*setup)(struct net_device *),
7050 unsigned int txqs, unsigned int rxqs)
7052 struct net_device *dev;
7054 struct net_device *p;
7056 BUG_ON(strlen(name) >= sizeof(dev->name));
7059 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7065 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7070 alloc_size = sizeof(struct net_device);
7072 /* ensure 32-byte alignment of private area */
7073 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7074 alloc_size += sizeof_priv;
7076 /* ensure 32-byte alignment of whole construct */
7077 alloc_size += NETDEV_ALIGN - 1;
7079 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7081 p = vzalloc(alloc_size);
7085 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7086 dev->padded = (char *)dev - (char *)p;
7088 dev->pcpu_refcnt = alloc_percpu(int);
7089 if (!dev->pcpu_refcnt)
7092 if (dev_addr_init(dev))
7098 dev_net_set(dev, &init_net);
7100 dev->gso_max_size = GSO_MAX_SIZE;
7101 dev->gso_max_segs = GSO_MAX_SEGS;
7102 dev->gso_min_segs = 0;
7104 INIT_LIST_HEAD(&dev->napi_list);
7105 INIT_LIST_HEAD(&dev->unreg_list);
7106 INIT_LIST_HEAD(&dev->close_list);
7107 INIT_LIST_HEAD(&dev->link_watch_list);
7108 INIT_LIST_HEAD(&dev->adj_list.upper);
7109 INIT_LIST_HEAD(&dev->adj_list.lower);
7110 INIT_LIST_HEAD(&dev->all_adj_list.upper);
7111 INIT_LIST_HEAD(&dev->all_adj_list.lower);
7112 INIT_LIST_HEAD(&dev->ptype_all);
7113 INIT_LIST_HEAD(&dev->ptype_specific);
7114 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7117 if (!dev->tx_queue_len)
7118 dev->priv_flags |= IFF_NO_QUEUE;
7120 dev->num_tx_queues = txqs;
7121 dev->real_num_tx_queues = txqs;
7122 if (netif_alloc_netdev_queues(dev))
7126 dev->num_rx_queues = rxqs;
7127 dev->real_num_rx_queues = rxqs;
7128 if (netif_alloc_rx_queues(dev))
7132 strcpy(dev->name, name);
7133 dev->name_assign_type = name_assign_type;
7134 dev->group = INIT_NETDEV_GROUP;
7135 if (!dev->ethtool_ops)
7136 dev->ethtool_ops = &default_ethtool_ops;
7138 nf_hook_ingress_init(dev);
7147 free_percpu(dev->pcpu_refcnt);
7149 netdev_freemem(dev);
7152 EXPORT_SYMBOL(alloc_netdev_mqs);
7155 * free_netdev - free network device
7158 * This function does the last stage of destroying an allocated device
7159 * interface. The reference to the device object is released.
7160 * If this is the last reference then it will be freed.
7162 void free_netdev(struct net_device *dev)
7164 struct napi_struct *p, *n;
7166 netif_free_tx_queues(dev);
7171 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7173 /* Flush device addresses */
7174 dev_addr_flush(dev);
7176 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7179 free_percpu(dev->pcpu_refcnt);
7180 dev->pcpu_refcnt = NULL;
7182 /* Compatibility with error handling in drivers */
7183 if (dev->reg_state == NETREG_UNINITIALIZED) {
7184 netdev_freemem(dev);
7188 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7189 dev->reg_state = NETREG_RELEASED;
7191 /* will free via device release */
7192 put_device(&dev->dev);
7194 EXPORT_SYMBOL(free_netdev);
7197 * synchronize_net - Synchronize with packet receive processing
7199 * Wait for packets currently being received to be done.
7200 * Does not block later packets from starting.
7202 void synchronize_net(void)
7205 if (rtnl_is_locked())
7206 synchronize_rcu_expedited();
7210 EXPORT_SYMBOL(synchronize_net);
7213 * unregister_netdevice_queue - remove device from the kernel
7217 * This function shuts down a device interface and removes it
7218 * from the kernel tables.
7219 * If head not NULL, device is queued to be unregistered later.
7221 * Callers must hold the rtnl semaphore. You may want
7222 * unregister_netdev() instead of this.
