2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
121 #include <linux/ipv6.h>
122 #include <linux/in.h>
123 #include <linux/jhash.h>
124 #include <linux/random.h>
125 #include <trace/events/napi.h>
126 #include <trace/events/net.h>
127 #include <trace/events/skb.h>
128 #include <linux/pci.h>
129 #include <linux/inetdevice.h>
130 #include <linux/cpu_rmap.h>
131 #include <linux/static_key.h>
132 #include <linux/hashtable.h>
133 #include <linux/vmalloc.h>
134 #include <linux/if_macvlan.h>
135 #include <linux/errqueue.h>
137 #include "net-sysfs.h"
139 /* Instead of increasing this, you should create a hash table. */
140 #define MAX_GRO_SKBS 8
142 /* This should be increased if a protocol with a bigger head is added. */
143 #define GRO_MAX_HEAD (MAX_HEADER + 128)
145 static DEFINE_SPINLOCK(ptype_lock);
146 static DEFINE_SPINLOCK(offload_lock);
147 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
148 struct list_head ptype_all __read_mostly; /* Taps */
149 static struct list_head offload_base __read_mostly;
151 static int netif_rx_internal(struct sk_buff *skb);
152 static int call_netdevice_notifiers_info(unsigned long val,
153 struct net_device *dev,
154 struct netdev_notifier_info *info);
157 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
160 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
162 * Writers must hold the rtnl semaphore while they loop through the
163 * dev_base_head list, and hold dev_base_lock for writing when they do the
164 * actual updates. This allows pure readers to access the list even
165 * while a writer is preparing to update it.
167 * To put it another way, dev_base_lock is held for writing only to
168 * protect against pure readers; the rtnl semaphore provides the
169 * protection against other writers.
171 * See, for example usages, register_netdevice() and
172 * unregister_netdevice(), which must be called with the rtnl
175 DEFINE_RWLOCK(dev_base_lock);
176 EXPORT_SYMBOL(dev_base_lock);
178 /* protects napi_hash addition/deletion and napi_gen_id */
179 static DEFINE_SPINLOCK(napi_hash_lock);
181 static unsigned int napi_gen_id;
182 static DEFINE_HASHTABLE(napi_hash, 8);
184 static seqcount_t devnet_rename_seq;
186 static inline void dev_base_seq_inc(struct net *net)
188 while (++net->dev_base_seq == 0);
191 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
193 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
195 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
198 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
200 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
203 static inline void rps_lock(struct softnet_data *sd)
206 spin_lock(&sd->input_pkt_queue.lock);
210 static inline void rps_unlock(struct softnet_data *sd)
213 spin_unlock(&sd->input_pkt_queue.lock);
217 /* Device list insertion */
218 static void list_netdevice(struct net_device *dev)
220 struct net *net = dev_net(dev);
224 write_lock_bh(&dev_base_lock);
225 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
226 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
227 hlist_add_head_rcu(&dev->index_hlist,
228 dev_index_hash(net, dev->ifindex));
229 write_unlock_bh(&dev_base_lock);
231 dev_base_seq_inc(net);
234 /* Device list removal
235 * caller must respect a RCU grace period before freeing/reusing dev
237 static void unlist_netdevice(struct net_device *dev)
241 /* Unlink dev from the device chain */
242 write_lock_bh(&dev_base_lock);
243 list_del_rcu(&dev->dev_list);
244 hlist_del_rcu(&dev->name_hlist);
245 hlist_del_rcu(&dev->index_hlist);
246 write_unlock_bh(&dev_base_lock);
248 dev_base_seq_inc(dev_net(dev));
255 static RAW_NOTIFIER_HEAD(netdev_chain);
258 * Device drivers call our routines to queue packets here. We empty the
259 * queue in the local softnet handler.
262 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
263 EXPORT_PER_CPU_SYMBOL(softnet_data);
265 #ifdef CONFIG_LOCKDEP
267 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
268 * according to dev->type
270 static const unsigned short netdev_lock_type[] =
271 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
272 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
273 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
274 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
275 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
276 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
277 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
278 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
279 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
280 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
281 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
282 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
283 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
284 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
285 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
287 static const char *const netdev_lock_name[] =
288 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
289 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
290 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
291 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
292 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
293 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
294 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
295 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
296 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
297 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
298 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
299 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
300 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
301 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
302 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
304 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
305 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
307 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
311 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
312 if (netdev_lock_type[i] == dev_type)
314 /* the last key is used by default */
315 return ARRAY_SIZE(netdev_lock_type) - 1;
318 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
319 unsigned short dev_type)
323 i = netdev_lock_pos(dev_type);
324 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
325 netdev_lock_name[i]);
328 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
332 i = netdev_lock_pos(dev->type);
333 lockdep_set_class_and_name(&dev->addr_list_lock,
334 &netdev_addr_lock_key[i],
335 netdev_lock_name[i]);
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
339 unsigned short dev_type)
342 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
347 /*******************************************************************************
349 Protocol management and registration routines
351 *******************************************************************************/
354 * Add a protocol ID to the list. Now that the input handler is
355 * smarter we can dispense with all the messy stuff that used to be
358 * BEWARE!!! Protocol handlers, mangling input packets,
359 * MUST BE last in hash buckets and checking protocol handlers
360 * MUST start from promiscuous ptype_all chain in net_bh.
361 * It is true now, do not change it.
362 * Explanation follows: if protocol handler, mangling packet, will
363 * be the first on list, it is not able to sense, that packet
364 * is cloned and should be copied-on-write, so that it will
365 * change it and subsequent readers will get broken packet.
369 static inline struct list_head *ptype_head(const struct packet_type *pt)
371 if (pt->type == htons(ETH_P_ALL))
374 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
378 * dev_add_pack - add packet handler
379 * @pt: packet type declaration
381 * Add a protocol handler to the networking stack. The passed &packet_type
382 * is linked into kernel lists and may not be freed until it has been
383 * removed from the kernel lists.
385 * This call does not sleep therefore it can not
386 * guarantee all CPU's that are in middle of receiving packets
387 * will see the new packet type (until the next received packet).
390 void dev_add_pack(struct packet_type *pt)
392 struct list_head *head = ptype_head(pt);
394 spin_lock(&ptype_lock);
395 list_add_rcu(&pt->list, head);
396 spin_unlock(&ptype_lock);
398 EXPORT_SYMBOL(dev_add_pack);
401 * __dev_remove_pack - remove packet handler
402 * @pt: packet type declaration
404 * Remove a protocol handler that was previously added to the kernel
405 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
406 * from the kernel lists and can be freed or reused once this function
409 * The packet type might still be in use by receivers
410 * and must not be freed until after all the CPU's have gone
411 * through a quiescent state.
413 void __dev_remove_pack(struct packet_type *pt)
415 struct list_head *head = ptype_head(pt);
416 struct packet_type *pt1;
418 spin_lock(&ptype_lock);
420 list_for_each_entry(pt1, head, list) {
422 list_del_rcu(&pt->list);
427 pr_warn("dev_remove_pack: %p not found\n", pt);
429 spin_unlock(&ptype_lock);
431 EXPORT_SYMBOL(__dev_remove_pack);
434 * dev_remove_pack - remove packet handler
435 * @pt: packet type declaration
437 * Remove a protocol handler that was previously added to the kernel
438 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
439 * from the kernel lists and can be freed or reused once this function
442 * This call sleeps to guarantee that no CPU is looking at the packet
445 void dev_remove_pack(struct packet_type *pt)
447 __dev_remove_pack(pt);
451 EXPORT_SYMBOL(dev_remove_pack);
455 * dev_add_offload - register offload handlers
456 * @po: protocol offload declaration
458 * Add protocol offload handlers to the networking stack. The passed
459 * &proto_offload is linked into kernel lists and may not be freed until
460 * it has been removed from the kernel lists.
462 * This call does not sleep therefore it can not
463 * guarantee all CPU's that are in middle of receiving packets
464 * will see the new offload handlers (until the next received packet).
466 void dev_add_offload(struct packet_offload *po)
468 struct list_head *head = &offload_base;
470 spin_lock(&offload_lock);
471 list_add_rcu(&po->list, head);
472 spin_unlock(&offload_lock);
474 EXPORT_SYMBOL(dev_add_offload);
477 * __dev_remove_offload - remove offload handler
478 * @po: packet offload declaration
480 * Remove a protocol offload handler that was previously added to the
481 * kernel offload handlers by dev_add_offload(). The passed &offload_type
482 * is removed from the kernel lists and can be freed or reused once this
485 * The packet type might still be in use by receivers
486 * and must not be freed until after all the CPU's have gone
487 * through a quiescent state.
489 static void __dev_remove_offload(struct packet_offload *po)
491 struct list_head *head = &offload_base;
492 struct packet_offload *po1;
494 spin_lock(&offload_lock);
496 list_for_each_entry(po1, head, list) {
498 list_del_rcu(&po->list);
503 pr_warn("dev_remove_offload: %p not found\n", po);
505 spin_unlock(&offload_lock);
509 * dev_remove_offload - remove packet offload handler
510 * @po: packet offload declaration
512 * Remove a packet offload handler that was previously added to the kernel
513 * offload handlers by dev_add_offload(). The passed &offload_type is
514 * removed from the kernel lists and can be freed or reused once this
517 * This call sleeps to guarantee that no CPU is looking at the packet
520 void dev_remove_offload(struct packet_offload *po)
522 __dev_remove_offload(po);
526 EXPORT_SYMBOL(dev_remove_offload);
528 /******************************************************************************
530 Device Boot-time Settings Routines
532 *******************************************************************************/
534 /* Boot time configuration table */
535 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
538 * netdev_boot_setup_add - add new setup entry
539 * @name: name of the device
540 * @map: configured settings for the device
542 * Adds new setup entry to the dev_boot_setup list. The function
543 * returns 0 on error and 1 on success. This is a generic routine to
546 static int netdev_boot_setup_add(char *name, struct ifmap *map)
548 struct netdev_boot_setup *s;
552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
553 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
554 memset(s[i].name, 0, sizeof(s[i].name));
555 strlcpy(s[i].name, name, IFNAMSIZ);
556 memcpy(&s[i].map, map, sizeof(s[i].map));
561 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
565 * netdev_boot_setup_check - check boot time settings
566 * @dev: the netdevice
568 * Check boot time settings for the device.
569 * The found settings are set for the device to be used
570 * later in the device probing.
571 * Returns 0 if no settings found, 1 if they are.
573 int netdev_boot_setup_check(struct net_device *dev)
575 struct netdev_boot_setup *s = dev_boot_setup;
578 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
579 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
580 !strcmp(dev->name, s[i].name)) {
581 dev->irq = s[i].map.irq;
582 dev->base_addr = s[i].map.base_addr;
583 dev->mem_start = s[i].map.mem_start;
584 dev->mem_end = s[i].map.mem_end;
590 EXPORT_SYMBOL(netdev_boot_setup_check);
594 * netdev_boot_base - get address from boot time settings
595 * @prefix: prefix for network device
596 * @unit: id for network device
598 * Check boot time settings for the base address of device.
599 * The found settings are set for the device to be used
600 * later in the device probing.
601 * Returns 0 if no settings found.
603 unsigned long netdev_boot_base(const char *prefix, int unit)
605 const struct netdev_boot_setup *s = dev_boot_setup;
609 sprintf(name, "%s%d", prefix, unit);
612 * If device already registered then return base of 1
613 * to indicate not to probe for this interface
615 if (__dev_get_by_name(&init_net, name))
618 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
619 if (!strcmp(name, s[i].name))
620 return s[i].map.base_addr;
625 * Saves at boot time configured settings for any netdevice.
627 int __init netdev_boot_setup(char *str)
632 str = get_options(str, ARRAY_SIZE(ints), ints);
637 memset(&map, 0, sizeof(map));
641 map.base_addr = ints[2];
643 map.mem_start = ints[3];
645 map.mem_end = ints[4];
647 /* Add new entry to the list */
648 return netdev_boot_setup_add(str, &map);
651 __setup("netdev=", netdev_boot_setup);
653 /*******************************************************************************
655 Device Interface Subroutines
657 *******************************************************************************/
660 * __dev_get_by_name - find a device by its name
661 * @net: the applicable net namespace
662 * @name: name to find
664 * Find an interface by name. Must be called under RTNL semaphore
665 * or @dev_base_lock. If the name is found a pointer to the device
666 * is returned. If the name is not found then %NULL is returned. The
667 * reference counters are not incremented so the caller must be
668 * careful with locks.
671 struct net_device *__dev_get_by_name(struct net *net, const char *name)
673 struct net_device *dev;
674 struct hlist_head *head = dev_name_hash(net, name);
676 hlist_for_each_entry(dev, head, name_hlist)
677 if (!strncmp(dev->name, name, IFNAMSIZ))
682 EXPORT_SYMBOL(__dev_get_by_name);
685 * dev_get_by_name_rcu - find a device by its name
686 * @net: the applicable net namespace
687 * @name: name to find
689 * Find an interface by name.
690 * If the name is found a pointer to the device is returned.
691 * If the name is not found then %NULL is returned.
692 * The reference counters are not incremented so the caller must be
693 * careful with locks. The caller must hold RCU lock.
696 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
698 struct net_device *dev;
699 struct hlist_head *head = dev_name_hash(net, name);
701 hlist_for_each_entry_rcu(dev, head, name_hlist)
702 if (!strncmp(dev->name, name, IFNAMSIZ))
707 EXPORT_SYMBOL(dev_get_by_name_rcu);
710 * dev_get_by_name - find a device by its name
711 * @net: the applicable net namespace
712 * @name: name to find
714 * Find an interface by name. This can be called from any
715 * context and does its own locking. The returned handle has
716 * the usage count incremented and the caller must use dev_put() to
717 * release it when it is no longer needed. %NULL is returned if no
718 * matching device is found.
721 struct net_device *dev_get_by_name(struct net *net, const char *name)
723 struct net_device *dev;
726 dev = dev_get_by_name_rcu(net, name);
732 EXPORT_SYMBOL(dev_get_by_name);
735 * __dev_get_by_index - find a device by its ifindex
736 * @net: the applicable net namespace
737 * @ifindex: index of device
739 * Search for an interface by index. Returns %NULL if the device
740 * is not found or a pointer to the device. The device has not
741 * had its reference counter increased so the caller must be careful
742 * about locking. The caller must hold either the RTNL semaphore
746 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
748 struct net_device *dev;
749 struct hlist_head *head = dev_index_hash(net, ifindex);
751 hlist_for_each_entry(dev, head, index_hlist)
752 if (dev->ifindex == ifindex)
757 EXPORT_SYMBOL(__dev_get_by_index);
760 * dev_get_by_index_rcu - find a device by its ifindex
761 * @net: the applicable net namespace
762 * @ifindex: index of device
764 * Search for an interface by index. Returns %NULL if the device
765 * is not found or a pointer to the device. The device has not
766 * had its reference counter increased so the caller must be careful
767 * about locking. The caller must hold RCU lock.
770 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
772 struct net_device *dev;
773 struct hlist_head *head = dev_index_hash(net, ifindex);
775 hlist_for_each_entry_rcu(dev, head, index_hlist)
776 if (dev->ifindex == ifindex)
781 EXPORT_SYMBOL(dev_get_by_index_rcu);
785 * dev_get_by_index - find a device by its ifindex
786 * @net: the applicable net namespace
787 * @ifindex: index of device
789 * Search for an interface by index. Returns NULL if the device
790 * is not found or a pointer to the device. The device returned has
791 * had a reference added and the pointer is safe until the user calls
792 * dev_put to indicate they have finished with it.
795 struct net_device *dev_get_by_index(struct net *net, int ifindex)
797 struct net_device *dev;
800 dev = dev_get_by_index_rcu(net, ifindex);
806 EXPORT_SYMBOL(dev_get_by_index);
809 * netdev_get_name - get a netdevice name, knowing its ifindex.
810 * @net: network namespace
811 * @name: a pointer to the buffer where the name will be stored.
812 * @ifindex: the ifindex of the interface to get the name from.
814 * The use of raw_seqcount_begin() and cond_resched() before
815 * retrying is required as we want to give the writers a chance
816 * to complete when CONFIG_PREEMPT is not set.
818 int netdev_get_name(struct net *net, char *name, int ifindex)
820 struct net_device *dev;
824 seq = raw_seqcount_begin(&devnet_rename_seq);
826 dev = dev_get_by_index_rcu(net, ifindex);
832 strcpy(name, dev->name);
834 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
843 * dev_getbyhwaddr_rcu - find a device by its hardware address
844 * @net: the applicable net namespace
845 * @type: media type of device
846 * @ha: hardware address
848 * Search for an interface by MAC address. Returns NULL if the device
849 * is not found or a pointer to the device.
850 * The caller must hold RCU or RTNL.
851 * The returned device has not had its ref count increased
852 * and the caller must therefore be careful about locking
856 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
859 struct net_device *dev;
861 for_each_netdev_rcu(net, dev)
862 if (dev->type == type &&
863 !memcmp(dev->dev_addr, ha, dev->addr_len))
868 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
870 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
872 struct net_device *dev;
875 for_each_netdev(net, dev)
876 if (dev->type == type)
881 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
883 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
885 struct net_device *dev, *ret = NULL;
888 for_each_netdev_rcu(net, dev)
889 if (dev->type == type) {
897 EXPORT_SYMBOL(dev_getfirstbyhwtype);
900 * dev_get_by_flags_rcu - find any device with given flags
901 * @net: the applicable net namespace
902 * @if_flags: IFF_* values
903 * @mask: bitmask of bits in if_flags to check
905 * Search for any interface with the given flags. Returns NULL if a device
906 * is not found or a pointer to the device. Must be called inside
907 * rcu_read_lock(), and result refcount is unchanged.
910 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
913 struct net_device *dev, *ret;
916 for_each_netdev_rcu(net, dev) {
917 if (((dev->flags ^ if_flags) & mask) == 0) {
924 EXPORT_SYMBOL(dev_get_by_flags_rcu);
927 * dev_valid_name - check if name is okay for network device
930 * Network device names need to be valid file names to
931 * to allow sysfs to work. We also disallow any kind of
934 bool dev_valid_name(const char *name)
938 if (strlen(name) >= IFNAMSIZ)
940 if (!strcmp(name, ".") || !strcmp(name, ".."))
944 if (*name == '/' || isspace(*name))
950 EXPORT_SYMBOL(dev_valid_name);
953 * __dev_alloc_name - allocate a name for a device
954 * @net: network namespace to allocate the device name in
955 * @name: name format string
956 * @buf: scratch buffer and result name string
958 * Passed a format string - eg "lt%d" it will try and find a suitable
959 * id. It scans list of devices to build up a free map, then chooses
960 * the first empty slot. The caller must hold the dev_base or rtnl lock
961 * while allocating the name and adding the device in order to avoid
963 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
964 * Returns the number of the unit assigned or a negative errno code.
