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>
136 #include "net-sysfs.h"
138 /* Instead of increasing this, you should create a hash table. */
139 #define MAX_GRO_SKBS 8
141 /* This should be increased if a protocol with a bigger head is added. */
142 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 static DEFINE_SPINLOCK(ptype_lock);
145 static DEFINE_SPINLOCK(offload_lock);
146 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
147 struct list_head ptype_all __read_mostly; /* Taps */
148 static struct list_head offload_base __read_mostly;
150 static int netif_rx_internal(struct sk_buff *skb);
153 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
156 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
158 * Writers must hold the rtnl semaphore while they loop through the
159 * dev_base_head list, and hold dev_base_lock for writing when they do the
160 * actual updates. This allows pure readers to access the list even
161 * while a writer is preparing to update it.
163 * To put it another way, dev_base_lock is held for writing only to
164 * protect against pure readers; the rtnl semaphore provides the
165 * protection against other writers.
167 * See, for example usages, register_netdevice() and
168 * unregister_netdevice(), which must be called with the rtnl
171 DEFINE_RWLOCK(dev_base_lock);
172 EXPORT_SYMBOL(dev_base_lock);
174 /* protects napi_hash addition/deletion and napi_gen_id */
175 static DEFINE_SPINLOCK(napi_hash_lock);
177 static unsigned int napi_gen_id;
178 static DEFINE_HASHTABLE(napi_hash, 8);
180 static seqcount_t devnet_rename_seq;
182 static inline void dev_base_seq_inc(struct net *net)
184 while (++net->dev_base_seq == 0);
187 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
189 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
191 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
194 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
196 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
199 static inline void rps_lock(struct softnet_data *sd)
202 spin_lock(&sd->input_pkt_queue.lock);
206 static inline void rps_unlock(struct softnet_data *sd)
209 spin_unlock(&sd->input_pkt_queue.lock);
213 /* Device list insertion */
214 static void list_netdevice(struct net_device *dev)
216 struct net *net = dev_net(dev);
220 write_lock_bh(&dev_base_lock);
221 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
222 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
223 hlist_add_head_rcu(&dev->index_hlist,
224 dev_index_hash(net, dev->ifindex));
225 write_unlock_bh(&dev_base_lock);
227 dev_base_seq_inc(net);
230 /* Device list removal
231 * caller must respect a RCU grace period before freeing/reusing dev
233 static void unlist_netdevice(struct net_device *dev)
237 /* Unlink dev from the device chain */
238 write_lock_bh(&dev_base_lock);
239 list_del_rcu(&dev->dev_list);
240 hlist_del_rcu(&dev->name_hlist);
241 hlist_del_rcu(&dev->index_hlist);
242 write_unlock_bh(&dev_base_lock);
244 dev_base_seq_inc(dev_net(dev));
251 static RAW_NOTIFIER_HEAD(netdev_chain);
254 * Device drivers call our routines to queue packets here. We empty the
255 * queue in the local softnet handler.
258 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
259 EXPORT_PER_CPU_SYMBOL(softnet_data);
261 #ifdef CONFIG_LOCKDEP
263 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
264 * according to dev->type
266 static const unsigned short netdev_lock_type[] =
267 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
268 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
269 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
270 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
271 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
272 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
273 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
274 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
275 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
276 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
277 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
278 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
279 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
280 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
281 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
283 static const char *const netdev_lock_name[] =
284 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
285 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
286 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
287 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
288 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
289 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
290 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
291 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
292 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
293 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
294 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
295 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
296 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
297 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
298 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
300 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
301 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
303 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
307 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
308 if (netdev_lock_type[i] == dev_type)
310 /* the last key is used by default */
311 return ARRAY_SIZE(netdev_lock_type) - 1;
314 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
315 unsigned short dev_type)
319 i = netdev_lock_pos(dev_type);
320 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
321 netdev_lock_name[i]);
324 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
328 i = netdev_lock_pos(dev->type);
329 lockdep_set_class_and_name(&dev->addr_list_lock,
330 &netdev_addr_lock_key[i],
331 netdev_lock_name[i]);
334 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
335 unsigned short dev_type)
338 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
343 /*******************************************************************************
345 Protocol management and registration routines
347 *******************************************************************************/
350 * Add a protocol ID to the list. Now that the input handler is
351 * smarter we can dispense with all the messy stuff that used to be
354 * BEWARE!!! Protocol handlers, mangling input packets,
355 * MUST BE last in hash buckets and checking protocol handlers
356 * MUST start from promiscuous ptype_all chain in net_bh.
