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 void netstamp_clear(struct work_struct *work)
1682 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1685 static_key_slow_dec(&netstamp_needed);
1687 static DECLARE_WORK(netstamp_work, netstamp_clear);
1690 void net_enable_timestamp(void)
1692 static_key_slow_inc(&netstamp_needed);
1694 EXPORT_SYMBOL(net_enable_timestamp);
1696 void net_disable_timestamp(void)
1698 #ifdef HAVE_JUMP_LABEL
1699 /* net_disable_timestamp() can be called from non process context */
1700 atomic_inc(&netstamp_needed_deferred);
1701 schedule_work(&netstamp_work);
1703 static_key_slow_dec(&netstamp_needed);
1706 EXPORT_SYMBOL(net_disable_timestamp);
1708 static inline void net_timestamp_set(struct sk_buff *skb)
1710 skb->tstamp.tv64 = 0;
1711 if (static_key_false(&netstamp_needed))
1712 __net_timestamp(skb);
1715 #define net_timestamp_check(COND, SKB) \
1716 if (static_key_false(&netstamp_needed)) { \
1717 if ((COND) && !(SKB)->tstamp.tv64) \
1718 __net_timestamp(SKB); \
1721 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1725 if (!(dev->flags & IFF_UP))
1728 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1729 if (skb->len <= len)
1732 /* if TSO is enabled, we don't care about the length as the packet
1733 * could be forwarded without being segmented before
1735 if (skb_is_gso(skb))
1740 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1742 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1744 if (skb_orphan_frags(skb, GFP_ATOMIC) ||
1745 unlikely(!is_skb_forwardable(dev, skb))) {
1746 atomic_long_inc(&dev->rx_dropped);
1751 skb_scrub_packet(skb, true);
1753 skb->protocol = eth_type_trans(skb, dev);
1754 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1758 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1761 * dev_forward_skb - loopback an skb to another netif
1763 * @dev: destination network device
1764 * @skb: buffer to forward
1767 * NET_RX_SUCCESS (no congestion)
1768 * NET_RX_DROP (packet was dropped, but freed)
1770 * dev_forward_skb can be used for injecting an skb from the
1771 * start_xmit function of one device into the receive queue
1772 * of another device.
1774 * The receiving device may be in another namespace, so
1775 * we have to clear all information in the skb that could
1776 * impact namespace isolation.
1778 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1780 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1782 EXPORT_SYMBOL_GPL(dev_forward_skb);
1784 static inline int deliver_skb(struct sk_buff *skb,
1785 struct packet_type *pt_prev,
1786 struct net_device *orig_dev)
1788 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1790 atomic_inc(&skb->users);
1791 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1794 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1795 struct packet_type **pt,
1796 struct net_device *orig_dev,
1798 struct list_head *ptype_list)
1800 struct packet_type *ptype, *pt_prev = *pt;
1802 list_for_each_entry_rcu(ptype, ptype_list, list) {
1803 if (ptype->type != type)
1806 deliver_skb(skb, pt_prev, orig_dev);
1812 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1814 if (!ptype->af_packet_priv || !skb->sk)
1817 if (ptype->id_match)
1818 return ptype->id_match(ptype, skb->sk);
1819 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1826 * Support routine. Sends outgoing frames to any network
1827 * taps currently in use.
1830 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1832 struct packet_type *ptype;
1833 struct sk_buff *skb2 = NULL;
1834 struct packet_type *pt_prev = NULL;
1835 struct list_head *ptype_list = &ptype_all;
1839 list_for_each_entry_rcu(ptype, ptype_list, list) {
1840 /* Never send packets back to the socket
1841 * they originated from - MvS (miquels@drinkel.ow.org)
1843 if (skb_loop_sk(ptype, skb))
1847 deliver_skb(skb2, pt_prev, skb->dev);
1852 /* need to clone skb, done only once */
1853 skb2 = skb_clone(skb, GFP_ATOMIC);
1857 net_timestamp_set(skb2);
1859 /* skb->nh should be correctly
1860 * set by sender, so that the second statement is
1861 * just protection against buggy protocols.
1863 skb_reset_mac_header(skb2);
1865 if (skb_network_header(skb2) < skb2->data ||
1866 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1867 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1868 ntohs(skb2->protocol),
1870 skb_reset_network_header(skb2);
1873 skb2->transport_header = skb2->network_header;
1874 skb2->pkt_type = PACKET_OUTGOING;
1878 if (ptype_list == &ptype_all) {
1879 ptype_list = &dev->ptype_all;
1884 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1889 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1890 * @dev: Network device
1891 * @txq: number of queues available
1893 * If real_num_tx_queues is changed the tc mappings may no longer be
1894 * valid. To resolve this verify the tc mapping remains valid and if
1895 * not NULL the mapping. With no priorities mapping to this
1896 * offset/count pair it will no longer be used. In the worst case TC0
1897 * is invalid nothing can be done so disable priority mappings. If is
1898 * expected that drivers will fix this mapping if they can before
1899 * calling netif_set_real_num_tx_queues.
1901 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1904 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1906 /* If TC0 is invalidated disable TC mapping */
1907 if (tc->offset + tc->count > txq) {
1908 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1913 /* Invalidated prio to tc mappings set to TC0 */
1914 for (i = 1; i < TC_BITMASK + 1; i++) {
1915 int q = netdev_get_prio_tc_map(dev, i);
1917 tc = &dev->tc_to_txq[q];
1918 if (tc->offset + tc->count > txq) {
1919 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1921 netdev_set_prio_tc_map(dev, i, 0);
1927 static DEFINE_MUTEX(xps_map_mutex);
1928 #define xmap_dereference(P) \
1929 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1931 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1934 struct xps_map *map = NULL;
1938 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1940 for (pos = 0; map && pos < map->len; pos++) {
1941 if (map->queues[pos] == index) {
1943 map->queues[pos] = map->queues[--map->len];
1945 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1946 kfree_rcu(map, rcu);
1956 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1958 struct xps_dev_maps *dev_maps;
1960 bool active = false;
1962 mutex_lock(&xps_map_mutex);
1963 dev_maps = xmap_dereference(dev->xps_maps);
1968 for_each_possible_cpu(cpu) {
1969 for (i = index; i < dev->num_tx_queues; i++) {
1970 if (!remove_xps_queue(dev_maps, cpu, i))
1973 if (i == dev->num_tx_queues)
1978 RCU_INIT_POINTER(dev->xps_maps, NULL);
1979 kfree_rcu(dev_maps, rcu);
1982 for (i = index; i < dev->num_tx_queues; i++)
1983 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1987 mutex_unlock(&xps_map_mutex);
1990 static struct xps_map *expand_xps_map(struct xps_map *map,
1993 struct xps_map *new_map;
1994 int alloc_len = XPS_MIN_MAP_ALLOC;
1997 for (pos = 0; map && pos < map->len; pos++) {
1998 if (map->queues[pos] != index)
2003 /* Need to add queue to this CPU's existing map */
2005 if (pos < map->alloc_len)
2008 alloc_len = map->alloc_len * 2;
2011 /* Need to allocate new map to store queue on this CPU's map */
2012 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2017 for (i = 0; i < pos; i++)
2018 new_map->queues[i] = map->queues[i];
2019 new_map->alloc_len = alloc_len;
2025 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2028 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2029 struct xps_map *map, *new_map;
2030 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
2031 int cpu, numa_node_id = -2;
2032 bool active = false;
2034 mutex_lock(&xps_map_mutex);
2036 dev_maps = xmap_dereference(dev->xps_maps);
2038 /* allocate memory for queue storage */
2039 for_each_online_cpu(cpu) {
2040 if (!cpumask_test_cpu(cpu, mask))
2044 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2045 if (!new_dev_maps) {
2046 mutex_unlock(&xps_map_mutex);
2050 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2053 map = expand_xps_map(map, cpu, index);
2057 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2061 goto out_no_new_maps;
2063 for_each_possible_cpu(cpu) {
2064 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2065 /* add queue to CPU maps */
2068 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2069 while ((pos < map->len) && (map->queues[pos] != index))
2072 if (pos == map->len)
2073 map->queues[map->len++] = index;
2075 if (numa_node_id == -2)
2076 numa_node_id = cpu_to_node(cpu);
2077 else if (numa_node_id != cpu_to_node(cpu))
2080 } else if (dev_maps) {
2081 /* fill in the new device map from the old device map */
2082 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2083 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2088 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2090 /* Cleanup old maps */
2092 for_each_possible_cpu(cpu) {
2093 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2094 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2095 if (map && map != new_map)
2096 kfree_rcu(map, rcu);
2099 kfree_rcu(dev_maps, rcu);
2102 dev_maps = new_dev_maps;
2106 /* update Tx queue numa node */
2107 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2108 (numa_node_id >= 0) ? numa_node_id :
2114 /* removes queue from unused CPUs */
2115 for_each_possible_cpu(cpu) {
2116 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2119 if (remove_xps_queue(dev_maps, cpu, index))
2123 /* free map if not active */
2125 RCU_INIT_POINTER(dev->xps_maps, NULL);
2126 kfree_rcu(dev_maps, rcu);
2130 mutex_unlock(&xps_map_mutex);
2134 /* remove any maps that we added */
2135 for_each_possible_cpu(cpu) {
2136 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2137 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2139 if (new_map && new_map != map)
2143 mutex_unlock(&xps_map_mutex);
2145 kfree(new_dev_maps);
2148 EXPORT_SYMBOL(netif_set_xps_queue);
2152 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2153 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2155 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2159 if (txq < 1 || txq > dev->num_tx_queues)
2162 if (dev->reg_state == NETREG_REGISTERED ||
2163 dev->reg_state == NETREG_UNREGISTERING) {
2166 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2172 netif_setup_tc(dev, txq);
2174 if (txq < dev->real_num_tx_queues) {
2175 qdisc_reset_all_tx_gt(dev, txq);
2177 netif_reset_xps_queues_gt(dev, txq);
2182 dev->real_num_tx_queues = txq;
2185 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2189 * netif_set_real_num_rx_queues - set actual number of RX queues used
2190 * @dev: Network device
2191 * @rxq: Actual number of RX queues
2193 * This must be called either with the rtnl_lock held or before
2194 * registration of the net device. Returns 0 on success, or a
2195 * negative error code. If called before registration, it always
2198 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2202 if (rxq < 1 || rxq > dev->num_rx_queues)
2205 if (dev->reg_state == NETREG_REGISTERED) {
2208 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2214 dev->real_num_rx_queues = rxq;
2217 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2221 * netif_get_num_default_rss_queues - default number of RSS queues
2223 * This routine should set an upper limit on the number of RSS queues
2224 * used by default by multiqueue devices.
2226 int netif_get_num_default_rss_queues(void)
2228 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2230 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2232 static inline void __netif_reschedule(struct Qdisc *q)
2234 struct softnet_data *sd;
2235 unsigned long flags;
2237 local_irq_save(flags);
2238 sd = this_cpu_ptr(&softnet_data);
2239 q->next_sched = NULL;
2240 *sd->output_queue_tailp = q;
2241 sd->output_queue_tailp = &q->next_sched;
2242 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2243 local_irq_restore(flags);
2246 void __netif_schedule(struct Qdisc *q)
2248 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2249 __netif_reschedule(q);
2251 EXPORT_SYMBOL(__netif_schedule);
2253 struct dev_kfree_skb_cb {
2254 enum skb_free_reason reason;
2257 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2259 return (struct dev_kfree_skb_cb *)skb->cb;
2262 void netif_schedule_queue(struct netdev_queue *txq)
2265 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2266 struct Qdisc *q = rcu_dereference(txq->qdisc);
2268 __netif_schedule(q);
2272 EXPORT_SYMBOL(netif_schedule_queue);
2275 * netif_wake_subqueue - allow sending packets on subqueue
2276 * @dev: network device
2277 * @queue_index: sub queue index
2279 * Resume individual transmit queue of a device with multiple transmit queues.
2281 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2283 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2285 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2289 q = rcu_dereference(txq->qdisc);
2290 __netif_schedule(q);
2294 EXPORT_SYMBOL(netif_wake_subqueue);
2296 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2298 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2302 q = rcu_dereference(dev_queue->qdisc);
2303 __netif_schedule(q);
2307 EXPORT_SYMBOL(netif_tx_wake_queue);
2309 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2311 unsigned long flags;
2313 if (likely(atomic_read(&skb->users) == 1)) {
2315 atomic_set(&skb->users, 0);
2316 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2319 get_kfree_skb_cb(skb)->reason = reason;
2320 local_irq_save(flags);
2321 skb->next = __this_cpu_read(softnet_data.completion_queue);
2322 __this_cpu_write(softnet_data.completion_queue, skb);
2323 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2324 local_irq_restore(flags);
2326 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2328 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2330 if (in_irq() || irqs_disabled())
2331 __dev_kfree_skb_irq(skb, reason);
2335 EXPORT_SYMBOL(__dev_kfree_skb_any);
2339 * netif_device_detach - mark device as removed
2340 * @dev: network device
2342 * Mark device as removed from system and therefore no longer available.
2344 void netif_device_detach(struct net_device *dev)
2346 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2347 netif_running(dev)) {
2348 netif_tx_stop_all_queues(dev);
2351 EXPORT_SYMBOL(netif_device_detach);
2354 * netif_device_attach - mark device as attached
2355 * @dev: network device
2357 * Mark device as attached from system and restart if needed.
2359 void netif_device_attach(struct net_device *dev)
2361 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2362 netif_running(dev)) {
2363 netif_tx_wake_all_queues(dev);
2364 __netdev_watchdog_up(dev);
2367 EXPORT_SYMBOL(netif_device_attach);
2370 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2371 * to be used as a distribution range.
2373 u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
2374 unsigned int num_tx_queues)
2378 u16 qcount = num_tx_queues;
2380 if (skb_rx_queue_recorded(skb)) {
2381 hash = skb_get_rx_queue(skb);
2382 while (unlikely(hash >= num_tx_queues))
2383 hash -= num_tx_queues;
2388 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2389 qoffset = dev->tc_to_txq[tc].offset;
2390 qcount = dev->tc_to_txq[tc].count;
2393 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2395 EXPORT_SYMBOL(__skb_tx_hash);
2397 static void skb_warn_bad_offload(const struct sk_buff *skb)
2399 static const netdev_features_t null_features = 0;
2400 struct net_device *dev = skb->dev;
2401 const char *name = "";
2403 if (!net_ratelimit())
2407 if (dev->dev.parent)
2408 name = dev_driver_string(dev->dev.parent);
2410 name = netdev_name(dev);
2412 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2413 "gso_type=%d ip_summed=%d\n",
2414 name, dev ? &dev->features : &null_features,
2415 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2416 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2417 skb_shinfo(skb)->gso_type, skb->ip_summed);
2421 * Invalidate hardware checksum when packet is to be mangled, and
2422 * complete checksum manually on outgoing path.
2424 int skb_checksum_help(struct sk_buff *skb)
2427 int ret = 0, offset;
2429 if (skb->ip_summed == CHECKSUM_COMPLETE)
2430 goto out_set_summed;
2432 if (unlikely(skb_shinfo(skb)->gso_size)) {
2433 skb_warn_bad_offload(skb);
2437 /* Before computing a checksum, we should make sure no frag could
2438 * be modified by an external entity : checksum could be wrong.
2440 if (skb_has_shared_frag(skb)) {
2441 ret = __skb_linearize(skb);
2446 offset = skb_checksum_start_offset(skb);
2447 BUG_ON(offset >= skb_headlen(skb));
2448 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2450 offset += skb->csum_offset;
2451 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2453 if (skb_cloned(skb) &&
2454 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2455 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2460 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2462 skb->ip_summed = CHECKSUM_NONE;
2466 EXPORT_SYMBOL(skb_checksum_help);
2468 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2470 __be16 type = skb->protocol;
2472 /* Tunnel gso handlers can set protocol to ethernet. */
2473 if (type == htons(ETH_P_TEB)) {
2476 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2479 eth = (struct ethhdr *)skb_mac_header(skb);
2480 type = eth->h_proto;
2483 return __vlan_get_protocol(skb, type, depth);
2487 * skb_mac_gso_segment - mac layer segmentation handler.
2488 * @skb: buffer to segment
2489 * @features: features for the output path (see dev->features)
2491 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2492 netdev_features_t features)
2494 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2495 struct packet_offload *ptype;
2496 int vlan_depth = skb->mac_len;
2497 __be16 type = skb_network_protocol(skb, &vlan_depth);
2499 if (unlikely(!type))
2500 return ERR_PTR(-EINVAL);
2502 __skb_pull(skb, vlan_depth);
2505 list_for_each_entry_rcu(ptype, &offload_base, list) {
2506 if (ptype->type == type && ptype->callbacks.gso_segment) {
2507 segs = ptype->callbacks.gso_segment(skb, features);
2513 __skb_push(skb, skb->data - skb_mac_header(skb));
2517 EXPORT_SYMBOL(skb_mac_gso_segment);
2520 /* openvswitch calls this on rx path, so we need a different check.
