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 - 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 * rtnl_lock(), and result refcount is unchanged.
910 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
913 struct net_device *dev, *ret;
918 for_each_netdev(net, dev) {
919 if (((dev->flags ^ if_flags) & mask) == 0) {
926 EXPORT_SYMBOL(__dev_get_by_flags);
929 * dev_valid_name - check if name is okay for network device
932 * Network device names need to be valid file names to
933 * to allow sysfs to work. We also disallow any kind of
936 bool dev_valid_name(const char *name)
940 if (strlen(name) >= IFNAMSIZ)
942 if (!strcmp(name, ".") || !strcmp(name, ".."))
946 if (*name == '/' || isspace(*name))
952 EXPORT_SYMBOL(dev_valid_name);
955 * __dev_alloc_name - allocate a name for a device
956 * @net: network namespace to allocate the device name in
957 * @name: name format string
958 * @buf: scratch buffer and result name string
960 * Passed a format string - eg "lt%d" it will try and find a suitable
961 * id. It scans list of devices to build up a free map, then chooses
962 * the first empty slot. The caller must hold the dev_base or rtnl lock
963 * while allocating the name and adding the device in order to avoid
965 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
966 * Returns the number of the unit assigned or a negative errno code.
969 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
973 const int max_netdevices = 8*PAGE_SIZE;
974 unsigned long *inuse;
975 struct net_device *d;
977 p = strnchr(name, IFNAMSIZ-1, '%');
980 * Verify the string as this thing may have come from
981 * the user. There must be either one "%d" and no other "%"
984 if (p[1] != 'd' || strchr(p + 2, '%'))
987 /* Use one page as a bit array of possible slots */
988 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
992 for_each_netdev(net, d) {
993 if (!sscanf(d->name, name, &i))
995 if (i < 0 || i >= max_netdevices)
998 /* avoid cases where sscanf is not exact inverse of printf */
999 snprintf(buf, IFNAMSIZ, name, i);
1000 if (!strncmp(buf, d->name, IFNAMSIZ))
1004 i = find_first_zero_bit(inuse, max_netdevices);
1005 free_page((unsigned long) inuse);
1009 snprintf(buf, IFNAMSIZ, name, i);
1010 if (!__dev_get_by_name(net, buf))
1013 /* It is possible to run out of possible slots
1014 * when the name is long and there isn't enough space left
1015 * for the digits, or if all bits are used.
1021 * dev_alloc_name - allocate a name for a device
1023 * @name: name format string
1025 * Passed a format string - eg "lt%d" it will try and find a suitable
1026 * id. It scans list of devices to build up a free map, then chooses
1027 * the first empty slot. The caller must hold the dev_base or rtnl lock
1028 * while allocating the name and adding the device in order to avoid
1030 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1031 * Returns the number of the unit assigned or a negative errno code.
1034 int dev_alloc_name(struct net_device *dev, const char *name)
1040 BUG_ON(!dev_net(dev));
1042 ret = __dev_alloc_name(net, name, buf);
1044 strlcpy(dev->name, buf, IFNAMSIZ);
1047 EXPORT_SYMBOL(dev_alloc_name);
1049 static int dev_alloc_name_ns(struct net *net,
1050 struct net_device *dev,
1056 ret = __dev_alloc_name(net, name, buf);
1058 strlcpy(dev->name, buf, IFNAMSIZ);
1062 static int dev_get_valid_name(struct net *net,
1063 struct net_device *dev,
1068 if (!dev_valid_name(name))
1071 if (strchr(name, '%'))
1072 return dev_alloc_name_ns(net, dev, name);
1073 else if (__dev_get_by_name(net, name))
1075 else if (dev->name != name)
1076 strlcpy(dev->name, name, IFNAMSIZ);
1082 * dev_change_name - change name of a device
1084 * @newname: name (or format string) must be at least IFNAMSIZ
1086 * Change name of a device, can pass format strings "eth%d".
1089 int dev_change_name(struct net_device *dev, const char *newname)
1091 unsigned char old_assign_type;
1092 char oldname[IFNAMSIZ];
1098 BUG_ON(!dev_net(dev));
1101 if (dev->flags & IFF_UP)
1104 write_seqcount_begin(&devnet_rename_seq);
1106 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1107 write_seqcount_end(&devnet_rename_seq);
1111 memcpy(oldname, dev->name, IFNAMSIZ);
1113 err = dev_get_valid_name(net, dev, newname);
1115 write_seqcount_end(&devnet_rename_seq);
1119 if (oldname[0] && !strchr(oldname, '%'))
1120 netdev_info(dev, "renamed from %s\n", oldname);
1122 old_assign_type = dev->name_assign_type;
1123 dev->name_assign_type = NET_NAME_RENAMED;
1126 ret = device_rename(&dev->dev, dev->name);
1128 memcpy(dev->name, oldname, IFNAMSIZ);
1129 dev->name_assign_type = old_assign_type;
1130 write_seqcount_end(&devnet_rename_seq);
1134 write_seqcount_end(&devnet_rename_seq);
1136 netdev_adjacent_rename_links(dev, oldname);
1138 write_lock_bh(&dev_base_lock);
1139 hlist_del_rcu(&dev->name_hlist);
1140 write_unlock_bh(&dev_base_lock);
1144 write_lock_bh(&dev_base_lock);
1145 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1146 write_unlock_bh(&dev_base_lock);
1148 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1149 ret = notifier_to_errno(ret);
1152 /* err >= 0 after dev_alloc_name() or stores the first errno */
1155 write_seqcount_begin(&devnet_rename_seq);
1156 memcpy(dev->name, oldname, IFNAMSIZ);
1157 memcpy(oldname, newname, IFNAMSIZ);
1158 dev->name_assign_type = old_assign_type;
1159 old_assign_type = NET_NAME_RENAMED;
1162 pr_err("%s: name change rollback failed: %d\n",
1171 * dev_set_alias - change ifalias of a device
1173 * @alias: name up to IFALIASZ
1174 * @len: limit of bytes to copy from info
1176 * Set ifalias for a device,
1178 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1184 if (len >= IFALIASZ)
1188 kfree(dev->ifalias);
1189 dev->ifalias = NULL;
1193 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1196 dev->ifalias = new_ifalias;
1198 strlcpy(dev->ifalias, alias, len+1);
1204 * netdev_features_change - device changes features
1205 * @dev: device to cause notification
1207 * Called to indicate a device has changed features.
1209 void netdev_features_change(struct net_device *dev)
1211 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1213 EXPORT_SYMBOL(netdev_features_change);
1216 * netdev_state_change - device changes state
1217 * @dev: device to cause notification
1219 * Called to indicate a device has changed state. This function calls
1220 * the notifier chains for netdev_chain and sends a NEWLINK message
1221 * to the routing socket.
1223 void netdev_state_change(struct net_device *dev)
1225 if (dev->flags & IFF_UP) {
1226 struct netdev_notifier_change_info change_info;
1228 change_info.flags_changed = 0;
1229 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1231 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1234 EXPORT_SYMBOL(netdev_state_change);
1237 * netdev_notify_peers - notify network peers about existence of @dev
1238 * @dev: network device
1240 * Generate traffic such that interested network peers are aware of
1241 * @dev, such as by generating a gratuitous ARP. This may be used when
1242 * a device wants to inform the rest of the network about some sort of
1243 * reconfiguration such as a failover event or virtual machine
1246 void netdev_notify_peers(struct net_device *dev)
1249 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1252 EXPORT_SYMBOL(netdev_notify_peers);
1254 static int __dev_open(struct net_device *dev)
1256 const struct net_device_ops *ops = dev->netdev_ops;
1261 if (!netif_device_present(dev))
1264 /* Block netpoll from trying to do any rx path servicing.
1265 * If we don't do this there is a chance ndo_poll_controller
1266 * or ndo_poll may be running while we open the device
1268 netpoll_poll_disable(dev);
1270 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1271 ret = notifier_to_errno(ret);
1275 set_bit(__LINK_STATE_START, &dev->state);
1277 if (ops->ndo_validate_addr)
1278 ret = ops->ndo_validate_addr(dev);
1280 if (!ret && ops->ndo_open)
1281 ret = ops->ndo_open(dev);
1283 netpoll_poll_enable(dev);
1286 clear_bit(__LINK_STATE_START, &dev->state);
1288 dev->flags |= IFF_UP;
1289 dev_set_rx_mode(dev);
1291 add_device_randomness(dev->dev_addr, dev->addr_len);
1298 * dev_open - prepare an interface for use.
1299 * @dev: device to open
1301 * Takes a device from down to up state. The device's private open
1302 * function is invoked and then the multicast lists are loaded. Finally
1303 * the device is moved into the up state and a %NETDEV_UP message is
1304 * sent to the netdev notifier chain.
1306 * Calling this function on an active interface is a nop. On a failure
1307 * a negative errno code is returned.
1309 int dev_open(struct net_device *dev)
1313 if (dev->flags & IFF_UP)
1316 ret = __dev_open(dev);
1320 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1321 call_netdevice_notifiers(NETDEV_UP, dev);
1325 EXPORT_SYMBOL(dev_open);
1327 static int __dev_close_many(struct list_head *head)
1329 struct net_device *dev;
1334 list_for_each_entry(dev, head, close_list) {
1335 /* Temporarily disable netpoll until the interface is down */
1336 netpoll_poll_disable(dev);
1338 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1340 clear_bit(__LINK_STATE_START, &dev->state);
1342 /* Synchronize to scheduled poll. We cannot touch poll list, it
1343 * can be even on different cpu. So just clear netif_running().
1345 * dev->stop() will invoke napi_disable() on all of it's
1346 * napi_struct instances on this device.
1348 smp_mb__after_atomic(); /* Commit netif_running(). */
1351 dev_deactivate_many(head);
1353 list_for_each_entry(dev, head, close_list) {
1354 const struct net_device_ops *ops = dev->netdev_ops;
1357 * Call the device specific close. This cannot fail.
1358 * Only if device is UP
1360 * We allow it to be called even after a DETACH hot-plug
1366 dev->flags &= ~IFF_UP;
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 = this_cpu_ptr(&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 netif_schedule_queue(struct netdev_queue *txq)
2183 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2184 struct Qdisc *q = rcu_dereference(txq->qdisc);
2186 __netif_schedule(q);
2190 EXPORT_SYMBOL(netif_schedule_queue);
2193 * netif_wake_subqueue - allow sending packets on subqueue
2194 * @dev: network device
2195 * @queue_index: sub queue index
2197 * Resume individual transmit queue of a device with multiple transmit queues.
2199 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2201 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2203 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2207 q = rcu_dereference(txq->qdisc);
2208 __netif_schedule(q);
2212 EXPORT_SYMBOL(netif_wake_subqueue);
2214 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2216 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2220 q = rcu_dereference(dev_queue->qdisc);
2221 __netif_schedule(q);
2225 EXPORT_SYMBOL(netif_tx_wake_queue);
2227 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2229 unsigned long flags;
2231 if (likely(atomic_read(&skb->users) == 1)) {
2233 atomic_set(&skb->users, 0);
2234 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2237 get_kfree_skb_cb(skb)->reason = reason;
2238 local_irq_save(flags);
2239 skb->next = __this_cpu_read(softnet_data.completion_queue);
2240 __this_cpu_write(softnet_data.completion_queue, skb);
2241 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2242 local_irq_restore(flags);
2244 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2246 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2248 if (in_irq() || irqs_disabled())
2249 __dev_kfree_skb_irq(skb, reason);
2253 EXPORT_SYMBOL(__dev_kfree_skb_any);
2257 * netif_device_detach - mark device as removed
2258 * @dev: network device
2260 * Mark device as removed from system and therefore no longer available.
2262 void netif_device_detach(struct net_device *dev)
2264 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2265 netif_running(dev)) {
2266 netif_tx_stop_all_queues(dev);
2269 EXPORT_SYMBOL(netif_device_detach);
2272 * netif_device_attach - mark device as attached
2273 * @dev: network device
2275 * Mark device as attached from system and restart if needed.
2277 void netif_device_attach(struct net_device *dev)
2279 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2280 netif_running(dev)) {
2281 netif_tx_wake_all_queues(dev);
2282 __netdev_watchdog_up(dev);
2285 EXPORT_SYMBOL(netif_device_attach);
2287 static void skb_warn_bad_offload(const struct sk_buff *skb)
2289 static const netdev_features_t null_features = 0;
2290 struct net_device *dev = skb->dev;
2291 const char *driver = "";
2293 if (!net_ratelimit())
2296 if (dev && dev->dev.parent)
2297 driver = dev_driver_string(dev->dev.parent);
2299 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2300 "gso_type=%d ip_summed=%d\n",
2301 driver, dev ? &dev->features : &null_features,
2302 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2303 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2304 skb_shinfo(skb)->gso_type, skb->ip_summed);
2308 * Invalidate hardware checksum when packet is to be mangled, and
2309 * complete checksum manually on outgoing path.
2311 int skb_checksum_help(struct sk_buff *skb)
2314 int ret = 0, offset;
2316 if (skb->ip_summed == CHECKSUM_COMPLETE)
2317 goto out_set_summed;
2319 if (unlikely(skb_shinfo(skb)->gso_size)) {
2320 skb_warn_bad_offload(skb);
2324 /* Before computing a checksum, we should make sure no frag could
2325 * be modified by an external entity : checksum could be wrong.
2327 if (skb_has_shared_frag(skb)) {
2328 ret = __skb_linearize(skb);
2333 offset = skb_checksum_start_offset(skb);
2334 BUG_ON(offset >= skb_headlen(skb));
2335 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2337 offset += skb->csum_offset;
2338 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2340 if (skb_cloned(skb) &&
2341 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2342 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2347 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2349 skb->ip_summed = CHECKSUM_NONE;
2353 EXPORT_SYMBOL(skb_checksum_help);
2355 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2357 unsigned int vlan_depth = skb->mac_len;
2358 __be16 type = skb->protocol;
2360 /* Tunnel gso handlers can set protocol to ethernet. */
2361 if (type == htons(ETH_P_TEB)) {
2364 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2367 eth = (struct ethhdr *)skb_mac_header(skb);
2368 type = eth->h_proto;
2371 /* if skb->protocol is 802.1Q/AD then the header should already be
2372 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at
2373 * ETH_HLEN otherwise
2375 if (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2377 if (WARN_ON(vlan_depth < VLAN_HLEN))
2379 vlan_depth -= VLAN_HLEN;
2381 vlan_depth = ETH_HLEN;
2384 struct vlan_hdr *vh;
2386 if (unlikely(!pskb_may_pull(skb,
2387 vlan_depth + VLAN_HLEN)))
2390 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2391 type = vh->h_vlan_encapsulated_proto;
2392 vlan_depth += VLAN_HLEN;
2393 } while (type == htons(ETH_P_8021Q) ||
2394 type == htons(ETH_P_8021AD));
2397 *depth = vlan_depth;
2403 * skb_mac_gso_segment - mac layer segmentation handler.
