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 net_dmaengine_get();
1290 dev_set_rx_mode(dev);
1292 add_device_randomness(dev->dev_addr, dev->addr_len);
1299 * dev_open - prepare an interface for use.
1300 * @dev: device to open
1302 * Takes a device from down to up state. The device's private open
1303 * function is invoked and then the multicast lists are loaded. Finally
1304 * the device is moved into the up state and a %NETDEV_UP message is
1305 * sent to the netdev notifier chain.
1307 * Calling this function on an active interface is a nop. On a failure
1308 * a negative errno code is returned.
1310 int dev_open(struct net_device *dev)
1314 if (dev->flags & IFF_UP)
1317 ret = __dev_open(dev);
1321 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1322 call_netdevice_notifiers(NETDEV_UP, dev);
1326 EXPORT_SYMBOL(dev_open);
1328 static int __dev_close_many(struct list_head *head)
1330 struct net_device *dev;
1335 list_for_each_entry(dev, head, close_list) {
1336 /* Temporarily disable netpoll until the interface is down */
1337 netpoll_poll_disable(dev);
1339 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1341 clear_bit(__LINK_STATE_START, &dev->state);
1343 /* Synchronize to scheduled poll. We cannot touch poll list, it
1344 * can be even on different cpu. So just clear netif_running().
1346 * dev->stop() will invoke napi_disable() on all of it's
1347 * napi_struct instances on this device.
1349 smp_mb__after_atomic(); /* Commit netif_running(). */
1352 dev_deactivate_many(head);
1354 list_for_each_entry(dev, head, close_list) {
1355 const struct net_device_ops *ops = dev->netdev_ops;
1358 * Call the device specific close. This cannot fail.
1359 * Only if device is UP
1361 * We allow it to be called even after a DETACH hot-plug
1367 dev->flags &= ~IFF_UP;
1368 net_dmaengine_put();
1369 netpoll_poll_enable(dev);
1375 static int __dev_close(struct net_device *dev)
1380 list_add(&dev->close_list, &single);
1381 retval = __dev_close_many(&single);
1387 static int dev_close_many(struct list_head *head)
1389 struct net_device *dev, *tmp;
1391 /* Remove the devices that don't need to be closed */
1392 list_for_each_entry_safe(dev, tmp, head, close_list)
1393 if (!(dev->flags & IFF_UP))
1394 list_del_init(&dev->close_list);
1396 __dev_close_many(head);
1398 list_for_each_entry_safe(dev, tmp, head, close_list) {
1399 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1400 call_netdevice_notifiers(NETDEV_DOWN, dev);
1401 list_del_init(&dev->close_list);
1408 * dev_close - shutdown an interface.
1409 * @dev: device to shutdown
1411 * This function moves an active device into down state. A
1412 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1413 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1416 int dev_close(struct net_device *dev)
1418 if (dev->flags & IFF_UP) {
1421 list_add(&dev->close_list, &single);
1422 dev_close_many(&single);
1427 EXPORT_SYMBOL(dev_close);
1431 * dev_disable_lro - disable Large Receive Offload on a device
1434 * Disable Large Receive Offload (LRO) on a net device. Must be
1435 * called under RTNL. This is needed if received packets may be
1436 * forwarded to another interface.
1438 void dev_disable_lro(struct net_device *dev)
1441 * If we're trying to disable lro on a vlan device
1442 * use the underlying physical device instead
1444 if (is_vlan_dev(dev))
1445 dev = vlan_dev_real_dev(dev);
1447 /* the same for macvlan devices */
1448 if (netif_is_macvlan(dev))
1449 dev = macvlan_dev_real_dev(dev);
1451 dev->wanted_features &= ~NETIF_F_LRO;
1452 netdev_update_features(dev);
1454 if (unlikely(dev->features & NETIF_F_LRO))
1455 netdev_WARN(dev, "failed to disable LRO!\n");
1457 EXPORT_SYMBOL(dev_disable_lro);
1459 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1460 struct net_device *dev)
1462 struct netdev_notifier_info info;
1464 netdev_notifier_info_init(&info, dev);
1465 return nb->notifier_call(nb, val, &info);
1468 static int dev_boot_phase = 1;
1471 * register_netdevice_notifier - register a network notifier block
1474 * Register a notifier to be called when network device events occur.
1475 * The notifier passed is linked into the kernel structures and must
1476 * not be reused until it has been unregistered. A negative errno code
1477 * is returned on a failure.
1479 * When registered all registration and up events are replayed
1480 * to the new notifier to allow device to have a race free
1481 * view of the network device list.
1484 int register_netdevice_notifier(struct notifier_block *nb)
1486 struct net_device *dev;
1487 struct net_device *last;
1492 err = raw_notifier_chain_register(&netdev_chain, nb);
1498 for_each_netdev(net, dev) {
1499 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1500 err = notifier_to_errno(err);
1504 if (!(dev->flags & IFF_UP))
1507 call_netdevice_notifier(nb, NETDEV_UP, dev);
1518 for_each_netdev(net, dev) {
1522 if (dev->flags & IFF_UP) {
1523 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1525 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1527 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1532 raw_notifier_chain_unregister(&netdev_chain, nb);
1535 EXPORT_SYMBOL(register_netdevice_notifier);
1538 * unregister_netdevice_notifier - unregister a network notifier block
1541 * Unregister a notifier previously registered by
1542 * register_netdevice_notifier(). The notifier is unlinked into the
1543 * kernel structures and may then be reused. A negative errno code
1544 * is returned on a failure.
1546 * After unregistering unregister and down device events are synthesized
1547 * for all devices on the device list to the removed notifier to remove
1548 * the need for special case cleanup code.
1551 int unregister_netdevice_notifier(struct notifier_block *nb)
1553 struct net_device *dev;
1558 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1563 for_each_netdev(net, dev) {
1564 if (dev->flags & IFF_UP) {
1565 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1567 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1569 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1576 EXPORT_SYMBOL(unregister_netdevice_notifier);
1579 * call_netdevice_notifiers_info - call all network notifier blocks
1580 * @val: value passed unmodified to notifier function
1581 * @dev: net_device pointer passed unmodified to notifier function
1582 * @info: notifier information data
1584 * Call all network notifier blocks. Parameters and return value
1585 * are as for raw_notifier_call_chain().
1588 static int call_netdevice_notifiers_info(unsigned long val,
1589 struct net_device *dev,
1590 struct netdev_notifier_info *info)
1593 netdev_notifier_info_init(info, dev);
1594 return raw_notifier_call_chain(&netdev_chain, val, info);
1598 * call_netdevice_notifiers - call all network notifier blocks
1599 * @val: value passed unmodified to notifier function
1600 * @dev: net_device pointer passed unmodified to notifier function
1602 * Call all network notifier blocks. Parameters and return value
1603 * are as for raw_notifier_call_chain().
1606 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1608 struct netdev_notifier_info info;
1610 return call_netdevice_notifiers_info(val, dev, &info);
1612 EXPORT_SYMBOL(call_netdevice_notifiers);
1614 static struct static_key netstamp_needed __read_mostly;
1615 #ifdef HAVE_JUMP_LABEL
1616 /* We are not allowed to call static_key_slow_dec() from irq context
1617 * If net_disable_timestamp() is called from irq context, defer the
1618 * static_key_slow_dec() calls.
1620 static atomic_t netstamp_needed_deferred;
1623 void net_enable_timestamp(void)
1625 #ifdef HAVE_JUMP_LABEL
1626 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1630 static_key_slow_dec(&netstamp_needed);
1634 static_key_slow_inc(&netstamp_needed);
1636 EXPORT_SYMBOL(net_enable_timestamp);
1638 void net_disable_timestamp(void)
1640 #ifdef HAVE_JUMP_LABEL
1641 if (in_interrupt()) {
1642 atomic_inc(&netstamp_needed_deferred);
1646 static_key_slow_dec(&netstamp_needed);
1648 EXPORT_SYMBOL(net_disable_timestamp);
1650 static inline void net_timestamp_set(struct sk_buff *skb)
1652 skb->tstamp.tv64 = 0;
1653 if (static_key_false(&netstamp_needed))
1654 __net_timestamp(skb);
1657 #define net_timestamp_check(COND, SKB) \
1658 if (static_key_false(&netstamp_needed)) { \
1659 if ((COND) && !(SKB)->tstamp.tv64) \
1660 __net_timestamp(SKB); \
1663 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1667 if (!(dev->flags & IFF_UP))
1670 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1671 if (skb->len <= len)
1674 /* if TSO is enabled, we don't care about the length as the packet
1675 * could be forwarded without being segmented before
1677 if (skb_is_gso(skb))
1682 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1684 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1686 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1687 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1688 atomic_long_inc(&dev->rx_dropped);
1694 if (unlikely(!is_skb_forwardable(dev, skb))) {
1695 atomic_long_inc(&dev->rx_dropped);
1700 skb_scrub_packet(skb, true);
1701 skb->protocol = eth_type_trans(skb, dev);
1705 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1708 * dev_forward_skb - loopback an skb to another netif
1710 * @dev: destination network device
1711 * @skb: buffer to forward
1714 * NET_RX_SUCCESS (no congestion)
1715 * NET_RX_DROP (packet was dropped, but freed)
1717 * dev_forward_skb can be used for injecting an skb from the
1718 * start_xmit function of one device into the receive queue
1719 * of another device.
1721 * The receiving device may be in another namespace, so
1722 * we have to clear all information in the skb that could
1723 * impact namespace isolation.
1725 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1727 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1729 EXPORT_SYMBOL_GPL(dev_forward_skb);
1731 static inline int deliver_skb(struct sk_buff *skb,
1732 struct packet_type *pt_prev,
1733 struct net_device *orig_dev)
1735 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1737 atomic_inc(&skb->users);
1738 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1741 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1743 if (!ptype->af_packet_priv || !skb->sk)
1746 if (ptype->id_match)
1747 return ptype->id_match(ptype, skb->sk);
1748 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1755 * Support routine. Sends outgoing frames to any network
1756 * taps currently in use.
1759 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1761 struct packet_type *ptype;
1762 struct sk_buff *skb2 = NULL;
1763 struct packet_type *pt_prev = NULL;
1766 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1767 /* Never send packets back to the socket
1768 * they originated from - MvS (miquels@drinkel.ow.org)
1770 if ((ptype->dev == dev || !ptype->dev) &&
1771 (!skb_loop_sk(ptype, skb))) {
1773 deliver_skb(skb2, pt_prev, skb->dev);
1778 skb2 = skb_clone(skb, GFP_ATOMIC);
1782 net_timestamp_set(skb2);
1784 /* skb->nh should be correctly
1785 set by sender, so that the second statement is
1786 just protection against buggy protocols.
1788 skb_reset_mac_header(skb2);
1790 if (skb_network_header(skb2) < skb2->data ||
1791 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1792 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1793 ntohs(skb2->protocol),
1795 skb_reset_network_header(skb2);
1798 skb2->transport_header = skb2->network_header;
1799 skb2->pkt_type = PACKET_OUTGOING;
1804 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1809 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1810 * @dev: Network device
1811 * @txq: number of queues available
1813 * If real_num_tx_queues is changed the tc mappings may no longer be
1814 * valid. To resolve this verify the tc mapping remains valid and if
1815 * not NULL the mapping. With no priorities mapping to this
1816 * offset/count pair it will no longer be used. In the worst case TC0
1817 * is invalid nothing can be done so disable priority mappings. If is
1818 * expected that drivers will fix this mapping if they can before
1819 * calling netif_set_real_num_tx_queues.
1821 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1824 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1826 /* If TC0 is invalidated disable TC mapping */
1827 if (tc->offset + tc->count > txq) {
1828 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1833 /* Invalidated prio to tc mappings set to TC0 */
1834 for (i = 1; i < TC_BITMASK + 1; i++) {
1835 int q = netdev_get_prio_tc_map(dev, i);
1837 tc = &dev->tc_to_txq[q];
1838 if (tc->offset + tc->count > txq) {
1839 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1841 netdev_set_prio_tc_map(dev, i, 0);
1847 static DEFINE_MUTEX(xps_map_mutex);
1848 #define xmap_dereference(P) \
1849 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1851 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1854 struct xps_map *map = NULL;
1858 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1860 for (pos = 0; map && pos < map->len; pos++) {
1861 if (map->queues[pos] == index) {
1863 map->queues[pos] = map->queues[--map->len];
1865 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1866 kfree_rcu(map, rcu);
1876 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1878 struct xps_dev_maps *dev_maps;
1880 bool active = false;
1882 mutex_lock(&xps_map_mutex);
1883 dev_maps = xmap_dereference(dev->xps_maps);
1888 for_each_possible_cpu(cpu) {
1889 for (i = index; i < dev->num_tx_queues; i++) {
1890 if (!remove_xps_queue(dev_maps, cpu, i))
1893 if (i == dev->num_tx_queues)
1898 RCU_INIT_POINTER(dev->xps_maps, NULL);
1899 kfree_rcu(dev_maps, rcu);
1902 for (i = index; i < dev->num_tx_queues; i++)
1903 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1907 mutex_unlock(&xps_map_mutex);
1910 static struct xps_map *expand_xps_map(struct xps_map *map,
1913 struct xps_map *new_map;
1914 int alloc_len = XPS_MIN_MAP_ALLOC;
1917 for (pos = 0; map && pos < map->len; pos++) {
1918 if (map->queues[pos] != index)
1923 /* Need to add queue to this CPU's existing map */
1925 if (pos < map->alloc_len)
1928 alloc_len = map->alloc_len * 2;
1931 /* Need to allocate new map to store queue on this CPU's map */
1932 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1937 for (i = 0; i < pos; i++)
1938 new_map->queues[i] = map->queues[i];
1939 new_map->alloc_len = alloc_len;
1945 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1948 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1949 struct xps_map *map, *new_map;
1950 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1951 int cpu, numa_node_id = -2;
1952 bool active = false;
1954 mutex_lock(&xps_map_mutex);
1956 dev_maps = xmap_dereference(dev->xps_maps);
1958 /* allocate memory for queue storage */
1959 for_each_online_cpu(cpu) {
1960 if (!cpumask_test_cpu(cpu, mask))
1964 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1965 if (!new_dev_maps) {
1966 mutex_unlock(&xps_map_mutex);
1970 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1973 map = expand_xps_map(map, cpu, index);
1977 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1981 goto out_no_new_maps;
1983 for_each_possible_cpu(cpu) {
1984 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1985 /* add queue to CPU maps */
1988 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1989 while ((pos < map->len) && (map->queues[pos] != index))
1992 if (pos == map->len)
1993 map->queues[map->len++] = index;
1995 if (numa_node_id == -2)
1996 numa_node_id = cpu_to_node(cpu);
1997 else if (numa_node_id != cpu_to_node(cpu))
2000 } else if (dev_maps) {
2001 /* fill in the new device map from the old device map */
2002 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2003 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2008 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2010 /* Cleanup old maps */
2012 for_each_possible_cpu(cpu) {
2013 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2014 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2015 if (map && map != new_map)
2016 kfree_rcu(map, rcu);
2019 kfree_rcu(dev_maps, rcu);
2022 dev_maps = new_dev_maps;
2026 /* update Tx queue numa node */
2027 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2028 (numa_node_id >= 0) ? numa_node_id :
2034 /* removes queue from unused CPUs */
2035 for_each_possible_cpu(cpu) {
2036 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2039 if (remove_xps_queue(dev_maps, cpu, index))
2043 /* free map if not active */
2045 RCU_INIT_POINTER(dev->xps_maps, NULL);
2046 kfree_rcu(dev_maps, rcu);
2050 mutex_unlock(&xps_map_mutex);
2054 /* remove any maps that we added */
2055 for_each_possible_cpu(cpu) {
2056 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2057 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2059 if (new_map && new_map != map)
2063 mutex_unlock(&xps_map_mutex);
2065 kfree(new_dev_maps);
2068 EXPORT_SYMBOL(netif_set_xps_queue);
2072 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2073 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2075 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2079 if (txq < 1 || txq > dev->num_tx_queues)
2082 if (dev->reg_state == NETREG_REGISTERED ||
2083 dev->reg_state == NETREG_UNREGISTERING) {
2086 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2092 netif_setup_tc(dev, txq);
2094 if (txq < dev->real_num_tx_queues) {
2095 qdisc_reset_all_tx_gt(dev, txq);
2097 netif_reset_xps_queues_gt(dev, txq);
2102 dev->real_num_tx_queues = txq;
2105 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2109 * netif_set_real_num_rx_queues - set actual number of RX queues used
2110 * @dev: Network device
2111 * @rxq: Actual number of RX queues
2113 * This must be called either with the rtnl_lock held or before
2114 * registration of the net device. Returns 0 on success, or a
2115 * negative error code. If called before registration, it always
2118 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2122 if (rxq < 1 || rxq > dev->num_rx_queues)
2125 if (dev->reg_state == NETREG_REGISTERED) {
2128 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2134 dev->real_num_rx_queues = rxq;
2137 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2141 * netif_get_num_default_rss_queues - default number of RSS queues
2143 * This routine should set an upper limit on the number of RSS queues
2144 * used by default by multiqueue devices.