967 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
971 const int max_netdevices = 8*PAGE_SIZE;
972 unsigned long *inuse;
973 struct net_device *d;
975 p = strnchr(name, IFNAMSIZ-1, '%');
978 * Verify the string as this thing may have come from
979 * the user. There must be either one "%d" and no other "%"
982 if (p[1] != 'd' || strchr(p + 2, '%'))
985 /* Use one page as a bit array of possible slots */
986 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
990 for_each_netdev(net, d) {
991 if (!sscanf(d->name, name, &i))
993 if (i < 0 || i >= max_netdevices)
996 /* avoid cases where sscanf is not exact inverse of printf */
997 snprintf(buf, IFNAMSIZ, name, i);
998 if (!strncmp(buf, d->name, IFNAMSIZ))
1002 i = find_first_zero_bit(inuse, max_netdevices);
1003 free_page((unsigned long) inuse);
1007 snprintf(buf, IFNAMSIZ, name, i);
1008 if (!__dev_get_by_name(net, buf))
1011 /* It is possible to run out of possible slots
1012 * when the name is long and there isn't enough space left
1013 * for the digits, or if all bits are used.
1019 * dev_alloc_name - allocate a name for a device
1021 * @name: name format string
1023 * Passed a format string - eg "lt%d" it will try and find a suitable
1024 * id. It scans list of devices to build up a free map, then chooses
1025 * the first empty slot. The caller must hold the dev_base or rtnl lock
1026 * while allocating the name and adding the device in order to avoid
1028 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1029 * Returns the number of the unit assigned or a negative errno code.
1032 int dev_alloc_name(struct net_device *dev, const char *name)
1038 BUG_ON(!dev_net(dev));
1040 ret = __dev_alloc_name(net, name, buf);
1042 strlcpy(dev->name, buf, IFNAMSIZ);
1045 EXPORT_SYMBOL(dev_alloc_name);
1047 static int dev_alloc_name_ns(struct net *net,
1048 struct net_device *dev,
1054 ret = __dev_alloc_name(net, name, buf);
1056 strlcpy(dev->name, buf, IFNAMSIZ);
1060 static int dev_get_valid_name(struct net *net,
1061 struct net_device *dev,
1066 if (!dev_valid_name(name))
1069 if (strchr(name, '%'))
1070 return dev_alloc_name_ns(net, dev, name);
1071 else if (__dev_get_by_name(net, name))
1073 else if (dev->name != name)
1074 strlcpy(dev->name, name, IFNAMSIZ);
1080 * dev_change_name - change name of a device
1082 * @newname: name (or format string) must be at least IFNAMSIZ
1084 * Change name of a device, can pass format strings "eth%d".
1087 int dev_change_name(struct net_device *dev, const char *newname)
1089 unsigned char old_assign_type;
1090 char oldname[IFNAMSIZ];
1096 BUG_ON(!dev_net(dev));
1099 if (dev->flags & IFF_UP)
1102 write_seqcount_begin(&devnet_rename_seq);
1104 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1105 write_seqcount_end(&devnet_rename_seq);
1109 memcpy(oldname, dev->name, IFNAMSIZ);
1111 err = dev_get_valid_name(net, dev, newname);
1113 write_seqcount_end(&devnet_rename_seq);
1117 if (oldname[0] && !strchr(oldname, '%'))
1118 netdev_info(dev, "renamed from %s\n", oldname);
1120 old_assign_type = dev->name_assign_type;
1121 dev->name_assign_type = NET_NAME_RENAMED;
1124 ret = device_rename(&dev->dev, dev->name);
1126 memcpy(dev->name, oldname, IFNAMSIZ);
1127 dev->name_assign_type = old_assign_type;
1128 write_seqcount_end(&devnet_rename_seq);
1132 write_seqcount_end(&devnet_rename_seq);
1134 netdev_adjacent_rename_links(dev, oldname);
1136 write_lock_bh(&dev_base_lock);
1137 hlist_del_rcu(&dev->name_hlist);
1138 write_unlock_bh(&dev_base_lock);
1142 write_lock_bh(&dev_base_lock);
1143 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1144 write_unlock_bh(&dev_base_lock);
1146 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1147 ret = notifier_to_errno(ret);
1150 /* err >= 0 after dev_alloc_name() or stores the first errno */
1153 write_seqcount_begin(&devnet_rename_seq);
1154 memcpy(dev->name, oldname, IFNAMSIZ);
1155 memcpy(oldname, newname, IFNAMSIZ);
1156 dev->name_assign_type = old_assign_type;
1157 old_assign_type = NET_NAME_RENAMED;
1160 pr_err("%s: name change rollback failed: %d\n",
1169 * dev_set_alias - change ifalias of a device
1171 * @alias: name up to IFALIASZ
1172 * @len: limit of bytes to copy from info
1174 * Set ifalias for a device,
1176 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1182 if (len >= IFALIASZ)
1186 kfree(dev->ifalias);
1187 dev->ifalias = NULL;
1191 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1194 dev->ifalias = new_ifalias;
1196 strlcpy(dev->ifalias, alias, len+1);
1202 * netdev_features_change - device changes features
1203 * @dev: device to cause notification
1205 * Called to indicate a device has changed features.
1207 void netdev_features_change(struct net_device *dev)
1209 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1211 EXPORT_SYMBOL(netdev_features_change);
1214 * netdev_state_change - device changes state
1215 * @dev: device to cause notification
1217 * Called to indicate a device has changed state. This function calls
1218 * the notifier chains for netdev_chain and sends a NEWLINK message
1219 * to the routing socket.
1221 void netdev_state_change(struct net_device *dev)
1223 if (dev->flags & IFF_UP) {
1224 struct netdev_notifier_change_info change_info;
1226 change_info.flags_changed = 0;
1227 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1229 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1232 EXPORT_SYMBOL(netdev_state_change);
1235 * netdev_notify_peers - notify network peers about existence of @dev
1236 * @dev: network device
1238 * Generate traffic such that interested network peers are aware of
1239 * @dev, such as by generating a gratuitous ARP. This may be used when
1240 * a device wants to inform the rest of the network about some sort of
1241 * reconfiguration such as a failover event or virtual machine
1244 void netdev_notify_peers(struct net_device *dev)
1247 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1250 EXPORT_SYMBOL(netdev_notify_peers);
1252 static int __dev_open(struct net_device *dev)
1254 const struct net_device_ops *ops = dev->netdev_ops;
1259 if (!netif_device_present(dev))
1262 /* Block netpoll from trying to do any rx path servicing.
1263 * If we don't do this there is a chance ndo_poll_controller
1264 * or ndo_poll may be running while we open the device
1266 netpoll_poll_disable(dev);
1268 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1269 ret = notifier_to_errno(ret);
1273 set_bit(__LINK_STATE_START, &dev->state);
1275 if (ops->ndo_validate_addr)
1276 ret = ops->ndo_validate_addr(dev);
1278 if (!ret && ops->ndo_open)
1279 ret = ops->ndo_open(dev);
1281 netpoll_poll_enable(dev);
1284 clear_bit(__LINK_STATE_START, &dev->state);
1286 dev->flags |= IFF_UP;
1287 net_dmaengine_get();
1288 dev_set_rx_mode(dev);
1290 add_device_randomness(dev->dev_addr, dev->addr_len);
1297 * dev_open - prepare an interface for use.
1298 * @dev: device to open
1300 * Takes a device from down to up state. The device's private open
1301 * function is invoked and then the multicast lists are loaded. Finally
1302 * the device is moved into the up state and a %NETDEV_UP message is
1303 * sent to the netdev notifier chain.
1305 * Calling this function on an active interface is a nop. On a failure
1306 * a negative errno code is returned.
1308 int dev_open(struct net_device *dev)
1312 if (dev->flags & IFF_UP)
1315 ret = __dev_open(dev);
1319 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1320 call_netdevice_notifiers(NETDEV_UP, dev);
1324 EXPORT_SYMBOL(dev_open);
1326 static int __dev_close_many(struct list_head *head)
1328 struct net_device *dev;
1333 list_for_each_entry(dev, head, close_list) {
1334 /* Temporarily disable netpoll until the interface is down */
1335 netpoll_poll_disable(dev);
1337 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1339 clear_bit(__LINK_STATE_START, &dev->state);
1341 /* Synchronize to scheduled poll. We cannot touch poll list, it
1342 * can be even on different cpu. So just clear netif_running().
1344 * dev->stop() will invoke napi_disable() on all of it's
1345 * napi_struct instances on this device.
1347 smp_mb__after_atomic(); /* Commit netif_running(). */
1350 dev_deactivate_many(head);
1352 list_for_each_entry(dev, head, close_list) {
1353 const struct net_device_ops *ops = dev->netdev_ops;
1356 * Call the device specific close. This cannot fail.
1357 * Only if device is UP
1359 * We allow it to be called even after a DETACH hot-plug
1365 dev->flags &= ~IFF_UP;
1366 net_dmaengine_put();
1367 netpoll_poll_enable(dev);
1373 static int __dev_close(struct net_device *dev)
1378 list_add(&dev->close_list, &single);
1379 retval = __dev_close_many(&single);
1385 static int dev_close_many(struct list_head *head)
1387 struct net_device *dev, *tmp;
1389 /* Remove the devices that don't need to be closed */
1390 list_for_each_entry_safe(dev, tmp, head, close_list)
1391 if (!(dev->flags & IFF_UP))
1392 list_del_init(&dev->close_list);
1394 __dev_close_many(head);
1396 list_for_each_entry_safe(dev, tmp, head, close_list) {
1397 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1398 call_netdevice_notifiers(NETDEV_DOWN, dev);
1399 list_del_init(&dev->close_list);
1406 * dev_close - shutdown an interface.
1407 * @dev: device to shutdown
1409 * This function moves an active device into down state. A
1410 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1411 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1414 int dev_close(struct net_device *dev)
1416 if (dev->flags & IFF_UP) {
1419 list_add(&dev->close_list, &single);
1420 dev_close_many(&single);
1425 EXPORT_SYMBOL(dev_close);
1429 * dev_disable_lro - disable Large Receive Offload on a device
1432 * Disable Large Receive Offload (LRO) on a net device. Must be
1433 * called under RTNL. This is needed if received packets may be
1434 * forwarded to another interface.
1436 void dev_disable_lro(struct net_device *dev)
1439 * If we're trying to disable lro on a vlan device
1440 * use the underlying physical device instead
1442 if (is_vlan_dev(dev))
1443 dev = vlan_dev_real_dev(dev);
1445 /* the same for macvlan devices */
1446 if (netif_is_macvlan(dev))
1447 dev = macvlan_dev_real_dev(dev);
1449 dev->wanted_features &= ~NETIF_F_LRO;
1450 netdev_update_features(dev);
1452 if (unlikely(dev->features & NETIF_F_LRO))
1453 netdev_WARN(dev, "failed to disable LRO!\n");
1455 EXPORT_SYMBOL(dev_disable_lro);
1457 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1458 struct net_device *dev)
1460 struct netdev_notifier_info info;
1462 netdev_notifier_info_init(&info, dev);
1463 return nb->notifier_call(nb, val, &info);
1466 static int dev_boot_phase = 1;
1469 * register_netdevice_notifier - register a network notifier block
1472 * Register a notifier to be called when network device events occur.
1473 * The notifier passed is linked into the kernel structures and must
1474 * not be reused until it has been unregistered. A negative errno code
1475 * is returned on a failure.
1477 * When registered all registration and up events are replayed
1478 * to the new notifier to allow device to have a race free
1479 * view of the network device list.
1482 int register_netdevice_notifier(struct notifier_block *nb)
1484 struct net_device *dev;
1485 struct net_device *last;
1490 err = raw_notifier_chain_register(&netdev_chain, nb);
1496 for_each_netdev(net, dev) {
1497 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1498 err = notifier_to_errno(err);
1502 if (!(dev->flags & IFF_UP))
1505 call_netdevice_notifier(nb, NETDEV_UP, dev);
1516 for_each_netdev(net, dev) {
1520 if (dev->flags & IFF_UP) {
1521 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1523 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1525 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1530 raw_notifier_chain_unregister(&netdev_chain, nb);
1533 EXPORT_SYMBOL(register_netdevice_notifier);
1536 * unregister_netdevice_notifier - unregister a network notifier block
1539 * Unregister a notifier previously registered by
1540 * register_netdevice_notifier(). The notifier is unlinked into the
1541 * kernel structures and may then be reused. A negative errno code
1542 * is returned on a failure.
1544 * After unregistering unregister and down device events are synthesized
1545 * for all devices on the device list to the removed notifier to remove
1546 * the need for special case cleanup code.
1549 int unregister_netdevice_notifier(struct notifier_block *nb)
1551 struct net_device *dev;
1556 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1561 for_each_netdev(net, dev) {
1562 if (dev->flags & IFF_UP) {
1563 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1565 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1567 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1574 EXPORT_SYMBOL(unregister_netdevice_notifier);
1577 * call_netdevice_notifiers_info - call all network notifier blocks
1578 * @val: value passed unmodified to notifier function
1579 * @dev: net_device pointer passed unmodified to notifier function
1580 * @info: notifier information data
1582 * Call all network notifier blocks. Parameters and return value
1583 * are as for raw_notifier_call_chain().
1586 static int call_netdevice_notifiers_info(unsigned long val,
1587 struct net_device *dev,
1588 struct netdev_notifier_info *info)
1591 netdev_notifier_info_init(info, dev);
1592 return raw_notifier_call_chain(&netdev_chain, val, info);
1596 * call_netdevice_notifiers - call all network notifier blocks
1597 * @val: value passed unmodified to notifier function
1598 * @dev: net_device pointer passed unmodified to notifier function
1600 * Call all network notifier blocks. Parameters and return value
1601 * are as for raw_notifier_call_chain().
1604 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1606 struct netdev_notifier_info info;
1608 return call_netdevice_notifiers_info(val, dev, &info);
1610 EXPORT_SYMBOL(call_netdevice_notifiers);
1612 static struct static_key netstamp_needed __read_mostly;
1613 #ifdef HAVE_JUMP_LABEL
1614 /* We are not allowed to call static_key_slow_dec() from irq context
1615 * If net_disable_timestamp() is called from irq context, defer the
1616 * static_key_slow_dec() calls.
1618 static atomic_t netstamp_needed_deferred;
1621 void net_enable_timestamp(void)
1623 #ifdef HAVE_JUMP_LABEL
1624 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1628 static_key_slow_dec(&netstamp_needed);
1632 static_key_slow_inc(&netstamp_needed);
1634 EXPORT_SYMBOL(net_enable_timestamp);
1636 void net_disable_timestamp(void)
1638 #ifdef HAVE_JUMP_LABEL
1639 if (in_interrupt()) {
1640 atomic_inc(&netstamp_needed_deferred);
1644 static_key_slow_dec(&netstamp_needed);
1646 EXPORT_SYMBOL(net_disable_timestamp);
1648 static inline void net_timestamp_set(struct sk_buff *skb)
1650 skb->tstamp.tv64 = 0;
1651 if (static_key_false(&netstamp_needed))
1652 __net_timestamp(skb);
1655 #define net_timestamp_check(COND, SKB) \
1656 if (static_key_false(&netstamp_needed)) { \
1657 if ((COND) && !(SKB)->tstamp.tv64) \
1658 __net_timestamp(SKB); \
1661 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1665 if (!(dev->flags & IFF_UP))
1668 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1669 if (skb->len <= len)
1672 /* if TSO is enabled, we don't care about the length as the packet
1673 * could be forwarded without being segmented before
1675 if (skb_is_gso(skb))
1680 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1682 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1684 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1685 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1686 atomic_long_inc(&dev->rx_dropped);
1692 if (unlikely(!is_skb_forwardable(dev, skb))) {
1693 atomic_long_inc(&dev->rx_dropped);
1698 skb_scrub_packet(skb, true);
1699 skb->protocol = eth_type_trans(skb, dev);
1703 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1706 * dev_forward_skb - loopback an skb to another netif
1708 * @dev: destination network device
1709 * @skb: buffer to forward
1712 * NET_RX_SUCCESS (no congestion)
1713 * NET_RX_DROP (packet was dropped, but freed)
1715 * dev_forward_skb can be used for injecting an skb from the
1716 * start_xmit function of one device into the receive queue
1717 * of another device.
1719 * The receiving device may be in another namespace, so
1720 * we have to clear all information in the skb that could
1721 * impact namespace isolation.
1723 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1725 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1727 EXPORT_SYMBOL_GPL(dev_forward_skb);
1729 static inline int deliver_skb(struct sk_buff *skb,
1730 struct packet_type *pt_prev,
1731 struct net_device *orig_dev)
1733 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1735 atomic_inc(&skb->users);
1736 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1739 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1741 if (!ptype->af_packet_priv || !skb->sk)
1744 if (ptype->id_match)
1745 return ptype->id_match(ptype, skb->sk);
1746 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1753 * Support routine. Sends outgoing frames to any network
1754 * taps currently in use.
1757 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1759 struct packet_type *ptype;
1760 struct sk_buff *skb2 = NULL;
1761 struct packet_type *pt_prev = NULL;
1764 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1765 /* Never send packets back to the socket
1766 * they originated from - MvS (miquels@drinkel.ow.org)
1768 if ((ptype->dev == dev || !ptype->dev) &&
1769 (!skb_loop_sk(ptype, skb))) {
1771 deliver_skb(skb2, pt_prev, skb->dev);
1776 skb2 = skb_clone(skb, GFP_ATOMIC);
1780 net_timestamp_set(skb2);
1782 /* skb->nh should be correctly
1783 set by sender, so that the second statement is
1784 just protection against buggy protocols.
1786 skb_reset_mac_header(skb2);
1788 if (skb_network_header(skb2) < skb2->data ||
1789 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1790 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1791 ntohs(skb2->protocol),
1793 skb_reset_network_header(skb2);
1796 skb2->transport_header = skb2->network_header;
1797 skb2->pkt_type = PACKET_OUTGOING;
1802 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1807 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1808 * @dev: Network device
1809 * @txq: number of queues available
1811 * If real_num_tx_queues is changed the tc mappings may no longer be
1812 * valid. To resolve this verify the tc mapping remains valid and if
1813 * not NULL the mapping. With no priorities mapping to this
1814 * offset/count pair it will no longer be used. In the worst case TC0
1815 * is invalid nothing can be done so disable priority mappings. If is
1816 * expected that drivers will fix this mapping if they can before
1817 * calling netif_set_real_num_tx_queues.