7225 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7230 list_move_tail(&dev->unreg_list, head);
7232 rollback_registered(dev);
7233 /* Finish processing unregister after unlock */
7237 EXPORT_SYMBOL(unregister_netdevice_queue);
7240 * unregister_netdevice_many - unregister many devices
7241 * @head: list of devices
7243 * Note: As most callers use a stack allocated list_head,
7244 * we force a list_del() to make sure stack wont be corrupted later.
7246 void unregister_netdevice_many(struct list_head *head)
7248 struct net_device *dev;
7250 if (!list_empty(head)) {
7251 rollback_registered_many(head);
7252 list_for_each_entry(dev, head, unreg_list)
7257 EXPORT_SYMBOL(unregister_netdevice_many);
7260 * unregister_netdev - remove device from the kernel
7263 * This function shuts down a device interface and removes it
7264 * from the kernel tables.
7266 * This is just a wrapper for unregister_netdevice that takes
7267 * the rtnl semaphore. In general you want to use this and not
7268 * unregister_netdevice.
7270 void unregister_netdev(struct net_device *dev)
7273 unregister_netdevice(dev);
7276 EXPORT_SYMBOL(unregister_netdev);
7279 * dev_change_net_namespace - move device to different nethost namespace
7281 * @net: network namespace
7282 * @pat: If not NULL name pattern to try if the current device name
7283 * is already taken in the destination network namespace.
7285 * This function shuts down a device interface and moves it
7286 * to a new network namespace. On success 0 is returned, on
7287 * a failure a netagive errno code is returned.
7289 * Callers must hold the rtnl semaphore.
7292 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7298 /* Don't allow namespace local devices to be moved. */
7300 if (dev->features & NETIF_F_NETNS_LOCAL)
7303 /* Ensure the device has been registrered */
7304 if (dev->reg_state != NETREG_REGISTERED)
7307 /* Get out if there is nothing todo */
7309 if (net_eq(dev_net(dev), net))
7312 /* Pick the destination device name, and ensure
7313 * we can use it in the destination network namespace.
7316 if (__dev_get_by_name(net, dev->name)) {
7317 /* We get here if we can't use the current device name */
7320 if (dev_get_valid_name(net, dev, pat) < 0)
7325 * And now a mini version of register_netdevice unregister_netdevice.
7328 /* If device is running close it first. */
7331 /* And unlink it from device chain */
7333 unlist_netdevice(dev);
7337 /* Shutdown queueing discipline. */
7340 /* Notify protocols, that we are about to destroy
7341 this device. They should clean all the things.
7343 Note that dev->reg_state stays at NETREG_REGISTERED.
7344 This is wanted because this way 8021q and macvlan know
7345 the device is just moving and can keep their slaves up.
7347 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7349 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7350 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7353 * Flush the unicast and multicast chains
7358 /* Send a netdev-removed uevent to the old namespace */
7359 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7360 netdev_adjacent_del_links(dev);
7362 /* Actually switch the network namespace */
7363 dev_net_set(dev, net);
7365 /* If there is an ifindex conflict assign a new one */
7366 if (__dev_get_by_index(net, dev->ifindex))
7367 dev->ifindex = dev_new_index(net);
7369 /* Send a netdev-add uevent to the new namespace */
7370 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7371 netdev_adjacent_add_links(dev);
7373 /* Fixup kobjects */
7374 err = device_rename(&dev->dev, dev->name);
7377 /* Add the device back in the hashes */
7378 list_netdevice(dev);
7380 /* Notify protocols, that a new device appeared. */
7381 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7384 * Prevent userspace races by waiting until the network
7385 * device is fully setup before sending notifications.
7387 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7394 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7396 static int dev_cpu_callback(struct notifier_block *nfb,
7397 unsigned long action,
7400 struct sk_buff **list_skb;
7401 struct sk_buff *skb;
7402 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7403 struct softnet_data *sd, *oldsd;
7405 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7408 local_irq_disable();
7409 cpu = smp_processor_id();
7410 sd = &per_cpu(softnet_data, cpu);
7411 oldsd = &per_cpu(softnet_data, oldcpu);
7413 /* Find end of our completion_queue. */
7414 list_skb = &sd->completion_queue;
7416 list_skb = &(*list_skb)->next;
7417 /* Append completion queue from offline CPU. */
7418 *list_skb = oldsd->completion_queue;
7419 oldsd->completion_queue = NULL;
7421 /* Append output queue from offline CPU. */
7422 if (oldsd->output_queue) {
7423 *sd->output_queue_tailp = oldsd->output_queue;
7424 sd->output_queue_tailp = oldsd->output_queue_tailp;
7425 oldsd->output_queue = NULL;
7426 oldsd->output_queue_tailp = &oldsd->output_queue;
7428 /* Append NAPI poll list from offline CPU, with one exception :
7429 * process_backlog() must be called by cpu owning percpu backlog.