357 * It is true now, do not change it.
358 * Explanation follows: if protocol handler, mangling packet, will
359 * be the first on list, it is not able to sense, that packet
360 * is cloned and should be copied-on-write, so that it will
361 * change it and subsequent readers will get broken packet.
365 static inline struct list_head *ptype_head(const struct packet_type *pt)
367 if (pt->type == htons(ETH_P_ALL))
370 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
374 * dev_add_pack - add packet handler
375 * @pt: packet type declaration
377 * Add a protocol handler to the networking stack. The passed &packet_type
378 * is linked into kernel lists and may not be freed until it has been
379 * removed from the kernel lists.
381 * This call does not sleep therefore it can not
382 * guarantee all CPU's that are in middle of receiving packets
383 * will see the new packet type (until the next received packet).
386 void dev_add_pack(struct packet_type *pt)
388 struct list_head *head = ptype_head(pt);
390 spin_lock(&ptype_lock);
391 list_add_rcu(&pt->list, head);
392 spin_unlock(&ptype_lock);
394 EXPORT_SYMBOL(dev_add_pack);
397 * __dev_remove_pack - remove packet handler
398 * @pt: packet type declaration
400 * Remove a protocol handler that was previously added to the kernel
401 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
402 * from the kernel lists and can be freed or reused once this function
405 * The packet type might still be in use by receivers
406 * and must not be freed until after all the CPU's have gone
407 * through a quiescent state.
409 void __dev_remove_pack(struct packet_type *pt)
411 struct list_head *head = ptype_head(pt);
412 struct packet_type *pt1;
414 spin_lock(&ptype_lock);
416 list_for_each_entry(pt1, head, list) {
418 list_del_rcu(&pt->list);
423 pr_warn("dev_remove_pack: %p not found\n", pt);
425 spin_unlock(&ptype_lock);
427 EXPORT_SYMBOL(__dev_remove_pack);
430 * dev_remove_pack - remove packet handler
431 * @pt: packet type declaration
433 * Remove a protocol handler that was previously added to the kernel
434 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
435 * from the kernel lists and can be freed or reused once this function
438 * This call sleeps to guarantee that no CPU is looking at the packet
441 void dev_remove_pack(struct packet_type *pt)
443 __dev_remove_pack(pt);
447 EXPORT_SYMBOL(dev_remove_pack);
451 * dev_add_offload - register offload handlers
452 * @po: protocol offload declaration
454 * Add protocol offload handlers to the networking stack. The passed
455 * &proto_offload is linked into kernel lists and may not be freed until
456 * it has been removed from the kernel lists.
458 * This call does not sleep therefore it can not
459 * guarantee all CPU's that are in middle of receiving packets
460 * will see the new offload handlers (until the next received packet).
462 void dev_add_offload(struct packet_offload *po)
464 struct list_head *head = &offload_base;
466 spin_lock(&offload_lock);
467 list_add_rcu(&po->list, head);
468 spin_unlock(&offload_lock);
470 EXPORT_SYMBOL(dev_add_offload);
473 * __dev_remove_offload - remove offload handler
474 * @po: packet offload declaration
476 * Remove a protocol offload handler that was previously added to the
477 * kernel offload handlers by dev_add_offload(). The passed &offload_type
478 * is removed from the kernel lists and can be freed or reused once this
481 * The packet type might still be in use by receivers
482 * and must not be freed until after all the CPU's have gone
483 * through a quiescent state.
485 static void __dev_remove_offload(struct packet_offload *po)
487 struct list_head *head = &offload_base;
488 struct packet_offload *po1;
490 spin_lock(&offload_lock);
492 list_for_each_entry(po1, head, list) {
494 list_del_rcu(&po->list);
499 pr_warn("dev_remove_offload: %p not found\n", po);
501 spin_unlock(&offload_lock);
505 * dev_remove_offload - remove packet offload handler
506 * @po: packet offload declaration
508 * Remove a packet offload handler that was previously added to the kernel
509 * offload handlers by dev_add_offload(). The passed &offload_type is
510 * removed from the kernel lists and can be freed or reused once this
513 * This call sleeps to guarantee that no CPU is looking at the packet
516 void dev_remove_offload(struct packet_offload *po)
518 __dev_remove_offload(po);
522 EXPORT_SYMBOL(dev_remove_offload);
524 /******************************************************************************
526 Device Boot-time Settings Routines
528 *******************************************************************************/
530 /* Boot time configuration table */
531 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
534 * netdev_boot_setup_add - add new setup entry
535 * @name: name of the device
536 * @map: configured settings for the device
538 * Adds new setup entry to the dev_boot_setup list. The function
539 * returns 0 on error and 1 on success. This is a generic routine to
542 static int netdev_boot_setup_add(char *name, struct ifmap *map)
544 struct netdev_boot_setup *s;
548 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
549 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
550 memset(s[i].name, 0, sizeof(s[i].name));
551 strlcpy(s[i].name, name, IFNAMSIZ);
552 memcpy(&s[i].map, map, sizeof(s[i].map));
557 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
561 * netdev_boot_setup_check - check boot time settings
562 * @dev: the netdevice
564 * Check boot time settings for the device.
