2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
121 #include <linux/ipv6.h>
122 #include <linux/in.h>
123 #include <linux/jhash.h>
124 #include <linux/random.h>
125 #include <trace/events/napi.h>
126 #include <trace/events/net.h>
127 #include <trace/events/skb.h>
128 #include <linux/pci.h>
129 #include <linux/inetdevice.h>
130 #include <linux/cpu_rmap.h>
131 #include <linux/static_key.h>
132 #include <linux/hashtable.h>
133 #include <linux/vmalloc.h>
134 #include <linux/if_macvlan.h>
135 #include <linux/errqueue.h>
137 #include "net-sysfs.h"
139 /* Instead of increasing this, you should create a hash table. */
140 #define MAX_GRO_SKBS 8
142 /* This should be increased if a protocol with a bigger head is added. */
143 #define GRO_MAX_HEAD (MAX_HEADER + 128)
145 static DEFINE_SPINLOCK(ptype_lock);
146 static DEFINE_SPINLOCK(offload_lock);
147 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
148 struct list_head ptype_all __read_mostly; /* Taps */
149 static struct list_head offload_base __read_mostly;
151 static int netif_rx_internal(struct sk_buff *skb);
152 static int call_netdevice_notifiers_info(unsigned long val,
153 struct net_device *dev,
154 struct netdev_notifier_info *info);
157 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
160 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
162 * Writers must hold the rtnl semaphore while they loop through the
163 * dev_base_head list, and hold dev_base_lock for writing when they do the
164 * actual updates. This allows pure readers to access the list even
165 * while a writer is preparing to update it.
167 * To put it another way, dev_base_lock is held for writing only to
168 * protect against pure readers; the rtnl semaphore provides the
169 * protection against other writers.
171 * See, for example usages, register_netdevice() and
172 * unregister_netdevice(), which must be called with the rtnl
175 DEFINE_RWLOCK(dev_base_lock);
176 EXPORT_SYMBOL(dev_base_lock);
178 /* protects napi_hash addition/deletion and napi_gen_id */
179 static DEFINE_SPINLOCK(napi_hash_lock);
181 static unsigned int napi_gen_id;
182 static DEFINE_HASHTABLE(napi_hash, 8);
184 static seqcount_t devnet_rename_seq;
186 static inline void dev_base_seq_inc(struct net *net)
188 while (++net->dev_base_seq == 0);
191 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
193 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
195 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
198 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
200 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
203 static inline void rps_lock(struct softnet_data *sd)
206 spin_lock(&sd->input_pkt_queue.lock);
210 static inline void rps_unlock(struct softnet_data *sd)
213 spin_unlock(&sd->input_pkt_queue.lock);
217 /* Device list insertion */
218 static void list_netdevice(struct net_device *dev)
220 struct net *net = dev_net(dev);
224 write_lock_bh(&dev_base_lock);
225 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
226 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
227 hlist_add_head_rcu(&dev->index_hlist,
228 dev_index_hash(net, dev->ifindex));
229 write_unlock_bh(&dev_base_lock);
231 dev_base_seq_inc(net);
234 /* Device list removal
235 * caller must respect a RCU grace period before freeing/reusing dev
237 static void unlist_netdevice(struct net_device *dev)
241 /* Unlink dev from the device chain */
242 write_lock_bh(&dev_base_lock);
243 list_del_rcu(&dev->dev_list);
244 hlist_del_rcu(&dev->name_hlist);
245 hlist_del_rcu(&dev->index_hlist);
246 write_unlock_bh(&dev_base_lock);
248 dev_base_seq_inc(dev_net(dev));
255 static RAW_NOTIFIER_HEAD(netdev_chain);
258 * Device drivers call our routines to queue packets here. We empty the
259 * queue in the local softnet handler.
262 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
263 EXPORT_PER_CPU_SYMBOL(softnet_data);
265 #ifdef CONFIG_LOCKDEP
267 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
268 * according to dev->type
270 static const unsigned short netdev_lock_type[] =
271 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
272 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
273 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
274 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
275 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
276 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
277 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
278 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
279 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
280 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
281 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
282 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
283 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
284 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
285 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
287 static const char *const netdev_lock_name[] =
288 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
289 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
290 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
291 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
292 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
293 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
294 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
295 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
296 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
297 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
298 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
299 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
300 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
301 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
302 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
304 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
305 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
307 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
311 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
312 if (netdev_lock_type[i] == dev_type)
314 /* the last key is used by default */
315 return ARRAY_SIZE(netdev_lock_type) - 1;
318 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
319 unsigned short dev_type)
323 i = netdev_lock_pos(dev_type);
324 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
325 netdev_lock_name[i]);
328 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
332 i = netdev_lock_pos(dev->type);
333 lockdep_set_class_and_name(&dev->addr_list_lock,
334 &netdev_addr_lock_key[i],
335 netdev_lock_name[i]);
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
339 unsigned short dev_type)
342 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
347 /*******************************************************************************
349 Protocol management and registration routines
351 *******************************************************************************/
354 * Add a protocol ID to the list. Now that the input handler is
355 * smarter we can dispense with all the messy stuff that used to be
358 * BEWARE!!! Protocol handlers, mangling input packets,
359 * MUST BE last in hash buckets and checking protocol handlers
360 * MUST start from promiscuous ptype_all chain in net_bh.
361 * It is true now, do not change it.
362 * Explanation follows: if protocol handler, mangling packet, will
363 * be the first on list, it is not able to sense, that packet
364 * is cloned and should be copied-on-write, so that it will
365 * change it and subsequent readers will get broken packet.
369 static inline struct list_head *ptype_head(const struct packet_type *pt)
371 if (pt->type == htons(ETH_P_ALL))
374 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
378 * dev_add_pack - add packet handler
379 * @pt: packet type declaration
381 * Add a protocol handler to the networking stack. The passed &packet_type
382 * is linked into kernel lists and may not be freed until it has been
383 * removed from the kernel lists.
385 * This call does not sleep therefore it can not
386 * guarantee all CPU's that are in middle of receiving packets
387 * will see the new packet type (until the next received packet).
390 void dev_add_pack(struct packet_type *pt)
392 struct list_head *head = ptype_head(pt);
394 spin_lock(&ptype_lock);
395 list_add_rcu(&pt->list, head);
396 spin_unlock(&ptype_lock);
398 EXPORT_SYMBOL(dev_add_pack);
401 * __dev_remove_pack - remove packet handler
402 * @pt: packet type declaration
404 * Remove a protocol handler that was previously added to the kernel
405 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
406 * from the kernel lists and can be freed or reused once this function
409 * The packet type might still be in use by receivers
410 * and must not be freed until after all the CPU's have gone
411 * through a quiescent state.
413 void __dev_remove_pack(struct packet_type *pt)
415 struct list_head *head = ptype_head(pt);
416 struct packet_type *pt1;
418 spin_lock(&ptype_lock);
420 list_for_each_entry(pt1, head, list) {
422 list_del_rcu(&pt->list);
427 pr_warn("dev_remove_pack: %p not found\n", pt);
429 spin_unlock(&ptype_lock);
431 EXPORT_SYMBOL(__dev_remove_pack);
434 * dev_remove_pack - remove packet handler
435 * @pt: packet type declaration
437 * Remove a protocol handler that was previously added to the kernel
438 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
439 * from the kernel lists and can be freed or reused once this function
442 * This call sleeps to guarantee that no CPU is looking at the packet
445 void dev_remove_pack(struct packet_type *pt)
447 __dev_remove_pack(pt);
451 EXPORT_SYMBOL(dev_remove_pack);
455 * dev_add_offload - register offload handlers
456 * @po: protocol offload declaration
458 * Add protocol offload handlers to the networking stack. The passed
459 * &proto_offload is linked into kernel lists and may not be freed until
460 * it has been removed from the kernel lists.
462 * This call does not sleep therefore it can not
463 * guarantee all CPU's that are in middle of receiving packets
464 * will see the new offload handlers (until the next received packet).
466 void dev_add_offload(struct packet_offload *po)
468 struct list_head *head = &offload_base;
470 spin_lock(&offload_lock);
471 list_add_rcu(&po->list, head);
472 spin_unlock(&offload_lock);
474 EXPORT_SYMBOL(dev_add_offload);
477 * __dev_remove_offload - remove offload handler
478 * @po: packet offload declaration
480 * Remove a protocol offload handler that was previously added to the
481 * kernel offload handlers by dev_add_offload(). The passed &offload_type
482 * is removed from the kernel lists and can be freed or reused once this
485 * The packet type might still be in use by receivers
486 * and must not be freed until after all the CPU's have gone
487 * through a quiescent state.
489 static void __dev_remove_offload(struct packet_offload *po)
491 struct list_head *head = &offload_base;
492 struct packet_offload *po1;
494 spin_lock(&offload_lock);
496 list_for_each_entry(po1, head, list) {
498 list_del_rcu(&po->list);
503 pr_warn("dev_remove_offload: %p not found\n", po);
505 spin_unlock(&offload_lock);
509 * dev_remove_offload - remove packet offload handler
510 * @po: packet offload declaration
512 * Remove a packet offload handler that was previously added to the kernel
513 * offload handlers by dev_add_offload(). The passed &offload_type is
514 * removed from the kernel lists and can be freed or reused once this
517 * This call sleeps to guarantee that no CPU is looking at the packet
520 void dev_remove_offload(struct packet_offload *po)
522 __dev_remove_offload(po);
526 EXPORT_SYMBOL(dev_remove_offload);
528 /******************************************************************************
530 Device Boot-time Settings Routines
532 *******************************************************************************/
534 /* Boot time configuration table */
535 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
538 * netdev_boot_setup_add - add new setup entry
539 * @name: name of the device
540 * @map: configured settings for the device
542 * Adds new setup entry to the dev_boot_setup list. The function
543 * returns 0 on error and 1 on success. This is a generic routine to
546 static int netdev_boot_setup_add(char *name, struct ifmap *map)
548 struct netdev_boot_setup *s;
552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
553 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
554 memset(s[i].name, 0, sizeof(s[i].name));
555 strlcpy(s[i].name, name, IFNAMSIZ);
556 memcpy(&s[i].map, map, sizeof(s[i].map));
561 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
565 * netdev_boot_setup_check - check boot time settings
566 * @dev: the netdevice
568 * Check boot time settings for the device.
569 * The found settings are set for the device to be used
570 * later in the device probing.
571 * Returns 0 if no settings found, 1 if they are.
573 int netdev_boot_setup_check(struct net_device *dev)
575 struct netdev_boot_setup *s = dev_boot_setup;
578 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
579 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
580 !strcmp(dev->name, s[i].name)) {
581 dev->irq = s[i].map.irq;
582 dev->base_addr = s[i].map.base_addr;
583 dev->mem_start = s[i].map.mem_start;
584 dev->mem_end = s[i].map.mem_end;
590 EXPORT_SYMBOL(netdev_boot_setup_check);
594 * netdev_boot_base - get address from boot time settings
595 * @prefix: prefix for network device
596 * @unit: id for network device
598 * Check boot time settings for the base address of device.
599 * The found settings are set for the device to be used
600 * later in the device probing.
601 * Returns 0 if no settings found.
603 unsigned long netdev_boot_base(const char *prefix, int unit)
605 const struct netdev_boot_setup *s = dev_boot_setup;
609 sprintf(name, "%s%d", prefix, unit);
612 * If device already registered then return base of 1
613 * to indicate not to probe for this interface
615 if (__dev_get_by_name(&init_net, name))
618 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
619 if (!strcmp(name, s[i].name))
620 return s[i].map.base_addr;
625 * Saves at boot time configured settings for any netdevice.
627 int __init netdev_boot_setup(char *str)
632 str = get_options(str, ARRAY_SIZE(ints), ints);
637 memset(&map, 0, sizeof(map));
641 map.base_addr = ints[2];
643 map.mem_start = ints[3];
645 map.mem_end = ints[4];
647 /* Add new entry to the list */
648 return netdev_boot_setup_add(str, &map);
651 __setup("netdev=", netdev_boot_setup);
653 /*******************************************************************************
655 Device Interface Subroutines
657 *******************************************************************************/
660 * __dev_get_by_name - find a device by its name
661 * @net: the applicable net namespace
662 * @name: name to find
664 * Find an interface by name. Must be called under RTNL semaphore
665 * or @dev_base_lock. If the name is found a pointer to the device
666 * is returned. If the name is not found then %NULL is returned. The
667 * reference counters are not incremented so the caller must be
668 * careful with locks.
671 struct net_device *__dev_get_by_name(struct net *net, const char *name)
673 struct net_device *dev;
674 struct hlist_head *head = dev_name_hash(net, name);
676 hlist_for_each_entry(dev, head, name_hlist)
677 if (!strncmp(dev->name, name, IFNAMSIZ))
682 EXPORT_SYMBOL(__dev_get_by_name);
685 * dev_get_by_name_rcu - find a device by its name
686 * @net: the applicable net namespace
687 * @name: name to find
689 * Find an interface by name.
690 * If the name is found a pointer to the device is returned.
691 * If the name is not found then %NULL is returned.
692 * The reference counters are not incremented so the caller must be
693 * careful with locks. The caller must hold RCU lock.
696 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
698 struct net_device *dev;
699 struct hlist_head *head = dev_name_hash(net, name);
701 hlist_for_each_entry_rcu(dev, head, name_hlist)
702 if (!strncmp(dev->name, name, IFNAMSIZ))
707 EXPORT_SYMBOL(dev_get_by_name_rcu);
710 * dev_get_by_name - find a device by its name
711 * @net: the applicable net namespace
712 * @name: name to find
714 * Find an interface by name. This can be called from any
715 * context and does its own locking. The returned handle has
716 * the usage count incremented and the caller must use dev_put() to
717 * release it when it is no longer needed. %NULL is returned if no
718 * matching device is found.
721 struct net_device *dev_get_by_name(struct net *net, const char *name)
723 struct net_device *dev;
726 dev = dev_get_by_name_rcu(net, name);
732 EXPORT_SYMBOL(dev_get_by_name);
735 * __dev_get_by_index - find a device by its ifindex
736 * @net: the applicable net namespace
737 * @ifindex: index of device
739 * Search for an interface by index. Returns %NULL if the device
740 * is not found or a pointer to the device. The device has not
741 * had its reference counter increased so the caller must be careful
742 * about locking. The caller must hold either the RTNL semaphore
746 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
748 struct net_device *dev;
749 struct hlist_head *head = dev_index_hash(net, ifindex);
751 hlist_for_each_entry(dev, head, index_hlist)
752 if (dev->ifindex == ifindex)
757 EXPORT_SYMBOL(__dev_get_by_index);
760 * dev_get_by_index_rcu - find a device by its ifindex
761 * @net: the applicable net namespace
762 * @ifindex: index of device
764 * Search for an interface by index. Returns %NULL if the device
765 * is not found or a pointer to the device. The device has not
766 * had its reference counter increased so the caller must be careful
767 * about locking. The caller must hold RCU lock.
770 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
772 struct net_device *dev;
773 struct hlist_head *head = dev_index_hash(net, ifindex);
775 hlist_for_each_entry_rcu(dev, head, index_hlist)
776 if (dev->ifindex == ifindex)
781 EXPORT_SYMBOL(dev_get_by_index_rcu);
785 * dev_get_by_index - find a device by its ifindex
786 * @net: the applicable net namespace
787 * @ifindex: index of device
789 * Search for an interface by index. Returns NULL if the device
790 * is not found or a pointer to the device. The device returned has
791 * had a reference added and the pointer is safe until the user calls
792 * dev_put to indicate they have finished with it.
795 struct net_device *dev_get_by_index(struct net *net, int ifindex)
797 struct net_device *dev;
800 dev = dev_get_by_index_rcu(net, ifindex);
806 EXPORT_SYMBOL(dev_get_by_index);
809 * netdev_get_name - get a netdevice name, knowing its ifindex.
810 * @net: network namespace
811 * @name: a pointer to the buffer where the name will be stored.
812 * @ifindex: the ifindex of the interface to get the name from.
814 * The use of raw_seqcount_begin() and cond_resched() before
815 * retrying is required as we want to give the writers a chance
816 * to complete when CONFIG_PREEMPT is not set.
818 int netdev_get_name(struct net *net, char *name, int ifindex)
820 struct net_device *dev;
824 seq = raw_seqcount_begin(&devnet_rename_seq);
826 dev = dev_get_by_index_rcu(net, ifindex);
832 strcpy(name, dev->name);
834 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
843 * dev_getbyhwaddr_rcu - find a device by its hardware address
844 * @net: the applicable net namespace
845 * @type: media type of device
846 * @ha: hardware address
848 * Search for an interface by MAC address. Returns NULL if the device
849 * is not found or a pointer to the device.
850 * The caller must hold RCU or RTNL.
851 * The returned device has not had its ref count increased
852 * and the caller must therefore be careful about locking
856 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
859 struct net_device *dev;
861 for_each_netdev_rcu(net, dev)
862 if (dev->type == type &&
863 !memcmp(dev->dev_addr, ha, dev->addr_len))
868 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
870 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
872 struct net_device *dev;
875 for_each_netdev(net, dev)
876 if (dev->type == type)
881 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
883 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
885 struct net_device *dev, *ret = NULL;
888 for_each_netdev_rcu(net, dev)
889 if (dev->type == type) {
897 EXPORT_SYMBOL(dev_getfirstbyhwtype);
900 * __dev_get_by_flags - find any device with given flags
901 * @net: the applicable net namespace
902 * @if_flags: IFF_* values
903 * @mask: bitmask of bits in if_flags to check
905 * Search for any interface with the given flags. Returns NULL if a device
906 * is not found or a pointer to the device. Must be called inside
907 * rtnl_lock(), and result refcount is unchanged.
910 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
913 struct net_device *dev, *ret;
918 for_each_netdev(net, dev) {
919 if (((dev->flags ^ if_flags) & mask) == 0) {
926 EXPORT_SYMBOL(__dev_get_by_flags);
929 * dev_valid_name - check if name is okay for network device
932 * Network device names need to be valid file names to
933 * to allow sysfs to work. We also disallow any kind of
936 bool dev_valid_name(const char *name)
940 if (strlen(name) >= IFNAMSIZ)
942 if (!strcmp(name, ".") || !strcmp(name, ".."))
946 if (*name == '/' || isspace(*name))
952 EXPORT_SYMBOL(dev_valid_name);
955 * __dev_alloc_name - allocate a name for a device
956 * @net: network namespace to allocate the device name in
957 * @name: name format string
958 * @buf: scratch buffer and result name string
960 * Passed a format string - eg "lt%d" it will try and find a suitable
961 * id. It scans list of devices to build up a free map, then chooses
962 * the first empty slot. The caller must hold the dev_base or rtnl lock
963 * while allocating the name and adding the device in order to avoid
965 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
966 * Returns the number of the unit assigned or a negative errno code.
969 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
973 const int max_netdevices = 8*PAGE_SIZE;
974 unsigned long *inuse;
975 struct net_device *d;
977 p = strnchr(name, IFNAMSIZ-1, '%');
980 * Verify the string as this thing may have come from
981 * the user. There must be either one "%d" and no other "%"
984 if (p[1] != 'd' || strchr(p + 2, '%'))
987 /* Use one page as a bit array of possible slots */
988 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
992 for_each_netdev(net, d) {
993 if (!sscanf(d->name, name, &i))
995 if (i < 0 || i >= max_netdevices)
998 /* avoid cases where sscanf is not exact inverse of printf */
999 snprintf(buf, IFNAMSIZ, name, i);
1000 if (!strncmp(buf, d->name, IFNAMSIZ))
1004 i = find_first_zero_bit(inuse, max_netdevices);
1005 free_page((unsigned long) inuse);
1009 snprintf(buf, IFNAMSIZ, name, i);
1010 if (!__dev_get_by_name(net, buf))
1013 /* It is possible to run out of possible slots
1014 * when the name is long and there isn't enough space left
1015 * for the digits, or if all bits are used.