2404 * @skb: buffer to segment
2405 * @features: features for the output path (see dev->features)
2407 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2408 netdev_features_t features)
2410 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2411 struct packet_offload *ptype;
2412 int vlan_depth = skb->mac_len;
2413 __be16 type = skb_network_protocol(skb, &vlan_depth);
2415 if (unlikely(!type))
2416 return ERR_PTR(-EINVAL);
2418 __skb_pull(skb, vlan_depth);
2421 list_for_each_entry_rcu(ptype, &offload_base, list) {
2422 if (ptype->type == type && ptype->callbacks.gso_segment) {
2423 segs = ptype->callbacks.gso_segment(skb, features);
2429 __skb_push(skb, skb->data - skb_mac_header(skb));
2433 EXPORT_SYMBOL(skb_mac_gso_segment);
2436 /* openvswitch calls this on rx path, so we need a different check.
2438 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2441 return skb->ip_summed != CHECKSUM_PARTIAL;
2443 return skb->ip_summed == CHECKSUM_NONE;
2447 * __skb_gso_segment - Perform segmentation on skb.
2448 * @skb: buffer to segment
2449 * @features: features for the output path (see dev->features)
2450 * @tx_path: whether it is called in TX path
2452 * This function segments the given skb and returns a list of segments.
2454 * It may return NULL if the skb requires no segmentation. This is
2455 * only possible when GSO is used for verifying header integrity.
2457 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2458 netdev_features_t features, bool tx_path)
2460 if (unlikely(skb_needs_check(skb, tx_path))) {
2463 skb_warn_bad_offload(skb);
2465 err = skb_cow_head(skb, 0);
2467 return ERR_PTR(err);
2470 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2471 SKB_GSO_CB(skb)->encap_level = 0;
2473 skb_reset_mac_header(skb);
2474 skb_reset_mac_len(skb);
2476 return skb_mac_gso_segment(skb, features);
2478 EXPORT_SYMBOL(__skb_gso_segment);
2480 /* Take action when hardware reception checksum errors are detected. */
2482 void netdev_rx_csum_fault(struct net_device *dev)
2484 if (net_ratelimit()) {
2485 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2489 EXPORT_SYMBOL(netdev_rx_csum_fault);
2492 /* Actually, we should eliminate this check as soon as we know, that:
2493 * 1. IOMMU is present and allows to map all the memory.
2494 * 2. No high memory really exists on this machine.
2497 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2499 #ifdef CONFIG_HIGHMEM
2501 if (!(dev->features & NETIF_F_HIGHDMA)) {
2502 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2503 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2504 if (PageHighMem(skb_frag_page(frag)))
2509 if (PCI_DMA_BUS_IS_PHYS) {
2510 struct device *pdev = dev->dev.parent;
2514 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2515 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2516 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2517 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2525 /* If MPLS offload request, verify we are testing hardware MPLS features
2526 * instead of standard features for the netdev.
2528 #ifdef CONFIG_NET_MPLS_GSO
2529 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2530 netdev_features_t features,
2533 if (type == htons(ETH_P_MPLS_UC) || type == htons(ETH_P_MPLS_MC))
2534 features &= skb->dev->mpls_features;
2539 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2540 netdev_features_t features,
2547 static netdev_features_t harmonize_features(struct sk_buff *skb,
2548 netdev_features_t features)
2553 type = skb_network_protocol(skb, &tmp);
2554 features = net_mpls_features(skb, features, type);
2556 if (skb->ip_summed != CHECKSUM_NONE &&
2557 !can_checksum_protocol(features, type)) {
2558 features &= ~NETIF_F_ALL_CSUM;
2559 } else if (illegal_highdma(skb->dev, skb)) {
2560 features &= ~NETIF_F_SG;
2566 netdev_features_t netif_skb_features(struct sk_buff *skb)
2568 const struct net_device *dev = skb->dev;
2569 netdev_features_t features = dev->features;
2570 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2571 __be16 protocol = skb->protocol;
2573 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2574 features &= ~NETIF_F_GSO_MASK;
2576 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2577 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2578 protocol = veh->h_vlan_encapsulated_proto;
2579 } else if (!vlan_tx_tag_present(skb)) {
2580 return harmonize_features(skb, features);
2583 features = netdev_intersect_features(features,
2584 dev->vlan_features |
2585 NETIF_F_HW_VLAN_CTAG_TX |
2586 NETIF_F_HW_VLAN_STAG_TX);
2588 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2589 features = netdev_intersect_features(features,
2594 NETIF_F_HW_VLAN_CTAG_TX |
2595 NETIF_F_HW_VLAN_STAG_TX);
2597 return harmonize_features(skb, features);
2599 EXPORT_SYMBOL(netif_skb_features);
2601 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2602 struct netdev_queue *txq, bool more)
2607 if (!list_empty(&ptype_all))
2608 dev_queue_xmit_nit(skb, dev);
2611 trace_net_dev_start_xmit(skb, dev);
2612 rc = netdev_start_xmit(skb, dev, txq, more);
2613 trace_net_dev_xmit(skb, rc, dev, len);
2618 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2619 struct netdev_queue *txq, int *ret)
2621 struct sk_buff *skb = first;
2622 int rc = NETDEV_TX_OK;
2625 struct sk_buff *next = skb->next;
2628 rc = xmit_one(skb, dev, txq, next != NULL);
2629 if (unlikely(!dev_xmit_complete(rc))) {
2635 if (netif_xmit_stopped(txq) && skb) {
2636 rc = NETDEV_TX_BUSY;
2646 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2647 netdev_features_t features)
2649 if (vlan_tx_tag_present(skb) &&
2650 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2651 skb = __vlan_put_tag(skb, skb->vlan_proto,
2652 vlan_tx_tag_get(skb));
2659 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2661 netdev_features_t features;
2666 features = netif_skb_features(skb);
2667 skb = validate_xmit_vlan(skb, features);
2671 /* If encapsulation offload request, verify we are testing
2672 * hardware encapsulation features instead of standard
2673 * features for the netdev
2675 if (skb->encapsulation)
2676 features &= dev->hw_enc_features;
2678 if (netif_needs_gso(dev, skb, features)) {
2679 struct sk_buff *segs;
2681 segs = skb_gso_segment(skb, features);
2689 if (skb_needs_linearize(skb, features) &&
2690 __skb_linearize(skb))
2693 /* If packet is not checksummed and device does not
2694 * support checksumming for this protocol, complete
2695 * checksumming here.
2697 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2698 if (skb->encapsulation)
2699 skb_set_inner_transport_header(skb,
2700 skb_checksum_start_offset(skb));
2702 skb_set_transport_header(skb,
2703 skb_checksum_start_offset(skb));
2704 if (!(features & NETIF_F_ALL_CSUM) &&
2705 skb_checksum_help(skb))
2718 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2720 struct sk_buff *next, *head = NULL, *tail;
2722 for (; skb != NULL; skb = next) {
2726 /* in case skb wont be segmented, point to itself */
2729 skb = validate_xmit_skb(skb, dev);
2737 /* If skb was segmented, skb->prev points to
2738 * the last segment. If not, it still contains skb.
2745 static void qdisc_pkt_len_init(struct sk_buff *skb)
2747 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2749 qdisc_skb_cb(skb)->pkt_len = skb->len;
2751 /* To get more precise estimation of bytes sent on wire,
2752 * we add to pkt_len the headers size of all segments
2754 if (shinfo->gso_size) {
2755 unsigned int hdr_len;
2756 u16 gso_segs = shinfo->gso_segs;
2758 /* mac layer + network layer */
2759 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2761 /* + transport layer */
2762 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2763 hdr_len += tcp_hdrlen(skb);
2765 hdr_len += sizeof(struct udphdr);
2767 if (shinfo->gso_type & SKB_GSO_DODGY)
2768 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2771 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2775 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2776 struct net_device *dev,
2777 struct netdev_queue *txq)
2779 spinlock_t *root_lock = qdisc_lock(q);
2783 qdisc_pkt_len_init(skb);
2784 qdisc_calculate_pkt_len(skb, q);
2786 * Heuristic to force contended enqueues to serialize on a
2787 * separate lock before trying to get qdisc main lock.
2788 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2789 * often and dequeue packets faster.
2791 contended = qdisc_is_running(q);
2792 if (unlikely(contended))
2793 spin_lock(&q->busylock);
2795 spin_lock(root_lock);
2796 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2799 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2800 qdisc_run_begin(q)) {
2802 * This is a work-conserving queue; there are no old skbs
2803 * waiting to be sent out; and the qdisc is not running -
2804 * xmit the skb directly.
2807 qdisc_bstats_update(q, skb);
2809 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2810 if (unlikely(contended)) {
2811 spin_unlock(&q->busylock);
2818 rc = NET_XMIT_SUCCESS;
2820 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2821 if (qdisc_run_begin(q)) {
2822 if (unlikely(contended)) {
2823 spin_unlock(&q->busylock);
2829 spin_unlock(root_lock);
2830 if (unlikely(contended))
2831 spin_unlock(&q->busylock);
2835 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2836 static void skb_update_prio(struct sk_buff *skb)
2838 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2840 if (!skb->priority && skb->sk && map) {
2841 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2843 if (prioidx < map->priomap_len)
2844 skb->priority = map->priomap[prioidx];
2848 #define skb_update_prio(skb)
2851 static DEFINE_PER_CPU(int, xmit_recursion);
2852 #define RECURSION_LIMIT 10
2855 * dev_loopback_xmit - loop back @skb
2856 * @skb: buffer to transmit
2858 int dev_loopback_xmit(struct sk_buff *skb)
2860 skb_reset_mac_header(skb);
2861 __skb_pull(skb, skb_network_offset(skb));
2862 skb->pkt_type = PACKET_LOOPBACK;
2863 skb->ip_summed = CHECKSUM_UNNECESSARY;
2864 WARN_ON(!skb_dst(skb));
2869 EXPORT_SYMBOL(dev_loopback_xmit);
2872 * __dev_queue_xmit - transmit a buffer
2873 * @skb: buffer to transmit
2874 * @accel_priv: private data used for L2 forwarding offload
2876 * Queue a buffer for transmission to a network device. The caller must
2877 * have set the device and priority and built the buffer before calling
2878 * this function. The function can be called from an interrupt.
2880 * A negative errno code is returned on a failure. A success does not
2881 * guarantee the frame will be transmitted as it may be dropped due
2882 * to congestion or traffic shaping.
2884 * -----------------------------------------------------------------------------------
2885 * I notice this method can also return errors from the queue disciplines,
2886 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2889 * Regardless of the return value, the skb is consumed, so it is currently
2890 * difficult to retry a send to this method. (You can bump the ref count
2891 * before sending to hold a reference for retry if you are careful.)
2893 * When calling this method, interrupts MUST be enabled. This is because
2894 * the BH enable code must have IRQs enabled so that it will not deadlock.
2897 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2899 struct net_device *dev = skb->dev;
2900 struct netdev_queue *txq;
2904 skb_reset_mac_header(skb);
2906 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
2907 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
2909 /* Disable soft irqs for various locks below. Also
2910 * stops preemption for RCU.
2914 skb_update_prio(skb);
2916 /* If device/qdisc don't need skb->dst, release it right now while
2917 * its hot in this cpu cache.
2919 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2924 txq = netdev_pick_tx(dev, skb, accel_priv);
2925 q = rcu_dereference_bh(txq->qdisc);
2927 #ifdef CONFIG_NET_CLS_ACT
2928 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2930 trace_net_dev_queue(skb);
2932 rc = __dev_xmit_skb(skb, q, dev, txq);
2936 /* The device has no queue. Common case for software devices:
2937 loopback, all the sorts of tunnels...
2939 Really, it is unlikely that netif_tx_lock protection is necessary
2940 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2942 However, it is possible, that they rely on protection
2945 Check this and shot the lock. It is not prone from deadlocks.
2946 Either shot noqueue qdisc, it is even simpler 8)
2948 if (dev->flags & IFF_UP) {
2949 int cpu = smp_processor_id(); /* ok because BHs are off */
2951 if (txq->xmit_lock_owner != cpu) {
2953 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2954 goto recursion_alert;
2956 skb = validate_xmit_skb(skb, dev);
2960 HARD_TX_LOCK(dev, txq, cpu);
2962 if (!netif_xmit_stopped(txq)) {
2963 __this_cpu_inc(xmit_recursion);
2964 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
2965 __this_cpu_dec(xmit_recursion);
2966 if (dev_xmit_complete(rc)) {
2967 HARD_TX_UNLOCK(dev, txq);
2971 HARD_TX_UNLOCK(dev, txq);
2972 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2975 /* Recursion is detected! It is possible,
2979 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2986 rcu_read_unlock_bh();
2988 atomic_long_inc(&dev->tx_dropped);
2989 kfree_skb_list(skb);
2992 rcu_read_unlock_bh();
2996 int dev_queue_xmit(struct sk_buff *skb)
2998 return __dev_queue_xmit(skb, NULL);
3000 EXPORT_SYMBOL(dev_queue_xmit);
3002 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3004 return __dev_queue_xmit(skb, accel_priv);
3006 EXPORT_SYMBOL(dev_queue_xmit_accel);
3009 /*=======================================================================
3011 =======================================================================*/
3013 int netdev_max_backlog __read_mostly = 1000;
3014 EXPORT_SYMBOL(netdev_max_backlog);
3016 int netdev_tstamp_prequeue __read_mostly = 1;
3017 int netdev_budget __read_mostly = 300;
3018 int weight_p __read_mostly = 64; /* old backlog weight */
3020 /* Called with irq disabled */
3021 static inline void ____napi_schedule(struct softnet_data *sd,
3022 struct napi_struct *napi)
3024 list_add_tail(&napi->poll_list, &sd->poll_list);
3025 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3030 /* One global table that all flow-based protocols share. */
3031 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3032 EXPORT_SYMBOL(rps_sock_flow_table);
3034 struct static_key rps_needed __read_mostly;
3036 static struct rps_dev_flow *
3037 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3038 struct rps_dev_flow *rflow, u16 next_cpu)
3040 if (next_cpu != RPS_NO_CPU) {
3041 #ifdef CONFIG_RFS_ACCEL
3042 struct netdev_rx_queue *rxqueue;
3043 struct rps_dev_flow_table *flow_table;
3044 struct rps_dev_flow *old_rflow;
3049 /* Should we steer this flow to a different hardware queue? */
3050 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3051 !(dev->features & NETIF_F_NTUPLE))
3053 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3054 if (rxq_index == skb_get_rx_queue(skb))
3057 rxqueue = dev->_rx + rxq_index;
3058 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3061 flow_id = skb_get_hash(skb) & flow_table->mask;
3062 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3063 rxq_index, flow_id);
3067 rflow = &flow_table->flows[flow_id];
3069 if (old_rflow->filter == rflow->filter)
3070 old_rflow->filter = RPS_NO_FILTER;
3074 per_cpu(softnet_data, next_cpu).input_queue_head;
3077 rflow->cpu = next_cpu;
3082 * get_rps_cpu is called from netif_receive_skb and returns the target
3083 * CPU from the RPS map of the receiving queue for a given skb.
