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 static atomic_t netstamp_needed_deferred;
1680 static atomic_t netstamp_wanted;
1681 static void netstamp_clear(struct work_struct *work)
1683 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1686 wanted = atomic_add_return(deferred, &netstamp_wanted);
1688 static_key_enable(&netstamp_needed);
1690 static_key_disable(&netstamp_needed);
1692 static DECLARE_WORK(netstamp_work, netstamp_clear);
1695 void net_enable_timestamp(void)
1697 #ifdef HAVE_JUMP_LABEL
1701 wanted = atomic_read(&netstamp_wanted);
1704 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1707 atomic_inc(&netstamp_needed_deferred);
1708 schedule_work(&netstamp_work);
1710 static_key_slow_inc(&netstamp_needed);
1713 EXPORT_SYMBOL(net_enable_timestamp);
1715 void net_disable_timestamp(void)
1717 #ifdef HAVE_JUMP_LABEL
1721 wanted = atomic_read(&netstamp_wanted);
1724 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
1727 atomic_dec(&netstamp_needed_deferred);
1728 schedule_work(&netstamp_work);
1730 static_key_slow_dec(&netstamp_needed);
1733 EXPORT_SYMBOL(net_disable_timestamp);
1735 static inline void net_timestamp_set(struct sk_buff *skb)
1737 skb->tstamp.tv64 = 0;
1738 if (static_key_false(&netstamp_needed))
1739 __net_timestamp(skb);
1742 #define net_timestamp_check(COND, SKB) \
1743 if (static_key_false(&netstamp_needed)) { \
1744 if ((COND) && !(SKB)->tstamp.tv64) \
1745 __net_timestamp(SKB); \
1748 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1752 if (!(dev->flags & IFF_UP))
1755 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1756 if (skb->len <= len)
1759 /* if TSO is enabled, we don't care about the length as the packet
1760 * could be forwarded without being segmented before
1762 if (skb_is_gso(skb))
1767 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1769 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1771 if (skb_orphan_frags(skb, GFP_ATOMIC) ||
1772 unlikely(!is_skb_forwardable(dev, skb))) {
1773 atomic_long_inc(&dev->rx_dropped);
1778 skb_scrub_packet(skb, true);
1780 skb->protocol = eth_type_trans(skb, dev);
1781 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1785 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1788 * dev_forward_skb - loopback an skb to another netif
1790 * @dev: destination network device
1791 * @skb: buffer to forward
1794 * NET_RX_SUCCESS (no congestion)
1795 * NET_RX_DROP (packet was dropped, but freed)
1797 * dev_forward_skb can be used for injecting an skb from the
1798 * start_xmit function of one device into the receive queue
1799 * of another device.
1801 * The receiving device may be in another namespace, so
1802 * we have to clear all information in the skb that could
1803 * impact namespace isolation.
1805 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1807 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1809 EXPORT_SYMBOL_GPL(dev_forward_skb);
1811 static inline int deliver_skb(struct sk_buff *skb,
1812 struct packet_type *pt_prev,
1813 struct net_device *orig_dev)
1815 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1817 atomic_inc(&skb->users);
1818 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1821 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1822 struct packet_type **pt,
1823 struct net_device *orig_dev,
1825 struct list_head *ptype_list)
1827 struct packet_type *ptype, *pt_prev = *pt;
1829 list_for_each_entry_rcu(ptype, ptype_list, list) {
1830 if (ptype->type != type)
1833 deliver_skb(skb, pt_prev, orig_dev);
1839 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1841 if (!ptype->af_packet_priv || !skb->sk)
1844 if (ptype->id_match)
1845 return ptype->id_match(ptype, skb->sk);
1846 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1853 * Support routine. Sends outgoing frames to any network
1854 * taps currently in use.
1857 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1859 struct packet_type *ptype;
1860 struct sk_buff *skb2 = NULL;
1861 struct packet_type *pt_prev = NULL;
1862 struct list_head *ptype_list = &ptype_all;
1866 list_for_each_entry_rcu(ptype, ptype_list, list) {
1867 /* Never send packets back to the socket
1868 * they originated from - MvS (miquels@drinkel.ow.org)
1870 if (skb_loop_sk(ptype, skb))
1874 deliver_skb(skb2, pt_prev, skb->dev);
1879 /* need to clone skb, done only once */
1880 skb2 = skb_clone(skb, GFP_ATOMIC);
1884 net_timestamp_set(skb2);
1886 /* skb->nh should be correctly
1887 * set by sender, so that the second statement is
1888 * just protection against buggy protocols.
1890 skb_reset_mac_header(skb2);
1892 if (skb_network_header(skb2) < skb2->data ||
1893 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1894 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1895 ntohs(skb2->protocol),
1897 skb_reset_network_header(skb2);
1900 skb2->transport_header = skb2->network_header;
1901 skb2->pkt_type = PACKET_OUTGOING;
1905 if (ptype_list == &ptype_all) {
1906 ptype_list = &dev->ptype_all;
1911 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1916 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1917 * @dev: Network device
1918 * @txq: number of queues available
1920 * If real_num_tx_queues is changed the tc mappings may no longer be
1921 * valid. To resolve this verify the tc mapping remains valid and if
1922 * not NULL the mapping. With no priorities mapping to this
1923 * offset/count pair it will no longer be used. In the worst case TC0
1924 * is invalid nothing can be done so disable priority mappings. If is
1925 * expected that drivers will fix this mapping if they can before
1926 * calling netif_set_real_num_tx_queues.
1928 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1931 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1933 /* If TC0 is invalidated disable TC mapping */
1934 if (tc->offset + tc->count > txq) {
1935 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1940 /* Invalidated prio to tc mappings set to TC0 */
1941 for (i = 1; i < TC_BITMASK + 1; i++) {
1942 int q = netdev_get_prio_tc_map(dev, i);
1944 tc = &dev->tc_to_txq[q];
1945 if (tc->offset + tc->count > txq) {
1946 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1948 netdev_set_prio_tc_map(dev, i, 0);
1954 static DEFINE_MUTEX(xps_map_mutex);
1955 #define xmap_dereference(P) \
1956 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1958 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1961 struct xps_map *map = NULL;
1965 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1967 for (pos = 0; map && pos < map->len; pos++) {
1968 if (map->queues[pos] == index) {
1970 map->queues[pos] = map->queues[--map->len];
1972 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1973 kfree_rcu(map, rcu);
1983 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1985 struct xps_dev_maps *dev_maps;
1987 bool active = false;
1989 mutex_lock(&xps_map_mutex);
1990 dev_maps = xmap_dereference(dev->xps_maps);
1995 for_each_possible_cpu(cpu) {
1996 for (i = index; i < dev->num_tx_queues; i++) {
1997 if (!remove_xps_queue(dev_maps, cpu, i))
2000 if (i == dev->num_tx_queues)
2005 RCU_INIT_POINTER(dev->xps_maps, NULL);
2006 kfree_rcu(dev_maps, rcu);
2009 for (i = index; i < dev->num_tx_queues; i++)
2010 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2014 mutex_unlock(&xps_map_mutex);
2017 static struct xps_map *expand_xps_map(struct xps_map *map,
2020 struct xps_map *new_map;
2021 int alloc_len = XPS_MIN_MAP_ALLOC;
2024 for (pos = 0; map && pos < map->len; pos++) {
2025 if (map->queues[pos] != index)
2030 /* Need to add queue to this CPU's existing map */
2032 if (pos < map->alloc_len)
2035 alloc_len = map->alloc_len * 2;
2038 /* Need to allocate new map to store queue on this CPU's map */
2039 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2044 for (i = 0; i < pos; i++)
2045 new_map->queues[i] = map->queues[i];
2046 new_map->alloc_len = alloc_len;
2052 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2055 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2056 struct xps_map *map, *new_map;
2057 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2058 int cpu, numa_node_id = -2;
2059 bool active = false;
2061 mutex_lock(&xps_map_mutex);
2063 dev_maps = xmap_dereference(dev->xps_maps);
2065 /* allocate memory for queue storage */
2066 for_each_online_cpu(cpu) {
2067 if (!cpumask_test_cpu(cpu, mask))
2071 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2072 if (!new_dev_maps) {
2073 mutex_unlock(&xps_map_mutex);
2077 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2080 map = expand_xps_map(map, cpu, index);
2084 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2088 goto out_no_new_maps;
2090 for_each_possible_cpu(cpu) {
2091 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2092 /* add queue to CPU maps */
2095 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2096 while ((pos < map->len) && (map->queues[pos] != index))
2099 if (pos == map->len)
2100 map->queues[map->len++] = index;
2102 if (numa_node_id == -2)
2103 numa_node_id = cpu_to_node(cpu);
2104 else if (numa_node_id != cpu_to_node(cpu))
2107 } else if (dev_maps) {
2108 /* fill in the new device map from the old device map */
2109 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2110 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2115 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2117 /* Cleanup old maps */
2119 for_each_possible_cpu(cpu) {
2120 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2121 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2122 if (map && map != new_map)
2123 kfree_rcu(map, rcu);
2126 kfree_rcu(dev_maps, rcu);
2129 dev_maps = new_dev_maps;
2133 /* update Tx queue numa node */
2134 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2135 (numa_node_id >= 0) ? numa_node_id :
2141 /* removes queue from unused CPUs */
2142 for_each_possible_cpu(cpu) {
2143 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2146 if (remove_xps_queue(dev_maps, cpu, index))
2150 /* free map if not active */
2152 RCU_INIT_POINTER(dev->xps_maps, NULL);
2153 kfree_rcu(dev_maps, rcu);
2157 mutex_unlock(&xps_map_mutex);
2161 /* remove any maps that we added */
2162 for_each_possible_cpu(cpu) {
2163 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2164 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2166 if (new_map && new_map != map)
2170 mutex_unlock(&xps_map_mutex);
2172 kfree(new_dev_maps);
2175 EXPORT_SYMBOL(netif_set_xps_queue);
2179 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2180 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2182 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2186 if (txq < 1 || txq > dev->num_tx_queues)
2189 if (dev->reg_state == NETREG_REGISTERED ||
2190 dev->reg_state == NETREG_UNREGISTERING) {
2193 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2199 netif_setup_tc(dev, txq);
2201 if (txq < dev->real_num_tx_queues) {
2202 qdisc_reset_all_tx_gt(dev, txq);
2204 netif_reset_xps_queues_gt(dev, txq);
2209 dev->real_num_tx_queues = txq;
2212 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2216 * netif_set_real_num_rx_queues - set actual number of RX queues used
2217 * @dev: Network device
2218 * @rxq: Actual number of RX queues
2220 * This must be called either with the rtnl_lock held or before
2221 * registration of the net device. Returns 0 on success, or a
2222 * negative error code. If called before registration, it always
2225 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2229 if (rxq < 1 || rxq > dev->num_rx_queues)
2232 if (dev->reg_state == NETREG_REGISTERED) {
2235 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2241 dev->real_num_rx_queues = rxq;
2244 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2248 * netif_get_num_default_rss_queues - default number of RSS queues
2250 * This routine should set an upper limit on the number of RSS queues
2251 * used by default by multiqueue devices.
2253 int netif_get_num_default_rss_queues(void)
2255 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2257 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2259 static inline void __netif_reschedule(struct Qdisc *q)
2261 struct softnet_data *sd;
2262 unsigned long flags;
2264 local_irq_save(flags);
2265 sd = this_cpu_ptr(&softnet_data);
2266 q->next_sched = NULL;
2267 *sd->output_queue_tailp = q;
2268 sd->output_queue_tailp = &q->next_sched;
2269 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2270 local_irq_restore(flags);
2273 void __netif_schedule(struct Qdisc *q)
2275 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2276 __netif_reschedule(q);
2278 EXPORT_SYMBOL(__netif_schedule);
2280 struct dev_kfree_skb_cb {
2281 enum skb_free_reason reason;
2284 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2286 return (struct dev_kfree_skb_cb *)skb->cb;
2289 void netif_schedule_queue(struct netdev_queue *txq)
2292 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2293 struct Qdisc *q = rcu_dereference(txq->qdisc);
2295 __netif_schedule(q);
2299 EXPORT_SYMBOL(netif_schedule_queue);
2302 * netif_wake_subqueue - allow sending packets on subqueue
2303 * @dev: network device
2304 * @queue_index: sub queue index
2306 * Resume individual transmit queue of a device with multiple transmit queues.
2308 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2310 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2312 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2316 q = rcu_dereference(txq->qdisc);
2317 __netif_schedule(q);
2321 EXPORT_SYMBOL(netif_wake_subqueue);
2323 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2325 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2329 q = rcu_dereference(dev_queue->qdisc);
2330 __netif_schedule(q);
2334 EXPORT_SYMBOL(netif_tx_wake_queue);
2336 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2338 unsigned long flags;
2340 if (likely(atomic_read(&skb->users) == 1)) {
2342 atomic_set(&skb->users, 0);
2343 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2346 get_kfree_skb_cb(skb)->reason = reason;
2347 local_irq_save(flags);
2348 skb->next = __this_cpu_read(softnet_data.completion_queue);
2349 __this_cpu_write(softnet_data.completion_queue, skb);
2350 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2351 local_irq_restore(flags);
2353 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2355 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2357 if (in_irq() || irqs_disabled())
2358 __dev_kfree_skb_irq(skb, reason);
2362 EXPORT_SYMBOL(__dev_kfree_skb_any);
2366 * netif_device_detach - mark device as removed
2367 * @dev: network device
2369 * Mark device as removed from system and therefore no longer available.
2371 void netif_device_detach(struct net_device *dev)
2373 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2374 netif_running(dev)) {
2375 netif_tx_stop_all_queues(dev);
2378 EXPORT_SYMBOL(netif_device_detach);
2381 * netif_device_attach - mark device as attached
2382 * @dev: network device
2384 * Mark device as attached from system and restart if needed.
2386 void netif_device_attach(struct net_device *dev)
2388 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2389 netif_running(dev)) {
2390 netif_tx_wake_all_queues(dev);
2391 __netdev_watchdog_up(dev);
2394 EXPORT_SYMBOL(netif_device_attach);
2397 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2398 * to be used as a distribution range.
2400 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2401 unsigned int num_tx_queues)
2405 u16 qcount = num_tx_queues;
2407 if (skb_rx_queue_recorded(skb)) {
2408 hash = skb_get_rx_queue(skb);
2409 while (unlikely(hash >= num_tx_queues))
2410 hash -= num_tx_queues;
2415 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2416 qoffset = dev->tc_to_txq[tc].offset;
2417 qcount = dev->tc_to_txq[tc].count;
2420 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2422 EXPORT_SYMBOL(__skb_tx_hash);
2424 static void skb_warn_bad_offload(const struct sk_buff *skb)
2426 static const netdev_features_t null_features = 0;
2427 struct net_device *dev = skb->dev;
2428 const char *name = "";
2430 if (!net_ratelimit())
2434 if (dev->dev.parent)
2435 name = dev_driver_string(dev->dev.parent);
2437 name = netdev_name(dev);
2439 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2440 "gso_type=%d ip_summed=%d\n",
2441 name, dev ? &dev->features : &null_features,
2442 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2443 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2444 skb_shinfo(skb)->gso_type, skb->ip_summed);
2448 * Invalidate hardware checksum when packet is to be mangled, and
2449 * complete checksum manually on outgoing path.
2451 int skb_checksum_help(struct sk_buff *skb)
2454 int ret = 0, offset;
2456 if (skb->ip_summed == CHECKSUM_COMPLETE)
2457 goto out_set_summed;
2459 if (unlikely(skb_shinfo(skb)->gso_size)) {
2460 skb_warn_bad_offload(skb);
2464 /* Before computing a checksum, we should make sure no frag could
2465 * be modified by an external entity : checksum could be wrong.
2467 if (skb_has_shared_frag(skb)) {
2468 ret = __skb_linearize(skb);
2473 offset = skb_checksum_start_offset(skb);
2474 BUG_ON(offset >= skb_headlen(skb));
2475 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2477 offset += skb->csum_offset;
2478 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2480 if (skb_cloned(skb) &&
2481 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2482 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2487 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2489 skb->ip_summed = CHECKSUM_NONE;
2493 EXPORT_SYMBOL(skb_checksum_help);
2495 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2497 __be16 type = skb->protocol;
2499 /* Tunnel gso handlers can set protocol to ethernet. */
2500 if (type == htons(ETH_P_TEB)) {
2503 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2506 eth = (struct ethhdr *)skb_mac_header(skb);
2507 type = eth->h_proto;
2510 return __vlan_get_protocol(skb, type, depth);
2514 * skb_mac_gso_segment - mac layer segmentation handler.