1021 * dev_alloc_name - allocate a name for a device
1023 * @name: name format string
1025 * Passed a format string - eg "lt%d" it will try and find a suitable
1026 * id. It scans list of devices to build up a free map, then chooses
1027 * the first empty slot. The caller must hold the dev_base or rtnl lock
1028 * while allocating the name and adding the device in order to avoid
1030 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1031 * Returns the number of the unit assigned or a negative errno code.
1034 int dev_alloc_name(struct net_device *dev, const char *name)
1040 BUG_ON(!dev_net(dev));
1042 ret = __dev_alloc_name(net, name, buf);
1044 strlcpy(dev->name, buf, IFNAMSIZ);
1047 EXPORT_SYMBOL(dev_alloc_name);
1049 static int dev_alloc_name_ns(struct net *net,
1050 struct net_device *dev,
1056 ret = __dev_alloc_name(net, name, buf);
1058 strlcpy(dev->name, buf, IFNAMSIZ);
1062 static int dev_get_valid_name(struct net *net,
1063 struct net_device *dev,
1068 if (!dev_valid_name(name))
1071 if (strchr(name, '%'))
1072 return dev_alloc_name_ns(net, dev, name);
1073 else if (__dev_get_by_name(net, name))
1075 else if (dev->name != name)
1076 strlcpy(dev->name, name, IFNAMSIZ);
1082 * dev_change_name - change name of a device
1084 * @newname: name (or format string) must be at least IFNAMSIZ
1086 * Change name of a device, can pass format strings "eth%d".
1089 int dev_change_name(struct net_device *dev, const char *newname)
1091 unsigned char old_assign_type;
1092 char oldname[IFNAMSIZ];
1098 BUG_ON(!dev_net(dev));
1101 if (dev->flags & IFF_UP)
1104 write_seqcount_begin(&devnet_rename_seq);
1106 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1107 write_seqcount_end(&devnet_rename_seq);
1111 memcpy(oldname, dev->name, IFNAMSIZ);
1113 err = dev_get_valid_name(net, dev, newname);
1115 write_seqcount_end(&devnet_rename_seq);
1119 if (oldname[0] && !strchr(oldname, '%'))
1120 netdev_info(dev, "renamed from %s\n", oldname);
1122 old_assign_type = dev->name_assign_type;
1123 dev->name_assign_type = NET_NAME_RENAMED;
1126 ret = device_rename(&dev->dev, dev->name);
1128 memcpy(dev->name, oldname, IFNAMSIZ);
1129 dev->name_assign_type = old_assign_type;
1130 write_seqcount_end(&devnet_rename_seq);
1134 write_seqcount_end(&devnet_rename_seq);
1136 netdev_adjacent_rename_links(dev, oldname);
1138 write_lock_bh(&dev_base_lock);
1139 hlist_del_rcu(&dev->name_hlist);
1140 write_unlock_bh(&dev_base_lock);
1144 write_lock_bh(&dev_base_lock);
1145 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1146 write_unlock_bh(&dev_base_lock);
1148 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1149 ret = notifier_to_errno(ret);
1152 /* err >= 0 after dev_alloc_name() or stores the first errno */
1155 write_seqcount_begin(&devnet_rename_seq);
1156 memcpy(dev->name, oldname, IFNAMSIZ);
1157 memcpy(oldname, newname, IFNAMSIZ);
1158 dev->name_assign_type = old_assign_type;
1159 old_assign_type = NET_NAME_RENAMED;
1162 pr_err("%s: name change rollback failed: %d\n",
1171 * dev_set_alias - change ifalias of a device
1173 * @alias: name up to IFALIASZ
1174 * @len: limit of bytes to copy from info
1176 * Set ifalias for a device,
1178 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1184 if (len >= IFALIASZ)
1188 kfree(dev->ifalias);
1189 dev->ifalias = NULL;
1193 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1196 dev->ifalias = new_ifalias;
1198 strlcpy(dev->ifalias, alias, len+1);
1204 * netdev_features_change - device changes features
1205 * @dev: device to cause notification
1207 * Called to indicate a device has changed features.
1209 void netdev_features_change(struct net_device *dev)
1211 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1213 EXPORT_SYMBOL(netdev_features_change);
1216 * netdev_state_change - device changes state
1217 * @dev: device to cause notification
1219 * Called to indicate a device has changed state. This function calls
1220 * the notifier chains for netdev_chain and sends a NEWLINK message
1221 * to the routing socket.
1223 void netdev_state_change(struct net_device *dev)
1225 if (dev->flags & IFF_UP) {
1226 struct netdev_notifier_change_info change_info;
1228 change_info.flags_changed = 0;
1229 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
1231 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1234 EXPORT_SYMBOL(netdev_state_change);
1237 * netdev_notify_peers - notify network peers about existence of @dev
1238 * @dev: network device
1240 * Generate traffic such that interested network peers are aware of
1241 * @dev, such as by generating a gratuitous ARP. This may be used when
1242 * a device wants to inform the rest of the network about some sort of
1243 * reconfiguration such as a failover event or virtual machine
1246 void netdev_notify_peers(struct net_device *dev)
1249 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1252 EXPORT_SYMBOL(netdev_notify_peers);
1254 static int __dev_open(struct net_device *dev)
1256 const struct net_device_ops *ops = dev->netdev_ops;
1261 if (!netif_device_present(dev))
1264 /* Block netpoll from trying to do any rx path servicing.
1265 * If we don't do this there is a chance ndo_poll_controller
1266 * or ndo_poll may be running while we open the device
1268 netpoll_poll_disable(dev);
1270 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1271 ret = notifier_to_errno(ret);
1275 set_bit(__LINK_STATE_START, &dev->state);
1277 if (ops->ndo_validate_addr)
1278 ret = ops->ndo_validate_addr(dev);
1280 if (!ret && ops->ndo_open)
1281 ret = ops->ndo_open(dev);
1283 netpoll_poll_enable(dev);
1286 clear_bit(__LINK_STATE_START, &dev->state);
1288 dev->flags |= IFF_UP;
1289 net_dmaengine_get();
1290 dev_set_rx_mode(dev);
1292 add_device_randomness(dev->dev_addr, dev->addr_len);
1299 * dev_open - prepare an interface for use.
1300 * @dev: device to open
1302 * Takes a device from down to up state. The device's private open
1303 * function is invoked and then the multicast lists are loaded. Finally
1304 * the device is moved into the up state and a %NETDEV_UP message is
1305 * sent to the netdev notifier chain.
1307 * Calling this function on an active interface is a nop. On a failure
1308 * a negative errno code is returned.
1310 int dev_open(struct net_device *dev)
1314 if (dev->flags & IFF_UP)
1317 ret = __dev_open(dev);
1321 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1322 call_netdevice_notifiers(NETDEV_UP, dev);
1326 EXPORT_SYMBOL(dev_open);
1328 static int __dev_close_many(struct list_head *head)
1330 struct net_device *dev;
1335 list_for_each_entry(dev, head, close_list) {
1336 /* Temporarily disable netpoll until the interface is down */
1337 netpoll_poll_disable(dev);
1339 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1341 clear_bit(__LINK_STATE_START, &dev->state);
1343 /* Synchronize to scheduled poll. We cannot touch poll list, it
1344 * can be even on different cpu. So just clear netif_running().
1346 * dev->stop() will invoke napi_disable() on all of it's
1347 * napi_struct instances on this device.
1349 smp_mb__after_atomic(); /* Commit netif_running(). */
1352 dev_deactivate_many(head);
1354 list_for_each_entry(dev, head, close_list) {
1355 const struct net_device_ops *ops = dev->netdev_ops;
1358 * Call the device specific close. This cannot fail.
1359 * Only if device is UP
1361 * We allow it to be called even after a DETACH hot-plug
1367 dev->flags &= ~IFF_UP;
1368 net_dmaengine_put();
1369 netpoll_poll_enable(dev);
1375 static int __dev_close(struct net_device *dev)
1380 list_add(&dev->close_list, &single);
1381 retval = __dev_close_many(&single);
1387 static int dev_close_many(struct list_head *head)
1389 struct net_device *dev, *tmp;
1391 /* Remove the devices that don't need to be closed */
1392 list_for_each_entry_safe(dev, tmp, head, close_list)
1393 if (!(dev->flags & IFF_UP))
1394 list_del_init(&dev->close_list);
1396 __dev_close_many(head);
1398 list_for_each_entry_safe(dev, tmp, head, close_list) {
1399 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1400 call_netdevice_notifiers(NETDEV_DOWN, dev);
1401 list_del_init(&dev->close_list);
1408 * dev_close - shutdown an interface.
1409 * @dev: device to shutdown
1411 * This function moves an active device into down state. A
1412 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1413 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1416 int dev_close(struct net_device *dev)
1418 if (dev->flags & IFF_UP) {
1421 list_add(&dev->close_list, &single);
1422 dev_close_many(&single);
1427 EXPORT_SYMBOL(dev_close);
1431 * dev_disable_lro - disable Large Receive Offload on a device
1434 * Disable Large Receive Offload (LRO) on a net device. Must be
1435 * called under RTNL. This is needed if received packets may be
1436 * forwarded to another interface.
1438 void dev_disable_lro(struct net_device *dev)
1441 * If we're trying to disable lro on a vlan device
1442 * use the underlying physical device instead
1444 if (is_vlan_dev(dev))
1445 dev = vlan_dev_real_dev(dev);
1447 /* the same for macvlan devices */
1448 if (netif_is_macvlan(dev))
1449 dev = macvlan_dev_real_dev(dev);
1451 dev->wanted_features &= ~NETIF_F_LRO;
1452 netdev_update_features(dev);
1454 if (unlikely(dev->features & NETIF_F_LRO))
1455 netdev_WARN(dev, "failed to disable LRO!\n");
1457 EXPORT_SYMBOL(dev_disable_lro);
1459 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1460 struct net_device *dev)
1462 struct netdev_notifier_info info;
1464 netdev_notifier_info_init(&info, dev);
1465 return nb->notifier_call(nb, val, &info);
1468 static int dev_boot_phase = 1;
1471 * register_netdevice_notifier - register a network notifier block
1474 * Register a notifier to be called when network device events occur.
1475 * The notifier passed is linked into the kernel structures and must
1476 * not be reused until it has been unregistered. A negative errno code
1477 * is returned on a failure.
1479 * When registered all registration and up events are replayed
1480 * to the new notifier to allow device to have a race free
1481 * view of the network device list.
1484 int register_netdevice_notifier(struct notifier_block *nb)
1486 struct net_device *dev;
1487 struct net_device *last;
1492 err = raw_notifier_chain_register(&netdev_chain, nb);
1498 for_each_netdev(net, dev) {
1499 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1500 err = notifier_to_errno(err);
1504 if (!(dev->flags & IFF_UP))
1507 call_netdevice_notifier(nb, NETDEV_UP, dev);
1518 for_each_netdev(net, dev) {
1522 if (dev->flags & IFF_UP) {
1523 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1525 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1527 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1532 raw_notifier_chain_unregister(&netdev_chain, nb);
1535 EXPORT_SYMBOL(register_netdevice_notifier);
1538 * unregister_netdevice_notifier - unregister a network notifier block
1541 * Unregister a notifier previously registered by
1542 * register_netdevice_notifier(). The notifier is unlinked into the
1543 * kernel structures and may then be reused. A negative errno code
1544 * is returned on a failure.
1546 * After unregistering unregister and down device events are synthesized
1547 * for all devices on the device list to the removed notifier to remove
1548 * the need for special case cleanup code.
1551 int unregister_netdevice_notifier(struct notifier_block *nb)
1553 struct net_device *dev;
1558 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1563 for_each_netdev(net, dev) {
1564 if (dev->flags & IFF_UP) {
1565 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1567 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1569 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1576 EXPORT_SYMBOL(unregister_netdevice_notifier);
1579 * call_netdevice_notifiers_info - call all network notifier blocks
1580 * @val: value passed unmodified to notifier function
1581 * @dev: net_device pointer passed unmodified to notifier function
1582 * @info: notifier information data
1584 * Call all network notifier blocks. Parameters and return value
1585 * are as for raw_notifier_call_chain().
1588 static int call_netdevice_notifiers_info(unsigned long val,
1589 struct net_device *dev,
1590 struct netdev_notifier_info *info)
1593 netdev_notifier_info_init(info, dev);
1594 return raw_notifier_call_chain(&netdev_chain, val, info);
1598 * call_netdevice_notifiers - call all network notifier blocks
1599 * @val: value passed unmodified to notifier function
1600 * @dev: net_device pointer passed unmodified to notifier function
1602 * Call all network notifier blocks. Parameters and return value
1603 * are as for raw_notifier_call_chain().
1606 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1608 struct netdev_notifier_info info;
1610 return call_netdevice_notifiers_info(val, dev, &info);
1612 EXPORT_SYMBOL(call_netdevice_notifiers);
1614 static struct static_key netstamp_needed __read_mostly;
1615 #ifdef HAVE_JUMP_LABEL
1616 /* We are not allowed to call static_key_slow_dec() from irq context
1617 * If net_disable_timestamp() is called from irq context, defer the
1618 * static_key_slow_dec() calls.
1620 static atomic_t netstamp_needed_deferred;
1623 void net_enable_timestamp(void)
1625 #ifdef HAVE_JUMP_LABEL
1626 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1630 static_key_slow_dec(&netstamp_needed);
1634 static_key_slow_inc(&netstamp_needed);
1636 EXPORT_SYMBOL(net_enable_timestamp);
1638 void net_disable_timestamp(void)
1640 #ifdef HAVE_JUMP_LABEL
1641 if (in_interrupt()) {
1642 atomic_inc(&netstamp_needed_deferred);
1646 static_key_slow_dec(&netstamp_needed);
1648 EXPORT_SYMBOL(net_disable_timestamp);
1650 static inline void net_timestamp_set(struct sk_buff *skb)
1652 skb->tstamp.tv64 = 0;
1653 if (static_key_false(&netstamp_needed))
1654 __net_timestamp(skb);
1657 #define net_timestamp_check(COND, SKB) \
1658 if (static_key_false(&netstamp_needed)) { \
1659 if ((COND) && !(SKB)->tstamp.tv64) \
1660 __net_timestamp(SKB); \
1663 bool is_skb_forwardable(struct net_device *dev, struct sk_buff *skb)
1667 if (!(dev->flags & IFF_UP))
1670 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1671 if (skb->len <= len)
1674 /* if TSO is enabled, we don't care about the length as the packet
1675 * could be forwarded without being segmented before
1677 if (skb_is_gso(skb))
1682 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1684 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1686 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1687 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1688 atomic_long_inc(&dev->rx_dropped);
1694 if (unlikely(!is_skb_forwardable(dev, skb))) {
1695 atomic_long_inc(&dev->rx_dropped);
1700 skb_scrub_packet(skb, true);
1701 skb->protocol = eth_type_trans(skb, dev);
1705 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1708 * dev_forward_skb - loopback an skb to another netif
1710 * @dev: destination network device
1711 * @skb: buffer to forward
1714 * NET_RX_SUCCESS (no congestion)
1715 * NET_RX_DROP (packet was dropped, but freed)
1717 * dev_forward_skb can be used for injecting an skb from the
1718 * start_xmit function of one device into the receive queue
1719 * of another device.
1721 * The receiving device may be in another namespace, so
1722 * we have to clear all information in the skb that could
1723 * impact namespace isolation.
1725 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1727 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1729 EXPORT_SYMBOL_GPL(dev_forward_skb);
1731 static inline int deliver_skb(struct sk_buff *skb,
1732 struct packet_type *pt_prev,
1733 struct net_device *orig_dev)
1735 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1737 atomic_inc(&skb->users);
1738 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1741 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1743 if (!ptype->af_packet_priv || !skb->sk)
1746 if (ptype->id_match)
1747 return ptype->id_match(ptype, skb->sk);
1748 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1755 * Support routine. Sends outgoing frames to any network
1756 * taps currently in use.
1759 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1761 struct packet_type *ptype;
1762 struct sk_buff *skb2 = NULL;
1763 struct packet_type *pt_prev = NULL;
1766 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1767 /* Never send packets back to the socket
1768 * they originated from - MvS (miquels@drinkel.ow.org)
1770 if ((ptype->dev == dev || !ptype->dev) &&
1771 (!skb_loop_sk(ptype, skb))) {
1773 deliver_skb(skb2, pt_prev, skb->dev);
1778 skb2 = skb_clone(skb, GFP_ATOMIC);
1782 net_timestamp_set(skb2);
1784 /* skb->nh should be correctly
1785 set by sender, so that the second statement is
1786 just protection against buggy protocols.
1788 skb_reset_mac_header(skb2);
1790 if (skb_network_header(skb2) < skb2->data ||
1791 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1792 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1793 ntohs(skb2->protocol),
1795 skb_reset_network_header(skb2);
1798 skb2->transport_header = skb2->network_header;
1799 skb2->pkt_type = PACKET_OUTGOING;
1804 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1809 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1810 * @dev: Network device
1811 * @txq: number of queues available
1813 * If real_num_tx_queues is changed the tc mappings may no longer be
1814 * valid. To resolve this verify the tc mapping remains valid and if
1815 * not NULL the mapping. With no priorities mapping to this
1816 * offset/count pair it will no longer be used. In the worst case TC0
1817 * is invalid nothing can be done so disable priority mappings. If is
1818 * expected that drivers will fix this mapping if they can before
1819 * calling netif_set_real_num_tx_queues.