3084 * rcu_read_lock must be held on entry.
3086 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3087 struct rps_dev_flow **rflowp)
3089 struct netdev_rx_queue *rxqueue;
3090 struct rps_map *map;
3091 struct rps_dev_flow_table *flow_table;
3092 struct rps_sock_flow_table *sock_flow_table;
3097 if (skb_rx_queue_recorded(skb)) {
3098 u16 index = skb_get_rx_queue(skb);
3099 if (unlikely(index >= dev->real_num_rx_queues)) {
3100 WARN_ONCE(dev->real_num_rx_queues > 1,
3101 "%s received packet on queue %u, but number "
3102 "of RX queues is %u\n",
3103 dev->name, index, dev->real_num_rx_queues);
3106 rxqueue = dev->_rx + index;
3110 map = rcu_dereference(rxqueue->rps_map);
3112 if (map->len == 1 &&
3113 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3114 tcpu = map->cpus[0];
3115 if (cpu_online(tcpu))
3119 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3123 skb_reset_network_header(skb);
3124 hash = skb_get_hash(skb);
3128 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3129 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3130 if (flow_table && sock_flow_table) {
3132 struct rps_dev_flow *rflow;
3134 rflow = &flow_table->flows[hash & flow_table->mask];
3137 next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask];
3140 * If the desired CPU (where last recvmsg was done) is
3141 * different from current CPU (one in the rx-queue flow
3142 * table entry), switch if one of the following holds:
3143 * - Current CPU is unset (equal to RPS_NO_CPU).
3144 * - Current CPU is offline.
3145 * - The current CPU's queue tail has advanced beyond the
3146 * last packet that was enqueued using this table entry.
3147 * This guarantees that all previous packets for the flow
3148 * have been dequeued, thus preserving in order delivery.
3150 if (unlikely(tcpu != next_cpu) &&
3151 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3152 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3153 rflow->last_qtail)) >= 0)) {
3155 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3158 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3166 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3167 if (cpu_online(tcpu)) {
3177 #ifdef CONFIG_RFS_ACCEL
3180 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3181 * @dev: Device on which the filter was set
3182 * @rxq_index: RX queue index
3183 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3184 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3186 * Drivers that implement ndo_rx_flow_steer() should periodically call
3187 * this function for each installed filter and remove the filters for
3188 * which it returns %true.
3190 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3191 u32 flow_id, u16 filter_id)
3193 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3194 struct rps_dev_flow_table *flow_table;
3195 struct rps_dev_flow *rflow;
3200 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3201 if (flow_table && flow_id <= flow_table->mask) {
3202 rflow = &flow_table->flows[flow_id];
3203 cpu = ACCESS_ONCE(rflow->cpu);
3204 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3205 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3206 rflow->last_qtail) <
3207 (int)(10 * flow_table->mask)))
3213 EXPORT_SYMBOL(rps_may_expire_flow);
3215 #endif /* CONFIG_RFS_ACCEL */
3217 /* Called from hardirq (IPI) context */
3218 static void rps_trigger_softirq(void *data)
3220 struct softnet_data *sd = data;
3222 ____napi_schedule(sd, &sd->backlog);
3226 #endif /* CONFIG_RPS */
3229 * Check if this softnet_data structure is another cpu one
3230 * If yes, queue it to our IPI list and return 1
3233 static int rps_ipi_queued(struct softnet_data *sd)
3236 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3239 sd->rps_ipi_next = mysd->rps_ipi_list;
3240 mysd->rps_ipi_list = sd;
3242 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3245 #endif /* CONFIG_RPS */
3249 #ifdef CONFIG_NET_FLOW_LIMIT
3250 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3253 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3255 #ifdef CONFIG_NET_FLOW_LIMIT
3256 struct sd_flow_limit *fl;
3257 struct softnet_data *sd;
3258 unsigned int old_flow, new_flow;
3260 if (qlen < (netdev_max_backlog >> 1))
3263 sd = this_cpu_ptr(&softnet_data);
3266 fl = rcu_dereference(sd->flow_limit);
3268 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3269 old_flow = fl->history[fl->history_head];
3270 fl->history[fl->history_head] = new_flow;
3273 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3275 if (likely(fl->buckets[old_flow]))
3276 fl->buckets[old_flow]--;
3278 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3290 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3291 * queue (may be a remote CPU queue).
3293 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3294 unsigned int *qtail)
3296 struct softnet_data *sd;
3297 unsigned long flags;
3300 sd = &per_cpu(softnet_data, cpu);
3302 local_irq_save(flags);
3305 qlen = skb_queue_len(&sd->input_pkt_queue);
3306 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3307 if (skb_queue_len(&sd->input_pkt_queue)) {
3309 __skb_queue_tail(&sd->input_pkt_queue, skb);
3310 input_queue_tail_incr_save(sd, qtail);
3312 local_irq_restore(flags);
3313 return NET_RX_SUCCESS;
3316 /* Schedule NAPI for backlog device
3317 * We can use non atomic operation since we own the queue lock
3319 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3320 if (!rps_ipi_queued(sd))
3321 ____napi_schedule(sd, &sd->backlog);
3329 local_irq_restore(flags);
3331 atomic_long_inc(&skb->dev->rx_dropped);
3336 static int netif_rx_internal(struct sk_buff *skb)
3340 net_timestamp_check(netdev_tstamp_prequeue, skb);
3342 trace_netif_rx(skb);
3344 if (static_key_false(&rps_needed)) {
3345 struct rps_dev_flow voidflow, *rflow = &voidflow;
3351 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3353 cpu = smp_processor_id();
3355 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3363 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3370 * netif_rx - post buffer to the network code
3371 * @skb: buffer to post
3373 * This function receives a packet from a device driver and queues it for
3374 * the upper (protocol) levels to process. It always succeeds. The buffer
3375 * may be dropped during processing for congestion control or by the
3379 * NET_RX_SUCCESS (no congestion)
3380 * NET_RX_DROP (packet was dropped)
3384 int netif_rx(struct sk_buff *skb)
3386 trace_netif_rx_entry(skb);
3388 return netif_rx_internal(skb);
3390 EXPORT_SYMBOL(netif_rx);
3392 int netif_rx_ni(struct sk_buff *skb)
3396 trace_netif_rx_ni_entry(skb);
3399 err = netif_rx_internal(skb);
3400 if (local_softirq_pending())
3406 EXPORT_SYMBOL(netif_rx_ni);
3408 static void net_tx_action(struct softirq_action *h)
3410 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3412 if (sd->completion_queue) {
3413 struct sk_buff *clist;
3415 local_irq_disable();
3416 clist = sd->completion_queue;
3417 sd->completion_queue = NULL;
3421 struct sk_buff *skb = clist;
3422 clist = clist->next;
3424 WARN_ON(atomic_read(&skb->users));
3425 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3426 trace_consume_skb(skb);
3428 trace_kfree_skb(skb, net_tx_action);
3433 if (sd->output_queue) {
3436 local_irq_disable();
3437 head = sd->output_queue;
3438 sd->output_queue = NULL;
3439 sd->output_queue_tailp = &sd->output_queue;
3443 struct Qdisc *q = head;
3444 spinlock_t *root_lock;
3446 head = head->next_sched;
3448 root_lock = qdisc_lock(q);
3449 if (spin_trylock(root_lock)) {
3450 smp_mb__before_atomic();
3451 clear_bit(__QDISC_STATE_SCHED,
3454 spin_unlock(root_lock);
3456 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3458 __netif_reschedule(q);
3460 smp_mb__before_atomic();
3461 clear_bit(__QDISC_STATE_SCHED,
3469 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3470 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3471 /* This hook is defined here for ATM LANE */
3472 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3473 unsigned char *addr) __read_mostly;
3474 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3477 #ifdef CONFIG_NET_CLS_ACT
3478 /* TODO: Maybe we should just force sch_ingress to be compiled in
3479 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3480 * a compare and 2 stores extra right now if we dont have it on
3481 * but have CONFIG_NET_CLS_ACT
3482 * NOTE: This doesn't stop any functionality; if you dont have
3483 * the ingress scheduler, you just can't add policies on ingress.
3486 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3488 struct net_device *dev = skb->dev;
3489 u32 ttl = G_TC_RTTL(skb->tc_verd);
3490 int result = TC_ACT_OK;
3493 if (unlikely(MAX_RED_LOOP < ttl++)) {
3494 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3495 skb->skb_iif, dev->ifindex);
3499 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3500 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3502 q = rcu_dereference(rxq->qdisc);
3503 if (q != &noop_qdisc) {
3504 spin_lock(qdisc_lock(q));
3505 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3506 result = qdisc_enqueue_root(skb, q);
3507 spin_unlock(qdisc_lock(q));
3513 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3514 struct packet_type **pt_prev,
3515 int *ret, struct net_device *orig_dev)
3517 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3519 if (!rxq || rcu_access_pointer(rxq->qdisc) == &noop_qdisc)
3523 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3527 switch (ing_filter(skb, rxq)) {
3541 * netdev_rx_handler_register - register receive handler
3542 * @dev: device to register a handler for
3543 * @rx_handler: receive handler to register
3544 * @rx_handler_data: data pointer that is used by rx handler
3546 * Register a receive handler for a device. This handler will then be
3547 * called from __netif_receive_skb. A negative errno code is returned
3550 * The caller must hold the rtnl_mutex.
3552 * For a general description of rx_handler, see enum rx_handler_result.
3554 int netdev_rx_handler_register(struct net_device *dev,
3555 rx_handler_func_t *rx_handler,
3556 void *rx_handler_data)
3560 if (dev->rx_handler)
3563 /* Note: rx_handler_data must be set before rx_handler */
3564 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3565 rcu_assign_pointer(dev->rx_handler, rx_handler);
3569 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3572 * netdev_rx_handler_unregister - unregister receive handler
3573 * @dev: device to unregister a handler from
3575 * Unregister a receive handler from a device.
3577 * The caller must hold the rtnl_mutex.
3579 void netdev_rx_handler_unregister(struct net_device *dev)
3583 RCU_INIT_POINTER(dev->rx_handler, NULL);
3584 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3585 * section has a guarantee to see a non NULL rx_handler_data
3589 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3591 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3594 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3595 * the special handling of PFMEMALLOC skbs.
3597 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3599 switch (skb->protocol) {
3600 case htons(ETH_P_ARP):
3601 case htons(ETH_P_IP):
3602 case htons(ETH_P_IPV6):
3603 case htons(ETH_P_8021Q):
3604 case htons(ETH_P_8021AD):
3611 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3613 struct packet_type *ptype, *pt_prev;
3614 rx_handler_func_t *rx_handler;
3615 struct net_device *orig_dev;
3616 struct net_device *null_or_dev;
3617 bool deliver_exact = false;
3618 int ret = NET_RX_DROP;
3621 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3623 trace_netif_receive_skb(skb);
3625 orig_dev = skb->dev;
3627 skb_reset_network_header(skb);
3628 if (!skb_transport_header_was_set(skb))
3629 skb_reset_transport_header(skb);
3630 skb_reset_mac_len(skb);
3637 skb->skb_iif = skb->dev->ifindex;
3639 __this_cpu_inc(softnet_data.processed);
3641 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3642 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3643 skb = skb_vlan_untag(skb);
3648 #ifdef CONFIG_NET_CLS_ACT
3649 if (skb->tc_verd & TC_NCLS) {
3650 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3658 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3659 if (!ptype->dev || ptype->dev == skb->dev) {
3661 ret = deliver_skb(skb, pt_prev, orig_dev);
3667 #ifdef CONFIG_NET_CLS_ACT
3668 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3674 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3677 if (vlan_tx_tag_present(skb)) {
3679 ret = deliver_skb(skb, pt_prev, orig_dev);
3682 if (vlan_do_receive(&skb))
3684 else if (unlikely(!skb))
3688 rx_handler = rcu_dereference(skb->dev->rx_handler);
3691 ret = deliver_skb(skb, pt_prev, orig_dev);
3694 switch (rx_handler(&skb)) {
3695 case RX_HANDLER_CONSUMED:
3696 ret = NET_RX_SUCCESS;
3698 case RX_HANDLER_ANOTHER:
3700 case RX_HANDLER_EXACT:
3701 deliver_exact = true;
3702 case RX_HANDLER_PASS:
3709 if (unlikely(vlan_tx_tag_present(skb))) {
3710 if (vlan_tx_tag_get_id(skb))
3711 skb->pkt_type = PACKET_OTHERHOST;
3712 /* Note: we might in the future use prio bits
3713 * and set skb->priority like in vlan_do_receive()
3714 * For the time being, just ignore Priority Code Point
3719 /* deliver only exact match when indicated */
3720 null_or_dev = deliver_exact ? skb->dev : NULL;
3722 type = skb->protocol;
3723 list_for_each_entry_rcu(ptype,
3724 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3725 if (ptype->type == type &&
3726 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3727 ptype->dev == orig_dev)) {
3729 ret = deliver_skb(skb, pt_prev, orig_dev);
3735 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3738 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3741 atomic_long_inc(&skb->dev->rx_dropped);
3743 /* Jamal, now you will not able to escape explaining
3744 * me how you were going to use this. :-)
3754 static int __netif_receive_skb(struct sk_buff *skb)
3758 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3759 unsigned long pflags = current->flags;
3762 * PFMEMALLOC skbs are special, they should
3763 * - be delivered to SOCK_MEMALLOC sockets only
3764 * - stay away from userspace
3765 * - have bounded memory usage
3767 * Use PF_MEMALLOC as this saves us from propagating the allocation
3768 * context down to all allocation sites.
3770 current->flags |= PF_MEMALLOC;
3771 ret = __netif_receive_skb_core(skb, true);
3772 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3774 ret = __netif_receive_skb_core(skb, false);
3779 static int netif_receive_skb_internal(struct sk_buff *skb)
3781 net_timestamp_check(netdev_tstamp_prequeue, skb);
3783 if (skb_defer_rx_timestamp(skb))
3784 return NET_RX_SUCCESS;
3787 if (static_key_false(&rps_needed)) {
3788 struct rps_dev_flow voidflow, *rflow = &voidflow;
3793 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3796 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3803 return __netif_receive_skb(skb);
3807 * netif_receive_skb - process receive buffer from network
3808 * @skb: buffer to process
3810 * netif_receive_skb() is the main receive data processing function.
3811 * It always succeeds. The buffer may be dropped during processing
3812 * for congestion control or by the protocol layers.
3814 * This function may only be called from softirq context and interrupts
3815 * should be enabled.
3817 * Return values (usually ignored):
3818 * NET_RX_SUCCESS: no congestion
3819 * NET_RX_DROP: packet was dropped
3821 int netif_receive_skb(struct sk_buff *skb)
3823 trace_netif_receive_skb_entry(skb);
3825 return netif_receive_skb_internal(skb);
3827 EXPORT_SYMBOL(netif_receive_skb);
3829 /* Network device is going away, flush any packets still pending
3830 * Called with irqs disabled.