2515 * @skb: buffer to segment
2516 * @features: features for the output path (see dev->features)
2518 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2519 netdev_features_t features)
2521 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2522 struct packet_offload *ptype;
2523 int vlan_depth = skb->mac_len;
2524 __be16 type = skb_network_protocol(skb, &vlan_depth);
2526 if (unlikely(!type))
2527 return ERR_PTR(-EINVAL);
2529 __skb_pull(skb, vlan_depth);
2532 list_for_each_entry_rcu(ptype, &offload_base, list) {
2533 if (ptype->type == type && ptype->callbacks.gso_segment) {
2534 segs = ptype->callbacks.gso_segment(skb, features);
2540 __skb_push(skb, skb->data - skb_mac_header(skb));
2544 EXPORT_SYMBOL(skb_mac_gso_segment);
2547 /* openvswitch calls this on rx path, so we need a different check.
2549 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2552 return skb->ip_summed != CHECKSUM_PARTIAL;
2554 return skb->ip_summed == CHECKSUM_NONE;
2558 * __skb_gso_segment - Perform segmentation on skb.
2559 * @skb: buffer to segment
2560 * @features: features for the output path (see dev->features)
2561 * @tx_path: whether it is called in TX path
2563 * This function segments the given skb and returns a list of segments.
2565 * It may return NULL if the skb requires no segmentation. This is
2566 * only possible when GSO is used for verifying header integrity.
2568 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2570 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2571 netdev_features_t features, bool tx_path)
2573 if (unlikely(skb_needs_check(skb, tx_path))) {
2576 skb_warn_bad_offload(skb);
2578 err = skb_cow_head(skb, 0);
2580 return ERR_PTR(err);
2583 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2584 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2586 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2587 SKB_GSO_CB(skb)->encap_level = 0;
2589 skb_reset_mac_header(skb);
2590 skb_reset_mac_len(skb);
2592 return skb_mac_gso_segment(skb, features);
2594 EXPORT_SYMBOL(__skb_gso_segment);
2596 /* Take action when hardware reception checksum errors are detected. */
2598 void netdev_rx_csum_fault(struct net_device *dev)
2600 if (net_ratelimit()) {
2601 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2605 EXPORT_SYMBOL(netdev_rx_csum_fault);
2608 /* Actually, we should eliminate this check as soon as we know, that:
2609 * 1. IOMMU is present and allows to map all the memory.
2610 * 2. No high memory really exists on this machine.
2613 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2615 #ifdef CONFIG_HIGHMEM
2617 if (!(dev->features & NETIF_F_HIGHDMA)) {
2618 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2619 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2620 if (PageHighMem(skb_frag_page(frag)))
2625 if (PCI_DMA_BUS_IS_PHYS) {
2626 struct device *pdev = dev->dev.parent;
2630 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2631 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2632 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2633 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2641 /* If MPLS offload request, verify we are testing hardware MPLS features
2642 * instead of standard features for the netdev.
2644 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2645 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2646 netdev_features_t features,
2649 if (eth_p_mpls(type))
2650 features &= skb->dev->mpls_features;
2655 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2656 netdev_features_t features,
2663 static netdev_features_t harmonize_features(struct sk_buff *skb,
2664 netdev_features_t features)
2669 type = skb_network_protocol(skb, &tmp);
2670 features = net_mpls_features(skb, features, type);
2672 if (skb->ip_summed != CHECKSUM_NONE &&
2673 !can_checksum_protocol(features, type)) {
2674 features &= ~NETIF_F_ALL_CSUM;
2676 if (illegal_highdma(skb->dev, skb))
2677 features &= ~NETIF_F_SG;
2682 netdev_features_t passthru_features_check(struct sk_buff *skb,
2683 struct net_device *dev,
2684 netdev_features_t features)
2688 EXPORT_SYMBOL(passthru_features_check);
2690 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2691 struct net_device *dev,
2692 netdev_features_t features)
2694 return vlan_features_check(skb, features);
2697 netdev_features_t netif_skb_features(struct sk_buff *skb)
2699 struct net_device *dev = skb->dev;
2700 netdev_features_t features = dev->features;
2701 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2703 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2704 features &= ~NETIF_F_GSO_MASK;
2706 /* If encapsulation offload request, verify we are testing
2707 * hardware encapsulation features instead of standard
2708 * features for the netdev
2710 if (skb->encapsulation)
2711 features &= dev->hw_enc_features;
2713 if (skb_vlan_tagged(skb))
2714 features = netdev_intersect_features(features,
2715 dev->vlan_features |
2716 NETIF_F_HW_VLAN_CTAG_TX |
2717 NETIF_F_HW_VLAN_STAG_TX);
2719 if (dev->netdev_ops->ndo_features_check)
2720 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2723 features &= dflt_features_check(skb, dev, features);
2725 return harmonize_features(skb, features);
2727 EXPORT_SYMBOL(netif_skb_features);
2729 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2730 struct netdev_queue *txq, bool more)
2735 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2736 dev_queue_xmit_nit(skb, dev);
2739 trace_net_dev_start_xmit(skb, dev);
2740 rc = netdev_start_xmit(skb, dev, txq, more);
2741 trace_net_dev_xmit(skb, rc, dev, len);
2746 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2747 struct netdev_queue *txq, int *ret)
2749 struct sk_buff *skb = first;
2750 int rc = NETDEV_TX_OK;
2753 struct sk_buff *next = skb->next;
2756 rc = xmit_one(skb, dev, txq, next != NULL);
2757 if (unlikely(!dev_xmit_complete(rc))) {
2763 if (netif_xmit_stopped(txq) && skb) {
2764 rc = NETDEV_TX_BUSY;
2774 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2775 netdev_features_t features)
2777 if (skb_vlan_tag_present(skb) &&
2778 !vlan_hw_offload_capable(features, skb->vlan_proto))
2779 skb = __vlan_hwaccel_push_inside(skb);
2783 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2785 netdev_features_t features;
2790 features = netif_skb_features(skb);
2791 skb = validate_xmit_vlan(skb, features);
2795 if (netif_needs_gso(skb, features)) {
2796 struct sk_buff *segs;
2798 segs = skb_gso_segment(skb, features);
2806 if (skb_needs_linearize(skb, features) &&
2807 __skb_linearize(skb))
2810 /* If packet is not checksummed and device does not
2811 * support checksumming for this protocol, complete
2812 * checksumming here.
2814 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2815 if (skb->encapsulation)
2816 skb_set_inner_transport_header(skb,
2817 skb_checksum_start_offset(skb));
2819 skb_set_transport_header(skb,
2820 skb_checksum_start_offset(skb));
2821 if (!(features & NETIF_F_ALL_CSUM) &&
2822 skb_checksum_help(skb))
2835 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2837 struct sk_buff *next, *head = NULL, *tail;
2839 for (; skb != NULL; skb = next) {
2843 /* in case skb wont be segmented, point to itself */
2846 skb = validate_xmit_skb(skb, dev);
2854 /* If skb was segmented, skb->prev points to
2855 * the last segment. If not, it still contains skb.
2861 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
2863 static void qdisc_pkt_len_init(struct sk_buff *skb)
2865 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2867 qdisc_skb_cb(skb)->pkt_len = skb->len;
2869 /* To get more precise estimation of bytes sent on wire,
2870 * we add to pkt_len the headers size of all segments
2872 if (shinfo->gso_size) {
2873 unsigned int hdr_len;
2874 u16 gso_segs = shinfo->gso_segs;
2876 /* mac layer + network layer */
2877 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2879 /* + transport layer */
2880 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2881 hdr_len += tcp_hdrlen(skb);
2883 hdr_len += sizeof(struct udphdr);
2885 if (shinfo->gso_type & SKB_GSO_DODGY)
2886 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2889 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2893 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2894 struct net_device *dev,
2895 struct netdev_queue *txq)
2897 spinlock_t *root_lock = qdisc_lock(q);
2901 qdisc_pkt_len_init(skb);
2902 qdisc_calculate_pkt_len(skb, q);
2904 * Heuristic to force contended enqueues to serialize on a
2905 * separate lock before trying to get qdisc main lock.
2906 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2907 * often and dequeue packets faster.
2909 contended = qdisc_is_running(q);
2910 if (unlikely(contended))
2911 spin_lock(&q->busylock);
2913 spin_lock(root_lock);
2914 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2917 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2918 qdisc_run_begin(q)) {
2920 * This is a work-conserving queue; there are no old skbs
2921 * waiting to be sent out; and the qdisc is not running -
2922 * xmit the skb directly.
2925 qdisc_bstats_update(q, skb);
2927 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2928 if (unlikely(contended)) {
2929 spin_unlock(&q->busylock);
2936 rc = NET_XMIT_SUCCESS;
2938 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2939 if (qdisc_run_begin(q)) {
2940 if (unlikely(contended)) {
2941 spin_unlock(&q->busylock);
2947 spin_unlock(root_lock);
2948 if (unlikely(contended))
2949 spin_unlock(&q->busylock);
2953 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2954 static void skb_update_prio(struct sk_buff *skb)
2956 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2958 if (!skb->priority && skb->sk && map) {
2959 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2961 if (prioidx < map->priomap_len)
2962 skb->priority = map->priomap[prioidx];
2966 #define skb_update_prio(skb)
2969 DEFINE_PER_CPU(int, xmit_recursion);
2970 EXPORT_SYMBOL(xmit_recursion);
2972 #define RECURSION_LIMIT 10
2975 * dev_loopback_xmit - loop back @skb
2976 * @net: network namespace this loopback is happening in
2977 * @sk: sk needed to be a netfilter okfn
2978 * @skb: buffer to transmit
2980 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
2982 skb_reset_mac_header(skb);
2983 __skb_pull(skb, skb_network_offset(skb));
2984 skb->pkt_type = PACKET_LOOPBACK;
2985 skb->ip_summed = CHECKSUM_UNNECESSARY;
2986 WARN_ON(!skb_dst(skb));
2991 EXPORT_SYMBOL(dev_loopback_xmit);
2993 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2996 struct xps_dev_maps *dev_maps;
2997 struct xps_map *map;
2998 int queue_index = -1;
3001 dev_maps = rcu_dereference(dev->xps_maps);
3003 map = rcu_dereference(
3004 dev_maps->cpu_map[skb->sender_cpu - 1]);
3007 queue_index = map->queues[0];
3009 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3011 if (unlikely(queue_index >= dev->real_num_tx_queues))
3023 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3025 struct sock *sk = skb->sk;
3026 int queue_index = sk_tx_queue_get(sk);
3028 if (queue_index < 0 || skb->ooo_okay ||
3029 queue_index >= dev->real_num_tx_queues) {
3030 int new_index = get_xps_queue(dev, skb);
3032 new_index = skb_tx_hash(dev, skb);
3034 if (queue_index != new_index && sk &&
3036 rcu_access_pointer(sk->sk_dst_cache))
3037 sk_tx_queue_set(sk, new_index);
3039 queue_index = new_index;
3045 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3046 struct sk_buff *skb,
3049 int queue_index = 0;
3052 if (skb->sender_cpu == 0)
3053 skb->sender_cpu = raw_smp_processor_id() + 1;
3056 if (dev->real_num_tx_queues != 1) {
3057 const struct net_device_ops *ops = dev->netdev_ops;
3058 if (ops->ndo_select_queue)
3059 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3062 queue_index = __netdev_pick_tx(dev, skb);
3065 queue_index = netdev_cap_txqueue(dev, queue_index);
3068 skb_set_queue_mapping(skb, queue_index);
3069 return netdev_get_tx_queue(dev, queue_index);
3073 * __dev_queue_xmit - transmit a buffer
3074 * @skb: buffer to transmit
3075 * @accel_priv: private data used for L2 forwarding offload
3077 * Queue a buffer for transmission to a network device. The caller must
3078 * have set the device and priority and built the buffer before calling
3079 * this function. The function can be called from an interrupt.
3081 * A negative errno code is returned on a failure. A success does not
3082 * guarantee the frame will be transmitted as it may be dropped due
3083 * to congestion or traffic shaping.
3085 * -----------------------------------------------------------------------------------
3086 * I notice this method can also return errors from the queue disciplines,
3087 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3090 * Regardless of the return value, the skb is consumed, so it is currently
3091 * difficult to retry a send to this method. (You can bump the ref count
3092 * before sending to hold a reference for retry if you are careful.)
3094 * When calling this method, interrupts MUST be enabled. This is because
3095 * the BH enable code must have IRQs enabled so that it will not deadlock.
3098 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3100 struct net_device *dev = skb->dev;
3101 struct netdev_queue *txq;
3105 skb_reset_mac_header(skb);
3107 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3108 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3110 /* Disable soft irqs for various locks below. Also
3111 * stops preemption for RCU.
3115 skb_update_prio(skb);
3117 /* If device/qdisc don't need skb->dst, release it right now while
3118 * its hot in this cpu cache.
3120 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3125 #ifdef CONFIG_NET_SWITCHDEV
3126 /* Don't forward if offload device already forwarded */
3127 if (skb->offload_fwd_mark &&
3128 skb->offload_fwd_mark == dev->offload_fwd_mark) {
3130 rc = NET_XMIT_SUCCESS;
3135 txq = netdev_pick_tx(dev, skb, accel_priv);
3136 q = rcu_dereference_bh(txq->qdisc);
3138 #ifdef CONFIG_NET_CLS_ACT
3139 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3141 trace_net_dev_queue(skb);
3143 rc = __dev_xmit_skb(skb, q, dev, txq);
3147 /* The device has no queue. Common case for software devices:
3148 loopback, all the sorts of tunnels...
3150 Really, it is unlikely that netif_tx_lock protection is necessary
3151 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3153 However, it is possible, that they rely on protection
3156 Check this and shot the lock. It is not prone from deadlocks.
3157 Either shot noqueue qdisc, it is even simpler 8)
3159 if (dev->flags & IFF_UP) {
3160 int cpu = smp_processor_id(); /* ok because BHs are off */
3162 if (txq->xmit_lock_owner != cpu) {
3164 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
3165 goto recursion_alert;
3167 skb = validate_xmit_skb(skb, dev);
3171 HARD_TX_LOCK(dev, txq, cpu);
3173 if (!netif_xmit_stopped(txq)) {
3174 __this_cpu_inc(xmit_recursion);
3175 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3176 __this_cpu_dec(xmit_recursion);
3177 if (dev_xmit_complete(rc)) {
3178 HARD_TX_UNLOCK(dev, txq);
3182 HARD_TX_UNLOCK(dev, txq);
3183 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3186 /* Recursion is detected! It is possible,
3190 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3197 rcu_read_unlock_bh();
3199 atomic_long_inc(&dev->tx_dropped);
3200 kfree_skb_list(skb);
3203 rcu_read_unlock_bh();
3207 int dev_queue_xmit(struct sk_buff *skb)
3209 return __dev_queue_xmit(skb, NULL);
3211 EXPORT_SYMBOL(dev_queue_xmit);
3213 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3215 return __dev_queue_xmit(skb, accel_priv);
3217 EXPORT_SYMBOL(dev_queue_xmit_accel);
3220 /*=======================================================================
3222 =======================================================================*/
3224 int netdev_max_backlog __read_mostly = 1000;
3225 EXPORT_SYMBOL(netdev_max_backlog);
3227 int netdev_tstamp_prequeue __read_mostly = 1;
3228 int netdev_budget __read_mostly = 300;
3229 int weight_p __read_mostly = 64; /* old backlog weight */
3231 /* Called with irq disabled */
3232 static inline void ____napi_schedule(struct softnet_data *sd,
3233 struct napi_struct *napi)
3235 list_add_tail(&napi->poll_list, &sd->poll_list);
3236 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3241 /* One global table that all flow-based protocols share. */
3242 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3243 EXPORT_SYMBOL(rps_sock_flow_table);
3244 u32 rps_cpu_mask __read_mostly;
3245 EXPORT_SYMBOL(rps_cpu_mask);
3247 struct static_key rps_needed __read_mostly;
3249 static struct rps_dev_flow *
3250 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3251 struct rps_dev_flow *rflow, u16 next_cpu)
3253 if (next_cpu < nr_cpu_ids) {
3254 #ifdef CONFIG_RFS_ACCEL
3255 struct netdev_rx_queue *rxqueue;
3256 struct rps_dev_flow_table *flow_table;
3257 struct rps_dev_flow *old_rflow;
3262 /* Should we steer this flow to a different hardware queue? */
3263 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3264 !(dev->features & NETIF_F_NTUPLE))
3266 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3267 if (rxq_index == skb_get_rx_queue(skb))
3270 rxqueue = dev->_rx + rxq_index;
3271 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3274 flow_id = skb_get_hash(skb) & flow_table->mask;
3275 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3276 rxq_index, flow_id);
3280 rflow = &flow_table->flows[flow_id];
3282 if (old_rflow->filter == rflow->filter)
3283 old_rflow->filter = RPS_NO_FILTER;
3287 per_cpu(softnet_data, next_cpu).input_queue_head;
3290 rflow->cpu = next_cpu;
3295 * get_rps_cpu is called from netif_receive_skb and returns the target
3296 * CPU from the RPS map of the receiving queue for a given skb.