1821 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1824 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1826 /* If TC0 is invalidated disable TC mapping */
1827 if (tc->offset + tc->count > txq) {
1828 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1833 /* Invalidated prio to tc mappings set to TC0 */
1834 for (i = 1; i < TC_BITMASK + 1; i++) {
1835 int q = netdev_get_prio_tc_map(dev, i);
1837 tc = &dev->tc_to_txq[q];
1838 if (tc->offset + tc->count > txq) {
1839 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1841 netdev_set_prio_tc_map(dev, i, 0);
1847 static DEFINE_MUTEX(xps_map_mutex);
1848 #define xmap_dereference(P) \
1849 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1851 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1854 struct xps_map *map = NULL;
1858 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1860 for (pos = 0; map && pos < map->len; pos++) {
1861 if (map->queues[pos] == index) {
1863 map->queues[pos] = map->queues[--map->len];
1865 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1866 kfree_rcu(map, rcu);
1876 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1878 struct xps_dev_maps *dev_maps;
1880 bool active = false;
1882 mutex_lock(&xps_map_mutex);
1883 dev_maps = xmap_dereference(dev->xps_maps);
1888 for_each_possible_cpu(cpu) {
1889 for (i = index; i < dev->num_tx_queues; i++) {
1890 if (!remove_xps_queue(dev_maps, cpu, i))
1893 if (i == dev->num_tx_queues)
1898 RCU_INIT_POINTER(dev->xps_maps, NULL);
1899 kfree_rcu(dev_maps, rcu);
1902 for (i = index; i < dev->num_tx_queues; i++)
1903 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1907 mutex_unlock(&xps_map_mutex);
1910 static struct xps_map *expand_xps_map(struct xps_map *map,
1913 struct xps_map *new_map;
1914 int alloc_len = XPS_MIN_MAP_ALLOC;
1917 for (pos = 0; map && pos < map->len; pos++) {
1918 if (map->queues[pos] != index)
1923 /* Need to add queue to this CPU's existing map */
1925 if (pos < map->alloc_len)
1928 alloc_len = map->alloc_len * 2;
1931 /* Need to allocate new map to store queue on this CPU's map */
1932 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1937 for (i = 0; i < pos; i++)
1938 new_map->queues[i] = map->queues[i];
1939 new_map->alloc_len = alloc_len;
1945 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1948 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1949 struct xps_map *map, *new_map;
1950 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1951 int cpu, numa_node_id = -2;
1952 bool active = false;
1954 mutex_lock(&xps_map_mutex);
1956 dev_maps = xmap_dereference(dev->xps_maps);
1958 /* allocate memory for queue storage */
1959 for_each_online_cpu(cpu) {
1960 if (!cpumask_test_cpu(cpu, mask))
1964 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1965 if (!new_dev_maps) {
1966 mutex_unlock(&xps_map_mutex);
1970 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1973 map = expand_xps_map(map, cpu, index);
1977 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1981 goto out_no_new_maps;
1983 for_each_possible_cpu(cpu) {
1984 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1985 /* add queue to CPU maps */
1988 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1989 while ((pos < map->len) && (map->queues[pos] != index))
1992 if (pos == map->len)
1993 map->queues[map->len++] = index;
1995 if (numa_node_id == -2)
1996 numa_node_id = cpu_to_node(cpu);
1997 else if (numa_node_id != cpu_to_node(cpu))
2000 } else if (dev_maps) {
2001 /* fill in the new device map from the old device map */
2002 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2003 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
2008 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2010 /* Cleanup old maps */
2012 for_each_possible_cpu(cpu) {
2013 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2014 map = xmap_dereference(dev_maps->cpu_map[cpu]);
2015 if (map && map != new_map)
2016 kfree_rcu(map, rcu);
2019 kfree_rcu(dev_maps, rcu);
2022 dev_maps = new_dev_maps;
2026 /* update Tx queue numa node */
2027 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2028 (numa_node_id >= 0) ? numa_node_id :
2034 /* removes queue from unused CPUs */
2035 for_each_possible_cpu(cpu) {
2036 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2039 if (remove_xps_queue(dev_maps, cpu, index))
2043 /* free map if not active */
2045 RCU_INIT_POINTER(dev->xps_maps, NULL);
2046 kfree_rcu(dev_maps, rcu);
2050 mutex_unlock(&xps_map_mutex);
2054 /* remove any maps that we added */
2055 for_each_possible_cpu(cpu) {
2056 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2057 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2059 if (new_map && new_map != map)
2063 mutex_unlock(&xps_map_mutex);
2065 kfree(new_dev_maps);
2068 EXPORT_SYMBOL(netif_set_xps_queue);
2072 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2073 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2075 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2079 if (txq < 1 || txq > dev->num_tx_queues)
2082 if (dev->reg_state == NETREG_REGISTERED ||
2083 dev->reg_state == NETREG_UNREGISTERING) {
2086 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2092 netif_setup_tc(dev, txq);
2094 if (txq < dev->real_num_tx_queues) {
2095 qdisc_reset_all_tx_gt(dev, txq);
2097 netif_reset_xps_queues_gt(dev, txq);
2102 dev->real_num_tx_queues = txq;
2105 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2109 * netif_set_real_num_rx_queues - set actual number of RX queues used
2110 * @dev: Network device
2111 * @rxq: Actual number of RX queues
2113 * This must be called either with the rtnl_lock held or before
2114 * registration of the net device. Returns 0 on success, or a
2115 * negative error code. If called before registration, it always
2118 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2122 if (rxq < 1 || rxq > dev->num_rx_queues)
2125 if (dev->reg_state == NETREG_REGISTERED) {
2128 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2134 dev->real_num_rx_queues = rxq;
2137 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2141 * netif_get_num_default_rss_queues - default number of RSS queues
2143 * This routine should set an upper limit on the number of RSS queues
2144 * used by default by multiqueue devices.
2146 int netif_get_num_default_rss_queues(void)
2148 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2150 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2152 static inline void __netif_reschedule(struct Qdisc *q)
2154 struct softnet_data *sd;
2155 unsigned long flags;
2157 local_irq_save(flags);
2158 sd = &__get_cpu_var(softnet_data);
2159 q->next_sched = NULL;
2160 *sd->output_queue_tailp = q;
2161 sd->output_queue_tailp = &q->next_sched;
2162 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2163 local_irq_restore(flags);
2166 void __netif_schedule(struct Qdisc *q)
2168 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2169 __netif_reschedule(q);
2171 EXPORT_SYMBOL(__netif_schedule);
2173 struct dev_kfree_skb_cb {
2174 enum skb_free_reason reason;
2177 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2179 return (struct dev_kfree_skb_cb *)skb->cb;
2182 void netif_schedule_queue(struct netdev_queue *txq)
2185 if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2186 struct Qdisc *q = rcu_dereference(txq->qdisc);
2188 __netif_schedule(q);
2192 EXPORT_SYMBOL(netif_schedule_queue);
2195 * netif_wake_subqueue - allow sending packets on subqueue
2196 * @dev: network device
2197 * @queue_index: sub queue index
2199 * Resume individual transmit queue of a device with multiple transmit queues.
2201 void netif_wake_subqueue(struct net_device *dev, u16 queue_index)
2203 struct netdev_queue *txq = netdev_get_tx_queue(dev, queue_index);
2205 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &txq->state)) {
2209 q = rcu_dereference(txq->qdisc);
2210 __netif_schedule(q);
2214 EXPORT_SYMBOL(netif_wake_subqueue);
2216 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2218 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2222 q = rcu_dereference(dev_queue->qdisc);
2223 __netif_schedule(q);
2227 EXPORT_SYMBOL(netif_tx_wake_queue);
2229 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2231 unsigned long flags;
2233 if (likely(atomic_read(&skb->users) == 1)) {
2235 atomic_set(&skb->users, 0);
2236 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2239 get_kfree_skb_cb(skb)->reason = reason;
2240 local_irq_save(flags);
2241 skb->next = __this_cpu_read(softnet_data.completion_queue);
2242 __this_cpu_write(softnet_data.completion_queue, skb);
2243 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2244 local_irq_restore(flags);
2246 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2248 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2250 if (in_irq() || irqs_disabled())
2251 __dev_kfree_skb_irq(skb, reason);
2255 EXPORT_SYMBOL(__dev_kfree_skb_any);
2259 * netif_device_detach - mark device as removed
2260 * @dev: network device
2262 * Mark device as removed from system and therefore no longer available.
2264 void netif_device_detach(struct net_device *dev)
2266 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2267 netif_running(dev)) {
2268 netif_tx_stop_all_queues(dev);
2271 EXPORT_SYMBOL(netif_device_detach);
2274 * netif_device_attach - mark device as attached
2275 * @dev: network device
2277 * Mark device as attached from system and restart if needed.
2279 void netif_device_attach(struct net_device *dev)
2281 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2282 netif_running(dev)) {
2283 netif_tx_wake_all_queues(dev);
2284 __netdev_watchdog_up(dev);
2287 EXPORT_SYMBOL(netif_device_attach);
2289 static void skb_warn_bad_offload(const struct sk_buff *skb)
2291 static const netdev_features_t null_features = 0;
2292 struct net_device *dev = skb->dev;
2293 const char *driver = "";
2295 if (!net_ratelimit())
2298 if (dev && dev->dev.parent)
2299 driver = dev_driver_string(dev->dev.parent);
2301 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2302 "gso_type=%d ip_summed=%d\n",
2303 driver, dev ? &dev->features : &null_features,
2304 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2305 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2306 skb_shinfo(skb)->gso_type, skb->ip_summed);
2310 * Invalidate hardware checksum when packet is to be mangled, and
2311 * complete checksum manually on outgoing path.
2313 int skb_checksum_help(struct sk_buff *skb)
2316 int ret = 0, offset;
2318 if (skb->ip_summed == CHECKSUM_COMPLETE)
2319 goto out_set_summed;
2321 if (unlikely(skb_shinfo(skb)->gso_size)) {
2322 skb_warn_bad_offload(skb);
2326 /* Before computing a checksum, we should make sure no frag could
2327 * be modified by an external entity : checksum could be wrong.
2329 if (skb_has_shared_frag(skb)) {
2330 ret = __skb_linearize(skb);
2335 offset = skb_checksum_start_offset(skb);
2336 BUG_ON(offset >= skb_headlen(skb));
2337 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2339 offset += skb->csum_offset;
2340 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2342 if (skb_cloned(skb) &&
2343 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2344 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2349 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2351 skb->ip_summed = CHECKSUM_NONE;
2355 EXPORT_SYMBOL(skb_checksum_help);
2357 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2359 unsigned int vlan_depth = skb->mac_len;
2360 __be16 type = skb->protocol;
2362 /* Tunnel gso handlers can set protocol to ethernet. */
2363 if (type == htons(ETH_P_TEB)) {
2366 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2369 eth = (struct ethhdr *)skb_mac_header(skb);
2370 type = eth->h_proto;
2373 /* if skb->protocol is 802.1Q/AD then the header should already be
2374 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at
2375 * ETH_HLEN otherwise
2377 if (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2379 if (WARN_ON(vlan_depth < VLAN_HLEN))
2381 vlan_depth -= VLAN_HLEN;
2383 vlan_depth = ETH_HLEN;
2386 struct vlan_hdr *vh;
2388 if (unlikely(!pskb_may_pull(skb,
2389 vlan_depth + VLAN_HLEN)))
2392 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2393 type = vh->h_vlan_encapsulated_proto;
2394 vlan_depth += VLAN_HLEN;
2395 } while (type == htons(ETH_P_8021Q) ||
2396 type == htons(ETH_P_8021AD));
2399 *depth = vlan_depth;
2405 * skb_mac_gso_segment - mac layer segmentation handler.
2406 * @skb: buffer to segment
2407 * @features: features for the output path (see dev->features)
2409 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2410 netdev_features_t features)
2412 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2413 struct packet_offload *ptype;
2414 int vlan_depth = skb->mac_len;
2415 __be16 type = skb_network_protocol(skb, &vlan_depth);
2417 if (unlikely(!type))
2418 return ERR_PTR(-EINVAL);
2420 __skb_pull(skb, vlan_depth);
2423 list_for_each_entry_rcu(ptype, &offload_base, list) {
2424 if (ptype->type == type && ptype->callbacks.gso_segment) {
2425 segs = ptype->callbacks.gso_segment(skb, features);
2431 __skb_push(skb, skb->data - skb_mac_header(skb));
2435 EXPORT_SYMBOL(skb_mac_gso_segment);
2438 /* openvswitch calls this on rx path, so we need a different check.
2440 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2443 return skb->ip_summed != CHECKSUM_PARTIAL;
2445 return skb->ip_summed == CHECKSUM_NONE;
2449 * __skb_gso_segment - Perform segmentation on skb.
2450 * @skb: buffer to segment
2451 * @features: features for the output path (see dev->features)
2452 * @tx_path: whether it is called in TX path
2454 * This function segments the given skb and returns a list of segments.
2456 * It may return NULL if the skb requires no segmentation. This is
2457 * only possible when GSO is used for verifying header integrity.
2459 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2460 netdev_features_t features, bool tx_path)
2462 if (unlikely(skb_needs_check(skb, tx_path))) {
2465 skb_warn_bad_offload(skb);
2467 err = skb_cow_head(skb, 0);
2469 return ERR_PTR(err);
2472 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2473 SKB_GSO_CB(skb)->encap_level = 0;
2475 skb_reset_mac_header(skb);
2476 skb_reset_mac_len(skb);
2478 return skb_mac_gso_segment(skb, features);
2480 EXPORT_SYMBOL(__skb_gso_segment);
2482 /* Take action when hardware reception checksum errors are detected. */
2484 void netdev_rx_csum_fault(struct net_device *dev)
2486 if (net_ratelimit()) {
2487 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2491 EXPORT_SYMBOL(netdev_rx_csum_fault);
2494 /* Actually, we should eliminate this check as soon as we know, that:
2495 * 1. IOMMU is present and allows to map all the memory.
2496 * 2. No high memory really exists on this machine.
2499 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2501 #ifdef CONFIG_HIGHMEM
2503 if (!(dev->features & NETIF_F_HIGHDMA)) {
2504 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2505 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2506 if (PageHighMem(skb_frag_page(frag)))
2511 if (PCI_DMA_BUS_IS_PHYS) {
2512 struct device *pdev = dev->dev.parent;
2516 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2517 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2518 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2519 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2527 /* If MPLS offload request, verify we are testing hardware MPLS features
2528 * instead of standard features for the netdev.
2530 #ifdef CONFIG_NET_MPLS_GSO
2531 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2532 netdev_features_t features,
2535 if (type == htons(ETH_P_MPLS_UC) || type == htons(ETH_P_MPLS_MC))
2536 features &= skb->dev->mpls_features;
2541 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2542 netdev_features_t features,
2549 static netdev_features_t harmonize_features(struct sk_buff *skb,
2550 netdev_features_t features)
2555 type = skb_network_protocol(skb, &tmp);
2556 features = net_mpls_features(skb, features, type);
2558 if (skb->ip_summed != CHECKSUM_NONE &&
2559 !can_checksum_protocol(features, type)) {
2560 features &= ~NETIF_F_ALL_CSUM;
2561 } else if (illegal_highdma(skb->dev, skb)) {
2562 features &= ~NETIF_F_SG;
2568 netdev_features_t netif_skb_features(struct sk_buff *skb)
2570 const struct net_device *dev = skb->dev;
2571 netdev_features_t features = dev->features;
2572 u16 gso_segs = skb_shinfo(skb)->gso_segs;
2573 __be16 protocol = skb->protocol;
2575 if (gso_segs > dev->gso_max_segs || gso_segs < dev->gso_min_segs)
2576 features &= ~NETIF_F_GSO_MASK;
2578 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2579 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2580 protocol = veh->h_vlan_encapsulated_proto;
2581 } else if (!vlan_tx_tag_present(skb)) {
2582 return harmonize_features(skb, features);
2585 features = netdev_intersect_features(features,
2586 dev->vlan_features |
2587 NETIF_F_HW_VLAN_CTAG_TX |
2588 NETIF_F_HW_VLAN_STAG_TX);
2590 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2591 features = netdev_intersect_features(features,
2596 NETIF_F_HW_VLAN_CTAG_TX |
2597 NETIF_F_HW_VLAN_STAG_TX);
2599 return harmonize_features(skb, features);
2601 EXPORT_SYMBOL(netif_skb_features);
2603 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
2604 struct netdev_queue *txq, bool more)
2609 if (!list_empty(&ptype_all))
2610 dev_queue_xmit_nit(skb, dev);
2613 trace_net_dev_start_xmit(skb, dev);
2614 rc = netdev_start_xmit(skb, dev, txq, more);
2615 trace_net_dev_xmit(skb, rc, dev, len);
2620 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
2621 struct netdev_queue *txq, int *ret)
2623 struct sk_buff *skb = first;
2624 int rc = NETDEV_TX_OK;
2627 struct sk_buff *next = skb->next;
2630 rc = xmit_one(skb, dev, txq, next != NULL);
2631 if (unlikely(!dev_xmit_complete(rc))) {
2637 if (netif_xmit_stopped(txq) && skb) {
2638 rc = NETDEV_TX_BUSY;
2648 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
2649 netdev_features_t features)
2651 if (vlan_tx_tag_present(skb) &&
2652 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2653 skb = __vlan_put_tag(skb, skb->vlan_proto,
2654 vlan_tx_tag_get(skb));
2661 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev)
2663 netdev_features_t features;
2668 features = netif_skb_features(skb);
2669 skb = validate_xmit_vlan(skb, features);
2673 /* If encapsulation offload request, verify we are testing
2674 * hardware encapsulation features instead of standard
2675 * features for the netdev
2677 if (skb->encapsulation)
2678 features &= dev->hw_enc_features;
2680 if (netif_needs_gso(skb, features)) {
2681 struct sk_buff *segs;
2683 segs = skb_gso_segment(skb, features);
2691 if (skb_needs_linearize(skb, features) &&
2692 __skb_linearize(skb))
2695 /* If packet is not checksummed and device does not
2696 * support checksumming for this protocol, complete
2697 * checksumming here.
2699 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2700 if (skb->encapsulation)
2701 skb_set_inner_transport_header(skb,
2702 skb_checksum_start_offset(skb));
2704 skb_set_transport_header(skb,
2705 skb_checksum_start_offset(skb));
2706 if (!(features & NETIF_F_ALL_CSUM) &&
2707 skb_checksum_help(skb))
2720 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev)
2722 struct sk_buff *next, *head = NULL, *tail;
2724 for (; skb != NULL; skb = next) {
2728 /* in case skb wont be segmented, point to itself */
2731 skb = validate_xmit_skb(skb, dev);
2739 /* If skb was segmented, skb->prev points to
2740 * the last segment. If not, it still contains skb.
2747 static void qdisc_pkt_len_init(struct sk_buff *skb)
2749 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2751 qdisc_skb_cb(skb)->pkt_len = skb->len;
2753 /* To get more precise estimation of bytes sent on wire,
2754 * we add to pkt_len the headers size of all segments
2756 if (shinfo->gso_size) {
2757 unsigned int hdr_len;
2758 u16 gso_segs = shinfo->gso_segs;
2760 /* mac layer + network layer */
2761 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2763 /* + transport layer */
2764 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2765 hdr_len += tcp_hdrlen(skb);
2767 hdr_len += sizeof(struct udphdr);
2769 if (shinfo->gso_type & SKB_GSO_DODGY)
2770 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2773 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2777 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2778 struct net_device *dev,
2779 struct netdev_queue *txq)
2781 spinlock_t *root_lock = qdisc_lock(q);
2785 qdisc_pkt_len_init(skb);
2786 qdisc_calculate_pkt_len(skb, q);
2788 * Heuristic to force contended enqueues to serialize on a
2789 * separate lock before trying to get qdisc main lock.
2790 * This permits __QDISC___STATE_RUNNING owner to get the lock more
2791 * often and dequeue packets faster.
2793 contended = qdisc_is_running(q);
2794 if (unlikely(contended))
2795 spin_lock(&q->busylock);
2797 spin_lock(root_lock);
2798 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2801 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2802 qdisc_run_begin(q)) {
2804 * This is a work-conserving queue; there are no old skbs
2805 * waiting to be sent out; and the qdisc is not running -
2806 * xmit the skb directly.