1819 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1822 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1824 /* If TC0 is invalidated disable TC mapping */
1825 if (tc->offset + tc->count > txq) {
1826 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1831 /* Invalidated prio to tc mappings set to TC0 */
1832 for (i = 1; i < TC_BITMASK + 1; i++) {
1833 int q = netdev_get_prio_tc_map(dev, i);
1835 tc = &dev->tc_to_txq[q];
1836 if (tc->offset + tc->count > txq) {
1837 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1839 netdev_set_prio_tc_map(dev, i, 0);
1845 static DEFINE_MUTEX(xps_map_mutex);
1846 #define xmap_dereference(P) \
1847 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1849 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1852 struct xps_map *map = NULL;
1856 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1858 for (pos = 0; map && pos < map->len; pos++) {
1859 if (map->queues[pos] == index) {
1861 map->queues[pos] = map->queues[--map->len];
1863 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1864 kfree_rcu(map, rcu);
1874 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1876 struct xps_dev_maps *dev_maps;
1878 bool active = false;
1880 mutex_lock(&xps_map_mutex);
1881 dev_maps = xmap_dereference(dev->xps_maps);
1886 for_each_possible_cpu(cpu) {
1887 for (i = index; i < dev->num_tx_queues; i++) {
1888 if (!remove_xps_queue(dev_maps, cpu, i))
1891 if (i == dev->num_tx_queues)
1896 RCU_INIT_POINTER(dev->xps_maps, NULL);
1897 kfree_rcu(dev_maps, rcu);
1900 for (i = index; i < dev->num_tx_queues; i++)
1901 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1905 mutex_unlock(&xps_map_mutex);
1908 static struct xps_map *expand_xps_map(struct xps_map *map,
1911 struct xps_map *new_map;
1912 int alloc_len = XPS_MIN_MAP_ALLOC;
1915 for (pos = 0; map && pos < map->len; pos++) {
1916 if (map->queues[pos] != index)
1921 /* Need to add queue to this CPU's existing map */
1923 if (pos < map->alloc_len)
1926 alloc_len = map->alloc_len * 2;
1929 /* Need to allocate new map to store queue on this CPU's map */
1930 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1935 for (i = 0; i < pos; i++)
1936 new_map->queues[i] = map->queues[i];
1937 new_map->alloc_len = alloc_len;
1943 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1946 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1947 struct xps_map *map, *new_map;
1948 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1949 int cpu, numa_node_id = -2;
1950 bool active = false;
1952 mutex_lock(&xps_map_mutex);
1954 dev_maps = xmap_dereference(dev->xps_maps);
1956 /* allocate memory for queue storage */
1957 for_each_online_cpu(cpu) {
1958 if (!cpumask_test_cpu(cpu, mask))
1962 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1963 if (!new_dev_maps) {
1964 mutex_unlock(&xps_map_mutex);
1968 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1971 map = expand_xps_map(map, cpu, index);
1975 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1979 goto out_no_new_maps;
1981 for_each_possible_cpu(cpu) {
1982 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1983 /* add queue to CPU maps */
1986 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1987 while ((pos < map->len) && (map->queues[pos] != index))
1990 if (pos == map->len)
1991 map->queues[map->len++] = index;
1993 if (numa_node_id == -2)
1994 numa_node_id = cpu_to_node(cpu);
1995 else if (numa_node_id != cpu_to_node(cpu))
1998 } else if (dev_maps) {
1999 /* fill in the new device map from the old device map */
2000 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2001 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2006 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2008 /* Cleanup old maps */
2010 for_each_possible_cpu(cpu) {
2011 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2012 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2013 if (map && map != new_map)
2014 kfree_rcu(map, rcu);
2017 kfree_rcu(dev_maps, rcu);
2020 dev_maps = new_dev_maps;
2024 /* update Tx queue numa node */
2025 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2026 (numa_node_id >= 0) ? numa_node_id :
2032 /* removes queue from unused CPUs */
2033 for_each_possible_cpu(cpu) {
2034 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2037 if (remove_xps_queue(dev_maps, cpu, index))
2041 /* free map if not active */
2043 RCU_INIT_POINTER(dev->xps_maps, NULL);
2044 kfree_rcu(dev_maps, rcu);
2048 mutex_unlock(&xps_map_mutex);
2052 /* remove any maps that we added */
2053 for_each_possible_cpu(cpu) {
2054 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2055 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2057 if (new_map && new_map != map)
2061 mutex_unlock(&xps_map_mutex);
2063 kfree(new_dev_maps);
2066 EXPORT_SYMBOL(netif_set_xps_queue);
2070 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2071 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2073 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2077 if (txq < 1 || txq > dev->num_tx_queues)
2080 if (dev->reg_state == NETREG_REGISTERED ||
2081 dev->reg_state == NETREG_UNREGISTERING) {
2084 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2090 netif_setup_tc(dev, txq);
2092 if (txq < dev->real_num_tx_queues) {
2093 qdisc_reset_all_tx_gt(dev, txq);
2095 netif_reset_xps_queues_gt(dev, txq);
2100 dev->real_num_tx_queues = txq;
2103 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2107 * netif_set_real_num_rx_queues - set actual number of RX queues used
2108 * @dev: Network device
2109 * @rxq: Actual number of RX queues
2111 * This must be called either with the rtnl_lock held or before
2112 * registration of the net device. Returns 0 on success, or a
2113 * negative error code. If called before registration, it always
2116 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2120 if (rxq < 1 || rxq > dev->num_rx_queues)
2123 if (dev->reg_state == NETREG_REGISTERED) {
2126 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2132 dev->real_num_rx_queues = rxq;
2135 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2139 * netif_get_num_default_rss_queues - default number of RSS queues
2141 * This routine should set an upper limit on the number of RSS queues
2142 * used by default by multiqueue devices.
2144 int netif_get_num_default_rss_queues(void)
2146 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2148 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2150 static inline void __netif_reschedule(struct Qdisc *q)
2152 struct softnet_data *sd;
2153 unsigned long flags;
2155 local_irq_save(flags);
2156 sd = &__get_cpu_var(softnet_data);
2157 q->next_sched = NULL;
2158 *sd->output_queue_tailp = q;
2159 sd->output_queue_tailp = &q->next_sched;
2160 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2161 local_irq_restore(flags);
2164 void __netif_schedule(struct Qdisc *q)
2166 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2167 __netif_reschedule(q);
2169 EXPORT_SYMBOL(__netif_schedule);
2171 struct dev_kfree_skb_cb {
2172 enum skb_free_reason reason;
2175 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2177 return (struct dev_kfree_skb_cb *)skb->cb;
2180 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2182 unsigned long flags;
2184 if (likely(atomic_read(&skb->users) == 1)) {
2186 atomic_set(&skb->users, 0);
2187 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2190 get_kfree_skb_cb(skb)->reason = reason;
2191 local_irq_save(flags);
2192 skb->next = __this_cpu_read(softnet_data.completion_queue);
2193 __this_cpu_write(softnet_data.completion_queue, skb);
2194 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2195 local_irq_restore(flags);
2197 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2199 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2201 if (in_irq() || irqs_disabled())
2202 __dev_kfree_skb_irq(skb, reason);
2206 EXPORT_SYMBOL(__dev_kfree_skb_any);
2210 * netif_device_detach - mark device as removed
2211 * @dev: network device
2213 * Mark device as removed from system and therefore no longer available.
2215 void netif_device_detach(struct net_device *dev)
2217 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2218 netif_running(dev)) {
2219 netif_tx_stop_all_queues(dev);
2222 EXPORT_SYMBOL(netif_device_detach);
2225 * netif_device_attach - mark device as attached
2226 * @dev: network device
2228 * Mark device as attached from system and restart if needed.
2230 void netif_device_attach(struct net_device *dev)
2232 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2233 netif_running(dev)) {
2234 netif_tx_wake_all_queues(dev);
2235 __netdev_watchdog_up(dev);
2238 EXPORT_SYMBOL(netif_device_attach);
2240 static void skb_warn_bad_offload(const struct sk_buff *skb)
2242 static const netdev_features_t null_features = 0;
2243 struct net_device *dev = skb->dev;
2244 const char *driver = "";
2246 if (!net_ratelimit())
2249 if (dev && dev->dev.parent)
2250 driver = dev_driver_string(dev->dev.parent);
2252 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2253 "gso_type=%d ip_summed=%d\n",
2254 driver, dev ? &dev->features : &null_features,
2255 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2256 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2257 skb_shinfo(skb)->gso_type, skb->ip_summed);
2261 * Invalidate hardware checksum when packet is to be mangled, and
2262 * complete checksum manually on outgoing path.
2264 int skb_checksum_help(struct sk_buff *skb)
2267 int ret = 0, offset;
2269 if (skb->ip_summed == CHECKSUM_COMPLETE)
2270 goto out_set_summed;
2272 if (unlikely(skb_shinfo(skb)->gso_size)) {
2273 skb_warn_bad_offload(skb);
2277 /* Before computing a checksum, we should make sure no frag could
2278 * be modified by an external entity : checksum could be wrong.
2280 if (skb_has_shared_frag(skb)) {
2281 ret = __skb_linearize(skb);
2286 offset = skb_checksum_start_offset(skb);
2287 BUG_ON(offset >= skb_headlen(skb));
2288 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2290 offset += skb->csum_offset;
2291 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2293 if (skb_cloned(skb) &&
2294 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2295 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2300 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2302 skb->ip_summed = CHECKSUM_NONE;
2306 EXPORT_SYMBOL(skb_checksum_help);
2308 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2310 unsigned int vlan_depth = skb->mac_len;
2311 __be16 type = skb->protocol;
2313 /* Tunnel gso handlers can set protocol to ethernet. */
2314 if (type == htons(ETH_P_TEB)) {
2317 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2320 eth = (struct ethhdr *)skb_mac_header(skb);
2321 type = eth->h_proto;
2324 /* if skb->protocol is 802.1Q/AD then the header should already be
2325 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at
2326 * ETH_HLEN otherwise
2328 if (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2330 if (WARN_ON(vlan_depth < VLAN_HLEN))
2332 vlan_depth -= VLAN_HLEN;
2334 vlan_depth = ETH_HLEN;
2337 struct vlan_hdr *vh;
2339 if (unlikely(!pskb_may_pull(skb,
2340 vlan_depth + VLAN_HLEN)))
2343 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2344 type = vh->h_vlan_encapsulated_proto;
2345 vlan_depth += VLAN_HLEN;
2346 } while (type == htons(ETH_P_8021Q) ||
2347 type == htons(ETH_P_8021AD));
2350 *depth = vlan_depth;
2356 * skb_mac_gso_segment - mac layer segmentation handler.
2357 * @skb: buffer to segment
2358 * @features: features for the output path (see dev->features)
2360 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2361 netdev_features_t features)
2363 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2364 struct packet_offload *ptype;
2365 int vlan_depth = skb->mac_len;
2366 __be16 type = skb_network_protocol(skb, &vlan_depth);
2368 if (unlikely(!type))
2369 return ERR_PTR(-EINVAL);
2371 __skb_pull(skb, vlan_depth);
2374 list_for_each_entry_rcu(ptype, &offload_base, list) {
2375 if (ptype->type == type && ptype->callbacks.gso_segment) {
2376 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2379 err = ptype->callbacks.gso_send_check(skb);
2380 segs = ERR_PTR(err);
2381 if (err || skb_gso_ok(skb, features))
2383 __skb_push(skb, (skb->data -
2384 skb_network_header(skb)));
2386 segs = ptype->callbacks.gso_segment(skb, features);
2392 __skb_push(skb, skb->data - skb_mac_header(skb));
2396 EXPORT_SYMBOL(skb_mac_gso_segment);
2399 /* openvswitch calls this on rx path, so we need a different check.
2401 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2404 return skb->ip_summed != CHECKSUM_PARTIAL;
2406 return skb->ip_summed == CHECKSUM_NONE;
2410 * __skb_gso_segment - Perform segmentation on skb.
2411 * @skb: buffer to segment
2412 * @features: features for the output path (see dev->features)
2413 * @tx_path: whether it is called in TX path
2415 * This function segments the given skb and returns a list of segments.
2417 * It may return NULL if the skb requires no segmentation. This is
2418 * only possible when GSO is used for verifying header integrity.
2420 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2421 netdev_features_t features, bool tx_path)
2423 if (unlikely(skb_needs_check(skb, tx_path))) {
2426 skb_warn_bad_offload(skb);
2428 err = skb_cow_head(skb, 0);
2430 return ERR_PTR(err);
2433 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2434 SKB_GSO_CB(skb)->encap_level = 0;
2436 skb_reset_mac_header(skb);
2437 skb_reset_mac_len(skb);
2439 return skb_mac_gso_segment(skb, features);
2441 EXPORT_SYMBOL(__skb_gso_segment);
2443 /* Take action when hardware reception checksum errors are detected. */
2445 void netdev_rx_csum_fault(struct net_device *dev)
2447 if (net_ratelimit()) {
2448 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2452 EXPORT_SYMBOL(netdev_rx_csum_fault);
2455 /* Actually, we should eliminate this check as soon as we know, that:
2456 * 1. IOMMU is present and allows to map all the memory.
2457 * 2. No high memory really exists on this machine.
2460 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2462 #ifdef CONFIG_HIGHMEM
2464 if (!(dev->features & NETIF_F_HIGHDMA)) {
2465 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2466 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2467 if (PageHighMem(skb_frag_page(frag)))
2472 if (PCI_DMA_BUS_IS_PHYS) {
2473 struct device *pdev = dev->dev.parent;
2477 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2478 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2479 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2480 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2489 void (*destructor)(struct sk_buff *skb);
2492 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2494 static void dev_gso_skb_destructor(struct sk_buff *skb)
2496 struct dev_gso_cb *cb;
2498 kfree_skb_list(skb->next);
2501 cb = DEV_GSO_CB(skb);
2503 cb->destructor(skb);
2507 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2508 * @skb: buffer to segment
2509 * @features: device features as applicable to this skb
2511 * This function segments the given skb and stores the list of segments
2514 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2516 struct sk_buff *segs;
2518 segs = skb_gso_segment(skb, features);
2520 /* Verifying header integrity only. */
2525 return PTR_ERR(segs);
2528 DEV_GSO_CB(skb)->destructor = skb->destructor;
2529 skb->destructor = dev_gso_skb_destructor;
2534 /* If MPLS offload request, verify we are testing hardware MPLS features
2535 * instead of standard features for the netdev.
2537 #ifdef CONFIG_NET_MPLS_GSO
2538 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2539 netdev_features_t features,
2542 if (type == htons(ETH_P_MPLS_UC) || type == htons(ETH_P_MPLS_MC))
2543 features &= skb->dev->mpls_features;
2548 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2549 netdev_features_t features,
2556 static netdev_features_t harmonize_features(struct sk_buff *skb,
2557 netdev_features_t features)
2562 type = skb_network_protocol(skb, &tmp);
2563 features = net_mpls_features(skb, features, type);
2565 if (skb->ip_summed != CHECKSUM_NONE &&
2566 !can_checksum_protocol(features, type)) {
2567 features &= ~NETIF_F_ALL_CSUM;
2568 } else if (illegal_highdma(skb->dev, skb)) {
2569 features &= ~NETIF_F_SG;
2575 netdev_features_t netif_skb_features(struct sk_buff *skb)
2577 __be16 protocol = skb->protocol;
2578 netdev_features_t features = skb->dev->features;
2580 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2581 features &= ~NETIF_F_GSO_MASK;
2583 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2584 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2585 protocol = veh->h_vlan_encapsulated_proto;
2586 } else if (!vlan_tx_tag_present(skb)) {
2587 return harmonize_features(skb, features);
2590 features = netdev_intersect_features(features,
2591 skb->dev->vlan_features |
2592 NETIF_F_HW_VLAN_CTAG_TX |
2593 NETIF_F_HW_VLAN_STAG_TX);
2595 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2596 features = netdev_intersect_features(features,
2601 NETIF_F_HW_VLAN_CTAG_TX |
2602 NETIF_F_HW_VLAN_STAG_TX);
2604 return harmonize_features(skb, features);
2606 EXPORT_SYMBOL(netif_skb_features);
2608 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2609 struct netdev_queue *txq)
2611 const struct net_device_ops *ops = dev->netdev_ops;
2612 int rc = NETDEV_TX_OK;
2613 unsigned int skb_len;
2615 if (likely(!skb->next)) {
2616 netdev_features_t features;
2619 * If device doesn't need skb->dst, release it right now while
2620 * its hot in this cpu cache
2622 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2625 features = netif_skb_features(skb);
2627 if (vlan_tx_tag_present(skb) &&
2628 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2629 skb = __vlan_put_tag(skb, skb->vlan_proto,
2630 vlan_tx_tag_get(skb));
2637 /* If encapsulation offload request, verify we are testing
2638 * hardware encapsulation features instead of standard
2639 * features for the netdev
2641 if (skb->encapsulation)
2642 features &= dev->hw_enc_features;
2644 if (netif_needs_gso(skb, features)) {
2645 if (unlikely(dev_gso_segment(skb, features)))
2650 if (skb_needs_linearize(skb, features) &&
2651 __skb_linearize(skb))
2654 /* If packet is not checksummed and device does not
2655 * support checksumming for this protocol, complete
2656 * checksumming here.
2658 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2659 if (skb->encapsulation)
2660 skb_set_inner_transport_header(skb,
2661 skb_checksum_start_offset(skb));
2663 skb_set_transport_header(skb,
2664 skb_checksum_start_offset(skb));
2665 if (!(features & NETIF_F_ALL_CSUM) &&
2666 skb_checksum_help(skb))
2671 if (!list_empty(&ptype_all))
2672 dev_queue_xmit_nit(skb, dev);
2675 trace_net_dev_start_xmit(skb, dev);
2676 rc = ops->ndo_start_xmit(skb, dev);
2677 trace_net_dev_xmit(skb, rc, dev, skb_len);
2678 if (rc == NETDEV_TX_OK)
2679 txq_trans_update(txq);
2685 struct sk_buff *nskb = skb->next;
2687 skb->next = nskb->next;
2690 if (!list_empty(&ptype_all))
2691 dev_queue_xmit_nit(nskb, dev);
2693 skb_len = nskb->len;
2694 trace_net_dev_start_xmit(nskb, dev);
2695 rc = ops->ndo_start_xmit(nskb, dev);
2696 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2697 if (unlikely(rc != NETDEV_TX_OK)) {
2698 if (rc & ~NETDEV_TX_MASK)
2699 goto out_kfree_gso_skb;
2700 nskb->next = skb->next;
2704 txq_trans_update(txq);
2705 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2706 return NETDEV_TX_BUSY;
2707 } while (skb->next);
2710 if (likely(skb->next == NULL)) {
2711 skb->destructor = DEV_GSO_CB(skb)->destructor;
2720 EXPORT_SYMBOL_GPL(dev_hard_start_xmit);
2722 static void qdisc_pkt_len_init(struct sk_buff *skb)
2724 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2726 qdisc_skb_cb(skb)->pkt_len = skb->len;
2728 /* To get more precise estimation of bytes sent on wire,
2729 * we add to pkt_len the headers size of all segments
2731 if (shinfo->gso_size) {
2732 unsigned int hdr_len;
2733 u16 gso_segs = shinfo->gso_segs;
2735 /* mac layer + network layer */
2736 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2738 /* + transport layer */
2739 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2740 hdr_len += tcp_hdrlen(skb);
2742 hdr_len += sizeof(struct udphdr);
2744 if (shinfo->gso_type & SKB_GSO_DODGY)
2745 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2748 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2752 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2753 struct net_device *dev,
2754 struct netdev_queue *txq)
2756 spinlock_t *root_lock = qdisc_lock(q);
2760 qdisc_pkt_len_init(skb);
2761 qdisc_calculate_pkt_len(skb, q);
2763 * Heuristic to force contended enqueues to serialize on a
2764 * separate lock before trying to get qdisc main lock.
