2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
121 #include <linux/ipv6.h>
122 #include <linux/in.h>
123 #include <linux/jhash.h>
124 #include <linux/random.h>
125 #include <trace/events/napi.h>
126 #include <trace/events/net.h>
127 #include <trace/events/skb.h>
128 #include <linux/pci.h>
129 #include <linux/inetdevice.h>
130 #include <linux/cpu_rmap.h>
131 #include <linux/static_key.h>
132 #include <linux/hashtable.h>
133 #include <linux/vmalloc.h>
134 #include <linux/if_macvlan.h>
135 #include <linux/errqueue.h>
137 #include "net-sysfs.h"
139 /* Instead of increasing this, you should create a hash table. */
140 #define MAX_GRO_SKBS 8
142 /* This should be increased if a protocol with a bigger head is added. */
143 #define GRO_MAX_HEAD (MAX_HEADER + 128)
145 static DEFINE_SPINLOCK(ptype_lock);
146 static DEFINE_SPINLOCK(offload_lock);
147 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
148 struct list_head ptype_all __read_mostly; /* Taps */
149 static struct list_head offload_base __read_mostly;
151 static int netif_rx_internal(struct sk_buff *skb);
152 static int call_netdevice_notifiers_info(unsigned long val,
153 struct net_device *dev,
154 struct netdev_notifier_info *info);
157 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
160 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
162 * Writers must hold the rtnl semaphore while they loop through the
163 * dev_base_head list, and hold dev_base_lock for writing when they do the
164 * actual updates. This allows pure readers to access the list even
165 * while a writer is preparing to update it.
167 * To put it another way, dev_base_lock is held for writing only to
168 * protect against pure readers; the rtnl semaphore provides the
169 * protection against other writers.
171 * See, for example usages, register_netdevice() and
172 * unregister_netdevice(), which must be called with the rtnl
175 DEFINE_RWLOCK(dev_base_lock);
176 EXPORT_SYMBOL(dev_base_lock);
178 /* protects napi_hash addition/deletion and napi_gen_id */
179 static DEFINE_SPINLOCK(napi_hash_lock);
181 static unsigned int napi_gen_id;
182 static DEFINE_HASHTABLE(napi_hash, 8);
184 static seqcount_t devnet_rename_seq;
186 static inline void dev_base_seq_inc(struct net *net)
188 while (++net->dev_base_seq == 0);
191 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
193 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
195 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
198 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
200 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
203 static inline void rps_lock(struct softnet_data *sd)
206 spin_lock(&sd->input_pkt_queue.lock);
210 static inline void rps_unlock(struct softnet_data *sd)
213 spin_unlock(&sd->input_pkt_queue.lock);
217 /* Device list insertion */
218 static void list_netdevice(struct net_device *dev)
220 struct net *net = dev_net(dev);
224 write_lock_bh(&dev_base_lock);
225 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
226 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
227 hlist_add_head_rcu(&dev->index_hlist,
228 dev_index_hash(net, dev->ifindex));
229 write_unlock_bh(&dev_base_lock);
231 dev_base_seq_inc(net);
234 /* Device list removal
235 * caller must respect a RCU grace period before freeing/reusing dev
237 static void unlist_netdevice(struct net_device *dev)
241 /* Unlink dev from the device chain */
242 write_lock_bh(&dev_base_lock);
243 list_del_rcu(&dev->dev_list);
244 hlist_del_rcu(&dev->name_hlist);
245 hlist_del_rcu(&dev->index_hlist);
246 write_unlock_bh(&dev_base_lock);
248 dev_base_seq_inc(dev_net(dev));
255 static RAW_NOTIFIER_HEAD(netdev_chain);
258 * Device drivers call our routines to queue packets here. We empty the
259 * queue in the local softnet handler.
262 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
263 EXPORT_PER_CPU_SYMBOL(softnet_data);
265 #ifdef CONFIG_LOCKDEP
267 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
268 * according to dev->type
270 static const unsigned short netdev_lock_type[] =
271 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
272 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
273 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
274 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
275 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
276 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
277 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
278 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
279 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
280 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
281 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
282 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
283 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
284 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
285 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
287 static const char *const netdev_lock_name[] =
288 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
289 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
290 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
291 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
292 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
293 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
294 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
295 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
296 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
297 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
298 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
299 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
300 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
301 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
302 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
304 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
305 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
307 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
311 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
312 if (netdev_lock_type[i] == dev_type)
314 /* the last key is used by default */
315 return ARRAY_SIZE(netdev_lock_type) - 1;
318 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
319 unsigned short dev_type)
323 i = netdev_lock_pos(dev_type);
324 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
325 netdev_lock_name[i]);
328 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
332 i = netdev_lock_pos(dev->type);
333 lockdep_set_class_and_name(&dev->addr_list_lock,
334 &netdev_addr_lock_key[i],
335 netdev_lock_name[i]);
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
339 unsigned short dev_type)
342 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
347 /*******************************************************************************
349 Protocol management and registration routines
351 *******************************************************************************/
354 * Add a protocol ID to the list. Now that the input handler is
355 * smarter we can dispense with all the messy stuff that used to be
358 * BEWARE!!! Protocol handlers, mangling input packets,
359 * MUST BE last in hash buckets and checking protocol handlers
360 * MUST start from promiscuous ptype_all chain in net_bh.
361 * It is true now, do not change it.
362 * Explanation follows: if protocol handler, mangling packet, will
363 * be the first on list, it is not able to sense, that packet
364 * is cloned and should be copied-on-write, so that it will
365 * change it and subsequent readers will get broken packet.
369 static inline struct list_head *ptype_head(const struct packet_type *pt)
371 if (pt->type == htons(ETH_P_ALL))
374 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
378 * dev_add_pack - add packet handler
379 * @pt: packet type declaration
381 * Add a protocol handler to the networking stack. The passed &packet_type
382 * is linked into kernel lists and may not be freed until it has been
383 * removed from the kernel lists.
385 * This call does not sleep therefore it can not
386 * guarantee all CPU's that are in middle of receiving packets
387 * will see the new packet type (until the next received packet).
390 void dev_add_pack(struct packet_type *pt)
392 struct list_head *head = ptype_head(pt);
394 spin_lock(&ptype_lock);
395 list_add_rcu(&pt->list, head);
396 spin_unlock(&ptype_lock);
398 EXPORT_SYMBOL(dev_add_pack);
401 * __dev_remove_pack - remove packet handler
402 * @pt: packet type declaration
404 * Remove a protocol handler that was previously added to the kernel
405 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
406 * from the kernel lists and can be freed or reused once this function
409 * The packet type might still be in use by receivers
410 * and must not be freed until after all the CPU's have gone
411 * through a quiescent state.
413 void __dev_remove_pack(struct packet_type *pt)
415 struct list_head *head = ptype_head(pt);
416 struct packet_type *pt1;
418 spin_lock(&ptype_lock);
420 list_for_each_entry(pt1, head, list) {
422 list_del_rcu(&pt->list);
427 pr_warn("dev_remove_pack: %p not found\n", pt);
429 spin_unlock(&ptype_lock);
431 EXPORT_SYMBOL(__dev_remove_pack);
434 * dev_remove_pack - remove packet handler
435 * @pt: packet type declaration
437 * Remove a protocol handler that was previously added to the kernel
438 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
439 * from the kernel lists and can be freed or reused once this function
442 * This call sleeps to guarantee that no CPU is looking at the packet
445 void dev_remove_pack(struct packet_type *pt)
447 __dev_remove_pack(pt);
451 EXPORT_SYMBOL(dev_remove_pack);
455 * dev_add_offload - register offload handlers
456 * @po: protocol offload declaration
458 * Add protocol offload handlers to the networking stack. The passed
459 * &proto_offload is linked into kernel lists and may not be freed until
460 * it has been removed from the kernel lists.
462 * This call does not sleep therefore it can not
463 * guarantee all CPU's that are in middle of receiving packets
464 * will see the new offload handlers (until the next received packet).
466 void dev_add_offload(struct packet_offload *po)
468 struct list_head *head = &offload_base;
470 spin_lock(&offload_lock);
471 list_add_rcu(&po->list, head);
472 spin_unlock(&offload_lock);
474 EXPORT_SYMBOL(dev_add_offload);
477 * __dev_remove_offload - remove offload handler
478 * @po: packet offload declaration
480 * Remove a protocol offload handler that was previously added to the
481 * kernel offload handlers by dev_add_offload(). The passed &offload_type
482 * is removed from the kernel lists and can be freed or reused once this
485 * The packet type might still be in use by receivers
486 * and must not be freed until after all the CPU's have gone
487 * through a quiescent state.
489 static void __dev_remove_offload(struct packet_offload *po)
491 struct list_head *head = &offload_base;
492 struct packet_offload *po1;
494 spin_lock(&offload_lock);
496 list_for_each_entry(po1, head, list) {
498 list_del_rcu(&po->list);
503 pr_warn("dev_remove_offload: %p not found\n", po);
505 spin_unlock(&offload_lock);
509 * dev_remove_offload - remove packet offload handler
510 * @po: packet offload declaration
512 * Remove a packet offload handler that was previously added to the kernel
513 * offload handlers by dev_add_offload(). The passed &offload_type is
514 * removed from the kernel lists and can be freed or reused once this
517 * This call sleeps to guarantee that no CPU is looking at the packet
520 void dev_remove_offload(struct packet_offload *po)
522 __dev_remove_offload(po);
526 EXPORT_SYMBOL(dev_remove_offload);
528 /******************************************************************************
530 Device Boot-time Settings Routines
532 *******************************************************************************/
534 /* Boot time configuration table */
535 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
538 * netdev_boot_setup_add - add new setup entry
539 * @name: name of the device
540 * @map: configured settings for the device
542 * Adds new setup entry to the dev_boot_setup list. The function
543 * returns 0 on error and 1 on success. This is a generic routine to
546 static int netdev_boot_setup_add(char *name, struct ifmap *map)
548 struct netdev_boot_setup *s;
552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
553 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
554 memset(s[i].name, 0, sizeof(s[i].name));
555 strlcpy(s[i].name, name, IFNAMSIZ);
556 memcpy(&s[i].map, map, sizeof(s[i].map));
561 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
565 * netdev_boot_setup_check - check boot time settings
566 * @dev: the netdevice
568 * Check boot time settings for the device.
569 * The found settings are set for the device to be used
570 * later in the device probing.
571 * Returns 0 if no settings found, 1 if they are.
573 int netdev_boot_setup_check(struct net_device *dev)
575 struct netdev_boot_setup *s = dev_boot_setup;
578 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
579 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
580 !strcmp(dev->name, s[i].name)) {
581 dev->irq = s[i].map.irq;
582 dev->base_addr = s[i].map.base_addr;
583 dev->mem_start = s[i].map.mem_start;
584 dev->mem_end = s[i].map.mem_end;
590 EXPORT_SYMBOL(netdev_boot_setup_check);
594 * netdev_boot_base - get address from boot time settings
595 * @prefix: prefix for network device
596 * @unit: id for network device
598 * Check boot time settings for the base address of device.
599 * The found settings are set for the device to be used
600 * later in the device probing.
601 * Returns 0 if no settings found.
603 unsigned long netdev_boot_base(const char *prefix, int unit)
605 const struct netdev_boot_setup *s = dev_boot_setup;
609 sprintf(name, "%s%d", prefix, unit);
612 * If device already registered then return base of 1
613 * to indicate not to probe for this interface
615 if (__dev_get_by_name(&init_net, name))
618 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
619 if (!strcmp(name, s[i].name))
620 return s[i].map.base_addr;
625 * Saves at boot time configured settings for any netdevice.
627 int __init netdev_boot_setup(char *str)
632 str = get_options(str, ARRAY_SIZE(ints), ints);
637 memset(&map, 0, sizeof(map));
641 map.base_addr = ints[2];
643 map.mem_start = ints[3];
645 map.mem_end = ints[4];
647 /* Add new entry to the list */
648 return netdev_boot_setup_add(str, &map);
651 __setup("netdev=", netdev_boot_setup);
653 /*******************************************************************************
655 Device Interface Subroutines
657 *******************************************************************************/
660 * __dev_get_by_name - find a device by its name
661 * @net: the applicable net namespace
662 * @name: name to find
664 * Find an interface by name. Must be called under RTNL semaphore
665 * or @dev_base_lock. If the name is found a pointer to the device
666 * is returned. If the name is not found then %NULL is returned. The
667 * reference counters are not incremented so the caller must be
668 * careful with locks.
671 struct net_device *__dev_get_by_name(struct net *net, const char *name)
673 struct net_device *dev;
674 struct hlist_head *head = dev_name_hash(net, name);
676 hlist_for_each_entry(dev, head, name_hlist)
677 if (!strncmp(dev->name, name, IFNAMSIZ))
682 EXPORT_SYMBOL(__dev_get_by_name);
685 * dev_get_by_name_rcu - find a device by its name
686 * @net: the applicable net namespace
687 * @name: name to find
689 * Find an interface by name.
690 * If the name is found a pointer to the device is returned.
691 * If the name is not found then %NULL is returned.
692 * The reference counters are not incremented so the caller must be
693 * careful with locks. The caller must hold RCU lock.
696 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
698 struct net_device *dev;
699 struct hlist_head *head = dev_name_hash(net, name);
701 hlist_for_each_entry_rcu(dev, head, name_hlist)
702 if (!strncmp(dev->name, name, IFNAMSIZ))
707 EXPORT_SYMBOL(dev_get_by_name_rcu);
710 * dev_get_by_name - find a device by its name
711 * @net: the applicable net namespace
712 * @name: name to find
714 * Find an interface by name. This can be called from any
715 * context and does its own locking. The returned handle has
716 * the usage count incremented and the caller must use dev_put() to
717 * release it when it is no longer needed. %NULL is returned if no
718 * matching device is found.
721 struct net_device *dev_get_by_name(struct net *net, const char *name)
723 struct net_device *dev;
726 dev = dev_get_by_name_rcu(net, name);
732 EXPORT_SYMBOL(dev_get_by_name);
735 * __dev_get_by_index - find a device by its ifindex
736 * @net: the applicable net namespace
737 * @ifindex: index of device
739 * Search for an interface by index. Returns %NULL if the device
740 * is not found or a pointer to the device. The device has not
741 * had its reference counter increased so the caller must be careful
742 * about locking. The caller must hold either the RTNL semaphore
746 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
748 struct net_device *dev;
749 struct hlist_head *head = dev_index_hash(net, ifindex);
751 hlist_for_each_entry(dev, head, index_hlist)
752 if (dev->ifindex == ifindex)
757 EXPORT_SYMBOL(__dev_get_by_index);
760 * dev_get_by_index_rcu - find a device by its ifindex
761 * @net: the applicable net namespace
762 * @ifindex: index of device
764 * Search for an interface by index. Returns %NULL if the device
765 * is not found or a pointer to the device. The device has not
766 * had its reference counter increased so the caller must be careful
767 * about locking. The caller must hold RCU lock.
770 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
772 struct net_device *dev;
773 struct hlist_head *head = dev_index_hash(net, ifindex);
775 hlist_for_each_entry_rcu(dev, head, index_hlist)
776 if (dev->ifindex == ifindex)
781 EXPORT_SYMBOL(dev_get_by_index_rcu);
785 * dev_get_by_index - find a device by its ifindex
786 * @net: the applicable net namespace
787 * @ifindex: index of device
789 * Search for an interface by index. Returns NULL if the device
790 * is not found or a pointer to the device. The device returned has
791 * had a reference added and the pointer is safe until the user calls
792 * dev_put to indicate they have finished with it.
795 struct net_device *dev_get_by_index(struct net *net, int ifindex)
797 struct net_device *dev;
800 dev = dev_get_by_index_rcu(net, ifindex);
806 EXPORT_SYMBOL(dev_get_by_index);
809 * netdev_get_name - get a netdevice name, knowing its ifindex.
810 * @net: network namespace
811 * @name: a pointer to the buffer where the name will be stored.
812 * @ifindex: the ifindex of the interface to get the name from.
814 * The use of raw_seqcount_begin() and cond_resched() before
815 * retrying is required as we want to give the writers a chance
816 * to complete when CONFIG_PREEMPT is not set.
818 int netdev_get_name(struct net *net, char *name, int ifindex)
820 struct net_device *dev;
824 seq = raw_seqcount_begin(&devnet_rename_seq);
826 dev = dev_get_by_index_rcu(net, ifindex);
832 strcpy(name, dev->name);
834 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
843 * dev_getbyhwaddr_rcu - find a device by its hardware address
844 * @net: the applicable net namespace
845 * @type: media type of device
846 * @ha: hardware address
848 * Search for an interface by MAC address. Returns NULL if the device
849 * is not found or a pointer to the device.
850 * The caller must hold RCU or RTNL.
851 * The returned device has not had its ref count increased
852 * and the caller must therefore be careful about locking
856 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
859 struct net_device *dev;
861 for_each_netdev_rcu(net, dev)
862 if (dev->type == type &&
863 !memcmp(dev->dev_addr, ha, dev->addr_len))
868 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
870 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
872 struct net_device *dev;
875 for_each_netdev(net, dev)
876 if (dev->type == type)
881 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
883 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
885 struct net_device *dev, *ret = NULL;
888 for_each_netdev_rcu(net, dev)
889 if (dev->type == type) {
897 EXPORT_SYMBOL(dev_getfirstbyhwtype);
900 * __dev_get_by_flags - find any device with given flags
901 * @net: the applicable net namespace
902 * @if_flags: IFF_* values
903 * @mask: bitmask of bits in if_flags to check
905 * Search for any interface with the given flags. Returns NULL if a device
906 * is not found or a pointer to the device. Must be called inside
907 * rtnl_lock(), and result refcount is unchanged.
910 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
913 struct net_device *dev, *ret;
918 for_each_netdev(net, dev) {
919 if (((dev->flags ^ if_flags) & mask) == 0) {
926 EXPORT_SYMBOL(__dev_get_by_flags);
929 * dev_valid_name - check if name is okay for network device
932 * Network device names need to be valid file names to
933 * to allow sysfs to work. We also disallow any kind of
936 bool dev_valid_name(const char *name)
940 if (strlen(name) >= IFNAMSIZ)
942 if (!strcmp(name, ".") || !strcmp(name, ".."))
946 if (*name == '/' || isspace(*name))
952 EXPORT_SYMBOL(dev_valid_name);
955 * __dev_alloc_name - allocate a name for a device
956 * @net: network namespace to allocate the device name in
957 * @name: name format string
958 * @buf: scratch buffer and result name string
960 * Passed a format string - eg "lt%d" it will try and find a suitable
961 * id. It scans list of devices to build up a free map, then chooses
962 * the first empty slot. The caller must hold the dev_base or rtnl lock
963 * while allocating the name and adding the device in order to avoid
965 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
966 * Returns the number of the unit assigned or a negative errno code.