2809 qdisc_bstats_update(q, skb);
2811 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
2812 if (unlikely(contended)) {
2813 spin_unlock(&q->busylock);
2820 rc = NET_XMIT_SUCCESS;
2822 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2823 if (qdisc_run_begin(q)) {
2824 if (unlikely(contended)) {
2825 spin_unlock(&q->busylock);
2831 spin_unlock(root_lock);
2832 if (unlikely(contended))
2833 spin_unlock(&q->busylock);
2837 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2838 static void skb_update_prio(struct sk_buff *skb)
2840 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2842 if (!skb->priority && skb->sk && map) {
2843 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2845 if (prioidx < map->priomap_len)
2846 skb->priority = map->priomap[prioidx];
2850 #define skb_update_prio(skb)
2853 static DEFINE_PER_CPU(int, xmit_recursion);
2854 #define RECURSION_LIMIT 10
2857 * dev_loopback_xmit - loop back @skb
2858 * @skb: buffer to transmit
2860 int dev_loopback_xmit(struct sk_buff *skb)
2862 skb_reset_mac_header(skb);
2863 __skb_pull(skb, skb_network_offset(skb));
2864 skb->pkt_type = PACKET_LOOPBACK;
2865 skb->ip_summed = CHECKSUM_UNNECESSARY;
2866 WARN_ON(!skb_dst(skb));
2871 EXPORT_SYMBOL(dev_loopback_xmit);
2874 * __dev_queue_xmit - transmit a buffer
2875 * @skb: buffer to transmit
2876 * @accel_priv: private data used for L2 forwarding offload
2878 * Queue a buffer for transmission to a network device. The caller must
2879 * have set the device and priority and built the buffer before calling
2880 * this function. The function can be called from an interrupt.
2882 * A negative errno code is returned on a failure. A success does not
2883 * guarantee the frame will be transmitted as it may be dropped due
2884 * to congestion or traffic shaping.
2886 * -----------------------------------------------------------------------------------
2887 * I notice this method can also return errors from the queue disciplines,
2888 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2891 * Regardless of the return value, the skb is consumed, so it is currently
2892 * difficult to retry a send to this method. (You can bump the ref count
2893 * before sending to hold a reference for retry if you are careful.)
2895 * When calling this method, interrupts MUST be enabled. This is because
2896 * the BH enable code must have IRQs enabled so that it will not deadlock.
2899 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2901 struct net_device *dev = skb->dev;
2902 struct netdev_queue *txq;
2906 skb_reset_mac_header(skb);
2908 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
2909 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
2911 /* Disable soft irqs for various locks below. Also
2912 * stops preemption for RCU.
2916 skb_update_prio(skb);
2918 /* If device/qdisc don't need skb->dst, release it right now while
2919 * its hot in this cpu cache.
2921 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2926 txq = netdev_pick_tx(dev, skb, accel_priv);
2927 q = rcu_dereference_bh(txq->qdisc);
2929 #ifdef CONFIG_NET_CLS_ACT
2930 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2932 trace_net_dev_queue(skb);
2934 rc = __dev_xmit_skb(skb, q, dev, txq);
2938 /* The device has no queue. Common case for software devices:
2939 loopback, all the sorts of tunnels...
2941 Really, it is unlikely that netif_tx_lock protection is necessary
2942 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2944 However, it is possible, that they rely on protection
2947 Check this and shot the lock. It is not prone from deadlocks.
2948 Either shot noqueue qdisc, it is even simpler 8)
2950 if (dev->flags & IFF_UP) {
2951 int cpu = smp_processor_id(); /* ok because BHs are off */
2953 if (txq->xmit_lock_owner != cpu) {
2955 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2956 goto recursion_alert;
2958 skb = validate_xmit_skb(skb, dev);
2962 HARD_TX_LOCK(dev, txq, cpu);
2964 if (!netif_xmit_stopped(txq)) {
2965 __this_cpu_inc(xmit_recursion);
2966 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
2967 __this_cpu_dec(xmit_recursion);
2968 if (dev_xmit_complete(rc)) {
2969 HARD_TX_UNLOCK(dev, txq);
2973 HARD_TX_UNLOCK(dev, txq);
2974 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2977 /* Recursion is detected! It is possible,
2981 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2988 rcu_read_unlock_bh();
2990 atomic_long_inc(&dev->tx_dropped);
2991 kfree_skb_list(skb);
2994 rcu_read_unlock_bh();
2998 int dev_queue_xmit(struct sk_buff *skb)
3000 return __dev_queue_xmit(skb, NULL);
3002 EXPORT_SYMBOL(dev_queue_xmit);
3004 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3006 return __dev_queue_xmit(skb, accel_priv);
3008 EXPORT_SYMBOL(dev_queue_xmit_accel);
3011 /*=======================================================================
3013 =======================================================================*/
3015 int netdev_max_backlog __read_mostly = 1000;
3016 EXPORT_SYMBOL(netdev_max_backlog);
3018 int netdev_tstamp_prequeue __read_mostly = 1;
3019 int netdev_budget __read_mostly = 300;
3020 int weight_p __read_mostly = 64; /* old backlog weight */
3022 /* Called with irq disabled */
3023 static inline void ____napi_schedule(struct softnet_data *sd,
3024 struct napi_struct *napi)
3026 list_add_tail(&napi->poll_list, &sd->poll_list);
3027 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3032 /* One global table that all flow-based protocols share. */
3033 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3034 EXPORT_SYMBOL(rps_sock_flow_table);
3036 struct static_key rps_needed __read_mostly;
3038 static struct rps_dev_flow *
3039 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3040 struct rps_dev_flow *rflow, u16 next_cpu)
3042 if (next_cpu != RPS_NO_CPU) {
3043 #ifdef CONFIG_RFS_ACCEL
3044 struct netdev_rx_queue *rxqueue;
3045 struct rps_dev_flow_table *flow_table;
3046 struct rps_dev_flow *old_rflow;
3051 /* Should we steer this flow to a different hardware queue? */
3052 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3053 !(dev->features & NETIF_F_NTUPLE))
3055 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3056 if (rxq_index == skb_get_rx_queue(skb))
3059 rxqueue = dev->_rx + rxq_index;
3060 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3063 flow_id = skb_get_hash(skb) & flow_table->mask;
3064 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3065 rxq_index, flow_id);
3069 rflow = &flow_table->flows[flow_id];
3071 if (old_rflow->filter == rflow->filter)
3072 old_rflow->filter = RPS_NO_FILTER;
3076 per_cpu(softnet_data, next_cpu).input_queue_head;
3079 rflow->cpu = next_cpu;
3084 * get_rps_cpu is called from netif_receive_skb and returns the target
3085 * CPU from the RPS map of the receiving queue for a given skb.
3086 * rcu_read_lock must be held on entry.
3088 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3089 struct rps_dev_flow **rflowp)
3091 struct netdev_rx_queue *rxqueue;
3092 struct rps_map *map;
3093 struct rps_dev_flow_table *flow_table;
3094 struct rps_sock_flow_table *sock_flow_table;
3099 if (skb_rx_queue_recorded(skb)) {
3100 u16 index = skb_get_rx_queue(skb);
3101 if (unlikely(index >= dev->real_num_rx_queues)) {
3102 WARN_ONCE(dev->real_num_rx_queues > 1,
3103 "%s received packet on queue %u, but number "
3104 "of RX queues is %u\n",
3105 dev->name, index, dev->real_num_rx_queues);
3108 rxqueue = dev->_rx + index;
3112 map = rcu_dereference(rxqueue->rps_map);
3114 if (map->len == 1 &&
3115 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3116 tcpu = map->cpus[0];
3117 if (cpu_online(tcpu))
3121 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3125 skb_reset_network_header(skb);
3126 hash = skb_get_hash(skb);
3130 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3131 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3132 if (flow_table && sock_flow_table) {
3134 struct rps_dev_flow *rflow;
3136 rflow = &flow_table->flows[hash & flow_table->mask];
3139 next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask];
3142 * If the desired CPU (where last recvmsg was done) is
3143 * different from current CPU (one in the rx-queue flow
3144 * table entry), switch if one of the following holds:
3145 * - Current CPU is unset (equal to RPS_NO_CPU).
3146 * - Current CPU is offline.
3147 * - The current CPU's queue tail has advanced beyond the
3148 * last packet that was enqueued using this table entry.
3149 * This guarantees that all previous packets for the flow
3150 * have been dequeued, thus preserving in order delivery.
3152 if (unlikely(tcpu != next_cpu) &&
3153 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3154 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3155 rflow->last_qtail)) >= 0)) {
3157 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3160 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3168 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3169 if (cpu_online(tcpu)) {
3179 #ifdef CONFIG_RFS_ACCEL
3182 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3183 * @dev: Device on which the filter was set
3184 * @rxq_index: RX queue index
3185 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3186 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3188 * Drivers that implement ndo_rx_flow_steer() should periodically call
3189 * this function for each installed filter and remove the filters for
3190 * which it returns %true.
3192 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3193 u32 flow_id, u16 filter_id)
3195 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3196 struct rps_dev_flow_table *flow_table;
3197 struct rps_dev_flow *rflow;
3202 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3203 if (flow_table && flow_id <= flow_table->mask) {
3204 rflow = &flow_table->flows[flow_id];
3205 cpu = ACCESS_ONCE(rflow->cpu);
3206 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3207 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3208 rflow->last_qtail) <
3209 (int)(10 * flow_table->mask)))
3215 EXPORT_SYMBOL(rps_may_expire_flow);
3217 #endif /* CONFIG_RFS_ACCEL */
3219 /* Called from hardirq (IPI) context */
3220 static void rps_trigger_softirq(void *data)
3222 struct softnet_data *sd = data;
3224 ____napi_schedule(sd, &sd->backlog);
3228 #endif /* CONFIG_RPS */
3231 * Check if this softnet_data structure is another cpu one
3232 * If yes, queue it to our IPI list and return 1
3235 static int rps_ipi_queued(struct softnet_data *sd)
3238 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3241 sd->rps_ipi_next = mysd->rps_ipi_list;
3242 mysd->rps_ipi_list = sd;
3244 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3247 #endif /* CONFIG_RPS */
3251 #ifdef CONFIG_NET_FLOW_LIMIT
3252 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3255 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3257 #ifdef CONFIG_NET_FLOW_LIMIT
3258 struct sd_flow_limit *fl;
3259 struct softnet_data *sd;
3260 unsigned int old_flow, new_flow;
3262 if (qlen < (netdev_max_backlog >> 1))
3265 sd = &__get_cpu_var(softnet_data);
3268 fl = rcu_dereference(sd->flow_limit);
3270 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3271 old_flow = fl->history[fl->history_head];
3272 fl->history[fl->history_head] = new_flow;
3275 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3277 if (likely(fl->buckets[old_flow]))
3278 fl->buckets[old_flow]--;
3280 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3292 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3293 * queue (may be a remote CPU queue).
3295 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3296 unsigned int *qtail)
3298 struct softnet_data *sd;
3299 unsigned long flags;
3302 sd = &per_cpu(softnet_data, cpu);
3304 local_irq_save(flags);
3307 qlen = skb_queue_len(&sd->input_pkt_queue);
3308 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3309 if (skb_queue_len(&sd->input_pkt_queue)) {
3311 __skb_queue_tail(&sd->input_pkt_queue, skb);
3312 input_queue_tail_incr_save(sd, qtail);
3314 local_irq_restore(flags);
3315 return NET_RX_SUCCESS;
3318 /* Schedule NAPI for backlog device
3319 * We can use non atomic operation since we own the queue lock
3321 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3322 if (!rps_ipi_queued(sd))
3323 ____napi_schedule(sd, &sd->backlog);
3331 local_irq_restore(flags);
3333 atomic_long_inc(&skb->dev->rx_dropped);
3338 static int netif_rx_internal(struct sk_buff *skb)
3342 net_timestamp_check(netdev_tstamp_prequeue, skb);
3344 trace_netif_rx(skb);
3346 if (static_key_false(&rps_needed)) {
3347 struct rps_dev_flow voidflow, *rflow = &voidflow;
3353 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3355 cpu = smp_processor_id();
3357 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3365 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3372 * netif_rx - post buffer to the network code
3373 * @skb: buffer to post
3375 * This function receives a packet from a device driver and queues it for
3376 * the upper (protocol) levels to process. It always succeeds. The buffer
3377 * may be dropped during processing for congestion control or by the
3381 * NET_RX_SUCCESS (no congestion)
3382 * NET_RX_DROP (packet was dropped)
3386 int netif_rx(struct sk_buff *skb)
3388 trace_netif_rx_entry(skb);
3390 return netif_rx_internal(skb);
3392 EXPORT_SYMBOL(netif_rx);
3394 int netif_rx_ni(struct sk_buff *skb)
3398 trace_netif_rx_ni_entry(skb);
3401 err = netif_rx_internal(skb);
3402 if (local_softirq_pending())
3408 EXPORT_SYMBOL(netif_rx_ni);
3410 static void net_tx_action(struct softirq_action *h)
3412 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3414 if (sd->completion_queue) {
3415 struct sk_buff *clist;
3417 local_irq_disable();
3418 clist = sd->completion_queue;
3419 sd->completion_queue = NULL;
3423 struct sk_buff *skb = clist;
3424 clist = clist->next;
3426 WARN_ON(atomic_read(&skb->users));
3427 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3428 trace_consume_skb(skb);
3430 trace_kfree_skb(skb, net_tx_action);
3435 if (sd->output_queue) {
3438 local_irq_disable();
3439 head = sd->output_queue;
3440 sd->output_queue = NULL;
3441 sd->output_queue_tailp = &sd->output_queue;
3445 struct Qdisc *q = head;
3446 spinlock_t *root_lock;
3448 head = head->next_sched;
3450 root_lock = qdisc_lock(q);
3451 if (spin_trylock(root_lock)) {
3452 smp_mb__before_atomic();
3453 clear_bit(__QDISC_STATE_SCHED,
3456 spin_unlock(root_lock);
3458 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3460 __netif_reschedule(q);
3462 smp_mb__before_atomic();
3463 clear_bit(__QDISC_STATE_SCHED,
3471 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3472 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3473 /* This hook is defined here for ATM LANE */
3474 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3475 unsigned char *addr) __read_mostly;
3476 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3479 #ifdef CONFIG_NET_CLS_ACT
3480 /* TODO: Maybe we should just force sch_ingress to be compiled in
3481 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3482 * a compare and 2 stores extra right now if we dont have it on
3483 * but have CONFIG_NET_CLS_ACT
3484 * NOTE: This doesn't stop any functionality; if you dont have
3485 * the ingress scheduler, you just can't add policies on ingress.
3488 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3490 struct net_device *dev = skb->dev;
3491 u32 ttl = G_TC_RTTL(skb->tc_verd);
3492 int result = TC_ACT_OK;
3495 if (unlikely(MAX_RED_LOOP < ttl++)) {
3496 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3497 skb->skb_iif, dev->ifindex);
3501 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3502 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3504 q = rcu_dereference(rxq->qdisc);
3505 if (q != &noop_qdisc) {
3506 spin_lock(qdisc_lock(q));
3507 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3508 result = qdisc_enqueue_root(skb, q);
3509 spin_unlock(qdisc_lock(q));
3515 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3516 struct packet_type **pt_prev,
3517 int *ret, struct net_device *orig_dev)
3519 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3521 if (!rxq || rcu_access_pointer(rxq->qdisc) == &noop_qdisc)
3525 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3529 switch (ing_filter(skb, rxq)) {
3543 * netdev_rx_handler_register - register receive handler
3544 * @dev: device to register a handler for
3545 * @rx_handler: receive handler to register
3546 * @rx_handler_data: data pointer that is used by rx handler
3548 * Register a receive handler for a device. This handler will then be
3549 * called from __netif_receive_skb. A negative errno code is returned
3552 * The caller must hold the rtnl_mutex.
3554 * For a general description of rx_handler, see enum rx_handler_result.
3556 int netdev_rx_handler_register(struct net_device *dev,
3557 rx_handler_func_t *rx_handler,
3558 void *rx_handler_data)
3562 if (dev->rx_handler)
3565 /* Note: rx_handler_data must be set before rx_handler */
3566 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3567 rcu_assign_pointer(dev->rx_handler, rx_handler);
3571 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3574 * netdev_rx_handler_unregister - unregister receive handler
3575 * @dev: device to unregister a handler from
3577 * Unregister a receive handler from a device.
3579 * The caller must hold the rtnl_mutex.
3581 void netdev_rx_handler_unregister(struct net_device *dev)
3585 RCU_INIT_POINTER(dev->rx_handler, NULL);
3586 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3587 * section has a guarantee to see a non NULL rx_handler_data
3591 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3593 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3596 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3597 * the special handling of PFMEMALLOC skbs.
3599 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3601 switch (skb->protocol) {
3602 case htons(ETH_P_ARP):
3603 case htons(ETH_P_IP):
3604 case htons(ETH_P_IPV6):
3605 case htons(ETH_P_8021Q):
3606 case htons(ETH_P_8021AD):
3613 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3615 struct packet_type *ptype, *pt_prev;
3616 rx_handler_func_t *rx_handler;
3617 struct net_device *orig_dev;
3618 struct net_device *null_or_dev;
3619 bool deliver_exact = false;
3620 int ret = NET_RX_DROP;
3623 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3625 trace_netif_receive_skb(skb);
3627 orig_dev = skb->dev;
3629 skb_reset_network_header(skb);
3630 if (!skb_transport_header_was_set(skb))
3631 skb_reset_transport_header(skb);
3632 skb_reset_mac_len(skb);
3639 skb->skb_iif = skb->dev->ifindex;
3641 __this_cpu_inc(softnet_data.processed);
3643 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3644 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3645 skb = skb_vlan_untag(skb);
3650 #ifdef CONFIG_NET_CLS_ACT
3651 if (skb->tc_verd & TC_NCLS) {
3652 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3660 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3661 if (!ptype->dev || ptype->dev == skb->dev) {
3663 ret = deliver_skb(skb, pt_prev, orig_dev);
3669 #ifdef CONFIG_NET_CLS_ACT
3670 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3676 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3679 if (vlan_tx_tag_present(skb)) {
3681 ret = deliver_skb(skb, pt_prev, orig_dev);
3684 if (vlan_do_receive(&skb))
3686 else if (unlikely(!skb))
3690 rx_handler = rcu_dereference(skb->dev->rx_handler);
3693 ret = deliver_skb(skb, pt_prev, orig_dev);
3696 switch (rx_handler(&skb)) {
3697 case RX_HANDLER_CONSUMED:
3698 ret = NET_RX_SUCCESS;
3700 case RX_HANDLER_ANOTHER:
3702 case RX_HANDLER_EXACT:
3703 deliver_exact = true;
3704 case RX_HANDLER_PASS:
3711 if (unlikely(vlan_tx_tag_present(skb))) {
3712 if (vlan_tx_tag_get_id(skb))
3713 skb->pkt_type = PACKET_OTHERHOST;
3714 /* Note: we might in the future use prio bits
3715 * and set skb->priority like in vlan_do_receive()
3716 * For the time being, just ignore Priority Code Point
3721 /* deliver only exact match when indicated */
3722 null_or_dev = deliver_exact ? skb->dev : NULL;
3724 type = skb->protocol;
3725 list_for_each_entry_rcu(ptype,
3726 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3727 if (ptype->type == type &&
3728 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3729 ptype->dev == orig_dev)) {
3731 ret = deliver_skb(skb, pt_prev, orig_dev);
3737 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3740 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3743 atomic_long_inc(&skb->dev->rx_dropped);
3745 /* Jamal, now you will not able to escape explaining
3746 * me how you were going to use this. :-)
3756 static int __netif_receive_skb(struct sk_buff *skb)
3760 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3761 unsigned long pflags = current->flags;
3764 * PFMEMALLOC skbs are special, they should
3765 * - be delivered to SOCK_MEMALLOC sockets only
3766 * - stay away from userspace
3767 * - have bounded memory usage
3769 * Use PF_MEMALLOC as this saves us from propagating the allocation
3770 * context down to all allocation sites.