565 * The found settings are set for the device to be used
566 * later in the device probing.
567 * Returns 0 if no settings found, 1 if they are.
569 int netdev_boot_setup_check(struct net_device *dev)
571 struct netdev_boot_setup *s = dev_boot_setup;
574 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
575 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
576 !strcmp(dev->name, s[i].name)) {
577 dev->irq = s[i].map.irq;
578 dev->base_addr = s[i].map.base_addr;
579 dev->mem_start = s[i].map.mem_start;
580 dev->mem_end = s[i].map.mem_end;
586 EXPORT_SYMBOL(netdev_boot_setup_check);
590 * netdev_boot_base - get address from boot time settings
591 * @prefix: prefix for network device
592 * @unit: id for network device
594 * Check boot time settings for the base address of device.
595 * The found settings are set for the device to be used
596 * later in the device probing.
597 * Returns 0 if no settings found.
599 unsigned long netdev_boot_base(const char *prefix, int unit)
601 const struct netdev_boot_setup *s = dev_boot_setup;
605 sprintf(name, "%s%d", prefix, unit);
608 * If device already registered then return base of 1
609 * to indicate not to probe for this interface
611 if (__dev_get_by_name(&init_net, name))
614 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
615 if (!strcmp(name, s[i].name))
616 return s[i].map.base_addr;
621 * Saves at boot time configured settings for any netdevice.
623 int __init netdev_boot_setup(char *str)
628 str = get_options(str, ARRAY_SIZE(ints), ints);
633 memset(&map, 0, sizeof(map));
637 map.base_addr = ints[2];
639 map.mem_start = ints[3];
641 map.mem_end = ints[4];
643 /* Add new entry to the list */
644 return netdev_boot_setup_add(str, &map);
647 __setup("netdev=", netdev_boot_setup);
649 /*******************************************************************************
651 Device Interface Subroutines
653 *******************************************************************************/
656 * __dev_get_by_name - find a device by its name
657 * @net: the applicable net namespace
658 * @name: name to find
660 * Find an interface by name. Must be called under RTNL semaphore
661 * or @dev_base_lock. If the name is found a pointer to the device
662 * is returned. If the name is not found then %NULL is returned. The
663 * reference counters are not incremented so the caller must be
664 * careful with locks.
667 struct net_device *__dev_get_by_name(struct net *net, const char *name)
669 struct net_device *dev;
670 struct hlist_head *head = dev_name_hash(net, name);
672 hlist_for_each_entry(dev, head, name_hlist)
673 if (!strncmp(dev->name, name, IFNAMSIZ))
678 EXPORT_SYMBOL(__dev_get_by_name);
681 * dev_get_by_name_rcu - find a device by its name
682 * @net: the applicable net namespace
683 * @name: name to find
685 * Find an interface by name.
686 * If the name is found a pointer to the device is returned.
687 * If the name is not found then %NULL is returned.
688 * The reference counters are not incremented so the caller must be
689 * careful with locks. The caller must hold RCU lock.
692 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
694 struct net_device *dev;
695 struct hlist_head *head = dev_name_hash(net, name);
697 hlist_for_each_entry_rcu(dev, head, name_hlist)
698 if (!strncmp(dev->name, name, IFNAMSIZ))
703 EXPORT_SYMBOL(dev_get_by_name_rcu);
706 * dev_get_by_name - find a device by its name
707 * @net: the applicable net namespace
708 * @name: name to find
710 * Find an interface by name. This can be called from any
711 * context and does its own locking. The returned handle has
712 * the usage count incremented and the caller must use dev_put() to
713 * release it when it is no longer needed. %NULL is returned if no
714 * matching device is found.