969 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
973 const int max_netdevices = 8*PAGE_SIZE;
974 unsigned long *inuse;
975 struct net_device *d;
977 p = strnchr(name, IFNAMSIZ-1, '%');
980 * Verify the string as this thing may have come from
981 * the user. There must be either one "%d" and no other "%"
984 if (p[1] != 'd' || strchr(p + 2, '%'))
987 /* Use one page as a bit array of possible slots */
988 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
992 for_each_netdev(net, d) {
993 if (!sscanf(d->name, name, &i))
995 if (i < 0 || i >= max_netdevices)
998 /* avoid cases where sscanf is not exact inverse of printf */
999 snprintf(buf, IFNAMSIZ, name, i);
1000 if (!strncmp(buf, d->name, IFNAMSIZ))
1004 i = find_first_zero_bit(inuse, max_netdevices);
1005 free_page((unsigned long) inuse);
1009 snprintf(buf, IFNAMSIZ, name, i);
1010 if (!__dev_get_by_name(net, buf))
1013 /* It is possible to run out of possible slots
1014 * when the name is long and there isn't enough space left
1015 * for the digits, or if all bits are used.
1021 * dev_alloc_name - allocate a name for a device
1023 * @name: name format string
1025 * Passed a format string - eg "lt%d" it will try and find a suitable
1026 * id. It scans list of devices to build up a free map, then chooses
1027 * the first empty slot. The caller must hold the dev_base or rtnl lock
1028 * while allocating the name and adding the device in order to avoid
1030 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1031 * Returns the number of the unit assigned or a negative errno code.
1034 int dev_alloc_name(struct net_device *dev, const char *name)
1040 BUG_ON(!dev_net(dev));
1042 ret = __dev_alloc_name(net, name, buf);
1044 strlcpy(dev->name, buf, IFNAMSIZ);
1047 EXPORT_SYMBOL(dev_alloc_name);
1049 static int dev_alloc_name_ns(struct net *net,
1050 struct net_device *dev,
1056 ret = __dev_alloc_name(net, name, buf);
1058 strlcpy(dev->name, buf, IFNAMSIZ);
1062 static int dev_get_valid_name(struct net *net,
1063 struct net_device *dev,
1068 if (!dev_valid_name(name))
1071 if (strchr(name, '%'))
1072 return dev_alloc_name_ns(net, dev, name);
1073 else if (__dev_get_by_name(net, name))
1075 else if (dev->name != name)
1076 strlcpy(dev->name, name, IFNAMSIZ);
1082 * dev_change_name - change name of a device
1084 * @newname: name (or format string) must be at least IFNAMSIZ
1086 * Change name of a device, can pass format strings "eth%d".
1089 int dev_change_name(struct net_device *dev, const char *newname)
1091 unsigned char old_assign_type;
1092 char oldname[IFNAMSIZ];
1098 BUG_ON(!dev_net(dev));
1101 if (dev->flags & IFF_UP)
1104 write_seqcount_begin(&devnet_rename_seq);
1106 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1107 write_seqcount_end(&devnet_rename_seq);
1111 memcpy(oldname, dev->name, IFNAMSIZ);
1113 err = dev_get_valid_name(net, dev, newname);
1115 write_seqcount_end(&devnet_rename_seq);
1119 if (oldname[0] && !strchr(oldname, '%'))
1120 netdev_info(dev, "renamed from %s\n", oldname);
1122 old_assign_type = dev->name_assign_type;
1123 dev->name_assign_type = NET_NAME_RENAMED;
1126 ret = device_rename(&dev->dev, dev->name);
1128 memcpy(dev->name, oldname, IFNAMSIZ);
1129 dev->name_assign_type = old_assign_type;
1130 write_seqcount_end(&devnet_rename_seq);
1134 write_seqcount_end(&devnet_rename_seq);
1136 netdev_adjacent_rename_links(dev, oldname);
1138 write_lock_bh(&dev_base_lock);
1139 hlist_del_rcu(&dev->name_hlist);
1140 write_unlock_bh(&dev_base_lock);
1144 write_lock_bh(&dev_base_lock);
1145 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1146 write_unlock_bh(&dev_base_lock);
1148 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1149 ret = notifier_to_errno(ret);
1152 /* err >= 0 after dev_alloc_name() or stores the first errno */
1155 write_seqcount_begin(&devnet_rename_seq);
1156 memcpy(dev->name, oldname, IFNAMSIZ);
1157 memcpy(oldname, newname, IFNAMSIZ);
1158 dev->name_assign_type = old_assign_type;
1159 old_assign_type = NET_NAME_RENAMED;
1162 pr_err("%s: name change rollback failed: %d\n",
1171 * dev_set_alias - change ifalias of a device
1173 * @alias: name up to IFALIASZ
1174 * @len: limit of bytes to copy from info
1176 * Set ifalias for a device,
1178 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1184 if (len >= IFALIASZ)
1188 kfree(dev->ifalias);
1189 dev->ifalias = NULL;
1193 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1196 dev->ifalias = new_ifalias;
1198 strlcpy(dev->ifalias, alias, len+1);
1204 * netdev_features_change - device changes features
1205 * @dev: device to cause notification
1207 * Called to indicate a device has changed features.
1209 void netdev_features_change(struct net_device *dev)
1211 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1213 EXPORT_SYMBOL(netdev_features_change);
1216 * netdev_state_change - device changes state
1217 * @dev: device to cause notification
1219 * Called to indicate a device has changed state. This function calls
1220 * the notifier chains for netdev_chain and sends a NEWLINK message
1221 * to the routing socket.
1223 void netdev_state_change(struct net_device *dev)
1225 if (dev->flags & IFF_UP) {
1226 struct netdev_notifier_change_info change_info;
1228 change_info.flags_changed = 0;
1229 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1231 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1234 EXPORT_SYMBOL(netdev_state_change);
1237 * netdev_notify_peers - notify network peers about existence of @dev
1238 * @dev: network device
1240 * Generate traffic such that interested network peers are aware of
1241 * @dev, such as by generating a gratuitous ARP. This may be used when
1242 * a device wants to inform the rest of the network about some sort of
1243 * reconfiguration such as a failover event or virtual machine
1246 void netdev_notify_peers(struct net_device *dev)
1249 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1252 EXPORT_SYMBOL(netdev_notify_peers);
1254 static int __dev_open(struct net_device *dev)
1256 const struct net_device_ops *ops = dev->netdev_ops;
1261 if (!netif_device_present(dev))
1264 /* Block netpoll from trying to do any rx path servicing.
1265 * If we don't do this there is a chance ndo_poll_controller
1266 * or ndo_poll may be running while we open the device
1268 netpoll_poll_disable(dev);
1270 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1271 ret = notifier_to_errno(ret);
1275 set_bit(__LINK_STATE_START, &dev->state);
1277 if (ops->ndo_validate_addr)
1278 ret = ops->ndo_validate_addr(dev);
1280 if (!ret && ops->ndo_open)
1281 ret = ops->ndo_open(dev);
1283 netpoll_poll_enable(dev);
1286 clear_bit(__LINK_STATE_START, &dev->state);
1288 dev->flags |= IFF_UP;
1289 net_dmaengine_get();
1290 dev_set_rx_mode(dev);
1292 add_device_randomness(dev->dev_addr, dev->addr_len);
1299 * dev_open - prepare an interface for use.
1300 * @dev: device to open
1302 * Takes a device from down to up state. The device's private open
1303 * function is invoked and then the multicast lists are loaded. Finally
1304 * the device is moved into the up state and a %NETDEV_UP message is
1305 * sent to the netdev notifier chain.
1307 * Calling this function on an active interface is a nop. On a failure
1308 * a negative errno code is returned.
1310 int dev_open(struct net_device *dev)
1314 if (dev->flags & IFF_UP)
1317 ret = __dev_open(dev);
1321 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1322 call_netdevice_notifiers(NETDEV_UP, dev);
1326 EXPORT_SYMBOL(dev_open);
1328 static int __dev_close_many(struct list_head *head)
1330 struct net_device *dev;
1335 list_for_each_entry(dev, head, close_list) {
1336 /* Temporarily disable netpoll until the interface is down */
1337 netpoll_poll_disable(dev);
1339 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1341 clear_bit(__LINK_STATE_START, &dev->state);
1343 /* Synchronize to scheduled poll. We cannot touch poll list, it
1344 * can be even on different cpu. So just clear netif_running().
1346 * dev->stop() will invoke napi_disable() on all of it's
1347 * napi_struct instances on this device.
1349 smp_mb__after_atomic(); /* Commit netif_running(). */
1352 dev_deactivate_many(head);
1354 list_for_each_entry(dev, head, close_list) {
1355 const struct net_device_ops *ops = dev->netdev_ops;
1358 * Call the device specific close. This cannot fail.
1359 * Only if device is UP
1361 * We allow it to be called even after a DETACH hot-plug
1367 dev->flags &= ~IFF_UP;
1368 net_dmaengine_put();
1369 netpoll_poll_enable(dev);
1375 static int __dev_close(struct net_device *dev)
1380 list_add(&dev->close_list, &single);
1381 retval = __dev_close_many(&single);
1387 static int dev_close_many(struct list_head *head)
1389 struct net_device *dev, *tmp;
1391 /* Remove the devices that don't need to be closed */
1392 list_for_each_entry_safe(dev, tmp, head, close_list)
1393 if (!(dev->flags & IFF_UP))
1394 list_del_init(&dev->close_list);
1396 __dev_close_many(head);
1398 list_for_each_entry_safe(dev, tmp, head, close_list) {
1399 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1400 call_netdevice_notifiers(NETDEV_DOWN, dev);
1401 list_del_init(&dev->close_list);
1408 * dev_close - shutdown an interface.
1409 * @dev: device to shutdown
1411 * This function moves an active device into down state. A
1412 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1413 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1416 int dev_close(struct net_device *dev)
1418 if (dev->flags & IFF_UP) {
1421 list_add(&dev->close_list, &single);
1422 dev_close_many(&single);
1427 EXPORT_SYMBOL(dev_close);
1431 * dev_disable_lro - disable Large Receive Offload on a device
1434 * Disable Large Receive Offload (LRO) on a net device. Must be
1435 * called under RTNL. This is needed if received packets may be
1436 * forwarded to another interface.
1438 void dev_disable_lro(struct net_device *dev)
1441 * If we're trying to disable lro on a vlan device
1442 * use the underlying physical device instead
1444 if (is_vlan_dev(dev))
1445 dev = vlan_dev_real_dev(dev);
1447 /* the same for macvlan devices */
1448 if (netif_is_macvlan(dev))
1449 dev = macvlan_dev_real_dev(dev);
1451 dev->wanted_features &= ~NETIF_F_LRO;
1452 netdev_update_features(dev);
1454 if (unlikely(dev->features & NETIF_F_LRO))
1455 netdev_WARN(dev, "failed to disable LRO!\n");
1457 EXPORT_SYMBOL(dev_disable_lro);
1459 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1460 struct net_device *dev)
1462 struct netdev_notifier_info info;
1464 netdev_notifier_info_init(&info, dev);
1465 return nb->notifier_call(nb, val, &info);
1468 static int dev_boot_phase = 1;
1471 * register_netdevice_notifier - register a network notifier block
1474 * Register a notifier to be called when network device events occur.
1475 * The notifier passed is linked into the kernel structures and must
1476 * not be reused until it has been unregistered. A negative errno code
1477 * is returned on a failure.
1479 * When registered all registration and up events are replayed
1480 * to the new notifier to allow device to have a race free
1481 * view of the network device list.
1484 int register_netdevice_notifier(struct notifier_block *nb)
1486 struct net_device *dev;
1487 struct net_device *last;
1492 err = raw_notifier_chain_register(&netdev_chain, nb);
1498 for_each_netdev(net, dev) {
1499 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1500 err = notifier_to_errno(err);
1504 if (!(dev->flags & IFF_UP))
1507 call_netdevice_notifier(nb, NETDEV_UP, dev);
1518 for_each_netdev(net, dev) {
1522 if (dev->flags & IFF_UP) {
1523 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1525 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1527 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1532 raw_notifier_chain_unregister(&netdev_chain, nb);
1535 EXPORT_SYMBOL(register_netdevice_notifier);
1538 * unregister_netdevice_notifier - unregister a network notifier block
1541 * Unregister a notifier previously registered by
1542 * register_netdevice_notifier(). The notifier is unlinked into the
1543 * kernel structures and may then be reused. A negative errno code
1544 * is returned on a failure.
1546 * After unregistering unregister and down device events are synthesized
1547 * for all devices on the device list to the removed notifier to remove
1548 * the need for special case cleanup code.
1551 int unregister_netdevice_notifier(struct notifier_block *nb)
1553 struct net_device *dev;
1558 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1563 for_each_netdev(net, dev) {
1564 if (dev->flags & IFF_UP) {
1565 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1567 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1569 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1576 EXPORT_SYMBOL(unregister_netdevice_notifier);
1579 * call_netdevice_notifiers_info - call all network notifier blocks
1580 * @val: value passed unmodified to notifier function
1581 * @dev: net_device pointer passed unmodified to notifier function
1582 * @info: notifier information data
1584 * Call all network notifier blocks. Parameters and return value
1585 * are as for raw_notifier_call_chain().
1588 static int call_netdevice_notifiers_info(unsigned long val,
1589 struct net_device *dev,
1590 struct netdev_notifier_info *info)
1593 netdev_notifier_info_init(info, dev);
1594 return raw_notifier_call_chain(&netdev_chain, val, info);
1598 * call_netdevice_notifiers - call all network notifier blocks
1599 * @val: value passed unmodified to notifier function
1600 * @dev: net_device pointer passed unmodified to notifier function
1602 * Call all network notifier blocks. Parameters and return value
1603 * are as for raw_notifier_call_chain().
1606 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1608 struct netdev_notifier_info info;
1610 return call_netdevice_notifiers_info(val, dev, &info);
1612 EXPORT_SYMBOL(call_netdevice_notifiers);
1614 static struct static_key netstamp_needed __read_mostly;
1615 #ifdef HAVE_JUMP_LABEL
1616 /* We are not allowed to call static_key_slow_dec() from irq context
1617 * If net_disable_timestamp() is called from irq context, defer the
1618 * static_key_slow_dec() calls.
1620 static atomic_t netstamp_needed_deferred;
1623 void net_enable_timestamp(void)
1625 #ifdef HAVE_JUMP_LABEL
1626 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1630 static_key_slow_dec(&netstamp_needed);
1634 static_key_slow_inc(&netstamp_needed);
1636 EXPORT_SYMBOL(net_enable_timestamp);
1638 void net_disable_timestamp(void)
1640 #ifdef HAVE_JUMP_LABEL
1641 if (in_interrupt()) {
1642 atomic_inc(&netstamp_needed_deferred);
1646 static_key_slow_dec(&netstamp_needed);
1648 EXPORT_SYMBOL(net_disable_timestamp);
1650 static inline void net_timestamp_set(struct sk_buff *skb)
1652 skb->tstamp.tv64 = 0;
1653 if (static_key_false(&netstamp_needed))
1654 __net_timestamp(skb);
1657 #define net_timestamp_check(COND, SKB) \
1658 if (static_key_false(&netstamp_needed)) { \
1659 if ((COND) && !(SKB)->tstamp.tv64) \
1660 __net_timestamp(SKB); \
1663 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1667 if (!(dev->flags & IFF_UP))
1670 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1671 if (skb->len <= len)
1674 /* if TSO is enabled, we don't care about the length as the packet
1675 * could be forwarded without being segmented before
1677 if (skb_is_gso(skb))
1682 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1684 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1686 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1687 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1688 atomic_long_inc(&dev->rx_dropped);
1694 if (unlikely(!is_skb_forwardable(dev, skb))) {
1695 atomic_long_inc(&dev->rx_dropped);
1700 skb_scrub_packet(skb, true);
1701 skb->protocol = eth_type_trans(skb, dev);
1705 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1708 * dev_forward_skb - loopback an skb to another netif
1710 * @dev: destination network device
1711 * @skb: buffer to forward
1714 * NET_RX_SUCCESS (no congestion)
1715 * NET_RX_DROP (packet was dropped, but freed)
1717 * dev_forward_skb can be used for injecting an skb from the
1718 * start_xmit function of one device into the receive queue
1719 * of another device.
1721 * The receiving device may be in another namespace, so
1722 * we have to clear all information in the skb that could
1723 * impact namespace isolation.
1725 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1727 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1729 EXPORT_SYMBOL_GPL(dev_forward_skb);
1731 static inline int deliver_skb(struct sk_buff *skb,
1732 struct packet_type *pt_prev,
1733 struct net_device *orig_dev)
1735 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1737 atomic_inc(&skb->users);
1738 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1741 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1743 if (!ptype->af_packet_priv || !skb->sk)
1746 if (ptype->id_match)
1747 return ptype->id_match(ptype, skb->sk);
1748 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1755 * Support routine. Sends outgoing frames to any network
1756 * taps currently in use.
1759 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1761 struct packet_type *ptype;
1762 struct sk_buff *skb2 = NULL;
1763 struct packet_type *pt_prev = NULL;
1766 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1767 /* Never send packets back to the socket
1768 * they originated from - MvS (miquels@drinkel.ow.org)
1770 if ((ptype->dev == dev || !ptype->dev) &&
1771 (!skb_loop_sk(ptype, skb))) {
1773 deliver_skb(skb2, pt_prev, skb->dev);
1778 skb2 = skb_clone(skb, GFP_ATOMIC);
1782 net_timestamp_set(skb2);
1784 /* skb->nh should be correctly
1785 set by sender, so that the second statement is
1786 just protection against buggy protocols.
1788 skb_reset_mac_header(skb2);
1790 if (skb_network_header(skb2) < skb2->data ||
1791 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1792 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1793 ntohs(skb2->protocol),
1795 skb_reset_network_header(skb2);
1798 skb2->transport_header = skb2->network_header;
1799 skb2->pkt_type = PACKET_OUTGOING;
1804 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1809 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1810 * @dev: Network device
1811 * @txq: number of queues available
1813 * If real_num_tx_queues is changed the tc mappings may no longer be
1814 * valid. To resolve this verify the tc mapping remains valid and if
1815 * not NULL the mapping. With no priorities mapping to this
1816 * offset/count pair it will no longer be used. In the worst case TC0
1817 * is invalid nothing can be done so disable priority mappings. If is
1818 * expected that drivers will fix this mapping if they can before
1819 * calling netif_set_real_num_tx_queues.