3297 * rcu_read_lock must be held on entry.
3299 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3300 struct rps_dev_flow **rflowp)
3302 const struct rps_sock_flow_table *sock_flow_table;
3303 struct netdev_rx_queue *rxqueue = dev->_rx;
3304 struct rps_dev_flow_table *flow_table;
3305 struct rps_map *map;
3310 if (skb_rx_queue_recorded(skb)) {
3311 u16 index = skb_get_rx_queue(skb);
3313 if (unlikely(index >= dev->real_num_rx_queues)) {
3314 WARN_ONCE(dev->real_num_rx_queues > 1,
3315 "%s received packet on queue %u, but number "
3316 "of RX queues is %u\n",
3317 dev->name, index, dev->real_num_rx_queues);
3323 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3325 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3326 map = rcu_dereference(rxqueue->rps_map);
3327 if (!flow_table && !map)
3330 skb_reset_network_header(skb);
3331 hash = skb_get_hash(skb);
3335 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3336 if (flow_table && sock_flow_table) {
3337 struct rps_dev_flow *rflow;
3341 /* First check into global flow table if there is a match */
3342 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3343 if ((ident ^ hash) & ~rps_cpu_mask)
3346 next_cpu = ident & rps_cpu_mask;
3348 /* OK, now we know there is a match,
3349 * we can look at the local (per receive queue) flow table
3351 rflow = &flow_table->flows[hash & flow_table->mask];
3355 * If the desired CPU (where last recvmsg was done) is
3356 * different from current CPU (one in the rx-queue flow
3357 * table entry), switch if one of the following holds:
3358 * - Current CPU is unset (>= nr_cpu_ids).
3359 * - Current CPU is offline.
3360 * - The current CPU's queue tail has advanced beyond the
3361 * last packet that was enqueued using this table entry.
3362 * This guarantees that all previous packets for the flow
3363 * have been dequeued, thus preserving in order delivery.
3365 if (unlikely(tcpu != next_cpu) &&
3366 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3367 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3368 rflow->last_qtail)) >= 0)) {
3370 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3373 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3383 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3384 if (cpu_online(tcpu)) {
3394 #ifdef CONFIG_RFS_ACCEL
3397 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3398 * @dev: Device on which the filter was set
3399 * @rxq_index: RX queue index
3400 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3401 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3403 * Drivers that implement ndo_rx_flow_steer() should periodically call
3404 * this function for each installed filter and remove the filters for
3405 * which it returns %true.
3407 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3408 u32 flow_id, u16 filter_id)
3410 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3411 struct rps_dev_flow_table *flow_table;
3412 struct rps_dev_flow *rflow;
3417 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3418 if (flow_table && flow_id <= flow_table->mask) {
3419 rflow = &flow_table->flows[flow_id];
3420 cpu = ACCESS_ONCE(rflow->cpu);
3421 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3422 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3423 rflow->last_qtail) <
3424 (int)(10 * flow_table->mask)))
3430 EXPORT_SYMBOL(rps_may_expire_flow);
3432 #endif /* CONFIG_RFS_ACCEL */
3434 /* Called from hardirq (IPI) context */
3435 static void rps_trigger_softirq(void *data)
3437 struct softnet_data *sd = data;
3439 ____napi_schedule(sd, &sd->backlog);
3443 #endif /* CONFIG_RPS */
3446 * Check if this softnet_data structure is another cpu one
3447 * If yes, queue it to our IPI list and return 1
3450 static int rps_ipi_queued(struct softnet_data *sd)
3453 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3456 sd->rps_ipi_next = mysd->rps_ipi_list;
3457 mysd->rps_ipi_list = sd;
3459 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3462 #endif /* CONFIG_RPS */
3466 #ifdef CONFIG_NET_FLOW_LIMIT
3467 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3470 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3472 #ifdef CONFIG_NET_FLOW_LIMIT
3473 struct sd_flow_limit *fl;
3474 struct softnet_data *sd;
3475 unsigned int old_flow, new_flow;
3477 if (qlen < (netdev_max_backlog >> 1))
3480 sd = this_cpu_ptr(&softnet_data);
3483 fl = rcu_dereference(sd->flow_limit);
3485 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3486 old_flow = fl->history[fl->history_head];
3487 fl->history[fl->history_head] = new_flow;
3490 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3492 if (likely(fl->buckets[old_flow]))
3493 fl->buckets[old_flow]--;
3495 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3507 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3508 * queue (may be a remote CPU queue).
3510 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3511 unsigned int *qtail)
3513 struct softnet_data *sd;
3514 unsigned long flags;
3517 sd = &per_cpu(softnet_data, cpu);
3519 local_irq_save(flags);
3522 if (!netif_running(skb->dev))
3524 qlen = skb_queue_len(&sd->input_pkt_queue);
3525 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3528 __skb_queue_tail(&sd->input_pkt_queue, skb);
3529 input_queue_tail_incr_save(sd, qtail);
3531 local_irq_restore(flags);
3532 return NET_RX_SUCCESS;
3535 /* Schedule NAPI for backlog device
3536 * We can use non atomic operation since we own the queue lock
3538 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3539 if (!rps_ipi_queued(sd))
3540 ____napi_schedule(sd, &sd->backlog);
3549 local_irq_restore(flags);
3551 atomic_long_inc(&skb->dev->rx_dropped);
3556 static int netif_rx_internal(struct sk_buff *skb)
3560 net_timestamp_check(netdev_tstamp_prequeue, skb);
3562 trace_netif_rx(skb);
3564 if (static_key_false(&rps_needed)) {
3565 struct rps_dev_flow voidflow, *rflow = &voidflow;
3571 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3573 cpu = smp_processor_id();
3575 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3583 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3590 * netif_rx - post buffer to the network code
3591 * @skb: buffer to post
3593 * This function receives a packet from a device driver and queues it for
3594 * the upper (protocol) levels to process. It always succeeds. The buffer
3595 * may be dropped during processing for congestion control or by the
3599 * NET_RX_SUCCESS (no congestion)
3600 * NET_RX_DROP (packet was dropped)
3604 int netif_rx(struct sk_buff *skb)
3606 trace_netif_rx_entry(skb);
3608 return netif_rx_internal(skb);
3610 EXPORT_SYMBOL(netif_rx);
3612 int netif_rx_ni(struct sk_buff *skb)
3616 trace_netif_rx_ni_entry(skb);
3619 err = netif_rx_internal(skb);
3620 if (local_softirq_pending())
3626 EXPORT_SYMBOL(netif_rx_ni);
3628 static void net_tx_action(struct softirq_action *h)
3630 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3632 if (sd->completion_queue) {
3633 struct sk_buff *clist;
3635 local_irq_disable();
3636 clist = sd->completion_queue;
3637 sd->completion_queue = NULL;
3641 struct sk_buff *skb = clist;
3642 clist = clist->next;
3644 WARN_ON(atomic_read(&skb->users));
3645 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3646 trace_consume_skb(skb);
3648 trace_kfree_skb(skb, net_tx_action);
3653 if (sd->output_queue) {
3656 local_irq_disable();
3657 head = sd->output_queue;
3658 sd->output_queue = NULL;
3659 sd->output_queue_tailp = &sd->output_queue;
3663 struct Qdisc *q = head;
3664 spinlock_t *root_lock;
3666 head = head->next_sched;
3668 root_lock = qdisc_lock(q);
3669 if (spin_trylock(root_lock)) {
3670 smp_mb__before_atomic();
3671 clear_bit(__QDISC_STATE_SCHED,
3674 spin_unlock(root_lock);
3676 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3678 __netif_reschedule(q);
3680 smp_mb__before_atomic();
3681 clear_bit(__QDISC_STATE_SCHED,
3689 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3690 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3691 /* This hook is defined here for ATM LANE */
3692 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3693 unsigned char *addr) __read_mostly;
3694 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3697 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3698 struct packet_type **pt_prev,
3699 int *ret, struct net_device *orig_dev)
3701 #ifdef CONFIG_NET_CLS_ACT
3702 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3703 struct tcf_result cl_res;
3705 /* If there's at least one ingress present somewhere (so
3706 * we get here via enabled static key), remaining devices
3707 * that are not configured with an ingress qdisc will bail
3713 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3717 qdisc_skb_cb(skb)->pkt_len = skb->len;
3718 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3719 qdisc_bstats_cpu_update(cl->q, skb);
3721 switch (tc_classify(skb, cl, &cl_res, false)) {
3723 case TC_ACT_RECLASSIFY:
3724 skb->tc_index = TC_H_MIN(cl_res.classid);
3727 qdisc_qstats_cpu_drop(cl->q);
3732 case TC_ACT_REDIRECT:
3733 /* skb_mac_header check was done by cls/act_bpf, so
3734 * we can safely push the L2 header back before
3735 * redirecting to another netdev
3737 __skb_push(skb, skb->mac_len);
3738 skb_do_redirect(skb);
3743 #endif /* CONFIG_NET_CLS_ACT */
3748 * netdev_is_rx_handler_busy - check if receive handler is registered
3749 * @dev: device to check
3751 * Check if a receive handler is already registered for a given device.
3752 * Return true if there one.
3754 * The caller must hold the rtnl_mutex.
3756 bool netdev_is_rx_handler_busy(struct net_device *dev)
3759 return dev && rtnl_dereference(dev->rx_handler);
3761 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
3764 * netdev_rx_handler_register - register receive handler
3765 * @dev: device to register a handler for
3766 * @rx_handler: receive handler to register
3767 * @rx_handler_data: data pointer that is used by rx handler
3769 * Register a receive handler for a device. This handler will then be
3770 * called from __netif_receive_skb. A negative errno code is returned
3773 * The caller must hold the rtnl_mutex.
3775 * For a general description of rx_handler, see enum rx_handler_result.
3777 int netdev_rx_handler_register(struct net_device *dev,
3778 rx_handler_func_t *rx_handler,
3779 void *rx_handler_data)
3783 if (dev->rx_handler)
3786 /* Note: rx_handler_data must be set before rx_handler */
3787 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3788 rcu_assign_pointer(dev->rx_handler, rx_handler);
3792 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3795 * netdev_rx_handler_unregister - unregister receive handler
3796 * @dev: device to unregister a handler from
3798 * Unregister a receive handler from a device.
3800 * The caller must hold the rtnl_mutex.
3802 void netdev_rx_handler_unregister(struct net_device *dev)
3806 RCU_INIT_POINTER(dev->rx_handler, NULL);
3807 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3808 * section has a guarantee to see a non NULL rx_handler_data
3812 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3814 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3817 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3818 * the special handling of PFMEMALLOC skbs.
3820 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3822 switch (skb->protocol) {
3823 case htons(ETH_P_ARP):
3824 case htons(ETH_P_IP):
3825 case htons(ETH_P_IPV6):
3826 case htons(ETH_P_8021Q):
3827 case htons(ETH_P_8021AD):
3834 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
3835 int *ret, struct net_device *orig_dev)
3837 #ifdef CONFIG_NETFILTER_INGRESS
3838 if (nf_hook_ingress_active(skb)) {
3840 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3844 return nf_hook_ingress(skb);
3846 #endif /* CONFIG_NETFILTER_INGRESS */
3850 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3852 struct packet_type *ptype, *pt_prev;
3853 rx_handler_func_t *rx_handler;
3854 struct net_device *orig_dev;
3855 bool deliver_exact = false;
3856 int ret = NET_RX_DROP;
3859 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3861 trace_netif_receive_skb(skb);
3863 orig_dev = skb->dev;
3865 skb_reset_network_header(skb);
3866 if (!skb_transport_header_was_set(skb))
3867 skb_reset_transport_header(skb);
3868 skb_reset_mac_len(skb);
3873 skb->skb_iif = skb->dev->ifindex;
3875 __this_cpu_inc(softnet_data.processed);
3877 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3878 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3879 skb = skb_vlan_untag(skb);
3884 #ifdef CONFIG_NET_CLS_ACT
3885 if (skb->tc_verd & TC_NCLS) {
3886 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3894 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3896 ret = deliver_skb(skb, pt_prev, orig_dev);
3900 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3902 ret = deliver_skb(skb, pt_prev, orig_dev);
3907 #ifdef CONFIG_NET_INGRESS
3908 if (static_key_false(&ingress_needed)) {
3909 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3913 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
3917 #ifdef CONFIG_NET_CLS_ACT
3921 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3924 if (skb_vlan_tag_present(skb)) {
3926 ret = deliver_skb(skb, pt_prev, orig_dev);
3929 if (vlan_do_receive(&skb))
3931 else if (unlikely(!skb))
3935 rx_handler = rcu_dereference(skb->dev->rx_handler);
3938 ret = deliver_skb(skb, pt_prev, orig_dev);
3941 switch (rx_handler(&skb)) {
3942 case RX_HANDLER_CONSUMED:
3943 ret = NET_RX_SUCCESS;
3945 case RX_HANDLER_ANOTHER:
3947 case RX_HANDLER_EXACT:
3948 deliver_exact = true;
3949 case RX_HANDLER_PASS:
3956 if (unlikely(skb_vlan_tag_present(skb))) {
3957 if (skb_vlan_tag_get_id(skb))
3958 skb->pkt_type = PACKET_OTHERHOST;
3959 /* Note: we might in the future use prio bits
3960 * and set skb->priority like in vlan_do_receive()
3961 * For the time being, just ignore Priority Code Point
3966 type = skb->protocol;
3968 /* deliver only exact match when indicated */
3969 if (likely(!deliver_exact)) {
3970 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3971 &ptype_base[ntohs(type) &
3975 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3976 &orig_dev->ptype_specific);
3978 if (unlikely(skb->dev != orig_dev)) {
3979 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3980 &skb->dev->ptype_specific);
3984 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3987 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3990 atomic_long_inc(&skb->dev->rx_dropped);
3992 /* Jamal, now you will not able to escape explaining
3993 * me how you were going to use this. :-)
4002 static int __netif_receive_skb(struct sk_buff *skb)
4006 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4007 unsigned long pflags = current->flags;
4010 * PFMEMALLOC skbs are special, they should
4011 * - be delivered to SOCK_MEMALLOC sockets only
4012 * - stay away from userspace
4013 * - have bounded memory usage
4015 * Use PF_MEMALLOC as this saves us from propagating the allocation
4016 * context down to all allocation sites.
4018 current->flags |= PF_MEMALLOC;
4019 ret = __netif_receive_skb_core(skb, true);
4020 tsk_restore_flags(current, pflags, PF_MEMALLOC);
4022 ret = __netif_receive_skb_core(skb, false);
4027 static int netif_receive_skb_internal(struct sk_buff *skb)
4031 net_timestamp_check(netdev_tstamp_prequeue, skb);
4033 if (skb_defer_rx_timestamp(skb))
4034 return NET_RX_SUCCESS;
4039 if (static_key_false(&rps_needed)) {
4040 struct rps_dev_flow voidflow, *rflow = &voidflow;
4041 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4044 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4050 ret = __netif_receive_skb(skb);
4056 * netif_receive_skb - process receive buffer from network
4057 * @skb: buffer to process
4059 * netif_receive_skb() is the main receive data processing function.
4060 * It always succeeds. The buffer may be dropped during processing
4061 * for congestion control or by the protocol layers.
4063 * This function may only be called from softirq context and interrupts
4064 * should be enabled.
4066 * Return values (usually ignored):
4067 * NET_RX_SUCCESS: no congestion
4068 * NET_RX_DROP: packet was dropped
4070 int netif_receive_skb(struct sk_buff *skb)
4072 trace_netif_receive_skb_entry(skb);
4074 return netif_receive_skb_internal(skb);
4076 EXPORT_SYMBOL(netif_receive_skb);
4078 /* Network device is going away, flush any packets still pending
4079 * Called with irqs disabled.