717 struct net_device *dev_get_by_name(struct net *net, const char *name)
719 struct net_device *dev;
722 dev = dev_get_by_name_rcu(net, name);
728 EXPORT_SYMBOL(dev_get_by_name);
731 * __dev_get_by_index - find a device by its ifindex
732 * @net: the applicable net namespace
733 * @ifindex: index of device
735 * Search for an interface by index. Returns %NULL if the device
736 * is not found or a pointer to the device. The device has not
737 * had its reference counter increased so the caller must be careful
738 * about locking. The caller must hold either the RTNL semaphore
742 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
744 struct net_device *dev;
745 struct hlist_head *head = dev_index_hash(net, ifindex);
747 hlist_for_each_entry(dev, head, index_hlist)
748 if (dev->ifindex == ifindex)
753 EXPORT_SYMBOL(__dev_get_by_index);
756 * dev_get_by_index_rcu - find a device by its ifindex
757 * @net: the applicable net namespace
758 * @ifindex: index of device
760 * Search for an interface by index. Returns %NULL if the device
761 * is not found or a pointer to the device. The device has not
762 * had its reference counter increased so the caller must be careful
763 * about locking. The caller must hold RCU lock.
766 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
768 struct net_device *dev;
769 struct hlist_head *head = dev_index_hash(net, ifindex);
771 hlist_for_each_entry_rcu(dev, head, index_hlist)
772 if (dev->ifindex == ifindex)
777 EXPORT_SYMBOL(dev_get_by_index_rcu);
781 * dev_get_by_index - find a device by its ifindex
782 * @net: the applicable net namespace
783 * @ifindex: index of device
785 * Search for an interface by index. Returns NULL if the device
786 * is not found or a pointer to the device. The device returned has
787 * had a reference added and the pointer is safe until the user calls
788 * dev_put to indicate they have finished with it.
791 struct net_device *dev_get_by_index(struct net *net, int ifindex)
793 struct net_device *dev;
796 dev = dev_get_by_index_rcu(net, ifindex);
802 EXPORT_SYMBOL(dev_get_by_index);
805 * netdev_get_name - get a netdevice name, knowing its ifindex.
806 * @net: network namespace
807 * @name: a pointer to the buffer where the name will be stored.
808 * @ifindex: the ifindex of the interface to get the name from.
810 * The use of raw_seqcount_begin() and cond_resched() before
811 * retrying is required as we want to give the writers a chance
812 * to complete when CONFIG_PREEMPT is not set.
814 int netdev_get_name(struct net *net, char *name, int ifindex)
816 struct net_device *dev;
820 seq = raw_seqcount_begin(&devnet_rename_seq);
822 dev = dev_get_by_index_rcu(net, ifindex);
828 strcpy(name, dev->name);
830 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
839 * dev_getbyhwaddr_rcu - find a device by its hardware address
840 * @net: the applicable net namespace
841 * @type: media type of device
842 * @ha: hardware address
844 * Search for an interface by MAC address. Returns NULL if the device
845 * is not found or a pointer to the device.
846 * The caller must hold RCU or RTNL.
847 * The returned device has not had its ref count increased
848 * and the caller must therefore be careful about locking
852 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
855 struct net_device *dev;
857 for_each_netdev_rcu(net, dev)
858 if (dev->type == type &&
859 !memcmp(dev->dev_addr, ha, dev->addr_len))
864 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
866 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
868 struct net_device *dev;
871 for_each_netdev(net, dev)
872 if (dev->type == type)
877 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
879 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
881 struct net_device *dev, *ret = NULL;
884 for_each_netdev_rcu(net, dev)
885 if (dev->type == type) {
893 EXPORT_SYMBOL(dev_getfirstbyhwtype);
896 * dev_get_by_flags_rcu - find any device with given flags
897 * @net: the applicable net namespace
898 * @if_flags: IFF_* values
899 * @mask: bitmask of bits in if_flags to check
901 * Search for any interface with the given flags. Returns NULL if a device
902 * is not found or a pointer to the device. Must be called inside
903 * rcu_read_lock(), and result refcount is unchanged.
906 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
909 struct net_device *dev, *ret;
912 for_each_netdev_rcu(net, dev) {
913 if (((dev->flags ^ if_flags) & mask) == 0) {
920 EXPORT_SYMBOL(dev_get_by_flags_rcu);
923 * dev_valid_name - check if name is okay for network device
926 * Network device names need to be valid file names to
927 * to allow sysfs to work. We also disallow any kind of
930 bool dev_valid_name(const char *name)
934 if (strlen(name) >= IFNAMSIZ)
936 if (!strcmp(name, ".") || !strcmp(name, ".."))