2146 int netif_get_num_default_rss_queues(void)
2148 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2150 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2152 static inline void __netif_reschedule(struct Qdisc *q)
2154 struct softnet_data *sd;
2155 unsigned long flags;
2157 local_irq_save(flags);
2158 sd = &__get_cpu_var(softnet_data);
2159 q->next_sched = NULL;
2160 *sd->output_queue_tailp = q;
2161 sd->output_queue_tailp = &q->next_sched;
2162 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2163 local_irq_restore(flags);
2166 void __netif_schedule(struct Qdisc *q)
2168 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2169 __netif_reschedule(q);
2171 EXPORT_SYMBOL(__netif_schedule);
2173 struct dev_kfree_skb_cb {
2174 enum skb_free_reason reason;
2177 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2179 return (struct dev_kfree_skb_cb *)skb->cb;
2182 void netif_schedule_queue(struct netdev_queue *txq)
2185 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2186 struct Qdisc *q = rcu_dereference(txq->qdisc);
2188 __netif_schedule(q);
2192 EXPORT_SYMBOL(netif_schedule_queue);
2195 * netif_wake_subqueue - allow sending packets on subqueue
2196 * @dev: network device
2197 * @queue_index: sub queue index
2199 * Resume individual transmit queue of a device with multiple transmit queues.
2201 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2203 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2205 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2209 q = rcu_dereference(txq->qdisc);
2210 __netif_schedule(q);
2214 EXPORT_SYMBOL(netif_wake_subqueue);
2216 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2218 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2222 q = rcu_dereference(dev_queue->qdisc);
2223 __netif_schedule(q);
2227 EXPORT_SYMBOL(netif_tx_wake_queue);
2229 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2231 unsigned long flags;
2233 if (likely(atomic_read(&skb->users) == 1)) {
2235 atomic_set(&skb->users, 0);
2236 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2239 get_kfree_skb_cb(skb)->reason = reason;
2240 local_irq_save(flags);
2241 skb->next = __this_cpu_read(softnet_data.completion_queue);
2242 __this_cpu_write(softnet_data.completion_queue, skb);
2243 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2244 local_irq_restore(flags);
2246 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2248 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2250 if (in_irq() || irqs_disabled())
2251 __dev_kfree_skb_irq(skb, reason);
2255 EXPORT_SYMBOL(__dev_kfree_skb_any);
2259 * netif_device_detach - mark device as removed
2260 * @dev: network device
2262 * Mark device as removed from system and therefore no longer available.
2264 void netif_device_detach(struct net_device *dev)
2266 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2267 netif_running(dev)) {
2268 netif_tx_stop_all_queues(dev);
2271 EXPORT_SYMBOL(netif_device_detach);
2274 * netif_device_attach - mark device as attached
2275 * @dev: network device
2277 * Mark device as attached from system and restart if needed.
2279 void netif_device_attach(struct net_device *dev)
2281 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2282 netif_running(dev)) {
2283 netif_tx_wake_all_queues(dev);
2284 __netdev_watchdog_up(dev);
2287 EXPORT_SYMBOL(netif_device_attach);
2289 static void skb_warn_bad_offload(const struct sk_buff *skb)
2291 static const netdev_features_t null_features = 0;
2292 struct net_device *dev = skb->dev;
2293 const char *driver = "";
2295 if (!net_ratelimit())
2298 if (dev && dev->dev.parent)
2299 driver = dev_driver_string(dev->dev.parent);
2301 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2302 "gso_type=%d ip_summed=%d\n",
2303 driver, dev ? &dev->features : &null_features,
2304 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2305 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2306 skb_shinfo(skb)->gso_type, skb->ip_summed);
2310 * Invalidate hardware checksum when packet is to be mangled, and
2311 * complete checksum manually on outgoing path.
2313 int skb_checksum_help(struct sk_buff *skb)
2316 int ret = 0, offset;
2318 if (skb->ip_summed == CHECKSUM_COMPLETE)
2319 goto out_set_summed;
2321 if (unlikely(skb_shinfo(skb)->gso_size)) {
2322 skb_warn_bad_offload(skb);
2326 /* Before computing a checksum, we should make sure no frag could
2327 * be modified by an external entity : checksum could be wrong.
2329 if (skb_has_shared_frag(skb)) {
2330 ret = __skb_linearize(skb);
2335 offset = skb_checksum_start_offset(skb);
2336 BUG_ON(offset >= skb_headlen(skb));
2337 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2339 offset += skb->csum_offset;
2340 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2342 if (skb_cloned(skb) &&
2343 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2344 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2349 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2351 skb->ip_summed = CHECKSUM_NONE;
2355 EXPORT_SYMBOL(skb_checksum_help);
2357 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2359 unsigned int vlan_depth = skb->mac_len;
2360 __be16 type = skb->protocol;
2362 /* Tunnel gso handlers can set protocol to ethernet. */
2363 if (type == htons(ETH_P_TEB)) {
2366 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2369 eth = (struct ethhdr *)skb_mac_header(skb);
2370 type = eth->h_proto;
2373 /* if skb->protocol is 802.1Q/AD then the header should already be
2374 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at
2375 * ETH_HLEN otherwise
2377 if (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2379 if (WARN_ON(vlan_depth < VLAN_HLEN))
2381 vlan_depth -= VLAN_HLEN;
2383 vlan_depth = ETH_HLEN;
2386 struct vlan_hdr *vh;
2388 if (unlikely(!pskb_may_pull(skb,
2389 vlan_depth + VLAN_HLEN)))
2392 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2393 type = vh->h_vlan_encapsulated_proto;
2394 vlan_depth += VLAN_HLEN;
2395 } while (type == htons(ETH_P_8021Q) ||
2396 type == htons(ETH_P_8021AD));
2399 *depth = vlan_depth;
2405 * skb_mac_gso_segment - mac layer segmentation handler.
2406 * @skb: buffer to segment
2407 * @features: features for the output path (see dev->features)
2409 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2410 netdev_features_t features)
2412 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2413 struct packet_offload *ptype;
2414 int vlan_depth = skb->mac_len;
2415 __be16 type = skb_network_protocol(skb, &vlan_depth);
2417 if (unlikely(!type))
2418 return ERR_PTR(-EINVAL);
2420 __skb_pull(skb, vlan_depth);
2423 list_for_each_entry_rcu(ptype, &offload_base, list) {
2424 if (ptype->type == type && ptype->callbacks.gso_segment) {
2425 segs = ptype->callbacks.gso_segment(skb, features);
2431 __skb_push(skb, skb->data - skb_mac_header(skb));
2435 EXPORT_SYMBOL(skb_mac_gso_segment);
2438 /* openvswitch calls this on rx path, so we need a different check.
2440 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2443 return skb->ip_summed != CHECKSUM_PARTIAL;
2445 return skb->ip_summed == CHECKSUM_NONE;
2449 * __skb_gso_segment - Perform segmentation on skb.
2450 * @skb: buffer to segment
2451 * @features: features for the output path (see dev->features)
2452 * @tx_path: whether it is called in TX path
2454 * This function segments the given skb and returns a list of segments.
2456 * It may return NULL if the skb requires no segmentation. This is
2457 * only possible when GSO is used for verifying header integrity.
2459 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2460 netdev_features_t features, bool tx_path)
2462 if (unlikely(skb_needs_check(skb, tx_path))) {
2465 skb_warn_bad_offload(skb);
2467 err = skb_cow_head(skb, 0);
2469 return ERR_PTR(err);
2472 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2473 SKB_GSO_CB(skb)->encap_level = 0;
2475 skb_reset_mac_header(skb);
2476 skb_reset_mac_len(skb);
2478 return skb_mac_gso_segment(skb, features);
2480 EXPORT_SYMBOL(__skb_gso_segment);
2482 /* Take action when hardware reception checksum errors are detected. */
2484 void netdev_rx_csum_fault(struct net_device *dev)
2486 if (net_ratelimit()) {
2487 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2491 EXPORT_SYMBOL(netdev_rx_csum_fault);
2494 /* Actually, we should eliminate this check as soon as we know, that:
2495 * 1. IOMMU is present and allows to map all the memory.
2496 * 2. No high memory really exists on this machine.
2499 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2501 #ifdef CONFIG_HIGHMEM
2503 if (!(dev->features & NETIF_F_HIGHDMA)) {
2504 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2505 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2506 if (PageHighMem(skb_frag_page(frag)))
2511 if (PCI_DMA_BUS_IS_PHYS) {
2512 struct device *pdev = dev->dev.parent;
2516 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2517 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2518 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2519 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2527 /* If MPLS offload request, verify we are testing hardware MPLS features
2528 * instead of standard features for the netdev.
2530 #ifdef CONFIG_NET_MPLS_GSO
2531 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2532 netdev_features_t features,
2535 if (type == htons(ETH_P_MPLS_UC) || type == htons(ETH_P_MPLS_MC))
2536 features &= skb->dev->mpls_features;
2541 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2542 netdev_features_t features,
2549 static netdev_features_t harmonize_features(struct sk_buff *skb,
2550 netdev_features_t features)
2555 type = skb_network_protocol(skb, &tmp);
2556 features = net_mpls_features(skb, features, type);
2558 if (skb->ip_summed != CHECKSUM_NONE &&
2559 !can_checksum_protocol(features, type)) {
2560 features &= ~NETIF_F_ALL_CSUM;
2561 } else if (illegal_highdma(skb->dev, skb)) {
2562 features &= ~NETIF_F_SG;
2568 netdev_features_t netif_skb_features(struct sk_buff *skb)
2570 const struct net_device *dev = skb->dev;
2571 netdev_features_t features = dev->features;
2572 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2573 __be16 protocol = skb->protocol;
2575 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2576 features &= ~NETIF_F_GSO_MASK;
2578 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2579 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2580 protocol = veh->h_vlan_encapsulated_proto;
2581 } else if (!vlan_tx_tag_present(skb)) {
2582 return harmonize_features(skb, features);
2585 features = netdev_intersect_features(features,
2586 dev->vlan_features |
2587 NETIF_F_HW_VLAN_CTAG_TX |
2588 NETIF_F_HW_VLAN_STAG_TX);
2590 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2591 features = netdev_intersect_features(features,
2596 NETIF_F_HW_VLAN_CTAG_TX |
2597 NETIF_F_HW_VLAN_STAG_TX);
2599 return harmonize_features(skb, features);
2601 EXPORT_SYMBOL(netif_skb_features);
2603 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2604 struct netdev_queue *txq, bool more)
2609 if (!list_empty(&ptype_all))
2610 dev_queue_xmit_nit(skb, dev);
2613 trace_net_dev_start_xmit(skb, dev);
2614 rc = netdev_start_xmit(skb, dev, txq, more);
2615 trace_net_dev_xmit(skb, rc, dev, len);
2620 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2621 struct netdev_queue *txq, int *ret)
2623 struct sk_buff *skb = first;
2624 int rc = NETDEV_TX_OK;
2627 struct sk_buff *next = skb->next;
2630 rc = xmit_one(skb, dev, txq, next != NULL);
2631 if (unlikely(!dev_xmit_complete(rc))) {
2637 if (netif_xmit_stopped(txq) && skb) {
2638 rc = NETDEV_TX_BUSY;
2648 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2649 netdev_features_t features)
2651 if (vlan_tx_tag_present(skb) &&
2652 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2653 skb = __vlan_put_tag(skb, skb->vlan_proto,
2654 vlan_tx_tag_get(skb));
2661 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2663 netdev_features_t features;
2668 /* If device doesn't need skb->dst, release it right now while
2669 * its hot in this cpu cache
2671 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2674 features = netif_skb_features(skb);
2675 skb = validate_xmit_vlan(skb, features);
2679 /* If encapsulation offload request, verify we are testing
2680 * hardware encapsulation features instead of standard
2681 * features for the netdev
2683 if (skb->encapsulation)
2684 features &= dev->hw_enc_features;
2686 if (netif_needs_gso(skb, features)) {
2687 struct sk_buff *segs;
2689 segs = skb_gso_segment(skb, features);
2697 if (skb_needs_linearize(skb, features) &&
2698 __skb_linearize(skb))
2701 /* If packet is not checksummed and device does not
2702 * support checksumming for this protocol, complete
2703 * checksumming here.
2705 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2706 if (skb->encapsulation)
2707 skb_set_inner_transport_header(skb,
2708 skb_checksum_start_offset(skb));
2710 skb_set_transport_header(skb,
2711 skb_checksum_start_offset(skb));
2712 if (!(features & NETIF_F_ALL_CSUM) &&
2713 skb_checksum_help(skb))
2726 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2728 struct sk_buff *next, *head = NULL, *tail;
2730 for (; skb != NULL; skb = next) {
2734 /* in case skb wont be segmented, point to itself */
2737 skb = validate_xmit_skb(skb, dev);
2745 /* If skb was segmented, skb->prev points to
2746 * the last segment. If not, it still contains skb.
2753 static void qdisc_pkt_len_init(struct sk_buff *skb)
2755 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2757 qdisc_skb_cb(skb)->pkt_len = skb->len;
2759 /* To get more precise estimation of bytes sent on wire,
2760 * we add to pkt_len the headers size of all segments
2762 if (shinfo->gso_size) {
2763 unsigned int hdr_len;
2764 u16 gso_segs = shinfo->gso_segs;
2766 /* mac layer + network layer */
2767 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2769 /* + transport layer */
2770 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2771 hdr_len += tcp_hdrlen(skb);
2773 hdr_len += sizeof(struct udphdr);
2775 if (shinfo->gso_type & SKB_GSO_DODGY)
2776 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2779 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2783 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2784 struct net_device *dev,
2785 struct netdev_queue *txq)
2787 spinlock_t *root_lock = qdisc_lock(q);
2791 qdisc_pkt_len_init(skb);
2792 qdisc_calculate_pkt_len(skb, q);
2794 * Heuristic to force contended enqueues to serialize on a
2795 * separate lock before trying to get qdisc main lock.
2796 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2797 * often and dequeue packets faster.
2799 contended = qdisc_is_running(q);
2800 if (unlikely(contended))
2801 spin_lock(&q->busylock);
2803 spin_lock(root_lock);
2804 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2807 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2808 qdisc_run_begin(q)) {
2810 * This is a work-conserving queue; there are no old skbs
2811 * waiting to be sent out; and the qdisc is not running -
2812 * xmit the skb directly.