2522 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2525 return skb->ip_summed != CHECKSUM_PARTIAL;
2527 return skb->ip_summed == CHECKSUM_NONE;
2531 * __skb_gso_segment - Perform segmentation on skb.
2532 * @skb: buffer to segment
2533 * @features: features for the output path (see dev->features)
2534 * @tx_path: whether it is called in TX path
2536 * This function segments the given skb and returns a list of segments.
2538 * It may return NULL if the skb requires no segmentation. This is
2539 * only possible when GSO is used for verifying header integrity.
2541 * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2543 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2544 netdev_features_t features, bool tx_path)
2546 if (unlikely(skb_needs_check(skb, tx_path))) {
2549 skb_warn_bad_offload(skb);
2551 err = skb_cow_head(skb, 0);
2553 return ERR_PTR(err);
2556 BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2557 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2559 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2560 SKB_GSO_CB(skb)->encap_level = 0;
2562 skb_reset_mac_header(skb);
2563 skb_reset_mac_len(skb);
2565 return skb_mac_gso_segment(skb, features);
2567 EXPORT_SYMBOL(__skb_gso_segment);
2569 /* Take action when hardware reception checksum errors are detected. */
2571 void netdev_rx_csum_fault(struct net_device *dev)
2573 if (net_ratelimit()) {
2574 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2578 EXPORT_SYMBOL(netdev_rx_csum_fault);
2581 /* Actually, we should eliminate this check as soon as we know, that:
2582 * 1. IOMMU is present and allows to map all the memory.
2583 * 2. No high memory really exists on this machine.
2586 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2588 #ifdef CONFIG_HIGHMEM
2590 if (!(dev->features & NETIF_F_HIGHDMA)) {
2591 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2592 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2593 if (PageHighMem(skb_frag_page(frag)))
2598 if (PCI_DMA_BUS_IS_PHYS) {
2599 struct device *pdev = dev->dev.parent;
2603 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2604 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2605 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2606 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2614 /* If MPLS offload request, verify we are testing hardware MPLS features
2615 * instead of standard features for the netdev.
2617 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2618 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2619 netdev_features_t features,
2622 if (eth_p_mpls(type))
2623 features &= skb->dev->mpls_features;
2628 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2629 netdev_features_t features,
2636 static netdev_features_t harmonize_features(struct sk_buff *skb,
2637 netdev_features_t features)
2642 type = skb_network_protocol(skb, &tmp);
2643 features = net_mpls_features(skb, features, type);
2645 if (skb->ip_summed != CHECKSUM_NONE &&
2646 !can_checksum_protocol(features, type)) {
2647 features &= ~NETIF_F_ALL_CSUM;
2649 if (illegal_highdma(skb->dev, skb))
2650 features &= ~NETIF_F_SG;
2655 netdev_features_t passthru_features_check(struct sk_buff *skb,
2656 struct net_device *dev,
2657 netdev_features_t features)
2661 EXPORT_SYMBOL(passthru_features_check);
2663 static netdev_features_t dflt_features_check(const struct sk_buff *skb,
2664 struct net_device *dev,
2665 netdev_features_t features)
2667 return vlan_features_check(skb, features);
2670 netdev_features_t netif_skb_features(struct sk_buff *skb)
2672 struct net_device *dev = skb->dev;
2673 netdev_features_t features = dev->features;
2674 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2676 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2677 features &= ~NETIF_F_GSO_MASK;
2679 /* If encapsulation offload request, verify we are testing
2680 * hardware encapsulation features instead of standard
2681 * features for the netdev
2683 if (skb->encapsulation)
2684 features &= dev->hw_enc_features;
2686 if (skb_vlan_tagged(skb))
2687 features = netdev_intersect_features(features,
2688 dev->vlan_features |
2689 NETIF_F_HW_VLAN_CTAG_TX |
2690 NETIF_F_HW_VLAN_STAG_TX);
2692 if (dev->netdev_ops->ndo_features_check)
2693 features &= dev->netdev_ops->ndo_features_check(skb, dev,
2696 features &= dflt_features_check(skb, dev, features);
2698 return harmonize_features(skb, features);
2700 EXPORT_SYMBOL(netif_skb_features);
2702 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2703 struct netdev_queue *txq, bool more)
2708 if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
2709 dev_queue_xmit_nit(skb, dev);
2712 trace_net_dev_start_xmit(skb, dev);
2713 rc = netdev_start_xmit(skb, dev, txq, more);
2714 trace_net_dev_xmit(skb, rc, dev, len);
2719 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2720 struct netdev_queue *txq, int *ret)
2722 struct sk_buff *skb = first;
2723 int rc = NETDEV_TX_OK;
2726 struct sk_buff *next = skb->next;
2729 rc = xmit_one(skb, dev, txq, next != NULL);
2730 if (unlikely(!dev_xmit_complete(rc))) {
2736 if (netif_xmit_stopped(txq) && skb) {
2737 rc = NETDEV_TX_BUSY;
2747 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2748 netdev_features_t features)
2750 if (skb_vlan_tag_present(skb) &&
2751 !vlan_hw_offload_capable(features, skb->vlan_proto))
2752 skb = __vlan_hwaccel_push_inside(skb);
2756 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2758 netdev_features_t features;
2763 features = netif_skb_features(skb);
2764 skb = validate_xmit_vlan(skb, features);
2768 if (netif_needs_gso(skb, features)) {
2769 struct sk_buff *segs;
2771 segs = skb_gso_segment(skb, features);
2779 if (skb_needs_linearize(skb, features) &&
2780 __skb_linearize(skb))
2783 /* If packet is not checksummed and device does not
2784 * support checksumming for this protocol, complete
2785 * checksumming here.
2787 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2788 if (skb->encapsulation)
2789 skb_set_inner_transport_header(skb,
2790 skb_checksum_start_offset(skb));
2792 skb_set_transport_header(skb,
2793 skb_checksum_start_offset(skb));
2794 if (!(features & NETIF_F_ALL_CSUM) &&
2795 skb_checksum_help(skb))
2808 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2810 struct sk_buff *next, *head = NULL, *tail;
2812 for (; skb != NULL; skb = next) {
2816 /* in case skb wont be segmented, point to itself */
2819 skb = validate_xmit_skb(skb, dev);
2827 /* If skb was segmented, skb->prev points to
2828 * the last segment. If not, it still contains skb.
2834 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
2836 static void qdisc_pkt_len_init(struct sk_buff *skb)
2838 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2840 qdisc_skb_cb(skb)->pkt_len = skb->len;
2842 /* To get more precise estimation of bytes sent on wire,
2843 * we add to pkt_len the headers size of all segments
2845 if (shinfo->gso_size) {
2846 unsigned int hdr_len;
2847 u16 gso_segs = shinfo->gso_segs;
2849 /* mac layer + network layer */
2850 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2852 /* + transport layer */
2853 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2854 hdr_len += tcp_hdrlen(skb);
2856 hdr_len += sizeof(struct udphdr);
2858 if (shinfo->gso_type & SKB_GSO_DODGY)
2859 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2862 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2866 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2867 struct net_device *dev,
2868 struct netdev_queue *txq)
2870 spinlock_t *root_lock = qdisc_lock(q);
2874 qdisc_pkt_len_init(skb);
2875 qdisc_calculate_pkt_len(skb, q);
2877 * Heuristic to force contended enqueues to serialize on a
2878 * separate lock before trying to get qdisc main lock.
2879 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2880 * often and dequeue packets faster.
2882 contended = qdisc_is_running(q);
2883 if (unlikely(contended))
2884 spin_lock(&q->busylock);
2886 spin_lock(root_lock);
2887 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2890 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2891 qdisc_run_begin(q)) {
2893 * This is a work-conserving queue; there are no old skbs
2894 * waiting to be sent out; and the qdisc is not running -
2895 * xmit the skb directly.
2898 qdisc_bstats_update(q, skb);
2900 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2901 if (unlikely(contended)) {
2902 spin_unlock(&q->busylock);
2909 rc = NET_XMIT_SUCCESS;
2911 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2912 if (qdisc_run_begin(q)) {
2913 if (unlikely(contended)) {
2914 spin_unlock(&q->busylock);
2920 spin_unlock(root_lock);
2921 if (unlikely(contended))
2922 spin_unlock(&q->busylock);
2926 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2927 static void skb_update_prio(struct sk_buff *skb)
2929 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2931 if (!skb->priority && skb->sk && map) {
2932 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2934 if (prioidx < map->priomap_len)
2935 skb->priority = map->priomap[prioidx];
2939 #define skb_update_prio(skb)
2942 DEFINE_PER_CPU(int, xmit_recursion);
2943 EXPORT_SYMBOL(xmit_recursion);
2945 #define RECURSION_LIMIT 10
2948 * dev_loopback_xmit - loop back @skb
2949 * @net: network namespace this loopback is happening in
2950 * @sk: sk needed to be a netfilter okfn
2951 * @skb: buffer to transmit
2953 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
2955 skb_reset_mac_header(skb);
2956 __skb_pull(skb, skb_network_offset(skb));
2957 skb->pkt_type = PACKET_LOOPBACK;
2958 skb->ip_summed = CHECKSUM_UNNECESSARY;
2959 WARN_ON(!skb_dst(skb));
2964 EXPORT_SYMBOL(dev_loopback_xmit);
2966 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2969 struct xps_dev_maps *dev_maps;
2970 struct xps_map *map;
2971 int queue_index = -1;
2974 dev_maps = rcu_dereference(dev->xps_maps);
2976 map = rcu_dereference(
2977 dev_maps->cpu_map[skb->sender_cpu - 1]);
2980 queue_index = map->queues[0];
2982 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
2984 if (unlikely(queue_index >= dev->real_num_tx_queues))
2996 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
2998 struct sock *sk = skb->sk;
2999 int queue_index = sk_tx_queue_get(sk);
3001 if (queue_index < 0 || skb->ooo_okay ||
3002 queue_index >= dev->real_num_tx_queues) {
3003 int new_index = get_xps_queue(dev, skb);
3005 new_index = skb_tx_hash(dev, skb);
3007 if (queue_index != new_index && sk &&
3009 rcu_access_pointer(sk->sk_dst_cache))
3010 sk_tx_queue_set(sk, new_index);
3012 queue_index = new_index;
3018 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3019 struct sk_buff *skb,
3022 int queue_index = 0;
3025 if (skb->sender_cpu == 0)
3026 skb->sender_cpu = raw_smp_processor_id() + 1;
3029 if (dev->real_num_tx_queues != 1) {
3030 const struct net_device_ops *ops = dev->netdev_ops;
3031 if (ops->ndo_select_queue)
3032 queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3035 queue_index = __netdev_pick_tx(dev, skb);
3038 queue_index = netdev_cap_txqueue(dev, queue_index);
3041 skb_set_queue_mapping(skb, queue_index);
3042 return netdev_get_tx_queue(dev, queue_index);
3046 * __dev_queue_xmit - transmit a buffer
3047 * @skb: buffer to transmit
3048 * @accel_priv: private data used for L2 forwarding offload
3050 * Queue a buffer for transmission to a network device. The caller must
3051 * have set the device and priority and built the buffer before calling
3052 * this function. The function can be called from an interrupt.
3054 * A negative errno code is returned on a failure. A success does not
3055 * guarantee the frame will be transmitted as it may be dropped due
3056 * to congestion or traffic shaping.
3058 * -----------------------------------------------------------------------------------
3059 * I notice this method can also return errors from the queue disciplines,
3060 * including NET_XMIT_DROP, which is a positive value. So, errors can also
3063 * Regardless of the return value, the skb is consumed, so it is currently
3064 * difficult to retry a send to this method. (You can bump the ref count
3065 * before sending to hold a reference for retry if you are careful.)
3067 * When calling this method, interrupts MUST be enabled. This is because
3068 * the BH enable code must have IRQs enabled so that it will not deadlock.
3071 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3073 struct net_device *dev = skb->dev;
3074 struct netdev_queue *txq;
3078 skb_reset_mac_header(skb);
3080 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3081 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3083 /* Disable soft irqs for various locks below. Also
3084 * stops preemption for RCU.
3088 skb_update_prio(skb);
3090 /* If device/qdisc don't need skb->dst, release it right now while
3091 * its hot in this cpu cache.
3093 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3098 #ifdef CONFIG_NET_SWITCHDEV
3099 /* Don't forward if offload device already forwarded */
3100 if (skb->offload_fwd_mark &&
3101 skb->offload_fwd_mark == dev->offload_fwd_mark) {
3103 rc = NET_XMIT_SUCCESS;
3108 txq = netdev_pick_tx(dev, skb, accel_priv);
3109 q = rcu_dereference_bh(txq->qdisc);
3111 #ifdef CONFIG_NET_CLS_ACT
3112 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
3114 trace_net_dev_queue(skb);
3116 rc = __dev_xmit_skb(skb, q, dev, txq);
3120 /* The device has no queue. Common case for software devices:
3121 loopback, all the sorts of tunnels...
3123 Really, it is unlikely that netif_tx_lock protection is necessary
3124 here. (f.e. loopback and IP tunnels are clean ignoring statistics
3126 However, it is possible, that they rely on protection
3129 Check this and shot the lock. It is not prone from deadlocks.
3130 Either shot noqueue qdisc, it is even simpler 8)
3132 if (dev->flags & IFF_UP) {
3133 int cpu = smp_processor_id(); /* ok because BHs are off */
3135 if (txq->xmit_lock_owner != cpu) {
3137 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
3138 goto recursion_alert;
3140 skb = validate_xmit_skb(skb, dev);
3144 HARD_TX_LOCK(dev, txq, cpu);
3146 if (!netif_xmit_stopped(txq)) {
3147 __this_cpu_inc(xmit_recursion);
3148 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3149 __this_cpu_dec(xmit_recursion);
3150 if (dev_xmit_complete(rc)) {
3151 HARD_TX_UNLOCK(dev, txq);
3155 HARD_TX_UNLOCK(dev, txq);
3156 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3159 /* Recursion is detected! It is possible,
3163 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3170 rcu_read_unlock_bh();
3172 atomic_long_inc(&dev->tx_dropped);
3173 kfree_skb_list(skb);
3176 rcu_read_unlock_bh();
3180 int dev_queue_xmit(struct sk_buff *skb)
3182 return __dev_queue_xmit(skb, NULL);
3184 EXPORT_SYMBOL(dev_queue_xmit);
3186 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3188 return __dev_queue_xmit(skb, accel_priv);
3190 EXPORT_SYMBOL(dev_queue_xmit_accel);
3193 /*=======================================================================
3195 =======================================================================*/
3197 int netdev_max_backlog __read_mostly = 1000;
3198 EXPORT_SYMBOL(netdev_max_backlog);
3200 int netdev_tstamp_prequeue __read_mostly = 1;
3201 int netdev_budget __read_mostly = 300;
3202 int weight_p __read_mostly = 64; /* old backlog weight */
3204 /* Called with irq disabled */
3205 static inline void ____napi_schedule(struct softnet_data *sd,
3206 struct napi_struct *napi)
3208 list_add_tail(&napi->poll_list, &sd->poll_list);
3209 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3214 /* One global table that all flow-based protocols share. */
3215 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3216 EXPORT_SYMBOL(rps_sock_flow_table);
3217 u32 rps_cpu_mask __read_mostly;
3218 EXPORT_SYMBOL(rps_cpu_mask);
3220 struct static_key rps_needed __read_mostly;
3222 static struct rps_dev_flow *
3223 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3224 struct rps_dev_flow *rflow, u16 next_cpu)
3226 if (next_cpu < nr_cpu_ids) {
3227 #ifdef CONFIG_RFS_ACCEL
3228 struct netdev_rx_queue *rxqueue;
3229 struct rps_dev_flow_table *flow_table;
3230 struct rps_dev_flow *old_rflow;
3235 /* Should we steer this flow to a different hardware queue? */
3236 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3237 !(dev->features & NETIF_F_NTUPLE))
3239 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3240 if (rxq_index == skb_get_rx_queue(skb))
3243 rxqueue = dev->_rx + rxq_index;
3244 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3247 flow_id = skb_get_hash(skb) & flow_table->mask;
3248 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3249 rxq_index, flow_id);
3253 rflow = &flow_table->flows[flow_id];
3255 if (old_rflow->filter == rflow->filter)
3256 old_rflow->filter = RPS_NO_FILTER;
3260 per_cpu(softnet_data, next_cpu).input_queue_head;
3263 rflow->cpu = next_cpu;
3268 * get_rps_cpu is called from netif_receive_skb and returns the target
3269 * CPU from the RPS map of the receiving queue for a given skb.