3832 static void flush_backlog(void *arg)
3834 struct net_device *dev = arg;
3835 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
3836 struct sk_buff *skb, *tmp;
3839 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3840 if (skb->dev == dev) {
3841 __skb_unlink(skb, &sd->input_pkt_queue);
3843 input_queue_head_incr(sd);
3848 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3849 if (skb->dev == dev) {
3850 __skb_unlink(skb, &sd->process_queue);
3852 input_queue_head_incr(sd);
3857 static int napi_gro_complete(struct sk_buff *skb)
3859 struct packet_offload *ptype;
3860 __be16 type = skb->protocol;
3861 struct list_head *head = &offload_base;
3864 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3866 if (NAPI_GRO_CB(skb)->count == 1) {
3867 skb_shinfo(skb)->gso_size = 0;
3872 list_for_each_entry_rcu(ptype, head, list) {
3873 if (ptype->type != type || !ptype->callbacks.gro_complete)
3876 err = ptype->callbacks.gro_complete(skb, 0);
3882 WARN_ON(&ptype->list == head);
3884 return NET_RX_SUCCESS;
3888 return netif_receive_skb_internal(skb);
3891 /* napi->gro_list contains packets ordered by age.
3892 * youngest packets at the head of it.
3893 * Complete skbs in reverse order to reduce latencies.
3895 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3897 struct sk_buff *skb, *prev = NULL;
3899 /* scan list and build reverse chain */
3900 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3905 for (skb = prev; skb; skb = prev) {
3908 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3912 napi_gro_complete(skb);
3916 napi->gro_list = NULL;
3918 EXPORT_SYMBOL(napi_gro_flush);
3920 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3923 unsigned int maclen = skb->dev->hard_header_len;
3924 u32 hash = skb_get_hash_raw(skb);
3926 for (p = napi->gro_list; p; p = p->next) {
3927 unsigned long diffs;
3929 NAPI_GRO_CB(p)->flush = 0;
3931 if (hash != skb_get_hash_raw(p)) {
3932 NAPI_GRO_CB(p)->same_flow = 0;
3936 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3937 diffs |= p->vlan_tci ^ skb->vlan_tci;
3938 if (maclen == ETH_HLEN)
3939 diffs |= compare_ether_header(skb_mac_header(p),
3940 skb_mac_header(skb));
3942 diffs = memcmp(skb_mac_header(p),
3943 skb_mac_header(skb),
3945 NAPI_GRO_CB(p)->same_flow = !diffs;
3949 static void skb_gro_reset_offset(struct sk_buff *skb)
3951 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3952 const skb_frag_t *frag0 = &pinfo->frags[0];
3954 NAPI_GRO_CB(skb)->data_offset = 0;
3955 NAPI_GRO_CB(skb)->frag0 = NULL;
3956 NAPI_GRO_CB(skb)->frag0_len = 0;
3958 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3960 !PageHighMem(skb_frag_page(frag0))) {
3961 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3962 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3966 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
3968 struct skb_shared_info *pinfo = skb_shinfo(skb);
3970 BUG_ON(skb->end - skb->tail < grow);
3972 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3974 skb->data_len -= grow;
3977 pinfo->frags[0].page_offset += grow;
3978 skb_frag_size_sub(&pinfo->frags[0], grow);
3980 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
3981 skb_frag_unref(skb, 0);
3982 memmove(pinfo->frags, pinfo->frags + 1,
3983 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
3987 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3989 struct sk_buff **pp = NULL;
3990 struct packet_offload *ptype;
3991 __be16 type = skb->protocol;
3992 struct list_head *head = &offload_base;
3994 enum gro_result ret;
3997 if (!(skb->dev->features & NETIF_F_GRO))
4000 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4003 gro_list_prepare(napi, skb);
4006 list_for_each_entry_rcu(ptype, head, list) {
4007 if (ptype->type != type || !ptype->callbacks.gro_receive)
4010 skb_set_network_header(skb, skb_gro_offset(skb));
4011 skb_reset_mac_len(skb);
4012 NAPI_GRO_CB(skb)->same_flow = 0;
4013 NAPI_GRO_CB(skb)->flush = 0;
4014 NAPI_GRO_CB(skb)->free = 0;
4015 NAPI_GRO_CB(skb)->udp_mark = 0;
4017 /* Setup for GRO checksum validation */
4018 switch (skb->ip_summed) {
4019 case CHECKSUM_COMPLETE:
4020 NAPI_GRO_CB(skb)->csum = skb->csum;
4021 NAPI_GRO_CB(skb)->csum_valid = 1;
4022 NAPI_GRO_CB(skb)->csum_cnt = 0;
4024 case CHECKSUM_UNNECESSARY:
4025 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4026 NAPI_GRO_CB(skb)->csum_valid = 0;
4029 NAPI_GRO_CB(skb)->csum_cnt = 0;
4030 NAPI_GRO_CB(skb)->csum_valid = 0;
4033 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4038 if (&ptype->list == head)
4041 same_flow = NAPI_GRO_CB(skb)->same_flow;
4042 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4045 struct sk_buff *nskb = *pp;
4049 napi_gro_complete(nskb);
4056 if (NAPI_GRO_CB(skb)->flush)
4059 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4060 struct sk_buff *nskb = napi->gro_list;
4062 /* locate the end of the list to select the 'oldest' flow */
4063 while (nskb->next) {
4069 napi_gro_complete(nskb);
4073 NAPI_GRO_CB(skb)->count = 1;
4074 NAPI_GRO_CB(skb)->age = jiffies;
4075 NAPI_GRO_CB(skb)->last = skb;
4076 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4077 skb->next = napi->gro_list;
4078 napi->gro_list = skb;
4082 grow = skb_gro_offset(skb) - skb_headlen(skb);
4084 gro_pull_from_frag0(skb, grow);
4093 struct packet_offload *gro_find_receive_by_type(__be16 type)
4095 struct list_head *offload_head = &offload_base;
4096 struct packet_offload *ptype;
4098 list_for_each_entry_rcu(ptype, offload_head, list) {
4099 if (ptype->type != type || !ptype->callbacks.gro_receive)
4105 EXPORT_SYMBOL(gro_find_receive_by_type);
4107 struct packet_offload *gro_find_complete_by_type(__be16 type)
4109 struct list_head *offload_head = &offload_base;
4110 struct packet_offload *ptype;
4112 list_for_each_entry_rcu(ptype, offload_head, list) {
4113 if (ptype->type != type || !ptype->callbacks.gro_complete)
4119 EXPORT_SYMBOL(gro_find_complete_by_type);
4121 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4125 if (netif_receive_skb_internal(skb))
4133 case GRO_MERGED_FREE:
4134 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4135 kmem_cache_free(skbuff_head_cache, skb);
4148 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4150 trace_napi_gro_receive_entry(skb);
4152 skb_gro_reset_offset(skb);
4154 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4156 EXPORT_SYMBOL(napi_gro_receive);
4158 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4160 __skb_pull(skb, skb_headlen(skb));
4161 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4162 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4164 skb->dev = napi->dev;
4166 skb->encapsulation = 0;
4167 skb_shinfo(skb)->gso_type = 0;
4168 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4173 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4175 struct sk_buff *skb = napi->skb;
4178 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4183 EXPORT_SYMBOL(napi_get_frags);
4185 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4186 struct sk_buff *skb,
4192 __skb_push(skb, ETH_HLEN);
4193 skb->protocol = eth_type_trans(skb, skb->dev);
4194 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4199 case GRO_MERGED_FREE:
4200 napi_reuse_skb(napi, skb);
4210 /* Upper GRO stack assumes network header starts at gro_offset=0
4211 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4212 * We copy ethernet header into skb->data to have a common layout.
4214 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4216 struct sk_buff *skb = napi->skb;
4217 const struct ethhdr *eth;
4218 unsigned int hlen = sizeof(*eth);
4222 skb_reset_mac_header(skb);
4223 skb_gro_reset_offset(skb);
4225 eth = skb_gro_header_fast(skb, 0);
4226 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4227 eth = skb_gro_header_slow(skb, hlen, 0);
4228 if (unlikely(!eth)) {
4229 napi_reuse_skb(napi, skb);
4233 gro_pull_from_frag0(skb, hlen);
4234 NAPI_GRO_CB(skb)->frag0 += hlen;
4235 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4237 __skb_pull(skb, hlen);
4240 * This works because the only protocols we care about don't require
4242 * We'll fix it up properly in napi_frags_finish()
4244 skb->protocol = eth->h_proto;
4249 gro_result_t napi_gro_frags(struct napi_struct *napi)
4251 struct sk_buff *skb = napi_frags_skb(napi);
4256 trace_napi_gro_frags_entry(skb);
4258 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4260 EXPORT_SYMBOL(napi_gro_frags);
4262 /* Compute the checksum from gro_offset and return the folded value
4263 * after adding in any pseudo checksum.
4265 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4270 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4272 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4273 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4275 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4276 !skb->csum_complete_sw)
4277 netdev_rx_csum_fault(skb->dev);
4280 NAPI_GRO_CB(skb)->csum = wsum;
4281 NAPI_GRO_CB(skb)->csum_valid = 1;
4285 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4288 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4289 * Note: called with local irq disabled, but exits with local irq enabled.
4291 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4294 struct softnet_data *remsd = sd->rps_ipi_list;
4297 sd->rps_ipi_list = NULL;
4301 /* Send pending IPI's to kick RPS processing on remote cpus. */
4303 struct softnet_data *next = remsd->rps_ipi_next;
4305 if (cpu_online(remsd->cpu))
4306 smp_call_function_single_async(remsd->cpu,
4315 static int process_backlog(struct napi_struct *napi, int quota)
4318 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4321 /* Check if we have pending ipi, its better to send them now,
4322 * not waiting net_rx_action() end.
4324 if (sd->rps_ipi_list) {
4325 local_irq_disable();
4326 net_rps_action_and_irq_enable(sd);
4329 napi->weight = weight_p;
4330 local_irq_disable();
4332 struct sk_buff *skb;
4334 while ((skb = __skb_dequeue(&sd->process_queue))) {
4336 __netif_receive_skb(skb);
4337 local_irq_disable();
4338 input_queue_head_incr(sd);
4339 if (++work >= quota) {
4346 if (skb_queue_empty(&sd->input_pkt_queue)) {
4348 * Inline a custom version of __napi_complete().
4349 * only current cpu owns and manipulates this napi,
4350 * and NAPI_STATE_SCHED is the only possible flag set
4352 * We can use a plain write instead of clear_bit(),
4353 * and we dont need an smp_mb() memory barrier.
4355 list_del(&napi->poll_list);
4362 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4363 &sd->process_queue);
4372 * __napi_schedule - schedule for receive
4373 * @n: entry to schedule
4375 * The entry's receive function will be scheduled to run
4377 void __napi_schedule(struct napi_struct *n)
4379 unsigned long flags;
4381 local_irq_save(flags);
4382 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
4383 local_irq_restore(flags);
4385 EXPORT_SYMBOL(__napi_schedule);
4387 void __napi_complete(struct napi_struct *n)
4389 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4390 BUG_ON(n->gro_list);
4392 list_del(&n->poll_list);
4393 smp_mb__before_atomic();
4394 clear_bit(NAPI_STATE_SCHED, &n->state);
4396 EXPORT_SYMBOL(__napi_complete);
4398 void napi_complete(struct napi_struct *n)
4400 unsigned long flags;
4403 * don't let napi dequeue from the cpu poll list
4404 * just in case its running on a different cpu
4406 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4409 napi_gro_flush(n, false);
4410 local_irq_save(flags);
4412 local_irq_restore(flags);
4414 EXPORT_SYMBOL(napi_complete);
4416 /* must be called under rcu_read_lock(), as we dont take a reference */
4417 struct napi_struct *napi_by_id(unsigned int napi_id)
4419 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4420 struct napi_struct *napi;
4422 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4423 if (napi->napi_id == napi_id)
4428 EXPORT_SYMBOL_GPL(napi_by_id);
4430 void napi_hash_add(struct napi_struct *napi)
4432 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4434 spin_lock(&napi_hash_lock);
4436 /* 0 is not a valid id, we also skip an id that is taken
4437 * we expect both events to be extremely rare
4440 while (!napi->napi_id) {
4441 napi->napi_id = ++napi_gen_id;
4442 if (napi_by_id(napi->napi_id))
4446 hlist_add_head_rcu(&napi->napi_hash_node,
4447 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4449 spin_unlock(&napi_hash_lock);
4452 EXPORT_SYMBOL_GPL(napi_hash_add);
4454 /* Warning : caller is responsible to make sure rcu grace period
4455 * is respected before freeing memory containing @napi
4457 void napi_hash_del(struct napi_struct *napi)
4459 spin_lock(&napi_hash_lock);
4461 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4462 hlist_del_rcu(&napi->napi_hash_node);
4464 spin_unlock(&napi_hash_lock);
4466 EXPORT_SYMBOL_GPL(napi_hash_del);
4468 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4469 int (*poll)(struct napi_struct *, int), int weight)
4471 INIT_LIST_HEAD(&napi->poll_list);
4472 napi->gro_count = 0;
4473 napi->gro_list = NULL;
4476 if (weight > NAPI_POLL_WEIGHT)
4477 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4479 napi->weight = weight;
4480 list_add(&napi->dev_list, &dev->napi_list);
4482 #ifdef CONFIG_NETPOLL
4483 spin_lock_init(&napi->poll_lock);
4484 napi->poll_owner = -1;
4486 set_bit(NAPI_STATE_SCHED, &napi->state);
4488 EXPORT_SYMBOL(netif_napi_add);
4490 void netif_napi_del(struct napi_struct *napi)
4492 list_del_init(&napi->dev_list);
4493 napi_free_frags(napi);
4495 kfree_skb_list(napi->gro_list);
4496 napi->gro_list = NULL;
4497 napi->gro_count = 0;
4499 EXPORT_SYMBOL(netif_napi_del);
4501 static void net_rx_action(struct softirq_action *h)
4503 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4504 unsigned long time_limit = jiffies + 2;
4505 int budget = netdev_budget;
4508 local_irq_disable();
4510 while (!list_empty(&sd->poll_list)) {
4511 struct napi_struct *n;
4514 /* If softirq window is exhuasted then punt.
4515 * Allow this to run for 2 jiffies since which will allow
4516 * an average latency of 1.5/HZ.
4518 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4523 /* Even though interrupts have been re-enabled, this
4524 * access is safe because interrupts can only add new
4525 * entries to the tail of this list, and only ->poll()
4526 * calls can remove this head entry from the list.
4528 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4530 have = netpoll_poll_lock(n);
4534 /* This NAPI_STATE_SCHED test is for avoiding a race
4535 * with netpoll's poll_napi(). Only the entity which
4536 * obtains the lock and sees NAPI_STATE_SCHED set will
4537 * actually make the ->poll() call. Therefore we avoid
4538 * accidentally calling ->poll() when NAPI is not scheduled.