2814 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2817 qdisc_bstats_update(q, skb);
2819 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2820 if (unlikely(contended)) {
2821 spin_unlock(&q->busylock);
2828 rc = NET_XMIT_SUCCESS;
2831 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2832 if (qdisc_run_begin(q)) {
2833 if (unlikely(contended)) {
2834 spin_unlock(&q->busylock);
2840 spin_unlock(root_lock);
2841 if (unlikely(contended))
2842 spin_unlock(&q->busylock);
2846 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2847 static void skb_update_prio(struct sk_buff *skb)
2849 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2851 if (!skb->priority && skb->sk && map) {
2852 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2854 if (prioidx < map->priomap_len)
2855 skb->priority = map->priomap[prioidx];
2859 #define skb_update_prio(skb)
2862 static DEFINE_PER_CPU(int, xmit_recursion);
2863 #define RECURSION_LIMIT 10
2866 * dev_loopback_xmit - loop back @skb
2867 * @skb: buffer to transmit
2869 int dev_loopback_xmit(struct sk_buff *skb)
2871 skb_reset_mac_header(skb);
2872 __skb_pull(skb, skb_network_offset(skb));
2873 skb->pkt_type = PACKET_LOOPBACK;
2874 skb->ip_summed = CHECKSUM_UNNECESSARY;
2875 WARN_ON(!skb_dst(skb));
2880 EXPORT_SYMBOL(dev_loopback_xmit);
2883 * __dev_queue_xmit - transmit a buffer
2884 * @skb: buffer to transmit
2885 * @accel_priv: private data used for L2 forwarding offload
2887 * Queue a buffer for transmission to a network device. The caller must
2888 * have set the device and priority and built the buffer before calling
2889 * this function. The function can be called from an interrupt.
2891 * A negative errno code is returned on a failure. A success does not
2892 * guarantee the frame will be transmitted as it may be dropped due
2893 * to congestion or traffic shaping.
2895 * -----------------------------------------------------------------------------------
2896 * I notice this method can also return errors from the queue disciplines,
2897 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2900 * Regardless of the return value, the skb is consumed, so it is currently
2901 * difficult to retry a send to this method. (You can bump the ref count
2902 * before sending to hold a reference for retry if you are careful.)
2904 * When calling this method, interrupts MUST be enabled. This is because
2905 * the BH enable code must have IRQs enabled so that it will not deadlock.
2908 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2910 struct net_device *dev = skb->dev;
2911 struct netdev_queue *txq;
2915 skb_reset_mac_header(skb);
2917 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
2918 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
2920 /* Disable soft irqs for various locks below. Also
2921 * stops preemption for RCU.
2925 skb_update_prio(skb);
2927 txq = netdev_pick_tx(dev, skb, accel_priv);
2928 q = rcu_dereference_bh(txq->qdisc);
2930 #ifdef CONFIG_NET_CLS_ACT
2931 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2933 trace_net_dev_queue(skb);
2935 rc = __dev_xmit_skb(skb, q, dev, txq);
2939 /* The device has no queue. Common case for software devices:
2940 loopback, all the sorts of tunnels...
2942 Really, it is unlikely that netif_tx_lock protection is necessary
2943 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2945 However, it is possible, that they rely on protection
2948 Check this and shot the lock. It is not prone from deadlocks.
2949 Either shot noqueue qdisc, it is even simpler 8)
2951 if (dev->flags & IFF_UP) {
2952 int cpu = smp_processor_id(); /* ok because BHs are off */
2954 if (txq->xmit_lock_owner != cpu) {
2956 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2957 goto recursion_alert;
2959 skb = validate_xmit_skb(skb, dev);
2963 HARD_TX_LOCK(dev, txq, cpu);
2965 if (!netif_xmit_stopped(txq)) {
2966 __this_cpu_inc(xmit_recursion);
2967 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
2968 __this_cpu_dec(xmit_recursion);
2969 if (dev_xmit_complete(rc)) {
2970 HARD_TX_UNLOCK(dev, txq);
2974 HARD_TX_UNLOCK(dev, txq);
2975 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2978 /* Recursion is detected! It is possible,
2982 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2989 rcu_read_unlock_bh();
2991 atomic_long_inc(&dev->tx_dropped);
2992 kfree_skb_list(skb);
2995 rcu_read_unlock_bh();
2999 int dev_queue_xmit(struct sk_buff *skb)
3001 return __dev_queue_xmit(skb, NULL);
3003 EXPORT_SYMBOL(dev_queue_xmit);
3005 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3007 return __dev_queue_xmit(skb, accel_priv);
3009 EXPORT_SYMBOL(dev_queue_xmit_accel);
3012 /*=======================================================================
3014 =======================================================================*/
3016 int netdev_max_backlog __read_mostly = 1000;
3017 EXPORT_SYMBOL(netdev_max_backlog);
3019 int netdev_tstamp_prequeue __read_mostly = 1;
3020 int netdev_budget __read_mostly = 300;
3021 int weight_p __read_mostly = 64; /* old backlog weight */
3023 /* Called with irq disabled */
3024 static inline void ____napi_schedule(struct softnet_data *sd,
3025 struct napi_struct *napi)
3027 list_add_tail(&napi->poll_list, &sd->poll_list);
3028 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3033 /* One global table that all flow-based protocols share. */
3034 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3035 EXPORT_SYMBOL(rps_sock_flow_table);
3037 struct static_key rps_needed __read_mostly;
3039 static struct rps_dev_flow *
3040 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3041 struct rps_dev_flow *rflow, u16 next_cpu)
3043 if (next_cpu != RPS_NO_CPU) {
3044 #ifdef CONFIG_RFS_ACCEL
3045 struct netdev_rx_queue *rxqueue;
3046 struct rps_dev_flow_table *flow_table;
3047 struct rps_dev_flow *old_rflow;
3052 /* Should we steer this flow to a different hardware queue? */
3053 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3054 !(dev->features & NETIF_F_NTUPLE))
3056 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3057 if (rxq_index == skb_get_rx_queue(skb))
3060 rxqueue = dev->_rx + rxq_index;
3061 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3064 flow_id = skb_get_hash(skb) & flow_table->mask;
3065 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3066 rxq_index, flow_id);
3070 rflow = &flow_table->flows[flow_id];
3072 if (old_rflow->filter == rflow->filter)
3073 old_rflow->filter = RPS_NO_FILTER;
3077 per_cpu(softnet_data, next_cpu).input_queue_head;
3080 rflow->cpu = next_cpu;
3085 * get_rps_cpu is called from netif_receive_skb and returns the target
3086 * CPU from the RPS map of the receiving queue for a given skb.
3087 * rcu_read_lock must be held on entry.
3089 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3090 struct rps_dev_flow **rflowp)
3092 struct netdev_rx_queue *rxqueue;
3093 struct rps_map *map;
3094 struct rps_dev_flow_table *flow_table;
3095 struct rps_sock_flow_table *sock_flow_table;
3100 if (skb_rx_queue_recorded(skb)) {
3101 u16 index = skb_get_rx_queue(skb);
3102 if (unlikely(index >= dev->real_num_rx_queues)) {
3103 WARN_ONCE(dev->real_num_rx_queues > 1,
3104 "%s received packet on queue %u, but number "
3105 "of RX queues is %u\n",
3106 dev->name, index, dev->real_num_rx_queues);
3109 rxqueue = dev->_rx + index;
3113 map = rcu_dereference(rxqueue->rps_map);
3115 if (map->len == 1 &&
3116 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3117 tcpu = map->cpus[0];
3118 if (cpu_online(tcpu))
3122 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3126 skb_reset_network_header(skb);
3127 hash = skb_get_hash(skb);
3131 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3132 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3133 if (flow_table && sock_flow_table) {
3135 struct rps_dev_flow *rflow;
3137 rflow = &flow_table->flows[hash & flow_table->mask];
3140 next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask];
3143 * If the desired CPU (where last recvmsg was done) is
3144 * different from current CPU (one in the rx-queue flow
3145 * table entry), switch if one of the following holds:
3146 * - Current CPU is unset (equal to RPS_NO_CPU).
3147 * - Current CPU is offline.
3148 * - The current CPU's queue tail has advanced beyond the
3149 * last packet that was enqueued using this table entry.
3150 * This guarantees that all previous packets for the flow
3151 * have been dequeued, thus preserving in order delivery.
3153 if (unlikely(tcpu != next_cpu) &&
3154 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3155 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3156 rflow->last_qtail)) >= 0)) {
3158 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3161 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3169 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3170 if (cpu_online(tcpu)) {
3180 #ifdef CONFIG_RFS_ACCEL
3183 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3184 * @dev: Device on which the filter was set
3185 * @rxq_index: RX queue index
3186 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3187 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3189 * Drivers that implement ndo_rx_flow_steer() should periodically call
3190 * this function for each installed filter and remove the filters for
3191 * which it returns %true.
3193 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3194 u32 flow_id, u16 filter_id)
3196 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3197 struct rps_dev_flow_table *flow_table;
3198 struct rps_dev_flow *rflow;
3203 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3204 if (flow_table && flow_id <= flow_table->mask) {
3205 rflow = &flow_table->flows[flow_id];
3206 cpu = ACCESS_ONCE(rflow->cpu);
3207 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3208 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3209 rflow->last_qtail) <
3210 (int)(10 * flow_table->mask)))
3216 EXPORT_SYMBOL(rps_may_expire_flow);
3218 #endif /* CONFIG_RFS_ACCEL */
3220 /* Called from hardirq (IPI) context */
3221 static void rps_trigger_softirq(void *data)
3223 struct softnet_data *sd = data;
3225 ____napi_schedule(sd, &sd->backlog);
3229 #endif /* CONFIG_RPS */
3232 * Check if this softnet_data structure is another cpu one
3233 * If yes, queue it to our IPI list and return 1
3236 static int rps_ipi_queued(struct softnet_data *sd)
3239 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3242 sd->rps_ipi_next = mysd->rps_ipi_list;
3243 mysd->rps_ipi_list = sd;
3245 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3248 #endif /* CONFIG_RPS */
3252 #ifdef CONFIG_NET_FLOW_LIMIT
3253 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3256 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3258 #ifdef CONFIG_NET_FLOW_LIMIT
3259 struct sd_flow_limit *fl;
3260 struct softnet_data *sd;
3261 unsigned int old_flow, new_flow;
3263 if (qlen < (netdev_max_backlog >> 1))
3266 sd = &__get_cpu_var(softnet_data);
3269 fl = rcu_dereference(sd->flow_limit);
3271 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3272 old_flow = fl->history[fl->history_head];
3273 fl->history[fl->history_head] = new_flow;
3276 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3278 if (likely(fl->buckets[old_flow]))
3279 fl->buckets[old_flow]--;
3281 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3293 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3294 * queue (may be a remote CPU queue).
3296 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3297 unsigned int *qtail)
3299 struct softnet_data *sd;
3300 unsigned long flags;
3303 sd = &per_cpu(softnet_data, cpu);
3305 local_irq_save(flags);
3308 qlen = skb_queue_len(&sd->input_pkt_queue);
3309 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3310 if (skb_queue_len(&sd->input_pkt_queue)) {
3312 __skb_queue_tail(&sd->input_pkt_queue, skb);
3313 input_queue_tail_incr_save(sd, qtail);
3315 local_irq_restore(flags);
3316 return NET_RX_SUCCESS;
3319 /* Schedule NAPI for backlog device
3320 * We can use non atomic operation since we own the queue lock
3322 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3323 if (!rps_ipi_queued(sd))
3324 ____napi_schedule(sd, &sd->backlog);
3332 local_irq_restore(flags);
3334 atomic_long_inc(&skb->dev->rx_dropped);
3339 static int netif_rx_internal(struct sk_buff *skb)
3343 net_timestamp_check(netdev_tstamp_prequeue, skb);
3345 trace_netif_rx(skb);
3347 if (static_key_false(&rps_needed)) {
3348 struct rps_dev_flow voidflow, *rflow = &voidflow;
3354 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3356 cpu = smp_processor_id();
3358 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3366 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3373 * netif_rx - post buffer to the network code
3374 * @skb: buffer to post
3376 * This function receives a packet from a device driver and queues it for
3377 * the upper (protocol) levels to process. It always succeeds. The buffer
3378 * may be dropped during processing for congestion control or by the
3382 * NET_RX_SUCCESS (no congestion)
3383 * NET_RX_DROP (packet was dropped)
3387 int netif_rx(struct sk_buff *skb)
3389 trace_netif_rx_entry(skb);
3391 return netif_rx_internal(skb);
3393 EXPORT_SYMBOL(netif_rx);
3395 int netif_rx_ni(struct sk_buff *skb)
3399 trace_netif_rx_ni_entry(skb);
3402 err = netif_rx_internal(skb);
3403 if (local_softirq_pending())
3409 EXPORT_SYMBOL(netif_rx_ni);
3411 static void net_tx_action(struct softirq_action *h)
3413 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3415 if (sd->completion_queue) {
3416 struct sk_buff *clist;
3418 local_irq_disable();
3419 clist = sd->completion_queue;
3420 sd->completion_queue = NULL;
3424 struct sk_buff *skb = clist;
3425 clist = clist->next;
3427 WARN_ON(atomic_read(&skb->users));
3428 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3429 trace_consume_skb(skb);
3431 trace_kfree_skb(skb, net_tx_action);
3436 if (sd->output_queue) {
3439 local_irq_disable();
3440 head = sd->output_queue;
3441 sd->output_queue = NULL;
3442 sd->output_queue_tailp = &sd->output_queue;
3446 struct Qdisc *q = head;
3447 spinlock_t *root_lock;
3449 head = head->next_sched;
3451 root_lock = qdisc_lock(q);
3452 if (spin_trylock(root_lock)) {
3453 smp_mb__before_atomic();
3454 clear_bit(__QDISC_STATE_SCHED,
3457 spin_unlock(root_lock);
3459 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3461 __netif_reschedule(q);
3463 smp_mb__before_atomic();
3464 clear_bit(__QDISC_STATE_SCHED,
3472 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3473 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3474 /* This hook is defined here for ATM LANE */
3475 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3476 unsigned char *addr) __read_mostly;
3477 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3480 #ifdef CONFIG_NET_CLS_ACT
3481 /* TODO: Maybe we should just force sch_ingress to be compiled in
3482 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3483 * a compare and 2 stores extra right now if we dont have it on
3484 * but have CONFIG_NET_CLS_ACT
3485 * NOTE: This doesn't stop any functionality; if you dont have
3486 * the ingress scheduler, you just can't add policies on ingress.
3489 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3491 struct net_device *dev = skb->dev;
3492 u32 ttl = G_TC_RTTL(skb->tc_verd);
3493 int result = TC_ACT_OK;
3496 if (unlikely(MAX_RED_LOOP < ttl++)) {
3497 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3498 skb->skb_iif, dev->ifindex);
3502 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3503 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3505 q = rcu_dereference(rxq->qdisc);
3506 if (q != &noop_qdisc) {
3507 spin_lock(qdisc_lock(q));
3508 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3509 result = qdisc_enqueue_root(skb, q);
3510 spin_unlock(qdisc_lock(q));
3516 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3517 struct packet_type **pt_prev,
3518 int *ret, struct net_device *orig_dev)
3520 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3522 if (!rxq || rcu_access_pointer(rxq->qdisc) == &noop_qdisc)
3526 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3530 switch (ing_filter(skb, rxq)) {
3544 * netdev_rx_handler_register - register receive handler
3545 * @dev: device to register a handler for
3546 * @rx_handler: receive handler to register
3547 * @rx_handler_data: data pointer that is used by rx handler
3549 * Register a receive handler for a device. This handler will then be
3550 * called from __netif_receive_skb. A negative errno code is returned
3553 * The caller must hold the rtnl_mutex.
3555 * For a general description of rx_handler, see enum rx_handler_result.
3557 int netdev_rx_handler_register(struct net_device *dev,
3558 rx_handler_func_t *rx_handler,
3559 void *rx_handler_data)
3563 if (dev->rx_handler)
3566 /* Note: rx_handler_data must be set before rx_handler */
3567 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3568 rcu_assign_pointer(dev->rx_handler, rx_handler);
3572 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3575 * netdev_rx_handler_unregister - unregister receive handler
3576 * @dev: device to unregister a handler from
3578 * Unregister a receive handler from a device.
3580 * The caller must hold the rtnl_mutex.