1821 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1824 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1826 /* If TC0 is invalidated disable TC mapping */
1827 if (tc->offset + tc->count > txq) {
1828 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1833 /* Invalidated prio to tc mappings set to TC0 */
1834 for (i = 1; i < TC_BITMASK + 1; i++) {
1835 int q = netdev_get_prio_tc_map(dev, i);
1837 tc = &dev->tc_to_txq[q];
1838 if (tc->offset + tc->count > txq) {
1839 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1841 netdev_set_prio_tc_map(dev, i, 0);
1847 static DEFINE_MUTEX(xps_map_mutex);
1848 #define xmap_dereference(P) \
1849 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1851 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1854 struct xps_map *map = NULL;
1858 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1860 for (pos = 0; map && pos < map->len; pos++) {
1861 if (map->queues[pos] == index) {
1863 map->queues[pos] = map->queues[--map->len];
1865 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1866 kfree_rcu(map, rcu);
1876 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1878 struct xps_dev_maps *dev_maps;
1880 bool active = false;
1882 mutex_lock(&xps_map_mutex);
1883 dev_maps = xmap_dereference(dev->xps_maps);
1888 for_each_possible_cpu(cpu) {
1889 for (i = index; i < dev->num_tx_queues; i++) {
1890 if (!remove_xps_queue(dev_maps, cpu, i))
1893 if (i == dev->num_tx_queues)
1898 RCU_INIT_POINTER(dev->xps_maps, NULL);
1899 kfree_rcu(dev_maps, rcu);
1902 for (i = index; i < dev->num_tx_queues; i++)
1903 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1907 mutex_unlock(&xps_map_mutex);
1910 static struct xps_map *expand_xps_map(struct xps_map *map,
1913 struct xps_map *new_map;
1914 int alloc_len = XPS_MIN_MAP_ALLOC;
1917 for (pos = 0; map && pos < map->len; pos++) {
1918 if (map->queues[pos] != index)
1923 /* Need to add queue to this CPU's existing map */
1925 if (pos < map->alloc_len)
1928 alloc_len = map->alloc_len * 2;
1931 /* Need to allocate new map to store queue on this CPU's map */
1932 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1937 for (i = 0; i < pos; i++)
1938 new_map->queues[i] = map->queues[i];
1939 new_map->alloc_len = alloc_len;
1945 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1948 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1949 struct xps_map *map, *new_map;
1950 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1951 int cpu, numa_node_id = -2;
1952 bool active = false;
1954 mutex_lock(&xps_map_mutex);
1956 dev_maps = xmap_dereference(dev->xps_maps);
1958 /* allocate memory for queue storage */
1959 for_each_online_cpu(cpu) {
1960 if (!cpumask_test_cpu(cpu, mask))
1964 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1965 if (!new_dev_maps) {
1966 mutex_unlock(&xps_map_mutex);
1970 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1973 map = expand_xps_map(map, cpu, index);
1977 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1981 goto out_no_new_maps;
1983 for_each_possible_cpu(cpu) {
1984 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1985 /* add queue to CPU maps */
1988 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1989 while ((pos < map->len) && (map->queues[pos] != index))
1992 if (pos == map->len)
1993 map->queues[map->len++] = index;
1995 if (numa_node_id == -2)
1996 numa_node_id = cpu_to_node(cpu);
1997 else if (numa_node_id != cpu_to_node(cpu))
2000 } else if (dev_maps) {
2001 /* fill in the new device map from the old device map */
2002 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2003 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2008 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2010 /* Cleanup old maps */
2012 for_each_possible_cpu(cpu) {
2013 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2014 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2015 if (map && map != new_map)
2016 kfree_rcu(map, rcu);
2019 kfree_rcu(dev_maps, rcu);
2022 dev_maps = new_dev_maps;
2026 /* update Tx queue numa node */
2027 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2028 (numa_node_id >= 0) ? numa_node_id :
2034 /* removes queue from unused CPUs */
2035 for_each_possible_cpu(cpu) {
2036 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2039 if (remove_xps_queue(dev_maps, cpu, index))
2043 /* free map if not active */
2045 RCU_INIT_POINTER(dev->xps_maps, NULL);
2046 kfree_rcu(dev_maps, rcu);
2050 mutex_unlock(&xps_map_mutex);
2054 /* remove any maps that we added */
2055 for_each_possible_cpu(cpu) {
2056 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2057 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2059 if (new_map && new_map != map)
2063 mutex_unlock(&xps_map_mutex);
2065 kfree(new_dev_maps);
2068 EXPORT_SYMBOL(netif_set_xps_queue);
2072 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2073 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2075 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2079 if (txq < 1 || txq > dev->num_tx_queues)
2082 if (dev->reg_state == NETREG_REGISTERED ||
2083 dev->reg_state == NETREG_UNREGISTERING) {
2086 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2092 netif_setup_tc(dev, txq);
2094 if (txq < dev->real_num_tx_queues) {
2095 qdisc_reset_all_tx_gt(dev, txq);
2097 netif_reset_xps_queues_gt(dev, txq);
2102 dev->real_num_tx_queues = txq;
2105 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2109 * netif_set_real_num_rx_queues - set actual number of RX queues used
2110 * @dev: Network device
2111 * @rxq: Actual number of RX queues
2113 * This must be called either with the rtnl_lock held or before
2114 * registration of the net device. Returns 0 on success, or a
2115 * negative error code. If called before registration, it always
2118 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2122 if (rxq < 1 || rxq > dev->num_rx_queues)
2125 if (dev->reg_state == NETREG_REGISTERED) {
2128 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2134 dev->real_num_rx_queues = rxq;
2137 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2141 * netif_get_num_default_rss_queues - default number of RSS queues
2143 * This routine should set an upper limit on the number of RSS queues
2144 * used by default by multiqueue devices.
2146 int netif_get_num_default_rss_queues(void)
2148 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2150 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2152 static inline void __netif_reschedule(struct Qdisc *q)
2154 struct softnet_data *sd;
2155 unsigned long flags;
2157 local_irq_save(flags);
2158 sd = &__get_cpu_var(softnet_data);
2159 q->next_sched = NULL;
2160 *sd->output_queue_tailp = q;
2161 sd->output_queue_tailp = &q->next_sched;
2162 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2163 local_irq_restore(flags);
2166 void __netif_schedule(struct Qdisc *q)
2168 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2169 __netif_reschedule(q);
2171 EXPORT_SYMBOL(__netif_schedule);
2173 struct dev_kfree_skb_cb {
2174 enum skb_free_reason reason;
2177 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2179 return (struct dev_kfree_skb_cb *)skb->cb;
2182 void netif_schedule_queue(struct netdev_queue *txq)
2185 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2186 struct Qdisc *q = rcu_dereference(txq->qdisc);
2188 __netif_schedule(q);
2192 EXPORT_SYMBOL(netif_schedule_queue);
2195 * netif_wake_subqueue - allow sending packets on subqueue
2196 * @dev: network device
2197 * @queue_index: sub queue index
2199 * Resume individual transmit queue of a device with multiple transmit queues.
2201 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2203 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2205 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2209 q = rcu_dereference(txq->qdisc);
2210 __netif_schedule(q);
2214 EXPORT_SYMBOL(netif_wake_subqueue);
2216 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2218 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2222 q = rcu_dereference(dev_queue->qdisc);
2223 __netif_schedule(q);
2227 EXPORT_SYMBOL(netif_tx_wake_queue);
2229 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2231 unsigned long flags;
2233 if (likely(atomic_read(&skb->users) == 1)) {
2235 atomic_set(&skb->users, 0);
2236 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2239 get_kfree_skb_cb(skb)->reason = reason;
2240 local_irq_save(flags);
2241 skb->next = __this_cpu_read(softnet_data.completion_queue);
2242 __this_cpu_write(softnet_data.completion_queue, skb);
2243 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2244 local_irq_restore(flags);
2246 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2248 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2250 if (in_irq() || irqs_disabled())
2251 __dev_kfree_skb_irq(skb, reason);
2255 EXPORT_SYMBOL(__dev_kfree_skb_any);
2259 * netif_device_detach - mark device as removed
2260 * @dev: network device
2262 * Mark device as removed from system and therefore no longer available.
2264 void netif_device_detach(struct net_device *dev)
2266 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2267 netif_running(dev)) {
2268 netif_tx_stop_all_queues(dev);
2271 EXPORT_SYMBOL(netif_device_detach);
2274 * netif_device_attach - mark device as attached
2275 * @dev: network device
2277 * Mark device as attached from system and restart if needed.
2279 void netif_device_attach(struct net_device *dev)
2281 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2282 netif_running(dev)) {
2283 netif_tx_wake_all_queues(dev);
2284 __netdev_watchdog_up(dev);
2287 EXPORT_SYMBOL(netif_device_attach);
2289 static void skb_warn_bad_offload(const struct sk_buff *skb)
2291 static const netdev_features_t null_features = 0;
2292 struct net_device *dev = skb->dev;
2293 const char *driver = "";
2295 if (!net_ratelimit())
2298 if (dev && dev->dev.parent)
2299 driver = dev_driver_string(dev->dev.parent);
2301 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2302 "gso_type=%d ip_summed=%d\n",
2303 driver, dev ? &dev->features : &null_features,
2304 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2305 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2306 skb_shinfo(skb)->gso_type, skb->ip_summed);
2310 * Invalidate hardware checksum when packet is to be mangled, and
2311 * complete checksum manually on outgoing path.
2313 int skb_checksum_help(struct sk_buff *skb)
2316 int ret = 0, offset;
2318 if (skb->ip_summed == CHECKSUM_COMPLETE)
2319 goto out_set_summed;
2321 if (unlikely(skb_shinfo(skb)->gso_size)) {
2322 skb_warn_bad_offload(skb);
2326 /* Before computing a checksum, we should make sure no frag could
2327 * be modified by an external entity : checksum could be wrong.
2329 if (skb_has_shared_frag(skb)) {
2330 ret = __skb_linearize(skb);
2335 offset = skb_checksum_start_offset(skb);
2336 BUG_ON(offset >= skb_headlen(skb));
2337 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2339 offset += skb->csum_offset;
2340 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2342 if (skb_cloned(skb) &&
2343 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2344 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2349 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2351 skb->ip_summed = CHECKSUM_NONE;
2355 EXPORT_SYMBOL(skb_checksum_help);
2357 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2359 unsigned int vlan_depth = skb->mac_len;
2360 __be16 type = skb->protocol;
2362 /* Tunnel gso handlers can set protocol to ethernet. */
2363 if (type == htons(ETH_P_TEB)) {
2366 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2369 eth = (struct ethhdr *)skb_mac_header(skb);
2370 type = eth->h_proto;
2373 /* if skb->protocol is 802.1Q/AD then the header should already be
2374 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at
2375 * ETH_HLEN otherwise
2377 if (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2379 if (WARN_ON(vlan_depth < VLAN_HLEN))
2381 vlan_depth -= VLAN_HLEN;
2383 vlan_depth = ETH_HLEN;
2386 struct vlan_hdr *vh;
2388 if (unlikely(!pskb_may_pull(skb,
2389 vlan_depth + VLAN_HLEN)))
2392 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2393 type = vh->h_vlan_encapsulated_proto;
2394 vlan_depth += VLAN_HLEN;
2395 } while (type == htons(ETH_P_8021Q) ||
2396 type == htons(ETH_P_8021AD));
2399 *depth = vlan_depth;
2405 * skb_mac_gso_segment - mac layer segmentation handler.
2406 * @skb: buffer to segment
2407 * @features: features for the output path (see dev->features)
2409 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2410 netdev_features_t features)
2412 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2413 struct packet_offload *ptype;
2414 int vlan_depth = skb->mac_len;
2415 __be16 type = skb_network_protocol(skb, &vlan_depth);
2417 if (unlikely(!type))
2418 return ERR_PTR(-EINVAL);
2420 __skb_pull(skb, vlan_depth);
2423 list_for_each_entry_rcu(ptype, &offload_base, list) {
2424 if (ptype->type == type && ptype->callbacks.gso_segment) {
2425 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2428 err = ptype->callbacks.gso_send_check(skb);
2429 segs = ERR_PTR(err);
2430 if (err || skb_gso_ok(skb, features))
2432 __skb_push(skb, (skb->data -
2433 skb_network_header(skb)));
2435 segs = ptype->callbacks.gso_segment(skb, features);
2441 __skb_push(skb, skb->data - skb_mac_header(skb));
2445 EXPORT_SYMBOL(skb_mac_gso_segment);
2448 /* openvswitch calls this on rx path, so we need a different check.
2450 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2453 return skb->ip_summed != CHECKSUM_PARTIAL;
2455 return skb->ip_summed == CHECKSUM_NONE;
2459 * __skb_gso_segment - Perform segmentation on skb.
2460 * @skb: buffer to segment
2461 * @features: features for the output path (see dev->features)
2462 * @tx_path: whether it is called in TX path
2464 * This function segments the given skb and returns a list of segments.
2466 * It may return NULL if the skb requires no segmentation. This is
2467 * only possible when GSO is used for verifying header integrity.
2469 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2470 netdev_features_t features, bool tx_path)
2472 if (unlikely(skb_needs_check(skb, tx_path))) {
2475 skb_warn_bad_offload(skb);
2477 err = skb_cow_head(skb, 0);
2479 return ERR_PTR(err);
2482 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2483 SKB_GSO_CB(skb)->encap_level = 0;
2485 skb_reset_mac_header(skb);
2486 skb_reset_mac_len(skb);
2488 return skb_mac_gso_segment(skb, features);
2490 EXPORT_SYMBOL(__skb_gso_segment);
2492 /* Take action when hardware reception checksum errors are detected. */
2494 void netdev_rx_csum_fault(struct net_device *dev)
2496 if (net_ratelimit()) {
2497 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2501 EXPORT_SYMBOL(netdev_rx_csum_fault);
2504 /* Actually, we should eliminate this check as soon as we know, that:
2505 * 1. IOMMU is present and allows to map all the memory.
2506 * 2. No high memory really exists on this machine.
2509 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2511 #ifdef CONFIG_HIGHMEM
2513 if (!(dev->features & NETIF_F_HIGHDMA)) {
2514 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2515 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2516 if (PageHighMem(skb_frag_page(frag)))
2521 if (PCI_DMA_BUS_IS_PHYS) {
2522 struct device *pdev = dev->dev.parent;
2526 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2527 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2528 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2529 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2537 /* If MPLS offload request, verify we are testing hardware MPLS features
2538 * instead of standard features for the netdev.
2540 #ifdef CONFIG_NET_MPLS_GSO
2541 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2542 netdev_features_t features,
2545 if (type == htons(ETH_P_MPLS_UC) || type == htons(ETH_P_MPLS_MC))
2546 features &= skb->dev->mpls_features;
2551 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2552 netdev_features_t features,
2559 static netdev_features_t harmonize_features(struct sk_buff *skb,
2560 netdev_features_t features)
2565 type = skb_network_protocol(skb, &tmp);
2566 features = net_mpls_features(skb, features, type);
2568 if (skb->ip_summed != CHECKSUM_NONE &&
2569 !can_checksum_protocol(features, type)) {
2570 features &= ~NETIF_F_ALL_CSUM;
2571 } else if (illegal_highdma(skb->dev, skb)) {
2572 features &= ~NETIF_F_SG;
2578 netdev_features_t netif_skb_features(struct sk_buff *skb)
2580 __be16 protocol = skb->protocol;
2581 netdev_features_t features = skb->dev->features;
2583 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2584 features &= ~NETIF_F_GSO_MASK;
2586 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2587 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2588 protocol = veh->h_vlan_encapsulated_proto;
2589 } else if (!vlan_tx_tag_present(skb)) {
2590 return harmonize_features(skb, features);
2593 features = netdev_intersect_features(features,
2594 skb->dev->vlan_features |
2595 NETIF_F_HW_VLAN_CTAG_TX |
2596 NETIF_F_HW_VLAN_STAG_TX);
2598 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2599 features = netdev_intersect_features(features,
2604 NETIF_F_HW_VLAN_CTAG_TX |
2605 NETIF_F_HW_VLAN_STAG_TX);
2607 return harmonize_features(skb, features);
2609 EXPORT_SYMBOL(netif_skb_features);
2611 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2612 struct netdev_queue *txq, bool more)
2617 if (!list_empty(&ptype_all))
2618 dev_queue_xmit_nit(skb, dev);
2621 trace_net_dev_start_xmit(skb, dev);
2622 rc = netdev_start_xmit(skb, dev, txq, more);
2623 trace_net_dev_xmit(skb, rc, dev, len);
2628 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2629 struct netdev_queue *txq, int *ret)
2631 struct sk_buff *skb = first;
2632 int rc = NETDEV_TX_OK;
2635 struct sk_buff *next = skb->next;
2638 rc = xmit_one(skb, dev, txq, next != NULL);
2639 if (unlikely(!dev_xmit_complete(rc))) {
2645 if (netif_xmit_stopped(txq) && skb) {
2646 rc = NETDEV_TX_BUSY;
2656 struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, netdev_features_t features)
2658 if (vlan_tx_tag_present(skb) &&
2659 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2660 skb = __vlan_put_tag(skb, skb->vlan_proto,
2661 vlan_tx_tag_get(skb));
2668 struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2670 netdev_features_t features;
2675 /* If device doesn't need skb->dst, release it right now while
2676 * its hot in this cpu cache
2678 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2681 features = netif_skb_features(skb);
2682 skb = validate_xmit_vlan(skb, features);
2686 /* If encapsulation offload request, verify we are testing
2687 * hardware encapsulation features instead of standard
2688 * features for the netdev
2690 if (skb->encapsulation)
2691 features &= dev->hw_enc_features;
2693 if (netif_needs_gso(skb, features)) {
2694 struct sk_buff *segs;
2696 segs = skb_gso_segment(skb, features);
2704 if (skb_needs_linearize(skb, features) &&
2705 __skb_linearize(skb))
2708 /* If packet is not checksummed and device does not
2709 * support checksumming for this protocol, complete
2710 * checksumming here.
2712 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2713 if (skb->encapsulation)
2714 skb_set_inner_transport_header(skb,
2715 skb_checksum_start_offset(skb));
2717 skb_set_transport_header(skb,
2718 skb_checksum_start_offset(skb));
2719 if (!(features & NETIF_F_ALL_CSUM) &&
2720 skb_checksum_help(skb))
2733 static void qdisc_pkt_len_init(struct sk_buff *skb)
2735 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2737 qdisc_skb_cb(skb)->pkt_len = skb->len;
2739 /* To get more precise estimation of bytes sent on wire,
2740 * we add to pkt_len the headers size of all segments
2742 if (shinfo->gso_size) {
2743 unsigned int hdr_len;
2744 u16 gso_segs = shinfo->gso_segs;
2746 /* mac layer + network layer */
2747 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2749 /* + transport layer */
2750 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2751 hdr_len += tcp_hdrlen(skb);
2753 hdr_len += sizeof(struct udphdr);
2755 if (shinfo->gso_type & SKB_GSO_DODGY)
2756 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2759 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2763 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2764 struct net_device *dev,
2765 struct netdev_queue *txq)
2767 spinlock_t *root_lock = qdisc_lock(q);
2771 qdisc_pkt_len_init(skb);
2772 qdisc_calculate_pkt_len(skb, q);
2774 * Heuristic to force contended enqueues to serialize on a
2775 * separate lock before trying to get qdisc main lock.
2776 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2777 * often and dequeue packets faster.
2779 contended = qdisc_is_running(q);
2780 if (unlikely(contended))
2781 spin_lock(&q->busylock);
2783 spin_lock(root_lock);
2784 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2787 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2788 qdisc_run_begin(q)) {
2790 * This is a work-conserving queue; there are no old skbs
2791 * waiting to be sent out; and the qdisc is not running -
2792 * xmit the skb directly.