2765 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2766 * often and dequeue packets faster.
2768 contended = qdisc_is_running(q);
2769 if (unlikely(contended))
2770 spin_lock(&q->busylock);
2772 spin_lock(root_lock);
2773 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2776 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2777 qdisc_run_begin(q)) {
2779 * This is a work-conserving queue; there are no old skbs
2780 * waiting to be sent out; and the qdisc is not running -
2781 * xmit the skb directly.
2783 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2786 qdisc_bstats_update(q, skb);
2788 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2789 if (unlikely(contended)) {
2790 spin_unlock(&q->busylock);
2797 rc = NET_XMIT_SUCCESS;
2800 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2801 if (qdisc_run_begin(q)) {
2802 if (unlikely(contended)) {
2803 spin_unlock(&q->busylock);
2809 spin_unlock(root_lock);
2810 if (unlikely(contended))
2811 spin_unlock(&q->busylock);
2815 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2816 static void skb_update_prio(struct sk_buff *skb)
2818 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2820 if (!skb->priority && skb->sk && map) {
2821 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2823 if (prioidx < map->priomap_len)
2824 skb->priority = map->priomap[prioidx];
2828 #define skb_update_prio(skb)
2831 static DEFINE_PER_CPU(int, xmit_recursion);
2832 #define RECURSION_LIMIT 10
2835 * dev_loopback_xmit - loop back @skb
2836 * @skb: buffer to transmit
2838 int dev_loopback_xmit(struct sk_buff *skb)
2840 skb_reset_mac_header(skb);
2841 __skb_pull(skb, skb_network_offset(skb));
2842 skb->pkt_type = PACKET_LOOPBACK;
2843 skb->ip_summed = CHECKSUM_UNNECESSARY;
2844 WARN_ON(!skb_dst(skb));
2849 EXPORT_SYMBOL(dev_loopback_xmit);
2852 * __dev_queue_xmit - transmit a buffer
2853 * @skb: buffer to transmit
2854 * @accel_priv: private data used for L2 forwarding offload
2856 * Queue a buffer for transmission to a network device. The caller must
2857 * have set the device and priority and built the buffer before calling
2858 * this function. The function can be called from an interrupt.
2860 * A negative errno code is returned on a failure. A success does not
2861 * guarantee the frame will be transmitted as it may be dropped due
2862 * to congestion or traffic shaping.
2864 * -----------------------------------------------------------------------------------
2865 * I notice this method can also return errors from the queue disciplines,
2866 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2869 * Regardless of the return value, the skb is consumed, so it is currently
2870 * difficult to retry a send to this method. (You can bump the ref count
2871 * before sending to hold a reference for retry if you are careful.)
2873 * When calling this method, interrupts MUST be enabled. This is because
2874 * the BH enable code must have IRQs enabled so that it will not deadlock.
2877 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2879 struct net_device *dev = skb->dev;
2880 struct netdev_queue *txq;
2884 skb_reset_mac_header(skb);
2886 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
2887 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
2889 /* Disable soft irqs for various locks below. Also
2890 * stops preemption for RCU.
2894 skb_update_prio(skb);
2896 txq = netdev_pick_tx(dev, skb, accel_priv);
2897 q = rcu_dereference_bh(txq->qdisc);
2899 #ifdef CONFIG_NET_CLS_ACT
2900 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2902 trace_net_dev_queue(skb);
2904 rc = __dev_xmit_skb(skb, q, dev, txq);
2908 /* The device has no queue. Common case for software devices:
2909 loopback, all the sorts of tunnels...
2911 Really, it is unlikely that netif_tx_lock protection is necessary
2912 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2914 However, it is possible, that they rely on protection
2917 Check this and shot the lock. It is not prone from deadlocks.
2918 Either shot noqueue qdisc, it is even simpler 8)
2920 if (dev->flags & IFF_UP) {
2921 int cpu = smp_processor_id(); /* ok because BHs are off */
2923 if (txq->xmit_lock_owner != cpu) {
2925 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2926 goto recursion_alert;
2928 HARD_TX_LOCK(dev, txq, cpu);
2930 if (!netif_xmit_stopped(txq)) {
2931 __this_cpu_inc(xmit_recursion);
2932 rc = dev_hard_start_xmit(skb, dev, txq);
2933 __this_cpu_dec(xmit_recursion);
2934 if (dev_xmit_complete(rc)) {
2935 HARD_TX_UNLOCK(dev, txq);
2939 HARD_TX_UNLOCK(dev, txq);
2940 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2943 /* Recursion is detected! It is possible,
2947 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2953 rcu_read_unlock_bh();
2955 atomic_long_inc(&dev->tx_dropped);
2959 rcu_read_unlock_bh();
2963 int dev_queue_xmit(struct sk_buff *skb)
2965 return __dev_queue_xmit(skb, NULL);
2967 EXPORT_SYMBOL(dev_queue_xmit);
2969 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
2971 return __dev_queue_xmit(skb, accel_priv);
2973 EXPORT_SYMBOL(dev_queue_xmit_accel);
2976 /*=======================================================================
2978 =======================================================================*/
2980 int netdev_max_backlog __read_mostly = 1000;
2981 EXPORT_SYMBOL(netdev_max_backlog);
2983 int netdev_tstamp_prequeue __read_mostly = 1;
2984 int netdev_budget __read_mostly = 300;
2985 int weight_p __read_mostly = 64; /* old backlog weight */
2987 /* Called with irq disabled */
2988 static inline void ____napi_schedule(struct softnet_data *sd,
2989 struct napi_struct *napi)
2991 list_add_tail(&napi->poll_list, &sd->poll_list);
2992 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2997 /* One global table that all flow-based protocols share. */
2998 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2999 EXPORT_SYMBOL(rps_sock_flow_table);
3001 struct static_key rps_needed __read_mostly;
3003 static struct rps_dev_flow *
3004 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3005 struct rps_dev_flow *rflow, u16 next_cpu)
3007 if (next_cpu != RPS_NO_CPU) {
3008 #ifdef CONFIG_RFS_ACCEL
3009 struct netdev_rx_queue *rxqueue;
3010 struct rps_dev_flow_table *flow_table;
3011 struct rps_dev_flow *old_rflow;
3016 /* Should we steer this flow to a different hardware queue? */
3017 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3018 !(dev->features & NETIF_F_NTUPLE))
3020 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3021 if (rxq_index == skb_get_rx_queue(skb))
3024 rxqueue = dev->_rx + rxq_index;
3025 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3028 flow_id = skb_get_hash(skb) & flow_table->mask;
3029 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3030 rxq_index, flow_id);
3034 rflow = &flow_table->flows[flow_id];
3036 if (old_rflow->filter == rflow->filter)
3037 old_rflow->filter = RPS_NO_FILTER;
3041 per_cpu(softnet_data, next_cpu).input_queue_head;
3044 rflow->cpu = next_cpu;
3049 * get_rps_cpu is called from netif_receive_skb and returns the target
3050 * CPU from the RPS map of the receiving queue for a given skb.
3051 * rcu_read_lock must be held on entry.
3053 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3054 struct rps_dev_flow **rflowp)
3056 struct netdev_rx_queue *rxqueue;
3057 struct rps_map *map;
3058 struct rps_dev_flow_table *flow_table;
3059 struct rps_sock_flow_table *sock_flow_table;
3064 if (skb_rx_queue_recorded(skb)) {
3065 u16 index = skb_get_rx_queue(skb);
3066 if (unlikely(index >= dev->real_num_rx_queues)) {
3067 WARN_ONCE(dev->real_num_rx_queues > 1,
3068 "%s received packet on queue %u, but number "
3069 "of RX queues is %u\n",
3070 dev->name, index, dev->real_num_rx_queues);
3073 rxqueue = dev->_rx + index;
3077 map = rcu_dereference(rxqueue->rps_map);
3079 if (map->len == 1 &&
3080 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3081 tcpu = map->cpus[0];
3082 if (cpu_online(tcpu))
3086 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3090 skb_reset_network_header(skb);
3091 hash = skb_get_hash(skb);
3095 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3096 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3097 if (flow_table && sock_flow_table) {
3099 struct rps_dev_flow *rflow;
3101 rflow = &flow_table->flows[hash & flow_table->mask];
3104 next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask];
3107 * If the desired CPU (where last recvmsg was done) is
3108 * different from current CPU (one in the rx-queue flow
3109 * table entry), switch if one of the following holds:
3110 * - Current CPU is unset (equal to RPS_NO_CPU).
3111 * - Current CPU is offline.
3112 * - The current CPU's queue tail has advanced beyond the
3113 * last packet that was enqueued using this table entry.
3114 * This guarantees that all previous packets for the flow
3115 * have been dequeued, thus preserving in order delivery.
3117 if (unlikely(tcpu != next_cpu) &&
3118 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3119 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3120 rflow->last_qtail)) >= 0)) {
3122 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3125 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3133 tcpu = map->cpus[((u64) hash * map->len) >> 32];
3135 if (cpu_online(tcpu)) {
3145 #ifdef CONFIG_RFS_ACCEL
3148 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3149 * @dev: Device on which the filter was set
3150 * @rxq_index: RX queue index
3151 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3152 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3154 * Drivers that implement ndo_rx_flow_steer() should periodically call
3155 * this function for each installed filter and remove the filters for
3156 * which it returns %true.
3158 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3159 u32 flow_id, u16 filter_id)
3161 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3162 struct rps_dev_flow_table *flow_table;
3163 struct rps_dev_flow *rflow;
3168 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3169 if (flow_table && flow_id <= flow_table->mask) {
3170 rflow = &flow_table->flows[flow_id];
3171 cpu = ACCESS_ONCE(rflow->cpu);
3172 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3173 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3174 rflow->last_qtail) <
3175 (int)(10 * flow_table->mask)))
3181 EXPORT_SYMBOL(rps_may_expire_flow);
3183 #endif /* CONFIG_RFS_ACCEL */
3185 /* Called from hardirq (IPI) context */
3186 static void rps_trigger_softirq(void *data)
3188 struct softnet_data *sd = data;
3190 ____napi_schedule(sd, &sd->backlog);
3194 #endif /* CONFIG_RPS */
3197 * Check if this softnet_data structure is another cpu one
3198 * If yes, queue it to our IPI list and return 1
3201 static int rps_ipi_queued(struct softnet_data *sd)
3204 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3207 sd->rps_ipi_next = mysd->rps_ipi_list;
3208 mysd->rps_ipi_list = sd;
3210 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3213 #endif /* CONFIG_RPS */
3217 #ifdef CONFIG_NET_FLOW_LIMIT
3218 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3221 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3223 #ifdef CONFIG_NET_FLOW_LIMIT
3224 struct sd_flow_limit *fl;
3225 struct softnet_data *sd;
3226 unsigned int old_flow, new_flow;
3228 if (qlen < (netdev_max_backlog >> 1))
3231 sd = &__get_cpu_var(softnet_data);
3234 fl = rcu_dereference(sd->flow_limit);
3236 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3237 old_flow = fl->history[fl->history_head];
3238 fl->history[fl->history_head] = new_flow;
3241 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3243 if (likely(fl->buckets[old_flow]))
3244 fl->buckets[old_flow]--;
3246 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3258 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3259 * queue (may be a remote CPU queue).
3261 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3262 unsigned int *qtail)
3264 struct softnet_data *sd;
3265 unsigned long flags;
3268 sd = &per_cpu(softnet_data, cpu);
3270 local_irq_save(flags);
3273 qlen = skb_queue_len(&sd->input_pkt_queue);
3274 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3275 if (skb_queue_len(&sd->input_pkt_queue)) {
3277 __skb_queue_tail(&sd->input_pkt_queue, skb);
3278 input_queue_tail_incr_save(sd, qtail);
3280 local_irq_restore(flags);
3281 return NET_RX_SUCCESS;
3284 /* Schedule NAPI for backlog device
3285 * We can use non atomic operation since we own the queue lock
3287 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3288 if (!rps_ipi_queued(sd))
3289 ____napi_schedule(sd, &sd->backlog);
3297 local_irq_restore(flags);
3299 atomic_long_inc(&skb->dev->rx_dropped);
3304 static int netif_rx_internal(struct sk_buff *skb)
3308 net_timestamp_check(netdev_tstamp_prequeue, skb);
3310 trace_netif_rx(skb);
3312 if (static_key_false(&rps_needed)) {
3313 struct rps_dev_flow voidflow, *rflow = &voidflow;
3319 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3321 cpu = smp_processor_id();
3323 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3331 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3338 * netif_rx - post buffer to the network code
3339 * @skb: buffer to post
3341 * This function receives a packet from a device driver and queues it for
3342 * the upper (protocol) levels to process. It always succeeds. The buffer
3343 * may be dropped during processing for congestion control or by the
3347 * NET_RX_SUCCESS (no congestion)
3348 * NET_RX_DROP (packet was dropped)
3352 int netif_rx(struct sk_buff *skb)
3354 trace_netif_rx_entry(skb);
3356 return netif_rx_internal(skb);
3358 EXPORT_SYMBOL(netif_rx);
3360 int netif_rx_ni(struct sk_buff *skb)
3364 trace_netif_rx_ni_entry(skb);
3367 err = netif_rx_internal(skb);
3368 if (local_softirq_pending())
3374 EXPORT_SYMBOL(netif_rx_ni);
3376 static void net_tx_action(struct softirq_action *h)
3378 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3380 if (sd->completion_queue) {
3381 struct sk_buff *clist;
3383 local_irq_disable();
3384 clist = sd->completion_queue;
3385 sd->completion_queue = NULL;
3389 struct sk_buff *skb = clist;
3390 clist = clist->next;
3392 WARN_ON(atomic_read(&skb->users));
3393 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3394 trace_consume_skb(skb);
3396 trace_kfree_skb(skb, net_tx_action);
3401 if (sd->output_queue) {
3404 local_irq_disable();
3405 head = sd->output_queue;
3406 sd->output_queue = NULL;
3407 sd->output_queue_tailp = &sd->output_queue;
3411 struct Qdisc *q = head;
3412 spinlock_t *root_lock;
3414 head = head->next_sched;
3416 root_lock = qdisc_lock(q);
3417 if (spin_trylock(root_lock)) {
3418 smp_mb__before_atomic();
3419 clear_bit(__QDISC_STATE_SCHED,
3422 spin_unlock(root_lock);
3424 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3426 __netif_reschedule(q);
3428 smp_mb__before_atomic();
3429 clear_bit(__QDISC_STATE_SCHED,
3437 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3438 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3439 /* This hook is defined here for ATM LANE */
3440 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3441 unsigned char *addr) __read_mostly;
3442 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3445 #ifdef CONFIG_NET_CLS_ACT
3446 /* TODO: Maybe we should just force sch_ingress to be compiled in
3447 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3448 * a compare and 2 stores extra right now if we dont have it on
3449 * but have CONFIG_NET_CLS_ACT
3450 * NOTE: This doesn't stop any functionality; if you dont have
3451 * the ingress scheduler, you just can't add policies on ingress.
3454 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3456 struct net_device *dev = skb->dev;
3457 u32 ttl = G_TC_RTTL(skb->tc_verd);
3458 int result = TC_ACT_OK;
3461 if (unlikely(MAX_RED_LOOP < ttl++)) {
3462 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3463 skb->skb_iif, dev->ifindex);
3467 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3468 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3471 if (q != &noop_qdisc) {
3472 spin_lock(qdisc_lock(q));
3473 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3474 result = qdisc_enqueue_root(skb, q);
3475 spin_unlock(qdisc_lock(q));
3481 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3482 struct packet_type **pt_prev,
3483 int *ret, struct net_device *orig_dev)
3485 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3487 if (!rxq || rxq->qdisc == &noop_qdisc)
3491 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3495 switch (ing_filter(skb, rxq)) {
3509 * netdev_rx_handler_register - register receive handler
3510 * @dev: device to register a handler for
3511 * @rx_handler: receive handler to register
3512 * @rx_handler_data: data pointer that is used by rx handler
3514 * Register a receive handler for a device. This handler will then be
3515 * called from __netif_receive_skb. A negative errno code is returned
3518 * The caller must hold the rtnl_mutex.
3520 * For a general description of rx_handler, see enum rx_handler_result.
3522 int netdev_rx_handler_register(struct net_device *dev,
3523 rx_handler_func_t *rx_handler,
3524 void *rx_handler_data)
3528 if (dev->rx_handler)
3531 /* Note: rx_handler_data must be set before rx_handler */
3532 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3533 rcu_assign_pointer(dev->rx_handler, rx_handler);
3537 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3540 * netdev_rx_handler_unregister - unregister receive handler
3541 * @dev: device to unregister a handler from
3543 * Unregister a receive handler from a device.
3545 * The caller must hold the rtnl_mutex.
3547 void netdev_rx_handler_unregister(struct net_device *dev)
3551 RCU_INIT_POINTER(dev->rx_handler, NULL);
3552 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3553 * section has a guarantee to see a non NULL rx_handler_data
3557 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3559 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3562 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3563 * the special handling of PFMEMALLOC skbs.
3565 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3567 switch (skb->protocol) {
3568 case htons(ETH_P_ARP):
3569 case htons(ETH_P_IP):
3570 case htons(ETH_P_IPV6):
3571 case htons(ETH_P_8021Q):
3572 case htons(ETH_P_8021AD):
3579 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3581 struct packet_type *ptype, *pt_prev;
3582 rx_handler_func_t *rx_handler;
3583 struct net_device *orig_dev;
3584 struct net_device *null_or_dev;
3585 bool deliver_exact = false;
3586 int ret = NET_RX_DROP;
3589 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3591 trace_netif_receive_skb(skb);
3593 orig_dev = skb->dev;
3595 skb_reset_network_header(skb);
3596 if (!skb_transport_header_was_set(skb))
3597 skb_reset_transport_header(skb);
3598 skb_reset_mac_len(skb);
3605 skb->skb_iif = skb->dev->ifindex;
3607 __this_cpu_inc(softnet_data.processed);
3609 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3610 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3611 skb = skb_vlan_untag(skb);
3616 #ifdef CONFIG_NET_CLS_ACT
3617 if (skb->tc_verd & TC_NCLS) {
3618 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3626 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3627 if (!ptype->dev || ptype->dev == skb->dev) {
3629 ret = deliver_skb(skb, pt_prev, orig_dev);
3635 #ifdef CONFIG_NET_CLS_ACT
3636 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3642 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3645 if (vlan_tx_tag_present(skb)) {
3647 ret = deliver_skb(skb, pt_prev, orig_dev);
3650 if (vlan_do_receive(&skb))
3652 else if (unlikely(!skb))
3656 rx_handler = rcu_dereference(skb->dev->rx_handler);
3659 ret = deliver_skb(skb, pt_prev, orig_dev);
3662 switch (rx_handler(&skb)) {
3663 case RX_HANDLER_CONSUMED:
3664 ret = NET_RX_SUCCESS;
3666 case RX_HANDLER_ANOTHER:
3668 case RX_HANDLER_EXACT:
3669 deliver_exact = true;
3670 case RX_HANDLER_PASS:
3677 if (unlikely(vlan_tx_tag_present(skb))) {
3678 if (vlan_tx_tag_get_id(skb))
3679 skb->pkt_type = PACKET_OTHERHOST;
3680 /* Note: we might in the future use prio bits
3681 * and set skb->priority like in vlan_do_receive()
3682 * For the time being, just ignore Priority Code Point
3687 /* deliver only exact match when indicated */
3688 null_or_dev = deliver_exact ? skb->dev : NULL;
3690 type = skb->protocol;
3691 list_for_each_entry_rcu(ptype,
3692 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3693 if (ptype->type == type &&
3694 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3695 ptype->dev == orig_dev)) {
3697 ret = deliver_skb(skb, pt_prev, orig_dev);
3703 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3706 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3709 atomic_long_inc(&skb->dev->rx_dropped);
3711 /* Jamal, now you will not able to escape explaining
3712 * me how you were going to use this. :-)
3722 static int __netif_receive_skb(struct sk_buff *skb)
3726 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3727 unsigned long pflags = current->flags;
3730 * PFMEMALLOC skbs are special, they should
3731 * - be delivered to SOCK_MEMALLOC sockets only
3732 * - stay away from userspace
3733 * - have bounded memory usage
3735 * Use PF_MEMALLOC as this saves us from propagating the allocation
3736 * context down to all allocation sites.