3582 void netdev_rx_handler_unregister(struct net_device *dev)
3586 RCU_INIT_POINTER(dev->rx_handler, NULL);
3587 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3588 * section has a guarantee to see a non NULL rx_handler_data
3592 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3594 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3597 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3598 * the special handling of PFMEMALLOC skbs.
3600 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3602 switch (skb->protocol) {
3603 case htons(ETH_P_ARP):
3604 case htons(ETH_P_IP):
3605 case htons(ETH_P_IPV6):
3606 case htons(ETH_P_8021Q):
3607 case htons(ETH_P_8021AD):
3614 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3616 struct packet_type *ptype, *pt_prev;
3617 rx_handler_func_t *rx_handler;
3618 struct net_device *orig_dev;
3619 struct net_device *null_or_dev;
3620 bool deliver_exact = false;
3621 int ret = NET_RX_DROP;
3624 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3626 trace_netif_receive_skb(skb);
3628 orig_dev = skb->dev;
3630 skb_reset_network_header(skb);
3631 if (!skb_transport_header_was_set(skb))
3632 skb_reset_transport_header(skb);
3633 skb_reset_mac_len(skb);
3640 skb->skb_iif = skb->dev->ifindex;
3642 __this_cpu_inc(softnet_data.processed);
3644 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3645 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3646 skb = skb_vlan_untag(skb);
3651 #ifdef CONFIG_NET_CLS_ACT
3652 if (skb->tc_verd & TC_NCLS) {
3653 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3661 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3662 if (!ptype->dev || ptype->dev == skb->dev) {
3664 ret = deliver_skb(skb, pt_prev, orig_dev);
3670 #ifdef CONFIG_NET_CLS_ACT
3671 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3677 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3680 if (vlan_tx_tag_present(skb)) {
3682 ret = deliver_skb(skb, pt_prev, orig_dev);
3685 if (vlan_do_receive(&skb))
3687 else if (unlikely(!skb))
3691 rx_handler = rcu_dereference(skb->dev->rx_handler);
3694 ret = deliver_skb(skb, pt_prev, orig_dev);
3697 switch (rx_handler(&skb)) {
3698 case RX_HANDLER_CONSUMED:
3699 ret = NET_RX_SUCCESS;
3701 case RX_HANDLER_ANOTHER:
3703 case RX_HANDLER_EXACT:
3704 deliver_exact = true;
3705 case RX_HANDLER_PASS:
3712 if (unlikely(vlan_tx_tag_present(skb))) {
3713 if (vlan_tx_tag_get_id(skb))
3714 skb->pkt_type = PACKET_OTHERHOST;
3715 /* Note: we might in the future use prio bits
3716 * and set skb->priority like in vlan_do_receive()
3717 * For the time being, just ignore Priority Code Point
3722 /* deliver only exact match when indicated */
3723 null_or_dev = deliver_exact ? skb->dev : NULL;
3725 type = skb->protocol;
3726 list_for_each_entry_rcu(ptype,
3727 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3728 if (ptype->type == type &&
3729 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3730 ptype->dev == orig_dev)) {
3732 ret = deliver_skb(skb, pt_prev, orig_dev);
3738 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3741 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3744 atomic_long_inc(&skb->dev->rx_dropped);
3746 /* Jamal, now you will not able to escape explaining
3747 * me how you were going to use this. :-)
3757 static int __netif_receive_skb(struct sk_buff *skb)
3761 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3762 unsigned long pflags = current->flags;
3765 * PFMEMALLOC skbs are special, they should
3766 * - be delivered to SOCK_MEMALLOC sockets only
3767 * - stay away from userspace
3768 * - have bounded memory usage
3770 * Use PF_MEMALLOC as this saves us from propagating the allocation
3771 * context down to all allocation sites.
3773 current->flags |= PF_MEMALLOC;
3774 ret = __netif_receive_skb_core(skb, true);
3775 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3777 ret = __netif_receive_skb_core(skb, false);
3782 static int netif_receive_skb_internal(struct sk_buff *skb)
3784 net_timestamp_check(netdev_tstamp_prequeue, skb);
3786 if (skb_defer_rx_timestamp(skb))
3787 return NET_RX_SUCCESS;
3790 if (static_key_false(&rps_needed)) {
3791 struct rps_dev_flow voidflow, *rflow = &voidflow;
3796 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3799 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3806 return __netif_receive_skb(skb);
3810 * netif_receive_skb - process receive buffer from network
3811 * @skb: buffer to process
3813 * netif_receive_skb() is the main receive data processing function.
3814 * It always succeeds. The buffer may be dropped during processing
3815 * for congestion control or by the protocol layers.
3817 * This function may only be called from softirq context and interrupts
3818 * should be enabled.
3820 * Return values (usually ignored):
3821 * NET_RX_SUCCESS: no congestion
3822 * NET_RX_DROP: packet was dropped
3824 int netif_receive_skb(struct sk_buff *skb)
3826 trace_netif_receive_skb_entry(skb);
3828 return netif_receive_skb_internal(skb);
3830 EXPORT_SYMBOL(netif_receive_skb);
3832 /* Network device is going away, flush any packets still pending
3833 * Called with irqs disabled.
3835 static void flush_backlog(void *arg)
3837 struct net_device *dev = arg;
3838 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3839 struct sk_buff *skb, *tmp;
3842 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3843 if (skb->dev == dev) {
3844 __skb_unlink(skb, &sd->input_pkt_queue);
3846 input_queue_head_incr(sd);
3851 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3852 if (skb->dev == dev) {
3853 __skb_unlink(skb, &sd->process_queue);
3855 input_queue_head_incr(sd);
3860 static int napi_gro_complete(struct sk_buff *skb)
3862 struct packet_offload *ptype;
3863 __be16 type = skb->protocol;
3864 struct list_head *head = &offload_base;
3867 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3869 if (NAPI_GRO_CB(skb)->count == 1) {
3870 skb_shinfo(skb)->gso_size = 0;
3875 list_for_each_entry_rcu(ptype, head, list) {
3876 if (ptype->type != type || !ptype->callbacks.gro_complete)
3879 err = ptype->callbacks.gro_complete(skb, 0);
3885 WARN_ON(&ptype->list == head);
3887 return NET_RX_SUCCESS;
3891 return netif_receive_skb_internal(skb);
3894 /* napi->gro_list contains packets ordered by age.
3895 * youngest packets at the head of it.
3896 * Complete skbs in reverse order to reduce latencies.
3898 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3900 struct sk_buff *skb, *prev = NULL;
3902 /* scan list and build reverse chain */
3903 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3908 for (skb = prev; skb; skb = prev) {
3911 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3915 napi_gro_complete(skb);
3919 napi->gro_list = NULL;
3921 EXPORT_SYMBOL(napi_gro_flush);
3923 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3926 unsigned int maclen = skb->dev->hard_header_len;
3927 u32 hash = skb_get_hash_raw(skb);
3929 for (p = napi->gro_list; p; p = p->next) {
3930 unsigned long diffs;
3932 NAPI_GRO_CB(p)->flush = 0;
3934 if (hash != skb_get_hash_raw(p)) {
3935 NAPI_GRO_CB(p)->same_flow = 0;
3939 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3940 diffs |= p->vlan_tci ^ skb->vlan_tci;
3941 if (maclen == ETH_HLEN)
3942 diffs |= compare_ether_header(skb_mac_header(p),
3943 skb_mac_header(skb));
3945 diffs = memcmp(skb_mac_header(p),
3946 skb_mac_header(skb),
3948 NAPI_GRO_CB(p)->same_flow = !diffs;
3952 static void skb_gro_reset_offset(struct sk_buff *skb)
3954 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3955 const skb_frag_t *frag0 = &pinfo->frags[0];
3957 NAPI_GRO_CB(skb)->data_offset = 0;
3958 NAPI_GRO_CB(skb)->frag0 = NULL;
3959 NAPI_GRO_CB(skb)->frag0_len = 0;
3961 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3963 !PageHighMem(skb_frag_page(frag0))) {
3964 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3965 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3969 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
3971 struct skb_shared_info *pinfo = skb_shinfo(skb);
3973 BUG_ON(skb->end - skb->tail < grow);
3975 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3977 skb->data_len -= grow;
3980 pinfo->frags[0].page_offset += grow;
3981 skb_frag_size_sub(&pinfo->frags[0], grow);
3983 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
3984 skb_frag_unref(skb, 0);
3985 memmove(pinfo->frags, pinfo->frags + 1,
3986 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
3990 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3992 struct sk_buff **pp = NULL;
3993 struct packet_offload *ptype;
3994 __be16 type = skb->protocol;
3995 struct list_head *head = &offload_base;
3997 enum gro_result ret;
4000 if (!(skb->dev->features & NETIF_F_GRO))
4003 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4006 gro_list_prepare(napi, skb);
4009 list_for_each_entry_rcu(ptype, head, list) {
4010 if (ptype->type != type || !ptype->callbacks.gro_receive)
4013 skb_set_network_header(skb, skb_gro_offset(skb));
4014 skb_reset_mac_len(skb);
4015 NAPI_GRO_CB(skb)->same_flow = 0;
4016 NAPI_GRO_CB(skb)->flush = 0;
4017 NAPI_GRO_CB(skb)->free = 0;
4018 NAPI_GRO_CB(skb)->udp_mark = 0;
4020 /* Setup for GRO checksum validation */
4021 switch (skb->ip_summed) {
4022 case CHECKSUM_COMPLETE:
4023 NAPI_GRO_CB(skb)->csum = skb->csum;
4024 NAPI_GRO_CB(skb)->csum_valid = 1;
4025 NAPI_GRO_CB(skb)->csum_cnt = 0;
4027 case CHECKSUM_UNNECESSARY:
4028 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4029 NAPI_GRO_CB(skb)->csum_valid = 0;
4032 NAPI_GRO_CB(skb)->csum_cnt = 0;
4033 NAPI_GRO_CB(skb)->csum_valid = 0;
4036 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4041 if (&ptype->list == head)
4044 same_flow = NAPI_GRO_CB(skb)->same_flow;
4045 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4048 struct sk_buff *nskb = *pp;
4052 napi_gro_complete(nskb);
4059 if (NAPI_GRO_CB(skb)->flush)
4062 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4063 struct sk_buff *nskb = napi->gro_list;
4065 /* locate the end of the list to select the 'oldest' flow */
4066 while (nskb->next) {
4072 napi_gro_complete(nskb);
4076 NAPI_GRO_CB(skb)->count = 1;
4077 NAPI_GRO_CB(skb)->age = jiffies;
4078 NAPI_GRO_CB(skb)->last = skb;
4079 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4080 skb->next = napi->gro_list;
4081 napi->gro_list = skb;
4085 grow = skb_gro_offset(skb) - skb_headlen(skb);
4087 gro_pull_from_frag0(skb, grow);
4096 struct packet_offload *gro_find_receive_by_type(__be16 type)
4098 struct list_head *offload_head = &offload_base;
4099 struct packet_offload *ptype;
4101 list_for_each_entry_rcu(ptype, offload_head, list) {
4102 if (ptype->type != type || !ptype->callbacks.gro_receive)
4108 EXPORT_SYMBOL(gro_find_receive_by_type);
4110 struct packet_offload *gro_find_complete_by_type(__be16 type)
4112 struct list_head *offload_head = &offload_base;
4113 struct packet_offload *ptype;
4115 list_for_each_entry_rcu(ptype, offload_head, list) {
4116 if (ptype->type != type || !ptype->callbacks.gro_complete)
4122 EXPORT_SYMBOL(gro_find_complete_by_type);
4124 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4128 if (netif_receive_skb_internal(skb))
4136 case GRO_MERGED_FREE:
4137 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4138 kmem_cache_free(skbuff_head_cache, skb);
4151 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4153 trace_napi_gro_receive_entry(skb);
4155 skb_gro_reset_offset(skb);
4157 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4159 EXPORT_SYMBOL(napi_gro_receive);
4161 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4163 __skb_pull(skb, skb_headlen(skb));
4164 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4165 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4167 skb->dev = napi->dev;
4169 skb->encapsulation = 0;
4170 skb_shinfo(skb)->gso_type = 0;
4171 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4176 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4178 struct sk_buff *skb = napi->skb;
4181 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4186 EXPORT_SYMBOL(napi_get_frags);
4188 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4189 struct sk_buff *skb,
4195 __skb_push(skb, ETH_HLEN);
4196 skb->protocol = eth_type_trans(skb, skb->dev);
4197 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4202 case GRO_MERGED_FREE:
4203 napi_reuse_skb(napi, skb);
4213 /* Upper GRO stack assumes network header starts at gro_offset=0
4214 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4215 * We copy ethernet header into skb->data to have a common layout.
4217 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4219 struct sk_buff *skb = napi->skb;
4220 const struct ethhdr *eth;
4221 unsigned int hlen = sizeof(*eth);
4225 skb_reset_mac_header(skb);
4226 skb_gro_reset_offset(skb);
4228 eth = skb_gro_header_fast(skb, 0);
4229 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4230 eth = skb_gro_header_slow(skb, hlen, 0);
4231 if (unlikely(!eth)) {
4232 napi_reuse_skb(napi, skb);
4236 gro_pull_from_frag0(skb, hlen);
4237 NAPI_GRO_CB(skb)->frag0 += hlen;
4238 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4240 __skb_pull(skb, hlen);
4243 * This works because the only protocols we care about don't require
4245 * We'll fix it up properly in napi_frags_finish()
4247 skb->protocol = eth->h_proto;
4252 gro_result_t napi_gro_frags(struct napi_struct *napi)
4254 struct sk_buff *skb = napi_frags_skb(napi);
4259 trace_napi_gro_frags_entry(skb);
4261 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4263 EXPORT_SYMBOL(napi_gro_frags);
4265 /* Compute the checksum from gro_offset and return the folded value
4266 * after adding in any pseudo checksum.
4268 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4273 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4275 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4276 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4278 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4279 !skb->csum_complete_sw)
4280 netdev_rx_csum_fault(skb->dev);
4283 NAPI_GRO_CB(skb)->csum = wsum;
4284 NAPI_GRO_CB(skb)->csum_valid = 1;
4288 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4291 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4292 * Note: called with local irq disabled, but exits with local irq enabled.
4294 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4297 struct softnet_data *remsd = sd->rps_ipi_list;
4300 sd->rps_ipi_list = NULL;
4304 /* Send pending IPI's to kick RPS processing on remote cpus. */
4306 struct softnet_data *next = remsd->rps_ipi_next;
4308 if (cpu_online(remsd->cpu))
4309 smp_call_function_single_async(remsd->cpu,
4318 static int process_backlog(struct napi_struct *napi, int quota)
4321 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4324 /* Check if we have pending ipi, its better to send them now,
4325 * not waiting net_rx_action() end.
4327 if (sd->rps_ipi_list) {
4328 local_irq_disable();
4329 net_rps_action_and_irq_enable(sd);
4332 napi->weight = weight_p;
4333 local_irq_disable();
4335 struct sk_buff *skb;
4337 while ((skb = __skb_dequeue(&sd->process_queue))) {
4339 __netif_receive_skb(skb);
4340 local_irq_disable();
4341 input_queue_head_incr(sd);
4342 if (++work >= quota) {
4349 if (skb_queue_empty(&sd->input_pkt_queue)) {
4351 * Inline a custom version of __napi_complete().
4352 * only current cpu owns and manipulates this napi,
4353 * and NAPI_STATE_SCHED is the only possible flag set
4355 * We can use a plain write instead of clear_bit(),
4356 * and we dont need an smp_mb() memory barrier.