2794 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2797 qdisc_bstats_update(q, skb);
2799 skb = validate_xmit_skb(skb, dev);
2800 if (skb && sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2801 if (unlikely(contended)) {
2802 spin_unlock(&q->busylock);
2809 rc = NET_XMIT_SUCCESS;
2812 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2813 if (qdisc_run_begin(q)) {
2814 if (unlikely(contended)) {
2815 spin_unlock(&q->busylock);
2821 spin_unlock(root_lock);
2822 if (unlikely(contended))
2823 spin_unlock(&q->busylock);
2827 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2828 static void skb_update_prio(struct sk_buff *skb)
2830 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2832 if (!skb->priority && skb->sk && map) {
2833 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2835 if (prioidx < map->priomap_len)
2836 skb->priority = map->priomap[prioidx];
2840 #define skb_update_prio(skb)
2843 static DEFINE_PER_CPU(int, xmit_recursion);
2844 #define RECURSION_LIMIT 10
2847 * dev_loopback_xmit - loop back @skb
2848 * @skb: buffer to transmit
2850 int dev_loopback_xmit(struct sk_buff *skb)
2852 skb_reset_mac_header(skb);
2853 __skb_pull(skb, skb_network_offset(skb));
2854 skb->pkt_type = PACKET_LOOPBACK;
2855 skb->ip_summed = CHECKSUM_UNNECESSARY;
2856 WARN_ON(!skb_dst(skb));
2861 EXPORT_SYMBOL(dev_loopback_xmit);
2864 * __dev_queue_xmit - transmit a buffer
2865 * @skb: buffer to transmit
2866 * @accel_priv: private data used for L2 forwarding offload
2868 * Queue a buffer for transmission to a network device. The caller must
2869 * have set the device and priority and built the buffer before calling
2870 * this function. The function can be called from an interrupt.
2872 * A negative errno code is returned on a failure. A success does not
2873 * guarantee the frame will be transmitted as it may be dropped due
2874 * to congestion or traffic shaping.
2876 * -----------------------------------------------------------------------------------
2877 * I notice this method can also return errors from the queue disciplines,
2878 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2881 * Regardless of the return value, the skb is consumed, so it is currently
2882 * difficult to retry a send to this method. (You can bump the ref count
2883 * before sending to hold a reference for retry if you are careful.)
2885 * When calling this method, interrupts MUST be enabled. This is because
2886 * the BH enable code must have IRQs enabled so that it will not deadlock.
2889 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2891 struct net_device *dev = skb->dev;
2892 struct netdev_queue *txq;
2896 skb_reset_mac_header(skb);
2898 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
2899 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
2901 /* Disable soft irqs for various locks below. Also
2902 * stops preemption for RCU.
2906 skb_update_prio(skb);
2908 txq = netdev_pick_tx(dev, skb, accel_priv);
2909 q = rcu_dereference_bh(txq->qdisc);
2911 #ifdef CONFIG_NET_CLS_ACT
2912 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2914 trace_net_dev_queue(skb);
2916 rc = __dev_xmit_skb(skb, q, dev, txq);
2920 /* The device has no queue. Common case for software devices:
2921 loopback, all the sorts of tunnels...
2923 Really, it is unlikely that netif_tx_lock protection is necessary
2924 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2926 However, it is possible, that they rely on protection
2929 Check this and shot the lock. It is not prone from deadlocks.
2930 Either shot noqueue qdisc, it is even simpler 8)
2932 if (dev->flags & IFF_UP) {
2933 int cpu = smp_processor_id(); /* ok because BHs are off */
2935 if (txq->xmit_lock_owner != cpu) {
2937 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2938 goto recursion_alert;
2940 skb = validate_xmit_skb(skb, dev);
2944 HARD_TX_LOCK(dev, txq, cpu);
2946 if (!netif_xmit_stopped(txq)) {
2947 __this_cpu_inc(xmit_recursion);
2948 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
2949 __this_cpu_dec(xmit_recursion);
2950 if (dev_xmit_complete(rc)) {
2951 HARD_TX_UNLOCK(dev, txq);
2955 HARD_TX_UNLOCK(dev, txq);
2956 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2959 /* Recursion is detected! It is possible,
2963 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2970 rcu_read_unlock_bh();
2972 atomic_long_inc(&dev->tx_dropped);
2973 kfree_skb_list(skb);
2976 rcu_read_unlock_bh();
2980 int dev_queue_xmit(struct sk_buff *skb)
2982 return __dev_queue_xmit(skb, NULL);
2984 EXPORT_SYMBOL(dev_queue_xmit);
2986 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
2988 return __dev_queue_xmit(skb, accel_priv);
2990 EXPORT_SYMBOL(dev_queue_xmit_accel);
2993 /*=======================================================================
2995 =======================================================================*/
2997 int netdev_max_backlog __read_mostly = 1000;
2998 EXPORT_SYMBOL(netdev_max_backlog);
3000 int netdev_tstamp_prequeue __read_mostly = 1;
3001 int netdev_budget __read_mostly = 300;
3002 int weight_p __read_mostly = 64; /* old backlog weight */
3004 /* Called with irq disabled */
3005 static inline void ____napi_schedule(struct softnet_data *sd,
3006 struct napi_struct *napi)
3008 list_add_tail(&napi->poll_list, &sd->poll_list);
3009 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3014 /* One global table that all flow-based protocols share. */
3015 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3016 EXPORT_SYMBOL(rps_sock_flow_table);
3018 struct static_key rps_needed __read_mostly;
3020 static struct rps_dev_flow *
3021 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3022 struct rps_dev_flow *rflow, u16 next_cpu)
3024 if (next_cpu != RPS_NO_CPU) {
3025 #ifdef CONFIG_RFS_ACCEL
3026 struct netdev_rx_queue *rxqueue;
3027 struct rps_dev_flow_table *flow_table;
3028 struct rps_dev_flow *old_rflow;
3033 /* Should we steer this flow to a different hardware queue? */
3034 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3035 !(dev->features & NETIF_F_NTUPLE))
3037 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3038 if (rxq_index == skb_get_rx_queue(skb))
3041 rxqueue = dev->_rx + rxq_index;
3042 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3045 flow_id = skb_get_hash(skb) & flow_table->mask;
3046 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3047 rxq_index, flow_id);
3051 rflow = &flow_table->flows[flow_id];
3053 if (old_rflow->filter == rflow->filter)
3054 old_rflow->filter = RPS_NO_FILTER;
3058 per_cpu(softnet_data, next_cpu).input_queue_head;
3061 rflow->cpu = next_cpu;
3066 * get_rps_cpu is called from netif_receive_skb and returns the target
3067 * CPU from the RPS map of the receiving queue for a given skb.
3068 * rcu_read_lock must be held on entry.
3070 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3071 struct rps_dev_flow **rflowp)
3073 struct netdev_rx_queue *rxqueue;
3074 struct rps_map *map;
3075 struct rps_dev_flow_table *flow_table;
3076 struct rps_sock_flow_table *sock_flow_table;
3081 if (skb_rx_queue_recorded(skb)) {
3082 u16 index = skb_get_rx_queue(skb);
3083 if (unlikely(index >= dev->real_num_rx_queues)) {
3084 WARN_ONCE(dev->real_num_rx_queues > 1,
3085 "%s received packet on queue %u, but number "
3086 "of RX queues is %u\n",
3087 dev->name, index, dev->real_num_rx_queues);
3090 rxqueue = dev->_rx + index;
3094 map = rcu_dereference(rxqueue->rps_map);
3096 if (map->len == 1 &&
3097 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3098 tcpu = map->cpus[0];
3099 if (cpu_online(tcpu))
3103 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3107 skb_reset_network_header(skb);
3108 hash = skb_get_hash(skb);
3112 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3113 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3114 if (flow_table && sock_flow_table) {
3116 struct rps_dev_flow *rflow;
3118 rflow = &flow_table->flows[hash & flow_table->mask];
3121 next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask];
3124 * If the desired CPU (where last recvmsg was done) is
3125 * different from current CPU (one in the rx-queue flow
3126 * table entry), switch if one of the following holds:
3127 * - Current CPU is unset (equal to RPS_NO_CPU).
3128 * - Current CPU is offline.
3129 * - The current CPU's queue tail has advanced beyond the
3130 * last packet that was enqueued using this table entry.
3131 * This guarantees that all previous packets for the flow
3132 * have been dequeued, thus preserving in order delivery.
3134 if (unlikely(tcpu != next_cpu) &&
3135 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3136 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3137 rflow->last_qtail)) >= 0)) {
3139 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3142 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3150 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3151 if (cpu_online(tcpu)) {
3161 #ifdef CONFIG_RFS_ACCEL
3164 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3165 * @dev: Device on which the filter was set
3166 * @rxq_index: RX queue index
3167 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3168 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3170 * Drivers that implement ndo_rx_flow_steer() should periodically call
3171 * this function for each installed filter and remove the filters for
3172 * which it returns %true.
3174 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3175 u32 flow_id, u16 filter_id)
3177 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3178 struct rps_dev_flow_table *flow_table;
3179 struct rps_dev_flow *rflow;
3184 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3185 if (flow_table && flow_id <= flow_table->mask) {
3186 rflow = &flow_table->flows[flow_id];
3187 cpu = ACCESS_ONCE(rflow->cpu);
3188 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3189 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3190 rflow->last_qtail) <
3191 (int)(10 * flow_table->mask)))
3197 EXPORT_SYMBOL(rps_may_expire_flow);
3199 #endif /* CONFIG_RFS_ACCEL */
3201 /* Called from hardirq (IPI) context */
3202 static void rps_trigger_softirq(void *data)
3204 struct softnet_data *sd = data;
3206 ____napi_schedule(sd, &sd->backlog);
3210 #endif /* CONFIG_RPS */
3213 * Check if this softnet_data structure is another cpu one
3214 * If yes, queue it to our IPI list and return 1
3217 static int rps_ipi_queued(struct softnet_data *sd)
3220 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3223 sd->rps_ipi_next = mysd->rps_ipi_list;
3224 mysd->rps_ipi_list = sd;
3226 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3229 #endif /* CONFIG_RPS */
3233 #ifdef CONFIG_NET_FLOW_LIMIT
3234 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3237 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3239 #ifdef CONFIG_NET_FLOW_LIMIT
3240 struct sd_flow_limit *fl;
3241 struct softnet_data *sd;
3242 unsigned int old_flow, new_flow;
3244 if (qlen < (netdev_max_backlog >> 1))
3247 sd = &__get_cpu_var(softnet_data);
3250 fl = rcu_dereference(sd->flow_limit);
3252 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3253 old_flow = fl->history[fl->history_head];
3254 fl->history[fl->history_head] = new_flow;
3257 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3259 if (likely(fl->buckets[old_flow]))
3260 fl->buckets[old_flow]--;
3262 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3274 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3275 * queue (may be a remote CPU queue).
3277 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3278 unsigned int *qtail)
3280 struct softnet_data *sd;
3281 unsigned long flags;
3284 sd = &per_cpu(softnet_data, cpu);
3286 local_irq_save(flags);
3289 qlen = skb_queue_len(&sd->input_pkt_queue);
3290 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3291 if (skb_queue_len(&sd->input_pkt_queue)) {
3293 __skb_queue_tail(&sd->input_pkt_queue, skb);
3294 input_queue_tail_incr_save(sd, qtail);
3296 local_irq_restore(flags);
3297 return NET_RX_SUCCESS;
3300 /* Schedule NAPI for backlog device
3301 * We can use non atomic operation since we own the queue lock
3303 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3304 if (!rps_ipi_queued(sd))
3305 ____napi_schedule(sd, &sd->backlog);
3313 local_irq_restore(flags);
3315 atomic_long_inc(&skb->dev->rx_dropped);
3320 static int netif_rx_internal(struct sk_buff *skb)
3324 net_timestamp_check(netdev_tstamp_prequeue, skb);
3326 trace_netif_rx(skb);
3328 if (static_key_false(&rps_needed)) {
3329 struct rps_dev_flow voidflow, *rflow = &voidflow;
3335 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3337 cpu = smp_processor_id();
3339 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3347 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3354 * netif_rx - post buffer to the network code
3355 * @skb: buffer to post
3357 * This function receives a packet from a device driver and queues it for
3358 * the upper (protocol) levels to process. It always succeeds. The buffer
3359 * may be dropped during processing for congestion control or by the
3363 * NET_RX_SUCCESS (no congestion)
3364 * NET_RX_DROP (packet was dropped)
3368 int netif_rx(struct sk_buff *skb)
3370 trace_netif_rx_entry(skb);
3372 return netif_rx_internal(skb);
3374 EXPORT_SYMBOL(netif_rx);
3376 int netif_rx_ni(struct sk_buff *skb)
3380 trace_netif_rx_ni_entry(skb);
3383 err = netif_rx_internal(skb);
3384 if (local_softirq_pending())
3390 EXPORT_SYMBOL(netif_rx_ni);
3392 static void net_tx_action(struct softirq_action *h)
3394 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3396 if (sd->completion_queue) {
3397 struct sk_buff *clist;
3399 local_irq_disable();
3400 clist = sd->completion_queue;
3401 sd->completion_queue = NULL;
3405 struct sk_buff *skb = clist;
3406 clist = clist->next;
3408 WARN_ON(atomic_read(&skb->users));
3409 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3410 trace_consume_skb(skb);
3412 trace_kfree_skb(skb, net_tx_action);
3417 if (sd->output_queue) {
3420 local_irq_disable();
3421 head = sd->output_queue;
3422 sd->output_queue = NULL;
3423 sd->output_queue_tailp = &sd->output_queue;
3427 struct Qdisc *q = head;
3428 spinlock_t *root_lock;
3430 head = head->next_sched;
3432 root_lock = qdisc_lock(q);
3433 if (spin_trylock(root_lock)) {
3434 smp_mb__before_atomic();
3435 clear_bit(__QDISC_STATE_SCHED,
3438 spin_unlock(root_lock);
3440 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3442 __netif_reschedule(q);
3444 smp_mb__before_atomic();
3445 clear_bit(__QDISC_STATE_SCHED,
3453 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3454 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3455 /* This hook is defined here for ATM LANE */
3456 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3457 unsigned char *addr) __read_mostly;
3458 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3461 #ifdef CONFIG_NET_CLS_ACT
3462 /* TODO: Maybe we should just force sch_ingress to be compiled in
3463 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3464 * a compare and 2 stores extra right now if we dont have it on
3465 * but have CONFIG_NET_CLS_ACT
3466 * NOTE: This doesn't stop any functionality; if you dont have
3467 * the ingress scheduler, you just can't add policies on ingress.
3470 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3472 struct net_device *dev = skb->dev;
3473 u32 ttl = G_TC_RTTL(skb->tc_verd);
3474 int result = TC_ACT_OK;
3477 if (unlikely(MAX_RED_LOOP < ttl++)) {
3478 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3479 skb->skb_iif, dev->ifindex);
3483 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3484 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3486 q = rcu_dereference(rxq->qdisc);
3487 if (q != &noop_qdisc) {
3488 spin_lock(qdisc_lock(q));
3489 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3490 result = qdisc_enqueue_root(skb, q);
3491 spin_unlock(qdisc_lock(q));
3497 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3498 struct packet_type **pt_prev,
3499 int *ret, struct net_device *orig_dev)
3501 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3503 if (!rxq || rcu_access_pointer(rxq->qdisc) == &noop_qdisc)
3507 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3511 switch (ing_filter(skb, rxq)) {
3525 * netdev_rx_handler_register - register receive handler
3526 * @dev: device to register a handler for
3527 * @rx_handler: receive handler to register
3528 * @rx_handler_data: data pointer that is used by rx handler
3530 * Register a receive handler for a device. This handler will then be
3531 * called from __netif_receive_skb. A negative errno code is returned
3534 * The caller must hold the rtnl_mutex.
3536 * For a general description of rx_handler, see enum rx_handler_result.
3538 int netdev_rx_handler_register(struct net_device *dev,
3539 rx_handler_func_t *rx_handler,
3540 void *rx_handler_data)
3544 if (dev->rx_handler)
3547 /* Note: rx_handler_data must be set before rx_handler */
3548 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3549 rcu_assign_pointer(dev->rx_handler, rx_handler);
3553 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3556 * netdev_rx_handler_unregister - unregister receive handler
3557 * @dev: device to unregister a handler from
3559 * Unregister a receive handler from a device.
3561 * The caller must hold the rtnl_mutex.
3563 void netdev_rx_handler_unregister(struct net_device *dev)
3567 RCU_INIT_POINTER(dev->rx_handler, NULL);
3568 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3569 * section has a guarantee to see a non NULL rx_handler_data
3573 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3575 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3578 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3579 * the special handling of PFMEMALLOC skbs.
3581 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3583 switch (skb->protocol) {
3584 case htons(ETH_P_ARP):
3585 case htons(ETH_P_IP):
3586 case htons(ETH_P_IPV6):
3587 case htons(ETH_P_8021Q):
3588 case htons(ETH_P_8021AD):
3595 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3597 struct packet_type *ptype, *pt_prev;
3598 rx_handler_func_t *rx_handler;
3599 struct net_device *orig_dev;
3600 struct net_device *null_or_dev;
3601 bool deliver_exact = false;
3602 int ret = NET_RX_DROP;
3605 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3607 trace_netif_receive_skb(skb);
3609 orig_dev = skb->dev;
3611 skb_reset_network_header(skb);
3612 if (!skb_transport_header_was_set(skb))
3613 skb_reset_transport_header(skb);
3614 skb_reset_mac_len(skb);
3621 skb->skb_iif = skb->dev->ifindex;
3623 __this_cpu_inc(softnet_data.processed);
3625 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3626 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3627 skb = skb_vlan_untag(skb);
3632 #ifdef CONFIG_NET_CLS_ACT
3633 if (skb->tc_verd & TC_NCLS) {
3634 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3642 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3643 if (!ptype->dev || ptype->dev == skb->dev) {
3645 ret = deliver_skb(skb, pt_prev, orig_dev);
3651 #ifdef CONFIG_NET_CLS_ACT
3652 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3658 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3661 if (vlan_tx_tag_present(skb)) {
3663 ret = deliver_skb(skb, pt_prev, orig_dev);
3666 if (vlan_do_receive(&skb))
3668 else if (unlikely(!skb))
3672 rx_handler = rcu_dereference(skb->dev->rx_handler);
3675 ret = deliver_skb(skb, pt_prev, orig_dev);
3678 switch (rx_handler(&skb)) {
3679 case RX_HANDLER_CONSUMED:
3680 ret = NET_RX_SUCCESS;
3682 case RX_HANDLER_ANOTHER:
3684 case RX_HANDLER_EXACT:
3685 deliver_exact = true;
3686 case RX_HANDLER_PASS:
3693 if (unlikely(vlan_tx_tag_present(skb))) {
3694 if (vlan_tx_tag_get_id(skb))
3695 skb->pkt_type = PACKET_OTHERHOST;
3696 /* Note: we might in the future use prio bits
3697 * and set skb->priority like in vlan_do_receive()
3698 * For the time being, just ignore Priority Code Point
3703 /* deliver only exact match when indicated */
3704 null_or_dev = deliver_exact ? skb->dev : NULL;
3706 type = skb->protocol;
3707 list_for_each_entry_rcu(ptype,
3708 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3709 if (ptype->type == type &&
3710 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3711 ptype->dev == orig_dev)) {
3713 ret = deliver_skb(skb, pt_prev, orig_dev);
3719 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3722 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3725 atomic_long_inc(&skb->dev->rx_dropped);
3727 /* Jamal, now you will not able to escape explaining
3728 * me how you were going to use this. :-)
3738 static int __netif_receive_skb(struct sk_buff *skb)
3742 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3743 unsigned long pflags = current->flags;
3746 * PFMEMALLOC skbs are special, they should
3747 * - be delivered to SOCK_MEMALLOC sockets only
3748 * - stay away from userspace
3749 * - have bounded memory usage
3751 * Use PF_MEMALLOC as this saves us from propagating the allocation
3752 * context down to all allocation sites.