4541 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4542 work = n->poll(n, weight);
4546 WARN_ON_ONCE(work > weight);
4550 local_irq_disable();
4552 /* Drivers must not modify the NAPI state if they
4553 * consume the entire weight. In such cases this code
4554 * still "owns" the NAPI instance and therefore can
4555 * move the instance around on the list at-will.
4557 if (unlikely(work == weight)) {
4558 if (unlikely(napi_disable_pending(n))) {
4561 local_irq_disable();
4564 /* flush too old packets
4565 * If HZ < 1000, flush all packets.
4568 napi_gro_flush(n, HZ >= 1000);
4569 local_irq_disable();
4571 list_move_tail(&n->poll_list, &sd->poll_list);
4575 netpoll_poll_unlock(have);
4578 net_rps_action_and_irq_enable(sd);
4584 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4588 struct netdev_adjacent {
4589 struct net_device *dev;
4591 /* upper master flag, there can only be one master device per list */
4594 /* counter for the number of times this device was added to us */
4597 /* private field for the users */
4600 struct list_head list;
4601 struct rcu_head rcu;
4604 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4605 struct net_device *adj_dev,
4606 struct list_head *adj_list)
4608 struct netdev_adjacent *adj;
4610 list_for_each_entry(adj, adj_list, list) {
4611 if (adj->dev == adj_dev)
4618 * netdev_has_upper_dev - Check if device is linked to an upper device
4620 * @upper_dev: upper device to check
4622 * Find out if a device is linked to specified upper device and return true
4623 * in case it is. Note that this checks only immediate upper device,
4624 * not through a complete stack of devices. The caller must hold the RTNL lock.
4626 bool netdev_has_upper_dev(struct net_device *dev,
4627 struct net_device *upper_dev)
4631 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4633 EXPORT_SYMBOL(netdev_has_upper_dev);
4636 * netdev_has_any_upper_dev - Check if device is linked to some device
4639 * Find out if a device is linked to an upper device and return true in case
4640 * it is. The caller must hold the RTNL lock.
4642 static bool netdev_has_any_upper_dev(struct net_device *dev)
4646 return !list_empty(&dev->all_adj_list.upper);
4650 * netdev_master_upper_dev_get - Get master upper device
4653 * Find a master upper device and return pointer to it or NULL in case
4654 * it's not there. The caller must hold the RTNL lock.
4656 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4658 struct netdev_adjacent *upper;
4662 if (list_empty(&dev->adj_list.upper))
4665 upper = list_first_entry(&dev->adj_list.upper,
4666 struct netdev_adjacent, list);
4667 if (likely(upper->master))
4671 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4673 void *netdev_adjacent_get_private(struct list_head *adj_list)
4675 struct netdev_adjacent *adj;
4677 adj = list_entry(adj_list, struct netdev_adjacent, list);
4679 return adj->private;
4681 EXPORT_SYMBOL(netdev_adjacent_get_private);
4684 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4686 * @iter: list_head ** of the current position
4688 * Gets the next device from the dev's upper list, starting from iter
4689 * position. The caller must hold RCU read lock.
4691 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4692 struct list_head **iter)
4694 struct netdev_adjacent *upper;
4696 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4698 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4700 if (&upper->list == &dev->adj_list.upper)
4703 *iter = &upper->list;
4707 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4710 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4712 * @iter: list_head ** of the current position
4714 * Gets the next device from the dev's upper list, starting from iter
4715 * position. The caller must hold RCU read lock.
4717 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4718 struct list_head **iter)
4720 struct netdev_adjacent *upper;
4722 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4724 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4726 if (&upper->list == &dev->all_adj_list.upper)
4729 *iter = &upper->list;
4733 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4736 * netdev_lower_get_next_private - Get the next ->private from the
4737 * lower neighbour list
4739 * @iter: list_head ** of the current position
4741 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4742 * list, starting from iter position. The caller must hold either hold the
4743 * RTNL lock or its own locking that guarantees that the neighbour lower
4744 * list will remain unchainged.
4746 void *netdev_lower_get_next_private(struct net_device *dev,
4747 struct list_head **iter)
4749 struct netdev_adjacent *lower;
4751 lower = list_entry(*iter, struct netdev_adjacent, list);
4753 if (&lower->list == &dev->adj_list.lower)
4756 *iter = lower->list.next;
4758 return lower->private;
4760 EXPORT_SYMBOL(netdev_lower_get_next_private);
4763 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4764 * lower neighbour list, RCU
4767 * @iter: list_head ** of the current position
4769 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4770 * list, starting from iter position. The caller must hold RCU read lock.
4772 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4773 struct list_head **iter)
4775 struct netdev_adjacent *lower;
4777 WARN_ON_ONCE(!rcu_read_lock_held());
4779 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4781 if (&lower->list == &dev->adj_list.lower)
4784 *iter = &lower->list;
4786 return lower->private;
4788 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4791 * netdev_lower_get_next - Get the next device from the lower neighbour
4794 * @iter: list_head ** of the current position
4796 * Gets the next netdev_adjacent from the dev's lower neighbour
4797 * list, starting from iter position. The caller must hold RTNL lock or
4798 * its own locking that guarantees that the neighbour lower
4799 * list will remain unchainged.
4801 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4803 struct netdev_adjacent *lower;
4805 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4807 if (&lower->list == &dev->adj_list.lower)
4810 *iter = &lower->list;
4814 EXPORT_SYMBOL(netdev_lower_get_next);
4817 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4818 * lower neighbour list, RCU
4822 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4823 * list. The caller must hold RCU read lock.
4825 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4827 struct netdev_adjacent *lower;
4829 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4830 struct netdev_adjacent, list);
4832 return lower->private;
4835 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4838 * netdev_master_upper_dev_get_rcu - Get master upper device
4841 * Find a master upper device and return pointer to it or NULL in case
4842 * it's not there. The caller must hold the RCU read lock.
4844 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4846 struct netdev_adjacent *upper;
4848 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4849 struct netdev_adjacent, list);
4850 if (upper && likely(upper->master))
4854 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4856 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4857 struct net_device *adj_dev,
4858 struct list_head *dev_list)
4860 char linkname[IFNAMSIZ+7];
4861 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4862 "upper_%s" : "lower_%s", adj_dev->name);
4863 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4866 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4868 struct list_head *dev_list)
4870 char linkname[IFNAMSIZ+7];
4871 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4872 "upper_%s" : "lower_%s", name);
4873 sysfs_remove_link(&(dev->dev.kobj), linkname);
4876 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
4877 struct net_device *adj_dev,
4878 struct list_head *dev_list)
4880 return (dev_list == &dev->adj_list.upper ||
4881 dev_list == &dev->adj_list.lower) &&
4882 net_eq(dev_net(dev), dev_net(adj_dev));
4885 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4886 struct net_device *adj_dev,
4887 struct list_head *dev_list,
4888 void *private, bool master)
4890 struct netdev_adjacent *adj;
4893 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4900 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4905 adj->master = master;
4907 adj->private = private;
4910 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4911 adj_dev->name, dev->name, adj_dev->name);
4913 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
4914 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
4919 /* Ensure that master link is always the first item in list. */
4921 ret = sysfs_create_link(&(dev->dev.kobj),
4922 &(adj_dev->dev.kobj), "master");
4924 goto remove_symlinks;
4926 list_add_rcu(&adj->list, dev_list);
4928 list_add_tail_rcu(&adj->list, dev_list);
4934 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
4935 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4943 static void __netdev_adjacent_dev_remove(struct net_device *dev,
4944 struct net_device *adj_dev,
4945 struct list_head *dev_list)
4947 struct netdev_adjacent *adj;
4949 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4952 pr_err("tried to remove device %s from %s\n",
4953 dev->name, adj_dev->name);
4957 if (adj->ref_nr > 1) {
4958 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4965 sysfs_remove_link(&(dev->dev.kobj), "master");
4967 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
4968 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4970 list_del_rcu(&adj->list);
4971 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4972 adj_dev->name, dev->name, adj_dev->name);
4974 kfree_rcu(adj, rcu);
4977 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4978 struct net_device *upper_dev,
4979 struct list_head *up_list,
4980 struct list_head *down_list,
4981 void *private, bool master)
4985 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4990 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4993 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5000 static int __netdev_adjacent_dev_link(struct net_device *dev,
5001 struct net_device *upper_dev)
5003 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5004 &dev->all_adj_list.upper,
5005 &upper_dev->all_adj_list.lower,
5009 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5010 struct net_device *upper_dev,
5011 struct list_head *up_list,
5012 struct list_head *down_list)
5014 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5015 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5018 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5019 struct net_device *upper_dev)
5021 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5022 &dev->all_adj_list.upper,
5023 &upper_dev->all_adj_list.lower);
5026 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5027 struct net_device *upper_dev,
5028 void *private, bool master)
5030 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5035 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5036 &dev->adj_list.upper,
5037 &upper_dev->adj_list.lower,
5040 __netdev_adjacent_dev_unlink(dev, upper_dev);
5047 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5048 struct net_device *upper_dev)
5050 __netdev_adjacent_dev_unlink(dev, upper_dev);
5051 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5052 &dev->adj_list.upper,
5053 &upper_dev->adj_list.lower);
5056 static int __netdev_upper_dev_link(struct net_device *dev,
5057 struct net_device *upper_dev, bool master,
5060 struct netdev_adjacent *i, *j, *to_i, *to_j;
5065 if (dev == upper_dev)
5068 /* To prevent loops, check if dev is not upper device to upper_dev. */
5069 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5072 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
5075 if (master && netdev_master_upper_dev_get(dev))
5078 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5083 /* Now that we linked these devs, make all the upper_dev's
5084 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5085 * versa, and don't forget the devices itself. All of these
5086 * links are non-neighbours.
5088 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5089 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5090 pr_debug("Interlinking %s with %s, non-neighbour\n",
5091 i->dev->name, j->dev->name);
5092 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5098 /* add dev to every upper_dev's upper device */
5099 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5100 pr_debug("linking %s's upper device %s with %s\n",
5101 upper_dev->name, i->dev->name, dev->name);
5102 ret = __netdev_adjacent_dev_link(dev, i->dev);
5104 goto rollback_upper_mesh;
5107 /* add upper_dev to every dev's lower device */
5108 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5109 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5110 i->dev->name, upper_dev->name);
5111 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5113 goto rollback_lower_mesh;
5116 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5119 rollback_lower_mesh:
5121 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5124 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5129 rollback_upper_mesh:
5131 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5134 __netdev_adjacent_dev_unlink(dev, i->dev);
5142 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5143 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5144 if (i == to_i && j == to_j)
5146 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5152 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5158 * netdev_upper_dev_link - Add a link to the upper device
5160 * @upper_dev: new upper device
5162 * Adds a link to device which is upper to this one. The caller must hold
5163 * the RTNL lock. On a failure a negative errno code is returned.
5164 * On success the reference counts are adjusted and the function
5167 int netdev_upper_dev_link(struct net_device *dev,
5168 struct net_device *upper_dev)
5170 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5172 EXPORT_SYMBOL(netdev_upper_dev_link);
5175 * netdev_master_upper_dev_link - Add a master link to the upper device
5177 * @upper_dev: new upper device
5179 * Adds a link to device which is upper to this one. In this case, only
5180 * one master upper device can be linked, although other non-master devices
5181 * might be linked as well. The caller must hold the RTNL lock.
5182 * On a failure a negative errno code is returned. On success the reference
5183 * counts are adjusted and the function returns zero.
5185 int netdev_master_upper_dev_link(struct net_device *dev,
5186 struct net_device *upper_dev)
5188 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5190 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5192 int netdev_master_upper_dev_link_private(struct net_device *dev,
5193 struct net_device *upper_dev,
5196 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5198 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5201 * netdev_upper_dev_unlink - Removes a link to upper device
5203 * @upper_dev: new upper device
5205 * Removes a link to device which is upper to this one. The caller must hold
5208 void netdev_upper_dev_unlink(struct net_device *dev,
5209 struct net_device *upper_dev)
5211 struct netdev_adjacent *i, *j;
5214 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5216 /* Here is the tricky part. We must remove all dev's lower
5217 * devices from all upper_dev's upper devices and vice
5218 * versa, to maintain the graph relationship.
5220 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5221 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5222 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5224 /* remove also the devices itself from lower/upper device
5227 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5228 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5230 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5231 __netdev_adjacent_dev_unlink(dev, i->dev);
5233 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5235 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5237 void netdev_adjacent_add_links(struct net_device *dev)
5239 struct netdev_adjacent *iter;
5241 struct net *net = dev_net(dev);
5243 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5244 if (!net_eq(net,dev_net(iter->dev)))
5246 netdev_adjacent_sysfs_add(iter->dev, dev,
5247 &iter->dev->adj_list.lower);
5248 netdev_adjacent_sysfs_add(dev, iter->dev,
5249 &dev->adj_list.upper);
5252 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5253 if (!net_eq(net,dev_net(iter->dev)))
5255 netdev_adjacent_sysfs_add(iter->dev, dev,
5256 &iter->dev->adj_list.upper);
5257 netdev_adjacent_sysfs_add(dev, iter->dev,
5258 &dev->adj_list.lower);
5262 void netdev_adjacent_del_links(struct net_device *dev)
5264 struct netdev_adjacent *iter;
5266 struct net *net = dev_net(dev);
5268 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5269 if (!net_eq(net,dev_net(iter->dev)))
5271 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5272 &iter->dev->adj_list.lower);
5273 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5274 &dev->adj_list.upper);
5277 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5278 if (!net_eq(net,dev_net(iter->dev)))
5280 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5281 &iter->dev->adj_list.upper);
5282 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5283 &dev->adj_list.lower);
5287 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5289 struct netdev_adjacent *iter;
5291 struct net *net = dev_net(dev);
5293 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5294 if (!net_eq(net,dev_net(iter->dev)))
5296 netdev_adjacent_sysfs_del(iter->dev, oldname,
5297 &iter->dev->adj_list.lower);
5298 netdev_adjacent_sysfs_add(iter->dev, dev,
5299 &iter->dev->adj_list.lower);
5302 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5303 if (!net_eq(net,dev_net(iter->dev)))
5305 netdev_adjacent_sysfs_del(iter->dev, oldname,
5306 &iter->dev->adj_list.upper);
5307 netdev_adjacent_sysfs_add(iter->dev, dev,
5308 &iter->dev->adj_list.upper);
5312 void *netdev_lower_dev_get_private(struct net_device *dev,
5313 struct net_device *lower_dev)
5315 struct netdev_adjacent *lower;
5319 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5323 return lower->private;
5325 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5328 int dev_get_nest_level(struct net_device *dev,
5329 bool (*type_check)(struct net_device *dev))
5331 struct net_device *lower = NULL;
5332 struct list_head *iter;
5338 netdev_for_each_lower_dev(dev, lower, iter) {
5339 nest = dev_get_nest_level(lower, type_check);
5340 if (max_nest < nest)
5344 if (type_check(dev))
5349 EXPORT_SYMBOL(dev_get_nest_level);
5351 static void dev_change_rx_flags(struct net_device *dev, int flags)
5353 const struct net_device_ops *ops = dev->netdev_ops;
5355 if (ops->ndo_change_rx_flags)
5356 ops->ndo_change_rx_flags(dev, flags);
5359 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5361 unsigned int old_flags = dev->flags;
5367 dev->flags |= IFF_PROMISC;
5368 dev->promiscuity += inc;
5369 if (dev->promiscuity == 0) {
5372 * If inc causes overflow, untouch promisc and return error.
