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/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
121 #include <linux/ipv6.h>
122 #include <linux/in.h>
123 #include <linux/jhash.h>
124 #include <linux/random.h>
125 #include <trace/events/napi.h>
126 #include <trace/events/net.h>
127 #include <trace/events/skb.h>
128 #include <linux/pci.h>
129 #include <linux/inetdevice.h>
130 #include <linux/cpu_rmap.h>
131 #include <linux/static_key.h>
132 #include <linux/hashtable.h>
133 #include <linux/vmalloc.h>
134 #include <linux/if_macvlan.h>
135 #include <linux/errqueue.h>
137 #include "net-sysfs.h"
139 /* Instead of increasing this, you should create a hash table. */
140 #define MAX_GRO_SKBS 8
142 /* This should be increased if a protocol with a bigger head is added. */
143 #define GRO_MAX_HEAD (MAX_HEADER + 128)
145 static DEFINE_SPINLOCK(ptype_lock);
146 static DEFINE_SPINLOCK(offload_lock);
147 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
148 struct list_head ptype_all __read_mostly; /* Taps */
149 static struct list_head offload_base __read_mostly;
151 static int netif_rx_internal(struct sk_buff *skb);
152 static int call_netdevice_notifiers_info(unsigned long val,
153 struct net_device *dev,
154 struct netdev_notifier_info *info);
157 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
160 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
162 * Writers must hold the rtnl semaphore while they loop through the
163 * dev_base_head list, and hold dev_base_lock for writing when they do the
164 * actual updates. This allows pure readers to access the list even
165 * while a writer is preparing to update it.
167 * To put it another way, dev_base_lock is held for writing only to
168 * protect against pure readers; the rtnl semaphore provides the
169 * protection against other writers.
171 * See, for example usages, register_netdevice() and
172 * unregister_netdevice(), which must be called with the rtnl
175 DEFINE_RWLOCK(dev_base_lock);
176 EXPORT_SYMBOL(dev_base_lock);
178 /* protects napi_hash addition/deletion and napi_gen_id */
179 static DEFINE_SPINLOCK(napi_hash_lock);
181 static unsigned int napi_gen_id;
182 static DEFINE_HASHTABLE(napi_hash, 8);
184 static seqcount_t devnet_rename_seq;
186 static inline void dev_base_seq_inc(struct net *net)
188 while (++net->dev_base_seq == 0);
191 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
193 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
195 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
198 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
200 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
203 static inline void rps_lock(struct softnet_data *sd)
206 spin_lock(&sd->input_pkt_queue.lock);
210 static inline void rps_unlock(struct softnet_data *sd)
213 spin_unlock(&sd->input_pkt_queue.lock);
217 /* Device list insertion */
218 static void list_netdevice(struct net_device *dev)
220 struct net *net = dev_net(dev);
224 write_lock_bh(&dev_base_lock);
225 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
226 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
227 hlist_add_head_rcu(&dev->index_hlist,
228 dev_index_hash(net, dev->ifindex));
229 write_unlock_bh(&dev_base_lock);
231 dev_base_seq_inc(net);
234 /* Device list removal
235 * caller must respect a RCU grace period before freeing/reusing dev
237 static void unlist_netdevice(struct net_device *dev)
241 /* Unlink dev from the device chain */
242 write_lock_bh(&dev_base_lock);
243 list_del_rcu(&dev->dev_list);
244 hlist_del_rcu(&dev->name_hlist);
245 hlist_del_rcu(&dev->index_hlist);
246 write_unlock_bh(&dev_base_lock);
248 dev_base_seq_inc(dev_net(dev));
255 static RAW_NOTIFIER_HEAD(netdev_chain);
258 * Device drivers call our routines to queue packets here. We empty the
259 * queue in the local softnet handler.
262 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
263 EXPORT_PER_CPU_SYMBOL(softnet_data);
265 #ifdef CONFIG_LOCKDEP
267 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
268 * according to dev->type
270 static const unsigned short netdev_lock_type[] =
271 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
272 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
273 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
274 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
275 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
276 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
277 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
278 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
279 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
280 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
281 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
282 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
283 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
284 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
285 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
287 static const char *const netdev_lock_name[] =
288 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
289 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
290 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
291 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
292 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
293 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
294 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
295 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
296 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
297 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
298 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
299 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
300 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
301 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
302 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
304 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
305 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
307 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
311 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
312 if (netdev_lock_type[i] == dev_type)
314 /* the last key is used by default */
315 return ARRAY_SIZE(netdev_lock_type) - 1;
318 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
319 unsigned short dev_type)
323 i = netdev_lock_pos(dev_type);
324 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
325 netdev_lock_name[i]);
328 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
332 i = netdev_lock_pos(dev->type);
333 lockdep_set_class_and_name(&dev->addr_list_lock,
334 &netdev_addr_lock_key[i],
335 netdev_lock_name[i]);
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
339 unsigned short dev_type)
342 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
347 /*******************************************************************************
349 Protocol management and registration routines
351 *******************************************************************************/
354 * Add a protocol ID to the list. Now that the input handler is
355 * smarter we can dispense with all the messy stuff that used to be
358 * BEWARE!!! Protocol handlers, mangling input packets,
359 * MUST BE last in hash buckets and checking protocol handlers
360 * MUST start from promiscuous ptype_all chain in net_bh.
361 * It is true now, do not change it.
362 * Explanation follows: if protocol handler, mangling packet, will
363 * be the first on list, it is not able to sense, that packet
364 * is cloned and should be copied-on-write, so that it will
365 * change it and subsequent readers will get broken packet.
369 static inline struct list_head *ptype_head(const struct packet_type *pt)
371 if (pt->type == htons(ETH_P_ALL))
374 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
378 * dev_add_pack - add packet handler
379 * @pt: packet type declaration
381 * Add a protocol handler to the networking stack. The passed &packet_type
382 * is linked into kernel lists and may not be freed until it has been
383 * removed from the kernel lists.
385 * This call does not sleep therefore it can not
386 * guarantee all CPU's that are in middle of receiving packets
387 * will see the new packet type (until the next received packet).
390 void dev_add_pack(struct packet_type *pt)
392 struct list_head *head = ptype_head(pt);
394 spin_lock(&ptype_lock);
395 list_add_rcu(&pt->list, head);
396 spin_unlock(&ptype_lock);
398 EXPORT_SYMBOL(dev_add_pack);
401 * __dev_remove_pack - remove packet handler
402 * @pt: packet type declaration
404 * Remove a protocol handler that was previously added to the kernel
405 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
406 * from the kernel lists and can be freed or reused once this function
409 * The packet type might still be in use by receivers
410 * and must not be freed until after all the CPU's have gone
411 * through a quiescent state.
413 void __dev_remove_pack(struct packet_type *pt)
415 struct list_head *head = ptype_head(pt);
416 struct packet_type *pt1;
418 spin_lock(&ptype_lock);
420 list_for_each_entry(pt1, head, list) {
422 list_del_rcu(&pt->list);
427 pr_warn("dev_remove_pack: %p not found\n", pt);
429 spin_unlock(&ptype_lock);
431 EXPORT_SYMBOL(__dev_remove_pack);
434 * dev_remove_pack - remove packet handler
435 * @pt: packet type declaration
437 * Remove a protocol handler that was previously added to the kernel
438 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
439 * from the kernel lists and can be freed or reused once this function
442 * This call sleeps to guarantee that no CPU is looking at the packet
445 void dev_remove_pack(struct packet_type *pt)
447 __dev_remove_pack(pt);
451 EXPORT_SYMBOL(dev_remove_pack);
455 * dev_add_offload - register offload handlers
456 * @po: protocol offload declaration
458 * Add protocol offload handlers to the networking stack. The passed
459 * &proto_offload is linked into kernel lists and may not be freed until
460 * it has been removed from the kernel lists.
462 * This call does not sleep therefore it can not
463 * guarantee all CPU's that are in middle of receiving packets
464 * will see the new offload handlers (until the next received packet).
466 void dev_add_offload(struct packet_offload *po)
468 struct list_head *head = &offload_base;
470 spin_lock(&offload_lock);
471 list_add_rcu(&po->list, head);
472 spin_unlock(&offload_lock);
474 EXPORT_SYMBOL(dev_add_offload);
477 * __dev_remove_offload - remove offload handler
478 * @po: packet offload declaration
480 * Remove a protocol offload handler that was previously added to the
481 * kernel offload handlers by dev_add_offload(). The passed &offload_type
482 * is removed from the kernel lists and can be freed or reused once this
485 * The packet type might still be in use by receivers
486 * and must not be freed until after all the CPU's have gone
487 * through a quiescent state.
489 static void __dev_remove_offload(struct packet_offload *po)
491 struct list_head *head = &offload_base;
492 struct packet_offload *po1;
494 spin_lock(&offload_lock);
496 list_for_each_entry(po1, head, list) {
498 list_del_rcu(&po->list);
503 pr_warn("dev_remove_offload: %p not found\n", po);
505 spin_unlock(&offload_lock);
509 * dev_remove_offload - remove packet offload handler
510 * @po: packet offload declaration
512 * Remove a packet offload handler that was previously added to the kernel
513 * offload handlers by dev_add_offload(). The passed &offload_type is
514 * removed from the kernel lists and can be freed or reused once this
517 * This call sleeps to guarantee that no CPU is looking at the packet
520 void dev_remove_offload(struct packet_offload *po)
522 __dev_remove_offload(po);
526 EXPORT_SYMBOL(dev_remove_offload);
528 /******************************************************************************
530 Device Boot-time Settings Routines
532 *******************************************************************************/
534 /* Boot time configuration table */
535 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
538 * netdev_boot_setup_add - add new setup entry
539 * @name: name of the device
540 * @map: configured settings for the device
542 * Adds new setup entry to the dev_boot_setup list. The function
543 * returns 0 on error and 1 on success. This is a generic routine to
546 static int netdev_boot_setup_add(char *name, struct ifmap *map)
548 struct netdev_boot_setup *s;
552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
553 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
554 memset(s[i].name, 0, sizeof(s[i].name));
555 strlcpy(s[i].name, name, IFNAMSIZ);
556 memcpy(&s[i].map, map, sizeof(s[i].map));
561 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
565 * netdev_boot_setup_check - check boot time settings
566 * @dev: the netdevice
568 * Check boot time settings for the device.
569 * The found settings are set for the device to be used
570 * later in the device probing.
571 * Returns 0 if no settings found, 1 if they are.
573 int netdev_boot_setup_check(struct net_device *dev)
575 struct netdev_boot_setup *s = dev_boot_setup;
578 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
579 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
580 !strcmp(dev->name, s[i].name)) {
581 dev->irq = s[i].map.irq;
582 dev->base_addr = s[i].map.base_addr;
583 dev->mem_start = s[i].map.mem_start;
584 dev->mem_end = s[i].map.mem_end;
590 EXPORT_SYMBOL(netdev_boot_setup_check);
594 * netdev_boot_base - get address from boot time settings
595 * @prefix: prefix for network device
596 * @unit: id for network device
598 * Check boot time settings for the base address of device.
599 * The found settings are set for the device to be used
600 * later in the device probing.
601 * Returns 0 if no settings found.
603 unsigned long netdev_boot_base(const char *prefix, int unit)
605 const struct netdev_boot_setup *s = dev_boot_setup;
609 sprintf(name, "%s%d", prefix, unit);
612 * If device already registered then return base of 1
613 * to indicate not to probe for this interface
615 if (__dev_get_by_name(&init_net, name))
618 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
619 if (!strcmp(name, s[i].name))
620 return s[i].map.base_addr;
625 * Saves at boot time configured settings for any netdevice.
627 int __init netdev_boot_setup(char *str)
632 str = get_options(str, ARRAY_SIZE(ints), ints);
637 memset(&map, 0, sizeof(map));
641 map.base_addr = ints[2];
643 map.mem_start = ints[3];
645 map.mem_end = ints[4];
647 /* Add new entry to the list */
648 return netdev_boot_setup_add(str, &map);
651 __setup("netdev=", netdev_boot_setup);
653 /*******************************************************************************
655 Device Interface Subroutines
657 *******************************************************************************/
660 * __dev_get_by_name - find a device by its name
661 * @net: the applicable net namespace
662 * @name: name to find
664 * Find an interface by name. Must be called under RTNL semaphore
665 * or @dev_base_lock. If the name is found a pointer to the device
666 * is returned. If the name is not found then %NULL is returned. The
667 * reference counters are not incremented so the caller must be
668 * careful with locks.
671 struct net_device *__dev_get_by_name(struct net *net, const char *name)
673 struct net_device *dev;
674 struct hlist_head *head = dev_name_hash(net, name);
676 hlist_for_each_entry(dev, head, name_hlist)
677 if (!strncmp(dev->name, name, IFNAMSIZ))
682 EXPORT_SYMBOL(__dev_get_by_name);
685 * dev_get_by_name_rcu - find a device by its name
686 * @net: the applicable net namespace
687 * @name: name to find
689 * Find an interface by name.
690 * If the name is found a pointer to the device is returned.
691 * If the name is not found then %NULL is returned.
692 * The reference counters are not incremented so the caller must be
693 * careful with locks. The caller must hold RCU lock.
696 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
698 struct net_device *dev;
699 struct hlist_head *head = dev_name_hash(net, name);
701 hlist_for_each_entry_rcu(dev, head, name_hlist)
702 if (!strncmp(dev->name, name, IFNAMSIZ))
707 EXPORT_SYMBOL(dev_get_by_name_rcu);
710 * dev_get_by_name - find a device by its name
711 * @net: the applicable net namespace
712 * @name: name to find
714 * Find an interface by name. This can be called from any
715 * context and does its own locking. The returned handle has
716 * the usage count incremented and the caller must use dev_put() to
717 * release it when it is no longer needed. %NULL is returned if no
718 * matching device is found.
721 struct net_device *dev_get_by_name(struct net *net, const char *name)
723 struct net_device *dev;
726 dev = dev_get_by_name_rcu(net, name);
732 EXPORT_SYMBOL(dev_get_by_name);
735 * __dev_get_by_index - find a device by its ifindex
736 * @net: the applicable net namespace
737 * @ifindex: index of device
739 * Search for an interface by index. Returns %NULL if the device
740 * is not found or a pointer to the device. The device has not
741 * had its reference counter increased so the caller must be careful
742 * about locking. The caller must hold either the RTNL semaphore
746 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
748 struct net_device *dev;
749 struct hlist_head *head = dev_index_hash(net, ifindex);
751 hlist_for_each_entry(dev, head, index_hlist)
752 if (dev->ifindex == ifindex)
757 EXPORT_SYMBOL(__dev_get_by_index);
760 * dev_get_by_index_rcu - find a device by its ifindex
761 * @net: the applicable net namespace
762 * @ifindex: index of device
764 * Search for an interface by index. Returns %NULL if the device
765 * is not found or a pointer to the device. The device has not
766 * had its reference counter increased so the caller must be careful
767 * about locking. The caller must hold RCU lock.
770 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
772 struct net_device *dev;
773 struct hlist_head *head = dev_index_hash(net, ifindex);
775 hlist_for_each_entry_rcu(dev, head, index_hlist)
776 if (dev->ifindex == ifindex)
781 EXPORT_SYMBOL(dev_get_by_index_rcu);
785 * dev_get_by_index - find a device by its ifindex
786 * @net: the applicable net namespace
787 * @ifindex: index of device
789 * Search for an interface by index. Returns NULL if the device
790 * is not found or a pointer to the device. The device returned has
791 * had a reference added and the pointer is safe until the user calls
792 * dev_put to indicate they have finished with it.
795 struct net_device *dev_get_by_index(struct net *net, int ifindex)
797 struct net_device *dev;
800 dev = dev_get_by_index_rcu(net, ifindex);
806 EXPORT_SYMBOL(dev_get_by_index);
809 * netdev_get_name - get a netdevice name, knowing its ifindex.
810 * @net: network namespace
811 * @name: a pointer to the buffer where the name will be stored.
812 * @ifindex: the ifindex of the interface to get the name from.
814 * The use of raw_seqcount_begin() and cond_resched() before
815 * retrying is required as we want to give the writers a chance
816 * to complete when CONFIG_PREEMPT is not set.
818 int netdev_get_name(struct net *net, char *name, int ifindex)
820 struct net_device *dev;
824 seq = raw_seqcount_begin(&devnet_rename_seq);
826 dev = dev_get_by_index_rcu(net, ifindex);
832 strcpy(name, dev->name);
834 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
843 * dev_getbyhwaddr_rcu - find a device by its hardware address
844 * @net: the applicable net namespace
845 * @type: media type of device
846 * @ha: hardware address
848 * Search for an interface by MAC address. Returns NULL if the device
849 * is not found or a pointer to the device.
850 * The caller must hold RCU or RTNL.
851 * The returned device has not had its ref count increased
852 * and the caller must therefore be careful about locking
856 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
859 struct net_device *dev;
861 for_each_netdev_rcu(net, dev)
862 if (dev->type == type &&
863 !memcmp(dev->dev_addr, ha, dev->addr_len))
868 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
870 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
872 struct net_device *dev;
875 for_each_netdev(net, dev)
876 if (dev->type == type)
881 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
883 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
885 struct net_device *dev, *ret = NULL;
888 for_each_netdev_rcu(net, dev)
889 if (dev->type == type) {
897 EXPORT_SYMBOL(dev_getfirstbyhwtype);
900 * dev_get_by_flags_rcu - find any device with given flags
901 * @net: the applicable net namespace
902 * @if_flags: IFF_* values
903 * @mask: bitmask of bits in if_flags to check
905 * Search for any interface with the given flags. Returns NULL if a device
906 * is not found or a pointer to the device. Must be called inside
907 * rcu_read_lock(), and result refcount is unchanged.
910 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
913 struct net_device *dev, *ret;
916 for_each_netdev_rcu(net, dev) {
917 if (((dev->flags ^ if_flags) & mask) == 0) {
924 EXPORT_SYMBOL(dev_get_by_flags_rcu);
927 * dev_valid_name - check if name is okay for network device
930 * Network device names need to be valid file names to
931 * to allow sysfs to work. We also disallow any kind of
934 bool dev_valid_name(const char *name)
938 if (strlen(name) >= IFNAMSIZ)
940 if (!strcmp(name, ".") || !strcmp(name, ".."))
944 if (*name == '/' || isspace(*name))
950 EXPORT_SYMBOL(dev_valid_name);
953 * __dev_alloc_name - allocate a name for a device
954 * @net: network namespace to allocate the device name in
955 * @name: name format string
956 * @buf: scratch buffer and result name string
958 * Passed a format string - eg "lt%d" it will try and find a suitable
959 * id. It scans list of devices to build up a free map, then chooses
960 * the first empty slot. The caller must hold the dev_base or rtnl lock
961 * while allocating the name and adding the device in order to avoid
963 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
964 * Returns the number of the unit assigned or a negative errno code.