940 if (*name == '/' || isspace(*name))
946 EXPORT_SYMBOL(dev_valid_name);
949 * __dev_alloc_name - allocate a name for a device
950 * @net: network namespace to allocate the device name in
951 * @name: name format string
952 * @buf: scratch buffer and result name string
954 * Passed a format string - eg "lt%d" it will try and find a suitable
955 * id. It scans list of devices to build up a free map, then chooses
956 * the first empty slot. The caller must hold the dev_base or rtnl lock
957 * while allocating the name and adding the device in order to avoid
959 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
960 * Returns the number of the unit assigned or a negative errno code.
963 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
967 const int max_netdevices = 8*PAGE_SIZE;
968 unsigned long *inuse;
969 struct net_device *d;
971 p = strnchr(name, IFNAMSIZ-1, '%');
974 * Verify the string as this thing may have come from
975 * the user. There must be either one "%d" and no other "%"
978 if (p[1] != 'd' || strchr(p + 2, '%'))
981 /* Use one page as a bit array of possible slots */
982 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
986 for_each_netdev(net, d) {
987 if (!sscanf(d->name, name, &i))
989 if (i < 0 || i >= max_netdevices)
992 /* avoid cases where sscanf is not exact inverse of printf */
993 snprintf(buf, IFNAMSIZ, name, i);
994 if (!strncmp(buf, d->name, IFNAMSIZ))
998 i = find_first_zero_bit(inuse, max_netdevices);
999 free_page((unsigned long) inuse);
1003 snprintf(buf, IFNAMSIZ, name, i);
1004 if (!__dev_get_by_name(net, buf))
1007 /* It is possible to run out of possible slots
1008 * when the name is long and there isn't enough space left
1009 * for the digits, or if all bits are used.
1015 * dev_alloc_name - allocate a name for a device
1017 * @name: name format string
1019 * Passed a format string - eg "lt%d" it will try and find a suitable
1020 * id. It scans list of devices to build up a free map, then chooses
1021 * the first empty slot. The caller must hold the dev_base or rtnl lock
1022 * while allocating the name and adding the device in order to avoid
1024 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1025 * Returns the number of the unit assigned or a negative errno code.
1028 int dev_alloc_name(struct net_device *dev, const char *name)
1034 BUG_ON(!dev_net(dev));
1036 ret = __dev_alloc_name(net, name, buf);
1038 strlcpy(dev->name, buf, IFNAMSIZ);
1041 EXPORT_SYMBOL(dev_alloc_name);
1043 static int dev_alloc_name_ns(struct net *net,
1044 struct net_device *dev,
1050 ret = __dev_alloc_name(net, name, buf);
1052 strlcpy(dev->name, buf, IFNAMSIZ);
1056 static int dev_get_valid_name(struct net *net,
1057 struct net_device *dev,
1062 if (!dev_valid_name(name))
1065 if (strchr(name, '%'))
1066 return dev_alloc_name_ns(net, dev, name);
1067 else if (__dev_get_by_name(net, name))
1069 else if (dev->name != name)
1070 strlcpy(dev->name, name, IFNAMSIZ);
1076 * dev_change_name - change name of a device
1078 * @newname: name (or format string) must be at least IFNAMSIZ
1080 * Change name of a device, can pass format strings "eth%d".