4081 static void flush_backlog(void *arg)
4083 struct net_device *dev = arg;
4084 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4085 struct sk_buff *skb, *tmp;
4088 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4089 if (skb->dev == dev) {
4090 __skb_unlink(skb, &sd->input_pkt_queue);
4092 input_queue_head_incr(sd);
4097 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4098 if (skb->dev == dev) {
4099 __skb_unlink(skb, &sd->process_queue);
4101 input_queue_head_incr(sd);
4106 static int napi_gro_complete(struct sk_buff *skb)
4108 struct packet_offload *ptype;
4109 __be16 type = skb->protocol;
4110 struct list_head *head = &offload_base;
4113 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4115 if (NAPI_GRO_CB(skb)->count == 1) {
4116 skb_shinfo(skb)->gso_size = 0;
4121 list_for_each_entry_rcu(ptype, head, list) {
4122 if (ptype->type != type || !ptype->callbacks.gro_complete)
4125 err = ptype->callbacks.gro_complete(skb, 0);
4131 WARN_ON(&ptype->list == head);
4133 return NET_RX_SUCCESS;
4137 return netif_receive_skb_internal(skb);
4140 /* napi->gro_list contains packets ordered by age.
4141 * youngest packets at the head of it.
4142 * Complete skbs in reverse order to reduce latencies.
4144 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4146 struct sk_buff *skb, *prev = NULL;
4148 /* scan list and build reverse chain */
4149 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4154 for (skb = prev; skb; skb = prev) {
4157 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4161 napi_gro_complete(skb);
4165 napi->gro_list = NULL;
4167 EXPORT_SYMBOL(napi_gro_flush);
4169 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4172 unsigned int maclen = skb->dev->hard_header_len;
4173 u32 hash = skb_get_hash_raw(skb);
4175 for (p = napi->gro_list; p; p = p->next) {
4176 unsigned long diffs;
4178 NAPI_GRO_CB(p)->flush = 0;
4180 if (hash != skb_get_hash_raw(p)) {
4181 NAPI_GRO_CB(p)->same_flow = 0;
4185 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4186 diffs |= p->vlan_tci ^ skb->vlan_tci;
4187 diffs |= skb_metadata_dst_cmp(p, skb);
4188 if (maclen == ETH_HLEN)
4189 diffs |= compare_ether_header(skb_mac_header(p),
4190 skb_mac_header(skb));
4192 diffs = memcmp(skb_mac_header(p),
4193 skb_mac_header(skb),
4195 NAPI_GRO_CB(p)->same_flow = !diffs;
4199 static void skb_gro_reset_offset(struct sk_buff *skb)
4201 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4202 const skb_frag_t *frag0 = &pinfo->frags[0];
4204 NAPI_GRO_CB(skb)->data_offset = 0;
4205 NAPI_GRO_CB(skb)->frag0 = NULL;
4206 NAPI_GRO_CB(skb)->frag0_len = 0;
4208 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4210 !PageHighMem(skb_frag_page(frag0))) {
4211 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4212 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
4213 skb_frag_size(frag0),
4214 skb->end - skb->tail);
4218 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4220 struct skb_shared_info *pinfo = skb_shinfo(skb);
4222 BUG_ON(skb->end - skb->tail < grow);
4224 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4226 skb->data_len -= grow;
4229 pinfo->frags[0].page_offset += grow;
4230 skb_frag_size_sub(&pinfo->frags[0], grow);
4232 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4233 skb_frag_unref(skb, 0);
4234 memmove(pinfo->frags, pinfo->frags + 1,
4235 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4239 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4241 struct sk_buff **pp = NULL;
4242 struct packet_offload *ptype;
4243 __be16 type = skb->protocol;
4244 struct list_head *head = &offload_base;
4246 enum gro_result ret;
4249 if (!(skb->dev->features & NETIF_F_GRO))
4252 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4255 gro_list_prepare(napi, skb);
4258 list_for_each_entry_rcu(ptype, head, list) {
4259 if (ptype->type != type || !ptype->callbacks.gro_receive)
4262 skb_set_network_header(skb, skb_gro_offset(skb));
4263 skb_reset_mac_len(skb);
4264 NAPI_GRO_CB(skb)->same_flow = 0;
4265 NAPI_GRO_CB(skb)->flush = 0;
4266 NAPI_GRO_CB(skb)->free = 0;
4267 NAPI_GRO_CB(skb)->encap_mark = 0;
4268 NAPI_GRO_CB(skb)->recursion_counter = 0;
4269 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4271 /* Setup for GRO checksum validation */
4272 switch (skb->ip_summed) {
4273 case CHECKSUM_COMPLETE:
4274 NAPI_GRO_CB(skb)->csum = skb->csum;
4275 NAPI_GRO_CB(skb)->csum_valid = 1;
4276 NAPI_GRO_CB(skb)->csum_cnt = 0;
4278 case CHECKSUM_UNNECESSARY:
4279 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4280 NAPI_GRO_CB(skb)->csum_valid = 0;
4283 NAPI_GRO_CB(skb)->csum_cnt = 0;
4284 NAPI_GRO_CB(skb)->csum_valid = 0;
4287 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4292 if (&ptype->list == head)
4295 same_flow = NAPI_GRO_CB(skb)->same_flow;
4296 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4299 struct sk_buff *nskb = *pp;
4303 napi_gro_complete(nskb);
4310 if (NAPI_GRO_CB(skb)->flush)
4313 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4314 struct sk_buff *nskb = napi->gro_list;
4316 /* locate the end of the list to select the 'oldest' flow */
4317 while (nskb->next) {
4323 napi_gro_complete(nskb);
4327 NAPI_GRO_CB(skb)->count = 1;
4328 NAPI_GRO_CB(skb)->age = jiffies;
4329 NAPI_GRO_CB(skb)->last = skb;
4330 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4331 skb->next = napi->gro_list;
4332 napi->gro_list = skb;
4336 grow = skb_gro_offset(skb) - skb_headlen(skb);
4338 gro_pull_from_frag0(skb, grow);
4347 struct packet_offload *gro_find_receive_by_type(__be16 type)
4349 struct list_head *offload_head = &offload_base;
4350 struct packet_offload *ptype;
4352 list_for_each_entry_rcu(ptype, offload_head, list) {
4353 if (ptype->type != type || !ptype->callbacks.gro_receive)
4359 EXPORT_SYMBOL(gro_find_receive_by_type);
4361 struct packet_offload *gro_find_complete_by_type(__be16 type)
4363 struct list_head *offload_head = &offload_base;
4364 struct packet_offload *ptype;
4366 list_for_each_entry_rcu(ptype, offload_head, list) {
4367 if (ptype->type != type || !ptype->callbacks.gro_complete)
4373 EXPORT_SYMBOL(gro_find_complete_by_type);
4375 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4379 if (netif_receive_skb_internal(skb))
4387 case GRO_MERGED_FREE:
4388 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) {
4390 kmem_cache_free(skbuff_head_cache, skb);
4404 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4406 trace_napi_gro_receive_entry(skb);
4408 skb_gro_reset_offset(skb);
4410 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4412 EXPORT_SYMBOL(napi_gro_receive);
4414 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4416 if (unlikely(skb->pfmemalloc)) {
4420 __skb_pull(skb, skb_headlen(skb));
4421 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4422 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4424 skb->dev = napi->dev;
4426 skb->encapsulation = 0;
4427 skb_shinfo(skb)->gso_type = 0;
4428 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4433 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4435 struct sk_buff *skb = napi->skb;
4438 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4443 EXPORT_SYMBOL(napi_get_frags);
4445 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4446 struct sk_buff *skb,
4452 __skb_push(skb, ETH_HLEN);
4453 skb->protocol = eth_type_trans(skb, skb->dev);
4454 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4459 case GRO_MERGED_FREE:
4460 napi_reuse_skb(napi, skb);
4470 /* Upper GRO stack assumes network header starts at gro_offset=0
4471 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4472 * We copy ethernet header into skb->data to have a common layout.
4474 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4476 struct sk_buff *skb = napi->skb;
4477 const struct ethhdr *eth;
4478 unsigned int hlen = sizeof(*eth);
4482 skb_reset_mac_header(skb);
4483 skb_gro_reset_offset(skb);
4485 eth = skb_gro_header_fast(skb, 0);
4486 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4487 eth = skb_gro_header_slow(skb, hlen, 0);
4488 if (unlikely(!eth)) {
4489 napi_reuse_skb(napi, skb);
4493 gro_pull_from_frag0(skb, hlen);
4494 NAPI_GRO_CB(skb)->frag0 += hlen;
4495 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4497 __skb_pull(skb, hlen);
4500 * This works because the only protocols we care about don't require
4502 * We'll fix it up properly in napi_frags_finish()
4504 skb->protocol = eth->h_proto;
4509 gro_result_t napi_gro_frags(struct napi_struct *napi)
4511 struct sk_buff *skb = napi_frags_skb(napi);
4516 trace_napi_gro_frags_entry(skb);
4518 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4520 EXPORT_SYMBOL(napi_gro_frags);
4522 /* Compute the checksum from gro_offset and return the folded value
4523 * after adding in any pseudo checksum.
4525 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4530 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4532 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4533 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4535 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4536 !skb->csum_complete_sw)
4537 netdev_rx_csum_fault(skb->dev);
4540 NAPI_GRO_CB(skb)->csum = wsum;
4541 NAPI_GRO_CB(skb)->csum_valid = 1;
4545 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4548 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4549 * Note: called with local irq disabled, but exits with local irq enabled.
4551 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4554 struct softnet_data *remsd = sd->rps_ipi_list;
4557 sd->rps_ipi_list = NULL;
4561 /* Send pending IPI's to kick RPS processing on remote cpus. */
4563 struct softnet_data *next = remsd->rps_ipi_next;
4565 if (cpu_online(remsd->cpu))
4566 smp_call_function_single_async(remsd->cpu,
4575 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4578 return sd->rps_ipi_list != NULL;
4584 static int process_backlog(struct napi_struct *napi, int quota)
4587 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4589 /* Check if we have pending ipi, its better to send them now,
4590 * not waiting net_rx_action() end.
4592 if (sd_has_rps_ipi_waiting(sd)) {
4593 local_irq_disable();
4594 net_rps_action_and_irq_enable(sd);
4597 napi->weight = weight_p;
4598 local_irq_disable();
4600 struct sk_buff *skb;
4602 while ((skb = __skb_dequeue(&sd->process_queue))) {
4605 __netif_receive_skb(skb);
4607 local_irq_disable();
4608 input_queue_head_incr(sd);
4609 if (++work >= quota) {
4616 if (skb_queue_empty(&sd->input_pkt_queue)) {
4618 * Inline a custom version of __napi_complete().
4619 * only current cpu owns and manipulates this napi,
4620 * and NAPI_STATE_SCHED is the only possible flag set
4622 * We can use a plain write instead of clear_bit(),
4623 * and we dont need an smp_mb() memory barrier.
4631 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4632 &sd->process_queue);
4641 * __napi_schedule - schedule for receive
4642 * @n: entry to schedule
4644 * The entry's receive function will be scheduled to run.
4645 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4647 void __napi_schedule(struct napi_struct *n)
4649 unsigned long flags;
4651 local_irq_save(flags);
4652 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4653 local_irq_restore(flags);
4655 EXPORT_SYMBOL(__napi_schedule);
4658 * __napi_schedule_irqoff - schedule for receive
4659 * @n: entry to schedule
4661 * Variant of __napi_schedule() assuming hard irqs are masked
4663 void __napi_schedule_irqoff(struct napi_struct *n)
4665 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4667 EXPORT_SYMBOL(__napi_schedule_irqoff);
4669 void __napi_complete(struct napi_struct *n)
4671 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4673 list_del_init(&n->poll_list);
4674 smp_mb__before_atomic();
4675 clear_bit(NAPI_STATE_SCHED, &n->state);
4677 EXPORT_SYMBOL(__napi_complete);
4679 void napi_complete_done(struct napi_struct *n, int work_done)
4681 unsigned long flags;
4684 * don't let napi dequeue from the cpu poll list
4685 * just in case its running on a different cpu
4687 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4691 unsigned long timeout = 0;
4694 timeout = n->dev->gro_flush_timeout;
4697 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4698 HRTIMER_MODE_REL_PINNED);
4700 napi_gro_flush(n, false);
4702 if (likely(list_empty(&n->poll_list))) {
4703 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4705 /* If n->poll_list is not empty, we need to mask irqs */
4706 local_irq_save(flags);
4708 local_irq_restore(flags);
4711 EXPORT_SYMBOL(napi_complete_done);
4713 /* must be called under rcu_read_lock(), as we dont take a reference */
4714 struct napi_struct *napi_by_id(unsigned int napi_id)
4716 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4717 struct napi_struct *napi;
4719 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4720 if (napi->napi_id == napi_id)
4725 EXPORT_SYMBOL_GPL(napi_by_id);
4727 void napi_hash_add(struct napi_struct *napi)
4729 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4731 spin_lock(&napi_hash_lock);
4733 /* 0 is not a valid id, we also skip an id that is taken
4734 * we expect both events to be extremely rare
4737 while (!napi->napi_id) {
4738 napi->napi_id = ++napi_gen_id;
4739 if (napi_by_id(napi->napi_id))
4743 hlist_add_head_rcu(&napi->napi_hash_node,
4744 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4746 spin_unlock(&napi_hash_lock);
4749 EXPORT_SYMBOL_GPL(napi_hash_add);
4751 /* Warning : caller is responsible to make sure rcu grace period
4752 * is respected before freeing memory containing @napi
4754 void napi_hash_del(struct napi_struct *napi)
4756 spin_lock(&napi_hash_lock);
4758 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4759 hlist_del_rcu(&napi->napi_hash_node);
4761 spin_unlock(&napi_hash_lock);
4763 EXPORT_SYMBOL_GPL(napi_hash_del);
4765 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4767 struct napi_struct *napi;
4769 napi = container_of(timer, struct napi_struct, timer);
4771 napi_schedule(napi);
4773 return HRTIMER_NORESTART;
4776 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4777 int (*poll)(struct napi_struct *, int), int weight)
4779 INIT_LIST_HEAD(&napi->poll_list);
4780 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4781 napi->timer.function = napi_watchdog;
4782 napi->gro_count = 0;
4783 napi->gro_list = NULL;
4786 if (weight > NAPI_POLL_WEIGHT)
4787 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4789 napi->weight = weight;
4790 list_add(&napi->dev_list, &dev->napi_list);
4792 #ifdef CONFIG_NETPOLL
4793 spin_lock_init(&napi->poll_lock);
4794 napi->poll_owner = -1;
4796 set_bit(NAPI_STATE_SCHED, &napi->state);
4798 EXPORT_SYMBOL(netif_napi_add);
4800 void napi_disable(struct napi_struct *n)
4803 set_bit(NAPI_STATE_DISABLE, &n->state);
4805 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4807 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
4810 hrtimer_cancel(&n->timer);
4812 clear_bit(NAPI_STATE_DISABLE, &n->state);
4814 EXPORT_SYMBOL(napi_disable);
4816 void netif_napi_del(struct napi_struct *napi)
4818 list_del_init(&napi->dev_list);
4819 napi_free_frags(napi);
4821 kfree_skb_list(napi->gro_list);
4822 napi->gro_list = NULL;
4823 napi->gro_count = 0;
4825 EXPORT_SYMBOL(netif_napi_del);
4827 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4832 list_del_init(&n->poll_list);
4834 have = netpoll_poll_lock(n);
4838 /* This NAPI_STATE_SCHED test is for avoiding a race
4839 * with netpoll's poll_napi(). Only the entity which
4840 * obtains the lock and sees NAPI_STATE_SCHED set will
4841 * actually make the ->poll() call. Therefore we avoid
4842 * accidentally calling ->poll() when NAPI is not scheduled.
4845 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4846 work = n->poll(n, weight);
4850 WARN_ON_ONCE(work > weight);
4852 if (likely(work < weight))
4855 /* Drivers must not modify the NAPI state if they
4856 * consume the entire weight. In such cases this code
4857 * still "owns" the NAPI instance and therefore can
4858 * move the instance around on the list at-will.
4860 if (unlikely(napi_disable_pending(n))) {
4866 /* flush too old packets
4867 * If HZ < 1000, flush all packets.
4869 napi_gro_flush(n, HZ >= 1000);
4872 /* Some drivers may have called napi_schedule
4873 * prior to exhausting their budget.