5375 dev->flags &= ~IFF_PROMISC;
5377 dev->promiscuity -= inc;
5378 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5383 if (dev->flags != old_flags) {
5384 pr_info("device %s %s promiscuous mode\n",
5386 dev->flags & IFF_PROMISC ? "entered" : "left");
5387 if (audit_enabled) {
5388 current_uid_gid(&uid, &gid);
5389 audit_log(current->audit_context, GFP_ATOMIC,
5390 AUDIT_ANOM_PROMISCUOUS,
5391 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5392 dev->name, (dev->flags & IFF_PROMISC),
5393 (old_flags & IFF_PROMISC),
5394 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5395 from_kuid(&init_user_ns, uid),
5396 from_kgid(&init_user_ns, gid),
5397 audit_get_sessionid(current));
5400 dev_change_rx_flags(dev, IFF_PROMISC);
5403 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5408 * dev_set_promiscuity - update promiscuity count on a device
5412 * Add or remove promiscuity from a device. While the count in the device
5413 * remains above zero the interface remains promiscuous. Once it hits zero
5414 * the device reverts back to normal filtering operation. A negative inc
5415 * value is used to drop promiscuity on the device.
5416 * Return 0 if successful or a negative errno code on error.
5418 int dev_set_promiscuity(struct net_device *dev, int inc)
5420 unsigned int old_flags = dev->flags;
5423 err = __dev_set_promiscuity(dev, inc, true);
5426 if (dev->flags != old_flags)
5427 dev_set_rx_mode(dev);
5430 EXPORT_SYMBOL(dev_set_promiscuity);
5432 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5434 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5438 dev->flags |= IFF_ALLMULTI;
5439 dev->allmulti += inc;
5440 if (dev->allmulti == 0) {
5443 * If inc causes overflow, untouch allmulti and return error.
5446 dev->flags &= ~IFF_ALLMULTI;
5448 dev->allmulti -= inc;
5449 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5454 if (dev->flags ^ old_flags) {
5455 dev_change_rx_flags(dev, IFF_ALLMULTI);
5456 dev_set_rx_mode(dev);
5458 __dev_notify_flags(dev, old_flags,
5459 dev->gflags ^ old_gflags);
5465 * dev_set_allmulti - update allmulti count on a device
5469 * Add or remove reception of all multicast frames to a device. While the
5470 * count in the device remains above zero the interface remains listening
5471 * to all interfaces. Once it hits zero the device reverts back to normal
5472 * filtering operation. A negative @inc value is used to drop the counter
5473 * when releasing a resource needing all multicasts.
5474 * Return 0 if successful or a negative errno code on error.
5477 int dev_set_allmulti(struct net_device *dev, int inc)
5479 return __dev_set_allmulti(dev, inc, true);
5481 EXPORT_SYMBOL(dev_set_allmulti);
5484 * Upload unicast and multicast address lists to device and
5485 * configure RX filtering. When the device doesn't support unicast
5486 * filtering it is put in promiscuous mode while unicast addresses
5489 void __dev_set_rx_mode(struct net_device *dev)
5491 const struct net_device_ops *ops = dev->netdev_ops;
5493 /* dev_open will call this function so the list will stay sane. */
5494 if (!(dev->flags&IFF_UP))
5497 if (!netif_device_present(dev))
5500 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5501 /* Unicast addresses changes may only happen under the rtnl,
5502 * therefore calling __dev_set_promiscuity here is safe.
5504 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5505 __dev_set_promiscuity(dev, 1, false);
5506 dev->uc_promisc = true;
5507 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5508 __dev_set_promiscuity(dev, -1, false);
5509 dev->uc_promisc = false;
5513 if (ops->ndo_set_rx_mode)
5514 ops->ndo_set_rx_mode(dev);
5517 void dev_set_rx_mode(struct net_device *dev)
5519 netif_addr_lock_bh(dev);
5520 __dev_set_rx_mode(dev);
5521 netif_addr_unlock_bh(dev);
5525 * dev_get_flags - get flags reported to userspace
5528 * Get the combination of flag bits exported through APIs to userspace.
5530 unsigned int dev_get_flags(const struct net_device *dev)
5534 flags = (dev->flags & ~(IFF_PROMISC |
5539 (dev->gflags & (IFF_PROMISC |
5542 if (netif_running(dev)) {
5543 if (netif_oper_up(dev))
5544 flags |= IFF_RUNNING;
5545 if (netif_carrier_ok(dev))
5546 flags |= IFF_LOWER_UP;
5547 if (netif_dormant(dev))
5548 flags |= IFF_DORMANT;
5553 EXPORT_SYMBOL(dev_get_flags);
5555 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5557 unsigned int old_flags = dev->flags;
5563 * Set the flags on our device.
5566 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5567 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5569 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5573 * Load in the correct multicast list now the flags have changed.
5576 if ((old_flags ^ flags) & IFF_MULTICAST)
5577 dev_change_rx_flags(dev, IFF_MULTICAST);
5579 dev_set_rx_mode(dev);
5582 * Have we downed the interface. We handle IFF_UP ourselves
5583 * according to user attempts to set it, rather than blindly
5588 if ((old_flags ^ flags) & IFF_UP)
5589 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5591 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5592 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5593 unsigned int old_flags = dev->flags;
5595 dev->gflags ^= IFF_PROMISC;
5597 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5598 if (dev->flags != old_flags)
5599 dev_set_rx_mode(dev);
5602 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5603 is important. Some (broken) drivers set IFF_PROMISC, when
5604 IFF_ALLMULTI is requested not asking us and not reporting.
5606 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5607 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5609 dev->gflags ^= IFF_ALLMULTI;
5610 __dev_set_allmulti(dev, inc, false);
5616 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5617 unsigned int gchanges)
5619 unsigned int changes = dev->flags ^ old_flags;
5622 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5624 if (changes & IFF_UP) {
5625 if (dev->flags & IFF_UP)
5626 call_netdevice_notifiers(NETDEV_UP, dev);
5628 call_netdevice_notifiers(NETDEV_DOWN, dev);
5631 if (dev->flags & IFF_UP &&
5632 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5633 struct netdev_notifier_change_info change_info;
5635 change_info.flags_changed = changes;
5636 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5642 * dev_change_flags - change device settings
5644 * @flags: device state flags
5646 * Change settings on device based state flags. The flags are
5647 * in the userspace exported format.
5649 int dev_change_flags(struct net_device *dev, unsigned int flags)
5652 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5654 ret = __dev_change_flags(dev, flags);
5658 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5659 __dev_notify_flags(dev, old_flags, changes);
5662 EXPORT_SYMBOL(dev_change_flags);
5664 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5666 const struct net_device_ops *ops = dev->netdev_ops;
5668 if (ops->ndo_change_mtu)
5669 return ops->ndo_change_mtu(dev, new_mtu);
5676 * dev_set_mtu - Change maximum transfer unit
5678 * @new_mtu: new transfer unit
5680 * Change the maximum transfer size of the network device.
5682 int dev_set_mtu(struct net_device *dev, int new_mtu)
5686 if (new_mtu == dev->mtu)
5689 /* MTU must be positive. */
5693 if (!netif_device_present(dev))
5696 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5697 err = notifier_to_errno(err);
5701 orig_mtu = dev->mtu;
5702 err = __dev_set_mtu(dev, new_mtu);
5705 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5706 err = notifier_to_errno(err);
5708 /* setting mtu back and notifying everyone again,
5709 * so that they have a chance to revert changes.
5711 __dev_set_mtu(dev, orig_mtu);
5712 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5717 EXPORT_SYMBOL(dev_set_mtu);
5720 * dev_set_group - Change group this device belongs to
5722 * @new_group: group this device should belong to
5724 void dev_set_group(struct net_device *dev, int new_group)
5726 dev->group = new_group;
5728 EXPORT_SYMBOL(dev_set_group);
5731 * dev_set_mac_address - Change Media Access Control Address
5735 * Change the hardware (MAC) address of the device
5737 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5739 const struct net_device_ops *ops = dev->netdev_ops;
5742 if (!ops->ndo_set_mac_address)
5744 if (sa->sa_family != dev->type)
5746 if (!netif_device_present(dev))
5748 err = ops->ndo_set_mac_address(dev, sa);
5751 dev->addr_assign_type = NET_ADDR_SET;
5752 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5753 add_device_randomness(dev->dev_addr, dev->addr_len);
5756 EXPORT_SYMBOL(dev_set_mac_address);
5759 * dev_change_carrier - Change device carrier
5761 * @new_carrier: new value
5763 * Change device carrier
5765 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5767 const struct net_device_ops *ops = dev->netdev_ops;
5769 if (!ops->ndo_change_carrier)
5771 if (!netif_device_present(dev))
5773 return ops->ndo_change_carrier(dev, new_carrier);
5775 EXPORT_SYMBOL(dev_change_carrier);
5778 * dev_get_phys_port_id - Get device physical port ID
5782 * Get device physical port ID
5784 int dev_get_phys_port_id(struct net_device *dev,
5785 struct netdev_phys_port_id *ppid)
5787 const struct net_device_ops *ops = dev->netdev_ops;
5789 if (!ops->ndo_get_phys_port_id)
5791 return ops->ndo_get_phys_port_id(dev, ppid);
5793 EXPORT_SYMBOL(dev_get_phys_port_id);
5796 * dev_new_index - allocate an ifindex
5797 * @net: the applicable net namespace
5799 * Returns a suitable unique value for a new device interface
5800 * number. The caller must hold the rtnl semaphore or the
5801 * dev_base_lock to be sure it remains unique.
5803 static int dev_new_index(struct net *net)
5805 int ifindex = net->ifindex;
5809 if (!__dev_get_by_index(net, ifindex))
5810 return net->ifindex = ifindex;
5814 /* Delayed registration/unregisteration */
5815 static LIST_HEAD(net_todo_list);
5816 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5818 static void net_set_todo(struct net_device *dev)
5820 list_add_tail(&dev->todo_list, &net_todo_list);
5821 dev_net(dev)->dev_unreg_count++;
5824 static void rollback_registered_many(struct list_head *head)
5826 struct net_device *dev, *tmp;
5827 LIST_HEAD(close_head);
5829 BUG_ON(dev_boot_phase);
5832 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5833 /* Some devices call without registering
5834 * for initialization unwind. Remove those
5835 * devices and proceed with the remaining.
5837 if (dev->reg_state == NETREG_UNINITIALIZED) {
5838 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5842 list_del(&dev->unreg_list);
5845 dev->dismantle = true;
5846 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5849 /* If device is running, close it first. */
5850 list_for_each_entry(dev, head, unreg_list)
5851 list_add_tail(&dev->close_list, &close_head);
5852 dev_close_many(&close_head);
5854 list_for_each_entry(dev, head, unreg_list) {
5855 /* And unlink it from device chain. */
5856 unlist_netdevice(dev);
5858 dev->reg_state = NETREG_UNREGISTERING;
5863 list_for_each_entry(dev, head, unreg_list) {
5864 /* Shutdown queueing discipline. */
5868 /* Notify protocols, that we are about to destroy
5869 this device. They should clean all the things.
5871 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5874 * Flush the unicast and multicast chains
5879 if (dev->netdev_ops->ndo_uninit)
5880 dev->netdev_ops->ndo_uninit(dev);
5882 if (!dev->rtnl_link_ops ||
5883 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5884 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5886 /* Notifier chain MUST detach us all upper devices. */
5887 WARN_ON(netdev_has_any_upper_dev(dev));
5889 /* Remove entries from kobject tree */
5890 netdev_unregister_kobject(dev);
5892 /* Remove XPS queueing entries */
5893 netif_reset_xps_queues_gt(dev, 0);
5899 list_for_each_entry(dev, head, unreg_list)
5903 static void rollback_registered(struct net_device *dev)
5907 list_add(&dev->unreg_list, &single);
5908 rollback_registered_many(&single);
5912 static netdev_features_t netdev_fix_features(struct net_device *dev,
5913 netdev_features_t features)
5915 /* Fix illegal checksum combinations */
5916 if ((features & NETIF_F_HW_CSUM) &&
5917 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5918 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5919 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5922 /* TSO requires that SG is present as well. */
5923 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5924 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5925 features &= ~NETIF_F_ALL_TSO;
5928 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5929 !(features & NETIF_F_IP_CSUM)) {
5930 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5931 features &= ~NETIF_F_TSO;
5932 features &= ~NETIF_F_TSO_ECN;
5935 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5936 !(features & NETIF_F_IPV6_CSUM)) {
5937 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5938 features &= ~NETIF_F_TSO6;
5941 /* TSO ECN requires that TSO is present as well. */
5942 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5943 features &= ~NETIF_F_TSO_ECN;
5945 /* Software GSO depends on SG. */
5946 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5947 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5948 features &= ~NETIF_F_GSO;
5951 /* UFO needs SG and checksumming */
5952 if (features & NETIF_F_UFO) {
5953 /* maybe split UFO into V4 and V6? */
5954 if (!((features & NETIF_F_GEN_CSUM) ||
5955 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5956 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5958 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5959 features &= ~NETIF_F_UFO;
5962 if (!(features & NETIF_F_SG)) {
5964 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5965 features &= ~NETIF_F_UFO;
5969 #ifdef CONFIG_NET_RX_BUSY_POLL
5970 if (dev->netdev_ops->ndo_busy_poll)
5971 features |= NETIF_F_BUSY_POLL;
5974 features &= ~NETIF_F_BUSY_POLL;
5979 int __netdev_update_features(struct net_device *dev)
5981 netdev_features_t features;
5986 features = netdev_get_wanted_features(dev);
5988 if (dev->netdev_ops->ndo_fix_features)
5989 features = dev->netdev_ops->ndo_fix_features(dev, features);
5991 /* driver might be less strict about feature dependencies */
5992 features = netdev_fix_features(dev, features);
5994 if (dev->features == features)
5997 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5998 &dev->features, &features);
6000 if (dev->netdev_ops->ndo_set_features)
6001 err = dev->netdev_ops->ndo_set_features(dev, features);
6003 if (unlikely(err < 0)) {
6005 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6006 err, &features, &dev->features);
6011 dev->features = features;
6017 * netdev_update_features - recalculate device features
6018 * @dev: the device to check
6020 * Recalculate dev->features set and send notifications if it
6021 * has changed. Should be called after driver or hardware dependent
6022 * conditions might have changed that influence the features.