1083 int dev_change_name(struct net_device *dev, const char *newname)
1085 char oldname[IFNAMSIZ];
1091 BUG_ON(!dev_net(dev));
1094 if (dev->flags & IFF_UP)
1097 write_seqcount_begin(&devnet_rename_seq);
1099 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1100 write_seqcount_end(&devnet_rename_seq);
1104 memcpy(oldname, dev->name, IFNAMSIZ);
1106 err = dev_get_valid_name(net, dev, newname);
1108 write_seqcount_end(&devnet_rename_seq);
1113 ret = device_rename(&dev->dev, dev->name);
1115 memcpy(dev->name, oldname, IFNAMSIZ);
1116 write_seqcount_end(&devnet_rename_seq);
1120 write_seqcount_end(&devnet_rename_seq);
1122 netdev_adjacent_rename_links(dev, oldname);
1124 write_lock_bh(&dev_base_lock);
1125 hlist_del_rcu(&dev->name_hlist);
1126 write_unlock_bh(&dev_base_lock);
1130 write_lock_bh(&dev_base_lock);
1131 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1132 write_unlock_bh(&dev_base_lock);
1134 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1135 ret = notifier_to_errno(ret);
1138 /* err >= 0 after dev_alloc_name() or stores the first errno */
1141 write_seqcount_begin(&devnet_rename_seq);
1142 memcpy(dev->name, oldname, IFNAMSIZ);
1143 memcpy(oldname, newname, IFNAMSIZ);
1146 pr_err("%s: name change rollback failed: %d\n",
1155 * dev_set_alias - change ifalias of a device
1157 * @alias: name up to IFALIASZ
1158 * @len: limit of bytes to copy from info
1160 * Set ifalias for a device,
1162 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1168 if (len >= IFALIASZ)
1172 kfree(dev->ifalias);
1173 dev->ifalias = NULL;
1177 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1180 dev->ifalias = new_ifalias;
1182 strlcpy(dev->ifalias, alias, len+1);
1188 * netdev_features_change - device changes features
1189 * @dev: device to cause notification
1191 * Called to indicate a device has changed features.
1193 void netdev_features_change(struct net_device *dev)
1195 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1197 EXPORT_SYMBOL(netdev_features_change);
1200 * netdev_state_change - device changes state
1201 * @dev: device to cause notification
1203 * Called to indicate a device has changed state. This function calls
1204 * the notifier chains for netdev_chain and sends a NEWLINK message
1205 * to the routing socket.
1207 void netdev_state_change(struct net_device *dev)
1209 if (dev->flags & IFF_UP) {
1210 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1211 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1214 EXPORT_SYMBOL(netdev_state_change);
1217 * netdev_notify_peers - notify network peers about existence of @dev
1218 * @dev: network device
1220 * Generate traffic such that interested network peers are aware of
1221 * @dev, such as by generating a gratuitous ARP. This may be used when
1222 * a device wants to inform the rest of the network about some sort of
1223 * reconfiguration such as a failover event or virtual machine
1226 void netdev_notify_peers(struct net_device *dev)
1229 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1232 EXPORT_SYMBOL(netdev_notify_peers);
1234 static int __dev_open(struct net_device *dev)
1236 const struct net_device_ops *ops = dev->netdev_ops;
1241 if (!netif_device_present(dev))
1244 /* Block netpoll from trying to do any rx path servicing.
1245 * If we don't do this there is a chance ndo_poll_controller
1246 * or ndo_poll may be running while we open the device
1248 netpoll_poll_disable(dev);
1250 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1251 ret = notifier_to_errno(ret);
1255 set_bit(__LINK_STATE_START, &dev->state);
1257 if (ops->ndo_validate_addr)
1258 ret = ops->ndo_validate_addr(dev);
1260 if (!ret && ops->ndo_open)
1261 ret = ops->ndo_open(dev);
1263 netpoll_poll_enable(dev);
1266 clear_bit(__LINK_STATE_START, &dev->state);
1268 dev->flags |= IFF_UP;
1269 net_dmaengine_get();
1270 dev_set_rx_mode(dev);
1272 add_device_randomness(dev->dev_addr, dev->addr_len);
1279 * dev_open - prepare an interface for use.
1280 * @dev: device to open
1282 * Takes a device from down to up state. The device's private open
1283 * function is invoked and then the multicast lists are loaded. Finally
1284 * the device is moved into the up state and a %NETDEV_UP message is
1285 * sent to the netdev notifier chain.
1287 * Calling this function on an active interface is a nop. On a failure
1288 * a negative errno code is returned.
1290 int dev_open(struct net_device *dev)
1294 if (dev->flags & IFF_UP)
1297 ret = __dev_open(dev);
1301 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1302 call_netdevice_notifiers(NETDEV_UP, dev);
1306 EXPORT_SYMBOL(dev_open);
1308 static int __dev_close_many(struct list_head *head)
1310 struct net_device *dev;
1315 list_for_each_entry(dev, head, close_list) {
1316 /* Temporarily disable netpoll until the interface is down */
1317 netpoll_poll_disable(dev);
1319 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1321 clear_bit(__LINK_STATE_START, &dev->state);
1323 /* Synchronize to scheduled poll. We cannot touch poll list, it
1324 * can be even on different cpu. So just clear netif_running().
1326 * dev->stop() will invoke napi_disable() on all of it's
1327 * napi_struct instances on this device.