967 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
971 const int max_netdevices = 8*PAGE_SIZE;
972 unsigned long *inuse;
973 struct net_device *d;
975 p = strnchr(name, IFNAMSIZ-1, '%');
978 * Verify the string as this thing may have come from
979 * the user. There must be either one "%d" and no other "%"
982 if (p[1] != 'd' || strchr(p + 2, '%'))
985 /* Use one page as a bit array of possible slots */
986 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
990 for_each_netdev(net, d) {
991 if (!sscanf(d->name, name, &i))
993 if (i < 0 || i >= max_netdevices)
996 /* avoid cases where sscanf is not exact inverse of printf */
997 snprintf(buf, IFNAMSIZ, name, i);
998 if (!strncmp(buf, d->name, IFNAMSIZ))
1002 i = find_first_zero_bit(inuse, max_netdevices);
1003 free_page((unsigned long) inuse);
1007 snprintf(buf, IFNAMSIZ, name, i);
1008 if (!__dev_get_by_name(net, buf))
1011 /* It is possible to run out of possible slots
1012 * when the name is long and there isn't enough space left
1013 * for the digits, or if all bits are used.
1019 * dev_alloc_name - allocate a name for a device
1021 * @name: name format string
1023 * Passed a format string - eg "lt%d" it will try and find a suitable
1024 * id. It scans list of devices to build up a free map, then chooses
1025 * the first empty slot. The caller must hold the dev_base or rtnl lock
1026 * while allocating the name and adding the device in order to avoid
1028 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1029 * Returns the number of the unit assigned or a negative errno code.
1032 int dev_alloc_name(struct net_device *dev, const char *name)
1038 BUG_ON(!dev_net(dev));
1040 ret = __dev_alloc_name(net, name, buf);
1042 strlcpy(dev->name, buf, IFNAMSIZ);
1045 EXPORT_SYMBOL(dev_alloc_name);
1047 static int dev_alloc_name_ns(struct net *net,
1048 struct net_device *dev,
1054 ret = __dev_alloc_name(net, name, buf);
1056 strlcpy(dev->name, buf, IFNAMSIZ);
1060 static int dev_get_valid_name(struct net *net,
1061 struct net_device *dev,
1066 if (!dev_valid_name(name))
1069 if (strchr(name, '%'))
1070 return dev_alloc_name_ns(net, dev, name);
1071 else if (__dev_get_by_name(net, name))
1073 else if (dev->name != name)
1074 strlcpy(dev->name, name, IFNAMSIZ);
1080 * dev_change_name - change name of a device
1082 * @newname: name (or format string) must be at least IFNAMSIZ
1084 * Change name of a device, can pass format strings "eth%d".
1087 int dev_change_name(struct net_device *dev, const char *newname)
1089 unsigned char old_assign_type;
1090 char oldname[IFNAMSIZ];
1096 BUG_ON(!dev_net(dev));
1099 if (dev->flags & IFF_UP)
1102 write_seqcount_begin(&devnet_rename_seq);
1104 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1105 write_seqcount_end(&devnet_rename_seq);
1109 memcpy(oldname, dev->name, IFNAMSIZ);
1111 err = dev_get_valid_name(net, dev, newname);
1113 write_seqcount_end(&devnet_rename_seq);
1117 if (oldname[0] && !strchr(oldname, '%'))
1118 netdev_info(dev, "renamed from %s\n", oldname);
1120 old_assign_type = dev->name_assign_type;
1121 dev->name_assign_type = NET_NAME_RENAMED;
1124 ret = device_rename(&dev->dev, dev->name);
1126 memcpy(dev->name, oldname, IFNAMSIZ);
1127 dev->name_assign_type = old_assign_type;
1128 write_seqcount_end(&devnet_rename_seq);
1132 write_seqcount_end(&devnet_rename_seq);
1134 netdev_adjacent_rename_links(dev, oldname);
1136 write_lock_bh(&dev_base_lock);
1137 hlist_del_rcu(&dev->name_hlist);
1138 write_unlock_bh(&dev_base_lock);
1142 write_lock_bh(&dev_base_lock);
1143 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1144 write_unlock_bh(&dev_base_lock);
1146 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1147 ret = notifier_to_errno(ret);
1150 /* err >= 0 after dev_alloc_name() or stores the first errno */
1153 write_seqcount_begin(&devnet_rename_seq);
1154 memcpy(dev->name, oldname, IFNAMSIZ);
1155 memcpy(oldname, newname, IFNAMSIZ);
1156 dev->name_assign_type = old_assign_type;
1157 old_assign_type = NET_NAME_RENAMED;
1160 pr_err("%s: name change rollback failed: %d\n",
1169 * dev_set_alias - change ifalias of a device
1171 * @alias: name up to IFALIASZ
1172 * @len: limit of bytes to copy from info
1174 * Set ifalias for a device,
1176 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1182 if (len >= IFALIASZ)
1186 kfree(dev->ifalias);
1187 dev->ifalias = NULL;
1191 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1194 dev->ifalias = new_ifalias;
1196 strlcpy(dev->ifalias, alias, len+1);
1202 * netdev_features_change - device changes features
1203 * @dev: device to cause notification
1205 * Called to indicate a device has changed features.
1207 void netdev_features_change(struct net_device *dev)
1209 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1211 EXPORT_SYMBOL(netdev_features_change);
1214 * netdev_state_change - device changes state
1215 * @dev: device to cause notification
1217 * Called to indicate a device has changed state. This function calls
1218 * the notifier chains for netdev_chain and sends a NEWLINK message
1219 * to the routing socket.
1221 void netdev_state_change(struct net_device *dev)
1223 if (dev->flags & IFF_UP) {
1224 struct netdev_notifier_change_info change_info;
1226 change_info.flags_changed = 0;
1227 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1229 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1232 EXPORT_SYMBOL(netdev_state_change);
1235 * netdev_notify_peers - notify network peers about existence of @dev
1236 * @dev: network device
1238 * Generate traffic such that interested network peers are aware of
1239 * @dev, such as by generating a gratuitous ARP. This may be used when
1240 * a device wants to inform the rest of the network about some sort of
1241 * reconfiguration such as a failover event or virtual machine
1244 void netdev_notify_peers(struct net_device *dev)
1247 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1250 EXPORT_SYMBOL(netdev_notify_peers);
1252 static int __dev_open(struct net_device *dev)
1254 const struct net_device_ops *ops = dev->netdev_ops;
1259 if (!netif_device_present(dev))
1262 /* Block netpoll from trying to do any rx path servicing.
1263 * If we don't do this there is a chance ndo_poll_controller
1264 * or ndo_poll may be running while we open the device
1266 netpoll_poll_disable(dev);
1268 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1269 ret = notifier_to_errno(ret);
1273 set_bit(__LINK_STATE_START, &dev->state);
1275 if (ops->ndo_validate_addr)
1276 ret = ops->ndo_validate_addr(dev);
1278 if (!ret && ops->ndo_open)
1279 ret = ops->ndo_open(dev);
1281 netpoll_poll_enable(dev);
1284 clear_bit(__LINK_STATE_START, &dev->state);
1286 dev->flags |= IFF_UP;
1287 net_dmaengine_get();
1288 dev_set_rx_mode(dev);
1290 add_device_randomness(dev->dev_addr, dev->addr_len);
1297 * dev_open - prepare an interface for use.
1298 * @dev: device to open
1300 * Takes a device from down to up state. The device's private open
1301 * function is invoked and then the multicast lists are loaded. Finally
1302 * the device is moved into the up state and a %NETDEV_UP message is
1303 * sent to the netdev notifier chain.
1305 * Calling this function on an active interface is a nop. On a failure
1306 * a negative errno code is returned.
1308 int dev_open(struct net_device *dev)
1312 if (dev->flags & IFF_UP)
1315 ret = __dev_open(dev);
1319 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1320 call_netdevice_notifiers(NETDEV_UP, dev);
1324 EXPORT_SYMBOL(dev_open);
1326 static int __dev_close_many(struct list_head *head)
1328 struct net_device *dev;
1333 list_for_each_entry(dev, head, close_list) {
1334 /* Temporarily disable netpoll until the interface is down */
1335 netpoll_poll_disable(dev);
1337 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1339 clear_bit(__LINK_STATE_START, &dev->state);
1341 /* Synchronize to scheduled poll. We cannot touch poll list, it
1342 * can be even on different cpu. So just clear netif_running().
1344 * dev->stop() will invoke napi_disable() on all of it's
1345 * napi_struct instances on this device.
1347 smp_mb__after_atomic(); /* Commit netif_running(). */
1350 dev_deactivate_many(head);
1352 list_for_each_entry(dev, head, close_list) {
1353 const struct net_device_ops *ops = dev->netdev_ops;
1356 * Call the device specific close. This cannot fail.
1357 * Only if device is UP
1359 * We allow it to be called even after a DETACH hot-plug
1365 dev->flags &= ~IFF_UP;
1366 net_dmaengine_put();
1367 netpoll_poll_enable(dev);
1373 static int __dev_close(struct net_device *dev)
1378 list_add(&dev->close_list, &single);
1379 retval = __dev_close_many(&single);
1385 static int dev_close_many(struct list_head *head)
1387 struct net_device *dev, *tmp;
1389 /* Remove the devices that don't need to be closed */
1390 list_for_each_entry_safe(dev, tmp, head, close_list)
1391 if (!(dev->flags & IFF_UP))
1392 list_del_init(&dev->close_list);
1394 __dev_close_many(head);
1396 list_for_each_entry_safe(dev, tmp, head, close_list) {
1397 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1398 call_netdevice_notifiers(NETDEV_DOWN, dev);
1399 list_del_init(&dev->close_list);
1406 * dev_close - shutdown an interface.
1407 * @dev: device to shutdown
1409 * This function moves an active device into down state. A
1410 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1411 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1414 int dev_close(struct net_device *dev)
1416 if (dev->flags & IFF_UP) {
1419 list_add(&dev->close_list, &single);
1420 dev_close_many(&single);
1425 EXPORT_SYMBOL(dev_close);
1429 * dev_disable_lro - disable Large Receive Offload on a device
1432 * Disable Large Receive Offload (LRO) on a net device. Must be
1433 * called under RTNL. This is needed if received packets may be
1434 * forwarded to another interface.
1436 void dev_disable_lro(struct net_device *dev)
1439 * If we're trying to disable lro on a vlan device
1440 * use the underlying physical device instead
1442 if (is_vlan_dev(dev))
1443 dev = vlan_dev_real_dev(dev);
1445 /* the same for macvlan devices */
1446 if (netif_is_macvlan(dev))
1447 dev = macvlan_dev_real_dev(dev);
1449 dev->wanted_features &= ~NETIF_F_LRO;
1450 netdev_update_features(dev);
1452 if (unlikely(dev->features & NETIF_F_LRO))
1453 netdev_WARN(dev, "failed to disable LRO!\n");
1455 EXPORT_SYMBOL(dev_disable_lro);
1457 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1458 struct net_device *dev)
1460 struct netdev_notifier_info info;
1462 netdev_notifier_info_init(&info, dev);
1463 return nb->notifier_call(nb, val, &info);
1466 static int dev_boot_phase = 1;
1469 * register_netdevice_notifier - register a network notifier block
1472 * Register a notifier to be called when network device events occur.
1473 * The notifier passed is linked into the kernel structures and must
1474 * not be reused until it has been unregistered. A negative errno code
1475 * is returned on a failure.
1477 * When registered all registration and up events are replayed
1478 * to the new notifier to allow device to have a race free
1479 * view of the network device list.
1482 int register_netdevice_notifier(struct notifier_block *nb)
1484 struct net_device *dev;
1485 struct net_device *last;
1490 err = raw_notifier_chain_register(&netdev_chain, nb);
1496 for_each_netdev(net, dev) {
1497 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1498 err = notifier_to_errno(err);
1502 if (!(dev->flags & IFF_UP))
1505 call_netdevice_notifier(nb, NETDEV_UP, dev);
1516 for_each_netdev(net, dev) {
1520 if (dev->flags & IFF_UP) {
1521 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1523 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1525 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1530 raw_notifier_chain_unregister(&netdev_chain, nb);
1533 EXPORT_SYMBOL(register_netdevice_notifier);
1536 * unregister_netdevice_notifier - unregister a network notifier block
1539 * Unregister a notifier previously registered by
1540 * register_netdevice_notifier(). The notifier is unlinked into the
1541 * kernel structures and may then be reused. A negative errno code
1542 * is returned on a failure.
1544 * After unregistering unregister and down device events are synthesized
1545 * for all devices on the device list to the removed notifier to remove
1546 * the need for special case cleanup code.
1549 int unregister_netdevice_notifier(struct notifier_block *nb)
1551 struct net_device *dev;
1556 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1561 for_each_netdev(net, dev) {
1562 if (dev->flags & IFF_UP) {
1563 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1565 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1567 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1574 EXPORT_SYMBOL(unregister_netdevice_notifier);
1577 * call_netdevice_notifiers_info - call all network notifier blocks
1578 * @val: value passed unmodified to notifier function
1579 * @dev: net_device pointer passed unmodified to notifier function
1580 * @info: notifier information data
1582 * Call all network notifier blocks. Parameters and return value
1583 * are as for raw_notifier_call_chain().
1586 static int call_netdevice_notifiers_info(unsigned long val,
1587 struct net_device *dev,
1588 struct netdev_notifier_info *info)
1591 netdev_notifier_info_init(info, dev);
1592 return raw_notifier_call_chain(&netdev_chain, val, info);
1596 * call_netdevice_notifiers - call all network notifier blocks
1597 * @val: value passed unmodified to notifier function
1598 * @dev: net_device pointer passed unmodified to notifier function
1600 * Call all network notifier blocks. Parameters and return value
1601 * are as for raw_notifier_call_chain().
1604 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1606 struct netdev_notifier_info info;
1608 return call_netdevice_notifiers_info(val, dev, &info);
1610 EXPORT_SYMBOL(call_netdevice_notifiers);
1612 static struct static_key netstamp_needed __read_mostly;
1613 #ifdef HAVE_JUMP_LABEL
1614 /* We are not allowed to call static_key_slow_dec() from irq context
1615 * If net_disable_timestamp() is called from irq context, defer the
1616 * static_key_slow_dec() calls.
1618 static atomic_t netstamp_needed_deferred;
1621 void net_enable_timestamp(void)
1623 #ifdef HAVE_JUMP_LABEL
1624 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1628 static_key_slow_dec(&netstamp_needed);
1632 static_key_slow_inc(&netstamp_needed);
1634 EXPORT_SYMBOL(net_enable_timestamp);
1636 void net_disable_timestamp(void)
1638 #ifdef HAVE_JUMP_LABEL
1639 if (in_interrupt()) {
1640 atomic_inc(&netstamp_needed_deferred);
1644 static_key_slow_dec(&netstamp_needed);
1646 EXPORT_SYMBOL(net_disable_timestamp);
1648 static inline void net_timestamp_set(struct sk_buff *skb)
1650 skb->tstamp.tv64 = 0;
1651 if (static_key_false(&netstamp_needed))
1652 __net_timestamp(skb);
1655 #define net_timestamp_check(COND, SKB) \
1656 if (static_key_false(&netstamp_needed)) { \
1657 if ((COND) && !(SKB)->tstamp.tv64) \
1658 __net_timestamp(SKB); \
1661 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1665 if (!(dev->flags & IFF_UP))
1668 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1669 if (skb->len <= len)
1672 /* if TSO is enabled, we don't care about the length as the packet
1673 * could be forwarded without being segmented before
1675 if (skb_is_gso(skb))
1680 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1682 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1684 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1685 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1686 atomic_long_inc(&dev->rx_dropped);
1692 if (unlikely(!is_skb_forwardable(dev, skb))) {
1693 atomic_long_inc(&dev->rx_dropped);
1698 skb_scrub_packet(skb, true);
1699 skb->protocol = eth_type_trans(skb, dev);
1703 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1706 * dev_forward_skb - loopback an skb to another netif
1708 * @dev: destination network device
1709 * @skb: buffer to forward
1712 * NET_RX_SUCCESS (no congestion)
1713 * NET_RX_DROP (packet was dropped, but freed)
1715 * dev_forward_skb can be used for injecting an skb from the
1716 * start_xmit function of one device into the receive queue
1717 * of another device.
1719 * The receiving device may be in another namespace, so
1720 * we have to clear all information in the skb that could
1721 * impact namespace isolation.
1723 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1725 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1727 EXPORT_SYMBOL_GPL(dev_forward_skb);
1729 static inline int deliver_skb(struct sk_buff *skb,
1730 struct packet_type *pt_prev,
1731 struct net_device *orig_dev)
1733 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1735 atomic_inc(&skb->users);
1736 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1739 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1741 if (!ptype->af_packet_priv || !skb->sk)
1744 if (ptype->id_match)
1745 return ptype->id_match(ptype, skb->sk);
1746 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1753 * Support routine. Sends outgoing frames to any network
1754 * taps currently in use.
1757 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1759 struct packet_type *ptype;
1760 struct sk_buff *skb2 = NULL;
1761 struct packet_type *pt_prev = NULL;
1764 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1765 /* Never send packets back to the socket
1766 * they originated from - MvS (miquels@drinkel.ow.org)
1768 if ((ptype->dev == dev || !ptype->dev) &&
1769 (!skb_loop_sk(ptype, skb))) {
1771 deliver_skb(skb2, pt_prev, skb->dev);
1776 skb2 = skb_clone(skb, GFP_ATOMIC);
1780 net_timestamp_set(skb2);
1782 /* skb->nh should be correctly
1783 set by sender, so that the second statement is
1784 just protection against buggy protocols.
1786 skb_reset_mac_header(skb2);
1788 if (skb_network_header(skb2) < skb2->data ||
1789 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1790 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1791 ntohs(skb2->protocol),
1793 skb_reset_network_header(skb2);
1796 skb2->transport_header = skb2->network_header;
1797 skb2->pkt_type = PACKET_OUTGOING;
1802 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1807 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1808 * @dev: Network device
1809 * @txq: number of queues available
1811 * If real_num_tx_queues is changed the tc mappings may no longer be
1812 * valid. To resolve this verify the tc mapping remains valid and if
1813 * not NULL the mapping. With no priorities mapping to this
1814 * offset/count pair it will no longer be used. In the worst case TC0
1815 * is invalid nothing can be done so disable priority mappings. If is
1816 * expected that drivers will fix this mapping if they can before
1817 * calling netif_set_real_num_tx_queues.