6024 void netdev_update_features(struct net_device *dev)
6026 if (__netdev_update_features(dev))
6027 netdev_features_change(dev);
6029 EXPORT_SYMBOL(netdev_update_features);
6032 * netdev_change_features - recalculate device features
6033 * @dev: the device to check
6035 * Recalculate dev->features set and send notifications even
6036 * if they have not changed. Should be called instead of
6037 * netdev_update_features() if also dev->vlan_features might
6038 * have changed to allow the changes to be propagated to stacked
6041 void netdev_change_features(struct net_device *dev)
6043 __netdev_update_features(dev);
6044 netdev_features_change(dev);
6046 EXPORT_SYMBOL(netdev_change_features);
6049 * netif_stacked_transfer_operstate - transfer operstate
6050 * @rootdev: the root or lower level device to transfer state from
6051 * @dev: the device to transfer operstate to
6053 * Transfer operational state from root to device. This is normally
6054 * called when a stacking relationship exists between the root
6055 * device and the device(a leaf device).
6057 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6058 struct net_device *dev)
6060 if (rootdev->operstate == IF_OPER_DORMANT)
6061 netif_dormant_on(dev);
6063 netif_dormant_off(dev);
6065 if (netif_carrier_ok(rootdev)) {
6066 if (!netif_carrier_ok(dev))
6067 netif_carrier_on(dev);
6069 if (netif_carrier_ok(dev))
6070 netif_carrier_off(dev);
6073 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6076 static int netif_alloc_rx_queues(struct net_device *dev)
6078 unsigned int i, count = dev->num_rx_queues;
6079 struct netdev_rx_queue *rx;
6083 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
6089 for (i = 0; i < count; i++)
6095 static void netdev_init_one_queue(struct net_device *dev,
6096 struct netdev_queue *queue, void *_unused)
6098 /* Initialize queue lock */
6099 spin_lock_init(&queue->_xmit_lock);
6100 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6101 queue->xmit_lock_owner = -1;
6102 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6105 dql_init(&queue->dql, HZ);
6109 static void netif_free_tx_queues(struct net_device *dev)
6114 static int netif_alloc_netdev_queues(struct net_device *dev)
6116 unsigned int count = dev->num_tx_queues;
6117 struct netdev_queue *tx;
6118 size_t sz = count * sizeof(*tx);
6120 BUG_ON(count < 1 || count > 0xffff);
6122 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6130 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6131 spin_lock_init(&dev->tx_global_lock);
6137 * register_netdevice - register a network device
6138 * @dev: device to register
6140 * Take a completed network device structure and add it to the kernel
6141 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6142 * chain. 0 is returned on success. A negative errno code is returned
6143 * on a failure to set up the device, or if the name is a duplicate.
6145 * Callers must hold the rtnl semaphore. You may want
6146 * register_netdev() instead of this.
6149 * The locking appears insufficient to guarantee two parallel registers
6150 * will not get the same name.
6153 int register_netdevice(struct net_device *dev)
6156 struct net *net = dev_net(dev);
6158 BUG_ON(dev_boot_phase);
6163 /* When net_device's are persistent, this will be fatal. */
6164 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6167 spin_lock_init(&dev->addr_list_lock);
6168 netdev_set_addr_lockdep_class(dev);
6172 ret = dev_get_valid_name(net, dev, dev->name);
6176 /* Init, if this function is available */
6177 if (dev->netdev_ops->ndo_init) {
6178 ret = dev->netdev_ops->ndo_init(dev);
6186 if (((dev->hw_features | dev->features) &
6187 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6188 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6189 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6190 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6197 dev->ifindex = dev_new_index(net);
6198 else if (__dev_get_by_index(net, dev->ifindex))
6201 if (dev->iflink == -1)
6202 dev->iflink = dev->ifindex;
6204 /* Transfer changeable features to wanted_features and enable
6205 * software offloads (GSO and GRO).
6207 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6208 dev->features |= NETIF_F_SOFT_FEATURES;
6209 dev->wanted_features = dev->features & dev->hw_features;
6211 if (!(dev->flags & IFF_LOOPBACK)) {
6212 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6215 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6217 dev->vlan_features |= NETIF_F_HIGHDMA;
6219 /* Make NETIF_F_SG inheritable to tunnel devices.
6221 dev->hw_enc_features |= NETIF_F_SG;
6223 /* Make NETIF_F_SG inheritable to MPLS.
6225 dev->mpls_features |= NETIF_F_SG;
6227 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6228 ret = notifier_to_errno(ret);
6232 ret = netdev_register_kobject(dev);
6235 dev->reg_state = NETREG_REGISTERED;
6237 __netdev_update_features(dev);
6240 * Default initial state at registry is that the
6241 * device is present.
6244 set_bit(__LINK_STATE_PRESENT, &dev->state);
6246 linkwatch_init_dev(dev);
6248 dev_init_scheduler(dev);
6250 list_netdevice(dev);
6251 add_device_randomness(dev->dev_addr, dev->addr_len);
6253 /* If the device has permanent device address, driver should
6254 * set dev_addr and also addr_assign_type should be set to
6255 * NET_ADDR_PERM (default value).
6257 if (dev->addr_assign_type == NET_ADDR_PERM)
6258 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6260 /* Notify protocols, that a new device appeared. */
6261 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6262 ret = notifier_to_errno(ret);
6264 rollback_registered(dev);
6265 dev->reg_state = NETREG_UNREGISTERED;
6268 * Prevent userspace races by waiting until the network
6269 * device is fully setup before sending notifications.
6271 if (!dev->rtnl_link_ops ||
6272 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6273 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6279 if (dev->netdev_ops->ndo_uninit)
6280 dev->netdev_ops->ndo_uninit(dev);
6283 EXPORT_SYMBOL(register_netdevice);
6286 * init_dummy_netdev - init a dummy network device for NAPI
6287 * @dev: device to init
6289 * This takes a network device structure and initialize the minimum
6290 * amount of fields so it can be used to schedule NAPI polls without
6291 * registering a full blown interface. This is to be used by drivers
6292 * that need to tie several hardware interfaces to a single NAPI
6293 * poll scheduler due to HW limitations.
6295 int init_dummy_netdev(struct net_device *dev)
6297 /* Clear everything. Note we don't initialize spinlocks
6298 * are they aren't supposed to be taken by any of the
6299 * NAPI code and this dummy netdev is supposed to be
6300 * only ever used for NAPI polls
6302 memset(dev, 0, sizeof(struct net_device));
6304 /* make sure we BUG if trying to hit standard
6305 * register/unregister code path
6307 dev->reg_state = NETREG_DUMMY;
6309 /* NAPI wants this */
6310 INIT_LIST_HEAD(&dev->napi_list);
6312 /* a dummy interface is started by default */
6313 set_bit(__LINK_STATE_PRESENT, &dev->state);
6314 set_bit(__LINK_STATE_START, &dev->state);
6316 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6317 * because users of this 'device' dont need to change
6323 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6327 * register_netdev - register a network device
6328 * @dev: device to register
6330 * Take a completed network device structure and add it to the kernel
6331 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6332 * chain. 0 is returned on success. A negative errno code is returned
6333 * on a failure to set up the device, or if the name is a duplicate.
6335 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6336 * and expands the device name if you passed a format string to
6339 int register_netdev(struct net_device *dev)
6344 err = register_netdevice(dev);
6348 EXPORT_SYMBOL(register_netdev);
6350 int netdev_refcnt_read(const struct net_device *dev)
6354 for_each_possible_cpu(i)
6355 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6358 EXPORT_SYMBOL(netdev_refcnt_read);
6361 * netdev_wait_allrefs - wait until all references are gone.
6362 * @dev: target net_device
6364 * This is called when unregistering network devices.
6366 * Any protocol or device that holds a reference should register
6367 * for netdevice notification, and cleanup and put back the
6368 * reference if they receive an UNREGISTER event.
6369 * We can get stuck here if buggy protocols don't correctly
6372 static void netdev_wait_allrefs(struct net_device *dev)
6374 unsigned long rebroadcast_time, warning_time;
6377 linkwatch_forget_dev(dev);
6379 rebroadcast_time = warning_time = jiffies;
6380 refcnt = netdev_refcnt_read(dev);
6382 while (refcnt != 0) {
6383 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6386 /* Rebroadcast unregister notification */
6387 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6393 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6394 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6396 /* We must not have linkwatch events
6397 * pending on unregister. If this
6398 * happens, we simply run the queue
6399 * unscheduled, resulting in a noop
6402 linkwatch_run_queue();
6407 rebroadcast_time = jiffies;
6412 refcnt = netdev_refcnt_read(dev);
6414 if (time_after(jiffies, warning_time + 10 * HZ)) {
6415 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6417 warning_time = jiffies;
6426 * register_netdevice(x1);
6427 * register_netdevice(x2);
6429 * unregister_netdevice(y1);
6430 * unregister_netdevice(y2);
6436 * We are invoked by rtnl_unlock().
6437 * This allows us to deal with problems:
6438 * 1) We can delete sysfs objects which invoke hotplug
6439 * without deadlocking with linkwatch via keventd.
6440 * 2) Since we run with the RTNL semaphore not held, we can sleep
6441 * safely in order to wait for the netdev refcnt to drop to zero.
6443 * We must not return until all unregister events added during
6444 * the interval the lock was held have been completed.
6446 void netdev_run_todo(void)
6448 struct list_head list;
6450 /* Snapshot list, allow later requests */
6451 list_replace_init(&net_todo_list, &list);
6456 /* Wait for rcu callbacks to finish before next phase */
6457 if (!list_empty(&list))
6460 while (!list_empty(&list)) {
6461 struct net_device *dev
6462 = list_first_entry(&list, struct net_device, todo_list);
6463 list_del(&dev->todo_list);
6466 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6469 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6470 pr_err("network todo '%s' but state %d\n",
6471 dev->name, dev->reg_state);
6476 dev->reg_state = NETREG_UNREGISTERED;
6478 on_each_cpu(flush_backlog, dev, 1);
6480 netdev_wait_allrefs(dev);
6483 BUG_ON(netdev_refcnt_read(dev));
6484 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6485 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6486 WARN_ON(dev->dn_ptr);
6488 if (dev->destructor)
6489 dev->destructor(dev);
6491 /* Report a network device has been unregistered */
6493 dev_net(dev)->dev_unreg_count--;
6495 wake_up(&netdev_unregistering_wq);
6497 /* Free network device */
6498 kobject_put(&dev->dev.kobj);
6502 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6503 * fields in the same order, with only the type differing.
6505 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6506 const struct net_device_stats *netdev_stats)
6508 #if BITS_PER_LONG == 64
6509 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6510 memcpy(stats64, netdev_stats, sizeof(*stats64));
6512 size_t i, n = sizeof(*stats64) / sizeof(u64);
6513 const unsigned long *src = (const unsigned long *)netdev_stats;
6514 u64 *dst = (u64 *)stats64;
6516 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6517 sizeof(*stats64) / sizeof(u64));
6518 for (i = 0; i < n; i++)
6522 EXPORT_SYMBOL(netdev_stats_to_stats64);
6525 * dev_get_stats - get network device statistics
6526 * @dev: device to get statistics from
6527 * @storage: place to store stats
6529 * Get network statistics from device. Return @storage.
6530 * The device driver may provide its own method by setting
6531 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6532 * otherwise the internal statistics structure is used.
6534 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6535 struct rtnl_link_stats64 *storage)
6537 const struct net_device_ops *ops = dev->netdev_ops;
6539 if (ops->ndo_get_stats64) {
6540 memset(storage, 0, sizeof(*storage));
6541 ops->ndo_get_stats64(dev, storage);
6542 } else if (ops->ndo_get_stats) {
6543 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6545 netdev_stats_to_stats64(storage, &dev->stats);
6547 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6548 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6551 EXPORT_SYMBOL(dev_get_stats);
6553 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6555 struct netdev_queue *queue = dev_ingress_queue(dev);
6557 #ifdef CONFIG_NET_CLS_ACT
6560 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6563 netdev_init_one_queue(dev, queue, NULL);
6564 queue->qdisc = &noop_qdisc;
6565 queue->qdisc_sleeping = &noop_qdisc;
6566 rcu_assign_pointer(dev->ingress_queue, queue);
6571 static const struct ethtool_ops default_ethtool_ops;
6573 void netdev_set_default_ethtool_ops(struct net_device *dev,
6574 const struct ethtool_ops *ops)
6576 if (dev->ethtool_ops == &default_ethtool_ops)
6577 dev->ethtool_ops = ops;
6579 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6581 void netdev_freemem(struct net_device *dev)
6583 char *addr = (char *)dev - dev->padded;
6589 * alloc_netdev_mqs - allocate network device
6590 * @sizeof_priv: size of private data to allocate space for
6591 * @name: device name format string
6592 * @name_assign_type: origin of device name
6593 * @setup: callback to initialize device
6594 * @txqs: the number of TX subqueues to allocate
6595 * @rxqs: the number of RX subqueues to allocate
6597 * Allocates a struct net_device with private data area for driver use
6598 * and performs basic initialization. Also allocates subqueue structs
6599 * for each queue on the device.
6601 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6602 unsigned char name_assign_type,
6603 void (*setup)(struct net_device *),
6604 unsigned int txqs, unsigned int rxqs)
6606 struct net_device *dev;
6608 struct net_device *p;
6610 BUG_ON(strlen(name) >= sizeof(dev->name));
6613 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6619 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6624 alloc_size = sizeof(struct net_device);
6626 /* ensure 32-byte alignment of private area */
6627 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6628 alloc_size += sizeof_priv;
6630 /* ensure 32-byte alignment of whole construct */
6631 alloc_size += NETDEV_ALIGN - 1;
6633 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6635 p = vzalloc(alloc_size);
6639 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6640 dev->padded = (char *)dev - (char *)p;
6642 dev->pcpu_refcnt = alloc_percpu(int);
6643 if (!dev->pcpu_refcnt)
6646 if (dev_addr_init(dev))
6652 dev_net_set(dev, &init_net);
6654 dev->gso_max_size = GSO_MAX_SIZE;
6655 dev->gso_max_segs = GSO_MAX_SEGS;
6656 dev->gso_min_segs = 0;
6658 INIT_LIST_HEAD(&dev->napi_list);
6659 INIT_LIST_HEAD(&dev->unreg_list);
6660 INIT_LIST_HEAD(&dev->close_list);
6661 INIT_LIST_HEAD(&dev->link_watch_list);
6662 INIT_LIST_HEAD(&dev->adj_list.upper);
6663 INIT_LIST_HEAD(&dev->adj_list.lower);
6664 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6665 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6666 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
6669 dev->num_tx_queues = txqs;
6670 dev->real_num_tx_queues = txqs;
6671 if (netif_alloc_netdev_queues(dev))
6675 dev->num_rx_queues = rxqs;
6676 dev->real_num_rx_queues = rxqs;
6677 if (netif_alloc_rx_queues(dev))
6681 strcpy(dev->name, name);
6682 dev->name_assign_type = name_assign_type;
6683 dev->group = INIT_NETDEV_GROUP;
6684 if (!dev->ethtool_ops)
6685 dev->ethtool_ops = &default_ethtool_ops;
6693 free_percpu(dev->pcpu_refcnt);
6695 netdev_freemem(dev);
6698 EXPORT_SYMBOL(alloc_netdev_mqs);
6701 * free_netdev - free network device
6704 * This function does the last stage of destroying an allocated device
6705 * interface. The reference to the device object is released.