1329 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1332 dev_deactivate_many(head);
1334 list_for_each_entry(dev, head, close_list) {
1335 const struct net_device_ops *ops = dev->netdev_ops;
1338 * Call the device specific close. This cannot fail.
1339 * Only if device is UP
1341 * We allow it to be called even after a DETACH hot-plug
1347 dev->flags &= ~IFF_UP;
1348 net_dmaengine_put();
1349 netpoll_poll_enable(dev);
1355 static int __dev_close(struct net_device *dev)
1360 list_add(&dev->close_list, &single);
1361 retval = __dev_close_many(&single);
1367 static int dev_close_many(struct list_head *head)
1369 struct net_device *dev, *tmp;
1371 /* Remove the devices that don't need to be closed */
1372 list_for_each_entry_safe(dev, tmp, head, close_list)
1373 if (!(dev->flags & IFF_UP))
1374 list_del_init(&dev->close_list);
1376 __dev_close_many(head);
1378 list_for_each_entry_safe(dev, tmp, head, close_list) {
1379 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1380 call_netdevice_notifiers(NETDEV_DOWN, dev);
1381 list_del_init(&dev->close_list);
1388 * dev_close - shutdown an interface.
1389 * @dev: device to shutdown
1391 * This function moves an active device into down state. A
1392 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1393 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1396 int dev_close(struct net_device *dev)
1398 if (dev->flags & IFF_UP) {
1401 list_add(&dev->close_list, &single);
1402 dev_close_many(&single);
1407 EXPORT_SYMBOL(dev_close);
1411 * dev_disable_lro - disable Large Receive Offload on a device
1414 * Disable Large Receive Offload (LRO) on a net device. Must be
1415 * called under RTNL. This is needed if received packets may be
1416 * forwarded to another interface.
1418 void dev_disable_lro(struct net_device *dev)
1421 * If we're trying to disable lro on a vlan device
1422 * use the underlying physical device instead
1424 if (is_vlan_dev(dev))
1425 dev = vlan_dev_real_dev(dev);
1427 /* the same for macvlan devices */
1428 if (netif_is_macvlan(dev))
1429 dev = macvlan_dev_real_dev(dev);
1431 dev->wanted_features &= ~NETIF_F_LRO;
1432 netdev_update_features(dev);
1434 if (unlikely(dev->features & NETIF_F_LRO))
1435 netdev_WARN(dev, "failed to disable LRO!\n");
1437 EXPORT_SYMBOL(dev_disable_lro);
1439 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1440 struct net_device *dev)
1442 struct netdev_notifier_info info;
1444 netdev_notifier_info_init(&info, dev);
1445 return nb->notifier_call(nb, val, &info);
1448 static int dev_boot_phase = 1;
1451 * register_netdevice_notifier - register a network notifier block
1454 * Register a notifier to be called when network device events occur.
1455 * The notifier passed is linked into the kernel structures and must
1456 * not be reused until it has been unregistered. A negative errno code
1457 * is returned on a failure.
1459 * When registered all registration and up events are replayed
1460 * to the new notifier to allow device to have a race free
1461 * view of the network device list.
1464 int register_netdevice_notifier(struct notifier_block *nb)
1466 struct net_device *dev;
1467 struct net_device *last;
1472 err = raw_notifier_chain_register(&netdev_chain, nb);
1478 for_each_netdev(net, dev) {
1479 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1480 err = notifier_to_errno(err);
1484 if (!(dev->flags & IFF_UP))
1487 call_netdevice_notifier(nb, NETDEV_UP, dev);
1498 for_each_netdev(net, dev) {
1502 if (dev->flags & IFF_UP) {
1503 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1505 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1507 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1512 raw_notifier_chain_unregister(&netdev_chain, nb);
1515 EXPORT_SYMBOL(register_netdevice_notifier);
1518 * unregister_netdevice_notifier - unregister a network notifier block
1521 * Unregister a notifier previously registered by
1522 * register_netdevice_notifier(). The notifier is unlinked into the
1523 * kernel structures and may then be reused. A negative errno code
1524 * is returned on a failure.
1526 * After unregistering unregister and down device events are synthesized
1527 * for all devices on the device list to the removed notifier to remove
1528 * the need for special case cleanup code.