4875 if (unlikely(!list_empty(&n->poll_list))) {
4876 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4877 n->dev ? n->dev->name : "backlog");
4881 list_add_tail(&n->poll_list, repoll);
4884 netpoll_poll_unlock(have);
4889 static void net_rx_action(struct softirq_action *h)
4891 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4892 unsigned long time_limit = jiffies + 2;
4893 int budget = netdev_budget;
4897 local_irq_disable();
4898 list_splice_init(&sd->poll_list, &list);
4902 struct napi_struct *n;
4904 if (list_empty(&list)) {
4905 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4910 n = list_first_entry(&list, struct napi_struct, poll_list);
4911 budget -= napi_poll(n, &repoll);
4913 /* If softirq window is exhausted then punt.
4914 * Allow this to run for 2 jiffies since which will allow
4915 * an average latency of 1.5/HZ.
4917 if (unlikely(budget <= 0 ||
4918 time_after_eq(jiffies, time_limit))) {
4924 local_irq_disable();
4926 list_splice_tail_init(&sd->poll_list, &list);
4927 list_splice_tail(&repoll, &list);
4928 list_splice(&list, &sd->poll_list);
4929 if (!list_empty(&sd->poll_list))
4930 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4932 net_rps_action_and_irq_enable(sd);
4935 struct netdev_adjacent {
4936 struct net_device *dev;
4938 /* upper master flag, there can only be one master device per list */
4941 /* counter for the number of times this device was added to us */
4944 /* private field for the users */
4947 struct list_head list;
4948 struct rcu_head rcu;
4951 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
4952 struct list_head *adj_list)
4954 struct netdev_adjacent *adj;
4956 list_for_each_entry(adj, adj_list, list) {
4957 if (adj->dev == adj_dev)
4964 * netdev_has_upper_dev - Check if device is linked to an upper device
4966 * @upper_dev: upper device to check
4968 * Find out if a device is linked to specified upper device and return true
4969 * in case it is. Note that this checks only immediate upper device,
4970 * not through a complete stack of devices. The caller must hold the RTNL lock.
4972 bool netdev_has_upper_dev(struct net_device *dev,
4973 struct net_device *upper_dev)
4977 return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper);
4979 EXPORT_SYMBOL(netdev_has_upper_dev);
4982 * netdev_has_any_upper_dev - Check if device is linked to some device
4985 * Find out if a device is linked to an upper device and return true in case
4986 * it is. The caller must hold the RTNL lock.
4988 static bool netdev_has_any_upper_dev(struct net_device *dev)
4992 return !list_empty(&dev->all_adj_list.upper);
4996 * netdev_master_upper_dev_get - Get master upper device
4999 * Find a master upper device and return pointer to it or NULL in case
5000 * it's not there. The caller must hold the RTNL lock.
5002 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5004 struct netdev_adjacent *upper;
5008 if (list_empty(&dev->adj_list.upper))
5011 upper = list_first_entry(&dev->adj_list.upper,
5012 struct netdev_adjacent, list);
5013 if (likely(upper->master))
5017 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5019 void *netdev_adjacent_get_private(struct list_head *adj_list)
5021 struct netdev_adjacent *adj;
5023 adj = list_entry(adj_list, struct netdev_adjacent, list);
5025 return adj->private;
5027 EXPORT_SYMBOL(netdev_adjacent_get_private);
5030 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5032 * @iter: list_head ** of the current position
5034 * Gets the next device from the dev's upper list, starting from iter
5035 * position. The caller must hold RCU read lock.
5037 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5038 struct list_head **iter)
5040 struct netdev_adjacent *upper;
5042 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5044 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5046 if (&upper->list == &dev->adj_list.upper)
5049 *iter = &upper->list;
5053 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5056 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
5058 * @iter: list_head ** of the current position
5060 * Gets the next device from the dev's upper list, starting from iter
5061 * position. The caller must hold RCU read lock.
5063 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
5064 struct list_head **iter)
5066 struct netdev_adjacent *upper;
5068 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5070 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5072 if (&upper->list == &dev->all_adj_list.upper)
5075 *iter = &upper->list;
5079 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
5082 * netdev_lower_get_next_private - Get the next ->private from the
5083 * lower neighbour list
5085 * @iter: list_head ** of the current position
5087 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5088 * list, starting from iter position. The caller must hold either hold the
5089 * RTNL lock or its own locking that guarantees that the neighbour lower
5090 * list will remain unchanged.
5092 void *netdev_lower_get_next_private(struct net_device *dev,
5093 struct list_head **iter)
5095 struct netdev_adjacent *lower;
5097 lower = list_entry(*iter, struct netdev_adjacent, list);
5099 if (&lower->list == &dev->adj_list.lower)
5102 *iter = lower->list.next;
5104 return lower->private;
5106 EXPORT_SYMBOL(netdev_lower_get_next_private);
5109 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5110 * lower neighbour list, RCU
5113 * @iter: list_head ** of the current position
5115 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5116 * list, starting from iter position. The caller must hold RCU read lock.
5118 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5119 struct list_head **iter)
5121 struct netdev_adjacent *lower;
5123 WARN_ON_ONCE(!rcu_read_lock_held());
5125 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5127 if (&lower->list == &dev->adj_list.lower)
5130 *iter = &lower->list;
5132 return lower->private;
5134 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5137 * netdev_lower_get_next - Get the next device from the lower neighbour
5140 * @iter: list_head ** of the current position
5142 * Gets the next netdev_adjacent from the dev's lower neighbour
5143 * list, starting from iter position. The caller must hold RTNL lock or
5144 * its own locking that guarantees that the neighbour lower
5145 * list will remain unchanged.
5147 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5149 struct netdev_adjacent *lower;
5151 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5153 if (&lower->list == &dev->adj_list.lower)
5156 *iter = &lower->list;
5160 EXPORT_SYMBOL(netdev_lower_get_next);
5163 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5164 * lower neighbour list, RCU
5168 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5169 * list. The caller must hold RCU read lock.
5171 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5173 struct netdev_adjacent *lower;
5175 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5176 struct netdev_adjacent, list);
5178 return lower->private;
5181 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5184 * netdev_master_upper_dev_get_rcu - Get master upper device
5187 * Find a master upper device and return pointer to it or NULL in case
5188 * it's not there. The caller must hold the RCU read lock.
5190 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5192 struct netdev_adjacent *upper;
5194 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5195 struct netdev_adjacent, list);
5196 if (upper && likely(upper->master))
5200 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5202 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5203 struct net_device *adj_dev,
5204 struct list_head *dev_list)
5206 char linkname[IFNAMSIZ+7];
5207 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5208 "upper_%s" : "lower_%s", adj_dev->name);
5209 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5212 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5214 struct list_head *dev_list)
5216 char linkname[IFNAMSIZ+7];
5217 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5218 "upper_%s" : "lower_%s", name);
5219 sysfs_remove_link(&(dev->dev.kobj), linkname);
5222 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5223 struct net_device *adj_dev,
5224 struct list_head *dev_list)
5226 return (dev_list == &dev->adj_list.upper ||
5227 dev_list == &dev->adj_list.lower) &&
5228 net_eq(dev_net(dev), dev_net(adj_dev));
5231 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5232 struct net_device *adj_dev,
5234 struct list_head *dev_list,
5235 void *private, bool master)
5237 struct netdev_adjacent *adj;
5240 adj = __netdev_find_adj(adj_dev, dev_list);
5243 adj->ref_nr += ref_nr;
5247 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5252 adj->master = master;
5253 adj->ref_nr = ref_nr;
5254 adj->private = private;
5257 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5258 adj_dev->name, dev->name, adj_dev->name);
5260 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5261 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5266 /* Ensure that master link is always the first item in list. */
5268 ret = sysfs_create_link(&(dev->dev.kobj),
5269 &(adj_dev->dev.kobj), "master");
5271 goto remove_symlinks;
5273 list_add_rcu(&adj->list, dev_list);
5275 list_add_tail_rcu(&adj->list, dev_list);
5281 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5282 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5290 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5291 struct net_device *adj_dev,
5293 struct list_head *dev_list)
5295 struct netdev_adjacent *adj;
5297 adj = __netdev_find_adj(adj_dev, dev_list);
5300 pr_err("tried to remove device %s from %s\n",
5301 dev->name, adj_dev->name);
5305 if (adj->ref_nr > ref_nr) {
5306 pr_debug("%s to %s ref_nr-%d = %d\n", dev->name, adj_dev->name,
5307 ref_nr, adj->ref_nr-ref_nr);
5308 adj->ref_nr -= ref_nr;
5313 sysfs_remove_link(&(dev->dev.kobj), "master");
5315 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5316 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5318 list_del_rcu(&adj->list);
5319 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5320 adj_dev->name, dev->name, adj_dev->name);
5322 kfree_rcu(adj, rcu);
5325 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5326 struct net_device *upper_dev,
5328 struct list_head *up_list,
5329 struct list_head *down_list,
5330 void *private, bool master)
5334 ret = __netdev_adjacent_dev_insert(dev, upper_dev, ref_nr, up_list,
5339 ret = __netdev_adjacent_dev_insert(upper_dev, dev, ref_nr, down_list,
5342 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
5349 static int __netdev_adjacent_dev_link(struct net_device *dev,
5350 struct net_device *upper_dev,
5353 return __netdev_adjacent_dev_link_lists(dev, upper_dev, ref_nr,
5354 &dev->all_adj_list.upper,
5355 &upper_dev->all_adj_list.lower,
5359 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5360 struct net_device *upper_dev,
5362 struct list_head *up_list,
5363 struct list_head *down_list)
5365 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
5366 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
5369 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5370 struct net_device *upper_dev,
5373 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, ref_nr,
5374 &dev->all_adj_list.upper,
5375 &upper_dev->all_adj_list.lower);
5378 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5379 struct net_device *upper_dev,
5380 void *private, bool master)
5382 int ret = __netdev_adjacent_dev_link(dev, upper_dev, 1);
5387 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, 1,
5388 &dev->adj_list.upper,
5389 &upper_dev->adj_list.lower,
5392 __netdev_adjacent_dev_unlink(dev, upper_dev, 1);
5399 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5400 struct net_device *upper_dev)
5402 __netdev_adjacent_dev_unlink(dev, upper_dev, 1);
5403 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
5404 &dev->adj_list.upper,
5405 &upper_dev->adj_list.lower);
5408 static int __netdev_upper_dev_link(struct net_device *dev,
5409 struct net_device *upper_dev, bool master,
5412 struct netdev_notifier_changeupper_info changeupper_info;
5413 struct netdev_adjacent *i, *j, *to_i, *to_j;
5418 if (dev == upper_dev)
5421 /* To prevent loops, check if dev is not upper device to upper_dev. */
5422 if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper))
5425 if (__netdev_find_adj(upper_dev, &dev->adj_list.upper))
5428 if (master && netdev_master_upper_dev_get(dev))
5431 changeupper_info.upper_dev = upper_dev;
5432 changeupper_info.master = master;
5433 changeupper_info.linking = true;
5435 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5436 &changeupper_info.info);
5437 ret = notifier_to_errno(ret);
5441 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5446 /* Now that we linked these devs, make all the upper_dev's
5447 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5448 * versa, and don't forget the devices itself. All of these
5449 * links are non-neighbours.
5451 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5452 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5453 pr_debug("Interlinking %s with %s, non-neighbour\n",
5454 i->dev->name, j->dev->name);
5455 ret = __netdev_adjacent_dev_link(i->dev, j->dev, i->ref_nr);
5461 /* add dev to every upper_dev's upper device */
5462 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5463 pr_debug("linking %s's upper device %s with %s\n",
5464 upper_dev->name, i->dev->name, dev->name);
5465 ret = __netdev_adjacent_dev_link(dev, i->dev, i->ref_nr);
5467 goto rollback_upper_mesh;
5470 /* add upper_dev to every dev's lower device */
5471 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5472 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5473 i->dev->name, upper_dev->name);
5474 ret = __netdev_adjacent_dev_link(i->dev, upper_dev, i->ref_nr);
5476 goto rollback_lower_mesh;
5479 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5480 &changeupper_info.info);
5483 rollback_lower_mesh:
5485 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5488 __netdev_adjacent_dev_unlink(i->dev, upper_dev, i->ref_nr);
5493 rollback_upper_mesh:
5495 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5498 __netdev_adjacent_dev_unlink(dev, i->dev, i->ref_nr);
5506 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5507 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5508 if (i == to_i && j == to_j)
5510 __netdev_adjacent_dev_unlink(i->dev, j->dev, i->ref_nr);
5516 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5522 * netdev_upper_dev_link - Add a link to the upper device
5524 * @upper_dev: new upper device
5526 * Adds a link to device which is upper to this one. The caller must hold
5527 * the RTNL lock. On a failure a negative errno code is returned.
5528 * On success the reference counts are adjusted and the function
5531 int netdev_upper_dev_link(struct net_device *dev,
5532 struct net_device *upper_dev)
5534 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5536 EXPORT_SYMBOL(netdev_upper_dev_link);
5539 * netdev_master_upper_dev_link - Add a master link to the upper device
5541 * @upper_dev: new upper device
5543 * Adds a link to device which is upper to this one. In this case, only
5544 * one master upper device can be linked, although other non-master devices
5545 * might be linked as well. The caller must hold the RTNL lock.
5546 * On a failure a negative errno code is returned. On success the reference
5547 * counts are adjusted and the function returns zero.
5549 int netdev_master_upper_dev_link(struct net_device *dev,
5550 struct net_device *upper_dev)
5552 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5554 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5556 int netdev_master_upper_dev_link_private(struct net_device *dev,
5557 struct net_device *upper_dev,
5560 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5562 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5565 * netdev_upper_dev_unlink - Removes a link to upper device
5567 * @upper_dev: new upper device
5569 * Removes a link to device which is upper to this one. The caller must hold
5572 void netdev_upper_dev_unlink(struct net_device *dev,
5573 struct net_device *upper_dev)
5575 struct netdev_notifier_changeupper_info changeupper_info;
5576 struct netdev_adjacent *i, *j;
5579 changeupper_info.upper_dev = upper_dev;
5580 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
5581 changeupper_info.linking = false;
5583 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5584 &changeupper_info.info);
5586 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5588 /* Here is the tricky part. We must remove all dev's lower
5589 * devices from all upper_dev's upper devices and vice
5590 * versa, to maintain the graph relationship.
5592 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5593 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5594 __netdev_adjacent_dev_unlink(i->dev, j->dev, i->ref_nr);
5596 /* remove also the devices itself from lower/upper device
5599 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5600 __netdev_adjacent_dev_unlink(i->dev, upper_dev, i->ref_nr);
5602 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5603 __netdev_adjacent_dev_unlink(dev, i->dev, i->ref_nr);
5605 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5606 &changeupper_info.info);
5608 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5611 * netdev_bonding_info_change - Dispatch event about slave change
5613 * @bonding_info: info to dispatch
5615 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5616 * The caller must hold the RTNL lock.