3738 current->flags |= PF_MEMALLOC;
3739 ret = __netif_receive_skb_core(skb, true);
3740 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3742 ret = __netif_receive_skb_core(skb, false);
3747 static int netif_receive_skb_internal(struct sk_buff *skb)
3749 net_timestamp_check(netdev_tstamp_prequeue, skb);
3751 if (skb_defer_rx_timestamp(skb))
3752 return NET_RX_SUCCESS;
3755 if (static_key_false(&rps_needed)) {
3756 struct rps_dev_flow voidflow, *rflow = &voidflow;
3761 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3764 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3771 return __netif_receive_skb(skb);
3775 * netif_receive_skb - process receive buffer from network
3776 * @skb: buffer to process
3778 * netif_receive_skb() is the main receive data processing function.
3779 * It always succeeds. The buffer may be dropped during processing
3780 * for congestion control or by the protocol layers.
3782 * This function may only be called from softirq context and interrupts
3783 * should be enabled.
3785 * Return values (usually ignored):
3786 * NET_RX_SUCCESS: no congestion
3787 * NET_RX_DROP: packet was dropped
3789 int netif_receive_skb(struct sk_buff *skb)
3791 trace_netif_receive_skb_entry(skb);
3793 return netif_receive_skb_internal(skb);
3795 EXPORT_SYMBOL(netif_receive_skb);
3797 /* Network device is going away, flush any packets still pending
3798 * Called with irqs disabled.
3800 static void flush_backlog(void *arg)
3802 struct net_device *dev = arg;
3803 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3804 struct sk_buff *skb, *tmp;
3807 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3808 if (skb->dev == dev) {
3809 __skb_unlink(skb, &sd->input_pkt_queue);
3811 input_queue_head_incr(sd);
3816 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3817 if (skb->dev == dev) {
3818 __skb_unlink(skb, &sd->process_queue);
3820 input_queue_head_incr(sd);
3825 static int napi_gro_complete(struct sk_buff *skb)
3827 struct packet_offload *ptype;
3828 __be16 type = skb->protocol;
3829 struct list_head *head = &offload_base;
3832 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3834 if (NAPI_GRO_CB(skb)->count == 1) {
3835 skb_shinfo(skb)->gso_size = 0;
3840 list_for_each_entry_rcu(ptype, head, list) {
3841 if (ptype->type != type || !ptype->callbacks.gro_complete)
3844 err = ptype->callbacks.gro_complete(skb, 0);
3850 WARN_ON(&ptype->list == head);
3852 return NET_RX_SUCCESS;
3856 return netif_receive_skb_internal(skb);
3859 /* napi->gro_list contains packets ordered by age.
3860 * youngest packets at the head of it.
3861 * Complete skbs in reverse order to reduce latencies.
3863 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3865 struct sk_buff *skb, *prev = NULL;
3867 /* scan list and build reverse chain */
3868 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3873 for (skb = prev; skb; skb = prev) {
3876 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3880 napi_gro_complete(skb);
3884 napi->gro_list = NULL;
3886 EXPORT_SYMBOL(napi_gro_flush);
3888 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3891 unsigned int maclen = skb->dev->hard_header_len;
3892 u32 hash = skb_get_hash_raw(skb);
3894 for (p = napi->gro_list; p; p = p->next) {
3895 unsigned long diffs;
3897 NAPI_GRO_CB(p)->flush = 0;
3899 if (hash != skb_get_hash_raw(p)) {
3900 NAPI_GRO_CB(p)->same_flow = 0;
3904 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3905 diffs |= p->vlan_tci ^ skb->vlan_tci;
3906 if (maclen == ETH_HLEN)
3907 diffs |= compare_ether_header(skb_mac_header(p),
3908 skb_mac_header(skb));
3910 diffs = memcmp(skb_mac_header(p),
3911 skb_mac_header(skb),
3913 NAPI_GRO_CB(p)->same_flow = !diffs;
3917 static void skb_gro_reset_offset(struct sk_buff *skb)
3919 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3920 const skb_frag_t *frag0 = &pinfo->frags[0];
3922 NAPI_GRO_CB(skb)->data_offset = 0;
3923 NAPI_GRO_CB(skb)->frag0 = NULL;
3924 NAPI_GRO_CB(skb)->frag0_len = 0;
3926 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3928 !PageHighMem(skb_frag_page(frag0))) {
3929 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3930 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3934 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
3936 struct skb_shared_info *pinfo = skb_shinfo(skb);
3938 BUG_ON(skb->end - skb->tail < grow);
3940 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3942 skb->data_len -= grow;
3945 pinfo->frags[0].page_offset += grow;
3946 skb_frag_size_sub(&pinfo->frags[0], grow);
3948 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
3949 skb_frag_unref(skb, 0);
3950 memmove(pinfo->frags, pinfo->frags + 1,
3951 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
3955 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3957 struct sk_buff **pp = NULL;
3958 struct packet_offload *ptype;
3959 __be16 type = skb->protocol;
3960 struct list_head *head = &offload_base;
3962 enum gro_result ret;
3965 if (!(skb->dev->features & NETIF_F_GRO))
3968 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3971 gro_list_prepare(napi, skb);
3972 NAPI_GRO_CB(skb)->csum = skb->csum; /* Needed for CHECKSUM_COMPLETE */
3975 list_for_each_entry_rcu(ptype, head, list) {
3976 if (ptype->type != type || !ptype->callbacks.gro_receive)
3979 skb_set_network_header(skb, skb_gro_offset(skb));
3980 skb_reset_mac_len(skb);
3981 NAPI_GRO_CB(skb)->same_flow = 0;
3982 NAPI_GRO_CB(skb)->flush = 0;
3983 NAPI_GRO_CB(skb)->free = 0;
3984 NAPI_GRO_CB(skb)->udp_mark = 0;
3986 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3991 if (&ptype->list == head)
3994 same_flow = NAPI_GRO_CB(skb)->same_flow;
3995 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3998 struct sk_buff *nskb = *pp;
4002 napi_gro_complete(nskb);
4009 if (NAPI_GRO_CB(skb)->flush)
4012 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4013 struct sk_buff *nskb = napi->gro_list;
4015 /* locate the end of the list to select the 'oldest' flow */
4016 while (nskb->next) {
4022 napi_gro_complete(nskb);
4026 NAPI_GRO_CB(skb)->count = 1;
4027 NAPI_GRO_CB(skb)->age = jiffies;
4028 NAPI_GRO_CB(skb)->last = skb;
4029 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4030 skb->next = napi->gro_list;
4031 napi->gro_list = skb;
4035 grow = skb_gro_offset(skb) - skb_headlen(skb);
4037 gro_pull_from_frag0(skb, grow);
4046 struct packet_offload *gro_find_receive_by_type(__be16 type)
4048 struct list_head *offload_head = &offload_base;
4049 struct packet_offload *ptype;
4051 list_for_each_entry_rcu(ptype, offload_head, list) {
4052 if (ptype->type != type || !ptype->callbacks.gro_receive)
4058 EXPORT_SYMBOL(gro_find_receive_by_type);
4060 struct packet_offload *gro_find_complete_by_type(__be16 type)
4062 struct list_head *offload_head = &offload_base;
4063 struct packet_offload *ptype;
4065 list_for_each_entry_rcu(ptype, offload_head, list) {
4066 if (ptype->type != type || !ptype->callbacks.gro_complete)
4072 EXPORT_SYMBOL(gro_find_complete_by_type);
4074 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4078 if (netif_receive_skb_internal(skb))
4086 case GRO_MERGED_FREE:
4087 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4088 kmem_cache_free(skbuff_head_cache, skb);
4101 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4103 trace_napi_gro_receive_entry(skb);
4105 skb_gro_reset_offset(skb);
4107 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4109 EXPORT_SYMBOL(napi_gro_receive);
4111 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4113 __skb_pull(skb, skb_headlen(skb));
4114 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4115 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4117 skb->dev = napi->dev;
4119 skb->encapsulation = 0;
4120 skb_shinfo(skb)->gso_type = 0;
4121 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4126 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4128 struct sk_buff *skb = napi->skb;
4131 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4136 EXPORT_SYMBOL(napi_get_frags);
4138 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4139 struct sk_buff *skb,
4145 __skb_push(skb, ETH_HLEN);
4146 skb->protocol = eth_type_trans(skb, skb->dev);
4147 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4152 case GRO_MERGED_FREE:
4153 napi_reuse_skb(napi, skb);
4163 /* Upper GRO stack assumes network header starts at gro_offset=0
4164 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4165 * We copy ethernet header into skb->data to have a common layout.
4167 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4169 struct sk_buff *skb = napi->skb;
4170 const struct ethhdr *eth;
4171 unsigned int hlen = sizeof(*eth);
4175 skb_reset_mac_header(skb);
4176 skb_gro_reset_offset(skb);
4178 eth = skb_gro_header_fast(skb, 0);
4179 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4180 eth = skb_gro_header_slow(skb, hlen, 0);
4181 if (unlikely(!eth)) {
4182 napi_reuse_skb(napi, skb);
4186 gro_pull_from_frag0(skb, hlen);
4187 NAPI_GRO_CB(skb)->frag0 += hlen;
4188 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4190 __skb_pull(skb, hlen);
4193 * This works because the only protocols we care about don't require
4195 * We'll fix it up properly in napi_frags_finish()
4197 skb->protocol = eth->h_proto;
4202 gro_result_t napi_gro_frags(struct napi_struct *napi)
4204 struct sk_buff *skb = napi_frags_skb(napi);
4209 trace_napi_gro_frags_entry(skb);
4211 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4213 EXPORT_SYMBOL(napi_gro_frags);
4216 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4217 * Note: called with local irq disabled, but exits with local irq enabled.
4219 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4222 struct softnet_data *remsd = sd->rps_ipi_list;
4225 sd->rps_ipi_list = NULL;
4229 /* Send pending IPI's to kick RPS processing on remote cpus. */
4231 struct softnet_data *next = remsd->rps_ipi_next;
4233 if (cpu_online(remsd->cpu))
4234 smp_call_function_single_async(remsd->cpu,
4243 static int process_backlog(struct napi_struct *napi, int quota)
4246 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4249 /* Check if we have pending ipi, its better to send them now,
4250 * not waiting net_rx_action() end.
4252 if (sd->rps_ipi_list) {
4253 local_irq_disable();
4254 net_rps_action_and_irq_enable(sd);
4257 napi->weight = weight_p;
4258 local_irq_disable();
4260 struct sk_buff *skb;
4262 while ((skb = __skb_dequeue(&sd->process_queue))) {
4264 __netif_receive_skb(skb);
4265 local_irq_disable();
4266 input_queue_head_incr(sd);
4267 if (++work >= quota) {
4274 if (skb_queue_empty(&sd->input_pkt_queue)) {
4276 * Inline a custom version of __napi_complete().
4277 * only current cpu owns and manipulates this napi,
4278 * and NAPI_STATE_SCHED is the only possible flag set
4280 * We can use a plain write instead of clear_bit(),
4281 * and we dont need an smp_mb() memory barrier.
4283 list_del(&napi->poll_list);
4290 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4291 &sd->process_queue);
4300 * __napi_schedule - schedule for receive
4301 * @n: entry to schedule
4303 * The entry's receive function will be scheduled to run
4305 void __napi_schedule(struct napi_struct *n)
4307 unsigned long flags;
4309 local_irq_save(flags);
4310 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4311 local_irq_restore(flags);
4313 EXPORT_SYMBOL(__napi_schedule);
4315 void __napi_complete(struct napi_struct *n)
4317 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4318 BUG_ON(n->gro_list);
4320 list_del(&n->poll_list);
4321 smp_mb__before_atomic();
4322 clear_bit(NAPI_STATE_SCHED, &n->state);
4324 EXPORT_SYMBOL(__napi_complete);
4326 void napi_complete(struct napi_struct *n)
4328 unsigned long flags;
4331 * don't let napi dequeue from the cpu poll list
4332 * just in case its running on a different cpu
4334 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4337 napi_gro_flush(n, false);
4338 local_irq_save(flags);
4340 local_irq_restore(flags);
4342 EXPORT_SYMBOL(napi_complete);
4344 /* must be called under rcu_read_lock(), as we dont take a reference */
4345 struct napi_struct *napi_by_id(unsigned int napi_id)
4347 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4348 struct napi_struct *napi;
4350 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4351 if (napi->napi_id == napi_id)
4356 EXPORT_SYMBOL_GPL(napi_by_id);
4358 void napi_hash_add(struct napi_struct *napi)
4360 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4362 spin_lock(&napi_hash_lock);
4364 /* 0 is not a valid id, we also skip an id that is taken
4365 * we expect both events to be extremely rare
4368 while (!napi->napi_id) {
4369 napi->napi_id = ++napi_gen_id;
4370 if (napi_by_id(napi->napi_id))
4374 hlist_add_head_rcu(&napi->napi_hash_node,
4375 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4377 spin_unlock(&napi_hash_lock);
4380 EXPORT_SYMBOL_GPL(napi_hash_add);
4382 /* Warning : caller is responsible to make sure rcu grace period
4383 * is respected before freeing memory containing @napi
4385 void napi_hash_del(struct napi_struct *napi)
4387 spin_lock(&napi_hash_lock);
4389 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4390 hlist_del_rcu(&napi->napi_hash_node);
4392 spin_unlock(&napi_hash_lock);
4394 EXPORT_SYMBOL_GPL(napi_hash_del);
4396 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4397 int (*poll)(struct napi_struct *, int), int weight)
4399 INIT_LIST_HEAD(&napi->poll_list);
4400 napi->gro_count = 0;
4401 napi->gro_list = NULL;
4404 if (weight > NAPI_POLL_WEIGHT)
4405 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4407 napi->weight = weight;
4408 list_add(&napi->dev_list, &dev->napi_list);
4410 #ifdef CONFIG_NETPOLL
4411 spin_lock_init(&napi->poll_lock);
4412 napi->poll_owner = -1;
4414 set_bit(NAPI_STATE_SCHED, &napi->state);
4416 EXPORT_SYMBOL(netif_napi_add);
4418 void netif_napi_del(struct napi_struct *napi)
4420 list_del_init(&napi->dev_list);
4421 napi_free_frags(napi);
4423 kfree_skb_list(napi->gro_list);
4424 napi->gro_list = NULL;
4425 napi->gro_count = 0;
4427 EXPORT_SYMBOL(netif_napi_del);
4429 static void net_rx_action(struct softirq_action *h)
4431 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4432 unsigned long time_limit = jiffies + 2;
4433 int budget = netdev_budget;
4436 local_irq_disable();
4438 while (!list_empty(&sd->poll_list)) {
4439 struct napi_struct *n;
4442 /* If softirq window is exhuasted then punt.
4443 * Allow this to run for 2 jiffies since which will allow
4444 * an average latency of 1.5/HZ.
4446 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4451 /* Even though interrupts have been re-enabled, this
4452 * access is safe because interrupts can only add new
4453 * entries to the tail of this list, and only ->poll()
4454 * calls can remove this head entry from the list.
4456 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4458 have = netpoll_poll_lock(n);
4462 /* This NAPI_STATE_SCHED test is for avoiding a race
4463 * with netpoll's poll_napi(). Only the entity which
4464 * obtains the lock and sees NAPI_STATE_SCHED set will
4465 * actually make the ->poll() call. Therefore we avoid
4466 * accidentally calling ->poll() when NAPI is not scheduled.
4469 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4470 work = n->poll(n, weight);
4474 WARN_ON_ONCE(work > weight);
4478 local_irq_disable();
4480 /* Drivers must not modify the NAPI state if they
4481 * consume the entire weight. In such cases this code
4482 * still "owns" the NAPI instance and therefore can
4483 * move the instance around on the list at-will.
4485 if (unlikely(work == weight)) {
4486 if (unlikely(napi_disable_pending(n))) {
4489 local_irq_disable();
4492 /* flush too old packets
4493 * If HZ < 1000, flush all packets.
4496 napi_gro_flush(n, HZ >= 1000);
4497 local_irq_disable();
4499 list_move_tail(&n->poll_list, &sd->poll_list);
4503 netpoll_poll_unlock(have);
4506 net_rps_action_and_irq_enable(sd);
4508 #ifdef CONFIG_NET_DMA
4510 * There may not be any more sk_buffs coming right now, so push
4511 * any pending DMA copies to hardware
4513 dma_issue_pending_all();
4520 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4524 struct netdev_adjacent {
4525 struct net_device *dev;
4527 /* upper master flag, there can only be one master device per list */
4530 /* counter for the number of times this device was added to us */
4533 /* private field for the users */
4536 struct list_head list;
4537 struct rcu_head rcu;
4540 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4541 struct net_device *adj_dev,
4542 struct list_head *adj_list)
4544 struct netdev_adjacent *adj;
4546 list_for_each_entry(adj, adj_list, list) {
4547 if (adj->dev == adj_dev)
4554 * netdev_has_upper_dev - Check if device is linked to an upper device
4556 * @upper_dev: upper device to check
4558 * Find out if a device is linked to specified upper device and return true
4559 * in case it is. Note that this checks only immediate upper device,
4560 * not through a complete stack of devices. The caller must hold the RTNL lock.
4562 bool netdev_has_upper_dev(struct net_device *dev,
4563 struct net_device *upper_dev)
4567 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4569 EXPORT_SYMBOL(netdev_has_upper_dev);
4572 * netdev_has_any_upper_dev - Check if device is linked to some device
4575 * Find out if a device is linked to an upper device and return true in case
4576 * it is. The caller must hold the RTNL lock.
4578 static bool netdev_has_any_upper_dev(struct net_device *dev)
4582 return !list_empty(&dev->all_adj_list.upper);
4586 * netdev_master_upper_dev_get - Get master upper device
4589 * Find a master upper device and return pointer to it or NULL in case
4590 * it's not there. The caller must hold the RTNL lock.