3772 current->flags |= PF_MEMALLOC;
3773 ret = __netif_receive_skb_core(skb, true);
3774 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3776 ret = __netif_receive_skb_core(skb, false);
3781 static int netif_receive_skb_internal(struct sk_buff *skb)
3783 net_timestamp_check(netdev_tstamp_prequeue, skb);
3785 if (skb_defer_rx_timestamp(skb))
3786 return NET_RX_SUCCESS;
3789 if (static_key_false(&rps_needed)) {
3790 struct rps_dev_flow voidflow, *rflow = &voidflow;
3795 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3798 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3805 return __netif_receive_skb(skb);
3809 * netif_receive_skb - process receive buffer from network
3810 * @skb: buffer to process
3812 * netif_receive_skb() is the main receive data processing function.
3813 * It always succeeds. The buffer may be dropped during processing
3814 * for congestion control or by the protocol layers.
3816 * This function may only be called from softirq context and interrupts
3817 * should be enabled.
3819 * Return values (usually ignored):
3820 * NET_RX_SUCCESS: no congestion
3821 * NET_RX_DROP: packet was dropped
3823 int netif_receive_skb(struct sk_buff *skb)
3825 trace_netif_receive_skb_entry(skb);
3827 return netif_receive_skb_internal(skb);
3829 EXPORT_SYMBOL(netif_receive_skb);
3831 /* Network device is going away, flush any packets still pending
3832 * Called with irqs disabled.
3834 static void flush_backlog(void *arg)
3836 struct net_device *dev = arg;
3837 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3838 struct sk_buff *skb, *tmp;
3841 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3842 if (skb->dev == dev) {
3843 __skb_unlink(skb, &sd->input_pkt_queue);
3845 input_queue_head_incr(sd);
3850 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3851 if (skb->dev == dev) {
3852 __skb_unlink(skb, &sd->process_queue);
3854 input_queue_head_incr(sd);
3859 static int napi_gro_complete(struct sk_buff *skb)
3861 struct packet_offload *ptype;
3862 __be16 type = skb->protocol;
3863 struct list_head *head = &offload_base;
3866 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3868 if (NAPI_GRO_CB(skb)->count == 1) {
3869 skb_shinfo(skb)->gso_size = 0;
3874 list_for_each_entry_rcu(ptype, head, list) {
3875 if (ptype->type != type || !ptype->callbacks.gro_complete)
3878 err = ptype->callbacks.gro_complete(skb, 0);
3884 WARN_ON(&ptype->list == head);
3886 return NET_RX_SUCCESS;
3890 return netif_receive_skb_internal(skb);
3893 /* napi->gro_list contains packets ordered by age.
3894 * youngest packets at the head of it.
3895 * Complete skbs in reverse order to reduce latencies.
3897 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3899 struct sk_buff *skb, *prev = NULL;
3901 /* scan list and build reverse chain */
3902 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3907 for (skb = prev; skb; skb = prev) {
3910 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3914 napi_gro_complete(skb);
3918 napi->gro_list = NULL;
3920 EXPORT_SYMBOL(napi_gro_flush);
3922 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3925 unsigned int maclen = skb->dev->hard_header_len;
3926 u32 hash = skb_get_hash_raw(skb);
3928 for (p = napi->gro_list; p; p = p->next) {
3929 unsigned long diffs;
3931 NAPI_GRO_CB(p)->flush = 0;
3933 if (hash != skb_get_hash_raw(p)) {
3934 NAPI_GRO_CB(p)->same_flow = 0;
3938 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3939 diffs |= p->vlan_tci ^ skb->vlan_tci;
3940 if (maclen == ETH_HLEN)
3941 diffs |= compare_ether_header(skb_mac_header(p),
3942 skb_mac_header(skb));
3944 diffs = memcmp(skb_mac_header(p),
3945 skb_mac_header(skb),
3947 NAPI_GRO_CB(p)->same_flow = !diffs;
3951 static void skb_gro_reset_offset(struct sk_buff *skb)
3953 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3954 const skb_frag_t *frag0 = &pinfo->frags[0];
3956 NAPI_GRO_CB(skb)->data_offset = 0;
3957 NAPI_GRO_CB(skb)->frag0 = NULL;
3958 NAPI_GRO_CB(skb)->frag0_len = 0;
3960 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3962 !PageHighMem(skb_frag_page(frag0))) {
3963 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3964 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3968 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
3970 struct skb_shared_info *pinfo = skb_shinfo(skb);
3972 BUG_ON(skb->end - skb->tail < grow);
3974 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3976 skb->data_len -= grow;
3979 pinfo->frags[0].page_offset += grow;
3980 skb_frag_size_sub(&pinfo->frags[0], grow);
3982 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
3983 skb_frag_unref(skb, 0);
3984 memmove(pinfo->frags, pinfo->frags + 1,
3985 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
3989 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3991 struct sk_buff **pp = NULL;
3992 struct packet_offload *ptype;
3993 __be16 type = skb->protocol;
3994 struct list_head *head = &offload_base;
3996 enum gro_result ret;
3999 if (!(skb->dev->features & NETIF_F_GRO))
4002 if (skb_is_gso(skb) || skb_has_frag_list(skb) || skb->csum_bad)
4005 gro_list_prepare(napi, skb);
4008 list_for_each_entry_rcu(ptype, head, list) {
4009 if (ptype->type != type || !ptype->callbacks.gro_receive)
4012 skb_set_network_header(skb, skb_gro_offset(skb));
4013 skb_reset_mac_len(skb);
4014 NAPI_GRO_CB(skb)->same_flow = 0;
4015 NAPI_GRO_CB(skb)->flush = 0;
4016 NAPI_GRO_CB(skb)->free = 0;
4017 NAPI_GRO_CB(skb)->udp_mark = 0;
4019 /* Setup for GRO checksum validation */
4020 switch (skb->ip_summed) {
4021 case CHECKSUM_COMPLETE:
4022 NAPI_GRO_CB(skb)->csum = skb->csum;
4023 NAPI_GRO_CB(skb)->csum_valid = 1;
4024 NAPI_GRO_CB(skb)->csum_cnt = 0;
4026 case CHECKSUM_UNNECESSARY:
4027 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
4028 NAPI_GRO_CB(skb)->csum_valid = 0;
4031 NAPI_GRO_CB(skb)->csum_cnt = 0;
4032 NAPI_GRO_CB(skb)->csum_valid = 0;
4035 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
4040 if (&ptype->list == head)
4043 same_flow = NAPI_GRO_CB(skb)->same_flow;
4044 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
4047 struct sk_buff *nskb = *pp;
4051 napi_gro_complete(nskb);
4058 if (NAPI_GRO_CB(skb)->flush)
4061 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
4062 struct sk_buff *nskb = napi->gro_list;
4064 /* locate the end of the list to select the 'oldest' flow */
4065 while (nskb->next) {
4071 napi_gro_complete(nskb);
4075 NAPI_GRO_CB(skb)->count = 1;
4076 NAPI_GRO_CB(skb)->age = jiffies;
4077 NAPI_GRO_CB(skb)->last = skb;
4078 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
4079 skb->next = napi->gro_list;
4080 napi->gro_list = skb;
4084 grow = skb_gro_offset(skb) - skb_headlen(skb);
4086 gro_pull_from_frag0(skb, grow);
4095 struct packet_offload *gro_find_receive_by_type(__be16 type)
4097 struct list_head *offload_head = &offload_base;
4098 struct packet_offload *ptype;
4100 list_for_each_entry_rcu(ptype, offload_head, list) {
4101 if (ptype->type != type || !ptype->callbacks.gro_receive)
4107 EXPORT_SYMBOL(gro_find_receive_by_type);
4109 struct packet_offload *gro_find_complete_by_type(__be16 type)
4111 struct list_head *offload_head = &offload_base;
4112 struct packet_offload *ptype;
4114 list_for_each_entry_rcu(ptype, offload_head, list) {
4115 if (ptype->type != type || !ptype->callbacks.gro_complete)
4121 EXPORT_SYMBOL(gro_find_complete_by_type);
4123 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4127 if (netif_receive_skb_internal(skb))
4135 case GRO_MERGED_FREE:
4136 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4137 kmem_cache_free(skbuff_head_cache, skb);
4150 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4152 trace_napi_gro_receive_entry(skb);
4154 skb_gro_reset_offset(skb);
4156 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4158 EXPORT_SYMBOL(napi_gro_receive);
4160 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4162 __skb_pull(skb, skb_headlen(skb));
4163 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4164 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4166 skb->dev = napi->dev;
4168 skb->encapsulation = 0;
4169 skb_shinfo(skb)->gso_type = 0;
4170 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4175 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4177 struct sk_buff *skb = napi->skb;
4180 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4185 EXPORT_SYMBOL(napi_get_frags);
4187 static gro_result_t napi_frags_finish(struct napi_struct *napi,
4188 struct sk_buff *skb,
4194 __skb_push(skb, ETH_HLEN);
4195 skb->protocol = eth_type_trans(skb, skb->dev);
4196 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
4201 case GRO_MERGED_FREE:
4202 napi_reuse_skb(napi, skb);
4212 /* Upper GRO stack assumes network header starts at gro_offset=0
4213 * Drivers could call both napi_gro_frags() and napi_gro_receive()
4214 * We copy ethernet header into skb->data to have a common layout.
4216 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4218 struct sk_buff *skb = napi->skb;
4219 const struct ethhdr *eth;
4220 unsigned int hlen = sizeof(*eth);
4224 skb_reset_mac_header(skb);
4225 skb_gro_reset_offset(skb);
4227 eth = skb_gro_header_fast(skb, 0);
4228 if (unlikely(skb_gro_header_hard(skb, hlen))) {
4229 eth = skb_gro_header_slow(skb, hlen, 0);
4230 if (unlikely(!eth)) {
4231 napi_reuse_skb(napi, skb);
4235 gro_pull_from_frag0(skb, hlen);
4236 NAPI_GRO_CB(skb)->frag0 += hlen;
4237 NAPI_GRO_CB(skb)->frag0_len -= hlen;
4239 __skb_pull(skb, hlen);
4242 * This works because the only protocols we care about don't require
4244 * We'll fix it up properly in napi_frags_finish()
4246 skb->protocol = eth->h_proto;
4251 gro_result_t napi_gro_frags(struct napi_struct *napi)
4253 struct sk_buff *skb = napi_frags_skb(napi);
4258 trace_napi_gro_frags_entry(skb);
4260 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4262 EXPORT_SYMBOL(napi_gro_frags);
4264 /* Compute the checksum from gro_offset and return the folded value
4265 * after adding in any pseudo checksum.
4267 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
4272 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
4274 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
4275 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
4277 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
4278 !skb->csum_complete_sw)
4279 netdev_rx_csum_fault(skb->dev);
4282 NAPI_GRO_CB(skb)->csum = wsum;
4283 NAPI_GRO_CB(skb)->csum_valid = 1;
4287 EXPORT_SYMBOL(__skb_gro_checksum_complete);
4290 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4291 * Note: called with local irq disabled, but exits with local irq enabled.
4293 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4296 struct softnet_data *remsd = sd->rps_ipi_list;
4299 sd->rps_ipi_list = NULL;
4303 /* Send pending IPI's to kick RPS processing on remote cpus. */
4305 struct softnet_data *next = remsd->rps_ipi_next;
4307 if (cpu_online(remsd->cpu))
4308 smp_call_function_single_async(remsd->cpu,
4317 static int process_backlog(struct napi_struct *napi, int quota)
4320 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4323 /* Check if we have pending ipi, its better to send them now,
4324 * not waiting net_rx_action() end.
4326 if (sd->rps_ipi_list) {
4327 local_irq_disable();
4328 net_rps_action_and_irq_enable(sd);
4331 napi->weight = weight_p;
4332 local_irq_disable();
4334 struct sk_buff *skb;
4336 while ((skb = __skb_dequeue(&sd->process_queue))) {
4338 __netif_receive_skb(skb);
4339 local_irq_disable();
4340 input_queue_head_incr(sd);
4341 if (++work >= quota) {
4348 if (skb_queue_empty(&sd->input_pkt_queue)) {
4350 * Inline a custom version of __napi_complete().
4351 * only current cpu owns and manipulates this napi,
4352 * and NAPI_STATE_SCHED is the only possible flag set
4354 * We can use a plain write instead of clear_bit(),
4355 * and we dont need an smp_mb() memory barrier.
4357 list_del(&napi->poll_list);
4364 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4365 &sd->process_queue);
4374 * __napi_schedule - schedule for receive
4375 * @n: entry to schedule
4377 * The entry's receive function will be scheduled to run
4379 void __napi_schedule(struct napi_struct *n)
4381 unsigned long flags;
4383 local_irq_save(flags);
4384 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4385 local_irq_restore(flags);
4387 EXPORT_SYMBOL(__napi_schedule);
4389 void __napi_complete(struct napi_struct *n)
4391 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4392 BUG_ON(n->gro_list);
4394 list_del(&n->poll_list);
4395 smp_mb__before_atomic();
4396 clear_bit(NAPI_STATE_SCHED, &n->state);
4398 EXPORT_SYMBOL(__napi_complete);
4400 void napi_complete(struct napi_struct *n)
4402 unsigned long flags;
4405 * don't let napi dequeue from the cpu poll list
4406 * just in case its running on a different cpu
4408 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4411 napi_gro_flush(n, false);
4412 local_irq_save(flags);
4414 local_irq_restore(flags);
4416 EXPORT_SYMBOL(napi_complete);
4418 /* must be called under rcu_read_lock(), as we dont take a reference */
4419 struct napi_struct *napi_by_id(unsigned int napi_id)
4421 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4422 struct napi_struct *napi;
4424 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4425 if (napi->napi_id == napi_id)
4430 EXPORT_SYMBOL_GPL(napi_by_id);
4432 void napi_hash_add(struct napi_struct *napi)
4434 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4436 spin_lock(&napi_hash_lock);
4438 /* 0 is not a valid id, we also skip an id that is taken
4439 * we expect both events to be extremely rare
4442 while (!napi->napi_id) {
4443 napi->napi_id = ++napi_gen_id;
4444 if (napi_by_id(napi->napi_id))
4448 hlist_add_head_rcu(&napi->napi_hash_node,
4449 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4451 spin_unlock(&napi_hash_lock);
4454 EXPORT_SYMBOL_GPL(napi_hash_add);
4456 /* Warning : caller is responsible to make sure rcu grace period
4457 * is respected before freeing memory containing @napi
4459 void napi_hash_del(struct napi_struct *napi)
4461 spin_lock(&napi_hash_lock);
4463 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4464 hlist_del_rcu(&napi->napi_hash_node);
4466 spin_unlock(&napi_hash_lock);
4468 EXPORT_SYMBOL_GPL(napi_hash_del);
4470 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4471 int (*poll)(struct napi_struct *, int), int weight)
4473 INIT_LIST_HEAD(&napi->poll_list);
4474 napi->gro_count = 0;
4475 napi->gro_list = NULL;
4478 if (weight > NAPI_POLL_WEIGHT)
4479 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4481 napi->weight = weight;
4482 list_add(&napi->dev_list, &dev->napi_list);
4484 #ifdef CONFIG_NETPOLL
4485 spin_lock_init(&napi->poll_lock);
4486 napi->poll_owner = -1;
4488 set_bit(NAPI_STATE_SCHED, &napi->state);
4490 EXPORT_SYMBOL(netif_napi_add);
4492 void netif_napi_del(struct napi_struct *napi)
4494 list_del_init(&napi->dev_list);
4495 napi_free_frags(napi);
4497 kfree_skb_list(napi->gro_list);
4498 napi->gro_list = NULL;
4499 napi->gro_count = 0;
4501 EXPORT_SYMBOL(netif_napi_del);
4503 static void net_rx_action(struct softirq_action *h)
4505 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4506 unsigned long time_limit = jiffies + 2;
4507 int budget = netdev_budget;
4510 local_irq_disable();
4512 while (!list_empty(&sd->poll_list)) {
4513 struct napi_struct *n;
4516 /* If softirq window is exhuasted then punt.
4517 * Allow this to run for 2 jiffies since which will allow
4518 * an average latency of 1.5/HZ.
4520 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4525 /* Even though interrupts have been re-enabled, this
4526 * access is safe because interrupts can only add new
4527 * entries to the tail of this list, and only ->poll()
4528 * calls can remove this head entry from the list.
4530 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4532 have = netpoll_poll_lock(n);
4536 /* This NAPI_STATE_SCHED test is for avoiding a race
4537 * with netpoll's poll_napi(). Only the entity which
4538 * obtains the lock and sees NAPI_STATE_SCHED set will
4539 * actually make the ->poll() call. Therefore we avoid
4540 * accidentally calling ->poll() when NAPI is not scheduled.
4543 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4544 work = n->poll(n, weight);
4548 WARN_ON_ONCE(work > weight);
4552 local_irq_disable();
4554 /* Drivers must not modify the NAPI state if they
4555 * consume the entire weight. In such cases this code
4556 * still "owns" the NAPI instance and therefore can
4557 * move the instance around on the list at-will.
4559 if (unlikely(work == weight)) {
4560 if (unlikely(napi_disable_pending(n))) {
4563 local_irq_disable();
4566 /* flush too old packets
4567 * If HZ < 1000, flush all packets.
4570 napi_gro_flush(n, HZ >= 1000);
4571 local_irq_disable();
4573 list_move_tail(&n->poll_list, &sd->poll_list);
4577 netpoll_poll_unlock(have);
4580 net_rps_action_and_irq_enable(sd);
4582 #ifdef CONFIG_NET_DMA
4584 * There may not be any more sk_buffs coming right now, so push
4585 * any pending DMA copies to hardware
4587 dma_issue_pending_all();
4594 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4598 struct netdev_adjacent {
4599 struct net_device *dev;
4601 /* upper master flag, there can only be one master device per list */
4604 /* counter for the number of times this device was added to us */
4607 /* private field for the users */
4610 struct list_head list;
4611 struct rcu_head rcu;
4614 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4615 struct net_device *adj_dev,
4616 struct list_head *adj_list)
4618 struct netdev_adjacent *adj;
4620 list_for_each_entry(adj, adj_list, list) {
4621 if (adj->dev == adj_dev)
4628 * netdev_has_upper_dev - Check if device is linked to an upper device
4630 * @upper_dev: upper device to check
4632 * Find out if a device is linked to specified upper device and return true
4633 * in case it is. Note that this checks only immediate upper device,
4634 * not through a complete stack of devices. The caller must hold the RTNL lock.
4636 bool netdev_has_upper_dev(struct net_device *dev,
4637 struct net_device *upper_dev)
4641 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4643 EXPORT_SYMBOL(netdev_has_upper_dev);
4646 * netdev_has_any_upper_dev - Check if device is linked to some device
4649 * Find out if a device is linked to an upper device and return true in case
4650 * it is. The caller must hold the RTNL lock.
4652 static bool netdev_has_any_upper_dev(struct net_device *dev)
4656 return !list_empty(&dev->all_adj_list.upper);
4660 * netdev_master_upper_dev_get - Get master upper device
4663 * Find a master upper device and return pointer to it or NULL in case
4664 * it's not there. The caller must hold the RTNL lock.