1819 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1822 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1824 /* If TC0 is invalidated disable TC mapping */
1825 if (tc->offset + tc->count > txq) {
1826 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1831 /* Invalidated prio to tc mappings set to TC0 */
1832 for (i = 1; i < TC_BITMASK + 1; i++) {
1833 int q = netdev_get_prio_tc_map(dev, i);
1835 tc = &dev->tc_to_txq[q];
1836 if (tc->offset + tc->count > txq) {
1837 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1839 netdev_set_prio_tc_map(dev, i, 0);
1845 static DEFINE_MUTEX(xps_map_mutex);
1846 #define xmap_dereference(P) \
1847 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1849 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1852 struct xps_map *map = NULL;
1856 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1858 for (pos = 0; map && pos < map->len; pos++) {
1859 if (map->queues[pos] == index) {
1861 map->queues[pos] = map->queues[--map->len];
1863 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1864 kfree_rcu(map, rcu);
1874 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1876 struct xps_dev_maps *dev_maps;
1878 bool active = false;
1880 mutex_lock(&xps_map_mutex);
1881 dev_maps = xmap_dereference(dev->xps_maps);
1886 for_each_possible_cpu(cpu) {
1887 for (i = index; i < dev->num_tx_queues; i++) {
1888 if (!remove_xps_queue(dev_maps, cpu, i))
1891 if (i == dev->num_tx_queues)
1896 RCU_INIT_POINTER(dev->xps_maps, NULL);
1897 kfree_rcu(dev_maps, rcu);
1900 for (i = index; i < dev->num_tx_queues; i++)
1901 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1905 mutex_unlock(&xps_map_mutex);
1908 static struct xps_map *expand_xps_map(struct xps_map *map,
1911 struct xps_map *new_map;
1912 int alloc_len = XPS_MIN_MAP_ALLOC;
1915 for (pos = 0; map && pos < map->len; pos++) {
1916 if (map->queues[pos] != index)
1921 /* Need to add queue to this CPU's existing map */
1923 if (pos < map->alloc_len)
1926 alloc_len = map->alloc_len * 2;
1929 /* Need to allocate new map to store queue on this CPU's map */
1930 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1935 for (i = 0; i < pos; i++)
1936 new_map->queues[i] = map->queues[i];
1937 new_map->alloc_len = alloc_len;
1943 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1946 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1947 struct xps_map *map, *new_map;
1948 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1949 int cpu, numa_node_id = -2;
1950 bool active = false;
1952 mutex_lock(&xps_map_mutex);
1954 dev_maps = xmap_dereference(dev->xps_maps);
1956 /* allocate memory for queue storage */
1957 for_each_online_cpu(cpu) {
1958 if (!cpumask_test_cpu(cpu, mask))
1962 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1963 if (!new_dev_maps) {
1964 mutex_unlock(&xps_map_mutex);
1968 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1971 map = expand_xps_map(map, cpu, index);
1975 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1979 goto out_no_new_maps;
1981 for_each_possible_cpu(cpu) {
1982 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1983 /* add queue to CPU maps */
1986 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1987 while ((pos < map->len) && (map->queues[pos] != index))
1990 if (pos == map->len)
1991 map->queues[map->len++] = index;
1993 if (numa_node_id == -2)
1994 numa_node_id = cpu_to_node(cpu);
1995 else if (numa_node_id != cpu_to_node(cpu))
1998 } else if (dev_maps) {
1999 /* fill in the new device map from the old device map */
2000 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2001 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2006 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2008 /* Cleanup old maps */
2010 for_each_possible_cpu(cpu) {
2011 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2012 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2013 if (map && map != new_map)
2014 kfree_rcu(map, rcu);
2017 kfree_rcu(dev_maps, rcu);
2020 dev_maps = new_dev_maps;
2024 /* update Tx queue numa node */
2025 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2026 (numa_node_id >= 0) ? numa_node_id :
2032 /* removes queue from unused CPUs */
2033 for_each_possible_cpu(cpu) {
2034 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2037 if (remove_xps_queue(dev_maps, cpu, index))
2041 /* free map if not active */
2043 RCU_INIT_POINTER(dev->xps_maps, NULL);
2044 kfree_rcu(dev_maps, rcu);
2048 mutex_unlock(&xps_map_mutex);
2052 /* remove any maps that we added */
2053 for_each_possible_cpu(cpu) {
2054 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2055 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2057 if (new_map && new_map != map)
2061 mutex_unlock(&xps_map_mutex);
2063 kfree(new_dev_maps);
2066 EXPORT_SYMBOL(netif_set_xps_queue);
2070 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2071 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2073 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2077 if (txq < 1 || txq > dev->num_tx_queues)
2080 if (dev->reg_state == NETREG_REGISTERED ||
2081 dev->reg_state == NETREG_UNREGISTERING) {
2084 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2090 netif_setup_tc(dev, txq);
2092 if (txq < dev->real_num_tx_queues) {
2093 qdisc_reset_all_tx_gt(dev, txq);
2095 netif_reset_xps_queues_gt(dev, txq);
2100 dev->real_num_tx_queues = txq;
2103 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2107 * netif_set_real_num_rx_queues - set actual number of RX queues used
2108 * @dev: Network device
2109 * @rxq: Actual number of RX queues
2111 * This must be called either with the rtnl_lock held or before
2112 * registration of the net device. Returns 0 on success, or a
2113 * negative error code. If called before registration, it always
2116 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2120 if (rxq < 1 || rxq > dev->num_rx_queues)
2123 if (dev->reg_state == NETREG_REGISTERED) {
2126 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2132 dev->real_num_rx_queues = rxq;
2135 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2139 * netif_get_num_default_rss_queues - default number of RSS queues
2141 * This routine should set an upper limit on the number of RSS queues
2142 * used by default by multiqueue devices.
2144 int netif_get_num_default_rss_queues(void)
2146 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2148 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2150 static inline void __netif_reschedule(struct Qdisc *q)
2152 struct softnet_data *sd;
2153 unsigned long flags;
2155 local_irq_save(flags);
2156 sd = &__get_cpu_var(softnet_data);
2157 q->next_sched = NULL;
2158 *sd->output_queue_tailp = q;
2159 sd->output_queue_tailp = &q->next_sched;
2160 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2161 local_irq_restore(flags);
2164 void __netif_schedule(struct Qdisc *q)
2166 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2167 __netif_reschedule(q);
2169 EXPORT_SYMBOL(__netif_schedule);
2171 struct dev_kfree_skb_cb {
2172 enum skb_free_reason reason;
2175 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2177 return (struct dev_kfree_skb_cb *)skb->cb;
2180 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2182 unsigned long flags;
2184 if (likely(atomic_read(&skb->users) == 1)) {
2186 atomic_set(&skb->users, 0);
2187 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2190 get_kfree_skb_cb(skb)->reason = reason;
2191 local_irq_save(flags);
2192 skb->next = __this_cpu_read(softnet_data.completion_queue);
2193 __this_cpu_write(softnet_data.completion_queue, skb);
2194 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2195 local_irq_restore(flags);
2197 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2199 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2201 if (in_irq() || irqs_disabled())
2202 __dev_kfree_skb_irq(skb, reason);
2206 EXPORT_SYMBOL(__dev_kfree_skb_any);
2210 * netif_device_detach - mark device as removed
2211 * @dev: network device
2213 * Mark device as removed from system and therefore no longer available.
2215 void netif_device_detach(struct net_device *dev)
2217 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2218 netif_running(dev)) {
2219 netif_tx_stop_all_queues(dev);
2222 EXPORT_SYMBOL(netif_device_detach);
2225 * netif_device_attach - mark device as attached
2226 * @dev: network device
2228 * Mark device as attached from system and restart if needed.
2230 void netif_device_attach(struct net_device *dev)
2232 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2233 netif_running(dev)) {
2234 netif_tx_wake_all_queues(dev);
2235 __netdev_watchdog_up(dev);
2238 EXPORT_SYMBOL(netif_device_attach);
2240 static void skb_warn_bad_offload(const struct sk_buff *skb)
2242 static const netdev_features_t null_features = 0;
2243 struct net_device *dev = skb->dev;
2244 const char *driver = "";
2246 if (!net_ratelimit())
2249 if (dev && dev->dev.parent)
2250 driver = dev_driver_string(dev->dev.parent);
2252 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2253 "gso_type=%d ip_summed=%d\n",
2254 driver, dev ? &dev->features : &null_features,
2255 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2256 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2257 skb_shinfo(skb)->gso_type, skb->ip_summed);
2261 * Invalidate hardware checksum when packet is to be mangled, and
2262 * complete checksum manually on outgoing path.
2264 int skb_checksum_help(struct sk_buff *skb)
2267 int ret = 0, offset;
2269 if (skb->ip_summed == CHECKSUM_COMPLETE)
2270 goto out_set_summed;
2272 if (unlikely(skb_shinfo(skb)->gso_size)) {
2273 skb_warn_bad_offload(skb);
2277 /* Before computing a checksum, we should make sure no frag could
2278 * be modified by an external entity : checksum could be wrong.
2280 if (skb_has_shared_frag(skb)) {
2281 ret = __skb_linearize(skb);
2286 offset = skb_checksum_start_offset(skb);
2287 BUG_ON(offset >= skb_headlen(skb));
2288 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2290 offset += skb->csum_offset;
2291 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2293 if (skb_cloned(skb) &&
2294 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2295 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2300 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2302 skb->ip_summed = CHECKSUM_NONE;
2306 EXPORT_SYMBOL(skb_checksum_help);
2308 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2310 unsigned int vlan_depth = skb->mac_len;
2311 __be16 type = skb->protocol;
2313 /* Tunnel gso handlers can set protocol to ethernet. */
2314 if (type == htons(ETH_P_TEB)) {
2317 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2320 eth = (struct ethhdr *)skb_mac_header(skb);
2321 type = eth->h_proto;
2324 /* if skb->protocol is 802.1Q/AD then the header should already be
2325 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at
2326 * ETH_HLEN otherwise
2328 if (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2330 if (WARN_ON(vlan_depth < VLAN_HLEN))
2332 vlan_depth -= VLAN_HLEN;
2334 vlan_depth = ETH_HLEN;
2337 struct vlan_hdr *vh;
2339 if (unlikely(!pskb_may_pull(skb,
2340 vlan_depth + VLAN_HLEN)))
2343 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2344 type = vh->h_vlan_encapsulated_proto;
2345 vlan_depth += VLAN_HLEN;
2346 } while (type == htons(ETH_P_8021Q) ||
2347 type == htons(ETH_P_8021AD));
2350 *depth = vlan_depth;
2356 * skb_mac_gso_segment - mac layer segmentation handler.
2357 * @skb: buffer to segment
2358 * @features: features for the output path (see dev->features)
2360 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2361 netdev_features_t features)
2363 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2364 struct packet_offload *ptype;
2365 int vlan_depth = skb->mac_len;
2366 __be16 type = skb_network_protocol(skb, &vlan_depth);
2368 if (unlikely(!type))
2369 return ERR_PTR(-EINVAL);
2371 __skb_pull(skb, vlan_depth);
2374 list_for_each_entry_rcu(ptype, &offload_base, list) {
2375 if (ptype->type == type && ptype->callbacks.gso_segment) {
2376 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2379 err = ptype->callbacks.gso_send_check(skb);
2380 segs = ERR_PTR(err);
2381 if (err || skb_gso_ok(skb, features))
2383 __skb_push(skb, (skb->data -
2384 skb_network_header(skb)));
2386 segs = ptype->callbacks.gso_segment(skb, features);
2392 __skb_push(skb, skb->data - skb_mac_header(skb));
2396 EXPORT_SYMBOL(skb_mac_gso_segment);
2399 /* openvswitch calls this on rx path, so we need a different check.
2401 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2404 return skb->ip_summed != CHECKSUM_PARTIAL;
2406 return skb->ip_summed == CHECKSUM_NONE;
2410 * __skb_gso_segment - Perform segmentation on skb.
2411 * @skb: buffer to segment
2412 * @features: features for the output path (see dev->features)
2413 * @tx_path: whether it is called in TX path
2415 * This function segments the given skb and returns a list of segments.
2417 * It may return NULL if the skb requires no segmentation. This is
2418 * only possible when GSO is used for verifying header integrity.
2420 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2421 netdev_features_t features, bool tx_path)
2423 if (unlikely(skb_needs_check(skb, tx_path))) {
2426 skb_warn_bad_offload(skb);
2428 err = skb_cow_head(skb, 0);
2430 return ERR_PTR(err);
2433 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2434 SKB_GSO_CB(skb)->encap_level = 0;
2436 skb_reset_mac_header(skb);
2437 skb_reset_mac_len(skb);
2439 return skb_mac_gso_segment(skb, features);
2441 EXPORT_SYMBOL(__skb_gso_segment);
2443 /* Take action when hardware reception checksum errors are detected. */
2445 void netdev_rx_csum_fault(struct net_device *dev)
2447 if (net_ratelimit()) {
2448 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2452 EXPORT_SYMBOL(netdev_rx_csum_fault);
2455 /* Actually, we should eliminate this check as soon as we know, that:
2456 * 1. IOMMU is present and allows to map all the memory.
2457 * 2. No high memory really exists on this machine.
2460 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2462 #ifdef CONFIG_HIGHMEM
2464 if (!(dev->features & NETIF_F_HIGHDMA)) {
2465 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2466 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2467 if (PageHighMem(skb_frag_page(frag)))
2472 if (PCI_DMA_BUS_IS_PHYS) {
2473 struct device *pdev = dev->dev.parent;
2477 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2478 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2479 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2480 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2488 /* If MPLS offload request, verify we are testing hardware MPLS features
2489 * instead of standard features for the netdev.
2491 #ifdef CONFIG_NET_MPLS_GSO
2492 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2493 netdev_features_t features,
2496 if (type == htons(ETH_P_MPLS_UC) || type == htons(ETH_P_MPLS_MC))
2497 features &= skb->dev->mpls_features;
2502 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2503 netdev_features_t features,
2510 static netdev_features_t harmonize_features(struct sk_buff *skb,
2511 netdev_features_t features)
2516 type = skb_network_protocol(skb, &tmp);
2517 features = net_mpls_features(skb, features, type);
2519 if (skb->ip_summed != CHECKSUM_NONE &&
2520 !can_checksum_protocol(features, type)) {
2521 features &= ~NETIF_F_ALL_CSUM;
2522 } else if (illegal_highdma(skb->dev, skb)) {
2523 features &= ~NETIF_F_SG;
2529 netdev_features_t netif_skb_features(struct sk_buff *skb)
2531 __be16 protocol = skb->protocol;
2532 netdev_features_t features = skb->dev->features;
2534 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2535 features &= ~NETIF_F_GSO_MASK;
2537 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2538 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2539 protocol = veh->h_vlan_encapsulated_proto;
2540 } else if (!vlan_tx_tag_present(skb)) {
2541 return harmonize_features(skb, features);
2544 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2545 NETIF_F_HW_VLAN_STAG_TX);
2547 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2548 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2549 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2550 NETIF_F_HW_VLAN_STAG_TX;
2552 return harmonize_features(skb, features);
2554 EXPORT_SYMBOL(netif_skb_features);
2556 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2557 struct netdev_queue *txq, bool more)
2562 if (!list_empty(&ptype_all))
2563 dev_queue_xmit_nit(skb, dev);
2566 trace_net_dev_start_xmit(skb, dev);
2567 rc = netdev_start_xmit(skb, dev, txq, more);
2568 trace_net_dev_xmit(skb, rc, dev, len);
2573 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2574 struct netdev_queue *txq, int *ret)
2576 struct sk_buff *skb = first;
2577 int rc = NETDEV_TX_OK;
2580 struct sk_buff *next = skb->next;
2583 rc = xmit_one(skb, dev, txq, next != NULL);
2584 if (unlikely(!dev_xmit_complete(rc))) {
2590 if (netif_xmit_stopped(txq) && skb) {
2591 rc = NETDEV_TX_BUSY;
2601 struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, netdev_features_t features)
2603 if (vlan_tx_tag_present(skb) &&
2604 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2605 skb = __vlan_put_tag(skb, skb->vlan_proto,
2606 vlan_tx_tag_get(skb));
2613 struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2615 netdev_features_t features;
2620 /* If device doesn't need skb->dst, release it right now while
2621 * its hot in this cpu cache
2623 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2626 features = netif_skb_features(skb);
2627 skb = validate_xmit_vlan(skb, features);
2631 /* If encapsulation offload request, verify we are testing
2632 * hardware encapsulation features instead of standard
2633 * features for the netdev
2635 if (skb->encapsulation)
2636 features &= dev->hw_enc_features;
2638 if (netif_needs_gso(skb, features)) {
2639 struct sk_buff *segs;
2641 segs = skb_gso_segment(skb, features);
2647 if (skb_needs_linearize(skb, features) &&
2648 __skb_linearize(skb))
2651 /* If packet is not checksummed and device does not
2652 * support checksumming for this protocol, complete
2653 * checksumming here.
2655 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2656 if (skb->encapsulation)
2657 skb_set_inner_transport_header(skb,
2658 skb_checksum_start_offset(skb));
2660 skb_set_transport_header(skb,
2661 skb_checksum_start_offset(skb));
2662 if (!(features & NETIF_F_ALL_CSUM) &&
2663 skb_checksum_help(skb))
2676 static void qdisc_pkt_len_init(struct sk_buff *skb)
2678 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2680 qdisc_skb_cb(skb)->pkt_len = skb->len;
2682 /* To get more precise estimation of bytes sent on wire,
2683 * we add to pkt_len the headers size of all segments
2685 if (shinfo->gso_size) {
2686 unsigned int hdr_len;
2687 u16 gso_segs = shinfo->gso_segs;
2689 /* mac layer + network layer */
2690 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2692 /* + transport layer */
2693 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2694 hdr_len += tcp_hdrlen(skb);
2696 hdr_len += sizeof(struct udphdr);
2698 if (shinfo->gso_type & SKB_GSO_DODGY)
2699 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2702 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2706 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2707 struct net_device *dev,
2708 struct netdev_queue *txq)
2710 spinlock_t *root_lock = qdisc_lock(q);
2714 qdisc_pkt_len_init(skb);
2715 qdisc_calculate_pkt_len(skb, q);
2717 * Heuristic to force contended enqueues to serialize on a
2718 * separate lock before trying to get qdisc main lock.
2719 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2720 * often and dequeue packets faster.
2722 contended = qdisc_is_running(q);
2723 if (unlikely(contended))
2724 spin_lock(&q->busylock);
2726 spin_lock(root_lock);
2727 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2730 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2731 qdisc_run_begin(q)) {
2733 * This is a work-conserving queue; there are no old skbs
2734 * waiting to be sent out; and the qdisc is not running -
2735 * xmit the skb directly.