4358 list_del(&napi->poll_list);
4365 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4366 &sd->process_queue);
4375 * __napi_schedule - schedule for receive
4376 * @n: entry to schedule
4378 * The entry's receive function will be scheduled to run
4380 void __napi_schedule(struct napi_struct *n)
4382 unsigned long flags;
4384 local_irq_save(flags);
4385 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4386 local_irq_restore(flags);
4388 EXPORT_SYMBOL(__napi_schedule);
4390 void __napi_complete(struct napi_struct *n)
4392 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4393 BUG_ON(n->gro_list);
4395 list_del(&n->poll_list);
4396 smp_mb__before_atomic();
4397 clear_bit(NAPI_STATE_SCHED, &n->state);
4399 EXPORT_SYMBOL(__napi_complete);
4401 void napi_complete(struct napi_struct *n)
4403 unsigned long flags;
4406 * don't let napi dequeue from the cpu poll list
4407 * just in case its running on a different cpu
4409 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4412 napi_gro_flush(n, false);
4413 local_irq_save(flags);
4415 local_irq_restore(flags);
4417 EXPORT_SYMBOL(napi_complete);
4419 /* must be called under rcu_read_lock(), as we dont take a reference */
4420 struct napi_struct *napi_by_id(unsigned int napi_id)
4422 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4423 struct napi_struct *napi;
4425 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4426 if (napi->napi_id == napi_id)
4431 EXPORT_SYMBOL_GPL(napi_by_id);
4433 void napi_hash_add(struct napi_struct *napi)
4435 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4437 spin_lock(&napi_hash_lock);
4439 /* 0 is not a valid id, we also skip an id that is taken
4440 * we expect both events to be extremely rare
4443 while (!napi->napi_id) {
4444 napi->napi_id = ++napi_gen_id;
4445 if (napi_by_id(napi->napi_id))
4449 hlist_add_head_rcu(&napi->napi_hash_node,
4450 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4452 spin_unlock(&napi_hash_lock);
4455 EXPORT_SYMBOL_GPL(napi_hash_add);
4457 /* Warning : caller is responsible to make sure rcu grace period
4458 * is respected before freeing memory containing @napi
4460 void napi_hash_del(struct napi_struct *napi)
4462 spin_lock(&napi_hash_lock);
4464 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4465 hlist_del_rcu(&napi->napi_hash_node);
4467 spin_unlock(&napi_hash_lock);
4469 EXPORT_SYMBOL_GPL(napi_hash_del);
4471 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4472 int (*poll)(struct napi_struct *, int), int weight)
4474 INIT_LIST_HEAD(&napi->poll_list);
4475 napi->gro_count = 0;
4476 napi->gro_list = NULL;
4479 if (weight > NAPI_POLL_WEIGHT)
4480 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4482 napi->weight = weight;
4483 list_add(&napi->dev_list, &dev->napi_list);
4485 #ifdef CONFIG_NETPOLL
4486 spin_lock_init(&napi->poll_lock);
4487 napi->poll_owner = -1;
4489 set_bit(NAPI_STATE_SCHED, &napi->state);
4491 EXPORT_SYMBOL(netif_napi_add);
4493 void netif_napi_del(struct napi_struct *napi)
4495 list_del_init(&napi->dev_list);
4496 napi_free_frags(napi);
4498 kfree_skb_list(napi->gro_list);
4499 napi->gro_list = NULL;
4500 napi->gro_count = 0;
4502 EXPORT_SYMBOL(netif_napi_del);
4504 static void net_rx_action(struct softirq_action *h)
4506 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4507 unsigned long time_limit = jiffies + 2;
4508 int budget = netdev_budget;
4511 local_irq_disable();
4513 while (!list_empty(&sd->poll_list)) {
4514 struct napi_struct *n;
4517 /* If softirq window is exhuasted then punt.
4518 * Allow this to run for 2 jiffies since which will allow
4519 * an average latency of 1.5/HZ.
4521 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4526 /* Even though interrupts have been re-enabled, this
4527 * access is safe because interrupts can only add new
4528 * entries to the tail of this list, and only ->poll()
4529 * calls can remove this head entry from the list.
4531 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4533 have = netpoll_poll_lock(n);
4537 /* This NAPI_STATE_SCHED test is for avoiding a race
4538 * with netpoll's poll_napi(). Only the entity which
4539 * obtains the lock and sees NAPI_STATE_SCHED set will
4540 * actually make the ->poll() call. Therefore we avoid
4541 * accidentally calling ->poll() when NAPI is not scheduled.
4544 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4545 work = n->poll(n, weight);
4549 WARN_ON_ONCE(work > weight);
4553 local_irq_disable();
4555 /* Drivers must not modify the NAPI state if they
4556 * consume the entire weight. In such cases this code
4557 * still "owns" the NAPI instance and therefore can
4558 * move the instance around on the list at-will.
4560 if (unlikely(work == weight)) {
4561 if (unlikely(napi_disable_pending(n))) {
4564 local_irq_disable();
4567 /* flush too old packets
4568 * If HZ < 1000, flush all packets.
4571 napi_gro_flush(n, HZ >= 1000);
4572 local_irq_disable();
4574 list_move_tail(&n->poll_list, &sd->poll_list);
4578 netpoll_poll_unlock(have);
4581 net_rps_action_and_irq_enable(sd);
4583 #ifdef CONFIG_NET_DMA
4585 * There may not be any more sk_buffs coming right now, so push
4586 * any pending DMA copies to hardware
4588 dma_issue_pending_all();
4595 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4599 struct netdev_adjacent {
4600 struct net_device *dev;
4602 /* upper master flag, there can only be one master device per list */
4605 /* counter for the number of times this device was added to us */
4608 /* private field for the users */
4611 struct list_head list;
4612 struct rcu_head rcu;
4615 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4616 struct net_device *adj_dev,
4617 struct list_head *adj_list)
4619 struct netdev_adjacent *adj;
4621 list_for_each_entry(adj, adj_list, list) {
4622 if (adj->dev == adj_dev)
4629 * netdev_has_upper_dev - Check if device is linked to an upper device
4631 * @upper_dev: upper device to check
4633 * Find out if a device is linked to specified upper device and return true
4634 * in case it is. Note that this checks only immediate upper device,
4635 * not through a complete stack of devices. The caller must hold the RTNL lock.
4637 bool netdev_has_upper_dev(struct net_device *dev,
4638 struct net_device *upper_dev)
4642 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4644 EXPORT_SYMBOL(netdev_has_upper_dev);
4647 * netdev_has_any_upper_dev - Check if device is linked to some device
4650 * Find out if a device is linked to an upper device and return true in case
4651 * it is. The caller must hold the RTNL lock.
4653 static bool netdev_has_any_upper_dev(struct net_device *dev)
4657 return !list_empty(&dev->all_adj_list.upper);
4661 * netdev_master_upper_dev_get - Get master upper device
4664 * Find a master upper device and return pointer to it or NULL in case
4665 * it's not there. The caller must hold the RTNL lock.
4667 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4669 struct netdev_adjacent *upper;
4673 if (list_empty(&dev->adj_list.upper))
4676 upper = list_first_entry(&dev->adj_list.upper,
4677 struct netdev_adjacent, list);
4678 if (likely(upper->master))
4682 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4684 void *netdev_adjacent_get_private(struct list_head *adj_list)
4686 struct netdev_adjacent *adj;
4688 adj = list_entry(adj_list, struct netdev_adjacent, list);
4690 return adj->private;
4692 EXPORT_SYMBOL(netdev_adjacent_get_private);
4695 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4697 * @iter: list_head ** of the current position
4699 * Gets the next device from the dev's upper list, starting from iter
4700 * position. The caller must hold RCU read lock.
4702 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4703 struct list_head **iter)
4705 struct netdev_adjacent *upper;
4707 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4709 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4711 if (&upper->list == &dev->adj_list.upper)
4714 *iter = &upper->list;
4718 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4721 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4723 * @iter: list_head ** of the current position
4725 * Gets the next device from the dev's upper list, starting from iter
4726 * position. The caller must hold RCU read lock.
4728 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4729 struct list_head **iter)
4731 struct netdev_adjacent *upper;
4733 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4735 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4737 if (&upper->list == &dev->all_adj_list.upper)
4740 *iter = &upper->list;
4744 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4747 * netdev_lower_get_next_private - Get the next ->private from the
4748 * lower neighbour list
4750 * @iter: list_head ** of the current position
4752 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4753 * list, starting from iter position. The caller must hold either hold the
4754 * RTNL lock or its own locking that guarantees that the neighbour lower
4755 * list will remain unchainged.
4757 void *netdev_lower_get_next_private(struct net_device *dev,
4758 struct list_head **iter)
4760 struct netdev_adjacent *lower;
4762 lower = list_entry(*iter, struct netdev_adjacent, list);
4764 if (&lower->list == &dev->adj_list.lower)
4767 *iter = lower->list.next;
4769 return lower->private;
4771 EXPORT_SYMBOL(netdev_lower_get_next_private);
4774 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4775 * lower neighbour list, RCU
4778 * @iter: list_head ** of the current position
4780 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4781 * list, starting from iter position. The caller must hold RCU read lock.
4783 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4784 struct list_head **iter)
4786 struct netdev_adjacent *lower;
4788 WARN_ON_ONCE(!rcu_read_lock_held());
4790 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4792 if (&lower->list == &dev->adj_list.lower)
4795 *iter = &lower->list;
4797 return lower->private;
4799 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4802 * netdev_lower_get_next - Get the next device from the lower neighbour
4805 * @iter: list_head ** of the current position
4807 * Gets the next netdev_adjacent from the dev's lower neighbour
4808 * list, starting from iter position. The caller must hold RTNL lock or
4809 * its own locking that guarantees that the neighbour lower
4810 * list will remain unchainged.
4812 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4814 struct netdev_adjacent *lower;
4816 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4818 if (&lower->list == &dev->adj_list.lower)
4821 *iter = &lower->list;
4825 EXPORT_SYMBOL(netdev_lower_get_next);
4828 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4829 * lower neighbour list, RCU
4833 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4834 * list. The caller must hold RCU read lock.
4836 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4838 struct netdev_adjacent *lower;
4840 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4841 struct netdev_adjacent, list);
4843 return lower->private;
4846 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4849 * netdev_master_upper_dev_get_rcu - Get master upper device
4852 * Find a master upper device and return pointer to it or NULL in case
4853 * it's not there. The caller must hold the RCU read lock.
4855 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4857 struct netdev_adjacent *upper;
4859 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4860 struct netdev_adjacent, list);
4861 if (upper && likely(upper->master))
4865 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4867 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4868 struct net_device *adj_dev,
4869 struct list_head *dev_list)
4871 char linkname[IFNAMSIZ+7];
4872 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4873 "upper_%s" : "lower_%s", adj_dev->name);
4874 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4877 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4879 struct list_head *dev_list)
4881 char linkname[IFNAMSIZ+7];
4882 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4883 "upper_%s" : "lower_%s", name);
4884 sysfs_remove_link(&(dev->dev.kobj), linkname);
4887 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
4888 struct net_device *adj_dev,
4889 struct list_head *dev_list)
4891 return (dev_list == &dev->adj_list.upper ||
4892 dev_list == &dev->adj_list.lower) &&
4893 net_eq(dev_net(dev), dev_net(adj_dev));
4896 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4897 struct net_device *adj_dev,
4898 struct list_head *dev_list,
4899 void *private, bool master)
4901 struct netdev_adjacent *adj;
4904 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4911 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4916 adj->master = master;
4918 adj->private = private;
4921 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4922 adj_dev->name, dev->name, adj_dev->name);
4924 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
4925 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
4930 /* Ensure that master link is always the first item in list. */
4932 ret = sysfs_create_link(&(dev->dev.kobj),
4933 &(adj_dev->dev.kobj), "master");
4935 goto remove_symlinks;
4937 list_add_rcu(&adj->list, dev_list);
4939 list_add_tail_rcu(&adj->list, dev_list);
4945 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
4946 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4954 static void __netdev_adjacent_dev_remove(struct net_device *dev,
4955 struct net_device *adj_dev,
4956 struct list_head *dev_list)
4958 struct netdev_adjacent *adj;
4960 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4963 pr_err("tried to remove device %s from %s\n",
4964 dev->name, adj_dev->name);
4968 if (adj->ref_nr > 1) {
4969 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4976 sysfs_remove_link(&(dev->dev.kobj), "master");
4978 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
4979 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4981 list_del_rcu(&adj->list);
4982 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4983 adj_dev->name, dev->name, adj_dev->name);
4985 kfree_rcu(adj, rcu);
4988 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4989 struct net_device *upper_dev,
4990 struct list_head *up_list,
4991 struct list_head *down_list,
4992 void *private, bool master)
4996 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5001 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5004 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5011 static int __netdev_adjacent_dev_link(struct net_device *dev,
5012 struct net_device *upper_dev)
5014 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5015 &dev->all_adj_list.upper,
5016 &upper_dev->all_adj_list.lower,
5020 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5021 struct net_device *upper_dev,
5022 struct list_head *up_list,
5023 struct list_head *down_list)
5025 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5026 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5029 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5030 struct net_device *upper_dev)
5032 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5033 &dev->all_adj_list.upper,
5034 &upper_dev->all_adj_list.lower);
5037 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5038 struct net_device *upper_dev,
5039 void *private, bool master)
5041 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5046 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5047 &dev->adj_list.upper,
5048 &upper_dev->adj_list.lower,
5051 __netdev_adjacent_dev_unlink(dev, upper_dev);
5058 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5059 struct net_device *upper_dev)
5061 __netdev_adjacent_dev_unlink(dev, upper_dev);
5062 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5063 &dev->adj_list.upper,
5064 &upper_dev->adj_list.lower);
5067 static int __netdev_upper_dev_link(struct net_device *dev,
5068 struct net_device *upper_dev, bool master,
5071 struct netdev_adjacent *i, *j, *to_i, *to_j;
5076 if (dev == upper_dev)
5079 /* To prevent loops, check if dev is not upper device to upper_dev. */
5080 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5083 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
5086 if (master && netdev_master_upper_dev_get(dev))
5089 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5094 /* Now that we linked these devs, make all the upper_dev's
5095 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5096 * versa, and don't forget the devices itself. All of these
5097 * links are non-neighbours.
5099 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5100 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5101 pr_debug("Interlinking %s with %s, non-neighbour\n",
5102 i->dev->name, j->dev->name);
5103 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5109 /* add dev to every upper_dev's upper device */
5110 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5111 pr_debug("linking %s's upper device %s with %s\n",
5112 upper_dev->name, i->dev->name, dev->name);
5113 ret = __netdev_adjacent_dev_link(dev, i->dev);
5115 goto rollback_upper_mesh;
5118 /* add upper_dev to every dev's lower device */
5119 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5120 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5121 i->dev->name, upper_dev->name);
5122 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5124 goto rollback_lower_mesh;
5127 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5130 rollback_lower_mesh:
5132 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5135 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5140 rollback_upper_mesh:
5142 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5145 __netdev_adjacent_dev_unlink(dev, i->dev);
5153 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5154 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5155 if (i == to_i && j == to_j)
5157 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5163 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5169 * netdev_upper_dev_link - Add a link to the upper device
5171 * @upper_dev: new upper device
5173 * Adds a link to device which is upper to this one. The caller must hold
5174 * the RTNL lock. On a failure a negative errno code is returned.
5175 * On success the reference counts are adjusted and the function
5178 int netdev_upper_dev_link(struct net_device *dev,
5179 struct net_device *upper_dev)
5181 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5183 EXPORT_SYMBOL(netdev_upper_dev_link);
5186 * netdev_master_upper_dev_link - Add a master link to the upper device
5188 * @upper_dev: new upper device
5190 * Adds a link to device which is upper to this one. In this case, only
5191 * one master upper device can be linked, although other non-master devices
5192 * might be linked as well. The caller must hold the RTNL lock.
5193 * On a failure a negative errno code is returned. On success the reference
5194 * counts are adjusted and the function returns zero.