4592 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4594 struct netdev_adjacent *upper;
4598 if (list_empty(&dev->adj_list.upper))
4601 upper = list_first_entry(&dev->adj_list.upper,
4602 struct netdev_adjacent, list);
4603 if (likely(upper->master))
4607 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4609 void *netdev_adjacent_get_private(struct list_head *adj_list)
4611 struct netdev_adjacent *adj;
4613 adj = list_entry(adj_list, struct netdev_adjacent, list);
4615 return adj->private;
4617 EXPORT_SYMBOL(netdev_adjacent_get_private);
4620 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4622 * @iter: list_head ** of the current position
4624 * Gets the next device from the dev's upper list, starting from iter
4625 * position. The caller must hold RCU read lock.
4627 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4628 struct list_head **iter)
4630 struct netdev_adjacent *upper;
4632 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4634 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4636 if (&upper->list == &dev->adj_list.upper)
4639 *iter = &upper->list;
4643 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4646 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4648 * @iter: list_head ** of the current position
4650 * Gets the next device from the dev's upper list, starting from iter
4651 * position. The caller must hold RCU read lock.
4653 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4654 struct list_head **iter)
4656 struct netdev_adjacent *upper;
4658 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4660 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4662 if (&upper->list == &dev->all_adj_list.upper)
4665 *iter = &upper->list;
4669 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4672 * netdev_lower_get_next_private - Get the next ->private from the
4673 * lower neighbour list
4675 * @iter: list_head ** of the current position
4677 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4678 * list, starting from iter position. The caller must hold either hold the
4679 * RTNL lock or its own locking that guarantees that the neighbour lower
4680 * list will remain unchainged.
4682 void *netdev_lower_get_next_private(struct net_device *dev,
4683 struct list_head **iter)
4685 struct netdev_adjacent *lower;
4687 lower = list_entry(*iter, struct netdev_adjacent, list);
4689 if (&lower->list == &dev->adj_list.lower)
4692 *iter = lower->list.next;
4694 return lower->private;
4696 EXPORT_SYMBOL(netdev_lower_get_next_private);
4699 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4700 * lower neighbour list, RCU
4703 * @iter: list_head ** of the current position
4705 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4706 * list, starting from iter position. The caller must hold RCU read lock.
4708 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4709 struct list_head **iter)
4711 struct netdev_adjacent *lower;
4713 WARN_ON_ONCE(!rcu_read_lock_held());
4715 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4717 if (&lower->list == &dev->adj_list.lower)
4720 *iter = &lower->list;
4722 return lower->private;
4724 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4727 * netdev_lower_get_next - Get the next device from the lower neighbour
4730 * @iter: list_head ** of the current position
4732 * Gets the next netdev_adjacent from the dev's lower neighbour
4733 * list, starting from iter position. The caller must hold RTNL lock or
4734 * its own locking that guarantees that the neighbour lower
4735 * list will remain unchainged.
4737 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4739 struct netdev_adjacent *lower;
4741 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4743 if (&lower->list == &dev->adj_list.lower)
4746 *iter = &lower->list;
4750 EXPORT_SYMBOL(netdev_lower_get_next);
4753 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4754 * lower neighbour list, RCU
4758 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4759 * list. The caller must hold RCU read lock.
4761 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4763 struct netdev_adjacent *lower;
4765 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4766 struct netdev_adjacent, list);
4768 return lower->private;
4771 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4774 * netdev_master_upper_dev_get_rcu - Get master upper device
4777 * Find a master upper device and return pointer to it or NULL in case
4778 * it's not there. The caller must hold the RCU read lock.
4780 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4782 struct netdev_adjacent *upper;
4784 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4785 struct netdev_adjacent, list);
4786 if (upper && likely(upper->master))
4790 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4792 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4793 struct net_device *adj_dev,
4794 struct list_head *dev_list)
4796 char linkname[IFNAMSIZ+7];
4797 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4798 "upper_%s" : "lower_%s", adj_dev->name);
4799 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4802 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4804 struct list_head *dev_list)
4806 char linkname[IFNAMSIZ+7];
4807 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4808 "upper_%s" : "lower_%s", name);
4809 sysfs_remove_link(&(dev->dev.kobj), linkname);
4812 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4813 (dev_list == &dev->adj_list.upper || \
4814 dev_list == &dev->adj_list.lower)
4816 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4817 struct net_device *adj_dev,
4818 struct list_head *dev_list,
4819 void *private, bool master)
4821 struct netdev_adjacent *adj;
4824 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4831 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4836 adj->master = master;
4838 adj->private = private;
4841 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4842 adj_dev->name, dev->name, adj_dev->name);
4844 if (netdev_adjacent_is_neigh_list(dev, dev_list)) {
4845 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
4850 /* Ensure that master link is always the first item in list. */
4852 ret = sysfs_create_link(&(dev->dev.kobj),
4853 &(adj_dev->dev.kobj), "master");
4855 goto remove_symlinks;
4857 list_add_rcu(&adj->list, dev_list);
4859 list_add_tail_rcu(&adj->list, dev_list);
4865 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4866 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4874 static void __netdev_adjacent_dev_remove(struct net_device *dev,
4875 struct net_device *adj_dev,
4876 struct list_head *dev_list)
4878 struct netdev_adjacent *adj;
4880 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4883 pr_err("tried to remove device %s from %s\n",
4884 dev->name, adj_dev->name);
4888 if (adj->ref_nr > 1) {
4889 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4896 sysfs_remove_link(&(dev->dev.kobj), "master");
4898 if (netdev_adjacent_is_neigh_list(dev, dev_list) &&
4899 net_eq(dev_net(dev),dev_net(adj_dev)))
4900 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4902 list_del_rcu(&adj->list);
4903 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4904 adj_dev->name, dev->name, adj_dev->name);
4906 kfree_rcu(adj, rcu);
4909 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4910 struct net_device *upper_dev,
4911 struct list_head *up_list,
4912 struct list_head *down_list,
4913 void *private, bool master)
4917 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4922 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4925 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4932 static int __netdev_adjacent_dev_link(struct net_device *dev,
4933 struct net_device *upper_dev)
4935 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4936 &dev->all_adj_list.upper,
4937 &upper_dev->all_adj_list.lower,
4941 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4942 struct net_device *upper_dev,
4943 struct list_head *up_list,
4944 struct list_head *down_list)
4946 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4947 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4950 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
4951 struct net_device *upper_dev)
4953 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4954 &dev->all_adj_list.upper,
4955 &upper_dev->all_adj_list.lower);
4958 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4959 struct net_device *upper_dev,
4960 void *private, bool master)
4962 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4967 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4968 &dev->adj_list.upper,
4969 &upper_dev->adj_list.lower,
4972 __netdev_adjacent_dev_unlink(dev, upper_dev);
4979 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4980 struct net_device *upper_dev)
4982 __netdev_adjacent_dev_unlink(dev, upper_dev);
4983 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4984 &dev->adj_list.upper,
4985 &upper_dev->adj_list.lower);
4988 static int __netdev_upper_dev_link(struct net_device *dev,
4989 struct net_device *upper_dev, bool master,
4992 struct netdev_adjacent *i, *j, *to_i, *to_j;
4997 if (dev == upper_dev)
5000 /* To prevent loops, check if dev is not upper device to upper_dev. */
5001 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5004 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
5007 if (master && netdev_master_upper_dev_get(dev))
5010 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5015 /* Now that we linked these devs, make all the upper_dev's
5016 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5017 * versa, and don't forget the devices itself. All of these
5018 * links are non-neighbours.
5020 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5021 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5022 pr_debug("Interlinking %s with %s, non-neighbour\n",
5023 i->dev->name, j->dev->name);
5024 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5030 /* add dev to every upper_dev's upper device */
5031 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5032 pr_debug("linking %s's upper device %s with %s\n",
5033 upper_dev->name, i->dev->name, dev->name);
5034 ret = __netdev_adjacent_dev_link(dev, i->dev);
5036 goto rollback_upper_mesh;
5039 /* add upper_dev to every dev's lower device */
5040 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5041 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5042 i->dev->name, upper_dev->name);
5043 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5045 goto rollback_lower_mesh;
5048 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5051 rollback_lower_mesh:
5053 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5056 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5061 rollback_upper_mesh:
5063 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5066 __netdev_adjacent_dev_unlink(dev, i->dev);
5074 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5075 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5076 if (i == to_i && j == to_j)
5078 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5084 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5090 * netdev_upper_dev_link - Add a link to the upper device
5092 * @upper_dev: new upper device
5094 * Adds a link to device which is upper to this one. The caller must hold
5095 * the RTNL lock. On a failure a negative errno code is returned.
5096 * On success the reference counts are adjusted and the function
5099 int netdev_upper_dev_link(struct net_device *dev,
5100 struct net_device *upper_dev)
5102 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5104 EXPORT_SYMBOL(netdev_upper_dev_link);
5107 * netdev_master_upper_dev_link - Add a master link to the upper device
5109 * @upper_dev: new upper device
5111 * Adds a link to device which is upper to this one. In this case, only
5112 * one master upper device can be linked, although other non-master devices
5113 * might be linked as well. The caller must hold the RTNL lock.
5114 * On a failure a negative errno code is returned. On success the reference
5115 * counts are adjusted and the function returns zero.
5117 int netdev_master_upper_dev_link(struct net_device *dev,
5118 struct net_device *upper_dev)
5120 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5122 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5124 int netdev_master_upper_dev_link_private(struct net_device *dev,
5125 struct net_device *upper_dev,
5128 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5130 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5133 * netdev_upper_dev_unlink - Removes a link to upper device
5135 * @upper_dev: new upper device
5137 * Removes a link to device which is upper to this one. The caller must hold
5140 void netdev_upper_dev_unlink(struct net_device *dev,
5141 struct net_device *upper_dev)
5143 struct netdev_adjacent *i, *j;
5146 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5148 /* Here is the tricky part. We must remove all dev's lower
5149 * devices from all upper_dev's upper devices and vice
5150 * versa, to maintain the graph relationship.
5152 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5153 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5154 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5156 /* remove also the devices itself from lower/upper device
5159 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5160 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5162 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5163 __netdev_adjacent_dev_unlink(dev, i->dev);
5165 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5167 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5169 void netdev_adjacent_add_links(struct net_device *dev)
5171 struct netdev_adjacent *iter;
5173 struct net *net = dev_net(dev);
5175 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5176 if (!net_eq(net,dev_net(iter->dev)))
5178 netdev_adjacent_sysfs_add(iter->dev, dev,
5179 &iter->dev->adj_list.lower);
5180 netdev_adjacent_sysfs_add(dev, iter->dev,
5181 &dev->adj_list.upper);
5184 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5185 if (!net_eq(net,dev_net(iter->dev)))
5187 netdev_adjacent_sysfs_add(iter->dev, dev,
5188 &iter->dev->adj_list.upper);
5189 netdev_adjacent_sysfs_add(dev, iter->dev,
5190 &dev->adj_list.lower);
5194 void netdev_adjacent_del_links(struct net_device *dev)
5196 struct netdev_adjacent *iter;
5198 struct net *net = dev_net(dev);
5200 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5201 if (!net_eq(net,dev_net(iter->dev)))
5203 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5204 &iter->dev->adj_list.lower);
5205 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5206 &dev->adj_list.upper);
5209 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5210 if (!net_eq(net,dev_net(iter->dev)))
5212 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5213 &iter->dev->adj_list.upper);
5214 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5215 &dev->adj_list.lower);
5219 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5221 struct netdev_adjacent *iter;
5223 struct net *net = dev_net(dev);
5225 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5226 if (!net_eq(net,dev_net(iter->dev)))
5228 netdev_adjacent_sysfs_del(iter->dev, oldname,
5229 &iter->dev->adj_list.lower);
5230 netdev_adjacent_sysfs_add(iter->dev, dev,
5231 &iter->dev->adj_list.lower);
5234 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5235 if (!net_eq(net,dev_net(iter->dev)))
5237 netdev_adjacent_sysfs_del(iter->dev, oldname,
5238 &iter->dev->adj_list.upper);
5239 netdev_adjacent_sysfs_add(iter->dev, dev,
5240 &iter->dev->adj_list.upper);
5244 void *netdev_lower_dev_get_private(struct net_device *dev,
5245 struct net_device *lower_dev)
5247 struct netdev_adjacent *lower;
5251 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5255 return lower->private;
5257 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5260 int dev_get_nest_level(struct net_device *dev,
5261 bool (*type_check)(struct net_device *dev))
5263 struct net_device *lower = NULL;
5264 struct list_head *iter;
5270 netdev_for_each_lower_dev(dev, lower, iter) {
5271 nest = dev_get_nest_level(lower, type_check);
5272 if (max_nest < nest)
5276 if (type_check(dev))
5281 EXPORT_SYMBOL(dev_get_nest_level);
5283 static void dev_change_rx_flags(struct net_device *dev, int flags)
5285 const struct net_device_ops *ops = dev->netdev_ops;
5287 if (ops->ndo_change_rx_flags)
5288 ops->ndo_change_rx_flags(dev, flags);
5291 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5293 unsigned int old_flags = dev->flags;
5299 dev->flags |= IFF_PROMISC;
5300 dev->promiscuity += inc;
5301 if (dev->promiscuity == 0) {
5304 * If inc causes overflow, untouch promisc and return error.
5307 dev->flags &= ~IFF_PROMISC;
5309 dev->promiscuity -= inc;
5310 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5315 if (dev->flags != old_flags) {
5316 pr_info("device %s %s promiscuous mode\n",
5318 dev->flags & IFF_PROMISC ? "entered" : "left");
5319 if (audit_enabled) {
5320 current_uid_gid(&uid, &gid);
5321 audit_log(current->audit_context, GFP_ATOMIC,
5322 AUDIT_ANOM_PROMISCUOUS,
5323 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5324 dev->name, (dev->flags & IFF_PROMISC),
5325 (old_flags & IFF_PROMISC),
5326 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5327 from_kuid(&init_user_ns, uid),
5328 from_kgid(&init_user_ns, gid),
5329 audit_get_sessionid(current));
5332 dev_change_rx_flags(dev, IFF_PROMISC);
5335 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5340 * dev_set_promiscuity - update promiscuity count on a device
5344 * Add or remove promiscuity from a device. While the count in the device
5345 * remains above zero the interface remains promiscuous. Once it hits zero
5346 * the device reverts back to normal filtering operation. A negative inc
5347 * value is used to drop promiscuity on the device.
5348 * Return 0 if successful or a negative errno code on error.
5350 int dev_set_promiscuity(struct net_device *dev, int inc)
5352 unsigned int old_flags = dev->flags;
5355 err = __dev_set_promiscuity(dev, inc, true);
5358 if (dev->flags != old_flags)
5359 dev_set_rx_mode(dev);
5362 EXPORT_SYMBOL(dev_set_promiscuity);
5364 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5366 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5370 dev->flags |= IFF_ALLMULTI;
5371 dev->allmulti += inc;
5372 if (dev->allmulti == 0) {
5375 * If inc causes overflow, untouch allmulti and return error.
5378 dev->flags &= ~IFF_ALLMULTI;
5380 dev->allmulti -= inc;
5381 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5386 if (dev->flags ^ old_flags) {
5387 dev_change_rx_flags(dev, IFF_ALLMULTI);
5388 dev_set_rx_mode(dev);
5390 __dev_notify_flags(dev, old_flags,
5391 dev->gflags ^ old_gflags);
5397 * dev_set_allmulti - update allmulti count on a device
5401 * Add or remove reception of all multicast frames to a device. While the
5402 * count in the device remains above zero the interface remains listening
5403 * to all interfaces. Once it hits zero the device reverts back to normal
5404 * filtering operation. A negative @inc value is used to drop the counter
5405 * when releasing a resource needing all multicasts.
5406 * Return 0 if successful or a negative errno code on error.
5409 int dev_set_allmulti(struct net_device *dev, int inc)
5411 return __dev_set_allmulti(dev, inc, true);
5413 EXPORT_SYMBOL(dev_set_allmulti);
5416 * Upload unicast and multicast address lists to device and
5417 * configure RX filtering. When the device doesn't support unicast
5418 * filtering it is put in promiscuous mode while unicast addresses
5421 void __dev_set_rx_mode(struct net_device *dev)
5423 const struct net_device_ops *ops = dev->netdev_ops;
5425 /* dev_open will call this function so the list will stay sane. */
5426 if (!(dev->flags&IFF_UP))
5429 if (!netif_device_present(dev))
5432 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5433 /* Unicast addresses changes may only happen under the rtnl,
5434 * therefore calling __dev_set_promiscuity here is safe.
5436 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5437 __dev_set_promiscuity(dev, 1, false);
5438 dev->uc_promisc = true;
5439 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5440 __dev_set_promiscuity(dev, -1, false);
5441 dev->uc_promisc = false;
5445 if (ops->ndo_set_rx_mode)
5446 ops->ndo_set_rx_mode(dev);
5449 void dev_set_rx_mode(struct net_device *dev)
5451 netif_addr_lock_bh(dev);
5452 __dev_set_rx_mode(dev);
5453 netif_addr_unlock_bh(dev);
5457 * dev_get_flags - get flags reported to userspace
5460 * Get the combination of flag bits exported through APIs to userspace.
5462 unsigned int dev_get_flags(const struct net_device *dev)
5466 flags = (dev->flags & ~(IFF_PROMISC |
5471 (dev->gflags & (IFF_PROMISC |
5474 if (netif_running(dev)) {
5475 if (netif_oper_up(dev))
5476 flags |= IFF_RUNNING;
5477 if (netif_carrier_ok(dev))
5478 flags |= IFF_LOWER_UP;
5479 if (netif_dormant(dev))
5480 flags |= IFF_DORMANT;
5485 EXPORT_SYMBOL(dev_get_flags);
5487 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5489 unsigned int old_flags = dev->flags;
5495 * Set the flags on our device.
5498 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5499 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5501 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5505 * Load in the correct multicast list now the flags have changed.
5508 if ((old_flags ^ flags) & IFF_MULTICAST)
5509 dev_change_rx_flags(dev, IFF_MULTICAST);
5511 dev_set_rx_mode(dev);
5514 * Have we downed the interface. We handle IFF_UP ourselves
5515 * according to user attempts to set it, rather than blindly
5520 if ((old_flags ^ flags) & IFF_UP)
5521 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5523 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5524 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5525 unsigned int old_flags = dev->flags;
5527 dev->gflags ^= IFF_PROMISC;
5529 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5530 if (dev->flags != old_flags)
5531 dev_set_rx_mode(dev);
5534 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5535 is important. Some (broken) drivers set IFF_PROMISC, when
5536 IFF_ALLMULTI is requested not asking us and not reporting.
5538 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5539 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5541 dev->gflags ^= IFF_ALLMULTI;
5542 __dev_set_allmulti(dev, inc, false);
5548 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5549 unsigned int gchanges)
5551 unsigned int changes = dev->flags ^ old_flags;
5554 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5556 if (changes & IFF_UP) {
5557 if (dev->flags & IFF_UP)
5558 call_netdevice_notifiers(NETDEV_UP, dev);
5560 call_netdevice_notifiers(NETDEV_DOWN, dev);
5563 if (dev->flags & IFF_UP &&
5564 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5565 struct netdev_notifier_change_info change_info;
5567 change_info.flags_changed = changes;
5568 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5574 * dev_change_flags - change device settings
5576 * @flags: device state flags
5578 * Change settings on device based state flags. The flags are
5579 * in the userspace exported format.