4666 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4668 struct netdev_adjacent *upper;
4672 if (list_empty(&dev->adj_list.upper))
4675 upper = list_first_entry(&dev->adj_list.upper,
4676 struct netdev_adjacent, list);
4677 if (likely(upper->master))
4681 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4683 void *netdev_adjacent_get_private(struct list_head *adj_list)
4685 struct netdev_adjacent *adj;
4687 adj = list_entry(adj_list, struct netdev_adjacent, list);
4689 return adj->private;
4691 EXPORT_SYMBOL(netdev_adjacent_get_private);
4694 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
4696 * @iter: list_head ** of the current position
4698 * Gets the next device from the dev's upper list, starting from iter
4699 * position. The caller must hold RCU read lock.
4701 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
4702 struct list_head **iter)
4704 struct netdev_adjacent *upper;
4706 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4708 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4710 if (&upper->list == &dev->adj_list.upper)
4713 *iter = &upper->list;
4717 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
4720 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4722 * @iter: list_head ** of the current position
4724 * Gets the next device from the dev's upper list, starting from iter
4725 * position. The caller must hold RCU read lock.
4727 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4728 struct list_head **iter)
4730 struct netdev_adjacent *upper;
4732 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4734 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4736 if (&upper->list == &dev->all_adj_list.upper)
4739 *iter = &upper->list;
4743 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4746 * netdev_lower_get_next_private - Get the next ->private from the
4747 * lower neighbour list
4749 * @iter: list_head ** of the current position
4751 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4752 * list, starting from iter position. The caller must hold either hold the
4753 * RTNL lock or its own locking that guarantees that the neighbour lower
4754 * list will remain unchainged.
4756 void *netdev_lower_get_next_private(struct net_device *dev,
4757 struct list_head **iter)
4759 struct netdev_adjacent *lower;
4761 lower = list_entry(*iter, struct netdev_adjacent, list);
4763 if (&lower->list == &dev->adj_list.lower)
4766 *iter = lower->list.next;
4768 return lower->private;
4770 EXPORT_SYMBOL(netdev_lower_get_next_private);
4773 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4774 * lower neighbour list, RCU
4777 * @iter: list_head ** of the current position
4779 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4780 * list, starting from iter position. The caller must hold RCU read lock.
4782 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4783 struct list_head **iter)
4785 struct netdev_adjacent *lower;
4787 WARN_ON_ONCE(!rcu_read_lock_held());
4789 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4791 if (&lower->list == &dev->adj_list.lower)
4794 *iter = &lower->list;
4796 return lower->private;
4798 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4801 * netdev_lower_get_next - Get the next device from the lower neighbour
4804 * @iter: list_head ** of the current position
4806 * Gets the next netdev_adjacent from the dev's lower neighbour
4807 * list, starting from iter position. The caller must hold RTNL lock or
4808 * its own locking that guarantees that the neighbour lower
4809 * list will remain unchainged.
4811 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
4813 struct netdev_adjacent *lower;
4815 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
4817 if (&lower->list == &dev->adj_list.lower)
4820 *iter = &lower->list;
4824 EXPORT_SYMBOL(netdev_lower_get_next);
4827 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4828 * lower neighbour list, RCU
4832 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4833 * list. The caller must hold RCU read lock.
4835 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4837 struct netdev_adjacent *lower;
4839 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4840 struct netdev_adjacent, list);
4842 return lower->private;
4845 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4848 * netdev_master_upper_dev_get_rcu - Get master upper device
4851 * Find a master upper device and return pointer to it or NULL in case
4852 * it's not there. The caller must hold the RCU read lock.
4854 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4856 struct netdev_adjacent *upper;
4858 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4859 struct netdev_adjacent, list);
4860 if (upper && likely(upper->master))
4864 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4866 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4867 struct net_device *adj_dev,
4868 struct list_head *dev_list)
4870 char linkname[IFNAMSIZ+7];
4871 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4872 "upper_%s" : "lower_%s", adj_dev->name);
4873 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4876 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4878 struct list_head *dev_list)
4880 char linkname[IFNAMSIZ+7];
4881 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4882 "upper_%s" : "lower_%s", name);
4883 sysfs_remove_link(&(dev->dev.kobj), linkname);
4886 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
4887 struct net_device *adj_dev,
4888 struct list_head *dev_list)
4890 return (dev_list == &dev->adj_list.upper ||
4891 dev_list == &dev->adj_list.lower) &&
4892 net_eq(dev_net(dev), dev_net(adj_dev));
4895 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4896 struct net_device *adj_dev,
4897 struct list_head *dev_list,
4898 void *private, bool master)
4900 struct netdev_adjacent *adj;
4903 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4910 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4915 adj->master = master;
4917 adj->private = private;
4920 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4921 adj_dev->name, dev->name, adj_dev->name);
4923 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
4924 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
4929 /* Ensure that master link is always the first item in list. */
4931 ret = sysfs_create_link(&(dev->dev.kobj),
4932 &(adj_dev->dev.kobj), "master");
4934 goto remove_symlinks;
4936 list_add_rcu(&adj->list, dev_list);
4938 list_add_tail_rcu(&adj->list, dev_list);
4944 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
4945 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4953 static void __netdev_adjacent_dev_remove(struct net_device *dev,
4954 struct net_device *adj_dev,
4955 struct list_head *dev_list)
4957 struct netdev_adjacent *adj;
4959 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4962 pr_err("tried to remove device %s from %s\n",
4963 dev->name, adj_dev->name);
4967 if (adj->ref_nr > 1) {
4968 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4975 sysfs_remove_link(&(dev->dev.kobj), "master");
4977 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
4978 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4980 list_del_rcu(&adj->list);
4981 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4982 adj_dev->name, dev->name, adj_dev->name);
4984 kfree_rcu(adj, rcu);
4987 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4988 struct net_device *upper_dev,
4989 struct list_head *up_list,
4990 struct list_head *down_list,
4991 void *private, bool master)
4995 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
5000 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
5003 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5010 static int __netdev_adjacent_dev_link(struct net_device *dev,
5011 struct net_device *upper_dev)
5013 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
5014 &dev->all_adj_list.upper,
5015 &upper_dev->all_adj_list.lower,
5019 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
5020 struct net_device *upper_dev,
5021 struct list_head *up_list,
5022 struct list_head *down_list)
5024 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
5025 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
5028 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
5029 struct net_device *upper_dev)
5031 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5032 &dev->all_adj_list.upper,
5033 &upper_dev->all_adj_list.lower);
5036 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
5037 struct net_device *upper_dev,
5038 void *private, bool master)
5040 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
5045 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
5046 &dev->adj_list.upper,
5047 &upper_dev->adj_list.lower,
5050 __netdev_adjacent_dev_unlink(dev, upper_dev);
5057 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
5058 struct net_device *upper_dev)
5060 __netdev_adjacent_dev_unlink(dev, upper_dev);
5061 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
5062 &dev->adj_list.upper,
5063 &upper_dev->adj_list.lower);
5066 static int __netdev_upper_dev_link(struct net_device *dev,
5067 struct net_device *upper_dev, bool master,
5070 struct netdev_adjacent *i, *j, *to_i, *to_j;
5075 if (dev == upper_dev)
5078 /* To prevent loops, check if dev is not upper device to upper_dev. */
5079 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
5082 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
5085 if (master && netdev_master_upper_dev_get(dev))
5088 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
5093 /* Now that we linked these devs, make all the upper_dev's
5094 * all_adj_list.upper visible to every dev's all_adj_list.lower an
5095 * versa, and don't forget the devices itself. All of these
5096 * links are non-neighbours.
5098 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5099 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5100 pr_debug("Interlinking %s with %s, non-neighbour\n",
5101 i->dev->name, j->dev->name);
5102 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
5108 /* add dev to every upper_dev's upper device */
5109 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5110 pr_debug("linking %s's upper device %s with %s\n",
5111 upper_dev->name, i->dev->name, dev->name);
5112 ret = __netdev_adjacent_dev_link(dev, i->dev);
5114 goto rollback_upper_mesh;
5117 /* add upper_dev to every dev's lower device */
5118 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5119 pr_debug("linking %s's lower device %s with %s\n", dev->name,
5120 i->dev->name, upper_dev->name);
5121 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
5123 goto rollback_lower_mesh;
5126 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5129 rollback_lower_mesh:
5131 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5134 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5139 rollback_upper_mesh:
5141 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
5144 __netdev_adjacent_dev_unlink(dev, i->dev);
5152 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
5153 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
5154 if (i == to_i && j == to_j)
5156 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5162 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5168 * netdev_upper_dev_link - Add a link to the upper device
5170 * @upper_dev: new upper device
5172 * Adds a link to device which is upper to this one. The caller must hold
5173 * the RTNL lock. On a failure a negative errno code is returned.
5174 * On success the reference counts are adjusted and the function
5177 int netdev_upper_dev_link(struct net_device *dev,
5178 struct net_device *upper_dev)
5180 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
5182 EXPORT_SYMBOL(netdev_upper_dev_link);
5185 * netdev_master_upper_dev_link - Add a master link to the upper device
5187 * @upper_dev: new upper device
5189 * Adds a link to device which is upper to this one. In this case, only
5190 * one master upper device can be linked, although other non-master devices
5191 * might be linked as well. The caller must hold the RTNL lock.
5192 * On a failure a negative errno code is returned. On success the reference
5193 * counts are adjusted and the function returns zero.
5195 int netdev_master_upper_dev_link(struct net_device *dev,
5196 struct net_device *upper_dev)
5198 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
5200 EXPORT_SYMBOL(netdev_master_upper_dev_link);
5202 int netdev_master_upper_dev_link_private(struct net_device *dev,
5203 struct net_device *upper_dev,
5206 return __netdev_upper_dev_link(dev, upper_dev, true, private);
5208 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
5211 * netdev_upper_dev_unlink - Removes a link to upper device
5213 * @upper_dev: new upper device
5215 * Removes a link to device which is upper to this one. The caller must hold
5218 void netdev_upper_dev_unlink(struct net_device *dev,
5219 struct net_device *upper_dev)
5221 struct netdev_adjacent *i, *j;
5224 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
5226 /* Here is the tricky part. We must remove all dev's lower
5227 * devices from all upper_dev's upper devices and vice
5228 * versa, to maintain the graph relationship.
5230 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5231 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
5232 __netdev_adjacent_dev_unlink(i->dev, j->dev);
5234 /* remove also the devices itself from lower/upper device
5237 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5238 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5240 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5241 __netdev_adjacent_dev_unlink(dev, i->dev);
5243 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5245 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5247 void netdev_adjacent_add_links(struct net_device *dev)
5249 struct netdev_adjacent *iter;
5251 struct net *net = dev_net(dev);
5253 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5254 if (!net_eq(net,dev_net(iter->dev)))
5256 netdev_adjacent_sysfs_add(iter->dev, dev,
5257 &iter->dev->adj_list.lower);
5258 netdev_adjacent_sysfs_add(dev, iter->dev,
5259 &dev->adj_list.upper);
5262 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5263 if (!net_eq(net,dev_net(iter->dev)))
5265 netdev_adjacent_sysfs_add(iter->dev, dev,
5266 &iter->dev->adj_list.upper);
5267 netdev_adjacent_sysfs_add(dev, iter->dev,
5268 &dev->adj_list.lower);
5272 void netdev_adjacent_del_links(struct net_device *dev)
5274 struct netdev_adjacent *iter;
5276 struct net *net = dev_net(dev);
5278 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5279 if (!net_eq(net,dev_net(iter->dev)))
5281 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5282 &iter->dev->adj_list.lower);
5283 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5284 &dev->adj_list.upper);
5287 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5288 if (!net_eq(net,dev_net(iter->dev)))
5290 netdev_adjacent_sysfs_del(iter->dev, dev->name,
5291 &iter->dev->adj_list.upper);
5292 netdev_adjacent_sysfs_del(dev, iter->dev->name,
5293 &dev->adj_list.lower);
5297 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5299 struct netdev_adjacent *iter;
5301 struct net *net = dev_net(dev);
5303 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5304 if (!net_eq(net,dev_net(iter->dev)))
5306 netdev_adjacent_sysfs_del(iter->dev, oldname,
5307 &iter->dev->adj_list.lower);
5308 netdev_adjacent_sysfs_add(iter->dev, dev,
5309 &iter->dev->adj_list.lower);
5312 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5313 if (!net_eq(net,dev_net(iter->dev)))
5315 netdev_adjacent_sysfs_del(iter->dev, oldname,
5316 &iter->dev->adj_list.upper);
5317 netdev_adjacent_sysfs_add(iter->dev, dev,
5318 &iter->dev->adj_list.upper);
5322 void *netdev_lower_dev_get_private(struct net_device *dev,
5323 struct net_device *lower_dev)
5325 struct netdev_adjacent *lower;
5329 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5333 return lower->private;
5335 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5338 int dev_get_nest_level(struct net_device *dev,
5339 bool (*type_check)(struct net_device *dev))
5341 struct net_device *lower = NULL;
5342 struct list_head *iter;
5348 netdev_for_each_lower_dev(dev, lower, iter) {
5349 nest = dev_get_nest_level(lower, type_check);
5350 if (max_nest < nest)
5354 if (type_check(dev))
5359 EXPORT_SYMBOL(dev_get_nest_level);
5361 static void dev_change_rx_flags(struct net_device *dev, int flags)
5363 const struct net_device_ops *ops = dev->netdev_ops;
5365 if (ops->ndo_change_rx_flags)
5366 ops->ndo_change_rx_flags(dev, flags);
5369 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5371 unsigned int old_flags = dev->flags;
5377 dev->flags |= IFF_PROMISC;
5378 dev->promiscuity += inc;
5379 if (dev->promiscuity == 0) {
5382 * If inc causes overflow, untouch promisc and return error.
5385 dev->flags &= ~IFF_PROMISC;
5387 dev->promiscuity -= inc;
5388 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5393 if (dev->flags != old_flags) {
5394 pr_info("device %s %s promiscuous mode\n",
5396 dev->flags & IFF_PROMISC ? "entered" : "left");
5397 if (audit_enabled) {
5398 current_uid_gid(&uid, &gid);
5399 audit_log(current->audit_context, GFP_ATOMIC,
5400 AUDIT_ANOM_PROMISCUOUS,
5401 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5402 dev->name, (dev->flags & IFF_PROMISC),
5403 (old_flags & IFF_PROMISC),
5404 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5405 from_kuid(&init_user_ns, uid),
5406 from_kgid(&init_user_ns, gid),
5407 audit_get_sessionid(current));
5410 dev_change_rx_flags(dev, IFF_PROMISC);
5413 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5418 * dev_set_promiscuity - update promiscuity count on a device
5422 * Add or remove promiscuity from a device. While the count in the device
5423 * remains above zero the interface remains promiscuous. Once it hits zero
5424 * the device reverts back to normal filtering operation. A negative inc
5425 * value is used to drop promiscuity on the device.
5426 * Return 0 if successful or a negative errno code on error.
5428 int dev_set_promiscuity(struct net_device *dev, int inc)
5430 unsigned int old_flags = dev->flags;
5433 err = __dev_set_promiscuity(dev, inc, true);
5436 if (dev->flags != old_flags)
5437 dev_set_rx_mode(dev);
5440 EXPORT_SYMBOL(dev_set_promiscuity);
5442 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5444 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5448 dev->flags |= IFF_ALLMULTI;
5449 dev->allmulti += inc;
5450 if (dev->allmulti == 0) {
5453 * If inc causes overflow, untouch allmulti and return error.
5456 dev->flags &= ~IFF_ALLMULTI;
5458 dev->allmulti -= inc;
5459 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5464 if (dev->flags ^ old_flags) {
5465 dev_change_rx_flags(dev, IFF_ALLMULTI);
5466 dev_set_rx_mode(dev);
5468 __dev_notify_flags(dev, old_flags,
5469 dev->gflags ^ old_gflags);
5475 * dev_set_allmulti - update allmulti count on a device
5479 * Add or remove reception of all multicast frames to a device. While the
5480 * count in the device remains above zero the interface remains listening
5481 * to all interfaces. Once it hits zero the device reverts back to normal
5482 * filtering operation. A negative @inc value is used to drop the counter
5483 * when releasing a resource needing all multicasts.
5484 * Return 0 if successful or a negative errno code on error.
5487 int dev_set_allmulti(struct net_device *dev, int inc)
5489 return __dev_set_allmulti(dev, inc, true);
5491 EXPORT_SYMBOL(dev_set_allmulti);
5494 * Upload unicast and multicast address lists to device and
5495 * configure RX filtering. When the device doesn't support unicast
5496 * filtering it is put in promiscuous mode while unicast addresses
5499 void __dev_set_rx_mode(struct net_device *dev)
5501 const struct net_device_ops *ops = dev->netdev_ops;
5503 /* dev_open will call this function so the list will stay sane. */
5504 if (!(dev->flags&IFF_UP))
5507 if (!netif_device_present(dev))
5510 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5511 /* Unicast addresses changes may only happen under the rtnl,
5512 * therefore calling __dev_set_promiscuity here is safe.
5514 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5515 __dev_set_promiscuity(dev, 1, false);
5516 dev->uc_promisc = true;
5517 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5518 __dev_set_promiscuity(dev, -1, false);
5519 dev->uc_promisc = false;
5523 if (ops->ndo_set_rx_mode)
5524 ops->ndo_set_rx_mode(dev);
5527 void dev_set_rx_mode(struct net_device *dev)
5529 netif_addr_lock_bh(dev);
5530 __dev_set_rx_mode(dev);
5531 netif_addr_unlock_bh(dev);
5535 * dev_get_flags - get flags reported to userspace
5538 * Get the combination of flag bits exported through APIs to userspace.
5540 unsigned int dev_get_flags(const struct net_device *dev)
5544 flags = (dev->flags & ~(IFF_PROMISC |
5549 (dev->gflags & (IFF_PROMISC |
5552 if (netif_running(dev)) {
5553 if (netif_oper_up(dev))
5554 flags |= IFF_RUNNING;
5555 if (netif_carrier_ok(dev))
5556 flags |= IFF_LOWER_UP;
5557 if (netif_dormant(dev))
5558 flags |= IFF_DORMANT;
5563 EXPORT_SYMBOL(dev_get_flags);
5565 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5567 unsigned int old_flags = dev->flags;
5573 * Set the flags on our device.
5576 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5577 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5579 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5583 * Load in the correct multicast list now the flags have changed.
5586 if ((old_flags ^ flags) & IFF_MULTICAST)
5587 dev_change_rx_flags(dev, IFF_MULTICAST);
5589 dev_set_rx_mode(dev);
5592 * Have we downed the interface. We handle IFF_UP ourselves
5593 * according to user attempts to set it, rather than blindly
5598 if ((old_flags ^ flags) & IFF_UP)
5599 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5601 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5602 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5603 unsigned int old_flags = dev->flags;
5605 dev->gflags ^= IFF_PROMISC;
5607 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5608 if (dev->flags != old_flags)
5609 dev_set_rx_mode(dev);
5612 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5613 is important. Some (broken) drivers set IFF_PROMISC, when
5614 IFF_ALLMULTI is requested not asking us and not reporting.
5616 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5617 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5619 dev->gflags ^= IFF_ALLMULTI;
5620 __dev_set_allmulti(dev, inc, false);
5626 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5627 unsigned int gchanges)
5629 unsigned int changes = dev->flags ^ old_flags;
5632 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5634 if (changes & IFF_UP) {
5635 if (dev->flags & IFF_UP)
5636 call_netdevice_notifiers(NETDEV_UP, dev);
5638 call_netdevice_notifiers(NETDEV_DOWN, dev);
5641 if (dev->flags & IFF_UP &&
5642 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5643 struct netdev_notifier_change_info change_info;
5645 change_info.flags_changed = changes;
5646 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5652 * dev_change_flags - change device settings
5654 * @flags: device state flags
5656 * Change settings on device based state flags. The flags are
5657 * in the userspace exported format.