5196 int netdev_master_upper_dev_link(struct net_device *dev,
5197 struct net_device *upper_dev)
5199 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5201 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5203 int netdev_master_upper_dev_link_private(struct net_device *dev,
5204 struct net_device *upper_dev,
5207 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5209 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5212 * netdev_upper_dev_unlink - Removes a link to upper device
5214 * @upper_dev: new upper device
5216 * Removes a link to device which is upper to this one. The caller must hold
5219 void netdev_upper_dev_unlink(struct net_device *dev,
5220 struct net_device *upper_dev)
5222 struct netdev_adjacent *i, *j;
5225 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5227 /* Here is the tricky part. We must remove all dev's lower
5228 * devices from all upper_dev's upper devices and vice
5229 * versa, to maintain the graph relationship.
5231 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5232 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5233 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5235 /* remove also the devices itself from lower/upper device
5238 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5239 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5241 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5242 __netdev_adjacent_dev_unlink(dev, i->dev);
5244 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5246 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5248 void netdev_adjacent_add_links(struct net_device *dev)
5250 struct netdev_adjacent *iter;
5252 struct net *net = dev_net(dev);
5254 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5255 if (!net_eq(net,dev_net(iter->dev)))
5257 netdev_adjacent_sysfs_add(iter->dev, dev,
5258 &iter->dev->adj_list.lower);
5259 netdev_adjacent_sysfs_add(dev, iter->dev,
5260 &dev->adj_list.upper);
5263 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5264 if (!net_eq(net,dev_net(iter->dev)))
5266 netdev_adjacent_sysfs_add(iter->dev, dev,
5267 &iter->dev->adj_list.upper);
5268 netdev_adjacent_sysfs_add(dev, iter->dev,
5269 &dev->adj_list.lower);
5273 void netdev_adjacent_del_links(struct net_device *dev)
5275 struct netdev_adjacent *iter;
5277 struct net *net = dev_net(dev);
5279 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5280 if (!net_eq(net,dev_net(iter->dev)))
5282 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5283 &iter->dev->adj_list.lower);
5284 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5285 &dev->adj_list.upper);
5288 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5289 if (!net_eq(net,dev_net(iter->dev)))
5291 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5292 &iter->dev->adj_list.upper);
5293 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5294 &dev->adj_list.lower);
5298 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5300 struct netdev_adjacent *iter;
5302 struct net *net = dev_net(dev);
5304 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5305 if (!net_eq(net,dev_net(iter->dev)))
5307 netdev_adjacent_sysfs_del(iter->dev, oldname,
5308 &iter->dev->adj_list.lower);
5309 netdev_adjacent_sysfs_add(iter->dev, dev,
5310 &iter->dev->adj_list.lower);
5313 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5314 if (!net_eq(net,dev_net(iter->dev)))
5316 netdev_adjacent_sysfs_del(iter->dev, oldname,
5317 &iter->dev->adj_list.upper);
5318 netdev_adjacent_sysfs_add(iter->dev, dev,
5319 &iter->dev->adj_list.upper);
5323 void *netdev_lower_dev_get_private(struct net_device *dev,
5324 struct net_device *lower_dev)
5326 struct netdev_adjacent *lower;
5330 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5334 return lower->private;
5336 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5339 int dev_get_nest_level(struct net_device *dev,
5340 bool (*type_check)(struct net_device *dev))
5342 struct net_device *lower = NULL;
5343 struct list_head *iter;
5349 netdev_for_each_lower_dev(dev, lower, iter) {
5350 nest = dev_get_nest_level(lower, type_check);
5351 if (max_nest < nest)
5355 if (type_check(dev))
5360 EXPORT_SYMBOL(dev_get_nest_level);
5362 static void dev_change_rx_flags(struct net_device *dev, int flags)
5364 const struct net_device_ops *ops = dev->netdev_ops;
5366 if (ops->ndo_change_rx_flags)
5367 ops->ndo_change_rx_flags(dev, flags);
5370 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5372 unsigned int old_flags = dev->flags;
5378 dev->flags |= IFF_PROMISC;
5379 dev->promiscuity += inc;
5380 if (dev->promiscuity == 0) {
5383 * If inc causes overflow, untouch promisc and return error.
5386 dev->flags &= ~IFF_PROMISC;
5388 dev->promiscuity -= inc;
5389 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5394 if (dev->flags != old_flags) {
5395 pr_info("device %s %s promiscuous mode\n",
5397 dev->flags & IFF_PROMISC ? "entered" : "left");
5398 if (audit_enabled) {
5399 current_uid_gid(&uid, &gid);
5400 audit_log(current->audit_context, GFP_ATOMIC,
5401 AUDIT_ANOM_PROMISCUOUS,
5402 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5403 dev->name, (dev->flags & IFF_PROMISC),
5404 (old_flags & IFF_PROMISC),
5405 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5406 from_kuid(&init_user_ns, uid),
5407 from_kgid(&init_user_ns, gid),
5408 audit_get_sessionid(current));
5411 dev_change_rx_flags(dev, IFF_PROMISC);
5414 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5419 * dev_set_promiscuity - update promiscuity count on a device
5423 * Add or remove promiscuity from a device. While the count in the device
5424 * remains above zero the interface remains promiscuous. Once it hits zero
5425 * the device reverts back to normal filtering operation. A negative inc
5426 * value is used to drop promiscuity on the device.
5427 * Return 0 if successful or a negative errno code on error.
5429 int dev_set_promiscuity(struct net_device *dev, int inc)
5431 unsigned int old_flags = dev->flags;
5434 err = __dev_set_promiscuity(dev, inc, true);
5437 if (dev->flags != old_flags)
5438 dev_set_rx_mode(dev);
5441 EXPORT_SYMBOL(dev_set_promiscuity);
5443 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5445 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5449 dev->flags |= IFF_ALLMULTI;
5450 dev->allmulti += inc;
5451 if (dev->allmulti == 0) {
5454 * If inc causes overflow, untouch allmulti and return error.
5457 dev->flags &= ~IFF_ALLMULTI;
5459 dev->allmulti -= inc;
5460 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5465 if (dev->flags ^ old_flags) {
5466 dev_change_rx_flags(dev, IFF_ALLMULTI);
5467 dev_set_rx_mode(dev);
5469 __dev_notify_flags(dev, old_flags,
5470 dev->gflags ^ old_gflags);
5476 * dev_set_allmulti - update allmulti count on a device
5480 * Add or remove reception of all multicast frames to a device. While the
5481 * count in the device remains above zero the interface remains listening
5482 * to all interfaces. Once it hits zero the device reverts back to normal
5483 * filtering operation. A negative @inc value is used to drop the counter
5484 * when releasing a resource needing all multicasts.
5485 * Return 0 if successful or a negative errno code on error.
5488 int dev_set_allmulti(struct net_device *dev, int inc)
5490 return __dev_set_allmulti(dev, inc, true);
5492 EXPORT_SYMBOL(dev_set_allmulti);
5495 * Upload unicast and multicast address lists to device and
5496 * configure RX filtering. When the device doesn't support unicast
5497 * filtering it is put in promiscuous mode while unicast addresses
5500 void __dev_set_rx_mode(struct net_device *dev)
5502 const struct net_device_ops *ops = dev->netdev_ops;
5504 /* dev_open will call this function so the list will stay sane. */
5505 if (!(dev->flags&IFF_UP))
5508 if (!netif_device_present(dev))
5511 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5512 /* Unicast addresses changes may only happen under the rtnl,
5513 * therefore calling __dev_set_promiscuity here is safe.
5515 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5516 __dev_set_promiscuity(dev, 1, false);
5517 dev->uc_promisc = true;
5518 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5519 __dev_set_promiscuity(dev, -1, false);
5520 dev->uc_promisc = false;
5524 if (ops->ndo_set_rx_mode)
5525 ops->ndo_set_rx_mode(dev);
5528 void dev_set_rx_mode(struct net_device *dev)
5530 netif_addr_lock_bh(dev);
5531 __dev_set_rx_mode(dev);
5532 netif_addr_unlock_bh(dev);
5536 * dev_get_flags - get flags reported to userspace
5539 * Get the combination of flag bits exported through APIs to userspace.
5541 unsigned int dev_get_flags(const struct net_device *dev)
5545 flags = (dev->flags & ~(IFF_PROMISC |
5550 (dev->gflags & (IFF_PROMISC |
5553 if (netif_running(dev)) {
5554 if (netif_oper_up(dev))
5555 flags |= IFF_RUNNING;
5556 if (netif_carrier_ok(dev))
5557 flags |= IFF_LOWER_UP;
5558 if (netif_dormant(dev))
5559 flags |= IFF_DORMANT;
5564 EXPORT_SYMBOL(dev_get_flags);
5566 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5568 unsigned int old_flags = dev->flags;
5574 * Set the flags on our device.
5577 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5578 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5580 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5584 * Load in the correct multicast list now the flags have changed.
5587 if ((old_flags ^ flags) & IFF_MULTICAST)
5588 dev_change_rx_flags(dev, IFF_MULTICAST);
5590 dev_set_rx_mode(dev);
5593 * Have we downed the interface. We handle IFF_UP ourselves
5594 * according to user attempts to set it, rather than blindly
5599 if ((old_flags ^ flags) & IFF_UP)
5600 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5602 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5603 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5604 unsigned int old_flags = dev->flags;
5606 dev->gflags ^= IFF_PROMISC;
5608 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5609 if (dev->flags != old_flags)
5610 dev_set_rx_mode(dev);
5613 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5614 is important. Some (broken) drivers set IFF_PROMISC, when
5615 IFF_ALLMULTI is requested not asking us and not reporting.
5617 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5618 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5620 dev->gflags ^= IFF_ALLMULTI;
5621 __dev_set_allmulti(dev, inc, false);
5627 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5628 unsigned int gchanges)
5630 unsigned int changes = dev->flags ^ old_flags;
5633 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5635 if (changes & IFF_UP) {
5636 if (dev->flags & IFF_UP)
5637 call_netdevice_notifiers(NETDEV_UP, dev);
5639 call_netdevice_notifiers(NETDEV_DOWN, dev);
5642 if (dev->flags & IFF_UP &&
5643 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5644 struct netdev_notifier_change_info change_info;
5646 change_info.flags_changed = changes;
5647 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5653 * dev_change_flags - change device settings
5655 * @flags: device state flags
5657 * Change settings on device based state flags. The flags are
5658 * in the userspace exported format.
5660 int dev_change_flags(struct net_device *dev, unsigned int flags)
5663 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5665 ret = __dev_change_flags(dev, flags);
5669 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5670 __dev_notify_flags(dev, old_flags, changes);
5673 EXPORT_SYMBOL(dev_change_flags);
5675 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5677 const struct net_device_ops *ops = dev->netdev_ops;
5679 if (ops->ndo_change_mtu)
5680 return ops->ndo_change_mtu(dev, new_mtu);
5687 * dev_set_mtu - Change maximum transfer unit
5689 * @new_mtu: new transfer unit
5691 * Change the maximum transfer size of the network device.
5693 int dev_set_mtu(struct net_device *dev, int new_mtu)
5697 if (new_mtu == dev->mtu)
5700 /* MTU must be positive. */
5704 if (!netif_device_present(dev))
5707 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5708 err = notifier_to_errno(err);
5712 orig_mtu = dev->mtu;
5713 err = __dev_set_mtu(dev, new_mtu);
5716 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5717 err = notifier_to_errno(err);
5719 /* setting mtu back and notifying everyone again,
5720 * so that they have a chance to revert changes.
5722 __dev_set_mtu(dev, orig_mtu);
5723 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5728 EXPORT_SYMBOL(dev_set_mtu);
5731 * dev_set_group - Change group this device belongs to
5733 * @new_group: group this device should belong to
5735 void dev_set_group(struct net_device *dev, int new_group)
5737 dev->group = new_group;
5739 EXPORT_SYMBOL(dev_set_group);
5742 * dev_set_mac_address - Change Media Access Control Address
5746 * Change the hardware (MAC) address of the device
5748 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5750 const struct net_device_ops *ops = dev->netdev_ops;
5753 if (!ops->ndo_set_mac_address)
5755 if (sa->sa_family != dev->type)
5757 if (!netif_device_present(dev))
5759 err = ops->ndo_set_mac_address(dev, sa);
5762 dev->addr_assign_type = NET_ADDR_SET;
5763 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5764 add_device_randomness(dev->dev_addr, dev->addr_len);
5767 EXPORT_SYMBOL(dev_set_mac_address);
5770 * dev_change_carrier - Change device carrier
5772 * @new_carrier: new value
5774 * Change device carrier
5776 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5778 const struct net_device_ops *ops = dev->netdev_ops;
5780 if (!ops->ndo_change_carrier)
5782 if (!netif_device_present(dev))
5784 return ops->ndo_change_carrier(dev, new_carrier);
5786 EXPORT_SYMBOL(dev_change_carrier);
5789 * dev_get_phys_port_id - Get device physical port ID
5793 * Get device physical port ID
5795 int dev_get_phys_port_id(struct net_device *dev,
5796 struct netdev_phys_port_id *ppid)
5798 const struct net_device_ops *ops = dev->netdev_ops;
5800 if (!ops->ndo_get_phys_port_id)
5802 return ops->ndo_get_phys_port_id(dev, ppid);
5804 EXPORT_SYMBOL(dev_get_phys_port_id);
5807 * dev_new_index - allocate an ifindex
5808 * @net: the applicable net namespace
5810 * Returns a suitable unique value for a new device interface
5811 * number. The caller must hold the rtnl semaphore or the
5812 * dev_base_lock to be sure it remains unique.
5814 static int dev_new_index(struct net *net)
5816 int ifindex = net->ifindex;
5820 if (!__dev_get_by_index(net, ifindex))
5821 return net->ifindex = ifindex;
5825 /* Delayed registration/unregisteration */
5826 static LIST_HEAD(net_todo_list);
5827 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5829 static void net_set_todo(struct net_device *dev)
5831 list_add_tail(&dev->todo_list, &net_todo_list);
5832 dev_net(dev)->dev_unreg_count++;
5835 static void rollback_registered_many(struct list_head *head)
5837 struct net_device *dev, *tmp;
5838 LIST_HEAD(close_head);
5840 BUG_ON(dev_boot_phase);
5843 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5844 /* Some devices call without registering
5845 * for initialization unwind. Remove those
5846 * devices and proceed with the remaining.
5848 if (dev->reg_state == NETREG_UNINITIALIZED) {
5849 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5853 list_del(&dev->unreg_list);
5856 dev->dismantle = true;
5857 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5860 /* If device is running, close it first. */
5861 list_for_each_entry(dev, head, unreg_list)
5862 list_add_tail(&dev->close_list, &close_head);
5863 dev_close_many(&close_head);
5865 list_for_each_entry(dev, head, unreg_list) {
5866 /* And unlink it from device chain. */
5867 unlist_netdevice(dev);
5869 dev->reg_state = NETREG_UNREGISTERING;
5874 list_for_each_entry(dev, head, unreg_list) {
5875 /* Shutdown queueing discipline. */
5879 /* Notify protocols, that we are about to destroy
5880 this device. They should clean all the things.