3270 * rcu_read_lock must be held on entry.
3272 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3273 struct rps_dev_flow **rflowp)
3275 const struct rps_sock_flow_table *sock_flow_table;
3276 struct netdev_rx_queue *rxqueue = dev->_rx;
3277 struct rps_dev_flow_table *flow_table;
3278 struct rps_map *map;
3283 if (skb_rx_queue_recorded(skb)) {
3284 u16 index = skb_get_rx_queue(skb);
3286 if (unlikely(index >= dev->real_num_rx_queues)) {
3287 WARN_ONCE(dev->real_num_rx_queues > 1,
3288 "%s received packet on queue %u, but number "
3289 "of RX queues is %u\n",
3290 dev->name, index, dev->real_num_rx_queues);
3296 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3298 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3299 map = rcu_dereference(rxqueue->rps_map);
3300 if (!flow_table && !map)
3303 skb_reset_network_header(skb);
3304 hash = skb_get_hash(skb);
3308 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3309 if (flow_table && sock_flow_table) {
3310 struct rps_dev_flow *rflow;
3314 /* First check into global flow table if there is a match */
3315 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3316 if ((ident ^ hash) & ~rps_cpu_mask)
3319 next_cpu = ident & rps_cpu_mask;
3321 /* OK, now we know there is a match,
3322 * we can look at the local (per receive queue) flow table
3324 rflow = &flow_table->flows[hash & flow_table->mask];
3328 * If the desired CPU (where last recvmsg was done) is
3329 * different from current CPU (one in the rx-queue flow
3330 * table entry), switch if one of the following holds:
3331 * - Current CPU is unset (>= nr_cpu_ids).
3332 * - Current CPU is offline.
3333 * - The current CPU's queue tail has advanced beyond the
3334 * last packet that was enqueued using this table entry.
3335 * This guarantees that all previous packets for the flow
3336 * have been dequeued, thus preserving in order delivery.
3338 if (unlikely(tcpu != next_cpu) &&
3339 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3340 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3341 rflow->last_qtail)) >= 0)) {
3343 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3346 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3356 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3357 if (cpu_online(tcpu)) {
3367 #ifdef CONFIG_RFS_ACCEL
3370 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3371 * @dev: Device on which the filter was set
3372 * @rxq_index: RX queue index
3373 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3374 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3376 * Drivers that implement ndo_rx_flow_steer() should periodically call
3377 * this function for each installed filter and remove the filters for
3378 * which it returns %true.
3380 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3381 u32 flow_id, u16 filter_id)
3383 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3384 struct rps_dev_flow_table *flow_table;
3385 struct rps_dev_flow *rflow;
3390 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3391 if (flow_table && flow_id <= flow_table->mask) {
3392 rflow = &flow_table->flows[flow_id];
3393 cpu = ACCESS_ONCE(rflow->cpu);
3394 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3395 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3396 rflow->last_qtail) <
3397 (int)(10 * flow_table->mask)))
3403 EXPORT_SYMBOL(rps_may_expire_flow);
3405 #endif /* CONFIG_RFS_ACCEL */
3407 /* Called from hardirq (IPI) context */
3408 static void rps_trigger_softirq(void *data)
3410 struct softnet_data *sd = data;
3412 ____napi_schedule(sd, &sd->backlog);
3416 #endif /* CONFIG_RPS */
3419 * Check if this softnet_data structure is another cpu one
3420 * If yes, queue it to our IPI list and return 1
3423 static int rps_ipi_queued(struct softnet_data *sd)
3426 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3429 sd->rps_ipi_next = mysd->rps_ipi_list;
3430 mysd->rps_ipi_list = sd;
3432 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3435 #endif /* CONFIG_RPS */
3439 #ifdef CONFIG_NET_FLOW_LIMIT
3440 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3443 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3445 #ifdef CONFIG_NET_FLOW_LIMIT
3446 struct sd_flow_limit *fl;
3447 struct softnet_data *sd;
3448 unsigned int old_flow, new_flow;
3450 if (qlen < (netdev_max_backlog >> 1))
3453 sd = this_cpu_ptr(&softnet_data);
3456 fl = rcu_dereference(sd->flow_limit);
3458 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3459 old_flow = fl->history[fl->history_head];
3460 fl->history[fl->history_head] = new_flow;
3463 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3465 if (likely(fl->buckets[old_flow]))
3466 fl->buckets[old_flow]--;
3468 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3480 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3481 * queue (may be a remote CPU queue).
3483 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3484 unsigned int *qtail)
3486 struct softnet_data *sd;
3487 unsigned long flags;
3490 sd = &per_cpu(softnet_data, cpu);
3492 local_irq_save(flags);
3495 if (!netif_running(skb->dev))
3497 qlen = skb_queue_len(&sd->input_pkt_queue);
3498 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3501 __skb_queue_tail(&sd->input_pkt_queue, skb);
3502 input_queue_tail_incr_save(sd, qtail);
3504 local_irq_restore(flags);
3505 return NET_RX_SUCCESS;
3508 /* Schedule NAPI for backlog device
3509 * We can use non atomic operation since we own the queue lock
3511 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3512 if (!rps_ipi_queued(sd))
3513 ____napi_schedule(sd, &sd->backlog);
3522 local_irq_restore(flags);
3524 atomic_long_inc(&skb->dev->rx_dropped);
3529 static int netif_rx_internal(struct sk_buff *skb)
3533 net_timestamp_check(netdev_tstamp_prequeue, skb);
3535 trace_netif_rx(skb);
3537 if (static_key_false(&rps_needed)) {
3538 struct rps_dev_flow voidflow, *rflow = &voidflow;
3544 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3546 cpu = smp_processor_id();
3548 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3556 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3563 * netif_rx - post buffer to the network code
3564 * @skb: buffer to post
3566 * This function receives a packet from a device driver and queues it for
3567 * the upper (protocol) levels to process. It always succeeds. The buffer
3568 * may be dropped during processing for congestion control or by the
3572 * NET_RX_SUCCESS (no congestion)
3573 * NET_RX_DROP (packet was dropped)
3577 int netif_rx(struct sk_buff *skb)
3579 trace_netif_rx_entry(skb);
3581 return netif_rx_internal(skb);
3583 EXPORT_SYMBOL(netif_rx);
3585 int netif_rx_ni(struct sk_buff *skb)
3589 trace_netif_rx_ni_entry(skb);
3592 err = netif_rx_internal(skb);
3593 if (local_softirq_pending())
3599 EXPORT_SYMBOL(netif_rx_ni);
3601 static void net_tx_action(struct softirq_action *h)
3603 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3605 if (sd->completion_queue) {
3606 struct sk_buff *clist;
3608 local_irq_disable();
3609 clist = sd->completion_queue;
3610 sd->completion_queue = NULL;
3614 struct sk_buff *skb = clist;
3615 clist = clist->next;
3617 WARN_ON(atomic_read(&skb->users));
3618 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3619 trace_consume_skb(skb);
3621 trace_kfree_skb(skb, net_tx_action);
3626 if (sd->output_queue) {
3629 local_irq_disable();
3630 head = sd->output_queue;
3631 sd->output_queue = NULL;
3632 sd->output_queue_tailp = &sd->output_queue;
3636 struct Qdisc *q = head;
3637 spinlock_t *root_lock;
3639 head = head->next_sched;
3641 root_lock = qdisc_lock(q);
3642 if (spin_trylock(root_lock)) {
3643 smp_mb__before_atomic();
3644 clear_bit(__QDISC_STATE_SCHED,
3647 spin_unlock(root_lock);
3649 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3651 __netif_reschedule(q);
3653 smp_mb__before_atomic();
3654 clear_bit(__QDISC_STATE_SCHED,
3662 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3663 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3664 /* This hook is defined here for ATM LANE */
3665 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3666 unsigned char *addr) __read_mostly;
3667 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3670 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3671 struct packet_type **pt_prev,
3672 int *ret, struct net_device *orig_dev)
3674 #ifdef CONFIG_NET_CLS_ACT
3675 struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list);
3676 struct tcf_result cl_res;
3678 /* If there's at least one ingress present somewhere (so
3679 * we get here via enabled static key), remaining devices
3680 * that are not configured with an ingress qdisc will bail
3686 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3690 qdisc_skb_cb(skb)->pkt_len = skb->len;
3691 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3692 qdisc_bstats_cpu_update(cl->q, skb);
3694 switch (tc_classify(skb, cl, &cl_res, false)) {
3696 case TC_ACT_RECLASSIFY:
3697 skb->tc_index = TC_H_MIN(cl_res.classid);
3700 qdisc_qstats_cpu_drop(cl->q);
3705 case TC_ACT_REDIRECT:
3706 /* skb_mac_header check was done by cls/act_bpf, so
3707 * we can safely push the L2 header back before
3708 * redirecting to another netdev
3710 __skb_push(skb, skb->mac_len);
3711 skb_do_redirect(skb);
3716 #endif /* CONFIG_NET_CLS_ACT */
3721 * netdev_is_rx_handler_busy - check if receive handler is registered
3722 * @dev: device to check
3724 * Check if a receive handler is already registered for a given device.
3725 * Return true if there one.
3727 * The caller must hold the rtnl_mutex.
3729 bool netdev_is_rx_handler_busy(struct net_device *dev)
3732 return dev && rtnl_dereference(dev->rx_handler);
3734 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
3737 * netdev_rx_handler_register - register receive handler
3738 * @dev: device to register a handler for
3739 * @rx_handler: receive handler to register
3740 * @rx_handler_data: data pointer that is used by rx handler
3742 * Register a receive handler for a device. This handler will then be
3743 * called from __netif_receive_skb. A negative errno code is returned
3746 * The caller must hold the rtnl_mutex.
3748 * For a general description of rx_handler, see enum rx_handler_result.
3750 int netdev_rx_handler_register(struct net_device *dev,
3751 rx_handler_func_t *rx_handler,
3752 void *rx_handler_data)
3756 if (dev->rx_handler)
3759 /* Note: rx_handler_data must be set before rx_handler */
3760 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3761 rcu_assign_pointer(dev->rx_handler, rx_handler);
3765 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3768 * netdev_rx_handler_unregister - unregister receive handler
3769 * @dev: device to unregister a handler from
3771 * Unregister a receive handler from a device.
3773 * The caller must hold the rtnl_mutex.
3775 void netdev_rx_handler_unregister(struct net_device *dev)
3779 RCU_INIT_POINTER(dev->rx_handler, NULL);
3780 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3781 * section has a guarantee to see a non NULL rx_handler_data
3785 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3787 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3790 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3791 * the special handling of PFMEMALLOC skbs.
3793 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3795 switch (skb->protocol) {
3796 case htons(ETH_P_ARP):
3797 case htons(ETH_P_IP):
3798 case htons(ETH_P_IPV6):
3799 case htons(ETH_P_8021Q):
3800 case htons(ETH_P_8021AD):
3807 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
3808 int *ret, struct net_device *orig_dev)
3810 #ifdef CONFIG_NETFILTER_INGRESS
3811 if (nf_hook_ingress_active(skb)) {
3813 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3817 return nf_hook_ingress(skb);
3819 #endif /* CONFIG_NETFILTER_INGRESS */
3823 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3825 struct packet_type *ptype, *pt_prev;
3826 rx_handler_func_t *rx_handler;
3827 struct net_device *orig_dev;
3828 bool deliver_exact = false;
3829 int ret = NET_RX_DROP;
3832 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3834 trace_netif_receive_skb(skb);
3836 orig_dev = skb->dev;
3838 skb_reset_network_header(skb);
3839 if (!skb_transport_header_was_set(skb))
3840 skb_reset_transport_header(skb);
3841 skb_reset_mac_len(skb);
3846 skb->skb_iif = skb->dev->ifindex;
3848 __this_cpu_inc(softnet_data.processed);
3850 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3851 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3852 skb = skb_vlan_untag(skb);
3857 #ifdef CONFIG_NET_CLS_ACT
3858 if (skb->tc_verd & TC_NCLS) {
3859 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3867 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3869 ret = deliver_skb(skb, pt_prev, orig_dev);
3873 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
3875 ret = deliver_skb(skb, pt_prev, orig_dev);
3880 #ifdef CONFIG_NET_INGRESS
3881 if (static_key_false(&ingress_needed)) {
3882 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3886 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
3890 #ifdef CONFIG_NET_CLS_ACT
3894 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3897 if (skb_vlan_tag_present(skb)) {
3899 ret = deliver_skb(skb, pt_prev, orig_dev);
3902 if (vlan_do_receive(&skb))
3904 else if (unlikely(!skb))
3908 rx_handler = rcu_dereference(skb->dev->rx_handler);
3911 ret = deliver_skb(skb, pt_prev, orig_dev);
3914 switch (rx_handler(&skb)) {
3915 case RX_HANDLER_CONSUMED:
3916 ret = NET_RX_SUCCESS;
3918 case RX_HANDLER_ANOTHER:
3920 case RX_HANDLER_EXACT:
3921 deliver_exact = true;
3922 case RX_HANDLER_PASS:
3929 if (unlikely(skb_vlan_tag_present(skb))) {
3930 if (skb_vlan_tag_get_id(skb))
3931 skb->pkt_type = PACKET_OTHERHOST;
3932 /* Note: we might in the future use prio bits
3933 * and set skb->priority like in vlan_do_receive()
3934 * For the time being, just ignore Priority Code Point
3939 type = skb->protocol;
3941 /* deliver only exact match when indicated */
3942 if (likely(!deliver_exact)) {
3943 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3944 &ptype_base[ntohs(type) &
3948 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3949 &orig_dev->ptype_specific);
3951 if (unlikely(skb->dev != orig_dev)) {
3952 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
3953 &skb->dev->ptype_specific);
3957 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3960 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3963 atomic_long_inc(&skb->dev->rx_dropped);
3965 /* Jamal, now you will not able to escape explaining
3966 * me how you were going to use this. :-)
3975 static int __netif_receive_skb(struct sk_buff *skb)
3979 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3980 unsigned long pflags = current->flags;
3983 * PFMEMALLOC skbs are special, they should
3984 * - be delivered to SOCK_MEMALLOC sockets only
3985 * - stay away from userspace
3986 * - have bounded memory usage
3988 * Use PF_MEMALLOC as this saves us from propagating the allocation
3989 * context down to all allocation sites.
3991 current->flags |= PF_MEMALLOC;
3992 ret = __netif_receive_skb_core(skb, true);
3993 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3995 ret = __netif_receive_skb_core(skb, false);
4000 static int netif_receive_skb_internal(struct sk_buff *skb)
4004 net_timestamp_check(netdev_tstamp_prequeue, skb);
4006 if (skb_defer_rx_timestamp(skb))
4007 return NET_RX_SUCCESS;
4012 if (static_key_false(&rps_needed)) {
4013 struct rps_dev_flow voidflow, *rflow = &voidflow;
4014 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4017 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4023 ret = __netif_receive_skb(skb);
4029 * netif_receive_skb - process receive buffer from network
4030 * @skb: buffer to process
4032 * netif_receive_skb() is the main receive data processing function.
4033 * It always succeeds. The buffer may be dropped during processing
4034 * for congestion control or by the protocol layers.
4036 * This function may only be called from softirq context and interrupts
4037 * should be enabled.
4039 * Return values (usually ignored):
4040 * NET_RX_SUCCESS: no congestion
4041 * NET_RX_DROP: packet was dropped
4043 int netif_receive_skb(struct sk_buff *skb)
4045 trace_netif_receive_skb_entry(skb);
4047 return netif_receive_skb_internal(skb);
4049 EXPORT_SYMBOL(netif_receive_skb);
4051 /* Network device is going away, flush any packets still pending
4052 * Called with irqs disabled.
4054 static void flush_backlog(void *arg)
4056 struct net_device *dev = arg;
4057 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4058 struct sk_buff *skb, *tmp;
4061 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4062 if (skb->dev == dev) {
4063 __skb_unlink(skb, &sd->input_pkt_queue);
4065 input_queue_head_incr(sd);
4070 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4071 if (skb->dev == dev) {
4072 __skb_unlink(skb, &sd->process_queue);
4074 input_queue_head_incr(sd);
4079 static int napi_gro_complete(struct sk_buff *skb)
4081 struct packet_offload *ptype;
4082 __be16 type = skb->protocol;
4083 struct list_head *head = &offload_base;
4086 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4088 if (NAPI_GRO_CB(skb)->count == 1) {
4089 skb_shinfo(skb)->gso_size = 0;
4094 list_for_each_entry_rcu(ptype, head, list) {
4095 if (ptype->type != type || !ptype->callbacks.gro_complete)
4098 err = ptype->callbacks.gro_complete(skb, 0);
4104 WARN_ON(&ptype->list == head);
4106 return NET_RX_SUCCESS;
4110 return netif_receive_skb_internal(skb);
4113 /* napi->gro_list contains packets ordered by age.