2737 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2740 qdisc_bstats_update(q, skb);
2742 skb = validate_xmit_skb(skb, dev);
2743 if (skb && sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2744 if (unlikely(contended)) {
2745 spin_unlock(&q->busylock);
2752 rc = NET_XMIT_SUCCESS;
2755 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2756 if (qdisc_run_begin(q)) {
2757 if (unlikely(contended)) {
2758 spin_unlock(&q->busylock);
2764 spin_unlock(root_lock);
2765 if (unlikely(contended))
2766 spin_unlock(&q->busylock);
2770 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2771 static void skb_update_prio(struct sk_buff *skb)
2773 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2775 if (!skb->priority && skb->sk && map) {
2776 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2778 if (prioidx < map->priomap_len)
2779 skb->priority = map->priomap[prioidx];
2783 #define skb_update_prio(skb)
2786 static DEFINE_PER_CPU(int, xmit_recursion);
2787 #define RECURSION_LIMIT 10
2790 * dev_loopback_xmit - loop back @skb
2791 * @skb: buffer to transmit
2793 int dev_loopback_xmit(struct sk_buff *skb)
2795 skb_reset_mac_header(skb);
2796 __skb_pull(skb, skb_network_offset(skb));
2797 skb->pkt_type = PACKET_LOOPBACK;
2798 skb->ip_summed = CHECKSUM_UNNECESSARY;
2799 WARN_ON(!skb_dst(skb));
2804 EXPORT_SYMBOL(dev_loopback_xmit);
2807 * __dev_queue_xmit - transmit a buffer
2808 * @skb: buffer to transmit
2809 * @accel_priv: private data used for L2 forwarding offload
2811 * Queue a buffer for transmission to a network device. The caller must
2812 * have set the device and priority and built the buffer before calling
2813 * this function. The function can be called from an interrupt.
2815 * A negative errno code is returned on a failure. A success does not
2816 * guarantee the frame will be transmitted as it may be dropped due
2817 * to congestion or traffic shaping.
2819 * -----------------------------------------------------------------------------------
2820 * I notice this method can also return errors from the queue disciplines,
2821 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2824 * Regardless of the return value, the skb is consumed, so it is currently
2825 * difficult to retry a send to this method. (You can bump the ref count
2826 * before sending to hold a reference for retry if you are careful.)
2828 * When calling this method, interrupts MUST be enabled. This is because
2829 * the BH enable code must have IRQs enabled so that it will not deadlock.
2832 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2834 struct net_device *dev = skb->dev;
2835 struct netdev_queue *txq;
2839 skb_reset_mac_header(skb);
2841 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
2842 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
2844 /* Disable soft irqs for various locks below. Also
2845 * stops preemption for RCU.
2849 skb_update_prio(skb);
2851 txq = netdev_pick_tx(dev, skb, accel_priv);
2852 q = rcu_dereference_bh(txq->qdisc);
2854 #ifdef CONFIG_NET_CLS_ACT
2855 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2857 trace_net_dev_queue(skb);
2859 rc = __dev_xmit_skb(skb, q, dev, txq);
2863 /* The device has no queue. Common case for software devices:
2864 loopback, all the sorts of tunnels...
2866 Really, it is unlikely that netif_tx_lock protection is necessary
2867 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2869 However, it is possible, that they rely on protection
2872 Check this and shot the lock. It is not prone from deadlocks.
2873 Either shot noqueue qdisc, it is even simpler 8)
2875 if (dev->flags & IFF_UP) {
2876 int cpu = smp_processor_id(); /* ok because BHs are off */
2878 if (txq->xmit_lock_owner != cpu) {
2880 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2881 goto recursion_alert;
2883 skb = validate_xmit_skb(skb, dev);
2887 HARD_TX_LOCK(dev, txq, cpu);
2889 if (!netif_xmit_stopped(txq)) {
2890 __this_cpu_inc(xmit_recursion);
2891 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
2892 __this_cpu_dec(xmit_recursion);
2893 if (dev_xmit_complete(rc)) {
2894 HARD_TX_UNLOCK(dev, txq);
2898 HARD_TX_UNLOCK(dev, txq);
2899 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2902 /* Recursion is detected! It is possible,
2906 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2913 rcu_read_unlock_bh();
2915 atomic_long_inc(&dev->tx_dropped);
2916 kfree_skb_list(skb);
2919 rcu_read_unlock_bh();
2923 int dev_queue_xmit(struct sk_buff *skb)
2925 return __dev_queue_xmit(skb, NULL);
2927 EXPORT_SYMBOL(dev_queue_xmit);
2929 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
2931 return __dev_queue_xmit(skb, accel_priv);
2933 EXPORT_SYMBOL(dev_queue_xmit_accel);
2936 /*=======================================================================
2938 =======================================================================*/
2940 int netdev_max_backlog __read_mostly = 1000;
2941 EXPORT_SYMBOL(netdev_max_backlog);
2943 int netdev_tstamp_prequeue __read_mostly = 1;
2944 int netdev_budget __read_mostly = 300;
2945 int weight_p __read_mostly = 64; /* old backlog weight */
2947 /* Called with irq disabled */
2948 static inline void ____napi_schedule(struct softnet_data *sd,
2949 struct napi_struct *napi)
2951 list_add_tail(&napi->poll_list, &sd->poll_list);
2952 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2957 /* One global table that all flow-based protocols share. */
2958 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2959 EXPORT_SYMBOL(rps_sock_flow_table);
2961 struct static_key rps_needed __read_mostly;
2963 static struct rps_dev_flow *
2964 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2965 struct rps_dev_flow *rflow, u16 next_cpu)
2967 if (next_cpu != RPS_NO_CPU) {
2968 #ifdef CONFIG_RFS_ACCEL
2969 struct netdev_rx_queue *rxqueue;
2970 struct rps_dev_flow_table *flow_table;
2971 struct rps_dev_flow *old_rflow;
2976 /* Should we steer this flow to a different hardware queue? */
2977 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2978 !(dev->features & NETIF_F_NTUPLE))
2980 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2981 if (rxq_index == skb_get_rx_queue(skb))
2984 rxqueue = dev->_rx + rxq_index;
2985 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2988 flow_id = skb_get_hash(skb) & flow_table->mask;
2989 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2990 rxq_index, flow_id);
2994 rflow = &flow_table->flows[flow_id];
2996 if (old_rflow->filter == rflow->filter)
2997 old_rflow->filter = RPS_NO_FILTER;
3001 per_cpu(softnet_data, next_cpu).input_queue_head;
3004 rflow->cpu = next_cpu;
3009 * get_rps_cpu is called from netif_receive_skb and returns the target
3010 * CPU from the RPS map of the receiving queue for a given skb.
3011 * rcu_read_lock must be held on entry.
3013 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3014 struct rps_dev_flow **rflowp)
3016 struct netdev_rx_queue *rxqueue;
3017 struct rps_map *map;
3018 struct rps_dev_flow_table *flow_table;
3019 struct rps_sock_flow_table *sock_flow_table;
3024 if (skb_rx_queue_recorded(skb)) {
3025 u16 index = skb_get_rx_queue(skb);
3026 if (unlikely(index >= dev->real_num_rx_queues)) {
3027 WARN_ONCE(dev->real_num_rx_queues > 1,
3028 "%s received packet on queue %u, but number "
3029 "of RX queues is %u\n",
3030 dev->name, index, dev->real_num_rx_queues);
3033 rxqueue = dev->_rx + index;
3037 map = rcu_dereference(rxqueue->rps_map);
3039 if (map->len == 1 &&
3040 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3041 tcpu = map->cpus[0];
3042 if (cpu_online(tcpu))
3046 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3050 skb_reset_network_header(skb);
3051 hash = skb_get_hash(skb);
3055 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3056 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3057 if (flow_table && sock_flow_table) {
3059 struct rps_dev_flow *rflow;
3061 rflow = &flow_table->flows[hash & flow_table->mask];
3064 next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask];
3067 * If the desired CPU (where last recvmsg was done) is
3068 * different from current CPU (one in the rx-queue flow
3069 * table entry), switch if one of the following holds:
3070 * - Current CPU is unset (equal to RPS_NO_CPU).
3071 * - Current CPU is offline.
3072 * - The current CPU's queue tail has advanced beyond the
3073 * last packet that was enqueued using this table entry.
3074 * This guarantees that all previous packets for the flow
3075 * have been dequeued, thus preserving in order delivery.
3077 if (unlikely(tcpu != next_cpu) &&
3078 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3079 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3080 rflow->last_qtail)) >= 0)) {
3082 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3085 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3093 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3094 if (cpu_online(tcpu)) {
3104 #ifdef CONFIG_RFS_ACCEL
3107 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3108 * @dev: Device on which the filter was set
3109 * @rxq_index: RX queue index
3110 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3111 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3113 * Drivers that implement ndo_rx_flow_steer() should periodically call
3114 * this function for each installed filter and remove the filters for
3115 * which it returns %true.
3117 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3118 u32 flow_id, u16 filter_id)
3120 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3121 struct rps_dev_flow_table *flow_table;
3122 struct rps_dev_flow *rflow;
3127 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3128 if (flow_table && flow_id <= flow_table->mask) {
3129 rflow = &flow_table->flows[flow_id];
3130 cpu = ACCESS_ONCE(rflow->cpu);
3131 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3132 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3133 rflow->last_qtail) <
3134 (int)(10 * flow_table->mask)))
3140 EXPORT_SYMBOL(rps_may_expire_flow);
3142 #endif /* CONFIG_RFS_ACCEL */
3144 /* Called from hardirq (IPI) context */
3145 static void rps_trigger_softirq(void *data)
3147 struct softnet_data *sd = data;
3149 ____napi_schedule(sd, &sd->backlog);
3153 #endif /* CONFIG_RPS */
3156 * Check if this softnet_data structure is another cpu one
3157 * If yes, queue it to our IPI list and return 1
3160 static int rps_ipi_queued(struct softnet_data *sd)
3163 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3166 sd->rps_ipi_next = mysd->rps_ipi_list;
3167 mysd->rps_ipi_list = sd;
3169 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3172 #endif /* CONFIG_RPS */
3176 #ifdef CONFIG_NET_FLOW_LIMIT
3177 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3180 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3182 #ifdef CONFIG_NET_FLOW_LIMIT
3183 struct sd_flow_limit *fl;
3184 struct softnet_data *sd;
3185 unsigned int old_flow, new_flow;
3187 if (qlen < (netdev_max_backlog >> 1))
3190 sd = &__get_cpu_var(softnet_data);
3193 fl = rcu_dereference(sd->flow_limit);
3195 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3196 old_flow = fl->history[fl->history_head];
3197 fl->history[fl->history_head] = new_flow;
3200 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3202 if (likely(fl->buckets[old_flow]))
3203 fl->buckets[old_flow]--;
3205 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3217 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3218 * queue (may be a remote CPU queue).
3220 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3221 unsigned int *qtail)
3223 struct softnet_data *sd;
3224 unsigned long flags;
3227 sd = &per_cpu(softnet_data, cpu);
3229 local_irq_save(flags);
3232 qlen = skb_queue_len(&sd->input_pkt_queue);
3233 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3234 if (skb_queue_len(&sd->input_pkt_queue)) {
3236 __skb_queue_tail(&sd->input_pkt_queue, skb);
3237 input_queue_tail_incr_save(sd, qtail);
3239 local_irq_restore(flags);
3240 return NET_RX_SUCCESS;
3243 /* Schedule NAPI for backlog device
3244 * We can use non atomic operation since we own the queue lock
3246 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3247 if (!rps_ipi_queued(sd))
3248 ____napi_schedule(sd, &sd->backlog);
3256 local_irq_restore(flags);
3258 atomic_long_inc(&skb->dev->rx_dropped);
3263 static int netif_rx_internal(struct sk_buff *skb)
3267 net_timestamp_check(netdev_tstamp_prequeue, skb);
3269 trace_netif_rx(skb);
3271 if (static_key_false(&rps_needed)) {
3272 struct rps_dev_flow voidflow, *rflow = &voidflow;
3278 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3280 cpu = smp_processor_id();
3282 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3290 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3297 * netif_rx - post buffer to the network code
3298 * @skb: buffer to post
3300 * This function receives a packet from a device driver and queues it for
3301 * the upper (protocol) levels to process. It always succeeds. The buffer
3302 * may be dropped during processing for congestion control or by the
3306 * NET_RX_SUCCESS (no congestion)
3307 * NET_RX_DROP (packet was dropped)
3311 int netif_rx(struct sk_buff *skb)
3313 trace_netif_rx_entry(skb);
3315 return netif_rx_internal(skb);
3317 EXPORT_SYMBOL(netif_rx);
3319 int netif_rx_ni(struct sk_buff *skb)
3323 trace_netif_rx_ni_entry(skb);
3326 err = netif_rx_internal(skb);
3327 if (local_softirq_pending())
3333 EXPORT_SYMBOL(netif_rx_ni);
3335 static void net_tx_action(struct softirq_action *h)
3337 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3339 if (sd->completion_queue) {
3340 struct sk_buff *clist;
3342 local_irq_disable();
3343 clist = sd->completion_queue;
3344 sd->completion_queue = NULL;
3348 struct sk_buff *skb = clist;
3349 clist = clist->next;
3351 WARN_ON(atomic_read(&skb->users));
3352 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3353 trace_consume_skb(skb);
3355 trace_kfree_skb(skb, net_tx_action);
3360 if (sd->output_queue) {
3363 local_irq_disable();
3364 head = sd->output_queue;
3365 sd->output_queue = NULL;
3366 sd->output_queue_tailp = &sd->output_queue;
3370 struct Qdisc *q = head;
3371 spinlock_t *root_lock;
3373 head = head->next_sched;
3375 root_lock = qdisc_lock(q);
3376 if (spin_trylock(root_lock)) {
3377 smp_mb__before_atomic();
3378 clear_bit(__QDISC_STATE_SCHED,
3381 spin_unlock(root_lock);
3383 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3385 __netif_reschedule(q);
3387 smp_mb__before_atomic();
3388 clear_bit(__QDISC_STATE_SCHED,
3396 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3397 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3398 /* This hook is defined here for ATM LANE */
3399 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3400 unsigned char *addr) __read_mostly;
3401 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3404 #ifdef CONFIG_NET_CLS_ACT
3405 /* TODO: Maybe we should just force sch_ingress to be compiled in
3406 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3407 * a compare and 2 stores extra right now if we dont have it on
3408 * but have CONFIG_NET_CLS_ACT
3409 * NOTE: This doesn't stop any functionality; if you dont have
3410 * the ingress scheduler, you just can't add policies on ingress.
3413 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3415 struct net_device *dev = skb->dev;
3416 u32 ttl = G_TC_RTTL(skb->tc_verd);
3417 int result = TC_ACT_OK;
3420 if (unlikely(MAX_RED_LOOP < ttl++)) {
3421 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3422 skb->skb_iif, dev->ifindex);
3426 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3427 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3430 if (q != &noop_qdisc) {
3431 spin_lock(qdisc_lock(q));
3432 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3433 result = qdisc_enqueue_root(skb, q);
3434 spin_unlock(qdisc_lock(q));
3440 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3441 struct packet_type **pt_prev,
3442 int *ret, struct net_device *orig_dev)
3444 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3446 if (!rxq || rxq->qdisc == &noop_qdisc)
3450 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3454 switch (ing_filter(skb, rxq)) {
3468 * netdev_rx_handler_register - register receive handler
3469 * @dev: device to register a handler for
3470 * @rx_handler: receive handler to register
3471 * @rx_handler_data: data pointer that is used by rx handler
3473 * Register a receive handler for a device. This handler will then be
3474 * called from __netif_receive_skb. A negative errno code is returned
3477 * The caller must hold the rtnl_mutex.
3479 * For a general description of rx_handler, see enum rx_handler_result.
3481 int netdev_rx_handler_register(struct net_device *dev,
3482 rx_handler_func_t *rx_handler,
3483 void *rx_handler_data)
3487 if (dev->rx_handler)
3490 /* Note: rx_handler_data must be set before rx_handler */
3491 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3492 rcu_assign_pointer(dev->rx_handler, rx_handler);
3496 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3499 * netdev_rx_handler_unregister - unregister receive handler
3500 * @dev: device to unregister a handler from
3502 * Unregister a receive handler from a device.
3504 * The caller must hold the rtnl_mutex.
3506 void netdev_rx_handler_unregister(struct net_device *dev)
3510 RCU_INIT_POINTER(dev->rx_handler, NULL);
3511 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3512 * section has a guarantee to see a non NULL rx_handler_data
3516 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3518 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3521 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3522 * the special handling of PFMEMALLOC skbs.
3524 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3526 switch (skb->protocol) {
3527 case htons(ETH_P_ARP):
3528 case htons(ETH_P_IP):
3529 case htons(ETH_P_IPV6):
3530 case htons(ETH_P_8021Q):
3531 case htons(ETH_P_8021AD):
3538 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3540 struct packet_type *ptype, *pt_prev;
3541 rx_handler_func_t *rx_handler;
3542 struct net_device *orig_dev;
3543 struct net_device *null_or_dev;
3544 bool deliver_exact = false;
3545 int ret = NET_RX_DROP;
3548 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3550 trace_netif_receive_skb(skb);
3552 orig_dev = skb->dev;
3554 skb_reset_network_header(skb);
3555 if (!skb_transport_header_was_set(skb))
3556 skb_reset_transport_header(skb);
3557 skb_reset_mac_len(skb);
3564 skb->skb_iif = skb->dev->ifindex;
3566 __this_cpu_inc(softnet_data.processed);
3568 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3569 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3570 skb = skb_vlan_untag(skb);
3575 #ifdef CONFIG_NET_CLS_ACT
3576 if (skb->tc_verd & TC_NCLS) {
3577 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3585 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3586 if (!ptype->dev || ptype->dev == skb->dev) {
3588 ret = deliver_skb(skb, pt_prev, orig_dev);
3594 #ifdef CONFIG_NET_CLS_ACT
3595 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3601 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3604 if (vlan_tx_tag_present(skb)) {
3606 ret = deliver_skb(skb, pt_prev, orig_dev);
3609 if (vlan_do_receive(&skb))
3611 else if (unlikely(!skb))
3615 rx_handler = rcu_dereference(skb->dev->rx_handler);
3618 ret = deliver_skb(skb, pt_prev, orig_dev);
3621 switch (rx_handler(&skb)) {
3622 case RX_HANDLER_CONSUMED:
3623 ret = NET_RX_SUCCESS;
3625 case RX_HANDLER_ANOTHER:
3627 case RX_HANDLER_EXACT:
3628 deliver_exact = true;
3629 case RX_HANDLER_PASS:
3636 if (unlikely(vlan_tx_tag_present(skb))) {
3637 if (vlan_tx_tag_get_id(skb))
3638 skb->pkt_type = PACKET_OTHERHOST;
3639 /* Note: we might in the future use prio bits
3640 * and set skb->priority like in vlan_do_receive()
3641 * For the time being, just ignore Priority Code Point
3646 /* deliver only exact match when indicated */
3647 null_or_dev = deliver_exact ? skb->dev : NULL;
3649 type = skb->protocol;
3650 list_for_each_entry_rcu(ptype,
3651 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3652 if (ptype->type == type &&
3653 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3654 ptype->dev == orig_dev)) {
3656 ret = deliver_skb(skb, pt_prev, orig_dev);
3662 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3665 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3668 atomic_long_inc(&skb->dev->rx_dropped);
3670 /* Jamal, now you will not able to escape explaining
3671 * me how you were going to use this. :-)
3681 static int __netif_receive_skb(struct sk_buff *skb)
3685 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3686 unsigned long pflags = current->flags;
3689 * PFMEMALLOC skbs are special, they should
3690 * - be delivered to SOCK_MEMALLOC sockets only
3691 * - stay away from userspace
3692 * - have bounded memory usage
3694 * Use PF_MEMALLOC as this saves us from propagating the allocation
3695 * context down to all allocation sites.