7430 * We properly handle process_queue & input_pkt_queue later.
7432 while (!list_empty(&oldsd->poll_list)) {
7433 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7437 list_del_init(&napi->poll_list);
7438 if (napi->poll == process_backlog)
7441 ____napi_schedule(sd, napi);
7444 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7447 /* Process offline CPU's input_pkt_queue */
7448 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7450 input_queue_head_incr(oldsd);
7452 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7454 input_queue_head_incr(oldsd);
7462 * netdev_increment_features - increment feature set by one
7463 * @all: current feature set
7464 * @one: new feature set
7465 * @mask: mask feature set
7467 * Computes a new feature set after adding a device with feature set
7468 * @one to the master device with current feature set @all. Will not
7469 * enable anything that is off in @mask. Returns the new feature set.
7471 netdev_features_t netdev_increment_features(netdev_features_t all,
7472 netdev_features_t one, netdev_features_t mask)
7474 if (mask & NETIF_F_GEN_CSUM)
7475 mask |= NETIF_F_ALL_CSUM;
7476 mask |= NETIF_F_VLAN_CHALLENGED;
7478 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7479 all &= one | ~NETIF_F_ALL_FOR_ALL;
7481 /* If one device supports hw checksumming, set for all. */
7482 if (all & NETIF_F_GEN_CSUM)
7483 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7487 EXPORT_SYMBOL(netdev_increment_features);
7489 static struct hlist_head * __net_init netdev_create_hash(void)
7492 struct hlist_head *hash;
7494 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7496 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7497 INIT_HLIST_HEAD(&hash[i]);
7502 /* Initialize per network namespace state */
7503 static int __net_init netdev_init(struct net *net)
7505 if (net != &init_net)
7506 INIT_LIST_HEAD(&net->dev_base_head);
7508 net->dev_name_head = netdev_create_hash();
7509 if (net->dev_name_head == NULL)
7512 net->dev_index_head = netdev_create_hash();
7513 if (net->dev_index_head == NULL)
7519 kfree(net->dev_name_head);
7525 * netdev_drivername - network driver for the device
7526 * @dev: network device
7528 * Determine network driver for device.
7530 const char *netdev_drivername(const struct net_device *dev)
7532 const struct device_driver *driver;
7533 const struct device *parent;
7534 const char *empty = "";
7536 parent = dev->dev.parent;
7540 driver = parent->driver;
7541 if (driver && driver->name)
7542 return driver->name;
7546 static void __netdev_printk(const char *level, const struct net_device *dev,
7547 struct va_format *vaf)
7549 if (dev && dev->dev.parent) {
7550 dev_printk_emit(level[1] - '0',
7553 dev_driver_string(dev->dev.parent),
7554 dev_name(dev->dev.parent),
7555 netdev_name(dev), netdev_reg_state(dev),
7558 printk("%s%s%s: %pV",
7559 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7561 printk("%s(NULL net_device): %pV", level, vaf);
7565 void netdev_printk(const char *level, const struct net_device *dev,
7566 const char *format, ...)
7568 struct va_format vaf;
7571 va_start(args, format);
7576 __netdev_printk(level, dev, &vaf);
7580 EXPORT_SYMBOL(netdev_printk);
7582 #define define_netdev_printk_level(func, level) \
7583 void func(const struct net_device *dev, const char *fmt, ...) \
7585 struct va_format vaf; \
7588 va_start(args, fmt); \
7593 __netdev_printk(level, dev, &vaf); \
7597 EXPORT_SYMBOL(func);
7599 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7600 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7601 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7602 define_netdev_printk_level(netdev_err, KERN_ERR);
7603 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7604 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7605 define_netdev_printk_level(netdev_info, KERN_INFO);
7607 static void __net_exit netdev_exit(struct net *net)
7609 kfree(net->dev_name_head);
7610 kfree(net->dev_index_head);
7613 static struct pernet_operations __net_initdata netdev_net_ops = {
7614 .init = netdev_init,
7615 .exit = netdev_exit,
7618 static void __net_exit default_device_exit(struct net *net)
7620 struct net_device *dev, *aux;
7622 * Push all migratable network devices back to the
7623 * initial network namespace
7626 for_each_netdev_safe(net, dev, aux) {
7628 char fb_name[IFNAMSIZ];
7630 /* Ignore unmoveable devices (i.e. loopback) */
7631 if (dev->features & NETIF_F_NETNS_LOCAL)
7634 /* Leave virtual devices for the generic cleanup */
7635 if (dev->rtnl_link_ops)
7638 /* Push remaining network devices to init_net */
7639 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7640 err = dev_change_net_namespace(dev, &init_net, fb_name);
7642 pr_emerg("%s: failed to move %s to init_net: %d\n",
7643 __func__, dev->name, err);
7650 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7652 /* Return with the rtnl_lock held when there are no network
7653 * devices unregistering in any network namespace in net_list.