4114 * youngest packets at the head of it.
4115 * Complete skbs in reverse order to reduce latencies.
4117 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4119 struct sk_buff *skb, *prev = NULL;
4121 /* scan list and build reverse chain */
4122 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4127 for (skb = prev; skb; skb = prev) {
4130 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4134 napi_gro_complete(skb);
4138 napi->gro_list = NULL;
4140 EXPORT_SYMBOL(napi_gro_flush);
4142 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4145 unsigned int maclen = skb->dev->hard_header_len;
4146 u32 hash = skb_get_hash_raw(skb);
4148 for (p = napi->gro_list; p; p = p->next) {
4149 unsigned long diffs;
4151 NAPI_GRO_CB(p)->flush = 0;
4153 if (hash != skb_get_hash_raw(p)) {
4154 NAPI_GRO_CB(p)->same_flow = 0;
4158 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4159 diffs |= p->vlan_tci ^ skb->vlan_tci;
4160 diffs |= skb_metadata_dst_cmp(p, skb);
4161 if (maclen == ETH_HLEN)
4162 diffs |= compare_ether_header(skb_mac_header(p),
4163 skb_mac_header(skb));
4165 diffs = memcmp(skb_mac_header(p),
4166 skb_mac_header(skb),
4168 NAPI_GRO_CB(p)->same_flow = !diffs;
4172 static void skb_gro_reset_offset(struct sk_buff *skb)
4174 const struct skb_shared_info *pinfo = skb_shinfo(skb);
4175 const skb_frag_t *frag0 = &pinfo->frags[0];
4177 NAPI_GRO_CB(skb)->data_offset = 0;
4178 NAPI_GRO_CB(skb)->frag0 = NULL;
4179 NAPI_GRO_CB(skb)->frag0_len = 0;
4181 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4183 !PageHighMem(skb_frag_page(frag0))) {
4184 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4185 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
4186 skb_frag_size(frag0),
4187 skb->end - skb->tail);
4191 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4193 struct skb_shared_info *pinfo = skb_shinfo(skb);
4195 BUG_ON(skb->end - skb->tail < grow);
4197 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4199 skb->data_len -= grow;
4202 pinfo->frags[0].page_offset += grow;
4203 skb_frag_size_sub(&pinfo->frags[0], grow);
4205 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4206 skb_frag_unref(skb, 0);
4207 memmove(pinfo->frags, pinfo->frags + 1,
4208 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4212 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4214 struct sk_buff **pp = NULL;
4215 struct packet_offload *ptype;
4216 __be16 type = skb->protocol;
4217 struct list_head *head = &offload_base;
4219 enum gro_result ret;
4222 if (!(skb->dev->features & NETIF_F_GRO))
4225 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4228 gro_list_prepare(napi, skb);
4231 list_for_each_entry_rcu(ptype, head, list) {
4232 if (ptype->type != type || !ptype->callbacks.gro_receive)
4235 skb_set_network_header(skb, skb_gro_offset(skb));
4236 skb_reset_mac_len(skb);
4237 NAPI_GRO_CB(skb)->same_flow = 0;
4238 NAPI_GRO_CB(skb)->flush = 0;
4239 NAPI_GRO_CB(skb)->free = 0;
4240 NAPI_GRO_CB(skb)->encap_mark = 0;
4241 NAPI_GRO_CB(skb)->recursion_counter = 0;
4242 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
4244 /* Setup for GRO checksum validation */
4245 switch (skb->ip_summed) {
4246 case CHECKSUM_COMPLETE:
4247 NAPI_GRO_CB(skb)->csum = skb->csum;
4248 NAPI_GRO_CB(skb)->csum_valid = 1;
4249 NAPI_GRO_CB(skb)->csum_cnt = 0;
4251 case CHECKSUM_UNNECESSARY:
4252 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4253 NAPI_GRO_CB(skb)->csum_valid = 0;
4256 NAPI_GRO_CB(skb)->csum_cnt = 0;
4257 NAPI_GRO_CB(skb)->csum_valid = 0;
4260 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4265 if (&ptype->list == head)
4268 same_flow = NAPI_GRO_CB(skb)->same_flow;
4269 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4272 struct sk_buff *nskb = *pp;
4276 napi_gro_complete(nskb);
4283 if (NAPI_GRO_CB(skb)->flush)
4286 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4287 struct sk_buff *nskb = napi->gro_list;
4289 /* locate the end of the list to select the 'oldest' flow */
4290 while (nskb->next) {
4296 napi_gro_complete(nskb);
4300 NAPI_GRO_CB(skb)->count = 1;
4301 NAPI_GRO_CB(skb)->age = jiffies;
4302 NAPI_GRO_CB(skb)->last = skb;
4303 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4304 skb->next = napi->gro_list;
4305 napi->gro_list = skb;
4309 grow = skb_gro_offset(skb) - skb_headlen(skb);
4311 gro_pull_from_frag0(skb, grow);
4320 struct packet_offload *gro_find_receive_by_type(__be16 type)
4322 struct list_head *offload_head = &offload_base;
4323 struct packet_offload *ptype;
4325 list_for_each_entry_rcu(ptype, offload_head, list) {
4326 if (ptype->type != type || !ptype->callbacks.gro_receive)
4332 EXPORT_SYMBOL(gro_find_receive_by_type);
4334 struct packet_offload *gro_find_complete_by_type(__be16 type)
4336 struct list_head *offload_head = &offload_base;
4337 struct packet_offload *ptype;
4339 list_for_each_entry_rcu(ptype, offload_head, list) {
4340 if (ptype->type != type || !ptype->callbacks.gro_complete)
4346 EXPORT_SYMBOL(gro_find_complete_by_type);
4348 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4352 if (netif_receive_skb_internal(skb))
4360 case GRO_MERGED_FREE:
4361 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) {
4363 kmem_cache_free(skbuff_head_cache, skb);
4377 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4379 trace_napi_gro_receive_entry(skb);
4381 skb_gro_reset_offset(skb);
4383 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4385 EXPORT_SYMBOL(napi_gro_receive);
4387 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4389 if (unlikely(skb->pfmemalloc)) {
4393 __skb_pull(skb, skb_headlen(skb));
4394 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4395 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4397 skb->dev = napi->dev;
4399 skb->encapsulation = 0;
4400 skb_shinfo(skb)->gso_type = 0;
4401 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4406 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4408 struct sk_buff *skb = napi->skb;
4411 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
4416 EXPORT_SYMBOL(napi_get_frags);
4418 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4419 struct sk_buff *skb,
4425 __skb_push(skb, ETH_HLEN);
4426 skb->protocol = eth_type_trans(skb, skb->dev);
4427 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4432 case GRO_MERGED_FREE:
4433 napi_reuse_skb(napi, skb);
4443 /* Upper GRO stack assumes network header starts at gro_offset=0
4444 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4445 * We copy ethernet header into skb->data to have a common layout.
4447 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4449 struct sk_buff *skb = napi->skb;
4450 const struct ethhdr *eth;
4451 unsigned int hlen = sizeof(*eth);
4455 skb_reset_mac_header(skb);
4456 skb_gro_reset_offset(skb);
4458 eth = skb_gro_header_fast(skb, 0);
4459 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4460 eth = skb_gro_header_slow(skb, hlen, 0);
4461 if (unlikely(!eth)) {
4462 napi_reuse_skb(napi, skb);
4466 gro_pull_from_frag0(skb, hlen);
4467 NAPI_GRO_CB(skb)->frag0 += hlen;
4468 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4470 __skb_pull(skb, hlen);
4473 * This works because the only protocols we care about don't require
4475 * We'll fix it up properly in napi_frags_finish()
4477 skb->protocol = eth->h_proto;
4482 gro_result_t napi_gro_frags(struct napi_struct *napi)
4484 struct sk_buff *skb = napi_frags_skb(napi);
4489 trace_napi_gro_frags_entry(skb);
4491 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4493 EXPORT_SYMBOL(napi_gro_frags);
4495 /* Compute the checksum from gro_offset and return the folded value
4496 * after adding in any pseudo checksum.
4498 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4503 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4505 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4506 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4508 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4509 !skb->csum_complete_sw)
4510 netdev_rx_csum_fault(skb->dev);
4513 NAPI_GRO_CB(skb)->csum = wsum;
4514 NAPI_GRO_CB(skb)->csum_valid = 1;
4518 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4521 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4522 * Note: called with local irq disabled, but exits with local irq enabled.
4524 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4527 struct softnet_data *remsd = sd->rps_ipi_list;
4530 sd->rps_ipi_list = NULL;
4534 /* Send pending IPI's to kick RPS processing on remote cpus. */
4536 struct softnet_data *next = remsd->rps_ipi_next;
4538 if (cpu_online(remsd->cpu))
4539 smp_call_function_single_async(remsd->cpu,
4548 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
4551 return sd->rps_ipi_list != NULL;
4557 static int process_backlog(struct napi_struct *napi, int quota)
4560 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4562 /* Check if we have pending ipi, its better to send them now,
4563 * not waiting net_rx_action() end.
4565 if (sd_has_rps_ipi_waiting(sd)) {
4566 local_irq_disable();
4567 net_rps_action_and_irq_enable(sd);
4570 napi->weight = weight_p;
4571 local_irq_disable();
4573 struct sk_buff *skb;
4575 while ((skb = __skb_dequeue(&sd->process_queue))) {
4578 __netif_receive_skb(skb);
4580 local_irq_disable();
4581 input_queue_head_incr(sd);
4582 if (++work >= quota) {
4589 if (skb_queue_empty(&sd->input_pkt_queue)) {
4591 * Inline a custom version of __napi_complete().
4592 * only current cpu owns and manipulates this napi,
4593 * and NAPI_STATE_SCHED is the only possible flag set
4595 * We can use a plain write instead of clear_bit(),
4596 * and we dont need an smp_mb() memory barrier.
4604 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4605 &sd->process_queue);
4614 * __napi_schedule - schedule for receive
4615 * @n: entry to schedule
4617 * The entry's receive function will be scheduled to run.
4618 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
4620 void __napi_schedule(struct napi_struct *n)
4622 unsigned long flags;
4624 local_irq_save(flags);
4625 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4626 local_irq_restore(flags);
4628 EXPORT_SYMBOL(__napi_schedule);
4631 * __napi_schedule_irqoff - schedule for receive
4632 * @n: entry to schedule
4634 * Variant of __napi_schedule() assuming hard irqs are masked
4636 void __napi_schedule_irqoff(struct napi_struct *n)
4638 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4640 EXPORT_SYMBOL(__napi_schedule_irqoff);
4642 void __napi_complete(struct napi_struct *n)
4644 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4646 list_del_init(&n->poll_list);
4647 smp_mb__before_atomic();
4648 clear_bit(NAPI_STATE_SCHED, &n->state);
4650 EXPORT_SYMBOL(__napi_complete);
4652 void napi_complete_done(struct napi_struct *n, int work_done)
4654 unsigned long flags;
4657 * don't let napi dequeue from the cpu poll list
4658 * just in case its running on a different cpu
4660 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4664 unsigned long timeout = 0;
4667 timeout = n->dev->gro_flush_timeout;
4670 hrtimer_start(&n->timer, ns_to_ktime(timeout),
4671 HRTIMER_MODE_REL_PINNED);
4673 napi_gro_flush(n, false);
4675 if (likely(list_empty(&n->poll_list))) {
4676 WARN_ON_ONCE(!test_and_clear_bit(NAPI_STATE_SCHED, &n->state));
4678 /* If n->poll_list is not empty, we need to mask irqs */
4679 local_irq_save(flags);
4681 local_irq_restore(flags);
4684 EXPORT_SYMBOL(napi_complete_done);
4686 /* must be called under rcu_read_lock(), as we dont take a reference */
4687 struct napi_struct *napi_by_id(unsigned int napi_id)
4689 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4690 struct napi_struct *napi;
4692 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4693 if (napi->napi_id == napi_id)
4698 EXPORT_SYMBOL_GPL(napi_by_id);
4700 void napi_hash_add(struct napi_struct *napi)
4702 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4704 spin_lock(&napi_hash_lock);
4706 /* 0 is not a valid id, we also skip an id that is taken
4707 * we expect both events to be extremely rare
4710 while (!napi->napi_id) {
4711 napi->napi_id = ++napi_gen_id;
4712 if (napi_by_id(napi->napi_id))
4716 hlist_add_head_rcu(&napi->napi_hash_node,
4717 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4719 spin_unlock(&napi_hash_lock);
4722 EXPORT_SYMBOL_GPL(napi_hash_add);
4724 /* Warning : caller is responsible to make sure rcu grace period
4725 * is respected before freeing memory containing @napi
4727 void napi_hash_del(struct napi_struct *napi)
4729 spin_lock(&napi_hash_lock);
4731 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4732 hlist_del_rcu(&napi->napi_hash_node);
4734 spin_unlock(&napi_hash_lock);
4736 EXPORT_SYMBOL_GPL(napi_hash_del);
4738 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
4740 struct napi_struct *napi;
4742 napi = container_of(timer, struct napi_struct, timer);
4744 napi_schedule(napi);
4746 return HRTIMER_NORESTART;
4749 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4750 int (*poll)(struct napi_struct *, int), int weight)
4752 INIT_LIST_HEAD(&napi->poll_list);
4753 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
4754 napi->timer.function = napi_watchdog;
4755 napi->gro_count = 0;
4756 napi->gro_list = NULL;
4759 if (weight > NAPI_POLL_WEIGHT)
4760 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4762 napi->weight = weight;
4763 list_add(&napi->dev_list, &dev->napi_list);
4765 #ifdef CONFIG_NETPOLL
4766 spin_lock_init(&napi->poll_lock);
4767 napi->poll_owner = -1;
4769 set_bit(NAPI_STATE_SCHED, &napi->state);
4771 EXPORT_SYMBOL(netif_napi_add);
4773 void napi_disable(struct napi_struct *n)
4776 set_bit(NAPI_STATE_DISABLE, &n->state);
4778 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
4780 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
4783 hrtimer_cancel(&n->timer);
4785 clear_bit(NAPI_STATE_DISABLE, &n->state);
4787 EXPORT_SYMBOL(napi_disable);
4789 void netif_napi_del(struct napi_struct *napi)
4791 list_del_init(&napi->dev_list);
4792 napi_free_frags(napi);
4794 kfree_skb_list(napi->gro_list);
4795 napi->gro_list = NULL;
4796 napi->gro_count = 0;
4798 EXPORT_SYMBOL(netif_napi_del);
4800 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
4805 list_del_init(&n->poll_list);
4807 have = netpoll_poll_lock(n);
4811 /* This NAPI_STATE_SCHED test is for avoiding a race
4812 * with netpoll's poll_napi(). Only the entity which
4813 * obtains the lock and sees NAPI_STATE_SCHED set will
4814 * actually make the ->poll() call. Therefore we avoid
4815 * accidentally calling ->poll() when NAPI is not scheduled.
4818 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4819 work = n->poll(n, weight);
4823 WARN_ON_ONCE(work > weight);
4825 if (likely(work < weight))
4828 /* Drivers must not modify the NAPI state if they
4829 * consume the entire weight. In such cases this code
4830 * still "owns" the NAPI instance and therefore can
4831 * move the instance around on the list at-will.
4833 if (unlikely(napi_disable_pending(n))) {
4839 /* flush too old packets
4840 * If HZ < 1000, flush all packets.
4842 napi_gro_flush(n, HZ >= 1000);
4845 /* Some drivers may have called napi_schedule
4846 * prior to exhausting their budget.
4848 if (unlikely(!list_empty(&n->poll_list))) {
4849 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
4850 n->dev ? n->dev->name : "backlog");
4854 list_add_tail(&n->poll_list, repoll);
4857 netpoll_poll_unlock(have);
4862 static void net_rx_action(struct softirq_action *h)
4864 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4865 unsigned long time_limit = jiffies + 2;
4866 int budget = netdev_budget;
4870 local_irq_disable();
4871 list_splice_init(&sd->poll_list, &list);
4875 struct napi_struct *n;
4877 if (list_empty(&list)) {
4878 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
4883 n = list_first_entry(&list, struct napi_struct, poll_list);
4884 budget -= napi_poll(n, &repoll);
4886 /* If softirq window is exhausted then punt.
4887 * Allow this to run for 2 jiffies since which will allow
4888 * an average latency of 1.5/HZ.