3697 current->flags |= PF_MEMALLOC;
3698 ret = __netif_receive_skb_core(skb, true);
3699 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3701 ret = __netif_receive_skb_core(skb, false);
3706 static int netif_receive_skb_internal(struct sk_buff *skb)
3708 net_timestamp_check(netdev_tstamp_prequeue, skb);
3710 if (skb_defer_rx_timestamp(skb))
3711 return NET_RX_SUCCESS;
3714 if (static_key_false(&rps_needed)) {
3715 struct rps_dev_flow voidflow, *rflow = &voidflow;
3720 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3723 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3730 return __netif_receive_skb(skb);
3734 * netif_receive_skb - process receive buffer from network
3735 * @skb: buffer to process
3737 * netif_receive_skb() is the main receive data processing function.
3738 * It always succeeds. The buffer may be dropped during processing
3739 * for congestion control or by the protocol layers.
3741 * This function may only be called from softirq context and interrupts
3742 * should be enabled.
3744 * Return values (usually ignored):
3745 * NET_RX_SUCCESS: no congestion
3746 * NET_RX_DROP: packet was dropped
3748 int netif_receive_skb(struct sk_buff *skb)
3750 trace_netif_receive_skb_entry(skb);
3752 return netif_receive_skb_internal(skb);
3754 EXPORT_SYMBOL(netif_receive_skb);
3756 /* Network device is going away, flush any packets still pending
3757 * Called with irqs disabled.
3759 static void flush_backlog(void *arg)
3761 struct net_device *dev = arg;
3762 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3763 struct sk_buff *skb, *tmp;
3766 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3767 if (skb->dev == dev) {
3768 __skb_unlink(skb, &sd->input_pkt_queue);
3770 input_queue_head_incr(sd);
3775 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3776 if (skb->dev == dev) {
3777 __skb_unlink(skb, &sd->process_queue);
3779 input_queue_head_incr(sd);
3784 static int napi_gro_complete(struct sk_buff *skb)
3786 struct packet_offload *ptype;
3787 __be16 type = skb->protocol;
3788 struct list_head *head = &offload_base;
3791 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3793 if (NAPI_GRO_CB(skb)->count == 1) {
3794 skb_shinfo(skb)->gso_size = 0;
3799 list_for_each_entry_rcu(ptype, head, list) {
3800 if (ptype->type != type || !ptype->callbacks.gro_complete)
3803 err = ptype->callbacks.gro_complete(skb, 0);
3809 WARN_ON(&ptype->list == head);
3811 return NET_RX_SUCCESS;
3815 return netif_receive_skb_internal(skb);
3818 /* napi->gro_list contains packets ordered by age.
3819 * youngest packets at the head of it.
3820 * Complete skbs in reverse order to reduce latencies.
3822 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3824 struct sk_buff *skb, *prev = NULL;
3826 /* scan list and build reverse chain */
3827 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3832 for (skb = prev; skb; skb = prev) {
3835 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3839 napi_gro_complete(skb);
3843 napi->gro_list = NULL;
3845 EXPORT_SYMBOL(napi_gro_flush);
3847 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3850 unsigned int maclen = skb->dev->hard_header_len;
3851 u32 hash = skb_get_hash_raw(skb);
3853 for (p = napi->gro_list; p; p = p->next) {
3854 unsigned long diffs;
3856 NAPI_GRO_CB(p)->flush = 0;
3858 if (hash != skb_get_hash_raw(p)) {
3859 NAPI_GRO_CB(p)->same_flow = 0;
3863 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3864 diffs |= p->vlan_tci ^ skb->vlan_tci;
3865 if (maclen == ETH_HLEN)
3866 diffs |= compare_ether_header(skb_mac_header(p),
3867 skb_mac_header(skb));
3869 diffs = memcmp(skb_mac_header(p),
3870 skb_mac_header(skb),
3872 NAPI_GRO_CB(p)->same_flow = !diffs;
3876 static void skb_gro_reset_offset(struct sk_buff *skb)
3878 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3879 const skb_frag_t *frag0 = &pinfo->frags[0];
3881 NAPI_GRO_CB(skb)->data_offset = 0;
3882 NAPI_GRO_CB(skb)->frag0 = NULL;
3883 NAPI_GRO_CB(skb)->frag0_len = 0;
3885 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3887 !PageHighMem(skb_frag_page(frag0))) {
3888 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3889 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3893 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
3895 struct skb_shared_info *pinfo = skb_shinfo(skb);
3897 BUG_ON(skb->end - skb->tail < grow);
3899 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3901 skb->data_len -= grow;
3904 pinfo->frags[0].page_offset += grow;
3905 skb_frag_size_sub(&pinfo->frags[0], grow);
3907 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
3908 skb_frag_unref(skb, 0);
3909 memmove(pinfo->frags, pinfo->frags + 1,
3910 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
3914 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3916 struct sk_buff **pp = NULL;
3917 struct packet_offload *ptype;
3918 __be16 type = skb->protocol;
3919 struct list_head *head = &offload_base;
3921 enum gro_result ret;
3924 if (!(skb->dev->features & NETIF_F_GRO))
3927 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
3930 gro_list_prepare(napi, skb);
3933 list_for_each_entry_rcu(ptype, head, list) {
3934 if (ptype->type != type || !ptype->callbacks.gro_receive)
3937 skb_set_network_header(skb, skb_gro_offset(skb));
3938 skb_reset_mac_len(skb);
3939 NAPI_GRO_CB(skb)->same_flow = 0;
3940 NAPI_GRO_CB(skb)->flush = 0;
3941 NAPI_GRO_CB(skb)->free = 0;
3942 NAPI_GRO_CB(skb)->udp_mark = 0;
3944 /* Setup for GRO checksum validation */
3945 switch (skb->ip_summed) {
3946 case CHECKSUM_COMPLETE:
3947 NAPI_GRO_CB(skb)->csum = skb->csum;
3948 NAPI_GRO_CB(skb)->csum_valid = 1;
3949 NAPI_GRO_CB(skb)->csum_cnt = 0;
3951 case CHECKSUM_UNNECESSARY:
3952 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
3953 NAPI_GRO_CB(skb)->csum_valid = 0;
3956 NAPI_GRO_CB(skb)->csum_cnt = 0;
3957 NAPI_GRO_CB(skb)->csum_valid = 0;
3960 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3965 if (&ptype->list == head)
3968 same_flow = NAPI_GRO_CB(skb)->same_flow;
3969 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3972 struct sk_buff *nskb = *pp;
3976 napi_gro_complete(nskb);
3983 if (NAPI_GRO_CB(skb)->flush)
3986 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
3987 struct sk_buff *nskb = napi->gro_list;
3989 /* locate the end of the list to select the 'oldest' flow */
3990 while (nskb->next) {
3996 napi_gro_complete(nskb);
4000 NAPI_GRO_CB(skb)->count = 1;
4001 NAPI_GRO_CB(skb)->age = jiffies;
4002 NAPI_GRO_CB(skb)->last = skb;
4003 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4004 skb->next = napi->gro_list;
4005 napi->gro_list = skb;
4009 grow = skb_gro_offset(skb) - skb_headlen(skb);
4011 gro_pull_from_frag0(skb, grow);
4020 struct packet_offload *gro_find_receive_by_type(__be16 type)
4022 struct list_head *offload_head = &offload_base;
4023 struct packet_offload *ptype;
4025 list_for_each_entry_rcu(ptype, offload_head, list) {
4026 if (ptype->type != type || !ptype->callbacks.gro_receive)
4032 EXPORT_SYMBOL(gro_find_receive_by_type);
4034 struct packet_offload *gro_find_complete_by_type(__be16 type)
4036 struct list_head *offload_head = &offload_base;
4037 struct packet_offload *ptype;
4039 list_for_each_entry_rcu(ptype, offload_head, list) {
4040 if (ptype->type != type || !ptype->callbacks.gro_complete)
4046 EXPORT_SYMBOL(gro_find_complete_by_type);
4048 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4052 if (netif_receive_skb_internal(skb))
4060 case GRO_MERGED_FREE:
4061 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4062 kmem_cache_free(skbuff_head_cache, skb);
4075 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4077 trace_napi_gro_receive_entry(skb);
4079 skb_gro_reset_offset(skb);
4081 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4083 EXPORT_SYMBOL(napi_gro_receive);
4085 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4087 __skb_pull(skb, skb_headlen(skb));
4088 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4089 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4091 skb->dev = napi->dev;
4093 skb->encapsulation = 0;
4094 skb_shinfo(skb)->gso_type = 0;
4095 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4100 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4102 struct sk_buff *skb = napi->skb;
4105 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4110 EXPORT_SYMBOL(napi_get_frags);
4112 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4113 struct sk_buff *skb,
4119 __skb_push(skb, ETH_HLEN);
4120 skb->protocol = eth_type_trans(skb, skb->dev);
4121 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4126 case GRO_MERGED_FREE:
4127 napi_reuse_skb(napi, skb);
4137 /* Upper GRO stack assumes network header starts at gro_offset=0
4138 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4139 * We copy ethernet header into skb->data to have a common layout.
4141 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4143 struct sk_buff *skb = napi->skb;
4144 const struct ethhdr *eth;
4145 unsigned int hlen = sizeof(*eth);
4149 skb_reset_mac_header(skb);
4150 skb_gro_reset_offset(skb);
4152 eth = skb_gro_header_fast(skb, 0);
4153 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4154 eth = skb_gro_header_slow(skb, hlen, 0);
4155 if (unlikely(!eth)) {
4156 napi_reuse_skb(napi, skb);
4160 gro_pull_from_frag0(skb, hlen);
4161 NAPI_GRO_CB(skb)->frag0 += hlen;
4162 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4164 __skb_pull(skb, hlen);
4167 * This works because the only protocols we care about don't require
4169 * We'll fix it up properly in napi_frags_finish()
4171 skb->protocol = eth->h_proto;
4176 gro_result_t napi_gro_frags(struct napi_struct *napi)
4178 struct sk_buff *skb = napi_frags_skb(napi);
4183 trace_napi_gro_frags_entry(skb);
4185 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4187 EXPORT_SYMBOL(napi_gro_frags);
4189 /* Compute the checksum from gro_offset and return the folded value
4190 * after adding in any pseudo checksum.
4192 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4197 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4199 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4200 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4202 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4203 !skb->csum_complete_sw)
4204 netdev_rx_csum_fault(skb->dev);
4207 NAPI_GRO_CB(skb)->csum = wsum;
4208 NAPI_GRO_CB(skb)->csum_valid = 1;
4212 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4215 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4216 * Note: called with local irq disabled, but exits with local irq enabled.
4218 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4221 struct softnet_data *remsd = sd->rps_ipi_list;
4224 sd->rps_ipi_list = NULL;
4228 /* Send pending IPI's to kick RPS processing on remote cpus. */
4230 struct softnet_data *next = remsd->rps_ipi_next;
4232 if (cpu_online(remsd->cpu))
4233 smp_call_function_single_async(remsd->cpu,
4242 static int process_backlog(struct napi_struct *napi, int quota)
4245 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4248 /* Check if we have pending ipi, its better to send them now,
4249 * not waiting net_rx_action() end.
4251 if (sd->rps_ipi_list) {
4252 local_irq_disable();
4253 net_rps_action_and_irq_enable(sd);
4256 napi->weight = weight_p;
4257 local_irq_disable();
4259 struct sk_buff *skb;
4261 while ((skb = __skb_dequeue(&sd->process_queue))) {
4263 __netif_receive_skb(skb);
4264 local_irq_disable();
4265 input_queue_head_incr(sd);
4266 if (++work >= quota) {
4273 if (skb_queue_empty(&sd->input_pkt_queue)) {
4275 * Inline a custom version of __napi_complete().
4276 * only current cpu owns and manipulates this napi,
4277 * and NAPI_STATE_SCHED is the only possible flag set
4279 * We can use a plain write instead of clear_bit(),
4280 * and we dont need an smp_mb() memory barrier.
4282 list_del(&napi->poll_list);
4289 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4290 &sd->process_queue);
4299 * __napi_schedule - schedule for receive
4300 * @n: entry to schedule
4302 * The entry's receive function will be scheduled to run
4304 void __napi_schedule(struct napi_struct *n)
4306 unsigned long flags;
4308 local_irq_save(flags);
4309 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4310 local_irq_restore(flags);
4312 EXPORT_SYMBOL(__napi_schedule);
4314 void __napi_complete(struct napi_struct *n)
4316 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4317 BUG_ON(n->gro_list);
4319 list_del(&n->poll_list);
4320 smp_mb__before_atomic();
4321 clear_bit(NAPI_STATE_SCHED, &n->state);
4323 EXPORT_SYMBOL(__napi_complete);
4325 void napi_complete(struct napi_struct *n)
4327 unsigned long flags;
4330 * don't let napi dequeue from the cpu poll list
4331 * just in case its running on a different cpu
4333 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4336 napi_gro_flush(n, false);
4337 local_irq_save(flags);
4339 local_irq_restore(flags);
4341 EXPORT_SYMBOL(napi_complete);
4343 /* must be called under rcu_read_lock(), as we dont take a reference */
4344 struct napi_struct *napi_by_id(unsigned int napi_id)
4346 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4347 struct napi_struct *napi;
4349 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4350 if (napi->napi_id == napi_id)
4355 EXPORT_SYMBOL_GPL(napi_by_id);
4357 void napi_hash_add(struct napi_struct *napi)
4359 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4361 spin_lock(&napi_hash_lock);
4363 /* 0 is not a valid id, we also skip an id that is taken
4364 * we expect both events to be extremely rare
4367 while (!napi->napi_id) {
4368 napi->napi_id = ++napi_gen_id;
4369 if (napi_by_id(napi->napi_id))
4373 hlist_add_head_rcu(&napi->napi_hash_node,
4374 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4376 spin_unlock(&napi_hash_lock);
4379 EXPORT_SYMBOL_GPL(napi_hash_add);
4381 /* Warning : caller is responsible to make sure rcu grace period
4382 * is respected before freeing memory containing @napi
4384 void napi_hash_del(struct napi_struct *napi)
4386 spin_lock(&napi_hash_lock);
4388 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4389 hlist_del_rcu(&napi->napi_hash_node);
4391 spin_unlock(&napi_hash_lock);
4393 EXPORT_SYMBOL_GPL(napi_hash_del);
4395 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4396 int (*poll)(struct napi_struct *, int), int weight)
4398 INIT_LIST_HEAD(&napi->poll_list);
4399 napi->gro_count = 0;
4400 napi->gro_list = NULL;
4403 if (weight > NAPI_POLL_WEIGHT)
4404 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4406 napi->weight = weight;
4407 list_add(&napi->dev_list, &dev->napi_list);
4409 #ifdef CONFIG_NETPOLL
4410 spin_lock_init(&napi->poll_lock);
4411 napi->poll_owner = -1;
4413 set_bit(NAPI_STATE_SCHED, &napi->state);
4415 EXPORT_SYMBOL(netif_napi_add);
4417 void netif_napi_del(struct napi_struct *napi)
4419 list_del_init(&napi->dev_list);
4420 napi_free_frags(napi);
4422 kfree_skb_list(napi->gro_list);
4423 napi->gro_list = NULL;
4424 napi->gro_count = 0;
4426 EXPORT_SYMBOL(netif_napi_del);
4428 static void net_rx_action(struct softirq_action *h)
4430 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4431 unsigned long time_limit = jiffies + 2;
4432 int budget = netdev_budget;
4435 local_irq_disable();
4437 while (!list_empty(&sd->poll_list)) {
4438 struct napi_struct *n;
4441 /* If softirq window is exhuasted then punt.
4442 * Allow this to run for 2 jiffies since which will allow
4443 * an average latency of 1.5/HZ.
4445 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4450 /* Even though interrupts have been re-enabled, this
4451 * access is safe because interrupts can only add new
4452 * entries to the tail of this list, and only ->poll()
4453 * calls can remove this head entry from the list.
4455 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4457 have = netpoll_poll_lock(n);
4461 /* This NAPI_STATE_SCHED test is for avoiding a race
4462 * with netpoll's poll_napi(). Only the entity which
4463 * obtains the lock and sees NAPI_STATE_SCHED set will
4464 * actually make the ->poll() call. Therefore we avoid
4465 * accidentally calling ->poll() when NAPI is not scheduled.
4468 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4469 work = n->poll(n, weight);
4473 WARN_ON_ONCE(work > weight);
4477 local_irq_disable();
4479 /* Drivers must not modify the NAPI state if they
4480 * consume the entire weight. In such cases this code
4481 * still "owns" the NAPI instance and therefore can
4482 * move the instance around on the list at-will.
4484 if (unlikely(work == weight)) {
4485 if (unlikely(napi_disable_pending(n))) {
4488 local_irq_disable();
4491 /* flush too old packets
4492 * If HZ < 1000, flush all packets.
4495 napi_gro_flush(n, HZ >= 1000);
4496 local_irq_disable();
4498 list_move_tail(&n->poll_list, &sd->poll_list);
4502 netpoll_poll_unlock(have);
4505 net_rps_action_and_irq_enable(sd);
4507 #ifdef CONFIG_NET_DMA
4509 * There may not be any more sk_buffs coming right now, so push
4510 * any pending DMA copies to hardware
4512 dma_issue_pending_all();
4519 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4523 struct netdev_adjacent {
4524 struct net_device *dev;
4526 /* upper master flag, there can only be one master device per list */
4529 /* counter for the number of times this device was added to us */
4532 /* private field for the users */
4535 struct list_head list;
4536 struct rcu_head rcu;
4539 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4540 struct net_device *adj_dev,
4541 struct list_head *adj_list)
4543 struct netdev_adjacent *adj;
4545 list_for_each_entry(adj, adj_list, list) {
4546 if (adj->dev == adj_dev)
4553 * netdev_has_upper_dev - Check if device is linked to an upper device
4555 * @upper_dev: upper device to check
4557 * Find out if a device is linked to specified upper device and return true
4558 * in case it is. Note that this checks only immediate upper device,
4559 * not through a complete stack of devices. The caller must hold the RTNL lock.