3754 current->flags |= PF_MEMALLOC;
3755 ret = __netif_receive_skb_core(skb, true);
3756 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3758 ret = __netif_receive_skb_core(skb, false);
3763 static int netif_receive_skb_internal(struct sk_buff *skb)
3765 net_timestamp_check(netdev_tstamp_prequeue, skb);
3767 if (skb_defer_rx_timestamp(skb))
3768 return NET_RX_SUCCESS;
3771 if (static_key_false(&rps_needed)) {
3772 struct rps_dev_flow voidflow, *rflow = &voidflow;
3777 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3780 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3787 return __netif_receive_skb(skb);
3791 * netif_receive_skb - process receive buffer from network
3792 * @skb: buffer to process
3794 * netif_receive_skb() is the main receive data processing function.
3795 * It always succeeds. The buffer may be dropped during processing
3796 * for congestion control or by the protocol layers.
3798 * This function may only be called from softirq context and interrupts
3799 * should be enabled.
3801 * Return values (usually ignored):
3802 * NET_RX_SUCCESS: no congestion
3803 * NET_RX_DROP: packet was dropped
3805 int netif_receive_skb(struct sk_buff *skb)
3807 trace_netif_receive_skb_entry(skb);
3809 return netif_receive_skb_internal(skb);
3811 EXPORT_SYMBOL(netif_receive_skb);
3813 /* Network device is going away, flush any packets still pending
3814 * Called with irqs disabled.
3816 static void flush_backlog(void *arg)
3818 struct net_device *dev = arg;
3819 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3820 struct sk_buff *skb, *tmp;
3823 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3824 if (skb->dev == dev) {
3825 __skb_unlink(skb, &sd->input_pkt_queue);
3827 input_queue_head_incr(sd);
3832 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3833 if (skb->dev == dev) {
3834 __skb_unlink(skb, &sd->process_queue);
3836 input_queue_head_incr(sd);
3841 static int napi_gro_complete(struct sk_buff *skb)
3843 struct packet_offload *ptype;
3844 __be16 type = skb->protocol;
3845 struct list_head *head = &offload_base;
3848 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3850 if (NAPI_GRO_CB(skb)->count == 1) {
3851 skb_shinfo(skb)->gso_size = 0;
3856 list_for_each_entry_rcu(ptype, head, list) {
3857 if (ptype->type != type || !ptype->callbacks.gro_complete)
3860 err = ptype->callbacks.gro_complete(skb, 0);
3866 WARN_ON(&ptype->list == head);
3868 return NET_RX_SUCCESS;
3872 return netif_receive_skb_internal(skb);
3875 /* napi->gro_list contains packets ordered by age.
3876 * youngest packets at the head of it.
3877 * Complete skbs in reverse order to reduce latencies.
3879 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3881 struct sk_buff *skb, *prev = NULL;
3883 /* scan list and build reverse chain */
3884 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3889 for (skb = prev; skb; skb = prev) {
3892 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3896 napi_gro_complete(skb);
3900 napi->gro_list = NULL;
3902 EXPORT_SYMBOL(napi_gro_flush);
3904 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3907 unsigned int maclen = skb->dev->hard_header_len;
3908 u32 hash = skb_get_hash_raw(skb);
3910 for (p = napi->gro_list; p; p = p->next) {
3911 unsigned long diffs;
3913 NAPI_GRO_CB(p)->flush = 0;
3915 if (hash != skb_get_hash_raw(p)) {
3916 NAPI_GRO_CB(p)->same_flow = 0;
3920 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3921 diffs |= p->vlan_tci ^ skb->vlan_tci;
3922 if (maclen == ETH_HLEN)
3923 diffs |= compare_ether_header(skb_mac_header(p),
3924 skb_mac_header(skb));
3926 diffs = memcmp(skb_mac_header(p),
3927 skb_mac_header(skb),
3929 NAPI_GRO_CB(p)->same_flow = !diffs;
3933 static void skb_gro_reset_offset(struct sk_buff *skb)
3935 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3936 const skb_frag_t *frag0 = &pinfo->frags[0];
3938 NAPI_GRO_CB(skb)->data_offset = 0;
3939 NAPI_GRO_CB(skb)->frag0 = NULL;
3940 NAPI_GRO_CB(skb)->frag0_len = 0;
3942 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3944 !PageHighMem(skb_frag_page(frag0))) {
3945 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3946 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3950 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
3952 struct skb_shared_info *pinfo = skb_shinfo(skb);
3954 BUG_ON(skb->end - skb->tail < grow);
3956 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3958 skb->data_len -= grow;
3961 pinfo->frags[0].page_offset += grow;
3962 skb_frag_size_sub(&pinfo->frags[0], grow);
3964 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
3965 skb_frag_unref(skb, 0);
3966 memmove(pinfo->frags, pinfo->frags + 1,
3967 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
3971 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3973 struct sk_buff **pp = NULL;
3974 struct packet_offload *ptype;
3975 __be16 type = skb->protocol;
3976 struct list_head *head = &offload_base;
3978 enum gro_result ret;
3981 if (!(skb->dev->features & NETIF_F_GRO))
3984 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
3987 gro_list_prepare(napi, skb);
3990 list_for_each_entry_rcu(ptype, head, list) {
3991 if (ptype->type != type || !ptype->callbacks.gro_receive)
3994 skb_set_network_header(skb, skb_gro_offset(skb));
3995 skb_reset_mac_len(skb);
3996 NAPI_GRO_CB(skb)->same_flow = 0;
3997 NAPI_GRO_CB(skb)->flush = 0;
3998 NAPI_GRO_CB(skb)->free = 0;
3999 NAPI_GRO_CB(skb)->udp_mark = 0;
4001 /* Setup for GRO checksum validation */
4002 switch (skb->ip_summed) {
4003 case CHECKSUM_COMPLETE:
4004 NAPI_GRO_CB(skb)->csum = skb->csum;
4005 NAPI_GRO_CB(skb)->csum_valid = 1;
4006 NAPI_GRO_CB(skb)->csum_cnt = 0;
4008 case CHECKSUM_UNNECESSARY:
4009 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4010 NAPI_GRO_CB(skb)->csum_valid = 0;
4013 NAPI_GRO_CB(skb)->csum_cnt = 0;
4014 NAPI_GRO_CB(skb)->csum_valid = 0;
4017 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4022 if (&ptype->list == head)
4025 same_flow = NAPI_GRO_CB(skb)->same_flow;
4026 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4029 struct sk_buff *nskb = *pp;
4033 napi_gro_complete(nskb);
4040 if (NAPI_GRO_CB(skb)->flush)
4043 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4044 struct sk_buff *nskb = napi->gro_list;
4046 /* locate the end of the list to select the 'oldest' flow */
4047 while (nskb->next) {
4053 napi_gro_complete(nskb);
4057 NAPI_GRO_CB(skb)->count = 1;
4058 NAPI_GRO_CB(skb)->age = jiffies;
4059 NAPI_GRO_CB(skb)->last = skb;
4060 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4061 skb->next = napi->gro_list;
4062 napi->gro_list = skb;
4066 grow = skb_gro_offset(skb) - skb_headlen(skb);
4068 gro_pull_from_frag0(skb, grow);
4077 struct packet_offload *gro_find_receive_by_type(__be16 type)
4079 struct list_head *offload_head = &offload_base;
4080 struct packet_offload *ptype;
4082 list_for_each_entry_rcu(ptype, offload_head, list) {
4083 if (ptype->type != type || !ptype->callbacks.gro_receive)
4089 EXPORT_SYMBOL(gro_find_receive_by_type);
4091 struct packet_offload *gro_find_complete_by_type(__be16 type)
4093 struct list_head *offload_head = &offload_base;
4094 struct packet_offload *ptype;
4096 list_for_each_entry_rcu(ptype, offload_head, list) {
4097 if (ptype->type != type || !ptype->callbacks.gro_complete)
4103 EXPORT_SYMBOL(gro_find_complete_by_type);
4105 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4109 if (netif_receive_skb_internal(skb))
4117 case GRO_MERGED_FREE:
4118 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4119 kmem_cache_free(skbuff_head_cache, skb);
4132 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4134 trace_napi_gro_receive_entry(skb);
4136 skb_gro_reset_offset(skb);
4138 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4140 EXPORT_SYMBOL(napi_gro_receive);
4142 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4144 __skb_pull(skb, skb_headlen(skb));
4145 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4146 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4148 skb->dev = napi->dev;
4150 skb->encapsulation = 0;
4151 skb_shinfo(skb)->gso_type = 0;
4152 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4157 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4159 struct sk_buff *skb = napi->skb;
4162 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4167 EXPORT_SYMBOL(napi_get_frags);
4169 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4170 struct sk_buff *skb,
4176 __skb_push(skb, ETH_HLEN);
4177 skb->protocol = eth_type_trans(skb, skb->dev);
4178 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4183 case GRO_MERGED_FREE:
4184 napi_reuse_skb(napi, skb);
4194 /* Upper GRO stack assumes network header starts at gro_offset=0
4195 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4196 * We copy ethernet header into skb->data to have a common layout.
4198 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4200 struct sk_buff *skb = napi->skb;
4201 const struct ethhdr *eth;
4202 unsigned int hlen = sizeof(*eth);
4206 skb_reset_mac_header(skb);
4207 skb_gro_reset_offset(skb);
4209 eth = skb_gro_header_fast(skb, 0);
4210 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4211 eth = skb_gro_header_slow(skb, hlen, 0);
4212 if (unlikely(!eth)) {
4213 napi_reuse_skb(napi, skb);
4217 gro_pull_from_frag0(skb, hlen);
4218 NAPI_GRO_CB(skb)->frag0 += hlen;
4219 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4221 __skb_pull(skb, hlen);
4224 * This works because the only protocols we care about don't require
4226 * We'll fix it up properly in napi_frags_finish()
4228 skb->protocol = eth->h_proto;
4233 gro_result_t napi_gro_frags(struct napi_struct *napi)
4235 struct sk_buff *skb = napi_frags_skb(napi);
4240 trace_napi_gro_frags_entry(skb);
4242 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4244 EXPORT_SYMBOL(napi_gro_frags);
4246 /* Compute the checksum from gro_offset and return the folded value
4247 * after adding in any pseudo checksum.
4249 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4254 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4256 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4257 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4259 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4260 !skb->csum_complete_sw)
4261 netdev_rx_csum_fault(skb->dev);
4264 NAPI_GRO_CB(skb)->csum = wsum;
4265 NAPI_GRO_CB(skb)->csum_valid = 1;
4269 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4272 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4273 * Note: called with local irq disabled, but exits with local irq enabled.
4275 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4278 struct softnet_data *remsd = sd->rps_ipi_list;
4281 sd->rps_ipi_list = NULL;
4285 /* Send pending IPI's to kick RPS processing on remote cpus. */
4287 struct softnet_data *next = remsd->rps_ipi_next;
4289 if (cpu_online(remsd->cpu))
4290 smp_call_function_single_async(remsd->cpu,
4299 static int process_backlog(struct napi_struct *napi, int quota)
4302 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4305 /* Check if we have pending ipi, its better to send them now,
4306 * not waiting net_rx_action() end.
4308 if (sd->rps_ipi_list) {
4309 local_irq_disable();
4310 net_rps_action_and_irq_enable(sd);
4313 napi->weight = weight_p;
4314 local_irq_disable();
4316 struct sk_buff *skb;
4318 while ((skb = __skb_dequeue(&sd->process_queue))) {
4320 __netif_receive_skb(skb);
4321 local_irq_disable();
4322 input_queue_head_incr(sd);
4323 if (++work >= quota) {
4330 if (skb_queue_empty(&sd->input_pkt_queue)) {
4332 * Inline a custom version of __napi_complete().
4333 * only current cpu owns and manipulates this napi,
4334 * and NAPI_STATE_SCHED is the only possible flag set
4336 * We can use a plain write instead of clear_bit(),
4337 * and we dont need an smp_mb() memory barrier.
4339 list_del(&napi->poll_list);
4346 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4347 &sd->process_queue);
4356 * __napi_schedule - schedule for receive
4357 * @n: entry to schedule
4359 * The entry's receive function will be scheduled to run
4361 void __napi_schedule(struct napi_struct *n)
4363 unsigned long flags;
4365 local_irq_save(flags);
4366 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4367 local_irq_restore(flags);
4369 EXPORT_SYMBOL(__napi_schedule);
4371 void __napi_complete(struct napi_struct *n)
4373 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4374 BUG_ON(n->gro_list);
4376 list_del(&n->poll_list);
4377 smp_mb__before_atomic();
4378 clear_bit(NAPI_STATE_SCHED, &n->state);
4380 EXPORT_SYMBOL(__napi_complete);
4382 void napi_complete(struct napi_struct *n)
4384 unsigned long flags;
4387 * don't let napi dequeue from the cpu poll list
4388 * just in case its running on a different cpu
4390 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4393 napi_gro_flush(n, false);
4394 local_irq_save(flags);
4396 local_irq_restore(flags);
4398 EXPORT_SYMBOL(napi_complete);
4400 /* must be called under rcu_read_lock(), as we dont take a reference */
4401 struct napi_struct *napi_by_id(unsigned int napi_id)
4403 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4404 struct napi_struct *napi;
4406 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4407 if (napi->napi_id == napi_id)
4412 EXPORT_SYMBOL_GPL(napi_by_id);
4414 void napi_hash_add(struct napi_struct *napi)
4416 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4418 spin_lock(&napi_hash_lock);
4420 /* 0 is not a valid id, we also skip an id that is taken
4421 * we expect both events to be extremely rare
4424 while (!napi->napi_id) {
4425 napi->napi_id = ++napi_gen_id;
4426 if (napi_by_id(napi->napi_id))
4430 hlist_add_head_rcu(&napi->napi_hash_node,
4431 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4433 spin_unlock(&napi_hash_lock);
4436 EXPORT_SYMBOL_GPL(napi_hash_add);
4438 /* Warning : caller is responsible to make sure rcu grace period
4439 * is respected before freeing memory containing @napi
4441 void napi_hash_del(struct napi_struct *napi)
4443 spin_lock(&napi_hash_lock);
4445 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4446 hlist_del_rcu(&napi->napi_hash_node);
4448 spin_unlock(&napi_hash_lock);
4450 EXPORT_SYMBOL_GPL(napi_hash_del);
4452 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4453 int (*poll)(struct napi_struct *, int), int weight)
4455 INIT_LIST_HEAD(&napi->poll_list);
4456 napi->gro_count = 0;
4457 napi->gro_list = NULL;
4460 if (weight > NAPI_POLL_WEIGHT)
4461 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4463 napi->weight = weight;
4464 list_add(&napi->dev_list, &dev->napi_list);
4466 #ifdef CONFIG_NETPOLL
4467 spin_lock_init(&napi->poll_lock);
4468 napi->poll_owner = -1;
4470 set_bit(NAPI_STATE_SCHED, &napi->state);
4472 EXPORT_SYMBOL(netif_napi_add);
4474 void netif_napi_del(struct napi_struct *napi)
4476 list_del_init(&napi->dev_list);
4477 napi_free_frags(napi);
4479 kfree_skb_list(napi->gro_list);
4480 napi->gro_list = NULL;
4481 napi->gro_count = 0;
4483 EXPORT_SYMBOL(netif_napi_del);
4485 static void net_rx_action(struct softirq_action *h)
4487 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4488 unsigned long time_limit = jiffies + 2;
4489 int budget = netdev_budget;
4492 local_irq_disable();
4494 while (!list_empty(&sd->poll_list)) {
4495 struct napi_struct *n;
4498 /* If softirq window is exhuasted then punt.
4499 * Allow this to run for 2 jiffies since which will allow
4500 * an average latency of 1.5/HZ.
4502 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4507 /* Even though interrupts have been re-enabled, this
4508 * access is safe because interrupts can only add new
4509 * entries to the tail of this list, and only ->poll()
4510 * calls can remove this head entry from the list.
4512 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4514 have = netpoll_poll_lock(n);
4518 /* This NAPI_STATE_SCHED test is for avoiding a race
4519 * with netpoll's poll_napi(). Only the entity which
4520 * obtains the lock and sees NAPI_STATE_SCHED set will
4521 * actually make the ->poll() call. Therefore we avoid
4522 * accidentally calling ->poll() when NAPI is not scheduled.
4525 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4526 work = n->poll(n, weight);
4530 WARN_ON_ONCE(work > weight);
4534 local_irq_disable();
4536 /* Drivers must not modify the NAPI state if they
4537 * consume the entire weight. In such cases this code
4538 * still "owns" the NAPI instance and therefore can
4539 * move the instance around on the list at-will.
4541 if (unlikely(work == weight)) {
4542 if (unlikely(napi_disable_pending(n))) {
4545 local_irq_disable();
4548 /* flush too old packets
4549 * If HZ < 1000, flush all packets.
4552 napi_gro_flush(n, HZ >= 1000);
4553 local_irq_disable();
4555 list_move_tail(&n->poll_list, &sd->poll_list);
4559 netpoll_poll_unlock(have);
4562 net_rps_action_and_irq_enable(sd);
4564 #ifdef CONFIG_NET_DMA
4566 * There may not be any more sk_buffs coming right now, so push
4567 * any pending DMA copies to hardware
4569 dma_issue_pending_all();
4576 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4580 struct netdev_adjacent {
4581 struct net_device *dev;
4583 /* upper master flag, there can only be one master device per list */
4586 /* counter for the number of times this device was added to us */
4589 /* private field for the users */
4592 struct list_head list;
4593 struct rcu_head rcu;
4596 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4597 struct net_device *adj_dev,
4598 struct list_head *adj_list)
4600 struct netdev_adjacent *adj;
4602 list_for_each_entry(adj, adj_list, list) {
4603 if (adj->dev == adj_dev)
4610 * netdev_has_upper_dev - Check if device is linked to an upper device
4612 * @upper_dev: upper device to check
4614 * Find out if a device is linked to specified upper device and return true
4615 * in case it is. Note that this checks only immediate upper device,
4616 * not through a complete stack of devices. The caller must hold the RTNL lock.
4618 bool netdev_has_upper_dev(struct net_device *dev,
4619 struct net_device *upper_dev)
4623 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4625 EXPORT_SYMBOL(netdev_has_upper_dev);
4628 * netdev_has_any_upper_dev - Check if device is linked to some device
4631 * Find out if a device is linked to an upper device and return true in case
4632 * it is. The caller must hold the RTNL lock.