4890 if (unlikely(budget <= 0 ||
4891 time_after_eq(jiffies, time_limit))) {
4897 local_irq_disable();
4899 list_splice_tail_init(&sd->poll_list, &list);
4900 list_splice_tail(&repoll, &list);
4901 list_splice(&list, &sd->poll_list);
4902 if (!list_empty(&sd->poll_list))
4903 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4905 net_rps_action_and_irq_enable(sd);
4908 struct netdev_adjacent {
4909 struct net_device *dev;
4911 /* upper master flag, there can only be one master device per list */
4914 /* counter for the number of times this device was added to us */
4917 /* private field for the users */
4920 struct list_head list;
4921 struct rcu_head rcu;
4924 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
4925 struct list_head *adj_list)
4927 struct netdev_adjacent *adj;
4929 list_for_each_entry(adj, adj_list, list) {
4930 if (adj->dev == adj_dev)
4937 * netdev_has_upper_dev - Check if device is linked to an upper device
4939 * @upper_dev: upper device to check
4941 * Find out if a device is linked to specified upper device and return true
4942 * in case it is. Note that this checks only immediate upper device,
4943 * not through a complete stack of devices. The caller must hold the RTNL lock.
4945 bool netdev_has_upper_dev(struct net_device *dev,
4946 struct net_device *upper_dev)
4950 return __netdev_find_adj(upper_dev, &dev->all_adj_list.upper);
4952 EXPORT_SYMBOL(netdev_has_upper_dev);
4955 * netdev_has_any_upper_dev - Check if device is linked to some device
4958 * Find out if a device is linked to an upper device and return true in case
4959 * it is. The caller must hold the RTNL lock.
4961 static bool netdev_has_any_upper_dev(struct net_device *dev)
4965 return !list_empty(&dev->all_adj_list.upper);
4969 * netdev_master_upper_dev_get - Get master upper device
4972 * Find a master upper device and return pointer to it or NULL in case
4973 * it's not there. The caller must hold the RTNL lock.
4975 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4977 struct netdev_adjacent *upper;
4981 if (list_empty(&dev->adj_list.upper))
4984 upper = list_first_entry(&dev->adj_list.upper,
4985 struct netdev_adjacent, list);
4986 if (likely(upper->master))
4990 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4992 void *netdev_adjacent_get_private(struct list_head *adj_list)
4994 struct netdev_adjacent *adj;
4996 adj = list_entry(adj_list, struct netdev_adjacent, list);
4998 return adj->private;
5000 EXPORT_SYMBOL(netdev_adjacent_get_private);
5003 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
5005 * @iter: list_head ** of the current position
5007 * Gets the next device from the dev's upper list, starting from iter
5008 * position. The caller must hold RCU read lock.
5010 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
5011 struct list_head **iter)
5013 struct netdev_adjacent *upper;
5015 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5017 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5019 if (&upper->list == &dev->adj_list.upper)
5022 *iter = &upper->list;
5026 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
5029 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
5031 * @iter: list_head ** of the current position
5033 * Gets the next device from the dev's upper list, starting from iter
5034 * position. The caller must hold RCU read lock.
5036 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
5037 struct list_head **iter)
5039 struct netdev_adjacent *upper;
5041 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
5043 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5045 if (&upper->list == &dev->all_adj_list.upper)
5048 *iter = &upper->list;
5052 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
5055 * netdev_lower_get_next_private - Get the next ->private from the
5056 * lower neighbour list
5058 * @iter: list_head ** of the current position
5060 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5061 * list, starting from iter position. The caller must hold either hold the
5062 * RTNL lock or its own locking that guarantees that the neighbour lower
5063 * list will remain unchanged.
5065 void *netdev_lower_get_next_private(struct net_device *dev,
5066 struct list_head **iter)
5068 struct netdev_adjacent *lower;
5070 lower = list_entry(*iter, struct netdev_adjacent, list);
5072 if (&lower->list == &dev->adj_list.lower)
5075 *iter = lower->list.next;
5077 return lower->private;
5079 EXPORT_SYMBOL(netdev_lower_get_next_private);
5082 * netdev_lower_get_next_private_rcu - Get the next ->private from the
5083 * lower neighbour list, RCU
5086 * @iter: list_head ** of the current position
5088 * Gets the next netdev_adjacent->private from the dev's lower neighbour
5089 * list, starting from iter position. The caller must hold RCU read lock.
5091 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
5092 struct list_head **iter)
5094 struct netdev_adjacent *lower;
5096 WARN_ON_ONCE(!rcu_read_lock_held());
5098 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
5100 if (&lower->list == &dev->adj_list.lower)
5103 *iter = &lower->list;
5105 return lower->private;
5107 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
5110 * netdev_lower_get_next - Get the next device from the lower neighbour
5113 * @iter: list_head ** of the current position
5115 * Gets the next netdev_adjacent from the dev's lower neighbour
5116 * list, starting from iter position. The caller must hold RTNL lock or
5117 * its own locking that guarantees that the neighbour lower
5118 * list will remain unchanged.
5120 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
5122 struct netdev_adjacent *lower;
5124 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
5126 if (&lower->list == &dev->adj_list.lower)
5129 *iter = &lower->list;
5133 EXPORT_SYMBOL(netdev_lower_get_next);
5136 * netdev_lower_get_first_private_rcu - Get the first ->private from the
5137 * lower neighbour list, RCU
5141 * Gets the first netdev_adjacent->private from the dev's lower neighbour
5142 * list. The caller must hold RCU read lock.
5144 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
5146 struct netdev_adjacent *lower;
5148 lower = list_first_or_null_rcu(&dev->adj_list.lower,
5149 struct netdev_adjacent, list);
5151 return lower->private;
5154 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
5157 * netdev_master_upper_dev_get_rcu - Get master upper device
5160 * Find a master upper device and return pointer to it or NULL in case
5161 * it's not there. The caller must hold the RCU read lock.
5163 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
5165 struct netdev_adjacent *upper;
5167 upper = list_first_or_null_rcu(&dev->adj_list.upper,
5168 struct netdev_adjacent, list);
5169 if (upper && likely(upper->master))
5173 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
5175 static int netdev_adjacent_sysfs_add(struct net_device *dev,
5176 struct net_device *adj_dev,
5177 struct list_head *dev_list)
5179 char linkname[IFNAMSIZ+7];
5180 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5181 "upper_%s" : "lower_%s", adj_dev->name);
5182 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
5185 static void netdev_adjacent_sysfs_del(struct net_device *dev,
5187 struct list_head *dev_list)
5189 char linkname[IFNAMSIZ+7];
5190 sprintf(linkname, dev_list == &dev->adj_list.upper ?
5191 "upper_%s" : "lower_%s", name);
5192 sysfs_remove_link(&(dev->dev.kobj), linkname);
5195 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
5196 struct net_device *adj_dev,
5197 struct list_head *dev_list)
5199 return (dev_list == &dev->adj_list.upper ||
5200 dev_list == &dev->adj_list.lower) &&
5201 net_eq(dev_net(dev), dev_net(adj_dev));
5204 static int __netdev_adjacent_dev_insert(struct net_device *dev,
5205 struct net_device *adj_dev,
5207 struct list_head *dev_list,
5208 void *private, bool master)
5210 struct netdev_adjacent *adj;
5213 adj = __netdev_find_adj(adj_dev, dev_list);
5216 adj->ref_nr += ref_nr;
5220 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
5225 adj->master = master;
5226 adj->ref_nr = ref_nr;
5227 adj->private = private;
5230 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
5231 adj_dev->name, dev->name, adj_dev->name);
5233 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
5234 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
5239 /* Ensure that master link is always the first item in list. */
5241 ret = sysfs_create_link(&(dev->dev.kobj),
5242 &(adj_dev->dev.kobj), "master");
5244 goto remove_symlinks;
5246 list_add_rcu(&adj->list, dev_list);
5248 list_add_tail_rcu(&adj->list, dev_list);
5254 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5255 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5263 static void __netdev_adjacent_dev_remove(struct net_device *dev,
5264 struct net_device *adj_dev,
5266 struct list_head *dev_list)
5268 struct netdev_adjacent *adj;
5270 adj = __netdev_find_adj(adj_dev, dev_list);
5273 pr_err("tried to remove device %s from %s\n",
5274 dev->name, adj_dev->name);
5278 if (adj->ref_nr > ref_nr) {
5279 pr_debug("%s to %s ref_nr-%d = %d\n", dev->name, adj_dev->name,
5280 ref_nr, adj->ref_nr-ref_nr);
5281 adj->ref_nr -= ref_nr;
5286 sysfs_remove_link(&(dev->dev.kobj), "master");
5288 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
5289 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
5291 list_del_rcu(&adj->list);
5292 pr_debug("dev_put for %s, because link removed from %s to %s\n",
5293 adj_dev->name, dev->name, adj_dev->name);
5295 kfree_rcu(adj, rcu);
5298 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
5299 struct net_device *upper_dev,
5301 struct list_head *up_list,
5302 struct list_head *down_list,
5303 void *private, bool master)
5307 ret = __netdev_adjacent_dev_insert(dev, upper_dev, ref_nr, up_list,
5312 ret = __netdev_adjacent_dev_insert(upper_dev, dev, ref_nr, down_list,
5315 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
5322 static int __netdev_adjacent_dev_link(struct net_device *dev,
5323 struct net_device *upper_dev,
5326 return __netdev_adjacent_dev_link_lists(dev, upper_dev, ref_nr,
5327 &dev->all_adj_list.upper,
5328 &upper_dev->all_adj_list.lower,
5332 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5333 struct net_device *upper_dev,
5335 struct list_head *up_list,
5336 struct list_head *down_list)
5338 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
5339 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
5342 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5343 struct net_device *upper_dev,
5346 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, ref_nr,
5347 &dev->all_adj_list.upper,
5348 &upper_dev->all_adj_list.lower);
5351 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5352 struct net_device *upper_dev,
5353 void *private, bool master)
5355 int ret = __netdev_adjacent_dev_link(dev, upper_dev, 1);
5360 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev, 1,
5361 &dev->adj_list.upper,
5362 &upper_dev->adj_list.lower,
5365 __netdev_adjacent_dev_unlink(dev, upper_dev, 1);
5372 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5373 struct net_device *upper_dev)
5375 __netdev_adjacent_dev_unlink(dev, upper_dev, 1);
5376 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
5377 &dev->adj_list.upper,
5378 &upper_dev->adj_list.lower);
5381 static int __netdev_upper_dev_link(struct net_device *dev,
5382 struct net_device *upper_dev, bool master,
5385 struct netdev_notifier_changeupper_info changeupper_info;
5386 struct netdev_adjacent *i, *j, *to_i, *to_j;
5391 if (dev == upper_dev)
5394 /* To prevent loops, check if dev is not upper device to upper_dev. */
5395 if (__netdev_find_adj(dev, &upper_dev->all_adj_list.upper))
5398 if (__netdev_find_adj(upper_dev, &dev->adj_list.upper))
5401 if (master && netdev_master_upper_dev_get(dev))
5404 changeupper_info.upper_dev = upper_dev;
5405 changeupper_info.master = master;
5406 changeupper_info.linking = true;
5408 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5409 &changeupper_info.info);
5410 ret = notifier_to_errno(ret);
5414 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5419 /* Now that we linked these devs, make all the upper_dev's
5420 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5421 * versa, and don't forget the devices itself. All of these
5422 * links are non-neighbours.
5424 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5425 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5426 pr_debug("Interlinking %s with %s, non-neighbour\n",
5427 i->dev->name, j->dev->name);
5428 ret = __netdev_adjacent_dev_link(i->dev, j->dev, i->ref_nr);
5434 /* add dev to every upper_dev's upper device */
5435 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5436 pr_debug("linking %s's upper device %s with %s\n",
5437 upper_dev->name, i->dev->name, dev->name);
5438 ret = __netdev_adjacent_dev_link(dev, i->dev, i->ref_nr);
5440 goto rollback_upper_mesh;
5443 /* add upper_dev to every dev's lower device */
5444 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5445 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5446 i->dev->name, upper_dev->name);
5447 ret = __netdev_adjacent_dev_link(i->dev, upper_dev, i->ref_nr);
5449 goto rollback_lower_mesh;
5452 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5453 &changeupper_info.info);
5456 rollback_lower_mesh:
5458 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5461 __netdev_adjacent_dev_unlink(i->dev, upper_dev, i->ref_nr);
5466 rollback_upper_mesh:
5468 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5471 __netdev_adjacent_dev_unlink(dev, i->dev, i->ref_nr);
5479 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5480 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5481 if (i == to_i && j == to_j)
5483 __netdev_adjacent_dev_unlink(i->dev, j->dev, i->ref_nr);
5489 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5495 * netdev_upper_dev_link - Add a link to the upper device
5497 * @upper_dev: new upper device
5499 * Adds a link to device which is upper to this one. The caller must hold
5500 * the RTNL lock. On a failure a negative errno code is returned.
5501 * On success the reference counts are adjusted and the function
5504 int netdev_upper_dev_link(struct net_device *dev,
5505 struct net_device *upper_dev)
5507 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5509 EXPORT_SYMBOL(netdev_upper_dev_link);
5512 * netdev_master_upper_dev_link - Add a master link to the upper device
5514 * @upper_dev: new upper device
5516 * Adds a link to device which is upper to this one. In this case, only
5517 * one master upper device can be linked, although other non-master devices
5518 * might be linked as well. The caller must hold the RTNL lock.
5519 * On a failure a negative errno code is returned. On success the reference
5520 * counts are adjusted and the function returns zero.
5522 int netdev_master_upper_dev_link(struct net_device *dev,
5523 struct net_device *upper_dev)
5525 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5527 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5529 int netdev_master_upper_dev_link_private(struct net_device *dev,
5530 struct net_device *upper_dev,
5533 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5535 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5538 * netdev_upper_dev_unlink - Removes a link to upper device
5540 * @upper_dev: new upper device
5542 * Removes a link to device which is upper to this one. The caller must hold
5545 void netdev_upper_dev_unlink(struct net_device *dev,
5546 struct net_device *upper_dev)
5548 struct netdev_notifier_changeupper_info changeupper_info;
5549 struct netdev_adjacent *i, *j;
5552 changeupper_info.upper_dev = upper_dev;
5553 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
5554 changeupper_info.linking = false;
5556 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev,
5557 &changeupper_info.info);
5559 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5561 /* Here is the tricky part. We must remove all dev's lower
5562 * devices from all upper_dev's upper devices and vice
5563 * versa, to maintain the graph relationship.
5565 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5566 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5567 __netdev_adjacent_dev_unlink(i->dev, j->dev, i->ref_nr);
5569 /* remove also the devices itself from lower/upper device
5572 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5573 __netdev_adjacent_dev_unlink(i->dev, upper_dev, i->ref_nr);
5575 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5576 __netdev_adjacent_dev_unlink(dev, i->dev, i->ref_nr);
5578 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev,
5579 &changeupper_info.info);
5581 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5584 * netdev_bonding_info_change - Dispatch event about slave change
5586 * @bonding_info: info to dispatch
5588 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
5589 * The caller must hold the RTNL lock.