5581 int dev_change_flags(struct net_device *dev, unsigned int flags)
5584 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5586 ret = __dev_change_flags(dev, flags);
5590 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5591 __dev_notify_flags(dev, old_flags, changes);
5594 EXPORT_SYMBOL(dev_change_flags);
5596 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5598 const struct net_device_ops *ops = dev->netdev_ops;
5600 if (ops->ndo_change_mtu)
5601 return ops->ndo_change_mtu(dev, new_mtu);
5608 * dev_set_mtu - Change maximum transfer unit
5610 * @new_mtu: new transfer unit
5612 * Change the maximum transfer size of the network device.
5614 int dev_set_mtu(struct net_device *dev, int new_mtu)
5618 if (new_mtu == dev->mtu)
5621 /* MTU must be positive. */
5625 if (!netif_device_present(dev))
5628 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5629 err = notifier_to_errno(err);
5633 orig_mtu = dev->mtu;
5634 err = __dev_set_mtu(dev, new_mtu);
5637 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5638 err = notifier_to_errno(err);
5640 /* setting mtu back and notifying everyone again,
5641 * so that they have a chance to revert changes.
5643 __dev_set_mtu(dev, orig_mtu);
5644 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5649 EXPORT_SYMBOL(dev_set_mtu);
5652 * dev_set_group - Change group this device belongs to
5654 * @new_group: group this device should belong to
5656 void dev_set_group(struct net_device *dev, int new_group)
5658 dev->group = new_group;
5660 EXPORT_SYMBOL(dev_set_group);
5663 * dev_set_mac_address - Change Media Access Control Address
5667 * Change the hardware (MAC) address of the device
5669 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5671 const struct net_device_ops *ops = dev->netdev_ops;
5674 if (!ops->ndo_set_mac_address)
5676 if (sa->sa_family != dev->type)
5678 if (!netif_device_present(dev))
5680 err = ops->ndo_set_mac_address(dev, sa);
5683 dev->addr_assign_type = NET_ADDR_SET;
5684 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5685 add_device_randomness(dev->dev_addr, dev->addr_len);
5688 EXPORT_SYMBOL(dev_set_mac_address);
5691 * dev_change_carrier - Change device carrier
5693 * @new_carrier: new value
5695 * Change device carrier
5697 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5699 const struct net_device_ops *ops = dev->netdev_ops;
5701 if (!ops->ndo_change_carrier)
5703 if (!netif_device_present(dev))
5705 return ops->ndo_change_carrier(dev, new_carrier);
5707 EXPORT_SYMBOL(dev_change_carrier);
5710 * dev_get_phys_port_id - Get device physical port ID
5714 * Get device physical port ID
5716 int dev_get_phys_port_id(struct net_device *dev,
5717 struct netdev_phys_port_id *ppid)
5719 const struct net_device_ops *ops = dev->netdev_ops;
5721 if (!ops->ndo_get_phys_port_id)
5723 return ops->ndo_get_phys_port_id(dev, ppid);
5725 EXPORT_SYMBOL(dev_get_phys_port_id);
5728 * dev_new_index - allocate an ifindex
5729 * @net: the applicable net namespace
5731 * Returns a suitable unique value for a new device interface
5732 * number. The caller must hold the rtnl semaphore or the
5733 * dev_base_lock to be sure it remains unique.
5735 static int dev_new_index(struct net *net)
5737 int ifindex = net->ifindex;
5741 if (!__dev_get_by_index(net, ifindex))
5742 return net->ifindex = ifindex;
5746 /* Delayed registration/unregisteration */
5747 static LIST_HEAD(net_todo_list);
5748 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5750 static void net_set_todo(struct net_device *dev)
5752 list_add_tail(&dev->todo_list, &net_todo_list);
5753 dev_net(dev)->dev_unreg_count++;
5756 static void rollback_registered_many(struct list_head *head)
5758 struct net_device *dev, *tmp;
5759 LIST_HEAD(close_head);
5761 BUG_ON(dev_boot_phase);
5764 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5765 /* Some devices call without registering
5766 * for initialization unwind. Remove those
5767 * devices and proceed with the remaining.
5769 if (dev->reg_state == NETREG_UNINITIALIZED) {
5770 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5774 list_del(&dev->unreg_list);
5777 dev->dismantle = true;
5778 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5781 /* If device is running, close it first. */
5782 list_for_each_entry(dev, head, unreg_list)
5783 list_add_tail(&dev->close_list, &close_head);
5784 dev_close_many(&close_head);
5786 list_for_each_entry(dev, head, unreg_list) {
5787 /* And unlink it from device chain. */
5788 unlist_netdevice(dev);
5790 dev->reg_state = NETREG_UNREGISTERING;
5795 list_for_each_entry(dev, head, unreg_list) {
5796 /* Shutdown queueing discipline. */
5800 /* Notify protocols, that we are about to destroy
5801 this device. They should clean all the things.
5803 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5806 * Flush the unicast and multicast chains
5811 if (dev->netdev_ops->ndo_uninit)
5812 dev->netdev_ops->ndo_uninit(dev);
5814 if (!dev->rtnl_link_ops ||
5815 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5816 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5818 /* Notifier chain MUST detach us all upper devices. */
5819 WARN_ON(netdev_has_any_upper_dev(dev));
5821 /* Remove entries from kobject tree */
5822 netdev_unregister_kobject(dev);
5824 /* Remove XPS queueing entries */
5825 netif_reset_xps_queues_gt(dev, 0);
5831 list_for_each_entry(dev, head, unreg_list)
5835 static void rollback_registered(struct net_device *dev)
5839 list_add(&dev->unreg_list, &single);
5840 rollback_registered_many(&single);
5844 static netdev_features_t netdev_fix_features(struct net_device *dev,
5845 netdev_features_t features)
5847 /* Fix illegal checksum combinations */
5848 if ((features & NETIF_F_HW_CSUM) &&
5849 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5850 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5851 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5854 /* TSO requires that SG is present as well. */
5855 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5856 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5857 features &= ~NETIF_F_ALL_TSO;
5860 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5861 !(features & NETIF_F_IP_CSUM)) {
5862 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5863 features &= ~NETIF_F_TSO;
5864 features &= ~NETIF_F_TSO_ECN;
5867 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5868 !(features & NETIF_F_IPV6_CSUM)) {
5869 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5870 features &= ~NETIF_F_TSO6;
5873 /* TSO ECN requires that TSO is present as well. */
5874 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5875 features &= ~NETIF_F_TSO_ECN;
5877 /* Software GSO depends on SG. */
5878 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5879 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5880 features &= ~NETIF_F_GSO;
5883 /* UFO needs SG and checksumming */
5884 if (features & NETIF_F_UFO) {
5885 /* maybe split UFO into V4 and V6? */
5886 if (!((features & NETIF_F_GEN_CSUM) ||
5887 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5888 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5890 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5891 features &= ~NETIF_F_UFO;
5894 if (!(features & NETIF_F_SG)) {
5896 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5897 features &= ~NETIF_F_UFO;
5901 #ifdef CONFIG_NET_RX_BUSY_POLL
5902 if (dev->netdev_ops->ndo_busy_poll)
5903 features |= NETIF_F_BUSY_POLL;
5906 features &= ~NETIF_F_BUSY_POLL;
5911 int __netdev_update_features(struct net_device *dev)
5913 netdev_features_t features;
5918 features = netdev_get_wanted_features(dev);
5920 if (dev->netdev_ops->ndo_fix_features)
5921 features = dev->netdev_ops->ndo_fix_features(dev, features);
5923 /* driver might be less strict about feature dependencies */
5924 features = netdev_fix_features(dev, features);
5926 if (dev->features == features)
5929 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5930 &dev->features, &features);
5932 if (dev->netdev_ops->ndo_set_features)
5933 err = dev->netdev_ops->ndo_set_features(dev, features);
5935 if (unlikely(err < 0)) {
5937 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5938 err, &features, &dev->features);
5943 dev->features = features;
5949 * netdev_update_features - recalculate device features
5950 * @dev: the device to check
5952 * Recalculate dev->features set and send notifications if it
5953 * has changed. Should be called after driver or hardware dependent
5954 * conditions might have changed that influence the features.
5956 void netdev_update_features(struct net_device *dev)
5958 if (__netdev_update_features(dev))
5959 netdev_features_change(dev);
5961 EXPORT_SYMBOL(netdev_update_features);
5964 * netdev_change_features - recalculate device features
5965 * @dev: the device to check
5967 * Recalculate dev->features set and send notifications even
5968 * if they have not changed. Should be called instead of
5969 * netdev_update_features() if also dev->vlan_features might
5970 * have changed to allow the changes to be propagated to stacked
5973 void netdev_change_features(struct net_device *dev)
5975 __netdev_update_features(dev);
5976 netdev_features_change(dev);
5978 EXPORT_SYMBOL(netdev_change_features);
5981 * netif_stacked_transfer_operstate - transfer operstate
5982 * @rootdev: the root or lower level device to transfer state from
5983 * @dev: the device to transfer operstate to
5985 * Transfer operational state from root to device. This is normally
5986 * called when a stacking relationship exists between the root
5987 * device and the device(a leaf device).
5989 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5990 struct net_device *dev)
5992 if (rootdev->operstate == IF_OPER_DORMANT)
5993 netif_dormant_on(dev);
5995 netif_dormant_off(dev);
5997 if (netif_carrier_ok(rootdev)) {
5998 if (!netif_carrier_ok(dev))
5999 netif_carrier_on(dev);
6001 if (netif_carrier_ok(dev))
6002 netif_carrier_off(dev);
6005 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6008 static int netif_alloc_rx_queues(struct net_device *dev)
6010 unsigned int i, count = dev->num_rx_queues;
6011 struct netdev_rx_queue *rx;
6015 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
6021 for (i = 0; i < count; i++)
6027 static void netdev_init_one_queue(struct net_device *dev,
6028 struct netdev_queue *queue, void *_unused)
6030 /* Initialize queue lock */
6031 spin_lock_init(&queue->_xmit_lock);
6032 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6033 queue->xmit_lock_owner = -1;
6034 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6037 dql_init(&queue->dql, HZ);
6041 static void netif_free_tx_queues(struct net_device *dev)
6046 static int netif_alloc_netdev_queues(struct net_device *dev)
6048 unsigned int count = dev->num_tx_queues;
6049 struct netdev_queue *tx;
6050 size_t sz = count * sizeof(*tx);
6052 BUG_ON(count < 1 || count > 0xffff);
6054 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6062 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6063 spin_lock_init(&dev->tx_global_lock);
6069 * register_netdevice - register a network device
6070 * @dev: device to register
6072 * Take a completed network device structure and add it to the kernel
6073 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6074 * chain. 0 is returned on success. A negative errno code is returned
6075 * on a failure to set up the device, or if the name is a duplicate.
6077 * Callers must hold the rtnl semaphore. You may want
6078 * register_netdev() instead of this.
6081 * The locking appears insufficient to guarantee two parallel registers
6082 * will not get the same name.
6085 int register_netdevice(struct net_device *dev)
6088 struct net *net = dev_net(dev);
6090 BUG_ON(dev_boot_phase);
6095 /* When net_device's are persistent, this will be fatal. */
6096 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6099 spin_lock_init(&dev->addr_list_lock);
6100 netdev_set_addr_lockdep_class(dev);
6104 ret = dev_get_valid_name(net, dev, dev->name);
6108 /* Init, if this function is available */
6109 if (dev->netdev_ops->ndo_init) {
6110 ret = dev->netdev_ops->ndo_init(dev);
6118 if (((dev->hw_features | dev->features) &
6119 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6120 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6121 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6122 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6129 dev->ifindex = dev_new_index(net);
6130 else if (__dev_get_by_index(net, dev->ifindex))
6133 if (dev->iflink == -1)
6134 dev->iflink = dev->ifindex;
6136 /* Transfer changeable features to wanted_features and enable
6137 * software offloads (GSO and GRO).
6139 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6140 dev->features |= NETIF_F_SOFT_FEATURES;
6141 dev->wanted_features = dev->features & dev->hw_features;
6143 if (!(dev->flags & IFF_LOOPBACK)) {
6144 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6147 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6149 dev->vlan_features |= NETIF_F_HIGHDMA;
6151 /* Make NETIF_F_SG inheritable to tunnel devices.
6153 dev->hw_enc_features |= NETIF_F_SG;
6155 /* Make NETIF_F_SG inheritable to MPLS.
6157 dev->mpls_features |= NETIF_F_SG;
6159 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6160 ret = notifier_to_errno(ret);
6164 ret = netdev_register_kobject(dev);
6167 dev->reg_state = NETREG_REGISTERED;
6169 __netdev_update_features(dev);
6172 * Default initial state at registry is that the
6173 * device is present.
6176 set_bit(__LINK_STATE_PRESENT, &dev->state);
6178 linkwatch_init_dev(dev);
6180 dev_init_scheduler(dev);
6182 list_netdevice(dev);
6183 add_device_randomness(dev->dev_addr, dev->addr_len);
6185 /* If the device has permanent device address, driver should
6186 * set dev_addr and also addr_assign_type should be set to
6187 * NET_ADDR_PERM (default value).
6189 if (dev->addr_assign_type == NET_ADDR_PERM)
6190 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6192 /* Notify protocols, that a new device appeared. */
6193 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6194 ret = notifier_to_errno(ret);
6196 rollback_registered(dev);
6197 dev->reg_state = NETREG_UNREGISTERED;
6200 * Prevent userspace races by waiting until the network
6201 * device is fully setup before sending notifications.
6203 if (!dev->rtnl_link_ops ||
6204 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6205 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6211 if (dev->netdev_ops->ndo_uninit)
6212 dev->netdev_ops->ndo_uninit(dev);
6215 EXPORT_SYMBOL(register_netdevice);
6218 * init_dummy_netdev - init a dummy network device for NAPI
6219 * @dev: device to init
6221 * This takes a network device structure and initialize the minimum
6222 * amount of fields so it can be used to schedule NAPI polls without
6223 * registering a full blown interface. This is to be used by drivers
6224 * that need to tie several hardware interfaces to a single NAPI
6225 * poll scheduler due to HW limitations.
6227 int init_dummy_netdev(struct net_device *dev)
6229 /* Clear everything. Note we don't initialize spinlocks
6230 * are they aren't supposed to be taken by any of the
6231 * NAPI code and this dummy netdev is supposed to be
6232 * only ever used for NAPI polls
6234 memset(dev, 0, sizeof(struct net_device));
6236 /* make sure we BUG if trying to hit standard
6237 * register/unregister code path
6239 dev->reg_state = NETREG_DUMMY;
6241 /* NAPI wants this */
6242 INIT_LIST_HEAD(&dev->napi_list);
6244 /* a dummy interface is started by default */
6245 set_bit(__LINK_STATE_PRESENT, &dev->state);
6246 set_bit(__LINK_STATE_START, &dev->state);
6248 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6249 * because users of this 'device' dont need to change
6255 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6259 * register_netdev - register a network device
6260 * @dev: device to register
6262 * Take a completed network device structure and add it to the kernel
6263 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6264 * chain. 0 is returned on success. A negative errno code is returned
6265 * on a failure to set up the device, or if the name is a duplicate.
6267 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6268 * and expands the device name if you passed a format string to
6271 int register_netdev(struct net_device *dev)
6276 err = register_netdevice(dev);
6280 EXPORT_SYMBOL(register_netdev);
6282 int netdev_refcnt_read(const struct net_device *dev)
6286 for_each_possible_cpu(i)
6287 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6290 EXPORT_SYMBOL(netdev_refcnt_read);
6293 * netdev_wait_allrefs - wait until all references are gone.
6294 * @dev: target net_device
6296 * This is called when unregistering network devices.
6298 * Any protocol or device that holds a reference should register
6299 * for netdevice notification, and cleanup and put back the
6300 * reference if they receive an UNREGISTER event.
6301 * We can get stuck here if buggy protocols don't correctly
6304 static void netdev_wait_allrefs(struct net_device *dev)
6306 unsigned long rebroadcast_time, warning_time;
6309 linkwatch_forget_dev(dev);
6311 rebroadcast_time = warning_time = jiffies;
6312 refcnt = netdev_refcnt_read(dev);
6314 while (refcnt != 0) {
6315 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6318 /* Rebroadcast unregister notification */
6319 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6325 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6326 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6328 /* We must not have linkwatch events
6329 * pending on unregister. If this
6330 * happens, we simply run the queue
6331 * unscheduled, resulting in a noop
6334 linkwatch_run_queue();
6339 rebroadcast_time = jiffies;
6344 refcnt = netdev_refcnt_read(dev);
6346 if (time_after(jiffies, warning_time + 10 * HZ)) {
6347 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6349 warning_time = jiffies;
6358 * register_netdevice(x1);
6359 * register_netdevice(x2);
6361 * unregister_netdevice(y1);
6362 * unregister_netdevice(y2);
6368 * We are invoked by rtnl_unlock().
6369 * This allows us to deal with problems:
6370 * 1) We can delete sysfs objects which invoke hotplug
6371 * without deadlocking with linkwatch via keventd.
6372 * 2) Since we run with the RTNL semaphore not held, we can sleep
6373 * safely in order to wait for the netdev refcnt to drop to zero.
6375 * We must not return until all unregister events added during
6376 * the interval the lock was held have been completed.
6378 void netdev_run_todo(void)
6380 struct list_head list;
6382 /* Snapshot list, allow later requests */
6383 list_replace_init(&net_todo_list, &list);
6388 /* Wait for rcu callbacks to finish before next phase */
6389 if (!list_empty(&list))
6392 while (!list_empty(&list)) {
6393 struct net_device *dev
6394 = list_first_entry(&list, struct net_device, todo_list);
6395 list_del(&dev->todo_list);
6398 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6401 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6402 pr_err("network todo '%s' but state %d\n",
6403 dev->name, dev->reg_state);
6408 dev->reg_state = NETREG_UNREGISTERED;
6410 on_each_cpu(flush_backlog, dev, 1);
6412 netdev_wait_allrefs(dev);
6415 BUG_ON(netdev_refcnt_read(dev));
6416 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6417 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6418 WARN_ON(dev->dn_ptr);
6420 if (dev->destructor)
6421 dev->destructor(dev);
6423 /* Report a network device has been unregistered */
6425 dev_net(dev)->dev_unreg_count--;
6427 wake_up(&netdev_unregistering_wq);
6429 /* Free network device */
6430 kobject_put(&dev->dev.kobj);
6434 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6435 * fields in the same order, with only the type differing.