4634 static bool netdev_has_any_upper_dev(struct net_device *dev)
4638 return !list_empty(&dev->all_adj_list.upper);
4642 * netdev_master_upper_dev_get - Get master upper device
4645 * Find a master upper device and return pointer to it or NULL in case
4646 * it's not there. The caller must hold the RTNL lock.
4648 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4650 struct netdev_adjacent *upper;
4654 if (list_empty(&dev->adj_list.upper))
4657 upper = list_first_entry(&dev->adj_list.upper,
4658 struct netdev_adjacent, list);
4659 if (likely(upper->master))
4663 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4665 void *netdev_adjacent_get_private(struct list_head *adj_list)
4667 struct netdev_adjacent *adj;
4669 adj = list_entry(adj_list, struct netdev_adjacent, list);
4671 return adj->private;
4673 EXPORT_SYMBOL(netdev_adjacent_get_private);
4676 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4678 * @iter: list_head ** of the current position
4680 * Gets the next device from the dev's upper list, starting from iter
4681 * position. The caller must hold RCU read lock.
4683 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4684 struct list_head **iter)
4686 struct netdev_adjacent *upper;
4688 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4690 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4692 if (&upper->list == &dev->adj_list.upper)
4695 *iter = &upper->list;
4699 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4702 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4704 * @iter: list_head ** of the current position
4706 * Gets the next device from the dev's upper list, starting from iter
4707 * position. The caller must hold RCU read lock.
4709 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4710 struct list_head **iter)
4712 struct netdev_adjacent *upper;
4714 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4716 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4718 if (&upper->list == &dev->all_adj_list.upper)
4721 *iter = &upper->list;
4725 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4728 * netdev_lower_get_next_private - Get the next ->private from the
4729 * lower neighbour list
4731 * @iter: list_head ** of the current position
4733 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4734 * list, starting from iter position. The caller must hold either hold the
4735 * RTNL lock or its own locking that guarantees that the neighbour lower
4736 * list will remain unchainged.
4738 void *netdev_lower_get_next_private(struct net_device *dev,
4739 struct list_head **iter)
4741 struct netdev_adjacent *lower;
4743 lower = list_entry(*iter, struct netdev_adjacent, list);
4745 if (&lower->list == &dev->adj_list.lower)
4748 *iter = lower->list.next;
4750 return lower->private;
4752 EXPORT_SYMBOL(netdev_lower_get_next_private);
4755 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4756 * lower neighbour list, RCU
4759 * @iter: list_head ** of the current position
4761 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4762 * list, starting from iter position. The caller must hold RCU read lock.
4764 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4765 struct list_head **iter)
4767 struct netdev_adjacent *lower;
4769 WARN_ON_ONCE(!rcu_read_lock_held());
4771 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4773 if (&lower->list == &dev->adj_list.lower)
4776 *iter = &lower->list;
4778 return lower->private;
4780 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4783 * netdev_lower_get_next - Get the next device from the lower neighbour
4786 * @iter: list_head ** of the current position
4788 * Gets the next netdev_adjacent from the dev's lower neighbour
4789 * list, starting from iter position. The caller must hold RTNL lock or
4790 * its own locking that guarantees that the neighbour lower
4791 * list will remain unchainged.
4793 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4795 struct netdev_adjacent *lower;
4797 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4799 if (&lower->list == &dev->adj_list.lower)
4802 *iter = &lower->list;
4806 EXPORT_SYMBOL(netdev_lower_get_next);
4809 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4810 * lower neighbour list, RCU
4814 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4815 * list. The caller must hold RCU read lock.
4817 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4819 struct netdev_adjacent *lower;
4821 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4822 struct netdev_adjacent, list);
4824 return lower->private;
4827 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4830 * netdev_master_upper_dev_get_rcu - Get master upper device
4833 * Find a master upper device and return pointer to it or NULL in case
4834 * it's not there. The caller must hold the RCU read lock.
4836 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4838 struct netdev_adjacent *upper;
4840 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4841 struct netdev_adjacent, list);
4842 if (upper && likely(upper->master))
4846 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4848 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4849 struct net_device *adj_dev,
4850 struct list_head *dev_list)
4852 char linkname[IFNAMSIZ+7];
4853 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4854 "upper_%s" : "lower_%s", adj_dev->name);
4855 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4858 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4860 struct list_head *dev_list)
4862 char linkname[IFNAMSIZ+7];
4863 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4864 "upper_%s" : "lower_%s", name);
4865 sysfs_remove_link(&(dev->dev.kobj), linkname);
4868 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4869 (dev_list == &dev->adj_list.upper || \
4870 dev_list == &dev->adj_list.lower)
4872 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4873 struct net_device *adj_dev,
4874 struct list_head *dev_list,
4875 void *private, bool master)
4877 struct netdev_adjacent *adj;
4880 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4887 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4892 adj->master = master;
4894 adj->private = private;
4897 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4898 adj_dev->name, dev->name, adj_dev->name);
4900 if (netdev_adjacent_is_neigh_list(dev, dev_list)) {
4901 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
4906 /* Ensure that master link is always the first item in list. */
4908 ret = sysfs_create_link(&(dev->dev.kobj),
4909 &(adj_dev->dev.kobj), "master");
4911 goto remove_symlinks;
4913 list_add_rcu(&adj->list, dev_list);
4915 list_add_tail_rcu(&adj->list, dev_list);
4921 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4922 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4930 static void __netdev_adjacent_dev_remove(struct net_device *dev,
4931 struct net_device *adj_dev,
4932 struct list_head *dev_list)
4934 struct netdev_adjacent *adj;
4936 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4939 pr_err("tried to remove device %s from %s\n",
4940 dev->name, adj_dev->name);
4944 if (adj->ref_nr > 1) {
4945 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4952 sysfs_remove_link(&(dev->dev.kobj), "master");
4954 if (netdev_adjacent_is_neigh_list(dev, dev_list) &&
4955 net_eq(dev_net(dev),dev_net(adj_dev)))
4956 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4958 list_del_rcu(&adj->list);
4959 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4960 adj_dev->name, dev->name, adj_dev->name);
4962 kfree_rcu(adj, rcu);
4965 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4966 struct net_device *upper_dev,
4967 struct list_head *up_list,
4968 struct list_head *down_list,
4969 void *private, bool master)
4973 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4978 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4981 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4988 static int __netdev_adjacent_dev_link(struct net_device *dev,
4989 struct net_device *upper_dev)
4991 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4992 &dev->all_adj_list.upper,
4993 &upper_dev->all_adj_list.lower,
4997 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4998 struct net_device *upper_dev,
4999 struct list_head *up_list,
5000 struct list_head *down_list)
5002 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5003 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5006 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5007 struct net_device *upper_dev)
5009 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5010 &dev->all_adj_list.upper,
5011 &upper_dev->all_adj_list.lower);
5014 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5015 struct net_device *upper_dev,
5016 void *private, bool master)
5018 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5023 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5024 &dev->adj_list.upper,
5025 &upper_dev->adj_list.lower,
5028 __netdev_adjacent_dev_unlink(dev, upper_dev);
5035 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5036 struct net_device *upper_dev)
5038 __netdev_adjacent_dev_unlink(dev, upper_dev);
5039 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5040 &dev->adj_list.upper,
5041 &upper_dev->adj_list.lower);
5044 static int __netdev_upper_dev_link(struct net_device *dev,
5045 struct net_device *upper_dev, bool master,
5048 struct netdev_adjacent *i, *j, *to_i, *to_j;
5053 if (dev == upper_dev)
5056 /* To prevent loops, check if dev is not upper device to upper_dev. */
5057 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5060 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
5063 if (master && netdev_master_upper_dev_get(dev))
5066 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5071 /* Now that we linked these devs, make all the upper_dev's
5072 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5073 * versa, and don't forget the devices itself. All of these
5074 * links are non-neighbours.
5076 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5077 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5078 pr_debug("Interlinking %s with %s, non-neighbour\n",
5079 i->dev->name, j->dev->name);
5080 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5086 /* add dev to every upper_dev's upper device */
5087 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5088 pr_debug("linking %s's upper device %s with %s\n",
5089 upper_dev->name, i->dev->name, dev->name);
5090 ret = __netdev_adjacent_dev_link(dev, i->dev);
5092 goto rollback_upper_mesh;
5095 /* add upper_dev to every dev's lower device */
5096 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5097 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5098 i->dev->name, upper_dev->name);
5099 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5101 goto rollback_lower_mesh;
5104 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5107 rollback_lower_mesh:
5109 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5112 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5117 rollback_upper_mesh:
5119 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5122 __netdev_adjacent_dev_unlink(dev, i->dev);
5130 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5131 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5132 if (i == to_i && j == to_j)
5134 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5140 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5146 * netdev_upper_dev_link - Add a link to the upper device
5148 * @upper_dev: new upper device
5150 * Adds a link to device which is upper to this one. The caller must hold
5151 * the RTNL lock. On a failure a negative errno code is returned.
5152 * On success the reference counts are adjusted and the function
5155 int netdev_upper_dev_link(struct net_device *dev,
5156 struct net_device *upper_dev)
5158 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5160 EXPORT_SYMBOL(netdev_upper_dev_link);
5163 * netdev_master_upper_dev_link - Add a master link to the upper device
5165 * @upper_dev: new upper device
5167 * Adds a link to device which is upper to this one. In this case, only
5168 * one master upper device can be linked, although other non-master devices
5169 * might be linked as well. The caller must hold the RTNL lock.
5170 * On a failure a negative errno code is returned. On success the reference
5171 * counts are adjusted and the function returns zero.
5173 int netdev_master_upper_dev_link(struct net_device *dev,
5174 struct net_device *upper_dev)
5176 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5178 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5180 int netdev_master_upper_dev_link_private(struct net_device *dev,
5181 struct net_device *upper_dev,
5184 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5186 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5189 * netdev_upper_dev_unlink - Removes a link to upper device
5191 * @upper_dev: new upper device
5193 * Removes a link to device which is upper to this one. The caller must hold
5196 void netdev_upper_dev_unlink(struct net_device *dev,
5197 struct net_device *upper_dev)
5199 struct netdev_adjacent *i, *j;
5202 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5204 /* Here is the tricky part. We must remove all dev's lower
5205 * devices from all upper_dev's upper devices and vice
5206 * versa, to maintain the graph relationship.
5208 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5209 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5210 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5212 /* remove also the devices itself from lower/upper device
5215 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5216 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5218 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5219 __netdev_adjacent_dev_unlink(dev, i->dev);
5221 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5223 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5225 void netdev_adjacent_add_links(struct net_device *dev)
5227 struct netdev_adjacent *iter;
5229 struct net *net = dev_net(dev);
5231 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5232 if (!net_eq(net,dev_net(iter->dev)))
5234 netdev_adjacent_sysfs_add(iter->dev, dev,
5235 &iter->dev->adj_list.lower);
5236 netdev_adjacent_sysfs_add(dev, iter->dev,
5237 &dev->adj_list.upper);
5240 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5241 if (!net_eq(net,dev_net(iter->dev)))
5243 netdev_adjacent_sysfs_add(iter->dev, dev,
5244 &iter->dev->adj_list.upper);
5245 netdev_adjacent_sysfs_add(dev, iter->dev,
5246 &dev->adj_list.lower);
5250 void netdev_adjacent_del_links(struct net_device *dev)
5252 struct netdev_adjacent *iter;
5254 struct net *net = dev_net(dev);
5256 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5257 if (!net_eq(net,dev_net(iter->dev)))
5259 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5260 &iter->dev->adj_list.lower);
5261 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5262 &dev->adj_list.upper);
5265 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5266 if (!net_eq(net,dev_net(iter->dev)))
5268 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5269 &iter->dev->adj_list.upper);
5270 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5271 &dev->adj_list.lower);
5275 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5277 struct netdev_adjacent *iter;
5279 struct net *net = dev_net(dev);
5281 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5282 if (!net_eq(net,dev_net(iter->dev)))
5284 netdev_adjacent_sysfs_del(iter->dev, oldname,
5285 &iter->dev->adj_list.lower);
5286 netdev_adjacent_sysfs_add(iter->dev, dev,
5287 &iter->dev->adj_list.lower);
5290 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5291 if (!net_eq(net,dev_net(iter->dev)))
5293 netdev_adjacent_sysfs_del(iter->dev, oldname,
5294 &iter->dev->adj_list.upper);
5295 netdev_adjacent_sysfs_add(iter->dev, dev,
5296 &iter->dev->adj_list.upper);
5300 void *netdev_lower_dev_get_private(struct net_device *dev,
5301 struct net_device *lower_dev)
5303 struct netdev_adjacent *lower;
5307 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5311 return lower->private;
5313 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5316 int dev_get_nest_level(struct net_device *dev,
5317 bool (*type_check)(struct net_device *dev))
5319 struct net_device *lower = NULL;
5320 struct list_head *iter;
5326 netdev_for_each_lower_dev(dev, lower, iter) {
5327 nest = dev_get_nest_level(lower, type_check);
5328 if (max_nest < nest)
5332 if (type_check(dev))
5337 EXPORT_SYMBOL(dev_get_nest_level);
5339 static void dev_change_rx_flags(struct net_device *dev, int flags)
5341 const struct net_device_ops *ops = dev->netdev_ops;
5343 if (ops->ndo_change_rx_flags)
5344 ops->ndo_change_rx_flags(dev, flags);
5347 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5349 unsigned int old_flags = dev->flags;
5355 dev->flags |= IFF_PROMISC;
5356 dev->promiscuity += inc;
5357 if (dev->promiscuity == 0) {
5360 * If inc causes overflow, untouch promisc and return error.
5363 dev->flags &= ~IFF_PROMISC;
5365 dev->promiscuity -= inc;
5366 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5371 if (dev->flags != old_flags) {
5372 pr_info("device %s %s promiscuous mode\n",
5374 dev->flags & IFF_PROMISC ? "entered" : "left");
5375 if (audit_enabled) {
5376 current_uid_gid(&uid, &gid);
5377 audit_log(current->audit_context, GFP_ATOMIC,
5378 AUDIT_ANOM_PROMISCUOUS,
5379 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5380 dev->name, (dev->flags & IFF_PROMISC),
5381 (old_flags & IFF_PROMISC),
5382 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5383 from_kuid(&init_user_ns, uid),
5384 from_kgid(&init_user_ns, gid),
5385 audit_get_sessionid(current));
5388 dev_change_rx_flags(dev, IFF_PROMISC);
5391 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5396 * dev_set_promiscuity - update promiscuity count on a device
5400 * Add or remove promiscuity from a device. While the count in the device
5401 * remains above zero the interface remains promiscuous. Once it hits zero
5402 * the device reverts back to normal filtering operation. A negative inc
5403 * value is used to drop promiscuity on the device.
5404 * Return 0 if successful or a negative errno code on error.
5406 int dev_set_promiscuity(struct net_device *dev, int inc)
5408 unsigned int old_flags = dev->flags;
5411 err = __dev_set_promiscuity(dev, inc, true);
5414 if (dev->flags != old_flags)
5415 dev_set_rx_mode(dev);
5418 EXPORT_SYMBOL(dev_set_promiscuity);
5420 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5422 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5426 dev->flags |= IFF_ALLMULTI;
5427 dev->allmulti += inc;
5428 if (dev->allmulti == 0) {
5431 * If inc causes overflow, untouch allmulti and return error.
5434 dev->flags &= ~IFF_ALLMULTI;
5436 dev->allmulti -= inc;
5437 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5442 if (dev->flags ^ old_flags) {
5443 dev_change_rx_flags(dev, IFF_ALLMULTI);
5444 dev_set_rx_mode(dev);
5446 __dev_notify_flags(dev, old_flags,
5447 dev->gflags ^ old_gflags);
5453 * dev_set_allmulti - update allmulti count on a device
5457 * Add or remove reception of all multicast frames to a device. While the
5458 * count in the device remains above zero the interface remains listening
5459 * to all interfaces. Once it hits zero the device reverts back to normal
5460 * filtering operation. A negative @inc value is used to drop the counter
5461 * when releasing a resource needing all multicasts.
5462 * Return 0 if successful or a negative errno code on error.
5465 int dev_set_allmulti(struct net_device *dev, int inc)
5467 return __dev_set_allmulti(dev, inc, true);
5469 EXPORT_SYMBOL(dev_set_allmulti);
5472 * Upload unicast and multicast address lists to device and
5473 * configure RX filtering. When the device doesn't support unicast
5474 * filtering it is put in promiscuous mode while unicast addresses
5477 void __dev_set_rx_mode(struct net_device *dev)
5479 const struct net_device_ops *ops = dev->netdev_ops;
5481 /* dev_open will call this function so the list will stay sane. */
5482 if (!(dev->flags&IFF_UP))
5485 if (!netif_device_present(dev))
5488 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5489 /* Unicast addresses changes may only happen under the rtnl,
5490 * therefore calling __dev_set_promiscuity here is safe.
5492 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5493 __dev_set_promiscuity(dev, 1, false);
5494 dev->uc_promisc = true;
5495 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5496 __dev_set_promiscuity(dev, -1, false);
5497 dev->uc_promisc = false;
5501 if (ops->ndo_set_rx_mode)
5502 ops->ndo_set_rx_mode(dev);
5505 void dev_set_rx_mode(struct net_device *dev)
5507 netif_addr_lock_bh(dev);
5508 __dev_set_rx_mode(dev);
5509 netif_addr_unlock_bh(dev);
5513 * dev_get_flags - get flags reported to userspace
5516 * Get the combination of flag bits exported through APIs to userspace.
5518 unsigned int dev_get_flags(const struct net_device *dev)
5522 flags = (dev->flags & ~(IFF_PROMISC |
5527 (dev->gflags & (IFF_PROMISC |
5530 if (netif_running(dev)) {
5531 if (netif_oper_up(dev))
5532 flags |= IFF_RUNNING;
5533 if (netif_carrier_ok(dev))
5534 flags |= IFF_LOWER_UP;
5535 if (netif_dormant(dev))
5536 flags |= IFF_DORMANT;
5541 EXPORT_SYMBOL(dev_get_flags);
5543 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5545 unsigned int old_flags = dev->flags;
5551 * Set the flags on our device.
5554 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5555 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5557 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5561 * Load in the correct multicast list now the flags have changed.
5564 if ((old_flags ^ flags) & IFF_MULTICAST)
5565 dev_change_rx_flags(dev, IFF_MULTICAST);
5567 dev_set_rx_mode(dev);
5570 * Have we downed the interface. We handle IFF_UP ourselves
5571 * according to user attempts to set it, rather than blindly
5576 if ((old_flags ^ flags) & IFF_UP)
5577 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5579 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5580 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5581 unsigned int old_flags = dev->flags;
5583 dev->gflags ^= IFF_PROMISC;
5585 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5586 if (dev->flags != old_flags)
5587 dev_set_rx_mode(dev);
5590 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5591 is important. Some (broken) drivers set IFF_PROMISC, when
5592 IFF_ALLMULTI is requested not asking us and not reporting.
5594 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5595 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5597 dev->gflags ^= IFF_ALLMULTI;
5598 __dev_set_allmulti(dev, inc, false);
5604 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5605 unsigned int gchanges)
5607 unsigned int changes = dev->flags ^ old_flags;
5610 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5612 if (changes & IFF_UP) {
5613 if (dev->flags & IFF_UP)
5614 call_netdevice_notifiers(NETDEV_UP, dev);
5616 call_netdevice_notifiers(NETDEV_DOWN, dev);
5619 if (dev->flags & IFF_UP &&
5620 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5621 struct netdev_notifier_change_info change_info;
5623 change_info.flags_changed = changes;
5624 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5630 * dev_change_flags - change device settings
5632 * @flags: device state flags
5634 * Change settings on device based state flags. The flags are
5635 * in the userspace exported format.