5618 void netdev_bonding_info_change(struct net_device *dev,
5619 struct netdev_bonding_info *bonding_info)
5621 struct netdev_notifier_bonding_info info;
5623 memcpy(&info.bonding_info, bonding_info,
5624 sizeof(struct netdev_bonding_info));
5625 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5628 EXPORT_SYMBOL(netdev_bonding_info_change);
5630 static void netdev_adjacent_add_links(struct net_device *dev)
5632 struct netdev_adjacent *iter;
5634 struct net *net = dev_net(dev);
5636 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5637 if (!net_eq(net,dev_net(iter->dev)))
5639 netdev_adjacent_sysfs_add(iter->dev, dev,
5640 &iter->dev->adj_list.lower);
5641 netdev_adjacent_sysfs_add(dev, iter->dev,
5642 &dev->adj_list.upper);
5645 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5646 if (!net_eq(net,dev_net(iter->dev)))
5648 netdev_adjacent_sysfs_add(iter->dev, dev,
5649 &iter->dev->adj_list.upper);
5650 netdev_adjacent_sysfs_add(dev, iter->dev,
5651 &dev->adj_list.lower);
5655 static void netdev_adjacent_del_links(struct net_device *dev)
5657 struct netdev_adjacent *iter;
5659 struct net *net = dev_net(dev);
5661 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5662 if (!net_eq(net,dev_net(iter->dev)))
5664 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5665 &iter->dev->adj_list.lower);
5666 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5667 &dev->adj_list.upper);
5670 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5671 if (!net_eq(net,dev_net(iter->dev)))
5673 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5674 &iter->dev->adj_list.upper);
5675 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5676 &dev->adj_list.lower);
5680 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5682 struct netdev_adjacent *iter;
5684 struct net *net = dev_net(dev);
5686 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5687 if (!net_eq(net,dev_net(iter->dev)))
5689 netdev_adjacent_sysfs_del(iter->dev, oldname,
5690 &iter->dev->adj_list.lower);
5691 netdev_adjacent_sysfs_add(iter->dev, dev,
5692 &iter->dev->adj_list.lower);
5695 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5696 if (!net_eq(net,dev_net(iter->dev)))
5698 netdev_adjacent_sysfs_del(iter->dev, oldname,
5699 &iter->dev->adj_list.upper);
5700 netdev_adjacent_sysfs_add(iter->dev, dev,
5701 &iter->dev->adj_list.upper);
5705 void *netdev_lower_dev_get_private(struct net_device *dev,
5706 struct net_device *lower_dev)
5708 struct netdev_adjacent *lower;
5712 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
5716 return lower->private;
5718 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5721 int dev_get_nest_level(struct net_device *dev,
5722 bool (*type_check)(struct net_device *dev))
5724 struct net_device *lower = NULL;
5725 struct list_head *iter;
5731 netdev_for_each_lower_dev(dev, lower, iter) {
5732 nest = dev_get_nest_level(lower, type_check);
5733 if (max_nest < nest)
5737 if (type_check(dev))
5742 EXPORT_SYMBOL(dev_get_nest_level);
5744 static void dev_change_rx_flags(struct net_device *dev, int flags)
5746 const struct net_device_ops *ops = dev->netdev_ops;
5748 if (ops->ndo_change_rx_flags)
5749 ops->ndo_change_rx_flags(dev, flags);
5752 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5754 unsigned int old_flags = dev->flags;
5760 dev->flags |= IFF_PROMISC;
5761 dev->promiscuity += inc;
5762 if (dev->promiscuity == 0) {
5765 * If inc causes overflow, untouch promisc and return error.
5768 dev->flags &= ~IFF_PROMISC;
5770 dev->promiscuity -= inc;
5771 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5776 if (dev->flags != old_flags) {
5777 pr_info("device %s %s promiscuous mode\n",
5779 dev->flags & IFF_PROMISC ? "entered" : "left");
5780 if (audit_enabled) {
5781 current_uid_gid(&uid, &gid);
5782 audit_log(current->audit_context, GFP_ATOMIC,
5783 AUDIT_ANOM_PROMISCUOUS,
5784 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5785 dev->name, (dev->flags & IFF_PROMISC),
5786 (old_flags & IFF_PROMISC),
5787 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5788 from_kuid(&init_user_ns, uid),
5789 from_kgid(&init_user_ns, gid),
5790 audit_get_sessionid(current));
5793 dev_change_rx_flags(dev, IFF_PROMISC);
5796 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5801 * dev_set_promiscuity - update promiscuity count on a device
5805 * Add or remove promiscuity from a device. While the count in the device
5806 * remains above zero the interface remains promiscuous. Once it hits zero
5807 * the device reverts back to normal filtering operation. A negative inc
5808 * value is used to drop promiscuity on the device.
5809 * Return 0 if successful or a negative errno code on error.
5811 int dev_set_promiscuity(struct net_device *dev, int inc)
5813 unsigned int old_flags = dev->flags;
5816 err = __dev_set_promiscuity(dev, inc, true);
5819 if (dev->flags != old_flags)
5820 dev_set_rx_mode(dev);
5823 EXPORT_SYMBOL(dev_set_promiscuity);
5825 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5827 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5831 dev->flags |= IFF_ALLMULTI;
5832 dev->allmulti += inc;
5833 if (dev->allmulti == 0) {
5836 * If inc causes overflow, untouch allmulti and return error.
5839 dev->flags &= ~IFF_ALLMULTI;
5841 dev->allmulti -= inc;
5842 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5847 if (dev->flags ^ old_flags) {
5848 dev_change_rx_flags(dev, IFF_ALLMULTI);
5849 dev_set_rx_mode(dev);
5851 __dev_notify_flags(dev, old_flags,
5852 dev->gflags ^ old_gflags);
5858 * dev_set_allmulti - update allmulti count on a device
5862 * Add or remove reception of all multicast frames to a device. While the
5863 * count in the device remains above zero the interface remains listening
5864 * to all interfaces. Once it hits zero the device reverts back to normal
5865 * filtering operation. A negative @inc value is used to drop the counter
5866 * when releasing a resource needing all multicasts.
5867 * Return 0 if successful or a negative errno code on error.
5870 int dev_set_allmulti(struct net_device *dev, int inc)
5872 return __dev_set_allmulti(dev, inc, true);
5874 EXPORT_SYMBOL(dev_set_allmulti);
5877 * Upload unicast and multicast address lists to device and
5878 * configure RX filtering. When the device doesn't support unicast
5879 * filtering it is put in promiscuous mode while unicast addresses
5882 void __dev_set_rx_mode(struct net_device *dev)
5884 const struct net_device_ops *ops = dev->netdev_ops;
5886 /* dev_open will call this function so the list will stay sane. */
5887 if (!(dev->flags&IFF_UP))
5890 if (!netif_device_present(dev))
5893 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5894 /* Unicast addresses changes may only happen under the rtnl,
5895 * therefore calling __dev_set_promiscuity here is safe.
5897 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5898 __dev_set_promiscuity(dev, 1, false);
5899 dev->uc_promisc = true;
5900 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5901 __dev_set_promiscuity(dev, -1, false);
5902 dev->uc_promisc = false;
5906 if (ops->ndo_set_rx_mode)
5907 ops->ndo_set_rx_mode(dev);
5910 void dev_set_rx_mode(struct net_device *dev)
5912 netif_addr_lock_bh(dev);
5913 __dev_set_rx_mode(dev);
5914 netif_addr_unlock_bh(dev);
5918 * dev_get_flags - get flags reported to userspace
5921 * Get the combination of flag bits exported through APIs to userspace.
5923 unsigned int dev_get_flags(const struct net_device *dev)
5927 flags = (dev->flags & ~(IFF_PROMISC |
5932 (dev->gflags & (IFF_PROMISC |
5935 if (netif_running(dev)) {
5936 if (netif_oper_up(dev))
5937 flags |= IFF_RUNNING;
5938 if (netif_carrier_ok(dev))
5939 flags |= IFF_LOWER_UP;
5940 if (netif_dormant(dev))
5941 flags |= IFF_DORMANT;
5946 EXPORT_SYMBOL(dev_get_flags);
5948 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5950 unsigned int old_flags = dev->flags;
5956 * Set the flags on our device.
5959 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5960 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5962 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5966 * Load in the correct multicast list now the flags have changed.
5969 if ((old_flags ^ flags) & IFF_MULTICAST)
5970 dev_change_rx_flags(dev, IFF_MULTICAST);
5972 dev_set_rx_mode(dev);
5975 * Have we downed the interface. We handle IFF_UP ourselves
5976 * according to user attempts to set it, rather than blindly
5981 if ((old_flags ^ flags) & IFF_UP)
5982 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5984 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5985 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5986 unsigned int old_flags = dev->flags;
5988 dev->gflags ^= IFF_PROMISC;
5990 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5991 if (dev->flags != old_flags)
5992 dev_set_rx_mode(dev);
5995 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5996 is important. Some (broken) drivers set IFF_PROMISC, when
5997 IFF_ALLMULTI is requested not asking us and not reporting.
5999 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
6000 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
6002 dev->gflags ^= IFF_ALLMULTI;
6003 __dev_set_allmulti(dev, inc, false);
6009 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
6010 unsigned int gchanges)
6012 unsigned int changes = dev->flags ^ old_flags;
6015 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
6017 if (changes & IFF_UP) {
6018 if (dev->flags & IFF_UP)
6019 call_netdevice_notifiers(NETDEV_UP, dev);
6021 call_netdevice_notifiers(NETDEV_DOWN, dev);
6024 if (dev->flags & IFF_UP &&
6025 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
6026 struct netdev_notifier_change_info change_info;
6028 change_info.flags_changed = changes;
6029 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6035 * dev_change_flags - change device settings
6037 * @flags: device state flags
6039 * Change settings on device based state flags. The flags are
6040 * in the userspace exported format.
6042 int dev_change_flags(struct net_device *dev, unsigned int flags)
6045 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6047 ret = __dev_change_flags(dev, flags);
6051 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6052 __dev_notify_flags(dev, old_flags, changes);
6055 EXPORT_SYMBOL(dev_change_flags);
6057 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6059 const struct net_device_ops *ops = dev->netdev_ops;
6061 if (ops->ndo_change_mtu)
6062 return ops->ndo_change_mtu(dev, new_mtu);
6069 * dev_set_mtu - Change maximum transfer unit
6071 * @new_mtu: new transfer unit
6073 * Change the maximum transfer size of the network device.
6075 int dev_set_mtu(struct net_device *dev, int new_mtu)
6079 if (new_mtu == dev->mtu)
6082 /* MTU must be positive. */
6086 if (!netif_device_present(dev))
6089 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6090 err = notifier_to_errno(err);
6094 orig_mtu = dev->mtu;
6095 err = __dev_set_mtu(dev, new_mtu);
6098 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6099 err = notifier_to_errno(err);
6101 /* setting mtu back and notifying everyone again,
6102 * so that they have a chance to revert changes.
6104 __dev_set_mtu(dev, orig_mtu);
6105 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6110 EXPORT_SYMBOL(dev_set_mtu);
6113 * dev_set_group - Change group this device belongs to
6115 * @new_group: group this device should belong to
6117 void dev_set_group(struct net_device *dev, int new_group)
6119 dev->group = new_group;
6121 EXPORT_SYMBOL(dev_set_group);
6124 * dev_set_mac_address - Change Media Access Control Address
6128 * Change the hardware (MAC) address of the device
6130 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6132 const struct net_device_ops *ops = dev->netdev_ops;
6135 if (!ops->ndo_set_mac_address)
6137 if (sa->sa_family != dev->type)
6139 if (!netif_device_present(dev))
6141 err = ops->ndo_set_mac_address(dev, sa);
6144 dev->addr_assign_type = NET_ADDR_SET;
6145 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6146 add_device_randomness(dev->dev_addr, dev->addr_len);
6149 EXPORT_SYMBOL(dev_set_mac_address);
6152 * dev_change_carrier - Change device carrier
6154 * @new_carrier: new value
6156 * Change device carrier
6158 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6160 const struct net_device_ops *ops = dev->netdev_ops;
6162 if (!ops->ndo_change_carrier)
6164 if (!netif_device_present(dev))
6166 return ops->ndo_change_carrier(dev, new_carrier);
6168 EXPORT_SYMBOL(dev_change_carrier);
6171 * dev_get_phys_port_id - Get device physical port ID
6175 * Get device physical port ID
6177 int dev_get_phys_port_id(struct net_device *dev,
6178 struct netdev_phys_item_id *ppid)
6180 const struct net_device_ops *ops = dev->netdev_ops;
6182 if (!ops->ndo_get_phys_port_id)
6184 return ops->ndo_get_phys_port_id(dev, ppid);
6186 EXPORT_SYMBOL(dev_get_phys_port_id);
6189 * dev_get_phys_port_name - Get device physical port name
6193 * Get device physical port name
6195 int dev_get_phys_port_name(struct net_device *dev,
6196 char *name, size_t len)
6198 const struct net_device_ops *ops = dev->netdev_ops;
6200 if (!ops->ndo_get_phys_port_name)
6202 return ops->ndo_get_phys_port_name(dev, name, len);
6204 EXPORT_SYMBOL(dev_get_phys_port_name);
6207 * dev_change_proto_down - update protocol port state information
6209 * @proto_down: new value
6211 * This info can be used by switch drivers to set the phys state of the
6214 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6216 const struct net_device_ops *ops = dev->netdev_ops;
6218 if (!ops->ndo_change_proto_down)
6220 if (!netif_device_present(dev))
6222 return ops->ndo_change_proto_down(dev, proto_down);
6224 EXPORT_SYMBOL(dev_change_proto_down);
6227 * dev_new_index - allocate an ifindex
6228 * @net: the applicable net namespace
6230 * Returns a suitable unique value for a new device interface
6231 * number. The caller must hold the rtnl semaphore or the
6232 * dev_base_lock to be sure it remains unique.
6234 static int dev_new_index(struct net *net)
6236 int ifindex = net->ifindex;
6240 if (!__dev_get_by_index(net, ifindex))
6241 return net->ifindex = ifindex;
6245 /* Delayed registration/unregisteration */
6246 static LIST_HEAD(net_todo_list);
6247 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6249 static void net_set_todo(struct net_device *dev)
6251 list_add_tail(&dev->todo_list, &net_todo_list);
6252 dev_net(dev)->dev_unreg_count++;
6255 static void rollback_registered_many(struct list_head *head)
6257 struct net_device *dev, *tmp;
6258 LIST_HEAD(close_head);
6260 BUG_ON(dev_boot_phase);
6263 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6264 /* Some devices call without registering
6265 * for initialization unwind. Remove those
6266 * devices and proceed with the remaining.
6268 if (dev->reg_state == NETREG_UNINITIALIZED) {
6269 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6273 list_del(&dev->unreg_list);
6276 dev->dismantle = true;
6277 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6280 /* If device is running, close it first. */
6281 list_for_each_entry(dev, head, unreg_list)
6282 list_add_tail(&dev->close_list, &close_head);
6283 dev_close_many(&close_head, true);
6285 list_for_each_entry(dev, head, unreg_list) {
6286 /* And unlink it from device chain. */
6287 unlist_netdevice(dev);
6289 dev->reg_state = NETREG_UNREGISTERING;
6290 on_each_cpu(flush_backlog, dev, 1);
6295 list_for_each_entry(dev, head, unreg_list) {
6296 struct sk_buff *skb = NULL;
6298 /* Shutdown queueing discipline. */
6302 /* Notify protocols, that we are about to destroy
6303 this device. They should clean all the things.
6305 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6307 if (!dev->rtnl_link_ops ||
6308 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6309 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6313 * Flush the unicast and multicast chains
6318 if (dev->netdev_ops->ndo_uninit)
6319 dev->netdev_ops->ndo_uninit(dev);
6322 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6324 /* Notifier chain MUST detach us all upper devices. */
6325 WARN_ON(netdev_has_any_upper_dev(dev));
6327 /* Remove entries from kobject tree */
6328 netdev_unregister_kobject(dev);
6330 /* Remove XPS queueing entries */
6331 netif_reset_xps_queues_gt(dev, 0);
6337 list_for_each_entry(dev, head, unreg_list)
6341 static void rollback_registered(struct net_device *dev)
6345 list_add(&dev->unreg_list, &single);
6346 rollback_registered_many(&single);
6350 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6351 struct net_device *upper, netdev_features_t features)
6353 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6354 netdev_features_t feature;
6357 for_each_netdev_feature(&upper_disables, feature_bit) {
6358 feature = __NETIF_F_BIT(feature_bit);
6359 if (!(upper->wanted_features & feature)
6360 && (features & feature)) {
6361 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6362 &feature, upper->name);
6363 features &= ~feature;
6370 static void netdev_sync_lower_features(struct net_device *upper,
6371 struct net_device *lower, netdev_features_t features)
6373 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6374 netdev_features_t feature;
6377 for_each_netdev_feature(&upper_disables, feature_bit) {
6378 feature = __NETIF_F_BIT(feature_bit);
6379 if (!(features & feature) && (lower->features & feature)) {
6380 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6381 &feature, lower->name);
6382 lower->wanted_features &= ~feature;
6383 netdev_update_features(lower);
6385 if (unlikely(lower->features & feature))
6386 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6387 &feature, lower->name);
6392 static netdev_features_t netdev_fix_features(struct net_device *dev,
6393 netdev_features_t features)
6395 /* Fix illegal checksum combinations */
6396 if ((features & NETIF_F_HW_CSUM) &&
6397 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6398 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6399 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6402 /* TSO requires that SG is present as well. */
6403 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6404 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6405 features &= ~NETIF_F_ALL_TSO;
6408 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6409 !(features & NETIF_F_IP_CSUM)) {
6410 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6411 features &= ~NETIF_F_TSO;
6412 features &= ~NETIF_F_TSO_ECN;
6415 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6416 !(features & NETIF_F_IPV6_CSUM)) {
6417 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6418 features &= ~NETIF_F_TSO6;
6421 /* TSO ECN requires that TSO is present as well. */
6422 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6423 features &= ~NETIF_F_TSO_ECN;
6425 /* Software GSO depends on SG. */
6426 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6427 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6428 features &= ~NETIF_F_GSO;
6431 /* UFO needs SG and checksumming */
6432 if (features & NETIF_F_UFO) {
6433 /* maybe split UFO into V4 and V6? */
6434 if (!((features & NETIF_F_GEN_CSUM) ||
6435 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6436 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6438 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6439 features &= ~NETIF_F_UFO;
6442 if (!(features & NETIF_F_SG)) {
6444 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6445 features &= ~NETIF_F_UFO;
6449 #ifdef CONFIG_NET_RX_BUSY_POLL
6450 if (dev->netdev_ops->ndo_busy_poll)
6451 features |= NETIF_F_BUSY_POLL;
6454 features &= ~NETIF_F_BUSY_POLL;
6459 int __netdev_update_features(struct net_device *dev)
6461 struct net_device *upper, *lower;
6462 netdev_features_t features;
6463 struct list_head *iter;
6468 features = netdev_get_wanted_features(dev);
6470 if (dev->netdev_ops->ndo_fix_features)
6471 features = dev->netdev_ops->ndo_fix_features(dev, features);
6473 /* driver might be less strict about feature dependencies */
6474 features = netdev_fix_features(dev, features);
6476 /* some features can't be enabled if they're off an an upper device */
6477 netdev_for_each_upper_dev_rcu(dev, upper, iter)
6478 features = netdev_sync_upper_features(dev, upper, features);
6480 if (dev->features == features)
6483 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6484 &dev->features, &features);
6486 if (dev->netdev_ops->ndo_set_features)
6487 err = dev->netdev_ops->ndo_set_features(dev, features);
6491 if (unlikely(err < 0)) {
6493 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6494 err, &features, &dev->features);
6495 /* return non-0 since some features might have changed and
6496 * it's better to fire a spurious notification than miss it
6502 /* some features must be disabled on lower devices when disabled
6503 * on an upper device (think: bonding master or bridge)
6505 netdev_for_each_lower_dev(dev, lower, iter)
6506 netdev_sync_lower_features(dev, lower, features);
6509 dev->features = features;
6511 return err < 0 ? 0 : 1;
6515 * netdev_update_features - recalculate device features
6516 * @dev: the device to check
6518 * Recalculate dev->features set and send notifications if it
6519 * has changed. Should be called after driver or hardware dependent
6520 * conditions might have changed that influence the features.