6706 * If this is the last reference then it will be freed.
6708 void free_netdev(struct net_device *dev)
6710 struct napi_struct *p, *n;
6712 release_net(dev_net(dev));
6714 netif_free_tx_queues(dev);
6719 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6721 /* Flush device addresses */
6722 dev_addr_flush(dev);
6724 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6727 free_percpu(dev->pcpu_refcnt);
6728 dev->pcpu_refcnt = NULL;
6730 /* Compatibility with error handling in drivers */
6731 if (dev->reg_state == NETREG_UNINITIALIZED) {
6732 netdev_freemem(dev);
6736 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6737 dev->reg_state = NETREG_RELEASED;
6739 /* will free via device release */
6740 put_device(&dev->dev);
6742 EXPORT_SYMBOL(free_netdev);
6745 * synchronize_net - Synchronize with packet receive processing
6747 * Wait for packets currently being received to be done.
6748 * Does not block later packets from starting.
6750 void synchronize_net(void)
6753 if (rtnl_is_locked())
6754 synchronize_rcu_expedited();
6758 EXPORT_SYMBOL(synchronize_net);
6761 * unregister_netdevice_queue - remove device from the kernel
6765 * This function shuts down a device interface and removes it
6766 * from the kernel tables.
6767 * If head not NULL, device is queued to be unregistered later.
6769 * Callers must hold the rtnl semaphore. You may want
6770 * unregister_netdev() instead of this.
6773 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6778 list_move_tail(&dev->unreg_list, head);
6780 rollback_registered(dev);
6781 /* Finish processing unregister after unlock */
6785 EXPORT_SYMBOL(unregister_netdevice_queue);
6788 * unregister_netdevice_many - unregister many devices
6789 * @head: list of devices
6791 * Note: As most callers use a stack allocated list_head,
6792 * we force a list_del() to make sure stack wont be corrupted later.
6794 void unregister_netdevice_many(struct list_head *head)
6796 struct net_device *dev;
6798 if (!list_empty(head)) {
6799 rollback_registered_many(head);
6800 list_for_each_entry(dev, head, unreg_list)
6805 EXPORT_SYMBOL(unregister_netdevice_many);
6808 * unregister_netdev - remove device from the kernel
6811 * This function shuts down a device interface and removes it
6812 * from the kernel tables.
6814 * This is just a wrapper for unregister_netdevice that takes
6815 * the rtnl semaphore. In general you want to use this and not
6816 * unregister_netdevice.
6818 void unregister_netdev(struct net_device *dev)
6821 unregister_netdevice(dev);
6824 EXPORT_SYMBOL(unregister_netdev);
6827 * dev_change_net_namespace - move device to different nethost namespace
6829 * @net: network namespace
6830 * @pat: If not NULL name pattern to try if the current device name
6831 * is already taken in the destination network namespace.
6833 * This function shuts down a device interface and moves it
6834 * to a new network namespace. On success 0 is returned, on
6835 * a failure a netagive errno code is returned.
6837 * Callers must hold the rtnl semaphore.
6840 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6846 /* Don't allow namespace local devices to be moved. */
6848 if (dev->features & NETIF_F_NETNS_LOCAL)
6851 /* Ensure the device has been registrered */
6852 if (dev->reg_state != NETREG_REGISTERED)
6855 /* Get out if there is nothing todo */
6857 if (net_eq(dev_net(dev), net))
6860 /* Pick the destination device name, and ensure
6861 * we can use it in the destination network namespace.
6864 if (__dev_get_by_name(net, dev->name)) {
6865 /* We get here if we can't use the current device name */
6868 if (dev_get_valid_name(net, dev, pat) < 0)
6873 * And now a mini version of register_netdevice unregister_netdevice.
6876 /* If device is running close it first. */
6879 /* And unlink it from device chain */
6881 unlist_netdevice(dev);
6885 /* Shutdown queueing discipline. */
6888 /* Notify protocols, that we are about to destroy
6889 this device. They should clean all the things.
6891 Note that dev->reg_state stays at NETREG_REGISTERED.
6892 This is wanted because this way 8021q and macvlan know
6893 the device is just moving and can keep their slaves up.
6895 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6897 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6898 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6901 * Flush the unicast and multicast chains
6906 /* Send a netdev-removed uevent to the old namespace */
6907 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6908 netdev_adjacent_del_links(dev);
6910 /* Actually switch the network namespace */
6911 dev_net_set(dev, net);
6913 /* If there is an ifindex conflict assign a new one */
6914 if (__dev_get_by_index(net, dev->ifindex)) {
6915 int iflink = (dev->iflink == dev->ifindex);
6916 dev->ifindex = dev_new_index(net);
6918 dev->iflink = dev->ifindex;
6921 /* Send a netdev-add uevent to the new namespace */
6922 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6923 netdev_adjacent_add_links(dev);
6925 /* Fixup kobjects */
6926 err = device_rename(&dev->dev, dev->name);
6929 /* Add the device back in the hashes */
6930 list_netdevice(dev);
6932 /* Notify protocols, that a new device appeared. */
6933 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6936 * Prevent userspace races by waiting until the network
6937 * device is fully setup before sending notifications.
6939 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6946 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6948 static int dev_cpu_callback(struct notifier_block *nfb,
6949 unsigned long action,
6952 struct sk_buff **list_skb;
6953 struct sk_buff *skb;
6954 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6955 struct softnet_data *sd, *oldsd;
6957 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6960 local_irq_disable();
6961 cpu = smp_processor_id();
6962 sd = &per_cpu(softnet_data, cpu);
6963 oldsd = &per_cpu(softnet_data, oldcpu);
6965 /* Find end of our completion_queue. */
6966 list_skb = &sd->completion_queue;
6968 list_skb = &(*list_skb)->next;
6969 /* Append completion queue from offline CPU. */
6970 *list_skb = oldsd->completion_queue;
6971 oldsd->completion_queue = NULL;
6973 /* Append output queue from offline CPU. */
6974 if (oldsd->output_queue) {
6975 *sd->output_queue_tailp = oldsd->output_queue;
6976 sd->output_queue_tailp = oldsd->output_queue_tailp;
6977 oldsd->output_queue = NULL;
6978 oldsd->output_queue_tailp = &oldsd->output_queue;
6980 /* Append NAPI poll list from offline CPU. */
6981 if (!list_empty(&oldsd->poll_list)) {
6982 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6983 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6986 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6989 /* Process offline CPU's input_pkt_queue */
6990 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6991 netif_rx_internal(skb);
6992 input_queue_head_incr(oldsd);
6994 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6995 netif_rx_internal(skb);
6996 input_queue_head_incr(oldsd);
7004 * netdev_increment_features - increment feature set by one
7005 * @all: current feature set
7006 * @one: new feature set
7007 * @mask: mask feature set
7009 * Computes a new feature set after adding a device with feature set
7010 * @one to the master device with current feature set @all. Will not
7011 * enable anything that is off in @mask. Returns the new feature set.
7013 netdev_features_t netdev_increment_features(netdev_features_t all,
7014 netdev_features_t one, netdev_features_t mask)
7016 if (mask & NETIF_F_GEN_CSUM)
7017 mask |= NETIF_F_ALL_CSUM;
7018 mask |= NETIF_F_VLAN_CHALLENGED;
7020 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7021 all &= one | ~NETIF_F_ALL_FOR_ALL;
7023 /* If one device supports hw checksumming, set for all. */
7024 if (all & NETIF_F_GEN_CSUM)
7025 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7029 EXPORT_SYMBOL(netdev_increment_features);
7031 static struct hlist_head * __net_init netdev_create_hash(void)
7034 struct hlist_head *hash;
7036 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7038 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7039 INIT_HLIST_HEAD(&hash[i]);
7044 /* Initialize per network namespace state */
7045 static int __net_init netdev_init(struct net *net)
7047 if (net != &init_net)
7048 INIT_LIST_HEAD(&net->dev_base_head);
7050 net->dev_name_head = netdev_create_hash();
7051 if (net->dev_name_head == NULL)
7054 net->dev_index_head = netdev_create_hash();
7055 if (net->dev_index_head == NULL)
7061 kfree(net->dev_name_head);
7067 * netdev_drivername - network driver for the device
7068 * @dev: network device
7070 * Determine network driver for device.
7072 const char *netdev_drivername(const struct net_device *dev)
7074 const struct device_driver *driver;
7075 const struct device *parent;
7076 const char *empty = "";
7078 parent = dev->dev.parent;
7082 driver = parent->driver;
7083 if (driver && driver->name)
7084 return driver->name;
7088 static void __netdev_printk(const char *level, const struct net_device *dev,
7089 struct va_format *vaf)
7091 if (dev && dev->dev.parent) {
7092 dev_printk_emit(level[1] - '0',
7095 dev_driver_string(dev->dev.parent),
7096 dev_name(dev->dev.parent),
7097 netdev_name(dev), netdev_reg_state(dev),
7100 printk("%s%s%s: %pV",
7101 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7103 printk("%s(NULL net_device): %pV", level, vaf);
7107 void netdev_printk(const char *level, const struct net_device *dev,
7108 const char *format, ...)
7110 struct va_format vaf;
7113 va_start(args, format);
7118 __netdev_printk(level, dev, &vaf);
7122 EXPORT_SYMBOL(netdev_printk);
7124 #define define_netdev_printk_level(func, level) \
7125 void func(const struct net_device *dev, const char *fmt, ...) \
7127 struct va_format vaf; \
7130 va_start(args, fmt); \
7135 __netdev_printk(level, dev, &vaf); \
7139 EXPORT_SYMBOL(func);
7141 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7142 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7143 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7144 define_netdev_printk_level(netdev_err, KERN_ERR);
7145 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7146 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7147 define_netdev_printk_level(netdev_info, KERN_INFO);
7149 static void __net_exit netdev_exit(struct net *net)
7151 kfree(net->dev_name_head);
7152 kfree(net->dev_index_head);
7155 static struct pernet_operations __net_initdata netdev_net_ops = {
7156 .init = netdev_init,
7157 .exit = netdev_exit,
7160 static void __net_exit default_device_exit(struct net *net)
7162 struct net_device *dev, *aux;
7164 * Push all migratable network devices back to the
7165 * initial network namespace
7168 for_each_netdev_safe(net, dev, aux) {
7170 char fb_name[IFNAMSIZ];
7172 /* Ignore unmoveable devices (i.e. loopback) */
7173 if (dev->features & NETIF_F_NETNS_LOCAL)
7176 /* Leave virtual devices for the generic cleanup */
7177 if (dev->rtnl_link_ops)
7180 /* Push remaining network devices to init_net */
7181 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7182 err = dev_change_net_namespace(dev, &init_net, fb_name);
7184 pr_emerg("%s: failed to move %s to init_net: %d\n",
7185 __func__, dev->name, err);
7192 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7194 /* Return with the rtnl_lock held when there are no network
7195 * devices unregistering in any network namespace in net_list.
7202 prepare_to_wait(&netdev_unregistering_wq, &wait,
7203 TASK_UNINTERRUPTIBLE);
7204 unregistering = false;
7206 list_for_each_entry(net, net_list, exit_list) {
7207 if (net->dev_unreg_count > 0) {
7208 unregistering = true;
7217 finish_wait(&netdev_unregistering_wq, &wait);
7220 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7222 /* At exit all network devices most be removed from a network
7223 * namespace. Do this in the reverse order of registration.
7224 * Do this across as many network namespaces as possible to
7225 * improve batching efficiency.
7227 struct net_device *dev;
7229 LIST_HEAD(dev_kill_list);
7231 /* To prevent network device cleanup code from dereferencing
7232 * loopback devices or network devices that have been freed
7233 * wait here for all pending unregistrations to complete,
7234 * before unregistring the loopback device and allowing the
7235 * network namespace be freed.
7237 * The netdev todo list containing all network devices
7238 * unregistrations that happen in default_device_exit_batch
7239 * will run in the rtnl_unlock() at the end of
7240 * default_device_exit_batch.
7242 rtnl_lock_unregistering(net_list);
7243 list_for_each_entry(net, net_list, exit_list) {
7244 for_each_netdev_reverse(net, dev) {
7245 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7246 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7248 unregister_netdevice_queue(dev, &dev_kill_list);
7251 unregister_netdevice_many(&dev_kill_list);
7255 static struct pernet_operations __net_initdata default_device_ops = {
7256 .exit = default_device_exit,
7257 .exit_batch = default_device_exit_batch,
7261 * Initialize the DEV module. At boot time this walks the device list and
7262 * unhooks any devices that fail to initialise (normally hardware not
7263 * present) and leaves us with a valid list of present and active devices.
7268 * This is called single threaded during boot, so no need
7269 * to take the rtnl semaphore.
7271 static int __init net_dev_init(void)
7273 int i, rc = -ENOMEM;
7275 BUG_ON(!dev_boot_phase);
7277 if (dev_proc_init())
7280 if (netdev_kobject_init())
7283 INIT_LIST_HEAD(&ptype_all);
7284 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7285 INIT_LIST_HEAD(&ptype_base[i]);
7287 INIT_LIST_HEAD(&offload_base);
7289 if (register_pernet_subsys(&netdev_net_ops))
7293 * Initialise the packet receive queues.
7296 for_each_possible_cpu(i) {
7297 struct softnet_data *sd = &per_cpu(softnet_data, i);
7299 skb_queue_head_init(&sd->input_pkt_queue);
7300 skb_queue_head_init(&sd->process_queue);
7301 INIT_LIST_HEAD(&sd->poll_list);
7302 sd->output_queue_tailp = &sd->output_queue;
7304 sd->csd.func = rps_trigger_softirq;
7309 sd->backlog.poll = process_backlog;
7310 sd->backlog.weight = weight_p;
7315 /* The loopback device is special if any other network devices
7316 * is present in a network namespace the loopback device must
7317 * be present. Since we now dynamically allocate and free the
7318 * loopback device ensure this invariant is maintained by
7319 * keeping the loopback device as the first device on the
7320 * list of network devices. Ensuring the loopback devices
7321 * is the first device that appears and the last network device
7324 if (register_pernet_device(&loopback_net_ops))
7327 if (register_pernet_device(&default_device_ops))
7330 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7331 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7333 hotcpu_notifier(dev_cpu_callback, 0);
7340 subsys_initcall(net_dev_init);