6437 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6438 const struct net_device_stats *netdev_stats)
6440 #if BITS_PER_LONG == 64
6441 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6442 memcpy(stats64, netdev_stats, sizeof(*stats64));
6444 size_t i, n = sizeof(*stats64) / sizeof(u64);
6445 const unsigned long *src = (const unsigned long *)netdev_stats;
6446 u64 *dst = (u64 *)stats64;
6448 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6449 sizeof(*stats64) / sizeof(u64));
6450 for (i = 0; i < n; i++)
6454 EXPORT_SYMBOL(netdev_stats_to_stats64);
6457 * dev_get_stats - get network device statistics
6458 * @dev: device to get statistics from
6459 * @storage: place to store stats
6461 * Get network statistics from device. Return @storage.
6462 * The device driver may provide its own method by setting
6463 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6464 * otherwise the internal statistics structure is used.
6466 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6467 struct rtnl_link_stats64 *storage)
6469 const struct net_device_ops *ops = dev->netdev_ops;
6471 if (ops->ndo_get_stats64) {
6472 memset(storage, 0, sizeof(*storage));
6473 ops->ndo_get_stats64(dev, storage);
6474 } else if (ops->ndo_get_stats) {
6475 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6477 netdev_stats_to_stats64(storage, &dev->stats);
6479 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6480 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6483 EXPORT_SYMBOL(dev_get_stats);
6485 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6487 struct netdev_queue *queue = dev_ingress_queue(dev);
6489 #ifdef CONFIG_NET_CLS_ACT
6492 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6495 netdev_init_one_queue(dev, queue, NULL);
6496 queue->qdisc = &noop_qdisc;
6497 queue->qdisc_sleeping = &noop_qdisc;
6498 rcu_assign_pointer(dev->ingress_queue, queue);
6503 static const struct ethtool_ops default_ethtool_ops;
6505 void netdev_set_default_ethtool_ops(struct net_device *dev,
6506 const struct ethtool_ops *ops)
6508 if (dev->ethtool_ops == &default_ethtool_ops)
6509 dev->ethtool_ops = ops;
6511 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6513 void netdev_freemem(struct net_device *dev)
6515 char *addr = (char *)dev - dev->padded;
6521 * alloc_netdev_mqs - allocate network device
6522 * @sizeof_priv: size of private data to allocate space for
6523 * @name: device name format string
6524 * @name_assign_type: origin of device name
6525 * @setup: callback to initialize device
6526 * @txqs: the number of TX subqueues to allocate
6527 * @rxqs: the number of RX subqueues to allocate
6529 * Allocates a struct net_device with private data area for driver use
6530 * and performs basic initialization. Also allocates subqueue structs
6531 * for each queue on the device.
6533 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6534 unsigned char name_assign_type,
6535 void (*setup)(struct net_device *),
6536 unsigned int txqs, unsigned int rxqs)
6538 struct net_device *dev;
6540 struct net_device *p;
6542 BUG_ON(strlen(name) >= sizeof(dev->name));
6545 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6551 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6556 alloc_size = sizeof(struct net_device);
6558 /* ensure 32-byte alignment of private area */
6559 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6560 alloc_size += sizeof_priv;
6562 /* ensure 32-byte alignment of whole construct */
6563 alloc_size += NETDEV_ALIGN - 1;
6565 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6567 p = vzalloc(alloc_size);
6571 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6572 dev->padded = (char *)dev - (char *)p;
6574 dev->pcpu_refcnt = alloc_percpu(int);
6575 if (!dev->pcpu_refcnt)
6578 if (dev_addr_init(dev))
6584 dev_net_set(dev, &init_net);
6586 dev->gso_max_size = GSO_MAX_SIZE;
6587 dev->gso_max_segs = GSO_MAX_SEGS;
6589 INIT_LIST_HEAD(&dev->napi_list);
6590 INIT_LIST_HEAD(&dev->unreg_list);
6591 INIT_LIST_HEAD(&dev->close_list);
6592 INIT_LIST_HEAD(&dev->link_watch_list);
6593 INIT_LIST_HEAD(&dev->adj_list.upper);
6594 INIT_LIST_HEAD(&dev->adj_list.lower);
6595 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6596 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6597 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6600 dev->num_tx_queues = txqs;
6601 dev->real_num_tx_queues = txqs;
6602 if (netif_alloc_netdev_queues(dev))
6606 dev->num_rx_queues = rxqs;
6607 dev->real_num_rx_queues = rxqs;
6608 if (netif_alloc_rx_queues(dev))
6612 strcpy(dev->name, name);
6613 dev->name_assign_type = name_assign_type;
6614 dev->group = INIT_NETDEV_GROUP;
6615 if (!dev->ethtool_ops)
6616 dev->ethtool_ops = &default_ethtool_ops;
6624 free_percpu(dev->pcpu_refcnt);
6626 netdev_freemem(dev);
6629 EXPORT_SYMBOL(alloc_netdev_mqs);
6632 * free_netdev - free network device
6635 * This function does the last stage of destroying an allocated device
6636 * interface. The reference to the device object is released.
6637 * If this is the last reference then it will be freed.
6639 void free_netdev(struct net_device *dev)
6641 struct napi_struct *p, *n;
6643 release_net(dev_net(dev));
6645 netif_free_tx_queues(dev);
6650 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6652 /* Flush device addresses */
6653 dev_addr_flush(dev);
6655 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6658 free_percpu(dev->pcpu_refcnt);
6659 dev->pcpu_refcnt = NULL;
6661 /* Compatibility with error handling in drivers */
6662 if (dev->reg_state == NETREG_UNINITIALIZED) {
6663 netdev_freemem(dev);
6667 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6668 dev->reg_state = NETREG_RELEASED;
6670 /* will free via device release */
6671 put_device(&dev->dev);
6673 EXPORT_SYMBOL(free_netdev);
6676 * synchronize_net - Synchronize with packet receive processing
6678 * Wait for packets currently being received to be done.
6679 * Does not block later packets from starting.
6681 void synchronize_net(void)
6684 if (rtnl_is_locked())
6685 synchronize_rcu_expedited();
6689 EXPORT_SYMBOL(synchronize_net);
6692 * unregister_netdevice_queue - remove device from the kernel
6696 * This function shuts down a device interface and removes it
6697 * from the kernel tables.
6698 * If head not NULL, device is queued to be unregistered later.
6700 * Callers must hold the rtnl semaphore. You may want
6701 * unregister_netdev() instead of this.
6704 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6709 list_move_tail(&dev->unreg_list, head);
6711 rollback_registered(dev);
6712 /* Finish processing unregister after unlock */
6716 EXPORT_SYMBOL(unregister_netdevice_queue);
6719 * unregister_netdevice_many - unregister many devices
6720 * @head: list of devices
6722 * Note: As most callers use a stack allocated list_head,
6723 * we force a list_del() to make sure stack wont be corrupted later.
6725 void unregister_netdevice_many(struct list_head *head)
6727 struct net_device *dev;
6729 if (!list_empty(head)) {
6730 rollback_registered_many(head);
6731 list_for_each_entry(dev, head, unreg_list)
6736 EXPORT_SYMBOL(unregister_netdevice_many);
6739 * unregister_netdev - remove device from the kernel
6742 * This function shuts down a device interface and removes it
6743 * from the kernel tables.
6745 * This is just a wrapper for unregister_netdevice that takes
6746 * the rtnl semaphore. In general you want to use this and not
6747 * unregister_netdevice.
6749 void unregister_netdev(struct net_device *dev)
6752 unregister_netdevice(dev);
6755 EXPORT_SYMBOL(unregister_netdev);
6758 * dev_change_net_namespace - move device to different nethost namespace
6760 * @net: network namespace
6761 * @pat: If not NULL name pattern to try if the current device name
6762 * is already taken in the destination network namespace.
6764 * This function shuts down a device interface and moves it
6765 * to a new network namespace. On success 0 is returned, on
6766 * a failure a netagive errno code is returned.
6768 * Callers must hold the rtnl semaphore.
6771 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6777 /* Don't allow namespace local devices to be moved. */
6779 if (dev->features & NETIF_F_NETNS_LOCAL)
6782 /* Ensure the device has been registrered */
6783 if (dev->reg_state != NETREG_REGISTERED)
6786 /* Get out if there is nothing todo */
6788 if (net_eq(dev_net(dev), net))
6791 /* Pick the destination device name, and ensure
6792 * we can use it in the destination network namespace.
6795 if (__dev_get_by_name(net, dev->name)) {
6796 /* We get here if we can't use the current device name */
6799 if (dev_get_valid_name(net, dev, pat) < 0)
6804 * And now a mini version of register_netdevice unregister_netdevice.
6807 /* If device is running close it first. */
6810 /* And unlink it from device chain */
6812 unlist_netdevice(dev);
6816 /* Shutdown queueing discipline. */
6819 /* Notify protocols, that we are about to destroy
6820 this device. They should clean all the things.
6822 Note that dev->reg_state stays at NETREG_REGISTERED.
6823 This is wanted because this way 8021q and macvlan know
6824 the device is just moving and can keep their slaves up.
6826 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6828 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6829 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6832 * Flush the unicast and multicast chains
6837 /* Send a netdev-removed uevent to the old namespace */
6838 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6839 netdev_adjacent_del_links(dev);
6841 /* Actually switch the network namespace */
6842 dev_net_set(dev, net);
6844 /* If there is an ifindex conflict assign a new one */
6845 if (__dev_get_by_index(net, dev->ifindex)) {
6846 int iflink = (dev->iflink == dev->ifindex);
6847 dev->ifindex = dev_new_index(net);
6849 dev->iflink = dev->ifindex;
6852 /* Send a netdev-add uevent to the new namespace */
6853 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6854 netdev_adjacent_add_links(dev);
6856 /* Fixup kobjects */
6857 err = device_rename(&dev->dev, dev->name);
6860 /* Add the device back in the hashes */
6861 list_netdevice(dev);
6863 /* Notify protocols, that a new device appeared. */
6864 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6867 * Prevent userspace races by waiting until the network
6868 * device is fully setup before sending notifications.
6870 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6877 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6879 static int dev_cpu_callback(struct notifier_block *nfb,
6880 unsigned long action,
6883 struct sk_buff **list_skb;
6884 struct sk_buff *skb;
6885 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6886 struct softnet_data *sd, *oldsd;
6888 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6891 local_irq_disable();
6892 cpu = smp_processor_id();
6893 sd = &per_cpu(softnet_data, cpu);
6894 oldsd = &per_cpu(softnet_data, oldcpu);
6896 /* Find end of our completion_queue. */
6897 list_skb = &sd->completion_queue;
6899 list_skb = &(*list_skb)->next;
6900 /* Append completion queue from offline CPU. */
6901 *list_skb = oldsd->completion_queue;
6902 oldsd->completion_queue = NULL;
6904 /* Append output queue from offline CPU. */
6905 if (oldsd->output_queue) {
6906 *sd->output_queue_tailp = oldsd->output_queue;
6907 sd->output_queue_tailp = oldsd->output_queue_tailp;
6908 oldsd->output_queue = NULL;
6909 oldsd->output_queue_tailp = &oldsd->output_queue;
6911 /* Append NAPI poll list from offline CPU. */
6912 if (!list_empty(&oldsd->poll_list)) {
6913 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6914 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6917 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6920 /* Process offline CPU's input_pkt_queue */
6921 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6922 netif_rx_internal(skb);
6923 input_queue_head_incr(oldsd);
6925 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6926 netif_rx_internal(skb);
6927 input_queue_head_incr(oldsd);
6935 * netdev_increment_features - increment feature set by one
6936 * @all: current feature set
6937 * @one: new feature set
6938 * @mask: mask feature set
6940 * Computes a new feature set after adding a device with feature set
6941 * @one to the master device with current feature set @all. Will not
6942 * enable anything that is off in @mask. Returns the new feature set.
6944 netdev_features_t netdev_increment_features(netdev_features_t all,
6945 netdev_features_t one, netdev_features_t mask)
6947 if (mask & NETIF_F_GEN_CSUM)
6948 mask |= NETIF_F_ALL_CSUM;
6949 mask |= NETIF_F_VLAN_CHALLENGED;
6951 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6952 all &= one | ~NETIF_F_ALL_FOR_ALL;
6954 /* If one device supports hw checksumming, set for all. */
6955 if (all & NETIF_F_GEN_CSUM)
6956 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6960 EXPORT_SYMBOL(netdev_increment_features);
6962 static struct hlist_head * __net_init netdev_create_hash(void)
6965 struct hlist_head *hash;
6967 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6969 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6970 INIT_HLIST_HEAD(&hash[i]);
6975 /* Initialize per network namespace state */
6976 static int __net_init netdev_init(struct net *net)
6978 if (net != &init_net)
6979 INIT_LIST_HEAD(&net->dev_base_head);
6981 net->dev_name_head = netdev_create_hash();
6982 if (net->dev_name_head == NULL)
6985 net->dev_index_head = netdev_create_hash();
6986 if (net->dev_index_head == NULL)
6992 kfree(net->dev_name_head);
6998 * netdev_drivername - network driver for the device
6999 * @dev: network device
7001 * Determine network driver for device.
7003 const char *netdev_drivername(const struct net_device *dev)
7005 const struct device_driver *driver;
7006 const struct device *parent;
7007 const char *empty = "";
7009 parent = dev->dev.parent;
7013 driver = parent->driver;
7014 if (driver && driver->name)
7015 return driver->name;
7019 static int __netdev_printk(const char *level, const struct net_device *dev,
7020 struct va_format *vaf)
7024 if (dev && dev->dev.parent) {
7025 r = dev_printk_emit(level[1] - '0',
7028 dev_driver_string(dev->dev.parent),
7029 dev_name(dev->dev.parent),
7030 netdev_name(dev), netdev_reg_state(dev),
7033 r = printk("%s%s%s: %pV", level, netdev_name(dev),
7034 netdev_reg_state(dev), vaf);
7036 r = printk("%s(NULL net_device): %pV", level, vaf);
7042 int netdev_printk(const char *level, const struct net_device *dev,
7043 const char *format, ...)
7045 struct va_format vaf;
7049 va_start(args, format);
7054 r = __netdev_printk(level, dev, &vaf);
7060 EXPORT_SYMBOL(netdev_printk);
7062 #define define_netdev_printk_level(func, level) \
7063 int func(const struct net_device *dev, const char *fmt, ...) \
7066 struct va_format vaf; \
7069 va_start(args, fmt); \
7074 r = __netdev_printk(level, dev, &vaf); \
7080 EXPORT_SYMBOL(func);
7082 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7083 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7084 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7085 define_netdev_printk_level(netdev_err, KERN_ERR);
7086 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7087 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7088 define_netdev_printk_level(netdev_info, KERN_INFO);
7090 static void __net_exit netdev_exit(struct net *net)
7092 kfree(net->dev_name_head);
7093 kfree(net->dev_index_head);
7096 static struct pernet_operations __net_initdata netdev_net_ops = {
7097 .init = netdev_init,
7098 .exit = netdev_exit,
7101 static void __net_exit default_device_exit(struct net *net)
7103 struct net_device *dev, *aux;
7105 * Push all migratable network devices back to the
7106 * initial network namespace
7109 for_each_netdev_safe(net, dev, aux) {
7111 char fb_name[IFNAMSIZ];
7113 /* Ignore unmoveable devices (i.e. loopback) */
7114 if (dev->features & NETIF_F_NETNS_LOCAL)
7117 /* Leave virtual devices for the generic cleanup */
7118 if (dev->rtnl_link_ops)
7121 /* Push remaining network devices to init_net */
7122 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7123 err = dev_change_net_namespace(dev, &init_net, fb_name);
7125 pr_emerg("%s: failed to move %s to init_net: %d\n",
7126 __func__, dev->name, err);
7133 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7135 /* Return with the rtnl_lock held when there are no network
7136 * devices unregistering in any network namespace in net_list.
7143 prepare_to_wait(&netdev_unregistering_wq, &wait,
7144 TASK_UNINTERRUPTIBLE);
7145 unregistering = false;
7147 list_for_each_entry(net, net_list, exit_list) {
7148 if (net->dev_unreg_count > 0) {
7149 unregistering = true;
7158 finish_wait(&netdev_unregistering_wq, &wait);
7161 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7163 /* At exit all network devices most be removed from a network
7164 * namespace. Do this in the reverse order of registration.
7165 * Do this across as many network namespaces as possible to
7166 * improve batching efficiency.
7168 struct net_device *dev;
7170 LIST_HEAD(dev_kill_list);
7172 /* To prevent network device cleanup code from dereferencing
7173 * loopback devices or network devices that have been freed
7174 * wait here for all pending unregistrations to complete,
7175 * before unregistring the loopback device and allowing the
7176 * network namespace be freed.
7178 * The netdev todo list containing all network devices
7179 * unregistrations that happen in default_device_exit_batch
7180 * will run in the rtnl_unlock() at the end of
7181 * default_device_exit_batch.
7183 rtnl_lock_unregistering(net_list);
7184 list_for_each_entry(net, net_list, exit_list) {
7185 for_each_netdev_reverse(net, dev) {
7186 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7187 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7189 unregister_netdevice_queue(dev, &dev_kill_list);
7192 unregister_netdevice_many(&dev_kill_list);
7196 static struct pernet_operations __net_initdata default_device_ops = {
7197 .exit = default_device_exit,
7198 .exit_batch = default_device_exit_batch,
7202 * Initialize the DEV module. At boot time this walks the device list and
7203 * unhooks any devices that fail to initialise (normally hardware not
7204 * present) and leaves us with a valid list of present and active devices.
7209 * This is called single threaded during boot, so no need
7210 * to take the rtnl semaphore.
7212 static int __init net_dev_init(void)
7214 int i, rc = -ENOMEM;
7216 BUG_ON(!dev_boot_phase);
7218 if (dev_proc_init())
7221 if (netdev_kobject_init())
7224 INIT_LIST_HEAD(&ptype_all);
7225 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7226 INIT_LIST_HEAD(&ptype_base[i]);
7228 INIT_LIST_HEAD(&offload_base);
7230 if (register_pernet_subsys(&netdev_net_ops))
7234 * Initialise the packet receive queues.
7237 for_each_possible_cpu(i) {
7238 struct softnet_data *sd = &per_cpu(softnet_data, i);
7240 skb_queue_head_init(&sd->input_pkt_queue);
7241 skb_queue_head_init(&sd->process_queue);
7242 INIT_LIST_HEAD(&sd->poll_list);
7243 sd->output_queue_tailp = &sd->output_queue;
7245 sd->csd.func = rps_trigger_softirq;
7250 sd->backlog.poll = process_backlog;
7251 sd->backlog.weight = weight_p;
7256 /* The loopback device is special if any other network devices
7257 * is present in a network namespace the loopback device must
7258 * be present. Since we now dynamically allocate and free the
7259 * loopback device ensure this invariant is maintained by
7260 * keeping the loopback device as the first device on the
7261 * list of network devices. Ensuring the loopback devices
7262 * is the first device that appears and the last network device
7265 if (register_pernet_device(&loopback_net_ops))
7268 if (register_pernet_device(&default_device_ops))
7271 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7272 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7274 hotcpu_notifier(dev_cpu_callback, 0);
7281 subsys_initcall(net_dev_init);