5659 int dev_change_flags(struct net_device *dev, unsigned int flags)
5662 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5664 ret = __dev_change_flags(dev, flags);
5668 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5669 __dev_notify_flags(dev, old_flags, changes);
5672 EXPORT_SYMBOL(dev_change_flags);
5674 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5676 const struct net_device_ops *ops = dev->netdev_ops;
5678 if (ops->ndo_change_mtu)
5679 return ops->ndo_change_mtu(dev, new_mtu);
5686 * dev_set_mtu - Change maximum transfer unit
5688 * @new_mtu: new transfer unit
5690 * Change the maximum transfer size of the network device.
5692 int dev_set_mtu(struct net_device *dev, int new_mtu)
5696 if (new_mtu == dev->mtu)
5699 /* MTU must be positive. */
5703 if (!netif_device_present(dev))
5706 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5707 err = notifier_to_errno(err);
5711 orig_mtu = dev->mtu;
5712 err = __dev_set_mtu(dev, new_mtu);
5715 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5716 err = notifier_to_errno(err);
5718 /* setting mtu back and notifying everyone again,
5719 * so that they have a chance to revert changes.
5721 __dev_set_mtu(dev, orig_mtu);
5722 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5727 EXPORT_SYMBOL(dev_set_mtu);
5730 * dev_set_group - Change group this device belongs to
5732 * @new_group: group this device should belong to
5734 void dev_set_group(struct net_device *dev, int new_group)
5736 dev->group = new_group;
5738 EXPORT_SYMBOL(dev_set_group);
5741 * dev_set_mac_address - Change Media Access Control Address
5745 * Change the hardware (MAC) address of the device
5747 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5749 const struct net_device_ops *ops = dev->netdev_ops;
5752 if (!ops->ndo_set_mac_address)
5754 if (sa->sa_family != dev->type)
5756 if (!netif_device_present(dev))
5758 err = ops->ndo_set_mac_address(dev, sa);
5761 dev->addr_assign_type = NET_ADDR_SET;
5762 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5763 add_device_randomness(dev->dev_addr, dev->addr_len);
5766 EXPORT_SYMBOL(dev_set_mac_address);
5769 * dev_change_carrier - Change device carrier
5771 * @new_carrier: new value
5773 * Change device carrier
5775 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5777 const struct net_device_ops *ops = dev->netdev_ops;
5779 if (!ops->ndo_change_carrier)
5781 if (!netif_device_present(dev))
5783 return ops->ndo_change_carrier(dev, new_carrier);
5785 EXPORT_SYMBOL(dev_change_carrier);
5788 * dev_get_phys_port_id - Get device physical port ID
5792 * Get device physical port ID
5794 int dev_get_phys_port_id(struct net_device *dev,
5795 struct netdev_phys_port_id *ppid)
5797 const struct net_device_ops *ops = dev->netdev_ops;
5799 if (!ops->ndo_get_phys_port_id)
5801 return ops->ndo_get_phys_port_id(dev, ppid);
5803 EXPORT_SYMBOL(dev_get_phys_port_id);
5806 * dev_new_index - allocate an ifindex
5807 * @net: the applicable net namespace
5809 * Returns a suitable unique value for a new device interface
5810 * number. The caller must hold the rtnl semaphore or the
5811 * dev_base_lock to be sure it remains unique.
5813 static int dev_new_index(struct net *net)
5815 int ifindex = net->ifindex;
5819 if (!__dev_get_by_index(net, ifindex))
5820 return net->ifindex = ifindex;
5824 /* Delayed registration/unregisteration */
5825 static LIST_HEAD(net_todo_list);
5826 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5828 static void net_set_todo(struct net_device *dev)
5830 list_add_tail(&dev->todo_list, &net_todo_list);
5831 dev_net(dev)->dev_unreg_count++;
5834 static void rollback_registered_many(struct list_head *head)
5836 struct net_device *dev, *tmp;
5837 LIST_HEAD(close_head);
5839 BUG_ON(dev_boot_phase);
5842 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5843 /* Some devices call without registering
5844 * for initialization unwind. Remove those
5845 * devices and proceed with the remaining.
5847 if (dev->reg_state == NETREG_UNINITIALIZED) {
5848 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5852 list_del(&dev->unreg_list);
5855 dev->dismantle = true;
5856 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5859 /* If device is running, close it first. */
5860 list_for_each_entry(dev, head, unreg_list)
5861 list_add_tail(&dev->close_list, &close_head);
5862 dev_close_many(&close_head);
5864 list_for_each_entry(dev, head, unreg_list) {
5865 /* And unlink it from device chain. */
5866 unlist_netdevice(dev);
5868 dev->reg_state = NETREG_UNREGISTERING;
5873 list_for_each_entry(dev, head, unreg_list) {
5874 /* Shutdown queueing discipline. */
5878 /* Notify protocols, that we are about to destroy
5879 this device. They should clean all the things.
5881 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5884 * Flush the unicast and multicast chains
5889 if (dev->netdev_ops->ndo_uninit)
5890 dev->netdev_ops->ndo_uninit(dev);
5892 if (!dev->rtnl_link_ops ||
5893 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5894 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5896 /* Notifier chain MUST detach us all upper devices. */
5897 WARN_ON(netdev_has_any_upper_dev(dev));
5899 /* Remove entries from kobject tree */
5900 netdev_unregister_kobject(dev);
5902 /* Remove XPS queueing entries */
5903 netif_reset_xps_queues_gt(dev, 0);
5909 list_for_each_entry(dev, head, unreg_list)
5913 static void rollback_registered(struct net_device *dev)
5917 list_add(&dev->unreg_list, &single);
5918 rollback_registered_many(&single);
5922 static netdev_features_t netdev_fix_features(struct net_device *dev,
5923 netdev_features_t features)
5925 /* Fix illegal checksum combinations */
5926 if ((features & NETIF_F_HW_CSUM) &&
5927 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5928 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5929 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5932 /* TSO requires that SG is present as well. */
5933 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5934 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5935 features &= ~NETIF_F_ALL_TSO;
5938 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5939 !(features & NETIF_F_IP_CSUM)) {
5940 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5941 features &= ~NETIF_F_TSO;
5942 features &= ~NETIF_F_TSO_ECN;
5945 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5946 !(features & NETIF_F_IPV6_CSUM)) {
5947 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5948 features &= ~NETIF_F_TSO6;
5951 /* TSO ECN requires that TSO is present as well. */
5952 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5953 features &= ~NETIF_F_TSO_ECN;
5955 /* Software GSO depends on SG. */
5956 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5957 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5958 features &= ~NETIF_F_GSO;
5961 /* UFO needs SG and checksumming */
5962 if (features & NETIF_F_UFO) {
5963 /* maybe split UFO into V4 and V6? */
5964 if (!((features & NETIF_F_GEN_CSUM) ||
5965 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5966 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5968 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5969 features &= ~NETIF_F_UFO;
5972 if (!(features & NETIF_F_SG)) {
5974 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5975 features &= ~NETIF_F_UFO;
5979 #ifdef CONFIG_NET_RX_BUSY_POLL
5980 if (dev->netdev_ops->ndo_busy_poll)
5981 features |= NETIF_F_BUSY_POLL;
5984 features &= ~NETIF_F_BUSY_POLL;
5989 int __netdev_update_features(struct net_device *dev)
5991 netdev_features_t features;
5996 features = netdev_get_wanted_features(dev);
5998 if (dev->netdev_ops->ndo_fix_features)
5999 features = dev->netdev_ops->ndo_fix_features(dev, features);
6001 /* driver might be less strict about feature dependencies */
6002 features = netdev_fix_features(dev, features);
6004 if (dev->features == features)
6007 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
6008 &dev->features, &features);
6010 if (dev->netdev_ops->ndo_set_features)
6011 err = dev->netdev_ops->ndo_set_features(dev, features);
6013 if (unlikely(err < 0)) {
6015 "set_features() failed (%d); wanted %pNF, left %pNF\n",
6016 err, &features, &dev->features);
6021 dev->features = features;
6027 * netdev_update_features - recalculate device features
6028 * @dev: the device to check
6030 * Recalculate dev->features set and send notifications if it
6031 * has changed. Should be called after driver or hardware dependent
6032 * conditions might have changed that influence the features.
6034 void netdev_update_features(struct net_device *dev)
6036 if (__netdev_update_features(dev))
6037 netdev_features_change(dev);
6039 EXPORT_SYMBOL(netdev_update_features);
6042 * netdev_change_features - recalculate device features
6043 * @dev: the device to check
6045 * Recalculate dev->features set and send notifications even
6046 * if they have not changed. Should be called instead of
6047 * netdev_update_features() if also dev->vlan_features might
6048 * have changed to allow the changes to be propagated to stacked
6051 void netdev_change_features(struct net_device *dev)
6053 __netdev_update_features(dev);
6054 netdev_features_change(dev);
6056 EXPORT_SYMBOL(netdev_change_features);
6059 * netif_stacked_transfer_operstate - transfer operstate
6060 * @rootdev: the root or lower level device to transfer state from
6061 * @dev: the device to transfer operstate to
6063 * Transfer operational state from root to device. This is normally
6064 * called when a stacking relationship exists between the root
6065 * device and the device(a leaf device).
6067 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
6068 struct net_device *dev)
6070 if (rootdev->operstate == IF_OPER_DORMANT)
6071 netif_dormant_on(dev);
6073 netif_dormant_off(dev);
6075 if (netif_carrier_ok(rootdev)) {
6076 if (!netif_carrier_ok(dev))
6077 netif_carrier_on(dev);
6079 if (netif_carrier_ok(dev))
6080 netif_carrier_off(dev);
6083 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
6086 static int netif_alloc_rx_queues(struct net_device *dev)
6088 unsigned int i, count = dev->num_rx_queues;
6089 struct netdev_rx_queue *rx;
6093 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
6099 for (i = 0; i < count; i++)
6105 static void netdev_init_one_queue(struct net_device *dev,
6106 struct netdev_queue *queue, void *_unused)
6108 /* Initialize queue lock */
6109 spin_lock_init(&queue->_xmit_lock);
6110 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
6111 queue->xmit_lock_owner = -1;
6112 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
6115 dql_init(&queue->dql, HZ);
6119 static void netif_free_tx_queues(struct net_device *dev)
6124 static int netif_alloc_netdev_queues(struct net_device *dev)
6126 unsigned int count = dev->num_tx_queues;
6127 struct netdev_queue *tx;
6128 size_t sz = count * sizeof(*tx);
6130 BUG_ON(count < 1 || count > 0xffff);
6132 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6140 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
6141 spin_lock_init(&dev->tx_global_lock);
6147 * register_netdevice - register a network device
6148 * @dev: device to register
6150 * Take a completed network device structure and add it to the kernel
6151 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6152 * chain. 0 is returned on success. A negative errno code is returned
6153 * on a failure to set up the device, or if the name is a duplicate.
6155 * Callers must hold the rtnl semaphore. You may want
6156 * register_netdev() instead of this.
6159 * The locking appears insufficient to guarantee two parallel registers
6160 * will not get the same name.
6163 int register_netdevice(struct net_device *dev)
6166 struct net *net = dev_net(dev);
6168 BUG_ON(dev_boot_phase);
6173 /* When net_device's are persistent, this will be fatal. */
6174 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
6177 spin_lock_init(&dev->addr_list_lock);
6178 netdev_set_addr_lockdep_class(dev);
6182 ret = dev_get_valid_name(net, dev, dev->name);
6186 /* Init, if this function is available */
6187 if (dev->netdev_ops->ndo_init) {
6188 ret = dev->netdev_ops->ndo_init(dev);
6196 if (((dev->hw_features | dev->features) &
6197 NETIF_F_HW_VLAN_CTAG_FILTER) &&
6198 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
6199 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
6200 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
6207 dev->ifindex = dev_new_index(net);
6208 else if (__dev_get_by_index(net, dev->ifindex))
6211 if (dev->iflink == -1)
6212 dev->iflink = dev->ifindex;
6214 /* Transfer changeable features to wanted_features and enable
6215 * software offloads (GSO and GRO).
6217 dev->hw_features |= NETIF_F_SOFT_FEATURES;
6218 dev->features |= NETIF_F_SOFT_FEATURES;
6219 dev->wanted_features = dev->features & dev->hw_features;
6221 if (!(dev->flags & IFF_LOOPBACK)) {
6222 dev->hw_features |= NETIF_F_NOCACHE_COPY;
6225 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
6227 dev->vlan_features |= NETIF_F_HIGHDMA;
6229 /* Make NETIF_F_SG inheritable to tunnel devices.
6231 dev->hw_enc_features |= NETIF_F_SG;
6233 /* Make NETIF_F_SG inheritable to MPLS.
6235 dev->mpls_features |= NETIF_F_SG;
6237 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
6238 ret = notifier_to_errno(ret);
6242 ret = netdev_register_kobject(dev);
6245 dev->reg_state = NETREG_REGISTERED;
6247 __netdev_update_features(dev);
6250 * Default initial state at registry is that the
6251 * device is present.
6254 set_bit(__LINK_STATE_PRESENT, &dev->state);
6256 linkwatch_init_dev(dev);
6258 dev_init_scheduler(dev);
6260 list_netdevice(dev);
6261 add_device_randomness(dev->dev_addr, dev->addr_len);
6263 /* If the device has permanent device address, driver should
6264 * set dev_addr and also addr_assign_type should be set to
6265 * NET_ADDR_PERM (default value).
6267 if (dev->addr_assign_type == NET_ADDR_PERM)
6268 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
6270 /* Notify protocols, that a new device appeared. */
6271 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
6272 ret = notifier_to_errno(ret);
6274 rollback_registered(dev);
6275 dev->reg_state = NETREG_UNREGISTERED;
6278 * Prevent userspace races by waiting until the network
6279 * device is fully setup before sending notifications.
6281 if (!dev->rtnl_link_ops ||
6282 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
6283 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6289 if (dev->netdev_ops->ndo_uninit)
6290 dev->netdev_ops->ndo_uninit(dev);
6293 EXPORT_SYMBOL(register_netdevice);
6296 * init_dummy_netdev - init a dummy network device for NAPI
6297 * @dev: device to init
6299 * This takes a network device structure and initialize the minimum
6300 * amount of fields so it can be used to schedule NAPI polls without
6301 * registering a full blown interface. This is to be used by drivers
6302 * that need to tie several hardware interfaces to a single NAPI
6303 * poll scheduler due to HW limitations.
6305 int init_dummy_netdev(struct net_device *dev)
6307 /* Clear everything. Note we don't initialize spinlocks
6308 * are they aren't supposed to be taken by any of the
6309 * NAPI code and this dummy netdev is supposed to be
6310 * only ever used for NAPI polls
6312 memset(dev, 0, sizeof(struct net_device));
6314 /* make sure we BUG if trying to hit standard
6315 * register/unregister code path
6317 dev->reg_state = NETREG_DUMMY;
6319 /* NAPI wants this */
6320 INIT_LIST_HEAD(&dev->napi_list);
6322 /* a dummy interface is started by default */
6323 set_bit(__LINK_STATE_PRESENT, &dev->state);
6324 set_bit(__LINK_STATE_START, &dev->state);
6326 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6327 * because users of this 'device' dont need to change
6333 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6337 * register_netdev - register a network device
6338 * @dev: device to register
6340 * Take a completed network device structure and add it to the kernel
6341 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6342 * chain. 0 is returned on success. A negative errno code is returned
6343 * on a failure to set up the device, or if the name is a duplicate.
6345 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6346 * and expands the device name if you passed a format string to
6349 int register_netdev(struct net_device *dev)
6354 err = register_netdevice(dev);
6358 EXPORT_SYMBOL(register_netdev);
6360 int netdev_refcnt_read(const struct net_device *dev)
6364 for_each_possible_cpu(i)
6365 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6368 EXPORT_SYMBOL(netdev_refcnt_read);
6371 * netdev_wait_allrefs - wait until all references are gone.
6372 * @dev: target net_device
6374 * This is called when unregistering network devices.
6376 * Any protocol or device that holds a reference should register
6377 * for netdevice notification, and cleanup and put back the
6378 * reference if they receive an UNREGISTER event.
6379 * We can get stuck here if buggy protocols don't correctly
6382 static void netdev_wait_allrefs(struct net_device *dev)
6384 unsigned long rebroadcast_time, warning_time;
6387 linkwatch_forget_dev(dev);
6389 rebroadcast_time = warning_time = jiffies;
6390 refcnt = netdev_refcnt_read(dev);
6392 while (refcnt != 0) {
6393 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6396 /* Rebroadcast unregister notification */
6397 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6403 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6404 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6406 /* We must not have linkwatch events
6407 * pending on unregister. If this
6408 * happens, we simply run the queue
6409 * unscheduled, resulting in a noop
6412 linkwatch_run_queue();
6417 rebroadcast_time = jiffies;
6422 refcnt = netdev_refcnt_read(dev);
6424 if (time_after(jiffies, warning_time + 10 * HZ)) {
6425 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6427 warning_time = jiffies;
6436 * register_netdevice(x1);
6437 * register_netdevice(x2);
6439 * unregister_netdevice(y1);
6440 * unregister_netdevice(y2);
6446 * We are invoked by rtnl_unlock().
6447 * This allows us to deal with problems:
6448 * 1) We can delete sysfs objects which invoke hotplug
6449 * without deadlocking with linkwatch via keventd.
6450 * 2) Since we run with the RTNL semaphore not held, we can sleep
6451 * safely in order to wait for the netdev refcnt to drop to zero.
6453 * We must not return until all unregister events added during
6454 * the interval the lock was held have been completed.
6456 void netdev_run_todo(void)
6458 struct list_head list;
6460 /* Snapshot list, allow later requests */
6461 list_replace_init(&net_todo_list, &list);
6466 /* Wait for rcu callbacks to finish before next phase */
6467 if (!list_empty(&list))
6470 while (!list_empty(&list)) {
6471 struct net_device *dev
6472 = list_first_entry(&list, struct net_device, todo_list);
6473 list_del(&dev->todo_list);
6476 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6479 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6480 pr_err("network todo '%s' but state %d\n",
6481 dev->name, dev->reg_state);
6486 dev->reg_state = NETREG_UNREGISTERED;
6488 on_each_cpu(flush_backlog, dev, 1);
6490 netdev_wait_allrefs(dev);
6493 BUG_ON(netdev_refcnt_read(dev));
6494 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6495 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6496 WARN_ON(dev->dn_ptr);
6498 if (dev->destructor)
6499 dev->destructor(dev);
6501 /* Report a network device has been unregistered */
6503 dev_net(dev)->dev_unreg_count--;
6505 wake_up(&netdev_unregistering_wq);
6507 /* Free network device */
6508 kobject_put(&dev->dev.kobj);
6512 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6513 * fields in the same order, with only the type differing.