5591 void netdev_bonding_info_change(struct net_device *dev,
5592 struct netdev_bonding_info *bonding_info)
5594 struct netdev_notifier_bonding_info info;
5596 memcpy(&info.bonding_info, bonding_info,
5597 sizeof(struct netdev_bonding_info));
5598 call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev,
5601 EXPORT_SYMBOL(netdev_bonding_info_change);
5603 static void netdev_adjacent_add_links(struct net_device *dev)
5605 struct netdev_adjacent *iter;
5607 struct net *net = dev_net(dev);
5609 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5610 if (!net_eq(net,dev_net(iter->dev)))
5612 netdev_adjacent_sysfs_add(iter->dev, dev,
5613 &iter->dev->adj_list.lower);
5614 netdev_adjacent_sysfs_add(dev, iter->dev,
5615 &dev->adj_list.upper);
5618 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5619 if (!net_eq(net,dev_net(iter->dev)))
5621 netdev_adjacent_sysfs_add(iter->dev, dev,
5622 &iter->dev->adj_list.upper);
5623 netdev_adjacent_sysfs_add(dev, iter->dev,
5624 &dev->adj_list.lower);
5628 static void netdev_adjacent_del_links(struct net_device *dev)
5630 struct netdev_adjacent *iter;
5632 struct net *net = dev_net(dev);
5634 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5635 if (!net_eq(net,dev_net(iter->dev)))
5637 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5638 &iter->dev->adj_list.lower);
5639 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5640 &dev->adj_list.upper);
5643 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5644 if (!net_eq(net,dev_net(iter->dev)))
5646 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5647 &iter->dev->adj_list.upper);
5648 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5649 &dev->adj_list.lower);
5653 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5655 struct netdev_adjacent *iter;
5657 struct net *net = dev_net(dev);
5659 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5660 if (!net_eq(net,dev_net(iter->dev)))
5662 netdev_adjacent_sysfs_del(iter->dev, oldname,
5663 &iter->dev->adj_list.lower);
5664 netdev_adjacent_sysfs_add(iter->dev, dev,
5665 &iter->dev->adj_list.lower);
5668 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5669 if (!net_eq(net,dev_net(iter->dev)))
5671 netdev_adjacent_sysfs_del(iter->dev, oldname,
5672 &iter->dev->adj_list.upper);
5673 netdev_adjacent_sysfs_add(iter->dev, dev,
5674 &iter->dev->adj_list.upper);
5678 void *netdev_lower_dev_get_private(struct net_device *dev,
5679 struct net_device *lower_dev)
5681 struct netdev_adjacent *lower;
5685 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
5689 return lower->private;
5691 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5694 int dev_get_nest_level(struct net_device *dev,
5695 bool (*type_check)(struct net_device *dev))
5697 struct net_device *lower = NULL;
5698 struct list_head *iter;
5704 netdev_for_each_lower_dev(dev, lower, iter) {
5705 nest = dev_get_nest_level(lower, type_check);
5706 if (max_nest < nest)
5710 if (type_check(dev))
5715 EXPORT_SYMBOL(dev_get_nest_level);
5717 static void dev_change_rx_flags(struct net_device *dev, int flags)
5719 const struct net_device_ops *ops = dev->netdev_ops;
5721 if (ops->ndo_change_rx_flags)
5722 ops->ndo_change_rx_flags(dev, flags);
5725 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5727 unsigned int old_flags = dev->flags;
5733 dev->flags |= IFF_PROMISC;
5734 dev->promiscuity += inc;
5735 if (dev->promiscuity == 0) {
5738 * If inc causes overflow, untouch promisc and return error.
5741 dev->flags &= ~IFF_PROMISC;
5743 dev->promiscuity -= inc;
5744 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5749 if (dev->flags != old_flags) {
5750 pr_info("device %s %s promiscuous mode\n",
5752 dev->flags & IFF_PROMISC ? "entered" : "left");
5753 if (audit_enabled) {
5754 current_uid_gid(&uid, &gid);
5755 audit_log(current->audit_context, GFP_ATOMIC,
5756 AUDIT_ANOM_PROMISCUOUS,
5757 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5758 dev->name, (dev->flags & IFF_PROMISC),
5759 (old_flags & IFF_PROMISC),
5760 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5761 from_kuid(&init_user_ns, uid),
5762 from_kgid(&init_user_ns, gid),
5763 audit_get_sessionid(current));
5766 dev_change_rx_flags(dev, IFF_PROMISC);
5769 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5774 * dev_set_promiscuity - update promiscuity count on a device
5778 * Add or remove promiscuity from a device. While the count in the device
5779 * remains above zero the interface remains promiscuous. Once it hits zero
5780 * the device reverts back to normal filtering operation. A negative inc
5781 * value is used to drop promiscuity on the device.
5782 * Return 0 if successful or a negative errno code on error.
5784 int dev_set_promiscuity(struct net_device *dev, int inc)
5786 unsigned int old_flags = dev->flags;
5789 err = __dev_set_promiscuity(dev, inc, true);
5792 if (dev->flags != old_flags)
5793 dev_set_rx_mode(dev);
5796 EXPORT_SYMBOL(dev_set_promiscuity);
5798 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5800 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5804 dev->flags |= IFF_ALLMULTI;
5805 dev->allmulti += inc;
5806 if (dev->allmulti == 0) {
5809 * If inc causes overflow, untouch allmulti and return error.
5812 dev->flags &= ~IFF_ALLMULTI;
5814 dev->allmulti -= inc;
5815 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5820 if (dev->flags ^ old_flags) {
5821 dev_change_rx_flags(dev, IFF_ALLMULTI);
5822 dev_set_rx_mode(dev);
5824 __dev_notify_flags(dev, old_flags,
5825 dev->gflags ^ old_gflags);
5831 * dev_set_allmulti - update allmulti count on a device
5835 * Add or remove reception of all multicast frames to a device. While the
5836 * count in the device remains above zero the interface remains listening
5837 * to all interfaces. Once it hits zero the device reverts back to normal
5838 * filtering operation. A negative @inc value is used to drop the counter
5839 * when releasing a resource needing all multicasts.
5840 * Return 0 if successful or a negative errno code on error.
5843 int dev_set_allmulti(struct net_device *dev, int inc)
5845 return __dev_set_allmulti(dev, inc, true);
5847 EXPORT_SYMBOL(dev_set_allmulti);
5850 * Upload unicast and multicast address lists to device and
5851 * configure RX filtering. When the device doesn't support unicast
5852 * filtering it is put in promiscuous mode while unicast addresses
5855 void __dev_set_rx_mode(struct net_device *dev)
5857 const struct net_device_ops *ops = dev->netdev_ops;
5859 /* dev_open will call this function so the list will stay sane. */
5860 if (!(dev->flags&IFF_UP))
5863 if (!netif_device_present(dev))
5866 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5867 /* Unicast addresses changes may only happen under the rtnl,
5868 * therefore calling __dev_set_promiscuity here is safe.
5870 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5871 __dev_set_promiscuity(dev, 1, false);
5872 dev->uc_promisc = true;
5873 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5874 __dev_set_promiscuity(dev, -1, false);
5875 dev->uc_promisc = false;
5879 if (ops->ndo_set_rx_mode)
5880 ops->ndo_set_rx_mode(dev);
5883 void dev_set_rx_mode(struct net_device *dev)
5885 netif_addr_lock_bh(dev);
5886 __dev_set_rx_mode(dev);
5887 netif_addr_unlock_bh(dev);
5891 * dev_get_flags - get flags reported to userspace
5894 * Get the combination of flag bits exported through APIs to userspace.
5896 unsigned int dev_get_flags(const struct net_device *dev)
5900 flags = (dev->flags & ~(IFF_PROMISC |
5905 (dev->gflags & (IFF_PROMISC |
5908 if (netif_running(dev)) {
5909 if (netif_oper_up(dev))
5910 flags |= IFF_RUNNING;
5911 if (netif_carrier_ok(dev))
5912 flags |= IFF_LOWER_UP;
5913 if (netif_dormant(dev))
5914 flags |= IFF_DORMANT;
5919 EXPORT_SYMBOL(dev_get_flags);
5921 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5923 unsigned int old_flags = dev->flags;
5929 * Set the flags on our device.
5932 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5933 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5935 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5939 * Load in the correct multicast list now the flags have changed.
5942 if ((old_flags ^ flags) & IFF_MULTICAST)
5943 dev_change_rx_flags(dev, IFF_MULTICAST);
5945 dev_set_rx_mode(dev);
5948 * Have we downed the interface. We handle IFF_UP ourselves
5949 * according to user attempts to set it, rather than blindly
5954 if ((old_flags ^ flags) & IFF_UP)
5955 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5957 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5958 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5959 unsigned int old_flags = dev->flags;
5961 dev->gflags ^= IFF_PROMISC;
5963 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5964 if (dev->flags != old_flags)
5965 dev_set_rx_mode(dev);
5968 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5969 is important. Some (broken) drivers set IFF_PROMISC, when
5970 IFF_ALLMULTI is requested not asking us and not reporting.
5972 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5973 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5975 dev->gflags ^= IFF_ALLMULTI;
5976 __dev_set_allmulti(dev, inc, false);
5982 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5983 unsigned int gchanges)
5985 unsigned int changes = dev->flags ^ old_flags;
5988 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5990 if (changes & IFF_UP) {
5991 if (dev->flags & IFF_UP)
5992 call_netdevice_notifiers(NETDEV_UP, dev);
5994 call_netdevice_notifiers(NETDEV_DOWN, dev);
5997 if (dev->flags & IFF_UP &&
5998 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5999 struct netdev_notifier_change_info change_info;
6001 change_info.flags_changed = changes;
6002 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
6008 * dev_change_flags - change device settings
6010 * @flags: device state flags
6012 * Change settings on device based state flags. The flags are
6013 * in the userspace exported format.
6015 int dev_change_flags(struct net_device *dev, unsigned int flags)
6018 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
6020 ret = __dev_change_flags(dev, flags);
6024 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
6025 __dev_notify_flags(dev, old_flags, changes);
6028 EXPORT_SYMBOL(dev_change_flags);
6030 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
6032 const struct net_device_ops *ops = dev->netdev_ops;
6034 if (ops->ndo_change_mtu)
6035 return ops->ndo_change_mtu(dev, new_mtu);
6042 * dev_set_mtu - Change maximum transfer unit
6044 * @new_mtu: new transfer unit
6046 * Change the maximum transfer size of the network device.
6048 int dev_set_mtu(struct net_device *dev, int new_mtu)
6052 if (new_mtu == dev->mtu)
6055 /* MTU must be positive. */
6059 if (!netif_device_present(dev))
6062 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
6063 err = notifier_to_errno(err);
6067 orig_mtu = dev->mtu;
6068 err = __dev_set_mtu(dev, new_mtu);
6071 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6072 err = notifier_to_errno(err);
6074 /* setting mtu back and notifying everyone again,
6075 * so that they have a chance to revert changes.
6077 __dev_set_mtu(dev, orig_mtu);
6078 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
6083 EXPORT_SYMBOL(dev_set_mtu);
6086 * dev_set_group - Change group this device belongs to
6088 * @new_group: group this device should belong to
6090 void dev_set_group(struct net_device *dev, int new_group)
6092 dev->group = new_group;
6094 EXPORT_SYMBOL(dev_set_group);
6097 * dev_set_mac_address - Change Media Access Control Address
6101 * Change the hardware (MAC) address of the device
6103 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
6105 const struct net_device_ops *ops = dev->netdev_ops;
6108 if (!ops->ndo_set_mac_address)
6110 if (sa->sa_family != dev->type)
6112 if (!netif_device_present(dev))
6114 err = ops->ndo_set_mac_address(dev, sa);
6117 dev->addr_assign_type = NET_ADDR_SET;
6118 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
6119 add_device_randomness(dev->dev_addr, dev->addr_len);
6122 EXPORT_SYMBOL(dev_set_mac_address);
6125 * dev_change_carrier - Change device carrier
6127 * @new_carrier: new value
6129 * Change device carrier
6131 int dev_change_carrier(struct net_device *dev, bool new_carrier)
6133 const struct net_device_ops *ops = dev->netdev_ops;
6135 if (!ops->ndo_change_carrier)
6137 if (!netif_device_present(dev))
6139 return ops->ndo_change_carrier(dev, new_carrier);
6141 EXPORT_SYMBOL(dev_change_carrier);
6144 * dev_get_phys_port_id - Get device physical port ID
6148 * Get device physical port ID
6150 int dev_get_phys_port_id(struct net_device *dev,
6151 struct netdev_phys_item_id *ppid)
6153 const struct net_device_ops *ops = dev->netdev_ops;
6155 if (!ops->ndo_get_phys_port_id)
6157 return ops->ndo_get_phys_port_id(dev, ppid);
6159 EXPORT_SYMBOL(dev_get_phys_port_id);
6162 * dev_get_phys_port_name - Get device physical port name
6166 * Get device physical port name
6168 int dev_get_phys_port_name(struct net_device *dev,
6169 char *name, size_t len)
6171 const struct net_device_ops *ops = dev->netdev_ops;
6173 if (!ops->ndo_get_phys_port_name)
6175 return ops->ndo_get_phys_port_name(dev, name, len);
6177 EXPORT_SYMBOL(dev_get_phys_port_name);
6180 * dev_change_proto_down - update protocol port state information
6182 * @proto_down: new value
6184 * This info can be used by switch drivers to set the phys state of the
6187 int dev_change_proto_down(struct net_device *dev, bool proto_down)
6189 const struct net_device_ops *ops = dev->netdev_ops;
6191 if (!ops->ndo_change_proto_down)
6193 if (!netif_device_present(dev))
6195 return ops->ndo_change_proto_down(dev, proto_down);
6197 EXPORT_SYMBOL(dev_change_proto_down);
6200 * dev_new_index - allocate an ifindex
6201 * @net: the applicable net namespace
6203 * Returns a suitable unique value for a new device interface
6204 * number. The caller must hold the rtnl semaphore or the
6205 * dev_base_lock to be sure it remains unique.
6207 static int dev_new_index(struct net *net)
6209 int ifindex = net->ifindex;
6213 if (!__dev_get_by_index(net, ifindex))
6214 return net->ifindex = ifindex;
6218 /* Delayed registration/unregisteration */
6219 static LIST_HEAD(net_todo_list);
6220 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
6222 static void net_set_todo(struct net_device *dev)
6224 list_add_tail(&dev->todo_list, &net_todo_list);
6225 dev_net(dev)->dev_unreg_count++;
6228 static void rollback_registered_many(struct list_head *head)
6230 struct net_device *dev, *tmp;
6231 LIST_HEAD(close_head);
6233 BUG_ON(dev_boot_phase);
6236 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
6237 /* Some devices call without registering
6238 * for initialization unwind. Remove those
6239 * devices and proceed with the remaining.
6241 if (dev->reg_state == NETREG_UNINITIALIZED) {
6242 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
6246 list_del(&dev->unreg_list);
6249 dev->dismantle = true;
6250 BUG_ON(dev->reg_state != NETREG_REGISTERED);
6253 /* If device is running, close it first. */
6254 list_for_each_entry(dev, head, unreg_list)
6255 list_add_tail(&dev->close_list, &close_head);
6256 dev_close_many(&close_head, true);
6258 list_for_each_entry(dev, head, unreg_list) {
6259 /* And unlink it from device chain. */
6260 unlist_netdevice(dev);
6262 dev->reg_state = NETREG_UNREGISTERING;
6263 on_each_cpu(flush_backlog, dev, 1);
6268 list_for_each_entry(dev, head, unreg_list) {
6269 struct sk_buff *skb = NULL;
6271 /* Shutdown queueing discipline. */
6275 /* Notify protocols, that we are about to destroy
6276 this device. They should clean all the things.