5882 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5885 * Flush the unicast and multicast chains
5890 if (dev->netdev_ops->ndo_uninit)
5891 dev->netdev_ops->ndo_uninit(dev);
5893 if (!dev->rtnl_link_ops ||
5894 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5895 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5897 /* Notifier chain MUST detach us all upper devices. */
5898 WARN_ON(netdev_has_any_upper_dev(dev));
5900 /* Remove entries from kobject tree */
5901 netdev_unregister_kobject(dev);
5903 /* Remove XPS queueing entries */
5904 netif_reset_xps_queues_gt(dev, 0);
5910 list_for_each_entry(dev, head, unreg_list)
5914 static void rollback_registered(struct net_device *dev)
5918 list_add(&dev->unreg_list, &single);
5919 rollback_registered_many(&single);
5923 static netdev_features_t netdev_fix_features(struct net_device *dev,
5924 netdev_features_t features)
5926 /* Fix illegal checksum combinations */
5927 if ((features & NETIF_F_HW_CSUM) &&
5928 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5929 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5930 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5933 /* TSO requires that SG is present as well. */
5934 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5935 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5936 features &= ~NETIF_F_ALL_TSO;
5939 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5940 !(features & NETIF_F_IP_CSUM)) {
5941 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5942 features &= ~NETIF_F_TSO;
5943 features &= ~NETIF_F_TSO_ECN;
5946 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5947 !(features & NETIF_F_IPV6_CSUM)) {
5948 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5949 features &= ~NETIF_F_TSO6;
5952 /* TSO ECN requires that TSO is present as well. */
5953 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5954 features &= ~NETIF_F_TSO_ECN;
5956 /* Software GSO depends on SG. */
5957 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5958 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5959 features &= ~NETIF_F_GSO;
5962 /* UFO needs SG and checksumming */
5963 if (features & NETIF_F_UFO) {
5964 /* maybe split UFO into V4 and V6? */
5965 if (!((features & NETIF_F_GEN_CSUM) ||
5966 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5967 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5969 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5970 features &= ~NETIF_F_UFO;
5973 if (!(features & NETIF_F_SG)) {
5975 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5976 features &= ~NETIF_F_UFO;
5980 #ifdef CONFIG_NET_RX_BUSY_POLL
5981 if (dev->netdev_ops->ndo_busy_poll)
5982 features |= NETIF_F_BUSY_POLL;
5985 features &= ~NETIF_F_BUSY_POLL;
5990 int __netdev_update_features(struct net_device *dev)
5992 netdev_features_t features;
5997 features = netdev_get_wanted_features(dev);
5999 if (dev->netdev_ops->ndo_fix_features)
6000 features = dev->netdev_ops->ndo_fix_features(dev, features);
6002 /* driver might be less strict about feature dependencies */
6003 features = netdev_fix_features(dev, features);
6005 if (dev->features == features)
6008 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6009 &dev->features, &features);
6011 if (dev->netdev_ops->ndo_set_features)
6012 err = dev->netdev_ops->ndo_set_features(dev, features);
6014 if (unlikely(err < 0)) {
6016 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6017 err, &features, &dev->features);
6022 dev->features = features;
6028 * netdev_update_features - recalculate device features
6029 * @dev: the device to check
6031 * Recalculate dev->features set and send notifications if it
6032 * has changed. Should be called after driver or hardware dependent
6033 * conditions might have changed that influence the features.
6035 void netdev_update_features(struct net_device *dev)
6037 if (__netdev_update_features(dev))
6038 netdev_features_change(dev);
6040 EXPORT_SYMBOL(netdev_update_features);
6043 * netdev_change_features - recalculate device features
6044 * @dev: the device to check
6046 * Recalculate dev->features set and send notifications even
6047 * if they have not changed. Should be called instead of
6048 * netdev_update_features() if also dev->vlan_features might
6049 * have changed to allow the changes to be propagated to stacked
6052 void netdev_change_features(struct net_device *dev)
6054 __netdev_update_features(dev);
6055 netdev_features_change(dev);
6057 EXPORT_SYMBOL(netdev_change_features);
6060 * netif_stacked_transfer_operstate - transfer operstate
6061 * @rootdev: the root or lower level device to transfer state from
6062 * @dev: the device to transfer operstate to
6064 * Transfer operational state from root to device. This is normally
6065 * called when a stacking relationship exists between the root
6066 * device and the device(a leaf device).
6068 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6069 struct net_device *dev)
6071 if (rootdev->operstate == IF_OPER_DORMANT)
6072 netif_dormant_on(dev);
6074 netif_dormant_off(dev);
6076 if (netif_carrier_ok(rootdev)) {
6077 if (!netif_carrier_ok(dev))
6078 netif_carrier_on(dev);
6080 if (netif_carrier_ok(dev))
6081 netif_carrier_off(dev);
6084 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6087 static int netif_alloc_rx_queues(struct net_device *dev)
6089 unsigned int i, count = dev->num_rx_queues;
6090 struct netdev_rx_queue *rx;
6094 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
6100 for (i = 0; i < count; i++)
6106 static void netdev_init_one_queue(struct net_device *dev,
6107 struct netdev_queue *queue, void *_unused)
6109 /* Initialize queue lock */
6110 spin_lock_init(&queue->_xmit_lock);
6111 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6112 queue->xmit_lock_owner = -1;
6113 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6116 dql_init(&queue->dql, HZ);
6120 static void netif_free_tx_queues(struct net_device *dev)
6125 static int netif_alloc_netdev_queues(struct net_device *dev)
6127 unsigned int count = dev->num_tx_queues;
6128 struct netdev_queue *tx;
6129 size_t sz = count * sizeof(*tx);
6131 BUG_ON(count < 1 || count > 0xffff);
6133 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6141 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6142 spin_lock_init(&dev->tx_global_lock);
6148 * register_netdevice - register a network device
6149 * @dev: device to register
6151 * Take a completed network device structure and add it to the kernel
6152 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6153 * chain. 0 is returned on success. A negative errno code is returned
6154 * on a failure to set up the device, or if the name is a duplicate.
6156 * Callers must hold the rtnl semaphore. You may want
6157 * register_netdev() instead of this.
6160 * The locking appears insufficient to guarantee two parallel registers
6161 * will not get the same name.
6164 int register_netdevice(struct net_device *dev)
6167 struct net *net = dev_net(dev);
6169 BUG_ON(dev_boot_phase);
6174 /* When net_device's are persistent, this will be fatal. */
6175 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6178 spin_lock_init(&dev->addr_list_lock);
6179 netdev_set_addr_lockdep_class(dev);
6183 ret = dev_get_valid_name(net, dev, dev->name);
6187 /* Init, if this function is available */
6188 if (dev->netdev_ops->ndo_init) {
6189 ret = dev->netdev_ops->ndo_init(dev);
6197 if (((dev->hw_features | dev->features) &
6198 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6199 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6200 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6201 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6208 dev->ifindex = dev_new_index(net);
6209 else if (__dev_get_by_index(net, dev->ifindex))
6212 if (dev->iflink == -1)
6213 dev->iflink = dev->ifindex;
6215 /* Transfer changeable features to wanted_features and enable
6216 * software offloads (GSO and GRO).
6218 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6219 dev->features |= NETIF_F_SOFT_FEATURES;
6220 dev->wanted_features = dev->features & dev->hw_features;
6222 if (!(dev->flags & IFF_LOOPBACK)) {
6223 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6226 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6228 dev->vlan_features |= NETIF_F_HIGHDMA;
6230 /* Make NETIF_F_SG inheritable to tunnel devices.
6232 dev->hw_enc_features |= NETIF_F_SG;
6234 /* Make NETIF_F_SG inheritable to MPLS.
6236 dev->mpls_features |= NETIF_F_SG;
6238 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6239 ret = notifier_to_errno(ret);
6243 ret = netdev_register_kobject(dev);
6246 dev->reg_state = NETREG_REGISTERED;
6248 __netdev_update_features(dev);
6251 * Default initial state at registry is that the
6252 * device is present.
6255 set_bit(__LINK_STATE_PRESENT, &dev->state);
6257 linkwatch_init_dev(dev);
6259 dev_init_scheduler(dev);
6261 list_netdevice(dev);
6262 add_device_randomness(dev->dev_addr, dev->addr_len);
6264 /* If the device has permanent device address, driver should
6265 * set dev_addr and also addr_assign_type should be set to
6266 * NET_ADDR_PERM (default value).
6268 if (dev->addr_assign_type == NET_ADDR_PERM)
6269 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6271 /* Notify protocols, that a new device appeared. */
6272 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6273 ret = notifier_to_errno(ret);
6275 rollback_registered(dev);
6276 dev->reg_state = NETREG_UNREGISTERED;
6279 * Prevent userspace races by waiting until the network
6280 * device is fully setup before sending notifications.
6282 if (!dev->rtnl_link_ops ||
6283 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6284 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6290 if (dev->netdev_ops->ndo_uninit)
6291 dev->netdev_ops->ndo_uninit(dev);
6294 EXPORT_SYMBOL(register_netdevice);
6297 * init_dummy_netdev - init a dummy network device for NAPI
6298 * @dev: device to init
6300 * This takes a network device structure and initialize the minimum
6301 * amount of fields so it can be used to schedule NAPI polls without
6302 * registering a full blown interface. This is to be used by drivers
6303 * that need to tie several hardware interfaces to a single NAPI
6304 * poll scheduler due to HW limitations.
6306 int init_dummy_netdev(struct net_device *dev)
6308 /* Clear everything. Note we don't initialize spinlocks
6309 * are they aren't supposed to be taken by any of the
6310 * NAPI code and this dummy netdev is supposed to be
6311 * only ever used for NAPI polls
6313 memset(dev, 0, sizeof(struct net_device));
6315 /* make sure we BUG if trying to hit standard
6316 * register/unregister code path
6318 dev->reg_state = NETREG_DUMMY;
6320 /* NAPI wants this */
6321 INIT_LIST_HEAD(&dev->napi_list);
6323 /* a dummy interface is started by default */
6324 set_bit(__LINK_STATE_PRESENT, &dev->state);
6325 set_bit(__LINK_STATE_START, &dev->state);
6327 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6328 * because users of this 'device' dont need to change
6334 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6338 * register_netdev - register a network device
6339 * @dev: device to register
6341 * Take a completed network device structure and add it to the kernel
6342 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6343 * chain. 0 is returned on success. A negative errno code is returned
6344 * on a failure to set up the device, or if the name is a duplicate.
6346 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6347 * and expands the device name if you passed a format string to
6350 int register_netdev(struct net_device *dev)
6355 err = register_netdevice(dev);
6359 EXPORT_SYMBOL(register_netdev);
6361 int netdev_refcnt_read(const struct net_device *dev)
6365 for_each_possible_cpu(i)
6366 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6369 EXPORT_SYMBOL(netdev_refcnt_read);
6372 * netdev_wait_allrefs - wait until all references are gone.
6373 * @dev: target net_device
6375 * This is called when unregistering network devices.
6377 * Any protocol or device that holds a reference should register
6378 * for netdevice notification, and cleanup and put back the
6379 * reference if they receive an UNREGISTER event.
6380 * We can get stuck here if buggy protocols don't correctly
6383 static void netdev_wait_allrefs(struct net_device *dev)
6385 unsigned long rebroadcast_time, warning_time;
6388 linkwatch_forget_dev(dev);
6390 rebroadcast_time = warning_time = jiffies;
6391 refcnt = netdev_refcnt_read(dev);
6393 while (refcnt != 0) {
6394 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6397 /* Rebroadcast unregister notification */
6398 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6404 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6405 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6407 /* We must not have linkwatch events
6408 * pending on unregister. If this
6409 * happens, we simply run the queue
6410 * unscheduled, resulting in a noop
6413 linkwatch_run_queue();
6418 rebroadcast_time = jiffies;
6423 refcnt = netdev_refcnt_read(dev);
6425 if (time_after(jiffies, warning_time + 10 * HZ)) {
6426 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6428 warning_time = jiffies;
6437 * register_netdevice(x1);
6438 * register_netdevice(x2);
6440 * unregister_netdevice(y1);
6441 * unregister_netdevice(y2);
6447 * We are invoked by rtnl_unlock().
6448 * This allows us to deal with problems:
6449 * 1) We can delete sysfs objects which invoke hotplug
6450 * without deadlocking with linkwatch via keventd.
6451 * 2) Since we run with the RTNL semaphore not held, we can sleep
6452 * safely in order to wait for the netdev refcnt to drop to zero.
6454 * We must not return until all unregister events added during
6455 * the interval the lock was held have been completed.
6457 void netdev_run_todo(void)
6459 struct list_head list;
6461 /* Snapshot list, allow later requests */
6462 list_replace_init(&net_todo_list, &list);
6467 /* Wait for rcu callbacks to finish before next phase */
6468 if (!list_empty(&list))
6471 while (!list_empty(&list)) {
6472 struct net_device *dev
6473 = list_first_entry(&list, struct net_device, todo_list);
6474 list_del(&dev->todo_list);
6477 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6480 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6481 pr_err("network todo '%s' but state %d\n",
6482 dev->name, dev->reg_state);
6487 dev->reg_state = NETREG_UNREGISTERED;
6489 on_each_cpu(flush_backlog, dev, 1);
6491 netdev_wait_allrefs(dev);
6494 BUG_ON(netdev_refcnt_read(dev));
6495 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6496 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6497 WARN_ON(dev->dn_ptr);
6499 if (dev->destructor)
6500 dev->destructor(dev);
6502 /* Report a network device has been unregistered */
6504 dev_net(dev)->dev_unreg_count--;
6506 wake_up(&netdev_unregistering_wq);
6508 /* Free network device */
6509 kobject_put(&dev->dev.kobj);
6513 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6514 * fields in the same order, with only the type differing.
6516 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6517 const struct net_device_stats *netdev_stats)
6519 #if BITS_PER_LONG == 64
6520 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6521 memcpy(stats64, netdev_stats, sizeof(*stats64));
6523 size_t i, n = sizeof(*stats64) / sizeof(u64);
6524 const unsigned long *src = (const unsigned long *)netdev_stats;
6525 u64 *dst = (u64 *)stats64;
6527 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6528 sizeof(*stats64) / sizeof(u64));
6529 for (i = 0; i < n; i++)
6533 EXPORT_SYMBOL(netdev_stats_to_stats64);
6536 * dev_get_stats - get network device statistics
6537 * @dev: device to get statistics from
6538 * @storage: place to store stats
6540 * Get network statistics from device. Return @storage.
6541 * The device driver may provide its own method by setting
6542 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6543 * otherwise the internal statistics structure is used.
6545 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6546 struct rtnl_link_stats64 *storage)
6548 const struct net_device_ops *ops = dev->netdev_ops;
6550 if (ops->ndo_get_stats64) {
6551 memset(storage, 0, sizeof(*storage));
6552 ops->ndo_get_stats64(dev, storage);
6553 } else if (ops->ndo_get_stats) {
6554 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6556 netdev_stats_to_stats64(storage, &dev->stats);
6558 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6559 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6562 EXPORT_SYMBOL(dev_get_stats);
6564 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6566 struct netdev_queue *queue = dev_ingress_queue(dev);
6568 #ifdef CONFIG_NET_CLS_ACT
6571 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6574 netdev_init_one_queue(dev, queue, NULL);
6575 queue->qdisc = &noop_qdisc;
6576 queue->qdisc_sleeping = &noop_qdisc;
6577 rcu_assign_pointer(dev->ingress_queue, queue);
6582 static const struct ethtool_ops default_ethtool_ops;
6584 void netdev_set_default_ethtool_ops(struct net_device *dev,
6585 const struct ethtool_ops *ops)
6587 if (dev->ethtool_ops == &default_ethtool_ops)
6588 dev->ethtool_ops = ops;
6590 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6592 void netdev_freemem(struct net_device *dev)
6594 char *addr = (char *)dev - dev->padded;
6600 * alloc_netdev_mqs - allocate network device
6601 * @sizeof_priv: size of private data to allocate space for
6602 * @name: device name format string
6603 * @name_assign_type: origin of device name
6604 * @setup: callback to initialize device
6605 * @txqs: the number of TX subqueues to allocate
6606 * @rxqs: the number of RX subqueues to allocate
6608 * Allocates a struct net_device with private data area for driver use
6609 * and performs basic initialization. Also allocates subqueue structs
6610 * for each queue on the device.