4561 bool netdev_has_upper_dev(struct net_device *dev,
4562 struct net_device *upper_dev)
4566 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4568 EXPORT_SYMBOL(netdev_has_upper_dev);
4571 * netdev_has_any_upper_dev - Check if device is linked to some device
4574 * Find out if a device is linked to an upper device and return true in case
4575 * it is. The caller must hold the RTNL lock.
4577 static bool netdev_has_any_upper_dev(struct net_device *dev)
4581 return !list_empty(&dev->all_adj_list.upper);
4585 * netdev_master_upper_dev_get - Get master upper device
4588 * Find a master upper device and return pointer to it or NULL in case
4589 * it's not there. The caller must hold the RTNL lock.
4591 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4593 struct netdev_adjacent *upper;
4597 if (list_empty(&dev->adj_list.upper))
4600 upper = list_first_entry(&dev->adj_list.upper,
4601 struct netdev_adjacent, list);
4602 if (likely(upper->master))
4606 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4608 void *netdev_adjacent_get_private(struct list_head *adj_list)
4610 struct netdev_adjacent *adj;
4612 adj = list_entry(adj_list, struct netdev_adjacent, list);
4614 return adj->private;
4616 EXPORT_SYMBOL(netdev_adjacent_get_private);
4619 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4621 * @iter: list_head ** of the current position
4623 * Gets the next device from the dev's upper list, starting from iter
4624 * position. The caller must hold RCU read lock.
4626 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4627 struct list_head **iter)
4629 struct netdev_adjacent *upper;
4631 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4633 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4635 if (&upper->list == &dev->adj_list.upper)
4638 *iter = &upper->list;
4642 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4645 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4647 * @iter: list_head ** of the current position
4649 * Gets the next device from the dev's upper list, starting from iter
4650 * position. The caller must hold RCU read lock.
4652 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4653 struct list_head **iter)
4655 struct netdev_adjacent *upper;
4657 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4659 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4661 if (&upper->list == &dev->all_adj_list.upper)
4664 *iter = &upper->list;
4668 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4671 * netdev_lower_get_next_private - Get the next ->private from the
4672 * lower neighbour list
4674 * @iter: list_head ** of the current position
4676 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4677 * list, starting from iter position. The caller must hold either hold the
4678 * RTNL lock or its own locking that guarantees that the neighbour lower
4679 * list will remain unchainged.
4681 void *netdev_lower_get_next_private(struct net_device *dev,
4682 struct list_head **iter)
4684 struct netdev_adjacent *lower;
4686 lower = list_entry(*iter, struct netdev_adjacent, list);
4688 if (&lower->list == &dev->adj_list.lower)
4691 *iter = lower->list.next;
4693 return lower->private;
4695 EXPORT_SYMBOL(netdev_lower_get_next_private);
4698 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4699 * lower neighbour list, RCU
4702 * @iter: list_head ** of the current position
4704 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4705 * list, starting from iter position. The caller must hold RCU read lock.
4707 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4708 struct list_head **iter)
4710 struct netdev_adjacent *lower;
4712 WARN_ON_ONCE(!rcu_read_lock_held());
4714 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4716 if (&lower->list == &dev->adj_list.lower)
4719 *iter = &lower->list;
4721 return lower->private;
4723 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4726 * netdev_lower_get_next - Get the next device from the lower neighbour
4729 * @iter: list_head ** of the current position
4731 * Gets the next netdev_adjacent from the dev's lower neighbour
4732 * list, starting from iter position. The caller must hold RTNL lock or
4733 * its own locking that guarantees that the neighbour lower
4734 * list will remain unchainged.
4736 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4738 struct netdev_adjacent *lower;
4740 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4742 if (&lower->list == &dev->adj_list.lower)
4745 *iter = &lower->list;
4749 EXPORT_SYMBOL(netdev_lower_get_next);
4752 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4753 * lower neighbour list, RCU
4757 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4758 * list. The caller must hold RCU read lock.
4760 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4762 struct netdev_adjacent *lower;
4764 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4765 struct netdev_adjacent, list);
4767 return lower->private;
4770 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4773 * netdev_master_upper_dev_get_rcu - Get master upper device
4776 * Find a master upper device and return pointer to it or NULL in case
4777 * it's not there. The caller must hold the RCU read lock.
4779 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4781 struct netdev_adjacent *upper;
4783 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4784 struct netdev_adjacent, list);
4785 if (upper && likely(upper->master))
4789 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4791 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4792 struct net_device *adj_dev,
4793 struct list_head *dev_list)
4795 char linkname[IFNAMSIZ+7];
4796 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4797 "upper_%s" : "lower_%s", adj_dev->name);
4798 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4801 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4803 struct list_head *dev_list)
4805 char linkname[IFNAMSIZ+7];
4806 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4807 "upper_%s" : "lower_%s", name);
4808 sysfs_remove_link(&(dev->dev.kobj), linkname);
4811 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4812 (dev_list == &dev->adj_list.upper || \
4813 dev_list == &dev->adj_list.lower)
4815 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4816 struct net_device *adj_dev,
4817 struct list_head *dev_list,
4818 void *private, bool master)
4820 struct netdev_adjacent *adj;
4823 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4830 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4835 adj->master = master;
4837 adj->private = private;
4840 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4841 adj_dev->name, dev->name, adj_dev->name);
4843 if (netdev_adjacent_is_neigh_list(dev, dev_list)) {
4844 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
4849 /* Ensure that master link is always the first item in list. */
4851 ret = sysfs_create_link(&(dev->dev.kobj),
4852 &(adj_dev->dev.kobj), "master");
4854 goto remove_symlinks;
4856 list_add_rcu(&adj->list, dev_list);
4858 list_add_tail_rcu(&adj->list, dev_list);
4864 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4865 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4873 static void __netdev_adjacent_dev_remove(struct net_device *dev,
4874 struct net_device *adj_dev,
4875 struct list_head *dev_list)
4877 struct netdev_adjacent *adj;
4879 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4882 pr_err("tried to remove device %s from %s\n",
4883 dev->name, adj_dev->name);
4887 if (adj->ref_nr > 1) {
4888 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4895 sysfs_remove_link(&(dev->dev.kobj), "master");
4897 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4898 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4900 list_del_rcu(&adj->list);
4901 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4902 adj_dev->name, dev->name, adj_dev->name);
4904 kfree_rcu(adj, rcu);
4907 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4908 struct net_device *upper_dev,
4909 struct list_head *up_list,
4910 struct list_head *down_list,
4911 void *private, bool master)
4915 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4920 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4923 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4930 static int __netdev_adjacent_dev_link(struct net_device *dev,
4931 struct net_device *upper_dev)
4933 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4934 &dev->all_adj_list.upper,
4935 &upper_dev->all_adj_list.lower,
4939 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4940 struct net_device *upper_dev,
4941 struct list_head *up_list,
4942 struct list_head *down_list)
4944 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4945 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4948 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
4949 struct net_device *upper_dev)
4951 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4952 &dev->all_adj_list.upper,
4953 &upper_dev->all_adj_list.lower);
4956 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4957 struct net_device *upper_dev,
4958 void *private, bool master)
4960 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4965 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4966 &dev->adj_list.upper,
4967 &upper_dev->adj_list.lower,
4970 __netdev_adjacent_dev_unlink(dev, upper_dev);
4977 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4978 struct net_device *upper_dev)
4980 __netdev_adjacent_dev_unlink(dev, upper_dev);
4981 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4982 &dev->adj_list.upper,
4983 &upper_dev->adj_list.lower);
4986 static int __netdev_upper_dev_link(struct net_device *dev,
4987 struct net_device *upper_dev, bool master,
4990 struct netdev_adjacent *i, *j, *to_i, *to_j;
4995 if (dev == upper_dev)
4998 /* To prevent loops, check if dev is not upper device to upper_dev. */
4999 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5002 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
5005 if (master && netdev_master_upper_dev_get(dev))
5008 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5013 /* Now that we linked these devs, make all the upper_dev's
5014 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5015 * versa, and don't forget the devices itself. All of these
5016 * links are non-neighbours.
5018 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5019 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5020 pr_debug("Interlinking %s with %s, non-neighbour\n",
5021 i->dev->name, j->dev->name);
5022 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5028 /* add dev to every upper_dev's upper device */
5029 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5030 pr_debug("linking %s's upper device %s with %s\n",
5031 upper_dev->name, i->dev->name, dev->name);
5032 ret = __netdev_adjacent_dev_link(dev, i->dev);
5034 goto rollback_upper_mesh;
5037 /* add upper_dev to every dev's lower device */
5038 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5039 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5040 i->dev->name, upper_dev->name);
5041 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5043 goto rollback_lower_mesh;
5046 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5049 rollback_lower_mesh:
5051 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5054 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5059 rollback_upper_mesh:
5061 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5064 __netdev_adjacent_dev_unlink(dev, i->dev);
5072 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5073 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5074 if (i == to_i && j == to_j)
5076 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5082 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5088 * netdev_upper_dev_link - Add a link to the upper device
5090 * @upper_dev: new upper device
5092 * Adds a link to device which is upper to this one. The caller must hold
5093 * the RTNL lock. On a failure a negative errno code is returned.
5094 * On success the reference counts are adjusted and the function
5097 int netdev_upper_dev_link(struct net_device *dev,
5098 struct net_device *upper_dev)
5100 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5102 EXPORT_SYMBOL(netdev_upper_dev_link);
5105 * netdev_master_upper_dev_link - Add a master link to the upper device
5107 * @upper_dev: new upper device
5109 * Adds a link to device which is upper to this one. In this case, only
5110 * one master upper device can be linked, although other non-master devices
5111 * might be linked as well. The caller must hold the RTNL lock.
5112 * On a failure a negative errno code is returned. On success the reference
5113 * counts are adjusted and the function returns zero.
5115 int netdev_master_upper_dev_link(struct net_device *dev,
5116 struct net_device *upper_dev)
5118 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5120 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5122 int netdev_master_upper_dev_link_private(struct net_device *dev,
5123 struct net_device *upper_dev,
5126 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5128 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5131 * netdev_upper_dev_unlink - Removes a link to upper device
5133 * @upper_dev: new upper device
5135 * Removes a link to device which is upper to this one. The caller must hold
5138 void netdev_upper_dev_unlink(struct net_device *dev,
5139 struct net_device *upper_dev)
5141 struct netdev_adjacent *i, *j;
5144 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5146 /* Here is the tricky part. We must remove all dev's lower
5147 * devices from all upper_dev's upper devices and vice
5148 * versa, to maintain the graph relationship.
5150 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5151 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5152 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5154 /* remove also the devices itself from lower/upper device
5157 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5158 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5160 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5161 __netdev_adjacent_dev_unlink(dev, i->dev);
5163 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5165 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5167 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5169 struct netdev_adjacent *iter;
5171 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5172 netdev_adjacent_sysfs_del(iter->dev, oldname,
5173 &iter->dev->adj_list.lower);
5174 netdev_adjacent_sysfs_add(iter->dev, dev,
5175 &iter->dev->adj_list.lower);
5178 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5179 netdev_adjacent_sysfs_del(iter->dev, oldname,
5180 &iter->dev->adj_list.upper);
5181 netdev_adjacent_sysfs_add(iter->dev, dev,
5182 &iter->dev->adj_list.upper);
5186 void *netdev_lower_dev_get_private(struct net_device *dev,
5187 struct net_device *lower_dev)
5189 struct netdev_adjacent *lower;
5193 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5197 return lower->private;
5199 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5202 int dev_get_nest_level(struct net_device *dev,
5203 bool (*type_check)(struct net_device *dev))
5205 struct net_device *lower = NULL;
5206 struct list_head *iter;
5212 netdev_for_each_lower_dev(dev, lower, iter) {
5213 nest = dev_get_nest_level(lower, type_check);
5214 if (max_nest < nest)
5218 if (type_check(dev))
5223 EXPORT_SYMBOL(dev_get_nest_level);
5225 static void dev_change_rx_flags(struct net_device *dev, int flags)
5227 const struct net_device_ops *ops = dev->netdev_ops;
5229 if (ops->ndo_change_rx_flags)
5230 ops->ndo_change_rx_flags(dev, flags);
5233 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5235 unsigned int old_flags = dev->flags;
5241 dev->flags |= IFF_PROMISC;
5242 dev->promiscuity += inc;
5243 if (dev->promiscuity == 0) {
5246 * If inc causes overflow, untouch promisc and return error.
5249 dev->flags &= ~IFF_PROMISC;
5251 dev->promiscuity -= inc;
5252 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5257 if (dev->flags != old_flags) {
5258 pr_info("device %s %s promiscuous mode\n",
5260 dev->flags & IFF_PROMISC ? "entered" : "left");
5261 if (audit_enabled) {
5262 current_uid_gid(&uid, &gid);
5263 audit_log(current->audit_context, GFP_ATOMIC,
5264 AUDIT_ANOM_PROMISCUOUS,
5265 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5266 dev->name, (dev->flags & IFF_PROMISC),
5267 (old_flags & IFF_PROMISC),
5268 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5269 from_kuid(&init_user_ns, uid),
5270 from_kgid(&init_user_ns, gid),
5271 audit_get_sessionid(current));
5274 dev_change_rx_flags(dev, IFF_PROMISC);
5277 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5282 * dev_set_promiscuity - update promiscuity count on a device
5286 * Add or remove promiscuity from a device. While the count in the device
5287 * remains above zero the interface remains promiscuous. Once it hits zero
5288 * the device reverts back to normal filtering operation. A negative inc
5289 * value is used to drop promiscuity on the device.
5290 * Return 0 if successful or a negative errno code on error.
5292 int dev_set_promiscuity(struct net_device *dev, int inc)
5294 unsigned int old_flags = dev->flags;
5297 err = __dev_set_promiscuity(dev, inc, true);
5300 if (dev->flags != old_flags)
5301 dev_set_rx_mode(dev);
5304 EXPORT_SYMBOL(dev_set_promiscuity);
5306 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5308 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5312 dev->flags |= IFF_ALLMULTI;
5313 dev->allmulti += inc;
5314 if (dev->allmulti == 0) {
5317 * If inc causes overflow, untouch allmulti and return error.
5320 dev->flags &= ~IFF_ALLMULTI;
5322 dev->allmulti -= inc;
5323 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5328 if (dev->flags ^ old_flags) {
5329 dev_change_rx_flags(dev, IFF_ALLMULTI);
5330 dev_set_rx_mode(dev);
5332 __dev_notify_flags(dev, old_flags,
5333 dev->gflags ^ old_gflags);
5339 * dev_set_allmulti - update allmulti count on a device
5343 * Add or remove reception of all multicast frames to a device. While the
5344 * count in the device remains above zero the interface remains listening
5345 * to all interfaces. Once it hits zero the device reverts back to normal
5346 * filtering operation. A negative @inc value is used to drop the counter
5347 * when releasing a resource needing all multicasts.
5348 * Return 0 if successful or a negative errno code on error.
5351 int dev_set_allmulti(struct net_device *dev, int inc)
5353 return __dev_set_allmulti(dev, inc, true);
5355 EXPORT_SYMBOL(dev_set_allmulti);
5358 * Upload unicast and multicast address lists to device and
5359 * configure RX filtering. When the device doesn't support unicast
5360 * filtering it is put in promiscuous mode while unicast addresses
5363 void __dev_set_rx_mode(struct net_device *dev)
5365 const struct net_device_ops *ops = dev->netdev_ops;
5367 /* dev_open will call this function so the list will stay sane. */
5368 if (!(dev->flags&IFF_UP))
5371 if (!netif_device_present(dev))
5374 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5375 /* Unicast addresses changes may only happen under the rtnl,
5376 * therefore calling __dev_set_promiscuity here is safe.
5378 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5379 __dev_set_promiscuity(dev, 1, false);
5380 dev->uc_promisc = true;
5381 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5382 __dev_set_promiscuity(dev, -1, false);
5383 dev->uc_promisc = false;
5387 if (ops->ndo_set_rx_mode)
5388 ops->ndo_set_rx_mode(dev);
5391 void dev_set_rx_mode(struct net_device *dev)
5393 netif_addr_lock_bh(dev);
5394 __dev_set_rx_mode(dev);
5395 netif_addr_unlock_bh(dev);
5399 * dev_get_flags - get flags reported to userspace
5402 * Get the combination of flag bits exported through APIs to userspace.
5404 unsigned int dev_get_flags(const struct net_device *dev)
5408 flags = (dev->flags & ~(IFF_PROMISC |
5413 (dev->gflags & (IFF_PROMISC |
5416 if (netif_running(dev)) {
5417 if (netif_oper_up(dev))
5418 flags |= IFF_RUNNING;
5419 if (netif_carrier_ok(dev))
5420 flags |= IFF_LOWER_UP;
5421 if (netif_dormant(dev))
5422 flags |= IFF_DORMANT;
5427 EXPORT_SYMBOL(dev_get_flags);
5429 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5431 unsigned int old_flags = dev->flags;
5437 * Set the flags on our device.
5440 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5441 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5443 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5447 * Load in the correct multicast list now the flags have changed.
5450 if ((old_flags ^ flags) & IFF_MULTICAST)
5451 dev_change_rx_flags(dev, IFF_MULTICAST);
5453 dev_set_rx_mode(dev);
5456 * Have we downed the interface. We handle IFF_UP ourselves
5457 * according to user attempts to set it, rather than blindly
5462 if ((old_flags ^ flags) & IFF_UP)
5463 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5465 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5466 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5467 unsigned int old_flags = dev->flags;
5469 dev->gflags ^= IFF_PROMISC;
5471 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5472 if (dev->flags != old_flags)
5473 dev_set_rx_mode(dev);
5476 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5477 is important. Some (broken) drivers set IFF_PROMISC, when
5478 IFF_ALLMULTI is requested not asking us and not reporting.
5480 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5481 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5483 dev->gflags ^= IFF_ALLMULTI;
5484 __dev_set_allmulti(dev, inc, false);
5490 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5491 unsigned int gchanges)
5493 unsigned int changes = dev->flags ^ old_flags;
5496 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5498 if (changes & IFF_UP) {
5499 if (dev->flags & IFF_UP)
5500 call_netdevice_notifiers(NETDEV_UP, dev);
5502 call_netdevice_notifiers(NETDEV_DOWN, dev);
5505 if (dev->flags & IFF_UP &&
5506 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5507 struct netdev_notifier_change_info change_info;
5509 change_info.flags_changed = changes;
5510 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5516 * dev_change_flags - change device settings
5518 * @flags: device state flags
5520 * Change settings on device based state flags. The flags are
5521 * in the userspace exported format.