6522 void netdev_update_features(struct net_device *dev)
6524 if (__netdev_update_features(dev))
6525 netdev_features_change(dev);
6527 EXPORT_SYMBOL(netdev_update_features);
6530 * netdev_change_features - recalculate device features
6531 * @dev: the device to check
6533 * Recalculate dev->features set and send notifications even
6534 * if they have not changed. Should be called instead of
6535 * netdev_update_features() if also dev->vlan_features might
6536 * have changed to allow the changes to be propagated to stacked
6539 void netdev_change_features(struct net_device *dev)
6541 __netdev_update_features(dev);
6542 netdev_features_change(dev);
6544 EXPORT_SYMBOL(netdev_change_features);
6547 * netif_stacked_transfer_operstate - transfer operstate
6548 * @rootdev: the root or lower level device to transfer state from
6549 * @dev: the device to transfer operstate to
6551 * Transfer operational state from root to device. This is normally
6552 * called when a stacking relationship exists between the root
6553 * device and the device(a leaf device).
6555 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6556 struct net_device *dev)
6558 if (rootdev->operstate == IF_OPER_DORMANT)
6559 netif_dormant_on(dev);
6561 netif_dormant_off(dev);
6563 if (netif_carrier_ok(rootdev)) {
6564 if (!netif_carrier_ok(dev))
6565 netif_carrier_on(dev);
6567 if (netif_carrier_ok(dev))
6568 netif_carrier_off(dev);
6571 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6574 static int netif_alloc_rx_queues(struct net_device *dev)
6576 unsigned int i, count = dev->num_rx_queues;
6577 struct netdev_rx_queue *rx;
6578 size_t sz = count * sizeof(*rx);
6582 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6590 for (i = 0; i < count; i++)
6596 static void netdev_init_one_queue(struct net_device *dev,
6597 struct netdev_queue *queue, void *_unused)
6599 /* Initialize queue lock */
6600 spin_lock_init(&queue->_xmit_lock);
6601 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6602 queue->xmit_lock_owner = -1;
6603 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6606 dql_init(&queue->dql, HZ);
6610 static void netif_free_tx_queues(struct net_device *dev)
6615 static int netif_alloc_netdev_queues(struct net_device *dev)
6617 unsigned int count = dev->num_tx_queues;
6618 struct netdev_queue *tx;
6619 size_t sz = count * sizeof(*tx);
6621 if (count < 1 || count > 0xffff)
6624 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6632 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6633 spin_lock_init(&dev->tx_global_lock);
6638 void netif_tx_stop_all_queues(struct net_device *dev)
6642 for (i = 0; i < dev->num_tx_queues; i++) {
6643 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
6644 netif_tx_stop_queue(txq);
6647 EXPORT_SYMBOL(netif_tx_stop_all_queues);
6650 * register_netdevice - register a network device
6651 * @dev: device to register
6653 * Take a completed network device structure and add it to the kernel
6654 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6655 * chain. 0 is returned on success. A negative errno code is returned
6656 * on a failure to set up the device, or if the name is a duplicate.
6658 * Callers must hold the rtnl semaphore. You may want
6659 * register_netdev() instead of this.
6662 * The locking appears insufficient to guarantee two parallel registers
6663 * will not get the same name.
6666 int register_netdevice(struct net_device *dev)
6669 struct net *net = dev_net(dev);
6671 BUG_ON(dev_boot_phase);
6676 /* When net_device's are persistent, this will be fatal. */
6677 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6680 spin_lock_init(&dev->addr_list_lock);
6681 netdev_set_addr_lockdep_class(dev);
6683 ret = dev_get_valid_name(net, dev, dev->name);
6687 /* Init, if this function is available */
6688 if (dev->netdev_ops->ndo_init) {
6689 ret = dev->netdev_ops->ndo_init(dev);
6697 if (((dev->hw_features | dev->features) &
6698 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6699 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6700 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6701 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6708 dev->ifindex = dev_new_index(net);
6709 else if (__dev_get_by_index(net, dev->ifindex))
6712 /* Transfer changeable features to wanted_features and enable
6713 * software offloads (GSO and GRO).
6715 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6716 dev->features |= NETIF_F_SOFT_FEATURES;
6717 dev->wanted_features = dev->features & dev->hw_features;
6719 if (!(dev->flags & IFF_LOOPBACK)) {
6720 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6723 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6725 dev->vlan_features |= NETIF_F_HIGHDMA;
6727 /* Make NETIF_F_SG inheritable to tunnel devices.
6729 dev->hw_enc_features |= NETIF_F_SG;
6731 /* Make NETIF_F_SG inheritable to MPLS.
6733 dev->mpls_features |= NETIF_F_SG;
6735 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6736 ret = notifier_to_errno(ret);
6740 ret = netdev_register_kobject(dev);
6743 dev->reg_state = NETREG_REGISTERED;
6745 __netdev_update_features(dev);
6748 * Default initial state at registry is that the
6749 * device is present.
6752 set_bit(__LINK_STATE_PRESENT, &dev->state);
6754 linkwatch_init_dev(dev);
6756 dev_init_scheduler(dev);
6758 list_netdevice(dev);
6759 add_device_randomness(dev->dev_addr, dev->addr_len);
6761 /* If the device has permanent device address, driver should
6762 * set dev_addr and also addr_assign_type should be set to
6763 * NET_ADDR_PERM (default value).
6765 if (dev->addr_assign_type == NET_ADDR_PERM)
6766 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6768 /* Notify protocols, that a new device appeared. */
6769 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6770 ret = notifier_to_errno(ret);
6772 rollback_registered(dev);
6773 dev->reg_state = NETREG_UNREGISTERED;
6776 * Prevent userspace races by waiting until the network
6777 * device is fully setup before sending notifications.
6779 if (!dev->rtnl_link_ops ||
6780 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6781 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6787 if (dev->netdev_ops->ndo_uninit)
6788 dev->netdev_ops->ndo_uninit(dev);
6791 EXPORT_SYMBOL(register_netdevice);
6794 * init_dummy_netdev - init a dummy network device for NAPI
6795 * @dev: device to init
6797 * This takes a network device structure and initialize the minimum
6798 * amount of fields so it can be used to schedule NAPI polls without
6799 * registering a full blown interface. This is to be used by drivers
6800 * that need to tie several hardware interfaces to a single NAPI
6801 * poll scheduler due to HW limitations.
6803 int init_dummy_netdev(struct net_device *dev)
6805 /* Clear everything. Note we don't initialize spinlocks
6806 * are they aren't supposed to be taken by any of the
6807 * NAPI code and this dummy netdev is supposed to be
6808 * only ever used for NAPI polls
6810 memset(dev, 0, sizeof(struct net_device));
6812 /* make sure we BUG if trying to hit standard
6813 * register/unregister code path
6815 dev->reg_state = NETREG_DUMMY;
6817 /* NAPI wants this */
6818 INIT_LIST_HEAD(&dev->napi_list);
6820 /* a dummy interface is started by default */
6821 set_bit(__LINK_STATE_PRESENT, &dev->state);
6822 set_bit(__LINK_STATE_START, &dev->state);
6824 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6825 * because users of this 'device' dont need to change
6831 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6835 * register_netdev - register a network device
6836 * @dev: device to register
6838 * Take a completed network device structure and add it to the kernel
6839 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6840 * chain. 0 is returned on success. A negative errno code is returned
6841 * on a failure to set up the device, or if the name is a duplicate.
6843 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6844 * and expands the device name if you passed a format string to
6847 int register_netdev(struct net_device *dev)
6852 err = register_netdevice(dev);
6856 EXPORT_SYMBOL(register_netdev);
6858 int netdev_refcnt_read(const struct net_device *dev)
6862 for_each_possible_cpu(i)
6863 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6866 EXPORT_SYMBOL(netdev_refcnt_read);
6869 * netdev_wait_allrefs - wait until all references are gone.
6870 * @dev: target net_device
6872 * This is called when unregistering network devices.
6874 * Any protocol or device that holds a reference should register
6875 * for netdevice notification, and cleanup and put back the
6876 * reference if they receive an UNREGISTER event.
6877 * We can get stuck here if buggy protocols don't correctly
6880 static void netdev_wait_allrefs(struct net_device *dev)
6882 unsigned long rebroadcast_time, warning_time;
6885 linkwatch_forget_dev(dev);
6887 rebroadcast_time = warning_time = jiffies;
6888 refcnt = netdev_refcnt_read(dev);
6890 while (refcnt != 0) {
6891 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6894 /* Rebroadcast unregister notification */
6895 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6901 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6902 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6904 /* We must not have linkwatch events
6905 * pending on unregister. If this
6906 * happens, we simply run the queue
6907 * unscheduled, resulting in a noop
6910 linkwatch_run_queue();
6915 rebroadcast_time = jiffies;
6920 refcnt = netdev_refcnt_read(dev);
6922 if (time_after(jiffies, warning_time + 10 * HZ)) {
6923 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6925 warning_time = jiffies;
6934 * register_netdevice(x1);
6935 * register_netdevice(x2);
6937 * unregister_netdevice(y1);
6938 * unregister_netdevice(y2);
6944 * We are invoked by rtnl_unlock().
6945 * This allows us to deal with problems:
6946 * 1) We can delete sysfs objects which invoke hotplug
6947 * without deadlocking with linkwatch via keventd.
6948 * 2) Since we run with the RTNL semaphore not held, we can sleep
6949 * safely in order to wait for the netdev refcnt to drop to zero.
6951 * We must not return until all unregister events added during
6952 * the interval the lock was held have been completed.
6954 void netdev_run_todo(void)
6956 struct list_head list;
6958 /* Snapshot list, allow later requests */
6959 list_replace_init(&net_todo_list, &list);
6964 /* Wait for rcu callbacks to finish before next phase */
6965 if (!list_empty(&list))
6968 while (!list_empty(&list)) {
6969 struct net_device *dev
6970 = list_first_entry(&list, struct net_device, todo_list);
6971 list_del(&dev->todo_list);
6974 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6977 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6978 pr_err("network todo '%s' but state %d\n",
6979 dev->name, dev->reg_state);
6984 dev->reg_state = NETREG_UNREGISTERED;
6986 netdev_wait_allrefs(dev);
6989 BUG_ON(netdev_refcnt_read(dev));
6990 BUG_ON(!list_empty(&dev->ptype_all));
6991 BUG_ON(!list_empty(&dev->ptype_specific));
6992 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6993 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6994 WARN_ON(dev->dn_ptr);
6996 if (dev->destructor)
6997 dev->destructor(dev);
6999 /* Report a network device has been unregistered */
7001 dev_net(dev)->dev_unreg_count--;
7003 wake_up(&netdev_unregistering_wq);
7005 /* Free network device */
7006 kobject_put(&dev->dev.kobj);
7010 /* Convert net_device_stats to rtnl_link_stats64. They have the same
7011 * fields in the same order, with only the type differing.
7013 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
7014 const struct net_device_stats *netdev_stats)
7016 #if BITS_PER_LONG == 64
7017 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
7018 memcpy(stats64, netdev_stats, sizeof(*stats64));
7020 size_t i, n = sizeof(*stats64) / sizeof(u64);
7021 const unsigned long *src = (const unsigned long *)netdev_stats;
7022 u64 *dst = (u64 *)stats64;
7024 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
7025 sizeof(*stats64) / sizeof(u64));
7026 for (i = 0; i < n; i++)
7030 EXPORT_SYMBOL(netdev_stats_to_stats64);
7033 * dev_get_stats - get network device statistics
7034 * @dev: device to get statistics from
7035 * @storage: place to store stats
7037 * Get network statistics from device. Return @storage.
7038 * The device driver may provide its own method by setting
7039 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7040 * otherwise the internal statistics structure is used.
7042 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
7043 struct rtnl_link_stats64 *storage)
7045 const struct net_device_ops *ops = dev->netdev_ops;
7047 if (ops->ndo_get_stats64) {
7048 memset(storage, 0, sizeof(*storage));
7049 ops->ndo_get_stats64(dev, storage);
7050 } else if (ops->ndo_get_stats) {
7051 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7053 netdev_stats_to_stats64(storage, &dev->stats);
7055 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
7056 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
7059 EXPORT_SYMBOL(dev_get_stats);
7061 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7063 struct netdev_queue *queue = dev_ingress_queue(dev);
7065 #ifdef CONFIG_NET_CLS_ACT
7068 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7071 netdev_init_one_queue(dev, queue, NULL);
7072 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7073 queue->qdisc_sleeping = &noop_qdisc;
7074 rcu_assign_pointer(dev->ingress_queue, queue);
7079 static const struct ethtool_ops default_ethtool_ops;
7081 void netdev_set_default_ethtool_ops(struct net_device *dev,
7082 const struct ethtool_ops *ops)
7084 if (dev->ethtool_ops == &default_ethtool_ops)
7085 dev->ethtool_ops = ops;
7087 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7089 void netdev_freemem(struct net_device *dev)
7091 char *addr = (char *)dev - dev->padded;
7097 * alloc_netdev_mqs - allocate network device
7098 * @sizeof_priv: size of private data to allocate space for
7099 * @name: device name format string
7100 * @name_assign_type: origin of device name
7101 * @setup: callback to initialize device
7102 * @txqs: the number of TX subqueues to allocate
7103 * @rxqs: the number of RX subqueues to allocate
7105 * Allocates a struct net_device with private data area for driver use
7106 * and performs basic initialization. Also allocates subqueue structs
7107 * for each queue on the device.