6612 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6613 unsigned char name_assign_type,
6614 void (*setup)(struct net_device *),
6615 unsigned int txqs, unsigned int rxqs)
6617 struct net_device *dev;
6619 struct net_device *p;
6621 BUG_ON(strlen(name) >= sizeof(dev->name));
6624 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6630 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6635 alloc_size = sizeof(struct net_device);
6637 /* ensure 32-byte alignment of private area */
6638 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6639 alloc_size += sizeof_priv;
6641 /* ensure 32-byte alignment of whole construct */
6642 alloc_size += NETDEV_ALIGN - 1;
6644 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6646 p = vzalloc(alloc_size);
6650 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6651 dev->padded = (char *)dev - (char *)p;
6653 dev->pcpu_refcnt = alloc_percpu(int);
6654 if (!dev->pcpu_refcnt)
6657 if (dev_addr_init(dev))
6663 dev_net_set(dev, &init_net);
6665 dev->gso_max_size = GSO_MAX_SIZE;
6666 dev->gso_max_segs = GSO_MAX_SEGS;
6667 dev->gso_min_segs = 0;
6669 INIT_LIST_HEAD(&dev->napi_list);
6670 INIT_LIST_HEAD(&dev->unreg_list);
6671 INIT_LIST_HEAD(&dev->close_list);
6672 INIT_LIST_HEAD(&dev->link_watch_list);
6673 INIT_LIST_HEAD(&dev->adj_list.upper);
6674 INIT_LIST_HEAD(&dev->adj_list.lower);
6675 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6676 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6677 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6680 dev->num_tx_queues = txqs;
6681 dev->real_num_tx_queues = txqs;
6682 if (netif_alloc_netdev_queues(dev))
6686 dev->num_rx_queues = rxqs;
6687 dev->real_num_rx_queues = rxqs;
6688 if (netif_alloc_rx_queues(dev))
6692 strcpy(dev->name, name);
6693 dev->name_assign_type = name_assign_type;
6694 dev->group = INIT_NETDEV_GROUP;
6695 if (!dev->ethtool_ops)
6696 dev->ethtool_ops = &default_ethtool_ops;
6704 free_percpu(dev->pcpu_refcnt);
6706 netdev_freemem(dev);
6709 EXPORT_SYMBOL(alloc_netdev_mqs);
6712 * free_netdev - free network device
6715 * This function does the last stage of destroying an allocated device
6716 * interface. The reference to the device object is released.
6717 * If this is the last reference then it will be freed.
6719 void free_netdev(struct net_device *dev)
6721 struct napi_struct *p, *n;
6723 release_net(dev_net(dev));
6725 netif_free_tx_queues(dev);
6730 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6732 /* Flush device addresses */
6733 dev_addr_flush(dev);
6735 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6738 free_percpu(dev->pcpu_refcnt);
6739 dev->pcpu_refcnt = NULL;
6741 /* Compatibility with error handling in drivers */
6742 if (dev->reg_state == NETREG_UNINITIALIZED) {
6743 netdev_freemem(dev);
6747 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6748 dev->reg_state = NETREG_RELEASED;
6750 /* will free via device release */
6751 put_device(&dev->dev);
6753 EXPORT_SYMBOL(free_netdev);
6756 * synchronize_net - Synchronize with packet receive processing
6758 * Wait for packets currently being received to be done.
6759 * Does not block later packets from starting.
6761 void synchronize_net(void)
6764 if (rtnl_is_locked())
6765 synchronize_rcu_expedited();
6769 EXPORT_SYMBOL(synchronize_net);
6772 * unregister_netdevice_queue - remove device from the kernel
6776 * This function shuts down a device interface and removes it
6777 * from the kernel tables.
6778 * If head not NULL, device is queued to be unregistered later.
6780 * Callers must hold the rtnl semaphore. You may want
6781 * unregister_netdev() instead of this.
6784 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6789 list_move_tail(&dev->unreg_list, head);
6791 rollback_registered(dev);
6792 /* Finish processing unregister after unlock */
6796 EXPORT_SYMBOL(unregister_netdevice_queue);
6799 * unregister_netdevice_many - unregister many devices
6800 * @head: list of devices
6802 * Note: As most callers use a stack allocated list_head,
6803 * we force a list_del() to make sure stack wont be corrupted later.
6805 void unregister_netdevice_many(struct list_head *head)
6807 struct net_device *dev;
6809 if (!list_empty(head)) {
6810 rollback_registered_many(head);
6811 list_for_each_entry(dev, head, unreg_list)
6816 EXPORT_SYMBOL(unregister_netdevice_many);
6819 * unregister_netdev - remove device from the kernel
6822 * This function shuts down a device interface and removes it
6823 * from the kernel tables.
6825 * This is just a wrapper for unregister_netdevice that takes
6826 * the rtnl semaphore. In general you want to use this and not
6827 * unregister_netdevice.
6829 void unregister_netdev(struct net_device *dev)
6832 unregister_netdevice(dev);
6835 EXPORT_SYMBOL(unregister_netdev);
6838 * dev_change_net_namespace - move device to different nethost namespace
6840 * @net: network namespace
6841 * @pat: If not NULL name pattern to try if the current device name
6842 * is already taken in the destination network namespace.
6844 * This function shuts down a device interface and moves it
6845 * to a new network namespace. On success 0 is returned, on
6846 * a failure a netagive errno code is returned.
6848 * Callers must hold the rtnl semaphore.
6851 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6857 /* Don't allow namespace local devices to be moved. */
6859 if (dev->features & NETIF_F_NETNS_LOCAL)
6862 /* Ensure the device has been registrered */
6863 if (dev->reg_state != NETREG_REGISTERED)
6866 /* Get out if there is nothing todo */
6868 if (net_eq(dev_net(dev), net))
6871 /* Pick the destination device name, and ensure
6872 * we can use it in the destination network namespace.
6875 if (__dev_get_by_name(net, dev->name)) {
6876 /* We get here if we can't use the current device name */
6879 if (dev_get_valid_name(net, dev, pat) < 0)
6884 * And now a mini version of register_netdevice unregister_netdevice.
6887 /* If device is running close it first. */
6890 /* And unlink it from device chain */
6892 unlist_netdevice(dev);
6896 /* Shutdown queueing discipline. */
6899 /* Notify protocols, that we are about to destroy
6900 this device. They should clean all the things.
6902 Note that dev->reg_state stays at NETREG_REGISTERED.
6903 This is wanted because this way 8021q and macvlan know
6904 the device is just moving and can keep their slaves up.
6906 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6908 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6909 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6912 * Flush the unicast and multicast chains
6917 /* Send a netdev-removed uevent to the old namespace */
6918 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6919 netdev_adjacent_del_links(dev);
6921 /* Actually switch the network namespace */
6922 dev_net_set(dev, net);
6924 /* If there is an ifindex conflict assign a new one */
6925 if (__dev_get_by_index(net, dev->ifindex)) {
6926 int iflink = (dev->iflink == dev->ifindex);
6927 dev->ifindex = dev_new_index(net);
6929 dev->iflink = dev->ifindex;
6932 /* Send a netdev-add uevent to the new namespace */
6933 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6934 netdev_adjacent_add_links(dev);
6936 /* Fixup kobjects */
6937 err = device_rename(&dev->dev, dev->name);
6940 /* Add the device back in the hashes */
6941 list_netdevice(dev);
6943 /* Notify protocols, that a new device appeared. */
6944 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6947 * Prevent userspace races by waiting until the network
6948 * device is fully setup before sending notifications.
6950 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6957 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6959 static int dev_cpu_callback(struct notifier_block *nfb,
6960 unsigned long action,
6963 struct sk_buff **list_skb;
6964 struct sk_buff *skb;
6965 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6966 struct softnet_data *sd, *oldsd;
6968 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6971 local_irq_disable();
6972 cpu = smp_processor_id();
6973 sd = &per_cpu(softnet_data, cpu);
6974 oldsd = &per_cpu(softnet_data, oldcpu);
6976 /* Find end of our completion_queue. */
6977 list_skb = &sd->completion_queue;
6979 list_skb = &(*list_skb)->next;
6980 /* Append completion queue from offline CPU. */
6981 *list_skb = oldsd->completion_queue;
6982 oldsd->completion_queue = NULL;
6984 /* Append output queue from offline CPU. */
6985 if (oldsd->output_queue) {
6986 *sd->output_queue_tailp = oldsd->output_queue;
6987 sd->output_queue_tailp = oldsd->output_queue_tailp;
6988 oldsd->output_queue = NULL;
6989 oldsd->output_queue_tailp = &oldsd->output_queue;
6991 /* Append NAPI poll list from offline CPU. */
6992 if (!list_empty(&oldsd->poll_list)) {
6993 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6994 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6997 raise_softirq_irqoff(NET_TX_SOFTIRQ);
7000 /* Process offline CPU's input_pkt_queue */
7001 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7002 netif_rx_internal(skb);
7003 input_queue_head_incr(oldsd);
7005 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
7006 netif_rx_internal(skb);
7007 input_queue_head_incr(oldsd);
7015 * netdev_increment_features - increment feature set by one
7016 * @all: current feature set
7017 * @one: new feature set
7018 * @mask: mask feature set
7020 * Computes a new feature set after adding a device with feature set
7021 * @one to the master device with current feature set @all. Will not
7022 * enable anything that is off in @mask. Returns the new feature set.
7024 netdev_features_t netdev_increment_features(netdev_features_t all,
7025 netdev_features_t one, netdev_features_t mask)
7027 if (mask & NETIF_F_GEN_CSUM)
7028 mask |= NETIF_F_ALL_CSUM;
7029 mask |= NETIF_F_VLAN_CHALLENGED;
7031 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7032 all &= one | ~NETIF_F_ALL_FOR_ALL;
7034 /* If one device supports hw checksumming, set for all. */
7035 if (all & NETIF_F_GEN_CSUM)
7036 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7040 EXPORT_SYMBOL(netdev_increment_features);
7042 static struct hlist_head * __net_init netdev_create_hash(void)
7045 struct hlist_head *hash;
7047 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7049 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7050 INIT_HLIST_HEAD(&hash[i]);
7055 /* Initialize per network namespace state */
7056 static int __net_init netdev_init(struct net *net)
7058 if (net != &init_net)
7059 INIT_LIST_HEAD(&net->dev_base_head);
7061 net->dev_name_head = netdev_create_hash();
7062 if (net->dev_name_head == NULL)
7065 net->dev_index_head = netdev_create_hash();
7066 if (net->dev_index_head == NULL)
7072 kfree(net->dev_name_head);
7078 * netdev_drivername - network driver for the device
7079 * @dev: network device
7081 * Determine network driver for device.
7083 const char *netdev_drivername(const struct net_device *dev)
7085 const struct device_driver *driver;
7086 const struct device *parent;
7087 const char *empty = "";
7089 parent = dev->dev.parent;
7093 driver = parent->driver;
7094 if (driver && driver->name)
7095 return driver->name;
7099 static void __netdev_printk(const char *level, const struct net_device *dev,
7100 struct va_format *vaf)
7102 if (dev && dev->dev.parent) {
7103 dev_printk_emit(level[1] - '0',
7106 dev_driver_string(dev->dev.parent),
7107 dev_name(dev->dev.parent),
7108 netdev_name(dev), netdev_reg_state(dev),
7111 printk("%s%s%s: %pV",
7112 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7114 printk("%s(NULL net_device): %pV", level, vaf);
7118 void netdev_printk(const char *level, const struct net_device *dev,
7119 const char *format, ...)
7121 struct va_format vaf;
7124 va_start(args, format);
7129 __netdev_printk(level, dev, &vaf);
7133 EXPORT_SYMBOL(netdev_printk);
7135 #define define_netdev_printk_level(func, level) \
7136 void func(const struct net_device *dev, const char *fmt, ...) \
7138 struct va_format vaf; \
7141 va_start(args, fmt); \
7146 __netdev_printk(level, dev, &vaf); \
7150 EXPORT_SYMBOL(func);
7152 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7153 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7154 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7155 define_netdev_printk_level(netdev_err, KERN_ERR);
7156 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7157 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7158 define_netdev_printk_level(netdev_info, KERN_INFO);
7160 static void __net_exit netdev_exit(struct net *net)
7162 kfree(net->dev_name_head);
7163 kfree(net->dev_index_head);
7166 static struct pernet_operations __net_initdata netdev_net_ops = {
7167 .init = netdev_init,
7168 .exit = netdev_exit,
7171 static void __net_exit default_device_exit(struct net *net)
7173 struct net_device *dev, *aux;
7175 * Push all migratable network devices back to the
7176 * initial network namespace
7179 for_each_netdev_safe(net, dev, aux) {
7181 char fb_name[IFNAMSIZ];
7183 /* Ignore unmoveable devices (i.e. loopback) */
7184 if (dev->features & NETIF_F_NETNS_LOCAL)
7187 /* Leave virtual devices for the generic cleanup */
7188 if (dev->rtnl_link_ops)
7191 /* Push remaining network devices to init_net */
7192 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7193 err = dev_change_net_namespace(dev, &init_net, fb_name);
7195 pr_emerg("%s: failed to move %s to init_net: %d\n",
7196 __func__, dev->name, err);
7203 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7205 /* Return with the rtnl_lock held when there are no network
7206 * devices unregistering in any network namespace in net_list.
7213 prepare_to_wait(&netdev_unregistering_wq, &wait,
7214 TASK_UNINTERRUPTIBLE);
7215 unregistering = false;
7217 list_for_each_entry(net, net_list, exit_list) {
7218 if (net->dev_unreg_count > 0) {
7219 unregistering = true;
7228 finish_wait(&netdev_unregistering_wq, &wait);
7231 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7233 /* At exit all network devices most be removed from a network
7234 * namespace. Do this in the reverse order of registration.
7235 * Do this across as many network namespaces as possible to
7236 * improve batching efficiency.
7238 struct net_device *dev;
7240 LIST_HEAD(dev_kill_list);
7242 /* To prevent network device cleanup code from dereferencing
7243 * loopback devices or network devices that have been freed
7244 * wait here for all pending unregistrations to complete,
7245 * before unregistring the loopback device and allowing the
7246 * network namespace be freed.
7248 * The netdev todo list containing all network devices
7249 * unregistrations that happen in default_device_exit_batch
7250 * will run in the rtnl_unlock() at the end of
7251 * default_device_exit_batch.
7253 rtnl_lock_unregistering(net_list);
7254 list_for_each_entry(net, net_list, exit_list) {
7255 for_each_netdev_reverse(net, dev) {
7256 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7257 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7259 unregister_netdevice_queue(dev, &dev_kill_list);
7262 unregister_netdevice_many(&dev_kill_list);
7266 static struct pernet_operations __net_initdata default_device_ops = {
7267 .exit = default_device_exit,
7268 .exit_batch = default_device_exit_batch,
7272 * Initialize the DEV module. At boot time this walks the device list and
7273 * unhooks any devices that fail to initialise (normally hardware not
7274 * present) and leaves us with a valid list of present and active devices.
7279 * This is called single threaded during boot, so no need
7280 * to take the rtnl semaphore.
7282 static int __init net_dev_init(void)
7284 int i, rc = -ENOMEM;
7286 BUG_ON(!dev_boot_phase);
7288 if (dev_proc_init())
7291 if (netdev_kobject_init())
7294 INIT_LIST_HEAD(&ptype_all);
7295 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7296 INIT_LIST_HEAD(&ptype_base[i]);
7298 INIT_LIST_HEAD(&offload_base);
7300 if (register_pernet_subsys(&netdev_net_ops))
7304 * Initialise the packet receive queues.
7307 for_each_possible_cpu(i) {
7308 struct softnet_data *sd = &per_cpu(softnet_data, i);
7310 skb_queue_head_init(&sd->input_pkt_queue);
7311 skb_queue_head_init(&sd->process_queue);
7312 INIT_LIST_HEAD(&sd->poll_list);
7313 sd->output_queue_tailp = &sd->output_queue;
7315 sd->csd.func = rps_trigger_softirq;
7320 sd->backlog.poll = process_backlog;
7321 sd->backlog.weight = weight_p;
7326 /* The loopback device is special if any other network devices
7327 * is present in a network namespace the loopback device must
7328 * be present. Since we now dynamically allocate and free the
7329 * loopback device ensure this invariant is maintained by
7330 * keeping the loopback device as the first device on the
7331 * list of network devices. Ensuring the loopback devices
7332 * is the first device that appears and the last network device
7335 if (register_pernet_device(&loopback_net_ops))
7338 if (register_pernet_device(&default_device_ops))
7341 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7342 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7344 hotcpu_notifier(dev_cpu_callback, 0);
7351 subsys_initcall(net_dev_init);