7657 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7659 add_wait_queue(&netdev_unregistering_wq, &wait);
7661 unregistering = false;
7663 list_for_each_entry(net, net_list, exit_list) {
7664 if (net->dev_unreg_count > 0) {
7665 unregistering = true;
7673 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7675 remove_wait_queue(&netdev_unregistering_wq, &wait);
7678 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7680 /* At exit all network devices most be removed from a network
7681 * namespace. Do this in the reverse order of registration.
7682 * Do this across as many network namespaces as possible to
7683 * improve batching efficiency.
7685 struct net_device *dev;
7687 LIST_HEAD(dev_kill_list);
7689 /* To prevent network device cleanup code from dereferencing
7690 * loopback devices or network devices that have been freed
7691 * wait here for all pending unregistrations to complete,
7692 * before unregistring the loopback device and allowing the
7693 * network namespace be freed.
7695 * The netdev todo list containing all network devices
7696 * unregistrations that happen in default_device_exit_batch
7697 * will run in the rtnl_unlock() at the end of
7698 * default_device_exit_batch.
7700 rtnl_lock_unregistering(net_list);
7701 list_for_each_entry(net, net_list, exit_list) {
7702 for_each_netdev_reverse(net, dev) {
7703 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7704 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7706 unregister_netdevice_queue(dev, &dev_kill_list);
7709 unregister_netdevice_many(&dev_kill_list);
7713 static struct pernet_operations __net_initdata default_device_ops = {
7714 .exit = default_device_exit,
7715 .exit_batch = default_device_exit_batch,
7719 * Initialize the DEV module. At boot time this walks the device list and
7720 * unhooks any devices that fail to initialise (normally hardware not
7721 * present) and leaves us with a valid list of present and active devices.
7726 * This is called single threaded during boot, so no need
7727 * to take the rtnl semaphore.
7729 static int __init net_dev_init(void)
7731 int i, rc = -ENOMEM;
7733 BUG_ON(!dev_boot_phase);
7735 if (dev_proc_init())
7738 if (netdev_kobject_init())
7741 INIT_LIST_HEAD(&ptype_all);
7742 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7743 INIT_LIST_HEAD(&ptype_base[i]);
7745 INIT_LIST_HEAD(&offload_base);
7747 if (register_pernet_subsys(&netdev_net_ops))
7751 * Initialise the packet receive queues.
7754 for_each_possible_cpu(i) {
7755 struct softnet_data *sd = &per_cpu(softnet_data, i);
7757 skb_queue_head_init(&sd->input_pkt_queue);
7758 skb_queue_head_init(&sd->process_queue);
7759 INIT_LIST_HEAD(&sd->poll_list);
7760 sd->output_queue_tailp = &sd->output_queue;
7762 sd->csd.func = rps_trigger_softirq;
7767 sd->backlog.poll = process_backlog;
7768 sd->backlog.weight = weight_p;
7773 /* The loopback device is special if any other network devices
7774 * is present in a network namespace the loopback device must
7775 * be present. Since we now dynamically allocate and free the
7776 * loopback device ensure this invariant is maintained by
7777 * keeping the loopback device as the first device on the
7778 * list of network devices. Ensuring the loopback devices
7779 * is the first device that appears and the last network device
7782 if (register_pernet_device(&loopback_net_ops))
7785 if (register_pernet_device(&default_device_ops))
7788 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7789 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7791 hotcpu_notifier(dev_cpu_callback, 0);
7798 subsys_initcall(net_dev_init);