5637 int dev_change_flags(struct net_device *dev, unsigned int flags)
5640 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5642 ret = __dev_change_flags(dev, flags);
5646 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5647 __dev_notify_flags(dev, old_flags, changes);
5650 EXPORT_SYMBOL(dev_change_flags);
5652 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5654 const struct net_device_ops *ops = dev->netdev_ops;
5656 if (ops->ndo_change_mtu)
5657 return ops->ndo_change_mtu(dev, new_mtu);
5664 * dev_set_mtu - Change maximum transfer unit
5666 * @new_mtu: new transfer unit
5668 * Change the maximum transfer size of the network device.
5670 int dev_set_mtu(struct net_device *dev, int new_mtu)
5674 if (new_mtu == dev->mtu)
5677 /* MTU must be positive. */
5681 if (!netif_device_present(dev))
5684 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5685 err = notifier_to_errno(err);
5689 orig_mtu = dev->mtu;
5690 err = __dev_set_mtu(dev, new_mtu);
5693 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5694 err = notifier_to_errno(err);
5696 /* setting mtu back and notifying everyone again,
5697 * so that they have a chance to revert changes.
5699 __dev_set_mtu(dev, orig_mtu);
5700 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5705 EXPORT_SYMBOL(dev_set_mtu);
5708 * dev_set_group - Change group this device belongs to
5710 * @new_group: group this device should belong to
5712 void dev_set_group(struct net_device *dev, int new_group)
5714 dev->group = new_group;
5716 EXPORT_SYMBOL(dev_set_group);
5719 * dev_set_mac_address - Change Media Access Control Address
5723 * Change the hardware (MAC) address of the device
5725 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5727 const struct net_device_ops *ops = dev->netdev_ops;
5730 if (!ops->ndo_set_mac_address)
5732 if (sa->sa_family != dev->type)
5734 if (!netif_device_present(dev))
5736 err = ops->ndo_set_mac_address(dev, sa);
5739 dev->addr_assign_type = NET_ADDR_SET;
5740 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5741 add_device_randomness(dev->dev_addr, dev->addr_len);
5744 EXPORT_SYMBOL(dev_set_mac_address);
5747 * dev_change_carrier - Change device carrier
5749 * @new_carrier: new value
5751 * Change device carrier
5753 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5755 const struct net_device_ops *ops = dev->netdev_ops;
5757 if (!ops->ndo_change_carrier)
5759 if (!netif_device_present(dev))
5761 return ops->ndo_change_carrier(dev, new_carrier);
5763 EXPORT_SYMBOL(dev_change_carrier);
5766 * dev_get_phys_port_id - Get device physical port ID
5770 * Get device physical port ID
5772 int dev_get_phys_port_id(struct net_device *dev,
5773 struct netdev_phys_port_id *ppid)
5775 const struct net_device_ops *ops = dev->netdev_ops;
5777 if (!ops->ndo_get_phys_port_id)
5779 return ops->ndo_get_phys_port_id(dev, ppid);
5781 EXPORT_SYMBOL(dev_get_phys_port_id);
5784 * dev_new_index - allocate an ifindex
5785 * @net: the applicable net namespace
5787 * Returns a suitable unique value for a new device interface
5788 * number. The caller must hold the rtnl semaphore or the
5789 * dev_base_lock to be sure it remains unique.
5791 static int dev_new_index(struct net *net)
5793 int ifindex = net->ifindex;
5797 if (!__dev_get_by_index(net, ifindex))
5798 return net->ifindex = ifindex;
5802 /* Delayed registration/unregisteration */
5803 static LIST_HEAD(net_todo_list);
5804 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5806 static void net_set_todo(struct net_device *dev)
5808 list_add_tail(&dev->todo_list, &net_todo_list);
5809 dev_net(dev)->dev_unreg_count++;
5812 static void rollback_registered_many(struct list_head *head)
5814 struct net_device *dev, *tmp;
5815 LIST_HEAD(close_head);
5817 BUG_ON(dev_boot_phase);
5820 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5821 /* Some devices call without registering
5822 * for initialization unwind. Remove those
5823 * devices and proceed with the remaining.
5825 if (dev->reg_state == NETREG_UNINITIALIZED) {
5826 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5830 list_del(&dev->unreg_list);
5833 dev->dismantle = true;
5834 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5837 /* If device is running, close it first. */
5838 list_for_each_entry(dev, head, unreg_list)
5839 list_add_tail(&dev->close_list, &close_head);
5840 dev_close_many(&close_head);
5842 list_for_each_entry(dev, head, unreg_list) {
5843 /* And unlink it from device chain. */
5844 unlist_netdevice(dev);
5846 dev->reg_state = NETREG_UNREGISTERING;
5851 list_for_each_entry(dev, head, unreg_list) {
5852 /* Shutdown queueing discipline. */
5856 /* Notify protocols, that we are about to destroy
5857 this device. They should clean all the things.
5859 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5862 * Flush the unicast and multicast chains
5867 if (dev->netdev_ops->ndo_uninit)
5868 dev->netdev_ops->ndo_uninit(dev);
5870 if (!dev->rtnl_link_ops ||
5871 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5872 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5874 /* Notifier chain MUST detach us all upper devices. */
5875 WARN_ON(netdev_has_any_upper_dev(dev));
5877 /* Remove entries from kobject tree */
5878 netdev_unregister_kobject(dev);
5880 /* Remove XPS queueing entries */
5881 netif_reset_xps_queues_gt(dev, 0);
5887 list_for_each_entry(dev, head, unreg_list)
5891 static void rollback_registered(struct net_device *dev)
5895 list_add(&dev->unreg_list, &single);
5896 rollback_registered_many(&single);
5900 static netdev_features_t netdev_fix_features(struct net_device *dev,
5901 netdev_features_t features)
5903 /* Fix illegal checksum combinations */
5904 if ((features & NETIF_F_HW_CSUM) &&
5905 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5906 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5907 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5910 /* TSO requires that SG is present as well. */
5911 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5912 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5913 features &= ~NETIF_F_ALL_TSO;
5916 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5917 !(features & NETIF_F_IP_CSUM)) {
5918 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5919 features &= ~NETIF_F_TSO;
5920 features &= ~NETIF_F_TSO_ECN;
5923 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5924 !(features & NETIF_F_IPV6_CSUM)) {
5925 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5926 features &= ~NETIF_F_TSO6;
5929 /* TSO ECN requires that TSO is present as well. */
5930 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5931 features &= ~NETIF_F_TSO_ECN;
5933 /* Software GSO depends on SG. */
5934 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5935 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5936 features &= ~NETIF_F_GSO;
5939 /* UFO needs SG and checksumming */
5940 if (features & NETIF_F_UFO) {
5941 /* maybe split UFO into V4 and V6? */
5942 if (!((features & NETIF_F_GEN_CSUM) ||
5943 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5944 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5946 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5947 features &= ~NETIF_F_UFO;
5950 if (!(features & NETIF_F_SG)) {
5952 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5953 features &= ~NETIF_F_UFO;
5957 #ifdef CONFIG_NET_RX_BUSY_POLL
5958 if (dev->netdev_ops->ndo_busy_poll)
5959 features |= NETIF_F_BUSY_POLL;
5962 features &= ~NETIF_F_BUSY_POLL;
5967 int __netdev_update_features(struct net_device *dev)
5969 netdev_features_t features;
5974 features = netdev_get_wanted_features(dev);
5976 if (dev->netdev_ops->ndo_fix_features)
5977 features = dev->netdev_ops->ndo_fix_features(dev, features);
5979 /* driver might be less strict about feature dependencies */
5980 features = netdev_fix_features(dev, features);
5982 if (dev->features == features)
5985 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5986 &dev->features, &features);
5988 if (dev->netdev_ops->ndo_set_features)
5989 err = dev->netdev_ops->ndo_set_features(dev, features);
5991 if (unlikely(err < 0)) {
5993 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5994 err, &features, &dev->features);
5999 dev->features = features;
6005 * netdev_update_features - recalculate device features
6006 * @dev: the device to check
6008 * Recalculate dev->features set and send notifications if it
6009 * has changed. Should be called after driver or hardware dependent
6010 * conditions might have changed that influence the features.
6012 void netdev_update_features(struct net_device *dev)
6014 if (__netdev_update_features(dev))
6015 netdev_features_change(dev);
6017 EXPORT_SYMBOL(netdev_update_features);
6020 * netdev_change_features - recalculate device features
6021 * @dev: the device to check
6023 * Recalculate dev->features set and send notifications even
6024 * if they have not changed. Should be called instead of
6025 * netdev_update_features() if also dev->vlan_features might
6026 * have changed to allow the changes to be propagated to stacked
6029 void netdev_change_features(struct net_device *dev)
6031 __netdev_update_features(dev);
6032 netdev_features_change(dev);
6034 EXPORT_SYMBOL(netdev_change_features);
6037 * netif_stacked_transfer_operstate - transfer operstate
6038 * @rootdev: the root or lower level device to transfer state from
6039 * @dev: the device to transfer operstate to
6041 * Transfer operational state from root to device. This is normally
6042 * called when a stacking relationship exists between the root
6043 * device and the device(a leaf device).
6045 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6046 struct net_device *dev)
6048 if (rootdev->operstate == IF_OPER_DORMANT)
6049 netif_dormant_on(dev);
6051 netif_dormant_off(dev);
6053 if (netif_carrier_ok(rootdev)) {
6054 if (!netif_carrier_ok(dev))
6055 netif_carrier_on(dev);
6057 if (netif_carrier_ok(dev))
6058 netif_carrier_off(dev);
6061 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6064 static int netif_alloc_rx_queues(struct net_device *dev)
6066 unsigned int i, count = dev->num_rx_queues;
6067 struct netdev_rx_queue *rx;
6071 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
6077 for (i = 0; i < count; i++)
6083 static void netdev_init_one_queue(struct net_device *dev,
6084 struct netdev_queue *queue, void *_unused)
6086 /* Initialize queue lock */
6087 spin_lock_init(&queue->_xmit_lock);
6088 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6089 queue->xmit_lock_owner = -1;
6090 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6093 dql_init(&queue->dql, HZ);
6097 static void netif_free_tx_queues(struct net_device *dev)
6102 static int netif_alloc_netdev_queues(struct net_device *dev)
6104 unsigned int count = dev->num_tx_queues;
6105 struct netdev_queue *tx;
6106 size_t sz = count * sizeof(*tx);
6108 BUG_ON(count < 1 || count > 0xffff);
6110 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6118 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6119 spin_lock_init(&dev->tx_global_lock);
6125 * register_netdevice - register a network device
6126 * @dev: device to register
6128 * Take a completed network device structure and add it to the kernel
6129 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6130 * chain. 0 is returned on success. A negative errno code is returned
6131 * on a failure to set up the device, or if the name is a duplicate.
6133 * Callers must hold the rtnl semaphore. You may want
6134 * register_netdev() instead of this.
6137 * The locking appears insufficient to guarantee two parallel registers
6138 * will not get the same name.
6141 int register_netdevice(struct net_device *dev)
6144 struct net *net = dev_net(dev);
6146 BUG_ON(dev_boot_phase);
6151 /* When net_device's are persistent, this will be fatal. */
6152 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6155 spin_lock_init(&dev->addr_list_lock);
6156 netdev_set_addr_lockdep_class(dev);
6160 ret = dev_get_valid_name(net, dev, dev->name);
6164 /* Init, if this function is available */
6165 if (dev->netdev_ops->ndo_init) {
6166 ret = dev->netdev_ops->ndo_init(dev);
6174 if (((dev->hw_features | dev->features) &
6175 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6176 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6177 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6178 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6185 dev->ifindex = dev_new_index(net);
6186 else if (__dev_get_by_index(net, dev->ifindex))
6189 if (dev->iflink == -1)
6190 dev->iflink = dev->ifindex;
6192 /* Transfer changeable features to wanted_features and enable
6193 * software offloads (GSO and GRO).
6195 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6196 dev->features |= NETIF_F_SOFT_FEATURES;
6197 dev->wanted_features = dev->features & dev->hw_features;
6199 if (!(dev->flags & IFF_LOOPBACK)) {
6200 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6203 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6205 dev->vlan_features |= NETIF_F_HIGHDMA;
6207 /* Make NETIF_F_SG inheritable to tunnel devices.
6209 dev->hw_enc_features |= NETIF_F_SG;
6211 /* Make NETIF_F_SG inheritable to MPLS.
6213 dev->mpls_features |= NETIF_F_SG;
6215 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6216 ret = notifier_to_errno(ret);
6220 ret = netdev_register_kobject(dev);
6223 dev->reg_state = NETREG_REGISTERED;
6225 __netdev_update_features(dev);
6228 * Default initial state at registry is that the
6229 * device is present.
6232 set_bit(__LINK_STATE_PRESENT, &dev->state);
6234 linkwatch_init_dev(dev);
6236 dev_init_scheduler(dev);
6238 list_netdevice(dev);
6239 add_device_randomness(dev->dev_addr, dev->addr_len);
6241 /* If the device has permanent device address, driver should
6242 * set dev_addr and also addr_assign_type should be set to
6243 * NET_ADDR_PERM (default value).
6245 if (dev->addr_assign_type == NET_ADDR_PERM)
6246 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6248 /* Notify protocols, that a new device appeared. */
6249 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6250 ret = notifier_to_errno(ret);
6252 rollback_registered(dev);
6253 dev->reg_state = NETREG_UNREGISTERED;
6256 * Prevent userspace races by waiting until the network
6257 * device is fully setup before sending notifications.
6259 if (!dev->rtnl_link_ops ||
6260 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6261 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6267 if (dev->netdev_ops->ndo_uninit)
6268 dev->netdev_ops->ndo_uninit(dev);
6271 EXPORT_SYMBOL(register_netdevice);
6274 * init_dummy_netdev - init a dummy network device for NAPI
6275 * @dev: device to init
6277 * This takes a network device structure and initialize the minimum
6278 * amount of fields so it can be used to schedule NAPI polls without
6279 * registering a full blown interface. This is to be used by drivers
6280 * that need to tie several hardware interfaces to a single NAPI
6281 * poll scheduler due to HW limitations.
6283 int init_dummy_netdev(struct net_device *dev)
6285 /* Clear everything. Note we don't initialize spinlocks
6286 * are they aren't supposed to be taken by any of the
6287 * NAPI code and this dummy netdev is supposed to be
6288 * only ever used for NAPI polls
6290 memset(dev, 0, sizeof(struct net_device));
6292 /* make sure we BUG if trying to hit standard
6293 * register/unregister code path
6295 dev->reg_state = NETREG_DUMMY;
6297 /* NAPI wants this */
6298 INIT_LIST_HEAD(&dev->napi_list);
6300 /* a dummy interface is started by default */
6301 set_bit(__LINK_STATE_PRESENT, &dev->state);
6302 set_bit(__LINK_STATE_START, &dev->state);
6304 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6305 * because users of this 'device' dont need to change
6311 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6315 * register_netdev - register a network device
6316 * @dev: device to register
6318 * Take a completed network device structure and add it to the kernel
6319 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6320 * chain. 0 is returned on success. A negative errno code is returned
6321 * on a failure to set up the device, or if the name is a duplicate.
6323 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6324 * and expands the device name if you passed a format string to
6327 int register_netdev(struct net_device *dev)
6332 err = register_netdevice(dev);
6336 EXPORT_SYMBOL(register_netdev);
6338 int netdev_refcnt_read(const struct net_device *dev)
6342 for_each_possible_cpu(i)
6343 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6346 EXPORT_SYMBOL(netdev_refcnt_read);
6349 * netdev_wait_allrefs - wait until all references are gone.
6350 * @dev: target net_device
6352 * This is called when unregistering network devices.
6354 * Any protocol or device that holds a reference should register
6355 * for netdevice notification, and cleanup and put back the
6356 * reference if they receive an UNREGISTER event.
6357 * We can get stuck here if buggy protocols don't correctly
6360 static void netdev_wait_allrefs(struct net_device *dev)
6362 unsigned long rebroadcast_time, warning_time;
6365 linkwatch_forget_dev(dev);
6367 rebroadcast_time = warning_time = jiffies;
6368 refcnt = netdev_refcnt_read(dev);
6370 while (refcnt != 0) {
6371 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6374 /* Rebroadcast unregister notification */
6375 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6381 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6382 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6384 /* We must not have linkwatch events
6385 * pending on unregister. If this
6386 * happens, we simply run the queue
6387 * unscheduled, resulting in a noop
6390 linkwatch_run_queue();
6395 rebroadcast_time = jiffies;
6400 refcnt = netdev_refcnt_read(dev);
6402 if (time_after(jiffies, warning_time + 10 * HZ)) {
6403 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6405 warning_time = jiffies;
6414 * register_netdevice(x1);
6415 * register_netdevice(x2);
6417 * unregister_netdevice(y1);
6418 * unregister_netdevice(y2);
6424 * We are invoked by rtnl_unlock().
6425 * This allows us to deal with problems:
6426 * 1) We can delete sysfs objects which invoke hotplug
6427 * without deadlocking with linkwatch via keventd.
6428 * 2) Since we run with the RTNL semaphore not held, we can sleep
6429 * safely in order to wait for the netdev refcnt to drop to zero.
6431 * We must not return until all unregister events added during
6432 * the interval the lock was held have been completed.
6434 void netdev_run_todo(void)
6436 struct list_head list;
6438 /* Snapshot list, allow later requests */
6439 list_replace_init(&net_todo_list, &list);
6444 /* Wait for rcu callbacks to finish before next phase */
6445 if (!list_empty(&list))
6448 while (!list_empty(&list)) {
6449 struct net_device *dev
6450 = list_first_entry(&list, struct net_device, todo_list);
6451 list_del(&dev->todo_list);
6454 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6457 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6458 pr_err("network todo '%s' but state %d\n",
6459 dev->name, dev->reg_state);
6464 dev->reg_state = NETREG_UNREGISTERED;
6466 on_each_cpu(flush_backlog, dev, 1);
6468 netdev_wait_allrefs(dev);
6471 BUG_ON(netdev_refcnt_read(dev));
6472 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6473 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6474 WARN_ON(dev->dn_ptr);
6476 if (dev->destructor)
6477 dev->destructor(dev);
6479 /* Report a network device has been unregistered */
6481 dev_net(dev)->dev_unreg_count--;
6483 wake_up(&netdev_unregistering_wq);
6485 /* Free network device */
6486 kobject_put(&dev->dev.kobj);
6490 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6491 * fields in the same order, with only the type differing.