5523 int dev_change_flags(struct net_device *dev, unsigned int flags)
5526 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5528 ret = __dev_change_flags(dev, flags);
5532 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5533 __dev_notify_flags(dev, old_flags, changes);
5536 EXPORT_SYMBOL(dev_change_flags);
5538 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5540 const struct net_device_ops *ops = dev->netdev_ops;
5542 if (ops->ndo_change_mtu)
5543 return ops->ndo_change_mtu(dev, new_mtu);
5550 * dev_set_mtu - Change maximum transfer unit
5552 * @new_mtu: new transfer unit
5554 * Change the maximum transfer size of the network device.
5556 int dev_set_mtu(struct net_device *dev, int new_mtu)
5560 if (new_mtu == dev->mtu)
5563 /* MTU must be positive. */
5567 if (!netif_device_present(dev))
5570 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5571 err = notifier_to_errno(err);
5575 orig_mtu = dev->mtu;
5576 err = __dev_set_mtu(dev, new_mtu);
5579 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5580 err = notifier_to_errno(err);
5582 /* setting mtu back and notifying everyone again,
5583 * so that they have a chance to revert changes.
5585 __dev_set_mtu(dev, orig_mtu);
5586 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5591 EXPORT_SYMBOL(dev_set_mtu);
5594 * dev_set_group - Change group this device belongs to
5596 * @new_group: group this device should belong to
5598 void dev_set_group(struct net_device *dev, int new_group)
5600 dev->group = new_group;
5602 EXPORT_SYMBOL(dev_set_group);
5605 * dev_set_mac_address - Change Media Access Control Address
5609 * Change the hardware (MAC) address of the device
5611 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5613 const struct net_device_ops *ops = dev->netdev_ops;
5616 if (!ops->ndo_set_mac_address)
5618 if (sa->sa_family != dev->type)
5620 if (!netif_device_present(dev))
5622 err = ops->ndo_set_mac_address(dev, sa);
5625 dev->addr_assign_type = NET_ADDR_SET;
5626 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5627 add_device_randomness(dev->dev_addr, dev->addr_len);
5630 EXPORT_SYMBOL(dev_set_mac_address);
5633 * dev_change_carrier - Change device carrier
5635 * @new_carrier: new value
5637 * Change device carrier
5639 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5641 const struct net_device_ops *ops = dev->netdev_ops;
5643 if (!ops->ndo_change_carrier)
5645 if (!netif_device_present(dev))
5647 return ops->ndo_change_carrier(dev, new_carrier);
5649 EXPORT_SYMBOL(dev_change_carrier);
5652 * dev_get_phys_port_id - Get device physical port ID
5656 * Get device physical port ID
5658 int dev_get_phys_port_id(struct net_device *dev,
5659 struct netdev_phys_port_id *ppid)
5661 const struct net_device_ops *ops = dev->netdev_ops;
5663 if (!ops->ndo_get_phys_port_id)
5665 return ops->ndo_get_phys_port_id(dev, ppid);
5667 EXPORT_SYMBOL(dev_get_phys_port_id);
5670 * dev_new_index - allocate an ifindex
5671 * @net: the applicable net namespace
5673 * Returns a suitable unique value for a new device interface
5674 * number. The caller must hold the rtnl semaphore or the
5675 * dev_base_lock to be sure it remains unique.
5677 static int dev_new_index(struct net *net)
5679 int ifindex = net->ifindex;
5683 if (!__dev_get_by_index(net, ifindex))
5684 return net->ifindex = ifindex;
5688 /* Delayed registration/unregisteration */
5689 static LIST_HEAD(net_todo_list);
5690 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5692 static void net_set_todo(struct net_device *dev)
5694 list_add_tail(&dev->todo_list, &net_todo_list);
5695 dev_net(dev)->dev_unreg_count++;
5698 static void rollback_registered_many(struct list_head *head)
5700 struct net_device *dev, *tmp;
5701 LIST_HEAD(close_head);
5703 BUG_ON(dev_boot_phase);
5706 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5707 /* Some devices call without registering
5708 * for initialization unwind. Remove those
5709 * devices and proceed with the remaining.
5711 if (dev->reg_state == NETREG_UNINITIALIZED) {
5712 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5716 list_del(&dev->unreg_list);
5719 dev->dismantle = true;
5720 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5723 /* If device is running, close it first. */
5724 list_for_each_entry(dev, head, unreg_list)
5725 list_add_tail(&dev->close_list, &close_head);
5726 dev_close_many(&close_head);
5728 list_for_each_entry(dev, head, unreg_list) {
5729 /* And unlink it from device chain. */
5730 unlist_netdevice(dev);
5732 dev->reg_state = NETREG_UNREGISTERING;
5737 list_for_each_entry(dev, head, unreg_list) {
5738 /* Shutdown queueing discipline. */
5742 /* Notify protocols, that we are about to destroy
5743 this device. They should clean all the things.
5745 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5748 * Flush the unicast and multicast chains
5753 if (dev->netdev_ops->ndo_uninit)
5754 dev->netdev_ops->ndo_uninit(dev);
5756 if (!dev->rtnl_link_ops ||
5757 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5758 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5760 /* Notifier chain MUST detach us all upper devices. */
5761 WARN_ON(netdev_has_any_upper_dev(dev));
5763 /* Remove entries from kobject tree */
5764 netdev_unregister_kobject(dev);
5766 /* Remove XPS queueing entries */
5767 netif_reset_xps_queues_gt(dev, 0);
5773 list_for_each_entry(dev, head, unreg_list)
5777 static void rollback_registered(struct net_device *dev)
5781 list_add(&dev->unreg_list, &single);
5782 rollback_registered_many(&single);
5786 static netdev_features_t netdev_fix_features(struct net_device *dev,
5787 netdev_features_t features)
5789 /* Fix illegal checksum combinations */
5790 if ((features & NETIF_F_HW_CSUM) &&
5791 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5792 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5793 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5796 /* TSO requires that SG is present as well. */
5797 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5798 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5799 features &= ~NETIF_F_ALL_TSO;
5802 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5803 !(features & NETIF_F_IP_CSUM)) {
5804 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5805 features &= ~NETIF_F_TSO;
5806 features &= ~NETIF_F_TSO_ECN;
5809 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5810 !(features & NETIF_F_IPV6_CSUM)) {
5811 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5812 features &= ~NETIF_F_TSO6;
5815 /* TSO ECN requires that TSO is present as well. */
5816 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5817 features &= ~NETIF_F_TSO_ECN;
5819 /* Software GSO depends on SG. */
5820 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5821 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5822 features &= ~NETIF_F_GSO;
5825 /* UFO needs SG and checksumming */
5826 if (features & NETIF_F_UFO) {
5827 /* maybe split UFO into V4 and V6? */
5828 if (!((features & NETIF_F_GEN_CSUM) ||
5829 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5830 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5832 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5833 features &= ~NETIF_F_UFO;
5836 if (!(features & NETIF_F_SG)) {
5838 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5839 features &= ~NETIF_F_UFO;
5843 #ifdef CONFIG_NET_RX_BUSY_POLL
5844 if (dev->netdev_ops->ndo_busy_poll)
5845 features |= NETIF_F_BUSY_POLL;
5848 features &= ~NETIF_F_BUSY_POLL;
5853 int __netdev_update_features(struct net_device *dev)
5855 netdev_features_t features;
5860 features = netdev_get_wanted_features(dev);
5862 if (dev->netdev_ops->ndo_fix_features)
5863 features = dev->netdev_ops->ndo_fix_features(dev, features);
5865 /* driver might be less strict about feature dependencies */
5866 features = netdev_fix_features(dev, features);
5868 if (dev->features == features)
5871 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5872 &dev->features, &features);
5874 if (dev->netdev_ops->ndo_set_features)
5875 err = dev->netdev_ops->ndo_set_features(dev, features);
5877 if (unlikely(err < 0)) {
5879 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5880 err, &features, &dev->features);
5885 dev->features = features;
5891 * netdev_update_features - recalculate device features
5892 * @dev: the device to check
5894 * Recalculate dev->features set and send notifications if it
5895 * has changed. Should be called after driver or hardware dependent
5896 * conditions might have changed that influence the features.
5898 void netdev_update_features(struct net_device *dev)
5900 if (__netdev_update_features(dev))
5901 netdev_features_change(dev);
5903 EXPORT_SYMBOL(netdev_update_features);
5906 * netdev_change_features - recalculate device features
5907 * @dev: the device to check
5909 * Recalculate dev->features set and send notifications even
5910 * if they have not changed. Should be called instead of
5911 * netdev_update_features() if also dev->vlan_features might
5912 * have changed to allow the changes to be propagated to stacked
5915 void netdev_change_features(struct net_device *dev)
5917 __netdev_update_features(dev);
5918 netdev_features_change(dev);
5920 EXPORT_SYMBOL(netdev_change_features);
5923 * netif_stacked_transfer_operstate - transfer operstate
5924 * @rootdev: the root or lower level device to transfer state from
5925 * @dev: the device to transfer operstate to
5927 * Transfer operational state from root to device. This is normally
5928 * called when a stacking relationship exists between the root
5929 * device and the device(a leaf device).
5931 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5932 struct net_device *dev)
5934 if (rootdev->operstate == IF_OPER_DORMANT)
5935 netif_dormant_on(dev);
5937 netif_dormant_off(dev);
5939 if (netif_carrier_ok(rootdev)) {
5940 if (!netif_carrier_ok(dev))
5941 netif_carrier_on(dev);
5943 if (netif_carrier_ok(dev))
5944 netif_carrier_off(dev);
5947 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5950 static int netif_alloc_rx_queues(struct net_device *dev)
5952 unsigned int i, count = dev->num_rx_queues;
5953 struct netdev_rx_queue *rx;
5957 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5963 for (i = 0; i < count; i++)
5969 static void netdev_init_one_queue(struct net_device *dev,
5970 struct netdev_queue *queue, void *_unused)
5972 /* Initialize queue lock */
5973 spin_lock_init(&queue->_xmit_lock);
5974 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5975 queue->xmit_lock_owner = -1;
5976 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5979 dql_init(&queue->dql, HZ);
5983 static void netif_free_tx_queues(struct net_device *dev)
5988 static int netif_alloc_netdev_queues(struct net_device *dev)
5990 unsigned int count = dev->num_tx_queues;
5991 struct netdev_queue *tx;
5992 size_t sz = count * sizeof(*tx);
5994 BUG_ON(count < 1 || count > 0xffff);
5996 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6004 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6005 spin_lock_init(&dev->tx_global_lock);
6011 * register_netdevice - register a network device
6012 * @dev: device to register
6014 * Take a completed network device structure and add it to the kernel
6015 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6016 * chain. 0 is returned on success. A negative errno code is returned
6017 * on a failure to set up the device, or if the name is a duplicate.
6019 * Callers must hold the rtnl semaphore. You may want
6020 * register_netdev() instead of this.
6023 * The locking appears insufficient to guarantee two parallel registers
6024 * will not get the same name.
6027 int register_netdevice(struct net_device *dev)
6030 struct net *net = dev_net(dev);
6032 BUG_ON(dev_boot_phase);
6037 /* When net_device's are persistent, this will be fatal. */
6038 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6041 spin_lock_init(&dev->addr_list_lock);
6042 netdev_set_addr_lockdep_class(dev);
6046 ret = dev_get_valid_name(net, dev, dev->name);
6050 /* Init, if this function is available */
6051 if (dev->netdev_ops->ndo_init) {
6052 ret = dev->netdev_ops->ndo_init(dev);
6060 if (((dev->hw_features | dev->features) &
6061 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6062 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6063 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6064 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6071 dev->ifindex = dev_new_index(net);
6072 else if (__dev_get_by_index(net, dev->ifindex))
6075 if (dev->iflink == -1)
6076 dev->iflink = dev->ifindex;
6078 /* Transfer changeable features to wanted_features and enable
6079 * software offloads (GSO and GRO).
6081 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6082 dev->features |= NETIF_F_SOFT_FEATURES;
6083 dev->wanted_features = dev->features & dev->hw_features;
6085 if (!(dev->flags & IFF_LOOPBACK)) {
6086 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6089 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6091 dev->vlan_features |= NETIF_F_HIGHDMA;
6093 /* Make NETIF_F_SG inheritable to tunnel devices.
6095 dev->hw_enc_features |= NETIF_F_SG;
6097 /* Make NETIF_F_SG inheritable to MPLS.
6099 dev->mpls_features |= NETIF_F_SG;
6101 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6102 ret = notifier_to_errno(ret);
6106 ret = netdev_register_kobject(dev);
6109 dev->reg_state = NETREG_REGISTERED;
6111 __netdev_update_features(dev);
6114 * Default initial state at registry is that the
6115 * device is present.
6118 set_bit(__LINK_STATE_PRESENT, &dev->state);
6120 linkwatch_init_dev(dev);
6122 dev_init_scheduler(dev);
6124 list_netdevice(dev);
6125 add_device_randomness(dev->dev_addr, dev->addr_len);
6127 /* If the device has permanent device address, driver should
6128 * set dev_addr and also addr_assign_type should be set to
6129 * NET_ADDR_PERM (default value).
6131 if (dev->addr_assign_type == NET_ADDR_PERM)
6132 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6134 /* Notify protocols, that a new device appeared. */
6135 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6136 ret = notifier_to_errno(ret);
6138 rollback_registered(dev);
6139 dev->reg_state = NETREG_UNREGISTERED;
6142 * Prevent userspace races by waiting until the network
6143 * device is fully setup before sending notifications.
6145 if (!dev->rtnl_link_ops ||
6146 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6147 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6153 if (dev->netdev_ops->ndo_uninit)
6154 dev->netdev_ops->ndo_uninit(dev);
6157 EXPORT_SYMBOL(register_netdevice);
6160 * init_dummy_netdev - init a dummy network device for NAPI
6161 * @dev: device to init
6163 * This takes a network device structure and initialize the minimum
6164 * amount of fields so it can be used to schedule NAPI polls without
6165 * registering a full blown interface. This is to be used by drivers
6166 * that need to tie several hardware interfaces to a single NAPI
6167 * poll scheduler due to HW limitations.
6169 int init_dummy_netdev(struct net_device *dev)
6171 /* Clear everything. Note we don't initialize spinlocks
6172 * are they aren't supposed to be taken by any of the
6173 * NAPI code and this dummy netdev is supposed to be
6174 * only ever used for NAPI polls
6176 memset(dev, 0, sizeof(struct net_device));
6178 /* make sure we BUG if trying to hit standard
6179 * register/unregister code path
6181 dev->reg_state = NETREG_DUMMY;
6183 /* NAPI wants this */
6184 INIT_LIST_HEAD(&dev->napi_list);
6186 /* a dummy interface is started by default */
6187 set_bit(__LINK_STATE_PRESENT, &dev->state);
6188 set_bit(__LINK_STATE_START, &dev->state);
6190 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6191 * because users of this 'device' dont need to change
6197 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6201 * register_netdev - register a network device
6202 * @dev: device to register
6204 * Take a completed network device structure and add it to the kernel
6205 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6206 * chain. 0 is returned on success. A negative errno code is returned
6207 * on a failure to set up the device, or if the name is a duplicate.
6209 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6210 * and expands the device name if you passed a format string to
6213 int register_netdev(struct net_device *dev)
6218 err = register_netdevice(dev);
6222 EXPORT_SYMBOL(register_netdev);
6224 int netdev_refcnt_read(const struct net_device *dev)
6228 for_each_possible_cpu(i)
6229 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6232 EXPORT_SYMBOL(netdev_refcnt_read);
6235 * netdev_wait_allrefs - wait until all references are gone.
6236 * @dev: target net_device
6238 * This is called when unregistering network devices.
6240 * Any protocol or device that holds a reference should register
6241 * for netdevice notification, and cleanup and put back the
6242 * reference if they receive an UNREGISTER event.
6243 * We can get stuck here if buggy protocols don't correctly
6246 static void netdev_wait_allrefs(struct net_device *dev)
6248 unsigned long rebroadcast_time, warning_time;
6251 linkwatch_forget_dev(dev);
6253 rebroadcast_time = warning_time = jiffies;
6254 refcnt = netdev_refcnt_read(dev);
6256 while (refcnt != 0) {
6257 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6260 /* Rebroadcast unregister notification */
6261 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6267 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6268 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6270 /* We must not have linkwatch events
6271 * pending on unregister. If this
6272 * happens, we simply run the queue
6273 * unscheduled, resulting in a noop
6276 linkwatch_run_queue();
6281 rebroadcast_time = jiffies;
6286 refcnt = netdev_refcnt_read(dev);
6288 if (time_after(jiffies, warning_time + 10 * HZ)) {
6289 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6291 warning_time = jiffies;
6300 * register_netdevice(x1);
6301 * register_netdevice(x2);
6303 * unregister_netdevice(y1);
6304 * unregister_netdevice(y2);
6310 * We are invoked by rtnl_unlock().
6311 * This allows us to deal with problems:
6312 * 1) We can delete sysfs objects which invoke hotplug
6313 * without deadlocking with linkwatch via keventd.
6314 * 2) Since we run with the RTNL semaphore not held, we can sleep
6315 * safely in order to wait for the netdev refcnt to drop to zero.
6317 * We must not return until all unregister events added during
6318 * the interval the lock was held have been completed.
6320 void netdev_run_todo(void)
6322 struct list_head list;
6324 /* Snapshot list, allow later requests */
6325 list_replace_init(&net_todo_list, &list);
6330 /* Wait for rcu callbacks to finish before next phase */
6331 if (!list_empty(&list))
6334 while (!list_empty(&list)) {
6335 struct net_device *dev
6336 = list_first_entry(&list, struct net_device, todo_list);
6337 list_del(&dev->todo_list);
6340 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6343 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6344 pr_err("network todo '%s' but state %d\n",
6345 dev->name, dev->reg_state);
6350 dev->reg_state = NETREG_UNREGISTERED;
6352 on_each_cpu(flush_backlog, dev, 1);
6354 netdev_wait_allrefs(dev);
6357 BUG_ON(netdev_refcnt_read(dev));
6358 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6359 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6360 WARN_ON(dev->dn_ptr);
6362 if (dev->destructor)
6363 dev->destructor(dev);
6365 /* Report a network device has been unregistered */
6367 dev_net(dev)->dev_unreg_count--;
6369 wake_up(&netdev_unregistering_wq);
6371 /* Free network device */
6372 kobject_put(&dev->dev.kobj);
6376 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6377 * fields in the same order, with only the type differing.