6515 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6516 const struct net_device_stats *netdev_stats)
6518 #if BITS_PER_LONG == 64
6519 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6520 memcpy(stats64, netdev_stats, sizeof(*stats64));
6522 size_t i, n = sizeof(*stats64) / sizeof(u64);
6523 const unsigned long *src = (const unsigned long *)netdev_stats;
6524 u64 *dst = (u64 *)stats64;
6526 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6527 sizeof(*stats64) / sizeof(u64));
6528 for (i = 0; i < n; i++)
6532 EXPORT_SYMBOL(netdev_stats_to_stats64);
6535 * dev_get_stats - get network device statistics
6536 * @dev: device to get statistics from
6537 * @storage: place to store stats
6539 * Get network statistics from device. Return @storage.
6540 * The device driver may provide its own method by setting
6541 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6542 * otherwise the internal statistics structure is used.
6544 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6545 struct rtnl_link_stats64 *storage)
6547 const struct net_device_ops *ops = dev->netdev_ops;
6549 if (ops->ndo_get_stats64) {
6550 memset(storage, 0, sizeof(*storage));
6551 ops->ndo_get_stats64(dev, storage);
6552 } else if (ops->ndo_get_stats) {
6553 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6555 netdev_stats_to_stats64(storage, &dev->stats);
6557 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6558 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6561 EXPORT_SYMBOL(dev_get_stats);
6563 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6565 struct netdev_queue *queue = dev_ingress_queue(dev);
6567 #ifdef CONFIG_NET_CLS_ACT
6570 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6573 netdev_init_one_queue(dev, queue, NULL);
6574 queue->qdisc = &noop_qdisc;
6575 queue->qdisc_sleeping = &noop_qdisc;
6576 rcu_assign_pointer(dev->ingress_queue, queue);
6581 static const struct ethtool_ops default_ethtool_ops;
6583 void netdev_set_default_ethtool_ops(struct net_device *dev,
6584 const struct ethtool_ops *ops)
6586 if (dev->ethtool_ops == &default_ethtool_ops)
6587 dev->ethtool_ops = ops;
6589 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6591 void netdev_freemem(struct net_device *dev)
6593 char *addr = (char *)dev - dev->padded;
6599 * alloc_netdev_mqs - allocate network device
6600 * @sizeof_priv: size of private data to allocate space for
6601 * @name: device name format string
6602 * @name_assign_type: origin of device name
6603 * @setup: callback to initialize device
6604 * @txqs: the number of TX subqueues to allocate
6605 * @rxqs: the number of RX subqueues to allocate
6607 * Allocates a struct net_device with private data area for driver use
6608 * and performs basic initialization. Also allocates subqueue structs
6609 * for each queue on the device.
6611 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6612 unsigned char name_assign_type,
6613 void (*setup)(struct net_device *),
6614 unsigned int txqs, unsigned int rxqs)
6616 struct net_device *dev;
6618 struct net_device *p;
6620 BUG_ON(strlen(name) >= sizeof(dev->name));
6623 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6629 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6634 alloc_size = sizeof(struct net_device);
6636 /* ensure 32-byte alignment of private area */
6637 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6638 alloc_size += sizeof_priv;
6640 /* ensure 32-byte alignment of whole construct */
6641 alloc_size += NETDEV_ALIGN - 1;
6643 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6645 p = vzalloc(alloc_size);
6649 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6650 dev->padded = (char *)dev - (char *)p;
6652 dev->pcpu_refcnt = alloc_percpu(int);
6653 if (!dev->pcpu_refcnt)
6656 if (dev_addr_init(dev))
6662 dev_net_set(dev, &init_net);
6664 dev->gso_max_size = GSO_MAX_SIZE;
6665 dev->gso_max_segs = GSO_MAX_SEGS;
6666 dev->gso_min_segs = 0;
6668 INIT_LIST_HEAD(&dev->napi_list);
6669 INIT_LIST_HEAD(&dev->unreg_list);
6670 INIT_LIST_HEAD(&dev->close_list);
6671 INIT_LIST_HEAD(&dev->link_watch_list);
6672 INIT_LIST_HEAD(&dev->adj_list.upper);
6673 INIT_LIST_HEAD(&dev->adj_list.lower);
6674 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6675 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6676 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
6679 dev->num_tx_queues = txqs;
6680 dev->real_num_tx_queues = txqs;
6681 if (netif_alloc_netdev_queues(dev))
6685 dev->num_rx_queues = rxqs;
6686 dev->real_num_rx_queues = rxqs;
6687 if (netif_alloc_rx_queues(dev))
6691 strcpy(dev->name, name);
6692 dev->name_assign_type = name_assign_type;
6693 dev->group = INIT_NETDEV_GROUP;
6694 if (!dev->ethtool_ops)
6695 dev->ethtool_ops = &default_ethtool_ops;
6703 free_percpu(dev->pcpu_refcnt);
6705 netdev_freemem(dev);
6708 EXPORT_SYMBOL(alloc_netdev_mqs);
6711 * free_netdev - free network device
6714 * This function does the last stage of destroying an allocated device
6715 * interface. The reference to the device object is released.
6716 * If this is the last reference then it will be freed.
6718 void free_netdev(struct net_device *dev)
6720 struct napi_struct *p, *n;
6722 release_net(dev_net(dev));
6724 netif_free_tx_queues(dev);
6729 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6731 /* Flush device addresses */
6732 dev_addr_flush(dev);
6734 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6737 free_percpu(dev->pcpu_refcnt);
6738 dev->pcpu_refcnt = NULL;
6740 /* Compatibility with error handling in drivers */
6741 if (dev->reg_state == NETREG_UNINITIALIZED) {
6742 netdev_freemem(dev);
6746 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6747 dev->reg_state = NETREG_RELEASED;
6749 /* will free via device release */
6750 put_device(&dev->dev);
6752 EXPORT_SYMBOL(free_netdev);
6755 * synchronize_net - Synchronize with packet receive processing
6757 * Wait for packets currently being received to be done.
6758 * Does not block later packets from starting.
6760 void synchronize_net(void)
6763 if (rtnl_is_locked())
6764 synchronize_rcu_expedited();
6768 EXPORT_SYMBOL(synchronize_net);
6771 * unregister_netdevice_queue - remove device from the kernel
6775 * This function shuts down a device interface and removes it
6776 * from the kernel tables.
6777 * If head not NULL, device is queued to be unregistered later.
6779 * Callers must hold the rtnl semaphore. You may want
6780 * unregister_netdev() instead of this.
6783 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6788 list_move_tail(&dev->unreg_list, head);
6790 rollback_registered(dev);
6791 /* Finish processing unregister after unlock */
6795 EXPORT_SYMBOL(unregister_netdevice_queue);
6798 * unregister_netdevice_many - unregister many devices
6799 * @head: list of devices
6801 * Note: As most callers use a stack allocated list_head,
6802 * we force a list_del() to make sure stack wont be corrupted later.
6804 void unregister_netdevice_many(struct list_head *head)
6806 struct net_device *dev;
6808 if (!list_empty(head)) {
6809 rollback_registered_many(head);
6810 list_for_each_entry(dev, head, unreg_list)
6815 EXPORT_SYMBOL(unregister_netdevice_many);
6818 * unregister_netdev - remove device from the kernel
6821 * This function shuts down a device interface and removes it
6822 * from the kernel tables.
6824 * This is just a wrapper for unregister_netdevice that takes
6825 * the rtnl semaphore. In general you want to use this and not
6826 * unregister_netdevice.
6828 void unregister_netdev(struct net_device *dev)
6831 unregister_netdevice(dev);
6834 EXPORT_SYMBOL(unregister_netdev);
6837 * dev_change_net_namespace - move device to different nethost namespace
6839 * @net: network namespace
6840 * @pat: If not NULL name pattern to try if the current device name
6841 * is already taken in the destination network namespace.
6843 * This function shuts down a device interface and moves it
6844 * to a new network namespace. On success 0 is returned, on
6845 * a failure a netagive errno code is returned.
6847 * Callers must hold the rtnl semaphore.
6850 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6856 /* Don't allow namespace local devices to be moved. */
6858 if (dev->features & NETIF_F_NETNS_LOCAL)
6861 /* Ensure the device has been registrered */
6862 if (dev->reg_state != NETREG_REGISTERED)
6865 /* Get out if there is nothing todo */
6867 if (net_eq(dev_net(dev), net))
6870 /* Pick the destination device name, and ensure
6871 * we can use it in the destination network namespace.
6874 if (__dev_get_by_name(net, dev->name)) {
6875 /* We get here if we can't use the current device name */
6878 if (dev_get_valid_name(net, dev, pat) < 0)
6883 * And now a mini version of register_netdevice unregister_netdevice.
6886 /* If device is running close it first. */
6889 /* And unlink it from device chain */
6891 unlist_netdevice(dev);
6895 /* Shutdown queueing discipline. */
6898 /* Notify protocols, that we are about to destroy
6899 this device. They should clean all the things.
6901 Note that dev->reg_state stays at NETREG_REGISTERED.
6902 This is wanted because this way 8021q and macvlan know
6903 the device is just moving and can keep their slaves up.
6905 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6907 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6908 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6911 * Flush the unicast and multicast chains
6916 /* Send a netdev-removed uevent to the old namespace */
6917 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6918 netdev_adjacent_del_links(dev);
6920 /* Actually switch the network namespace */
6921 dev_net_set(dev, net);
6923 /* If there is an ifindex conflict assign a new one */
6924 if (__dev_get_by_index(net, dev->ifindex)) {
6925 int iflink = (dev->iflink == dev->ifindex);
6926 dev->ifindex = dev_new_index(net);
6928 dev->iflink = dev->ifindex;
6931 /* Send a netdev-add uevent to the new namespace */
6932 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6933 netdev_adjacent_add_links(dev);
6935 /* Fixup kobjects */
6936 err = device_rename(&dev->dev, dev->name);
6939 /* Add the device back in the hashes */
6940 list_netdevice(dev);
6942 /* Notify protocols, that a new device appeared. */
6943 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6946 * Prevent userspace races by waiting until the network
6947 * device is fully setup before sending notifications.
6949 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6956 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6958 static int dev_cpu_callback(struct notifier_block *nfb,
6959 unsigned long action,
6962 struct sk_buff **list_skb;
6963 struct sk_buff *skb;
6964 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6965 struct softnet_data *sd, *oldsd;
6967 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6970 local_irq_disable();
6971 cpu = smp_processor_id();
6972 sd = &per_cpu(softnet_data, cpu);
6973 oldsd = &per_cpu(softnet_data, oldcpu);
6975 /* Find end of our completion_queue. */
6976 list_skb = &sd->completion_queue;
6978 list_skb = &(*list_skb)->next;
6979 /* Append completion queue from offline CPU. */
6980 *list_skb = oldsd->completion_queue;
6981 oldsd->completion_queue = NULL;
6983 /* Append output queue from offline CPU. */
6984 if (oldsd->output_queue) {
6985 *sd->output_queue_tailp = oldsd->output_queue;
6986 sd->output_queue_tailp = oldsd->output_queue_tailp;
6987 oldsd->output_queue = NULL;
6988 oldsd->output_queue_tailp = &oldsd->output_queue;
6990 /* Append NAPI poll list from offline CPU. */
6991 if (!list_empty(&oldsd->poll_list)) {
6992 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6993 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6996 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6999 /* Process offline CPU's input_pkt_queue */
7000 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
7001 netif_rx_internal(skb);
7002 input_queue_head_incr(oldsd);
7004 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
7005 netif_rx_internal(skb);
7006 input_queue_head_incr(oldsd);
7014 * netdev_increment_features - increment feature set by one
7015 * @all: current feature set
7016 * @one: new feature set
7017 * @mask: mask feature set
7019 * Computes a new feature set after adding a device with feature set
7020 * @one to the master device with current feature set @all. Will not
7021 * enable anything that is off in @mask. Returns the new feature set.
7023 netdev_features_t netdev_increment_features(netdev_features_t all,
7024 netdev_features_t one, netdev_features_t mask)
7026 if (mask & NETIF_F_GEN_CSUM)
7027 mask |= NETIF_F_ALL_CSUM;
7028 mask |= NETIF_F_VLAN_CHALLENGED;
7030 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
7031 all &= one | ~NETIF_F_ALL_FOR_ALL;
7033 /* If one device supports hw checksumming, set for all. */
7034 if (all & NETIF_F_GEN_CSUM)
7035 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
7039 EXPORT_SYMBOL(netdev_increment_features);
7041 static struct hlist_head * __net_init netdev_create_hash(void)
7044 struct hlist_head *hash;
7046 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
7048 for (i = 0; i < NETDEV_HASHENTRIES; i++)
7049 INIT_HLIST_HEAD(&hash[i]);
7054 /* Initialize per network namespace state */
7055 static int __net_init netdev_init(struct net *net)
7057 if (net != &init_net)
7058 INIT_LIST_HEAD(&net->dev_base_head);
7060 net->dev_name_head = netdev_create_hash();
7061 if (net->dev_name_head == NULL)
7064 net->dev_index_head = netdev_create_hash();
7065 if (net->dev_index_head == NULL)
7071 kfree(net->dev_name_head);
7077 * netdev_drivername - network driver for the device
7078 * @dev: network device
7080 * Determine network driver for device.
7082 const char *netdev_drivername(const struct net_device *dev)
7084 const struct device_driver *driver;
7085 const struct device *parent;
7086 const char *empty = "";
7088 parent = dev->dev.parent;
7092 driver = parent->driver;
7093 if (driver && driver->name)
7094 return driver->name;
7098 static void __netdev_printk(const char *level, const struct net_device *dev,
7099 struct va_format *vaf)
7101 if (dev && dev->dev.parent) {
7102 dev_printk_emit(level[1] - '0',
7105 dev_driver_string(dev->dev.parent),
7106 dev_name(dev->dev.parent),
7107 netdev_name(dev), netdev_reg_state(dev),
7110 printk("%s%s%s: %pV",
7111 level, netdev_name(dev), netdev_reg_state(dev), vaf);
7113 printk("%s(NULL net_device): %pV", level, vaf);
7117 void netdev_printk(const char *level, const struct net_device *dev,
7118 const char *format, ...)
7120 struct va_format vaf;
7123 va_start(args, format);
7128 __netdev_printk(level, dev, &vaf);
7132 EXPORT_SYMBOL(netdev_printk);
7134 #define define_netdev_printk_level(func, level) \
7135 void func(const struct net_device *dev, const char *fmt, ...) \
7137 struct va_format vaf; \
7140 va_start(args, fmt); \
7145 __netdev_printk(level, dev, &vaf); \
7149 EXPORT_SYMBOL(func);
7151 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
7152 define_netdev_printk_level(netdev_alert, KERN_ALERT);
7153 define_netdev_printk_level(netdev_crit, KERN_CRIT);
7154 define_netdev_printk_level(netdev_err, KERN_ERR);
7155 define_netdev_printk_level(netdev_warn, KERN_WARNING);
7156 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
7157 define_netdev_printk_level(netdev_info, KERN_INFO);
7159 static void __net_exit netdev_exit(struct net *net)
7161 kfree(net->dev_name_head);
7162 kfree(net->dev_index_head);
7165 static struct pernet_operations __net_initdata netdev_net_ops = {
7166 .init = netdev_init,
7167 .exit = netdev_exit,
7170 static void __net_exit default_device_exit(struct net *net)
7172 struct net_device *dev, *aux;
7174 * Push all migratable network devices back to the
7175 * initial network namespace
7178 for_each_netdev_safe(net, dev, aux) {
7180 char fb_name[IFNAMSIZ];
7182 /* Ignore unmoveable devices (i.e. loopback) */
7183 if (dev->features & NETIF_F_NETNS_LOCAL)
7186 /* Leave virtual devices for the generic cleanup */
7187 if (dev->rtnl_link_ops)
7190 /* Push remaining network devices to init_net */
7191 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
7192 err = dev_change_net_namespace(dev, &init_net, fb_name);
7194 pr_emerg("%s: failed to move %s to init_net: %d\n",
7195 __func__, dev->name, err);
7202 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
7204 /* Return with the rtnl_lock held when there are no network
7205 * devices unregistering in any network namespace in net_list.
7212 prepare_to_wait(&netdev_unregistering_wq, &wait,
7213 TASK_UNINTERRUPTIBLE);
7214 unregistering = false;
7216 list_for_each_entry(net, net_list, exit_list) {
7217 if (net->dev_unreg_count > 0) {
7218 unregistering = true;
7227 finish_wait(&netdev_unregistering_wq, &wait);
7230 static void __net_exit default_device_exit_batch(struct list_head *net_list)
7232 /* At exit all network devices most be removed from a network
7233 * namespace. Do this in the reverse order of registration.
7234 * Do this across as many network namespaces as possible to
7235 * improve batching efficiency.
7237 struct net_device *dev;
7239 LIST_HEAD(dev_kill_list);
7241 /* To prevent network device cleanup code from dereferencing
7242 * loopback devices or network devices that have been freed
7243 * wait here for all pending unregistrations to complete,
7244 * before unregistring the loopback device and allowing the
7245 * network namespace be freed.
7247 * The netdev todo list containing all network devices
7248 * unregistrations that happen in default_device_exit_batch
7249 * will run in the rtnl_unlock() at the end of
7250 * default_device_exit_batch.
7252 rtnl_lock_unregistering(net_list);
7253 list_for_each_entry(net, net_list, exit_list) {
7254 for_each_netdev_reverse(net, dev) {
7255 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
7256 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
7258 unregister_netdevice_queue(dev, &dev_kill_list);
7261 unregister_netdevice_many(&dev_kill_list);
7265 static struct pernet_operations __net_initdata default_device_ops = {
7266 .exit = default_device_exit,
7267 .exit_batch = default_device_exit_batch,
7271 * Initialize the DEV module. At boot time this walks the device list and
7272 * unhooks any devices that fail to initialise (normally hardware not
7273 * present) and leaves us with a valid list of present and active devices.
7278 * This is called single threaded during boot, so no need
7279 * to take the rtnl semaphore.
7281 static int __init net_dev_init(void)
7283 int i, rc = -ENOMEM;
7285 BUG_ON(!dev_boot_phase);
7287 if (dev_proc_init())
7290 if (netdev_kobject_init())
7293 INIT_LIST_HEAD(&ptype_all);
7294 for (i = 0; i < PTYPE_HASH_SIZE; i++)
7295 INIT_LIST_HEAD(&ptype_base[i]);
7297 INIT_LIST_HEAD(&offload_base);
7299 if (register_pernet_subsys(&netdev_net_ops))
7303 * Initialise the packet receive queues.
7306 for_each_possible_cpu(i) {
7307 struct softnet_data *sd = &per_cpu(softnet_data, i);
7309 skb_queue_head_init(&sd->input_pkt_queue);
7310 skb_queue_head_init(&sd->process_queue);
7311 INIT_LIST_HEAD(&sd->poll_list);
7312 sd->output_queue_tailp = &sd->output_queue;
7314 sd->csd.func = rps_trigger_softirq;
7319 sd->backlog.poll = process_backlog;
7320 sd->backlog.weight = weight_p;
7325 /* The loopback device is special if any other network devices
7326 * is present in a network namespace the loopback device must
7327 * be present. Since we now dynamically allocate and free the
7328 * loopback device ensure this invariant is maintained by
7329 * keeping the loopback device as the first device on the
7330 * list of network devices. Ensuring the loopback devices
7331 * is the first device that appears and the last network device
7334 if (register_pernet_device(&loopback_net_ops))
7337 if (register_pernet_device(&default_device_ops))
7340 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7341 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7343 hotcpu_notifier(dev_cpu_callback, 0);
7350 subsys_initcall(net_dev_init);