6278 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6280 if (!dev->rtnl_link_ops ||
6281 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6282 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U,
6286 * Flush the unicast and multicast chains
6291 if (dev->netdev_ops->ndo_uninit)
6292 dev->netdev_ops->ndo_uninit(dev);
6295 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
6297 /* Notifier chain MUST detach us all upper devices. */
6298 WARN_ON(netdev_has_any_upper_dev(dev));
6300 /* Remove entries from kobject tree */
6301 netdev_unregister_kobject(dev);
6303 /* Remove XPS queueing entries */
6304 netif_reset_xps_queues_gt(dev, 0);
6310 list_for_each_entry(dev, head, unreg_list)
6314 static void rollback_registered(struct net_device *dev)
6318 list_add(&dev->unreg_list, &single);
6319 rollback_registered_many(&single);
6323 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
6324 struct net_device *upper, netdev_features_t features)
6326 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6327 netdev_features_t feature;
6330 for_each_netdev_feature(&upper_disables, feature_bit) {
6331 feature = __NETIF_F_BIT(feature_bit);
6332 if (!(upper->wanted_features & feature)
6333 && (features & feature)) {
6334 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
6335 &feature, upper->name);
6336 features &= ~feature;
6343 static void netdev_sync_lower_features(struct net_device *upper,
6344 struct net_device *lower, netdev_features_t features)
6346 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
6347 netdev_features_t feature;
6350 for_each_netdev_feature(&upper_disables, feature_bit) {
6351 feature = __NETIF_F_BIT(feature_bit);
6352 if (!(features & feature) && (lower->features & feature)) {
6353 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
6354 &feature, lower->name);
6355 lower->wanted_features &= ~feature;
6356 netdev_update_features(lower);
6358 if (unlikely(lower->features & feature))
6359 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
6360 &feature, lower->name);
6365 static netdev_features_t netdev_fix_features(struct net_device *dev,
6366 netdev_features_t features)
6368 /* Fix illegal checksum combinations */
6369 if ((features & NETIF_F_HW_CSUM) &&
6370 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6371 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
6372 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
6375 /* TSO requires that SG is present as well. */
6376 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
6377 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
6378 features &= ~NETIF_F_ALL_TSO;
6381 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
6382 !(features & NETIF_F_IP_CSUM)) {
6383 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
6384 features &= ~NETIF_F_TSO;
6385 features &= ~NETIF_F_TSO_ECN;
6388 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
6389 !(features & NETIF_F_IPV6_CSUM)) {
6390 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
6391 features &= ~NETIF_F_TSO6;
6394 /* TSO ECN requires that TSO is present as well. */
6395 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
6396 features &= ~NETIF_F_TSO_ECN;
6398 /* Software GSO depends on SG. */
6399 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
6400 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
6401 features &= ~NETIF_F_GSO;
6404 /* UFO needs SG and checksumming */
6405 if (features & NETIF_F_UFO) {
6406 /* maybe split UFO into V4 and V6? */
6407 if (!((features & NETIF_F_GEN_CSUM) ||
6408 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
6409 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
6411 "Dropping NETIF_F_UFO since no checksum offload features.\n");
6412 features &= ~NETIF_F_UFO;
6415 if (!(features & NETIF_F_SG)) {
6417 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
6418 features &= ~NETIF_F_UFO;
6422 #ifdef CONFIG_NET_RX_BUSY_POLL
6423 if (dev->netdev_ops->ndo_busy_poll)
6424 features |= NETIF_F_BUSY_POLL;
6427 features &= ~NETIF_F_BUSY_POLL;
6432 int __netdev_update_features(struct net_device *dev)
6434 struct net_device *upper, *lower;
6435 netdev_features_t features;
6436 struct list_head *iter;
6441 features = netdev_get_wanted_features(dev);
6443 if (dev->netdev_ops->ndo_fix_features)
6444 features = dev->netdev_ops->ndo_fix_features(dev, features);
6446 /* driver might be less strict about feature dependencies */
6447 features = netdev_fix_features(dev, features);
6449 /* some features can't be enabled if they're off an an upper device */
6450 netdev_for_each_upper_dev_rcu(dev, upper, iter)
6451 features = netdev_sync_upper_features(dev, upper, features);
6453 if (dev->features == features)
6456 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6457 &dev->features, &features);
6459 if (dev->netdev_ops->ndo_set_features)
6460 err = dev->netdev_ops->ndo_set_features(dev, features);
6464 if (unlikely(err < 0)) {
6466 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6467 err, &features, &dev->features);
6468 /* return non-0 since some features might have changed and
6469 * it's better to fire a spurious notification than miss it
6475 /* some features must be disabled on lower devices when disabled
6476 * on an upper device (think: bonding master or bridge)
6478 netdev_for_each_lower_dev(dev, lower, iter)
6479 netdev_sync_lower_features(dev, lower, features);
6482 dev->features = features;
6484 return err < 0 ? 0 : 1;
6488 * netdev_update_features - recalculate device features
6489 * @dev: the device to check
6491 * Recalculate dev->features set and send notifications if it
6492 * has changed. Should be called after driver or hardware dependent
6493 * conditions might have changed that influence the features.
6495 void netdev_update_features(struct net_device *dev)
6497 if (__netdev_update_features(dev))
6498 netdev_features_change(dev);
6500 EXPORT_SYMBOL(netdev_update_features);
6503 * netdev_change_features - recalculate device features
6504 * @dev: the device to check
6506 * Recalculate dev->features set and send notifications even
6507 * if they have not changed. Should be called instead of
6508 * netdev_update_features() if also dev->vlan_features might
6509 * have changed to allow the changes to be propagated to stacked
6512 void netdev_change_features(struct net_device *dev)
6514 __netdev_update_features(dev);
6515 netdev_features_change(dev);
6517 EXPORT_SYMBOL(netdev_change_features);
6520 * netif_stacked_transfer_operstate - transfer operstate
6521 * @rootdev: the root or lower level device to transfer state from
6522 * @dev: the device to transfer operstate to
6524 * Transfer operational state from root to device. This is normally
6525 * called when a stacking relationship exists between the root
6526 * device and the device(a leaf device).
6528 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6529 struct net_device *dev)
6531 if (rootdev->operstate == IF_OPER_DORMANT)
6532 netif_dormant_on(dev);
6534 netif_dormant_off(dev);
6536 if (netif_carrier_ok(rootdev)) {
6537 if (!netif_carrier_ok(dev))
6538 netif_carrier_on(dev);
6540 if (netif_carrier_ok(dev))
6541 netif_carrier_off(dev);
6544 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6547 static int netif_alloc_rx_queues(struct net_device *dev)
6549 unsigned int i, count = dev->num_rx_queues;
6550 struct netdev_rx_queue *rx;
6551 size_t sz = count * sizeof(*rx);
6555 rx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6563 for (i = 0; i < count; i++)
6569 static void netdev_init_one_queue(struct net_device *dev,
6570 struct netdev_queue *queue, void *_unused)
6572 /* Initialize queue lock */
6573 spin_lock_init(&queue->_xmit_lock);
6574 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6575 queue->xmit_lock_owner = -1;
6576 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6579 dql_init(&queue->dql, HZ);
6583 static void netif_free_tx_queues(struct net_device *dev)
6588 static int netif_alloc_netdev_queues(struct net_device *dev)
6590 unsigned int count = dev->num_tx_queues;
6591 struct netdev_queue *tx;
6592 size_t sz = count * sizeof(*tx);
6594 if (count < 1 || count > 0xffff)
6597 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6605 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6606 spin_lock_init(&dev->tx_global_lock);
6611 void netif_tx_stop_all_queues(struct net_device *dev)
6615 for (i = 0; i < dev->num_tx_queues; i++) {
6616 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
6617 netif_tx_stop_queue(txq);
6620 EXPORT_SYMBOL(netif_tx_stop_all_queues);
6623 * register_netdevice - register a network device
6624 * @dev: device to register
6626 * Take a completed network device structure and add it to the kernel
6627 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6628 * chain. 0 is returned on success. A negative errno code is returned
6629 * on a failure to set up the device, or if the name is a duplicate.
6631 * Callers must hold the rtnl semaphore. You may want
6632 * register_netdev() instead of this.
6635 * The locking appears insufficient to guarantee two parallel registers
6636 * will not get the same name.
6639 int register_netdevice(struct net_device *dev)
6642 struct net *net = dev_net(dev);
6644 BUG_ON(dev_boot_phase);
6649 /* When net_device's are persistent, this will be fatal. */
6650 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6653 spin_lock_init(&dev->addr_list_lock);
6654 netdev_set_addr_lockdep_class(dev);
6656 ret = dev_get_valid_name(net, dev, dev->name);
6660 /* Init, if this function is available */
6661 if (dev->netdev_ops->ndo_init) {
6662 ret = dev->netdev_ops->ndo_init(dev);
6670 if (((dev->hw_features | dev->features) &
6671 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6672 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6673 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6674 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6681 dev->ifindex = dev_new_index(net);
6682 else if (__dev_get_by_index(net, dev->ifindex))
6685 /* Transfer changeable features to wanted_features and enable
6686 * software offloads (GSO and GRO).
6688 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6689 dev->features |= NETIF_F_SOFT_FEATURES;
6690 dev->wanted_features = dev->features & dev->hw_features;
6692 if (!(dev->flags & IFF_LOOPBACK)) {
6693 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6696 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6698 dev->vlan_features |= NETIF_F_HIGHDMA;
6700 /* Make NETIF_F_SG inheritable to tunnel devices.
6702 dev->hw_enc_features |= NETIF_F_SG;
6704 /* Make NETIF_F_SG inheritable to MPLS.
6706 dev->mpls_features |= NETIF_F_SG;
6708 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6709 ret = notifier_to_errno(ret);
6713 ret = netdev_register_kobject(dev);
6716 dev->reg_state = NETREG_REGISTERED;
6718 __netdev_update_features(dev);
6721 * Default initial state at registry is that the
6722 * device is present.
6725 set_bit(__LINK_STATE_PRESENT, &dev->state);
6727 linkwatch_init_dev(dev);
6729 dev_init_scheduler(dev);
6731 list_netdevice(dev);
6732 add_device_randomness(dev->dev_addr, dev->addr_len);
6734 /* If the device has permanent device address, driver should
6735 * set dev_addr and also addr_assign_type should be set to
6736 * NET_ADDR_PERM (default value).
6738 if (dev->addr_assign_type == NET_ADDR_PERM)
6739 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6741 /* Notify protocols, that a new device appeared. */
6742 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6743 ret = notifier_to_errno(ret);
6745 rollback_registered(dev);
6746 dev->reg_state = NETREG_UNREGISTERED;
6749 * Prevent userspace races by waiting until the network
6750 * device is fully setup before sending notifications.
6752 if (!dev->rtnl_link_ops ||
6753 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6754 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6760 if (dev->netdev_ops->ndo_uninit)
6761 dev->netdev_ops->ndo_uninit(dev);
6764 EXPORT_SYMBOL(register_netdevice);
6767 * init_dummy_netdev - init a dummy network device for NAPI
6768 * @dev: device to init
6770 * This takes a network device structure and initialize the minimum
6771 * amount of fields so it can be used to schedule NAPI polls without
6772 * registering a full blown interface. This is to be used by drivers
6773 * that need to tie several hardware interfaces to a single NAPI
6774 * poll scheduler due to HW limitations.
6776 int init_dummy_netdev(struct net_device *dev)
6778 /* Clear everything. Note we don't initialize spinlocks
6779 * are they aren't supposed to be taken by any of the
6780 * NAPI code and this dummy netdev is supposed to be
6781 * only ever used for NAPI polls
6783 memset(dev, 0, sizeof(struct net_device));
6785 /* make sure we BUG if trying to hit standard
6786 * register/unregister code path
6788 dev->reg_state = NETREG_DUMMY;
6790 /* NAPI wants this */
6791 INIT_LIST_HEAD(&dev->napi_list);
6793 /* a dummy interface is started by default */
6794 set_bit(__LINK_STATE_PRESENT, &dev->state);
6795 set_bit(__LINK_STATE_START, &dev->state);
6797 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6798 * because users of this 'device' dont need to change
6804 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6808 * register_netdev - register a network device
6809 * @dev: device to register
6811 * Take a completed network device structure and add it to the kernel
6812 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6813 * chain. 0 is returned on success. A negative errno code is returned
6814 * on a failure to set up the device, or if the name is a duplicate.
6816 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6817 * and expands the device name if you passed a format string to
6820 int register_netdev(struct net_device *dev)
6825 err = register_netdevice(dev);
6829 EXPORT_SYMBOL(register_netdev);
6831 int netdev_refcnt_read(const struct net_device *dev)
6835 for_each_possible_cpu(i)
6836 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6839 EXPORT_SYMBOL(netdev_refcnt_read);
6842 * netdev_wait_allrefs - wait until all references are gone.
6843 * @dev: target net_device
6845 * This is called when unregistering network devices.
6847 * Any protocol or device that holds a reference should register
6848 * for netdevice notification, and cleanup and put back the
6849 * reference if they receive an UNREGISTER event.
6850 * We can get stuck here if buggy protocols don't correctly
6853 static void netdev_wait_allrefs(struct net_device *dev)
6855 unsigned long rebroadcast_time, warning_time;
6858 linkwatch_forget_dev(dev);
6860 rebroadcast_time = warning_time = jiffies;
6861 refcnt = netdev_refcnt_read(dev);
6863 while (refcnt != 0) {
6864 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6867 /* Rebroadcast unregister notification */
6868 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6874 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6875 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6877 /* We must not have linkwatch events
6878 * pending on unregister. If this
6879 * happens, we simply run the queue
6880 * unscheduled, resulting in a noop
6883 linkwatch_run_queue();
6888 rebroadcast_time = jiffies;
6893 refcnt = netdev_refcnt_read(dev);
6895 if (time_after(jiffies, warning_time + 10 * HZ)) {
6896 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6898 warning_time = jiffies;
6907 * register_netdevice(x1);
6908 * register_netdevice(x2);
6910 * unregister_netdevice(y1);
6911 * unregister_netdevice(y2);
6917 * We are invoked by rtnl_unlock().
6918 * This allows us to deal with problems:
6919 * 1) We can delete sysfs objects which invoke hotplug
6920 * without deadlocking with linkwatch via keventd.
6921 * 2) Since we run with the RTNL semaphore not held, we can sleep
6922 * safely in order to wait for the netdev refcnt to drop to zero.
6924 * We must not return until all unregister events added during
6925 * the interval the lock was held have been completed.
6927 void netdev_run_todo(void)
6929 struct list_head list;
6931 /* Snapshot list, allow later requests */
6932 list_replace_init(&net_todo_list, &list);
6937 /* Wait for rcu callbacks to finish before next phase */
6938 if (!list_empty(&list))
6941 while (!list_empty(&list)) {
6942 struct net_device *dev
6943 = list_first_entry(&list, struct net_device, todo_list);
6944 list_del(&dev->todo_list);
6947 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6950 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6951 pr_err("network todo '%s' but state %d\n",
6952 dev->name, dev->reg_state);
6957 dev->reg_state = NETREG_UNREGISTERED;
6959 netdev_wait_allrefs(dev);
6962 BUG_ON(netdev_refcnt_read(dev));
6963 BUG_ON(!list_empty(&dev->ptype_all));
6964 BUG_ON(!list_empty(&dev->ptype_specific));
6965 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6966 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6967 WARN_ON(dev->dn_ptr);
6969 if (dev->destructor)
6970 dev->destructor(dev);
6972 /* Report a network device has been unregistered */
6974 dev_net(dev)->dev_unreg_count--;
6976 wake_up(&netdev_unregistering_wq);
6978 /* Free network device */
6979 kobject_put(&dev->dev.kobj);
6983 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6984 * fields in the same order, with only the type differing.
6986 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6987 const struct net_device_stats *netdev_stats)
6989 #if BITS_PER_LONG == 64
6990 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6991 memcpy(stats64, netdev_stats, sizeof(*stats64));
6993 size_t i, n = sizeof(*stats64) / sizeof(u64);
6994 const unsigned long *src = (const unsigned long *)netdev_stats;
6995 u64 *dst = (u64 *)stats64;
6997 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6998 sizeof(*stats64) / sizeof(u64));
6999 for (i = 0; i < n; i++)
7003 EXPORT_SYMBOL(netdev_stats_to_stats64);
7006 * dev_get_stats - get network device statistics
7007 * @dev: device to get statistics from
7008 * @storage: place to store stats
7010 * Get network statistics from device. Return @storage.
7011 * The device driver may provide its own method by setting
7012 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
7013 * otherwise the internal statistics structure is used.
7015 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
7016 struct rtnl_link_stats64 *storage)
7018 const struct net_device_ops *ops = dev->netdev_ops;
7020 if (ops->ndo_get_stats64) {
7021 memset(storage, 0, sizeof(*storage));
7022 ops->ndo_get_stats64(dev, storage);
7023 } else if (ops->ndo_get_stats) {
7024 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
7026 netdev_stats_to_stats64(storage, &dev->stats);
7028 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
7029 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
7032 EXPORT_SYMBOL(dev_get_stats);
7034 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
7036 struct netdev_queue *queue = dev_ingress_queue(dev);
7038 #ifdef CONFIG_NET_CLS_ACT
7041 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
7044 netdev_init_one_queue(dev, queue, NULL);
7045 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
7046 queue->qdisc_sleeping = &noop_qdisc;
7047 rcu_assign_pointer(dev->ingress_queue, queue);
7052 static const struct ethtool_ops default_ethtool_ops;
7054 void netdev_set_default_ethtool_ops(struct net_device *dev,
7055 const struct ethtool_ops *ops)
7057 if (dev->ethtool_ops == &default_ethtool_ops)
7058 dev->ethtool_ops = ops;
7060 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
7062 void netdev_freemem(struct net_device *dev)
7064 char *addr = (char *)dev - dev->padded;
7070 * alloc_netdev_mqs - allocate network device
7071 * @sizeof_priv: size of private data to allocate space for
7072 * @name: device name format string
7073 * @name_assign_type: origin of device name
7074 * @setup: callback to initialize device
7075 * @txqs: the number of TX subqueues to allocate
7076 * @rxqs: the number of RX subqueues to allocate
7078 * Allocates a struct net_device with private data area for driver use
7079 * and performs basic initialization. Also allocates subqueue structs
7080 * for each queue on the device.