6379 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6380 const struct net_device_stats *netdev_stats)
6382 #if BITS_PER_LONG == 64
6383 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6384 memcpy(stats64, netdev_stats, sizeof(*stats64));
6386 size_t i, n = sizeof(*stats64) / sizeof(u64);
6387 const unsigned long *src = (const unsigned long *)netdev_stats;
6388 u64 *dst = (u64 *)stats64;
6390 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6391 sizeof(*stats64) / sizeof(u64));
6392 for (i = 0; i < n; i++)
6396 EXPORT_SYMBOL(netdev_stats_to_stats64);
6399 * dev_get_stats - get network device statistics
6400 * @dev: device to get statistics from
6401 * @storage: place to store stats
6403 * Get network statistics from device. Return @storage.
6404 * The device driver may provide its own method by setting
6405 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6406 * otherwise the internal statistics structure is used.
6408 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6409 struct rtnl_link_stats64 *storage)
6411 const struct net_device_ops *ops = dev->netdev_ops;
6413 if (ops->ndo_get_stats64) {
6414 memset(storage, 0, sizeof(*storage));
6415 ops->ndo_get_stats64(dev, storage);
6416 } else if (ops->ndo_get_stats) {
6417 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6419 netdev_stats_to_stats64(storage, &dev->stats);
6421 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6422 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6425 EXPORT_SYMBOL(dev_get_stats);
6427 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6429 struct netdev_queue *queue = dev_ingress_queue(dev);
6431 #ifdef CONFIG_NET_CLS_ACT
6434 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6437 netdev_init_one_queue(dev, queue, NULL);
6438 queue->qdisc = &noop_qdisc;
6439 queue->qdisc_sleeping = &noop_qdisc;
6440 rcu_assign_pointer(dev->ingress_queue, queue);
6445 static const struct ethtool_ops default_ethtool_ops;
6447 void netdev_set_default_ethtool_ops(struct net_device *dev,
6448 const struct ethtool_ops *ops)
6450 if (dev->ethtool_ops == &default_ethtool_ops)
6451 dev->ethtool_ops = ops;
6453 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6455 void netdev_freemem(struct net_device *dev)
6457 char *addr = (char *)dev - dev->padded;
6463 * alloc_netdev_mqs - allocate network device
6464 * @sizeof_priv: size of private data to allocate space for
6465 * @name: device name format string
6466 * @name_assign_type: origin of device name
6467 * @setup: callback to initialize device
6468 * @txqs: the number of TX subqueues to allocate
6469 * @rxqs: the number of RX subqueues to allocate
6471 * Allocates a struct net_device with private data area for driver use
6472 * and performs basic initialization. Also allocates subqueue structs
6473 * for each queue on the device.
6475 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6476 unsigned char name_assign_type,
6477 void (*setup)(struct net_device *),
6478 unsigned int txqs, unsigned int rxqs)
6480 struct net_device *dev;
6482 struct net_device *p;
6484 BUG_ON(strlen(name) >= sizeof(dev->name));
6487 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6493 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6498 alloc_size = sizeof(struct net_device);
6500 /* ensure 32-byte alignment of private area */
6501 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6502 alloc_size += sizeof_priv;
6504 /* ensure 32-byte alignment of whole construct */
6505 alloc_size += NETDEV_ALIGN - 1;
6507 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6509 p = vzalloc(alloc_size);
6513 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6514 dev->padded = (char *)dev - (char *)p;
6516 dev->pcpu_refcnt = alloc_percpu(int);
6517 if (!dev->pcpu_refcnt)
6520 if (dev_addr_init(dev))
6526 dev_net_set(dev, &init_net);
6528 dev->gso_max_size = GSO_MAX_SIZE;
6529 dev->gso_max_segs = GSO_MAX_SEGS;
6531 INIT_LIST_HEAD(&dev->napi_list);
6532 INIT_LIST_HEAD(&dev->unreg_list);
6533 INIT_LIST_HEAD(&dev->close_list);
6534 INIT_LIST_HEAD(&dev->link_watch_list);
6535 INIT_LIST_HEAD(&dev->adj_list.upper);
6536 INIT_LIST_HEAD(&dev->adj_list.lower);
6537 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6538 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6539 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6542 dev->num_tx_queues = txqs;
6543 dev->real_num_tx_queues = txqs;
6544 if (netif_alloc_netdev_queues(dev))
6548 dev->num_rx_queues = rxqs;
6549 dev->real_num_rx_queues = rxqs;
6550 if (netif_alloc_rx_queues(dev))
6554 strcpy(dev->name, name);
6555 dev->name_assign_type = name_assign_type;
6556 dev->group = INIT_NETDEV_GROUP;
6557 if (!dev->ethtool_ops)
6558 dev->ethtool_ops = &default_ethtool_ops;
6566 free_percpu(dev->pcpu_refcnt);
6568 netdev_freemem(dev);
6571 EXPORT_SYMBOL(alloc_netdev_mqs);
6574 * free_netdev - free network device
6577 * This function does the last stage of destroying an allocated device
6578 * interface. The reference to the device object is released.
6579 * If this is the last reference then it will be freed.
6581 void free_netdev(struct net_device *dev)
6583 struct napi_struct *p, *n;
6585 release_net(dev_net(dev));
6587 netif_free_tx_queues(dev);
6592 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6594 /* Flush device addresses */
6595 dev_addr_flush(dev);
6597 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6600 free_percpu(dev->pcpu_refcnt);
6601 dev->pcpu_refcnt = NULL;
6603 /* Compatibility with error handling in drivers */
6604 if (dev->reg_state == NETREG_UNINITIALIZED) {
6605 netdev_freemem(dev);
6609 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6610 dev->reg_state = NETREG_RELEASED;
6612 /* will free via device release */
6613 put_device(&dev->dev);
6615 EXPORT_SYMBOL(free_netdev);
6618 * synchronize_net - Synchronize with packet receive processing
6620 * Wait for packets currently being received to be done.
6621 * Does not block later packets from starting.
6623 void synchronize_net(void)
6626 if (rtnl_is_locked())
6627 synchronize_rcu_expedited();
6631 EXPORT_SYMBOL(synchronize_net);
6634 * unregister_netdevice_queue - remove device from the kernel
6638 * This function shuts down a device interface and removes it
6639 * from the kernel tables.
6640 * If head not NULL, device is queued to be unregistered later.
6642 * Callers must hold the rtnl semaphore. You may want
6643 * unregister_netdev() instead of this.
6646 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6651 list_move_tail(&dev->unreg_list, head);
6653 rollback_registered(dev);
6654 /* Finish processing unregister after unlock */
6658 EXPORT_SYMBOL(unregister_netdevice_queue);
6661 * unregister_netdevice_many - unregister many devices
6662 * @head: list of devices
6664 * Note: As most callers use a stack allocated list_head,
6665 * we force a list_del() to make sure stack wont be corrupted later.
6667 void unregister_netdevice_many(struct list_head *head)
6669 struct net_device *dev;
6671 if (!list_empty(head)) {
6672 rollback_registered_many(head);
6673 list_for_each_entry(dev, head, unreg_list)
6678 EXPORT_SYMBOL(unregister_netdevice_many);
6681 * unregister_netdev - remove device from the kernel
6684 * This function shuts down a device interface and removes it
6685 * from the kernel tables.
6687 * This is just a wrapper for unregister_netdevice that takes
6688 * the rtnl semaphore. In general you want to use this and not
6689 * unregister_netdevice.
6691 void unregister_netdev(struct net_device *dev)
6694 unregister_netdevice(dev);
6697 EXPORT_SYMBOL(unregister_netdev);
6700 * dev_change_net_namespace - move device to different nethost namespace
6702 * @net: network namespace
6703 * @pat: If not NULL name pattern to try if the current device name
6704 * is already taken in the destination network namespace.
6706 * This function shuts down a device interface and moves it
6707 * to a new network namespace. On success 0 is returned, on
6708 * a failure a netagive errno code is returned.
6710 * Callers must hold the rtnl semaphore.
6713 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6719 /* Don't allow namespace local devices to be moved. */
6721 if (dev->features & NETIF_F_NETNS_LOCAL)
6724 /* Ensure the device has been registrered */
6725 if (dev->reg_state != NETREG_REGISTERED)
6728 /* Get out if there is nothing todo */
6730 if (net_eq(dev_net(dev), net))
6733 /* Pick the destination device name, and ensure
6734 * we can use it in the destination network namespace.
6737 if (__dev_get_by_name(net, dev->name)) {
6738 /* We get here if we can't use the current device name */
6741 if (dev_get_valid_name(net, dev, pat) < 0)
6746 * And now a mini version of register_netdevice unregister_netdevice.
6749 /* If device is running close it first. */
6752 /* And unlink it from device chain */
6754 unlist_netdevice(dev);
6758 /* Shutdown queueing discipline. */
6761 /* Notify protocols, that we are about to destroy
6762 this device. They should clean all the things.
6764 Note that dev->reg_state stays at NETREG_REGISTERED.
6765 This is wanted because this way 8021q and macvlan know
6766 the device is just moving and can keep their slaves up.
6768 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6770 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6771 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6774 * Flush the unicast and multicast chains
6779 /* Send a netdev-removed uevent to the old namespace */
6780 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6782 /* Actually switch the network namespace */
6783 dev_net_set(dev, net);
6785 /* If there is an ifindex conflict assign a new one */
6786 if (__dev_get_by_index(net, dev->ifindex)) {
6787 int iflink = (dev->iflink == dev->ifindex);
6788 dev->ifindex = dev_new_index(net);
6790 dev->iflink = dev->ifindex;
6793 /* Send a netdev-add uevent to the new namespace */
6794 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6796 /* Fixup kobjects */
6797 err = device_rename(&dev->dev, dev->name);
6800 /* Add the device back in the hashes */
6801 list_netdevice(dev);
6803 /* Notify protocols, that a new device appeared. */
6804 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6807 * Prevent userspace races by waiting until the network
6808 * device is fully setup before sending notifications.
6810 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6817 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6819 static int dev_cpu_callback(struct notifier_block *nfb,
6820 unsigned long action,
6823 struct sk_buff **list_skb;
6824 struct sk_buff *skb;
6825 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6826 struct softnet_data *sd, *oldsd;
6828 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6831 local_irq_disable();
6832 cpu = smp_processor_id();
6833 sd = &per_cpu(softnet_data, cpu);
6834 oldsd = &per_cpu(softnet_data, oldcpu);
6836 /* Find end of our completion_queue. */
6837 list_skb = &sd->completion_queue;
6839 list_skb = &(*list_skb)->next;
6840 /* Append completion queue from offline CPU. */
6841 *list_skb = oldsd->completion_queue;
6842 oldsd->completion_queue = NULL;
6844 /* Append output queue from offline CPU. */
6845 if (oldsd->output_queue) {
6846 *sd->output_queue_tailp = oldsd->output_queue;
6847 sd->output_queue_tailp = oldsd->output_queue_tailp;
6848 oldsd->output_queue = NULL;
6849 oldsd->output_queue_tailp = &oldsd->output_queue;
6851 /* Append NAPI poll list from offline CPU. */
6852 if (!list_empty(&oldsd->poll_list)) {
6853 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6854 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6857 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6860 /* Process offline CPU's input_pkt_queue */
6861 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6862 netif_rx_internal(skb);
6863 input_queue_head_incr(oldsd);
6865 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6866 netif_rx_internal(skb);
6867 input_queue_head_incr(oldsd);
6875 * netdev_increment_features - increment feature set by one
6876 * @all: current feature set
6877 * @one: new feature set
6878 * @mask: mask feature set
6880 * Computes a new feature set after adding a device with feature set
6881 * @one to the master device with current feature set @all. Will not
6882 * enable anything that is off in @mask. Returns the new feature set.
6884 netdev_features_t netdev_increment_features(netdev_features_t all,
6885 netdev_features_t one, netdev_features_t mask)
6887 if (mask & NETIF_F_GEN_CSUM)
6888 mask |= NETIF_F_ALL_CSUM;
6889 mask |= NETIF_F_VLAN_CHALLENGED;
6891 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6892 all &= one | ~NETIF_F_ALL_FOR_ALL;
6894 /* If one device supports hw checksumming, set for all. */
6895 if (all & NETIF_F_GEN_CSUM)
6896 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6900 EXPORT_SYMBOL(netdev_increment_features);
6902 static struct hlist_head * __net_init netdev_create_hash(void)
6905 struct hlist_head *hash;
6907 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6909 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6910 INIT_HLIST_HEAD(&hash[i]);
6915 /* Initialize per network namespace state */
6916 static int __net_init netdev_init(struct net *net)
6918 if (net != &init_net)
6919 INIT_LIST_HEAD(&net->dev_base_head);
6921 net->dev_name_head = netdev_create_hash();
6922 if (net->dev_name_head == NULL)
6925 net->dev_index_head = netdev_create_hash();
6926 if (net->dev_index_head == NULL)
6932 kfree(net->dev_name_head);
6938 * netdev_drivername - network driver for the device
6939 * @dev: network device
6941 * Determine network driver for device.
6943 const char *netdev_drivername(const struct net_device *dev)
6945 const struct device_driver *driver;
6946 const struct device *parent;
6947 const char *empty = "";
6949 parent = dev->dev.parent;
6953 driver = parent->driver;
6954 if (driver && driver->name)
6955 return driver->name;
6959 static int __netdev_printk(const char *level, const struct net_device *dev,
6960 struct va_format *vaf)
6964 if (dev && dev->dev.parent) {
6965 r = dev_printk_emit(level[1] - '0',
6968 dev_driver_string(dev->dev.parent),
6969 dev_name(dev->dev.parent),
6970 netdev_name(dev), netdev_reg_state(dev),
6973 r = printk("%s%s%s: %pV", level, netdev_name(dev),
6974 netdev_reg_state(dev), vaf);
6976 r = printk("%s(NULL net_device): %pV", level, vaf);
6982 int netdev_printk(const char *level, const struct net_device *dev,
6983 const char *format, ...)
6985 struct va_format vaf;
6989 va_start(args, format);
6994 r = __netdev_printk(level, dev, &vaf);
7000 EXPORT_SYMBOL(netdev_printk);
7002 #define define_netdev_printk_level(func, level) \
7003 int func(const struct net_device *dev, const char *fmt, ...) \
7006 struct va_format vaf; \
7009 va_start(args, fmt); \
7014 r = __netdev_printk(level, dev, &vaf); \
7020 EXPORT_SYMBOL(func);
7022 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7023 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7024 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7025 define_netdev_printk_level(netdev_err, KERN_ERR);
7026 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7027 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7028 define_netdev_printk_level(netdev_info, KERN_INFO);
7030 static void __net_exit netdev_exit(struct net *net)
7032 kfree(net->dev_name_head);
7033 kfree(net->dev_index_head);
7036 static struct pernet_operations __net_initdata netdev_net_ops = {
7037 .init = netdev_init,
7038 .exit = netdev_exit,
7041 static void __net_exit default_device_exit(struct net *net)
7043 struct net_device *dev, *aux;
7045 * Push all migratable network devices back to the
7046 * initial network namespace
7049 for_each_netdev_safe(net, dev, aux) {
7051 char fb_name[IFNAMSIZ];
7053 /* Ignore unmoveable devices (i.e. loopback) */
7054 if (dev->features & NETIF_F_NETNS_LOCAL)
7057 /* Leave virtual devices for the generic cleanup */
7058 if (dev->rtnl_link_ops)
7061 /* Push remaining network devices to init_net */
7062 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7063 err = dev_change_net_namespace(dev, &init_net, fb_name);
7065 pr_emerg("%s: failed to move %s to init_net: %d\n",
7066 __func__, dev->name, err);
7073 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7075 /* Return with the rtnl_lock held when there are no network
7076 * devices unregistering in any network namespace in net_list.
7083 prepare_to_wait(&netdev_unregistering_wq, &wait,
7084 TASK_UNINTERRUPTIBLE);
7085 unregistering = false;
7087 list_for_each_entry(net, net_list, exit_list) {
7088 if (net->dev_unreg_count > 0) {
7089 unregistering = true;
7098 finish_wait(&netdev_unregistering_wq, &wait);
7101 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7103 /* At exit all network devices most be removed from a network
7104 * namespace. Do this in the reverse order of registration.
7105 * Do this across as many network namespaces as possible to
7106 * improve batching efficiency.
7108 struct net_device *dev;
7110 LIST_HEAD(dev_kill_list);
7112 /* To prevent network device cleanup code from dereferencing
7113 * loopback devices or network devices that have been freed
7114 * wait here for all pending unregistrations to complete,
7115 * before unregistring the loopback device and allowing the
7116 * network namespace be freed.
7118 * The netdev todo list containing all network devices
7119 * unregistrations that happen in default_device_exit_batch
7120 * will run in the rtnl_unlock() at the end of
7121 * default_device_exit_batch.
7123 rtnl_lock_unregistering(net_list);
7124 list_for_each_entry(net, net_list, exit_list) {
7125 for_each_netdev_reverse(net, dev) {
7126 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7127 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7129 unregister_netdevice_queue(dev, &dev_kill_list);
7132 unregister_netdevice_many(&dev_kill_list);
7136 static struct pernet_operations __net_initdata default_device_ops = {
7137 .exit = default_device_exit,
7138 .exit_batch = default_device_exit_batch,
7142 * Initialize the DEV module. At boot time this walks the device list and
7143 * unhooks any devices that fail to initialise (normally hardware not
7144 * present) and leaves us with a valid list of present and active devices.
7149 * This is called single threaded during boot, so no need
7150 * to take the rtnl semaphore.
7152 static int __init net_dev_init(void)
7154 int i, rc = -ENOMEM;
7156 BUG_ON(!dev_boot_phase);
7158 if (dev_proc_init())
7161 if (netdev_kobject_init())
7164 INIT_LIST_HEAD(&ptype_all);
7165 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7166 INIT_LIST_HEAD(&ptype_base[i]);
7168 INIT_LIST_HEAD(&offload_base);
7170 if (register_pernet_subsys(&netdev_net_ops))
7174 * Initialise the packet receive queues.
7177 for_each_possible_cpu(i) {
7178 struct softnet_data *sd = &per_cpu(softnet_data, i);
7180 skb_queue_head_init(&sd->input_pkt_queue);
7181 skb_queue_head_init(&sd->process_queue);
7182 INIT_LIST_HEAD(&sd->poll_list);
7183 sd->output_queue_tailp = &sd->output_queue;
7185 sd->csd.func = rps_trigger_softirq;
7190 sd->backlog.poll = process_backlog;
7191 sd->backlog.weight = weight_p;
7196 /* The loopback device is special if any other network devices
7197 * is present in a network namespace the loopback device must
7198 * be present. Since we now dynamically allocate and free the
7199 * loopback device ensure this invariant is maintained by
7200 * keeping the loopback device as the first device on the
7201 * list of network devices. Ensuring the loopback devices
7202 * is the first device that appears and the last network device
7205 if (register_pernet_device(&loopback_net_ops))
7208 if (register_pernet_device(&default_device_ops))
7211 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7212 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7214 hotcpu_notifier(dev_cpu_callback, 0);
7221 subsys_initcall(net_dev_init);