7109 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7110 unsigned char name_assign_type,
7111 void (*setup)(struct net_device *),
7112 unsigned int txqs, unsigned int rxqs)
7114 struct net_device *dev;
7116 struct net_device *p;
7118 BUG_ON(strlen(name) >= sizeof(dev->name));
7121 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7127 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7132 alloc_size = sizeof(struct net_device);
7134 /* ensure 32-byte alignment of private area */
7135 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7136 alloc_size += sizeof_priv;
7138 /* ensure 32-byte alignment of whole construct */
7139 alloc_size += NETDEV_ALIGN - 1;
7141 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7143 p = vzalloc(alloc_size);
7147 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7148 dev->padded = (char *)dev - (char *)p;
7150 dev->pcpu_refcnt = alloc_percpu(int);
7151 if (!dev->pcpu_refcnt)
7154 if (dev_addr_init(dev))
7160 dev_net_set(dev, &init_net);
7162 dev->gso_max_size = GSO_MAX_SIZE;
7163 dev->gso_max_segs = GSO_MAX_SEGS;
7164 dev->gso_min_segs = 0;
7166 INIT_LIST_HEAD(&dev->napi_list);
7167 INIT_LIST_HEAD(&dev->unreg_list);
7168 INIT_LIST_HEAD(&dev->close_list);
7169 INIT_LIST_HEAD(&dev->link_watch_list);
7170 INIT_LIST_HEAD(&dev->adj_list.upper);
7171 INIT_LIST_HEAD(&dev->adj_list.lower);
7172 INIT_LIST_HEAD(&dev->all_adj_list.upper);
7173 INIT_LIST_HEAD(&dev->all_adj_list.lower);
7174 INIT_LIST_HEAD(&dev->ptype_all);
7175 INIT_LIST_HEAD(&dev->ptype_specific);
7176 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7179 if (!dev->tx_queue_len) {
7180 dev->priv_flags |= IFF_NO_QUEUE;
7181 dev->tx_queue_len = 1;
7184 dev->num_tx_queues = txqs;
7185 dev->real_num_tx_queues = txqs;
7186 if (netif_alloc_netdev_queues(dev))
7190 dev->num_rx_queues = rxqs;
7191 dev->real_num_rx_queues = rxqs;
7192 if (netif_alloc_rx_queues(dev))
7196 strcpy(dev->name, name);
7197 dev->name_assign_type = name_assign_type;
7198 dev->group = INIT_NETDEV_GROUP;
7199 if (!dev->ethtool_ops)
7200 dev->ethtool_ops = &default_ethtool_ops;
7202 nf_hook_ingress_init(dev);
7211 free_percpu(dev->pcpu_refcnt);
7213 netdev_freemem(dev);
7216 EXPORT_SYMBOL(alloc_netdev_mqs);
7219 * free_netdev - free network device
7222 * This function does the last stage of destroying an allocated device
7223 * interface. The reference to the device object is released.
7224 * If this is the last reference then it will be freed.
7226 void free_netdev(struct net_device *dev)
7228 struct napi_struct *p, *n;
7230 netif_free_tx_queues(dev);
7235 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7237 /* Flush device addresses */
7238 dev_addr_flush(dev);
7240 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7243 free_percpu(dev->pcpu_refcnt);
7244 dev->pcpu_refcnt = NULL;
7246 /* Compatibility with error handling in drivers */
7247 if (dev->reg_state == NETREG_UNINITIALIZED) {
7248 netdev_freemem(dev);
7252 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7253 dev->reg_state = NETREG_RELEASED;
7255 /* will free via device release */
7256 put_device(&dev->dev);
7258 EXPORT_SYMBOL(free_netdev);
7261 * synchronize_net - Synchronize with packet receive processing
7263 * Wait for packets currently being received to be done.
7264 * Does not block later packets from starting.
7266 void synchronize_net(void)
7269 if (rtnl_is_locked())
7270 synchronize_rcu_expedited();
7274 EXPORT_SYMBOL(synchronize_net);
7277 * unregister_netdevice_queue - remove device from the kernel
7281 * This function shuts down a device interface and removes it
7282 * from the kernel tables.
7283 * If head not NULL, device is queued to be unregistered later.
7285 * Callers must hold the rtnl semaphore. You may want
7286 * unregister_netdev() instead of this.
7289 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7294 list_move_tail(&dev->unreg_list, head);
7296 rollback_registered(dev);
7297 /* Finish processing unregister after unlock */
7301 EXPORT_SYMBOL(unregister_netdevice_queue);
7304 * unregister_netdevice_many - unregister many devices
7305 * @head: list of devices
7307 * Note: As most callers use a stack allocated list_head,
7308 * we force a list_del() to make sure stack wont be corrupted later.
7310 void unregister_netdevice_many(struct list_head *head)
7312 struct net_device *dev;
7314 if (!list_empty(head)) {
7315 rollback_registered_many(head);
7316 list_for_each_entry(dev, head, unreg_list)
7321 EXPORT_SYMBOL(unregister_netdevice_many);
7324 * unregister_netdev - remove device from the kernel
7327 * This function shuts down a device interface and removes it
7328 * from the kernel tables.
7330 * This is just a wrapper for unregister_netdevice that takes
7331 * the rtnl semaphore. In general you want to use this and not
7332 * unregister_netdevice.
7334 void unregister_netdev(struct net_device *dev)
7337 unregister_netdevice(dev);
7340 EXPORT_SYMBOL(unregister_netdev);
7343 * dev_change_net_namespace - move device to different nethost namespace
7345 * @net: network namespace
7346 * @pat: If not NULL name pattern to try if the current device name
7347 * is already taken in the destination network namespace.
7349 * This function shuts down a device interface and moves it
7350 * to a new network namespace. On success 0 is returned, on
7351 * a failure a netagive errno code is returned.
7353 * Callers must hold the rtnl semaphore.
7356 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7362 /* Don't allow namespace local devices to be moved. */
7364 if (dev->features & NETIF_F_NETNS_LOCAL)
7367 /* Ensure the device has been registrered */
7368 if (dev->reg_state != NETREG_REGISTERED)
7371 /* Get out if there is nothing todo */
7373 if (net_eq(dev_net(dev), net))
7376 /* Pick the destination device name, and ensure
7377 * we can use it in the destination network namespace.
7380 if (__dev_get_by_name(net, dev->name)) {
7381 /* We get here if we can't use the current device name */
7384 if (dev_get_valid_name(net, dev, pat) < 0)
7389 * And now a mini version of register_netdevice unregister_netdevice.
7392 /* If device is running close it first. */
7395 /* And unlink it from device chain */
7397 unlist_netdevice(dev);
7401 /* Shutdown queueing discipline. */
7404 /* Notify protocols, that we are about to destroy
7405 this device. They should clean all the things.
7407 Note that dev->reg_state stays at NETREG_REGISTERED.
7408 This is wanted because this way 8021q and macvlan know
7409 the device is just moving and can keep their slaves up.
7411 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7413 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7414 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7417 * Flush the unicast and multicast chains
7422 /* Send a netdev-removed uevent to the old namespace */
7423 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7424 netdev_adjacent_del_links(dev);
7426 /* Actually switch the network namespace */
7427 dev_net_set(dev, net);
7429 /* If there is an ifindex conflict assign a new one */
7430 if (__dev_get_by_index(net, dev->ifindex))
7431 dev->ifindex = dev_new_index(net);
7433 /* Send a netdev-add uevent to the new namespace */
7434 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7435 netdev_adjacent_add_links(dev);
7437 /* Fixup kobjects */
7438 err = device_rename(&dev->dev, dev->name);
7441 /* Add the device back in the hashes */
7442 list_netdevice(dev);
7444 /* Notify protocols, that a new device appeared. */
7445 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7448 * Prevent userspace races by waiting until the network
7449 * device is fully setup before sending notifications.
7451 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7458 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7460 static int dev_cpu_callback(struct notifier_block *nfb,
7461 unsigned long action,
7464 struct sk_buff **list_skb;
7465 struct sk_buff *skb;
7466 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7467 struct softnet_data *sd, *oldsd;
7469 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7472 local_irq_disable();
7473 cpu = smp_processor_id();
7474 sd = &per_cpu(softnet_data, cpu);
7475 oldsd = &per_cpu(softnet_data, oldcpu);
7477 /* Find end of our completion_queue. */
7478 list_skb = &sd->completion_queue;
7480 list_skb = &(*list_skb)->next;
7481 /* Append completion queue from offline CPU. */
7482 *list_skb = oldsd->completion_queue;
7483 oldsd->completion_queue = NULL;
7485 /* Append output queue from offline CPU. */
7486 if (oldsd->output_queue) {
7487 *sd->output_queue_tailp = oldsd->output_queue;
7488 sd->output_queue_tailp = oldsd->output_queue_tailp;
7489 oldsd->output_queue = NULL;
7490 oldsd->output_queue_tailp = &oldsd->output_queue;
7492 /* Append NAPI poll list from offline CPU, with one exception :
7493 * process_backlog() must be called by cpu owning percpu backlog.
7494 * We properly handle process_queue & input_pkt_queue later.
7496 while (!list_empty(&oldsd->poll_list)) {
7497 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7501 list_del_init(&napi->poll_list);
7502 if (napi->poll == process_backlog)
7505 ____napi_schedule(sd, napi);
7508 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7511 /* Process offline CPU's input_pkt_queue */
7512 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7514 input_queue_head_incr(oldsd);
7516 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7518 input_queue_head_incr(oldsd);
7526 * netdev_increment_features - increment feature set by one
7527 * @all: current feature set
7528 * @one: new feature set
7529 * @mask: mask feature set
7531 * Computes a new feature set after adding a device with feature set
7532 * @one to the master device with current feature set @all. Will not
7533 * enable anything that is off in @mask. Returns the new feature set.
7535 netdev_features_t netdev_increment_features(netdev_features_t all,
7536 netdev_features_t one, netdev_features_t mask)
7538 if (mask & NETIF_F_GEN_CSUM)
7539 mask |= NETIF_F_ALL_CSUM;
7540 mask |= NETIF_F_VLAN_CHALLENGED;
7542 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7543 all &= one | ~NETIF_F_ALL_FOR_ALL;
7545 /* If one device supports hw checksumming, set for all. */
7546 if (all & NETIF_F_GEN_CSUM)
7547 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7551 EXPORT_SYMBOL(netdev_increment_features);
7553 static struct hlist_head * __net_init netdev_create_hash(void)
7556 struct hlist_head *hash;
7558 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7560 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7561 INIT_HLIST_HEAD(&hash[i]);
7566 /* Initialize per network namespace state */
7567 static int __net_init netdev_init(struct net *net)
7569 if (net != &init_net)
7570 INIT_LIST_HEAD(&net->dev_base_head);
7572 net->dev_name_head = netdev_create_hash();
7573 if (net->dev_name_head == NULL)
7576 net->dev_index_head = netdev_create_hash();
7577 if (net->dev_index_head == NULL)
7583 kfree(net->dev_name_head);
7589 * netdev_drivername - network driver for the device
7590 * @dev: network device
7592 * Determine network driver for device.
7594 const char *netdev_drivername(const struct net_device *dev)
7596 const struct device_driver *driver;
7597 const struct device *parent;
7598 const char *empty = "";
7600 parent = dev->dev.parent;
7604 driver = parent->driver;
7605 if (driver && driver->name)
7606 return driver->name;
7610 static void __netdev_printk(const char *level, const struct net_device *dev,
7611 struct va_format *vaf)
7613 if (dev && dev->dev.parent) {
7614 dev_printk_emit(level[1] - '0',
7617 dev_driver_string(dev->dev.parent),
7618 dev_name(dev->dev.parent),
7619 netdev_name(dev), netdev_reg_state(dev),
7622 printk("%s%s%s: %pV",
7623 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7625 printk("%s(NULL net_device): %pV", level, vaf);
7629 void netdev_printk(const char *level, const struct net_device *dev,
7630 const char *format, ...)
7632 struct va_format vaf;
7635 va_start(args, format);
7640 __netdev_printk(level, dev, &vaf);
7644 EXPORT_SYMBOL(netdev_printk);
7646 #define define_netdev_printk_level(func, level) \
7647 void func(const struct net_device *dev, const char *fmt, ...) \
7649 struct va_format vaf; \
7652 va_start(args, fmt); \
7657 __netdev_printk(level, dev, &vaf); \
7661 EXPORT_SYMBOL(func);
7663 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7664 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7665 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7666 define_netdev_printk_level(netdev_err, KERN_ERR);
7667 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7668 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7669 define_netdev_printk_level(netdev_info, KERN_INFO);
7671 static void __net_exit netdev_exit(struct net *net)
7673 kfree(net->dev_name_head);
7674 kfree(net->dev_index_head);
7677 static struct pernet_operations __net_initdata netdev_net_ops = {
7678 .init = netdev_init,
7679 .exit = netdev_exit,
7682 static void __net_exit default_device_exit(struct net *net)
7684 struct net_device *dev, *aux;
7686 * Push all migratable network devices back to the
7687 * initial network namespace
7690 for_each_netdev_safe(net, dev, aux) {
7692 char fb_name[IFNAMSIZ];
7694 /* Ignore unmoveable devices (i.e. loopback) */
7695 if (dev->features & NETIF_F_NETNS_LOCAL)
7698 /* Leave virtual devices for the generic cleanup */
7699 if (dev->rtnl_link_ops)
7702 /* Push remaining network devices to init_net */
7703 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7704 err = dev_change_net_namespace(dev, &init_net, fb_name);
7706 pr_emerg("%s: failed to move %s to init_net: %d\n",
7707 __func__, dev->name, err);
7714 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7716 /* Return with the rtnl_lock held when there are no network
7717 * devices unregistering in any network namespace in net_list.
7721 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7723 add_wait_queue(&netdev_unregistering_wq, &wait);
7725 unregistering = false;
7727 list_for_each_entry(net, net_list, exit_list) {
7728 if (net->dev_unreg_count > 0) {
7729 unregistering = true;
7737 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7739 remove_wait_queue(&netdev_unregistering_wq, &wait);
7742 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7744 /* At exit all network devices most be removed from a network
7745 * namespace. Do this in the reverse order of registration.
7746 * Do this across as many network namespaces as possible to
7747 * improve batching efficiency.
7749 struct net_device *dev;
7751 LIST_HEAD(dev_kill_list);
7753 /* To prevent network device cleanup code from dereferencing
7754 * loopback devices or network devices that have been freed
7755 * wait here for all pending unregistrations to complete,
7756 * before unregistring the loopback device and allowing the
7757 * network namespace be freed.
7759 * The netdev todo list containing all network devices
7760 * unregistrations that happen in default_device_exit_batch
7761 * will run in the rtnl_unlock() at the end of
7762 * default_device_exit_batch.
7764 rtnl_lock_unregistering(net_list);
7765 list_for_each_entry(net, net_list, exit_list) {
7766 for_each_netdev_reverse(net, dev) {
7767 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7768 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7770 unregister_netdevice_queue(dev, &dev_kill_list);
7773 unregister_netdevice_many(&dev_kill_list);
7777 static struct pernet_operations __net_initdata default_device_ops = {
7778 .exit = default_device_exit,
7779 .exit_batch = default_device_exit_batch,
7783 * Initialize the DEV module. At boot time this walks the device list and
7784 * unhooks any devices that fail to initialise (normally hardware not
7785 * present) and leaves us with a valid list of present and active devices.
7790 * This is called single threaded during boot, so no need
7791 * to take the rtnl semaphore.
7793 static int __init net_dev_init(void)
7795 int i, rc = -ENOMEM;
7797 BUG_ON(!dev_boot_phase);
7799 if (dev_proc_init())
7802 if (netdev_kobject_init())
7805 INIT_LIST_HEAD(&ptype_all);
7806 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7807 INIT_LIST_HEAD(&ptype_base[i]);
7809 INIT_LIST_HEAD(&offload_base);
7811 if (register_pernet_subsys(&netdev_net_ops))
7815 * Initialise the packet receive queues.
7818 for_each_possible_cpu(i) {
7819 struct softnet_data *sd = &per_cpu(softnet_data, i);
7821 skb_queue_head_init(&sd->input_pkt_queue);
7822 skb_queue_head_init(&sd->process_queue);
7823 INIT_LIST_HEAD(&sd->poll_list);
7824 sd->output_queue_tailp = &sd->output_queue;
7826 sd->csd.func = rps_trigger_softirq;
7831 sd->backlog.poll = process_backlog;
7832 sd->backlog.weight = weight_p;
7837 /* The loopback device is special if any other network devices
7838 * is present in a network namespace the loopback device must
7839 * be present. Since we now dynamically allocate and free the
7840 * loopback device ensure this invariant is maintained by
7841 * keeping the loopback device as the first device on the
7842 * list of network devices. Ensuring the loopback devices
7843 * is the first device that appears and the last network device
7846 if (register_pernet_device(&loopback_net_ops))
7849 if (register_pernet_device(&default_device_ops))
7852 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7853 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7855 hotcpu_notifier(dev_cpu_callback, 0);
7862 subsys_initcall(net_dev_init);