7082 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
7083 unsigned char name_assign_type,
7084 void (*setup)(struct net_device *),
7085 unsigned int txqs, unsigned int rxqs)
7087 struct net_device *dev;
7089 struct net_device *p;
7091 BUG_ON(strlen(name) >= sizeof(dev->name));
7094 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
7100 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
7105 alloc_size = sizeof(struct net_device);
7107 /* ensure 32-byte alignment of private area */
7108 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
7109 alloc_size += sizeof_priv;
7111 /* ensure 32-byte alignment of whole construct */
7112 alloc_size += NETDEV_ALIGN - 1;
7114 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
7116 p = vzalloc(alloc_size);
7120 dev = PTR_ALIGN(p, NETDEV_ALIGN);
7121 dev->padded = (char *)dev - (char *)p;
7123 dev->pcpu_refcnt = alloc_percpu(int);
7124 if (!dev->pcpu_refcnt)
7127 if (dev_addr_init(dev))
7133 dev_net_set(dev, &init_net);
7135 dev->gso_max_size = GSO_MAX_SIZE;
7136 dev->gso_max_segs = GSO_MAX_SEGS;
7137 dev->gso_min_segs = 0;
7139 INIT_LIST_HEAD(&dev->napi_list);
7140 INIT_LIST_HEAD(&dev->unreg_list);
7141 INIT_LIST_HEAD(&dev->close_list);
7142 INIT_LIST_HEAD(&dev->link_watch_list);
7143 INIT_LIST_HEAD(&dev->adj_list.upper);
7144 INIT_LIST_HEAD(&dev->adj_list.lower);
7145 INIT_LIST_HEAD(&dev->all_adj_list.upper);
7146 INIT_LIST_HEAD(&dev->all_adj_list.lower);
7147 INIT_LIST_HEAD(&dev->ptype_all);
7148 INIT_LIST_HEAD(&dev->ptype_specific);
7149 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
7152 if (!dev->tx_queue_len) {
7153 dev->priv_flags |= IFF_NO_QUEUE;
7154 dev->tx_queue_len = 1;
7157 dev->num_tx_queues = txqs;
7158 dev->real_num_tx_queues = txqs;
7159 if (netif_alloc_netdev_queues(dev))
7163 dev->num_rx_queues = rxqs;
7164 dev->real_num_rx_queues = rxqs;
7165 if (netif_alloc_rx_queues(dev))
7169 strcpy(dev->name, name);
7170 dev->name_assign_type = name_assign_type;
7171 dev->group = INIT_NETDEV_GROUP;
7172 if (!dev->ethtool_ops)
7173 dev->ethtool_ops = &default_ethtool_ops;
7175 nf_hook_ingress_init(dev);
7184 free_percpu(dev->pcpu_refcnt);
7186 netdev_freemem(dev);
7189 EXPORT_SYMBOL(alloc_netdev_mqs);
7192 * free_netdev - free network device
7195 * This function does the last stage of destroying an allocated device
7196 * interface. The reference to the device object is released.
7197 * If this is the last reference then it will be freed.
7199 void free_netdev(struct net_device *dev)
7201 struct napi_struct *p, *n;
7203 netif_free_tx_queues(dev);
7208 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
7210 /* Flush device addresses */
7211 dev_addr_flush(dev);
7213 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
7216 free_percpu(dev->pcpu_refcnt);
7217 dev->pcpu_refcnt = NULL;
7219 /* Compatibility with error handling in drivers */
7220 if (dev->reg_state == NETREG_UNINITIALIZED) {
7221 netdev_freemem(dev);
7225 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
7226 dev->reg_state = NETREG_RELEASED;
7228 /* will free via device release */
7229 put_device(&dev->dev);
7231 EXPORT_SYMBOL(free_netdev);
7234 * synchronize_net - Synchronize with packet receive processing
7236 * Wait for packets currently being received to be done.
7237 * Does not block later packets from starting.
7239 void synchronize_net(void)
7242 if (rtnl_is_locked())
7243 synchronize_rcu_expedited();
7247 EXPORT_SYMBOL(synchronize_net);
7250 * unregister_netdevice_queue - remove device from the kernel
7254 * This function shuts down a device interface and removes it
7255 * from the kernel tables.
7256 * If head not NULL, device is queued to be unregistered later.
7258 * Callers must hold the rtnl semaphore. You may want
7259 * unregister_netdev() instead of this.
7262 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
7267 list_move_tail(&dev->unreg_list, head);
7269 rollback_registered(dev);
7270 /* Finish processing unregister after unlock */
7274 EXPORT_SYMBOL(unregister_netdevice_queue);
7277 * unregister_netdevice_many - unregister many devices
7278 * @head: list of devices
7280 * Note: As most callers use a stack allocated list_head,
7281 * we force a list_del() to make sure stack wont be corrupted later.
7283 void unregister_netdevice_many(struct list_head *head)
7285 struct net_device *dev;
7287 if (!list_empty(head)) {
7288 rollback_registered_many(head);
7289 list_for_each_entry(dev, head, unreg_list)
7294 EXPORT_SYMBOL(unregister_netdevice_many);
7297 * unregister_netdev - remove device from the kernel
7300 * This function shuts down a device interface and removes it
7301 * from the kernel tables.
7303 * This is just a wrapper for unregister_netdevice that takes
7304 * the rtnl semaphore. In general you want to use this and not
7305 * unregister_netdevice.
7307 void unregister_netdev(struct net_device *dev)
7310 unregister_netdevice(dev);
7313 EXPORT_SYMBOL(unregister_netdev);
7316 * dev_change_net_namespace - move device to different nethost namespace
7318 * @net: network namespace
7319 * @pat: If not NULL name pattern to try if the current device name
7320 * is already taken in the destination network namespace.
7322 * This function shuts down a device interface and moves it
7323 * to a new network namespace. On success 0 is returned, on
7324 * a failure a netagive errno code is returned.
7326 * Callers must hold the rtnl semaphore.
7329 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
7335 /* Don't allow namespace local devices to be moved. */
7337 if (dev->features & NETIF_F_NETNS_LOCAL)
7340 /* Ensure the device has been registrered */
7341 if (dev->reg_state != NETREG_REGISTERED)
7344 /* Get out if there is nothing todo */
7346 if (net_eq(dev_net(dev), net))
7349 /* Pick the destination device name, and ensure
7350 * we can use it in the destination network namespace.
7353 if (__dev_get_by_name(net, dev->name)) {
7354 /* We get here if we can't use the current device name */
7357 if (dev_get_valid_name(net, dev, pat) < 0)
7362 * And now a mini version of register_netdevice unregister_netdevice.
7365 /* If device is running close it first. */
7368 /* And unlink it from device chain */
7370 unlist_netdevice(dev);
7374 /* Shutdown queueing discipline. */
7377 /* Notify protocols, that we are about to destroy
7378 this device. They should clean all the things.
7380 Note that dev->reg_state stays at NETREG_REGISTERED.
7381 This is wanted because this way 8021q and macvlan know
7382 the device is just moving and can keep their slaves up.
7384 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7386 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
7387 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
7390 * Flush the unicast and multicast chains
7395 /* Send a netdev-removed uevent to the old namespace */
7396 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
7397 netdev_adjacent_del_links(dev);
7399 /* Actually switch the network namespace */
7400 dev_net_set(dev, net);
7402 /* If there is an ifindex conflict assign a new one */
7403 if (__dev_get_by_index(net, dev->ifindex))
7404 dev->ifindex = dev_new_index(net);
7406 /* Send a netdev-add uevent to the new namespace */
7407 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
7408 netdev_adjacent_add_links(dev);
7410 /* Fixup kobjects */
7411 err = device_rename(&dev->dev, dev->name);
7414 /* Add the device back in the hashes */
7415 list_netdevice(dev);
7417 /* Notify protocols, that a new device appeared. */
7418 call_netdevice_notifiers(NETDEV_REGISTER, dev);
7421 * Prevent userspace races by waiting until the network
7422 * device is fully setup before sending notifications.
7424 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
7431 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
7433 static int dev_cpu_callback(struct notifier_block *nfb,
7434 unsigned long action,
7437 struct sk_buff **list_skb;
7438 struct sk_buff *skb;
7439 unsigned int cpu, oldcpu = (unsigned long)ocpu;
7440 struct softnet_data *sd, *oldsd;
7442 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
7445 local_irq_disable();
7446 cpu = smp_processor_id();
7447 sd = &per_cpu(softnet_data, cpu);
7448 oldsd = &per_cpu(softnet_data, oldcpu);
7450 /* Find end of our completion_queue. */
7451 list_skb = &sd->completion_queue;
7453 list_skb = &(*list_skb)->next;
7454 /* Append completion queue from offline CPU. */
7455 *list_skb = oldsd->completion_queue;
7456 oldsd->completion_queue = NULL;
7458 /* Append output queue from offline CPU. */
7459 if (oldsd->output_queue) {
7460 *sd->output_queue_tailp = oldsd->output_queue;
7461 sd->output_queue_tailp = oldsd->output_queue_tailp;
7462 oldsd->output_queue = NULL;
7463 oldsd->output_queue_tailp = &oldsd->output_queue;
7465 /* Append NAPI poll list from offline CPU, with one exception :
7466 * process_backlog() must be called by cpu owning percpu backlog.
7467 * We properly handle process_queue & input_pkt_queue later.
7469 while (!list_empty(&oldsd->poll_list)) {
7470 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
7474 list_del_init(&napi->poll_list);
7475 if (napi->poll == process_backlog)
7478 ____napi_schedule(sd, napi);
7481 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7484 /* Process offline CPU's input_pkt_queue */
7485 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7487 input_queue_head_incr(oldsd);
7489 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
7491 input_queue_head_incr(oldsd);
7499 * netdev_increment_features - increment feature set by one
7500 * @all: current feature set
7501 * @one: new feature set
7502 * @mask: mask feature set
7504 * Computes a new feature set after adding a device with feature set
7505 * @one to the master device with current feature set @all. Will not
7506 * enable anything that is off in @mask. Returns the new feature set.
7508 netdev_features_t netdev_increment_features(netdev_features_t all,
7509 netdev_features_t one, netdev_features_t mask)
7511 if (mask & NETIF_F_GEN_CSUM)
7512 mask |= NETIF_F_ALL_CSUM;
7513 mask |= NETIF_F_VLAN_CHALLENGED;
7515 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7516 all &= one | ~NETIF_F_ALL_FOR_ALL;
7518 /* If one device supports hw checksumming, set for all. */
7519 if (all & NETIF_F_GEN_CSUM)
7520 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7524 EXPORT_SYMBOL(netdev_increment_features);
7526 static struct hlist_head * __net_init netdev_create_hash(void)
7529 struct hlist_head *hash;
7531 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7533 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7534 INIT_HLIST_HEAD(&hash[i]);
7539 /* Initialize per network namespace state */
7540 static int __net_init netdev_init(struct net *net)
7542 if (net != &init_net)
7543 INIT_LIST_HEAD(&net->dev_base_head);
7545 net->dev_name_head = netdev_create_hash();
7546 if (net->dev_name_head == NULL)
7549 net->dev_index_head = netdev_create_hash();
7550 if (net->dev_index_head == NULL)
7556 kfree(net->dev_name_head);
7562 * netdev_drivername - network driver for the device
7563 * @dev: network device
7565 * Determine network driver for device.
7567 const char *netdev_drivername(const struct net_device *dev)
7569 const struct device_driver *driver;
7570 const struct device *parent;
7571 const char *empty = "";
7573 parent = dev->dev.parent;
7577 driver = parent->driver;
7578 if (driver && driver->name)
7579 return driver->name;
7583 static void __netdev_printk(const char *level, const struct net_device *dev,
7584 struct va_format *vaf)
7586 if (dev && dev->dev.parent) {
7587 dev_printk_emit(level[1] - '0',
7590 dev_driver_string(dev->dev.parent),
7591 dev_name(dev->dev.parent),
7592 netdev_name(dev), netdev_reg_state(dev),
7595 printk("%s%s%s: %pV",
7596 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7598 printk("%s(NULL net_device): %pV", level, vaf);
7602 void netdev_printk(const char *level, const struct net_device *dev,
7603 const char *format, ...)
7605 struct va_format vaf;
7608 va_start(args, format);
7613 __netdev_printk(level, dev, &vaf);
7617 EXPORT_SYMBOL(netdev_printk);
7619 #define define_netdev_printk_level(func, level) \
7620 void func(const struct net_device *dev, const char *fmt, ...) \
7622 struct va_format vaf; \
7625 va_start(args, fmt); \
7630 __netdev_printk(level, dev, &vaf); \
7634 EXPORT_SYMBOL(func);
7636 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7637 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7638 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7639 define_netdev_printk_level(netdev_err, KERN_ERR);
7640 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7641 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7642 define_netdev_printk_level(netdev_info, KERN_INFO);
7644 static void __net_exit netdev_exit(struct net *net)
7646 kfree(net->dev_name_head);
7647 kfree(net->dev_index_head);
7650 static struct pernet_operations __net_initdata netdev_net_ops = {
7651 .init = netdev_init,
7652 .exit = netdev_exit,
7655 static void __net_exit default_device_exit(struct net *net)
7657 struct net_device *dev, *aux;
7659 * Push all migratable network devices back to the
7660 * initial network namespace
7663 for_each_netdev_safe(net, dev, aux) {
7665 char fb_name[IFNAMSIZ];
7667 /* Ignore unmoveable devices (i.e. loopback) */
7668 if (dev->features & NETIF_F_NETNS_LOCAL)
7671 /* Leave virtual devices for the generic cleanup */
7672 if (dev->rtnl_link_ops)
7675 /* Push remaining network devices to init_net */
7676 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7677 err = dev_change_net_namespace(dev, &init_net, fb_name);
7679 pr_emerg("%s: failed to move %s to init_net: %d\n",
7680 __func__, dev->name, err);
7687 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7689 /* Return with the rtnl_lock held when there are no network
7690 * devices unregistering in any network namespace in net_list.
7694 DEFINE_WAIT_FUNC(wait, woken_wake_function);
7696 add_wait_queue(&netdev_unregistering_wq, &wait);
7698 unregistering = false;
7700 list_for_each_entry(net, net_list, exit_list) {
7701 if (net->dev_unreg_count > 0) {
7702 unregistering = true;
7710 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
7712 remove_wait_queue(&netdev_unregistering_wq, &wait);
7715 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7717 /* At exit all network devices most be removed from a network
7718 * namespace. Do this in the reverse order of registration.
7719 * Do this across as many network namespaces as possible to
7720 * improve batching efficiency.
7722 struct net_device *dev;
7724 LIST_HEAD(dev_kill_list);
7726 /* To prevent network device cleanup code from dereferencing
7727 * loopback devices or network devices that have been freed
7728 * wait here for all pending unregistrations to complete,
7729 * before unregistring the loopback device and allowing the
7730 * network namespace be freed.
7732 * The netdev todo list containing all network devices
7733 * unregistrations that happen in default_device_exit_batch
7734 * will run in the rtnl_unlock() at the end of
7735 * default_device_exit_batch.
7737 rtnl_lock_unregistering(net_list);
7738 list_for_each_entry(net, net_list, exit_list) {
7739 for_each_netdev_reverse(net, dev) {
7740 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7741 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7743 unregister_netdevice_queue(dev, &dev_kill_list);
7746 unregister_netdevice_many(&dev_kill_list);
7750 static struct pernet_operations __net_initdata default_device_ops = {
7751 .exit = default_device_exit,
7752 .exit_batch = default_device_exit_batch,
7756 * Initialize the DEV module. At boot time this walks the device list and
7757 * unhooks any devices that fail to initialise (normally hardware not
7758 * present) and leaves us with a valid list of present and active devices.
7763 * This is called single threaded during boot, so no need
7764 * to take the rtnl semaphore.
7766 static int __init net_dev_init(void)
7768 int i, rc = -ENOMEM;
7770 BUG_ON(!dev_boot_phase);
7772 if (dev_proc_init())
7775 if (netdev_kobject_init())
7778 INIT_LIST_HEAD(&ptype_all);
7779 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7780 INIT_LIST_HEAD(&ptype_base[i]);
7782 INIT_LIST_HEAD(&offload_base);
7784 if (register_pernet_subsys(&netdev_net_ops))
7788 * Initialise the packet receive queues.
7791 for_each_possible_cpu(i) {
7792 struct softnet_data *sd = &per_cpu(softnet_data, i);
7794 skb_queue_head_init(&sd->input_pkt_queue);
7795 skb_queue_head_init(&sd->process_queue);
7796 INIT_LIST_HEAD(&sd->poll_list);
7797 sd->output_queue_tailp = &sd->output_queue;
7799 sd->csd.func = rps_trigger_softirq;
7804 sd->backlog.poll = process_backlog;
7805 sd->backlog.weight = weight_p;
7810 /* The loopback device is special if any other network devices
7811 * is present in a network namespace the loopback device must
7812 * be present. Since we now dynamically allocate and free the
7813 * loopback device ensure this invariant is maintained by
7814 * keeping the loopback device as the first device on the
7815 * list of network devices. Ensuring the loopback devices
7816 * is the first device that appears and the last network device
7819 if (register_pernet_device(&loopback_net_ops))
7822 if (register_pernet_device(&default_device_ops))
7825 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7826 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7828 hotcpu_notifier(dev_cpu_callback, 0);
7835 subsys_initcall(net_dev_init);