6493 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6494 const struct net_device_stats *netdev_stats)
6496 #if BITS_PER_LONG == 64
6497 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6498 memcpy(stats64, netdev_stats, sizeof(*stats64));
6500 size_t i, n = sizeof(*stats64) / sizeof(u64);
6501 const unsigned long *src = (const unsigned long *)netdev_stats;
6502 u64 *dst = (u64 *)stats64;
6504 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6505 sizeof(*stats64) / sizeof(u64));
6506 for (i = 0; i < n; i++)
6510 EXPORT_SYMBOL(netdev_stats_to_stats64);
6513 * dev_get_stats - get network device statistics
6514 * @dev: device to get statistics from
6515 * @storage: place to store stats
6517 * Get network statistics from device. Return @storage.
6518 * The device driver may provide its own method by setting
6519 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6520 * otherwise the internal statistics structure is used.
6522 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6523 struct rtnl_link_stats64 *storage)
6525 const struct net_device_ops *ops = dev->netdev_ops;
6527 if (ops->ndo_get_stats64) {
6528 memset(storage, 0, sizeof(*storage));
6529 ops->ndo_get_stats64(dev, storage);
6530 } else if (ops->ndo_get_stats) {
6531 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6533 netdev_stats_to_stats64(storage, &dev->stats);
6535 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6536 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6539 EXPORT_SYMBOL(dev_get_stats);
6541 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6543 struct netdev_queue *queue = dev_ingress_queue(dev);
6545 #ifdef CONFIG_NET_CLS_ACT
6548 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6551 netdev_init_one_queue(dev, queue, NULL);
6552 queue->qdisc = &noop_qdisc;
6553 queue->qdisc_sleeping = &noop_qdisc;
6554 rcu_assign_pointer(dev->ingress_queue, queue);
6559 static const struct ethtool_ops default_ethtool_ops;
6561 void netdev_set_default_ethtool_ops(struct net_device *dev,
6562 const struct ethtool_ops *ops)
6564 if (dev->ethtool_ops == &default_ethtool_ops)
6565 dev->ethtool_ops = ops;
6567 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6569 void netdev_freemem(struct net_device *dev)
6571 char *addr = (char *)dev - dev->padded;
6577 * alloc_netdev_mqs - allocate network device
6578 * @sizeof_priv: size of private data to allocate space for
6579 * @name: device name format string
6580 * @name_assign_type: origin of device name
6581 * @setup: callback to initialize device
6582 * @txqs: the number of TX subqueues to allocate
6583 * @rxqs: the number of RX subqueues to allocate
6585 * Allocates a struct net_device with private data area for driver use
6586 * and performs basic initialization. Also allocates subqueue structs
6587 * for each queue on the device.
6589 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6590 unsigned char name_assign_type,
6591 void (*setup)(struct net_device *),
6592 unsigned int txqs, unsigned int rxqs)
6594 struct net_device *dev;
6596 struct net_device *p;
6598 BUG_ON(strlen(name) >= sizeof(dev->name));
6601 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6607 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6612 alloc_size = sizeof(struct net_device);
6614 /* ensure 32-byte alignment of private area */
6615 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6616 alloc_size += sizeof_priv;
6618 /* ensure 32-byte alignment of whole construct */
6619 alloc_size += NETDEV_ALIGN - 1;
6621 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6623 p = vzalloc(alloc_size);
6627 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6628 dev->padded = (char *)dev - (char *)p;
6630 dev->pcpu_refcnt = alloc_percpu(int);
6631 if (!dev->pcpu_refcnt)
6634 if (dev_addr_init(dev))
6640 dev_net_set(dev, &init_net);
6642 dev->gso_max_size = GSO_MAX_SIZE;
6643 dev->gso_max_segs = GSO_MAX_SEGS;
6645 INIT_LIST_HEAD(&dev->napi_list);
6646 INIT_LIST_HEAD(&dev->unreg_list);
6647 INIT_LIST_HEAD(&dev->close_list);
6648 INIT_LIST_HEAD(&dev->link_watch_list);
6649 INIT_LIST_HEAD(&dev->adj_list.upper);
6650 INIT_LIST_HEAD(&dev->adj_list.lower);
6651 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6652 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6653 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6656 dev->num_tx_queues = txqs;
6657 dev->real_num_tx_queues = txqs;
6658 if (netif_alloc_netdev_queues(dev))
6662 dev->num_rx_queues = rxqs;
6663 dev->real_num_rx_queues = rxqs;
6664 if (netif_alloc_rx_queues(dev))
6668 strcpy(dev->name, name);
6669 dev->name_assign_type = name_assign_type;
6670 dev->group = INIT_NETDEV_GROUP;
6671 if (!dev->ethtool_ops)
6672 dev->ethtool_ops = &default_ethtool_ops;
6680 free_percpu(dev->pcpu_refcnt);
6682 netdev_freemem(dev);
6685 EXPORT_SYMBOL(alloc_netdev_mqs);
6688 * free_netdev - free network device
6691 * This function does the last stage of destroying an allocated device
6692 * interface. The reference to the device object is released.
6693 * If this is the last reference then it will be freed.
6695 void free_netdev(struct net_device *dev)
6697 struct napi_struct *p, *n;
6699 release_net(dev_net(dev));
6701 netif_free_tx_queues(dev);
6706 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6708 /* Flush device addresses */
6709 dev_addr_flush(dev);
6711 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6714 free_percpu(dev->pcpu_refcnt);
6715 dev->pcpu_refcnt = NULL;
6717 /* Compatibility with error handling in drivers */
6718 if (dev->reg_state == NETREG_UNINITIALIZED) {
6719 netdev_freemem(dev);
6723 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6724 dev->reg_state = NETREG_RELEASED;
6726 /* will free via device release */
6727 put_device(&dev->dev);
6729 EXPORT_SYMBOL(free_netdev);
6732 * synchronize_net - Synchronize with packet receive processing
6734 * Wait for packets currently being received to be done.
6735 * Does not block later packets from starting.
6737 void synchronize_net(void)
6740 if (rtnl_is_locked())
6741 synchronize_rcu_expedited();
6745 EXPORT_SYMBOL(synchronize_net);
6748 * unregister_netdevice_queue - remove device from the kernel
6752 * This function shuts down a device interface and removes it
6753 * from the kernel tables.
6754 * If head not NULL, device is queued to be unregistered later.
6756 * Callers must hold the rtnl semaphore. You may want
6757 * unregister_netdev() instead of this.
6760 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6765 list_move_tail(&dev->unreg_list, head);
6767 rollback_registered(dev);
6768 /* Finish processing unregister after unlock */
6772 EXPORT_SYMBOL(unregister_netdevice_queue);
6775 * unregister_netdevice_many - unregister many devices
6776 * @head: list of devices
6778 * Note: As most callers use a stack allocated list_head,
6779 * we force a list_del() to make sure stack wont be corrupted later.
6781 void unregister_netdevice_many(struct list_head *head)
6783 struct net_device *dev;
6785 if (!list_empty(head)) {
6786 rollback_registered_many(head);
6787 list_for_each_entry(dev, head, unreg_list)
6792 EXPORT_SYMBOL(unregister_netdevice_many);
6795 * unregister_netdev - remove device from the kernel
6798 * This function shuts down a device interface and removes it
6799 * from the kernel tables.
6801 * This is just a wrapper for unregister_netdevice that takes
6802 * the rtnl semaphore. In general you want to use this and not
6803 * unregister_netdevice.
6805 void unregister_netdev(struct net_device *dev)
6808 unregister_netdevice(dev);
6811 EXPORT_SYMBOL(unregister_netdev);
6814 * dev_change_net_namespace - move device to different nethost namespace
6816 * @net: network namespace
6817 * @pat: If not NULL name pattern to try if the current device name
6818 * is already taken in the destination network namespace.
6820 * This function shuts down a device interface and moves it
6821 * to a new network namespace. On success 0 is returned, on
6822 * a failure a netagive errno code is returned.
6824 * Callers must hold the rtnl semaphore.
6827 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6833 /* Don't allow namespace local devices to be moved. */
6835 if (dev->features & NETIF_F_NETNS_LOCAL)
6838 /* Ensure the device has been registrered */
6839 if (dev->reg_state != NETREG_REGISTERED)
6842 /* Get out if there is nothing todo */
6844 if (net_eq(dev_net(dev), net))
6847 /* Pick the destination device name, and ensure
6848 * we can use it in the destination network namespace.
6851 if (__dev_get_by_name(net, dev->name)) {
6852 /* We get here if we can't use the current device name */
6855 if (dev_get_valid_name(net, dev, pat) < 0)
6860 * And now a mini version of register_netdevice unregister_netdevice.
6863 /* If device is running close it first. */
6866 /* And unlink it from device chain */
6868 unlist_netdevice(dev);
6872 /* Shutdown queueing discipline. */
6875 /* Notify protocols, that we are about to destroy
6876 this device. They should clean all the things.
6878 Note that dev->reg_state stays at NETREG_REGISTERED.
6879 This is wanted because this way 8021q and macvlan know
6880 the device is just moving and can keep their slaves up.
6882 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6884 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6885 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6888 * Flush the unicast and multicast chains
6893 /* Send a netdev-removed uevent to the old namespace */
6894 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6895 netdev_adjacent_del_links(dev);
6897 /* Actually switch the network namespace */
6898 dev_net_set(dev, net);
6900 /* If there is an ifindex conflict assign a new one */
6901 if (__dev_get_by_index(net, dev->ifindex)) {
6902 int iflink = (dev->iflink == dev->ifindex);
6903 dev->ifindex = dev_new_index(net);
6905 dev->iflink = dev->ifindex;
6908 /* Send a netdev-add uevent to the new namespace */
6909 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6910 netdev_adjacent_add_links(dev);
6912 /* Fixup kobjects */
6913 err = device_rename(&dev->dev, dev->name);
6916 /* Add the device back in the hashes */
6917 list_netdevice(dev);
6919 /* Notify protocols, that a new device appeared. */
6920 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6923 * Prevent userspace races by waiting until the network
6924 * device is fully setup before sending notifications.
6926 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6933 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6935 static int dev_cpu_callback(struct notifier_block *nfb,
6936 unsigned long action,
6939 struct sk_buff **list_skb;
6940 struct sk_buff *skb;
6941 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6942 struct softnet_data *sd, *oldsd;
6944 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6947 local_irq_disable();
6948 cpu = smp_processor_id();
6949 sd = &per_cpu(softnet_data, cpu);
6950 oldsd = &per_cpu(softnet_data, oldcpu);
6952 /* Find end of our completion_queue. */
6953 list_skb = &sd->completion_queue;
6955 list_skb = &(*list_skb)->next;
6956 /* Append completion queue from offline CPU. */
6957 *list_skb = oldsd->completion_queue;
6958 oldsd->completion_queue = NULL;
6960 /* Append output queue from offline CPU. */
6961 if (oldsd->output_queue) {
6962 *sd->output_queue_tailp = oldsd->output_queue;
6963 sd->output_queue_tailp = oldsd->output_queue_tailp;
6964 oldsd->output_queue = NULL;
6965 oldsd->output_queue_tailp = &oldsd->output_queue;
6967 /* Append NAPI poll list from offline CPU. */
6968 if (!list_empty(&oldsd->poll_list)) {
6969 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6970 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6973 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6976 /* Process offline CPU's input_pkt_queue */
6977 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6978 netif_rx_internal(skb);
6979 input_queue_head_incr(oldsd);
6981 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6982 netif_rx_internal(skb);
6983 input_queue_head_incr(oldsd);
6991 * netdev_increment_features - increment feature set by one
6992 * @all: current feature set
6993 * @one: new feature set
6994 * @mask: mask feature set
6996 * Computes a new feature set after adding a device with feature set
6997 * @one to the master device with current feature set @all. Will not
6998 * enable anything that is off in @mask. Returns the new feature set.
7000 netdev_features_t netdev_increment_features(netdev_features_t all,
7001 netdev_features_t one, netdev_features_t mask)
7003 if (mask & NETIF_F_GEN_CSUM)
7004 mask |= NETIF_F_ALL_CSUM;
7005 mask |= NETIF_F_VLAN_CHALLENGED;
7007 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7008 all &= one | ~NETIF_F_ALL_FOR_ALL;
7010 /* If one device supports hw checksumming, set for all. */
7011 if (all & NETIF_F_GEN_CSUM)
7012 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7016 EXPORT_SYMBOL(netdev_increment_features);
7018 static struct hlist_head * __net_init netdev_create_hash(void)
7021 struct hlist_head *hash;
7023 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7025 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7026 INIT_HLIST_HEAD(&hash[i]);
7031 /* Initialize per network namespace state */
7032 static int __net_init netdev_init(struct net *net)
7034 if (net != &init_net)
7035 INIT_LIST_HEAD(&net->dev_base_head);
7037 net->dev_name_head = netdev_create_hash();
7038 if (net->dev_name_head == NULL)
7041 net->dev_index_head = netdev_create_hash();
7042 if (net->dev_index_head == NULL)
7048 kfree(net->dev_name_head);
7054 * netdev_drivername - network driver for the device
7055 * @dev: network device
7057 * Determine network driver for device.
7059 const char *netdev_drivername(const struct net_device *dev)
7061 const struct device_driver *driver;
7062 const struct device *parent;
7063 const char *empty = "";
7065 parent = dev->dev.parent;
7069 driver = parent->driver;
7070 if (driver && driver->name)
7071 return driver->name;
7075 static int __netdev_printk(const char *level, const struct net_device *dev,
7076 struct va_format *vaf)
7080 if (dev && dev->dev.parent) {
7081 r = dev_printk_emit(level[1] - '0',
7084 dev_driver_string(dev->dev.parent),
7085 dev_name(dev->dev.parent),
7086 netdev_name(dev), netdev_reg_state(dev),
7089 r = printk("%s%s%s: %pV", level, netdev_name(dev),
7090 netdev_reg_state(dev), vaf);
7092 r = printk("%s(NULL net_device): %pV", level, vaf);
7098 int netdev_printk(const char *level, const struct net_device *dev,
7099 const char *format, ...)
7101 struct va_format vaf;
7105 va_start(args, format);
7110 r = __netdev_printk(level, dev, &vaf);
7116 EXPORT_SYMBOL(netdev_printk);
7118 #define define_netdev_printk_level(func, level) \
7119 int func(const struct net_device *dev, const char *fmt, ...) \
7122 struct va_format vaf; \
7125 va_start(args, fmt); \
7130 r = __netdev_printk(level, dev, &vaf); \
7136 EXPORT_SYMBOL(func);
7138 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7139 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7140 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7141 define_netdev_printk_level(netdev_err, KERN_ERR);
7142 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7143 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7144 define_netdev_printk_level(netdev_info, KERN_INFO);
7146 static void __net_exit netdev_exit(struct net *net)
7148 kfree(net->dev_name_head);
7149 kfree(net->dev_index_head);
7152 static struct pernet_operations __net_initdata netdev_net_ops = {
7153 .init = netdev_init,
7154 .exit = netdev_exit,
7157 static void __net_exit default_device_exit(struct net *net)
7159 struct net_device *dev, *aux;
7161 * Push all migratable network devices back to the
7162 * initial network namespace
7165 for_each_netdev_safe(net, dev, aux) {
7167 char fb_name[IFNAMSIZ];
7169 /* Ignore unmoveable devices (i.e. loopback) */
7170 if (dev->features & NETIF_F_NETNS_LOCAL)
7173 /* Leave virtual devices for the generic cleanup */
7174 if (dev->rtnl_link_ops)
7177 /* Push remaining network devices to init_net */
7178 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7179 err = dev_change_net_namespace(dev, &init_net, fb_name);
7181 pr_emerg("%s: failed to move %s to init_net: %d\n",
7182 __func__, dev->name, err);
7189 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7191 /* Return with the rtnl_lock held when there are no network
7192 * devices unregistering in any network namespace in net_list.
7199 prepare_to_wait(&netdev_unregistering_wq, &wait,
7200 TASK_UNINTERRUPTIBLE);
7201 unregistering = false;
7203 list_for_each_entry(net, net_list, exit_list) {
7204 if (net->dev_unreg_count > 0) {
7205 unregistering = true;
7214 finish_wait(&netdev_unregistering_wq, &wait);
7217 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7219 /* At exit all network devices most be removed from a network
7220 * namespace. Do this in the reverse order of registration.
7221 * Do this across as many network namespaces as possible to
7222 * improve batching efficiency.
7224 struct net_device *dev;
7226 LIST_HEAD(dev_kill_list);
7228 /* To prevent network device cleanup code from dereferencing
7229 * loopback devices or network devices that have been freed
7230 * wait here for all pending unregistrations to complete,
7231 * before unregistring the loopback device and allowing the
7232 * network namespace be freed.
7234 * The netdev todo list containing all network devices
7235 * unregistrations that happen in default_device_exit_batch
7236 * will run in the rtnl_unlock() at the end of
7237 * default_device_exit_batch.
7239 rtnl_lock_unregistering(net_list);
7240 list_for_each_entry(net, net_list, exit_list) {
7241 for_each_netdev_reverse(net, dev) {
7242 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7243 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7245 unregister_netdevice_queue(dev, &dev_kill_list);
7248 unregister_netdevice_many(&dev_kill_list);
7252 static struct pernet_operations __net_initdata default_device_ops = {
7253 .exit = default_device_exit,
7254 .exit_batch = default_device_exit_batch,
7258 * Initialize the DEV module. At boot time this walks the device list and
7259 * unhooks any devices that fail to initialise (normally hardware not
7260 * present) and leaves us with a valid list of present and active devices.
7265 * This is called single threaded during boot, so no need
7266 * to take the rtnl semaphore.
7268 static int __init net_dev_init(void)
7270 int i, rc = -ENOMEM;
7272 BUG_ON(!dev_boot_phase);
7274 if (dev_proc_init())
7277 if (netdev_kobject_init())
7280 INIT_LIST_HEAD(&ptype_all);
7281 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7282 INIT_LIST_HEAD(&ptype_base[i]);
7284 INIT_LIST_HEAD(&offload_base);
7286 if (register_pernet_subsys(&netdev_net_ops))
7290 * Initialise the packet receive queues.
7293 for_each_possible_cpu(i) {
7294 struct softnet_data *sd = &per_cpu(softnet_data, i);
7296 skb_queue_head_init(&sd->input_pkt_queue);
7297 skb_queue_head_init(&sd->process_queue);
7298 INIT_LIST_HEAD(&sd->poll_list);
7299 sd->output_queue_tailp = &sd->output_queue;
7301 sd->csd.func = rps_trigger_softirq;
7306 sd->backlog.poll = process_backlog;
7307 sd->backlog.weight = weight_p;
7312 /* The loopback device is special if any other network devices
7313 * is present in a network namespace the loopback device must
7314 * be present. Since we now dynamically allocate and free the
7315 * loopback device ensure this invariant is maintained by
7316 * keeping the loopback device as the first device on the
7317 * list of network devices. Ensuring the loopback devices
7318 * is the first device that appears and the last network device
7321 if (register_pernet_device(&loopback_net_ops))
7324 if (register_pernet_device(&default_device_ops))
7327 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7328 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7330 hotcpu_notifier(dev_cpu_callback, 0);
7337 subsys_initcall(net_dev_init);