1531 int unregister_netdevice_notifier(struct notifier_block *nb)
1533 struct net_device *dev;
1538 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1543 for_each_netdev(net, dev) {
1544 if (dev->flags & IFF_UP) {
1545 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1547 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1549 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1556 EXPORT_SYMBOL(unregister_netdevice_notifier);
1559 * call_netdevice_notifiers_info - call all network notifier blocks
1560 * @val: value passed unmodified to notifier function
1561 * @dev: net_device pointer passed unmodified to notifier function
1562 * @info: notifier information data
1564 * Call all network notifier blocks. Parameters and return value
1565 * are as for raw_notifier_call_chain().
1568 static int call_netdevice_notifiers_info(unsigned long val,
1569 struct net_device *dev,
1570 struct netdev_notifier_info *info)
1573 netdev_notifier_info_init(info, dev);
1574 return raw_notifier_call_chain(&netdev_chain, val, info);
1578 * call_netdevice_notifiers - call all network notifier blocks
1579 * @val: value passed unmodified to notifier function
1580 * @dev: net_device pointer passed unmodified to notifier function
1582 * Call all network notifier blocks. Parameters and return value
1583 * are as for raw_notifier_call_chain().
1586 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1588 struct netdev_notifier_info info;
1590 return call_netdevice_notifiers_info(val, dev, &info);
1592 EXPORT_SYMBOL(call_netdevice_notifiers);
1594 static struct static_key netstamp_needed __read_mostly;
1595 #ifdef HAVE_JUMP_LABEL
1596 /* We are not allowed to call static_key_slow_dec() from irq context
1597 * If net_disable_timestamp() is called from irq context, defer the
1598 * static_key_slow_dec() calls.
1600 static atomic_t netstamp_needed_deferred;
1603 void net_enable_timestamp(void)
1605 #ifdef HAVE_JUMP_LABEL
1606 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1610 static_key_slow_dec(&netstamp_needed);
1614 static_key_slow_inc(&netstamp_needed);
1616 EXPORT_SYMBOL(net_enable_timestamp);
1618 void net_disable_timestamp(void)
1620 #ifdef HAVE_JUMP_LABEL
1621 if (in_interrupt()) {
1622 atomic_inc(&netstamp_needed_deferred);
1626 static_key_slow_dec(&netstamp_needed);
1628 EXPORT_SYMBOL(net_disable_timestamp);
1630 static inline void net_timestamp_set(struct sk_buff *skb)
1632 skb->tstamp.tv64 = 0;
1633 if (static_key_false(&netstamp_needed))
1634 __net_timestamp(skb);
1637 #define net_timestamp_check(COND, SKB) \
1638 if (static_key_false(&netstamp_needed)) { \
1639 if ((COND) && !(SKB)->tstamp.tv64) \
1640 __net_timestamp(SKB); \
1643 static inline bool is_skb_forwardable(struct net_device *dev,
1644 struct sk_buff *skb)
1648 if (!(dev->flags & IFF_UP))
1651 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1652 if (skb->len <= len)
1655 /* if TSO is enabled, we don't care about the length as the packet
1656 * could be forwarded without being segmented before
1658 if (skb_is_gso(skb))
1665 * dev_forward_skb - loopback an skb to another netif
1667 * @dev: destination network device
1668 * @skb: buffer to forward
1671 * NET_RX_SUCCESS (no congestion)
1672 * NET_RX_DROP (packet was dropped, but freed)
1674 * dev_forward_skb can be used for injecting an skb from the
1675 * start_xmit function of one device into the receive queue
1676 * of another device.
1678 * The receiving device may be in another namespace, so
1679 * we have to clear all information in the skb that could
1680 * impact namespace isolation.
1682 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1684 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1685 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1686 atomic_long_inc(&dev->rx_dropped);
1692 if (unlikely(!is_skb_forwardable(dev, skb))) {
1693 atomic_long_inc(&dev->rx_dropped);
1698 skb_scrub_packet(skb, true);
1699 skb->protocol = eth_type_trans(skb, dev);
1701 return netif_rx_internal(skb);
1703 EXPORT_SYMBOL_GPL(dev_forward_skb);
1705 static inline int deliver_skb(struct sk_buff *skb,
1706 struct packet_type *pt_prev,
1707 struct net_device *orig_dev)
1709 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1711 atomic_inc(&skb->users);
1712 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1715 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1717 if (!ptype->af_packet_priv || !skb->sk)
1720 if (ptype->id_match)
1721 return ptype->id_match(ptype, skb->sk);
1722 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1729 * Support routine. Sends outgoing frames to any network
1730 * taps currently in use.
1733 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1735 struct packet_type *ptype;
1736 struct sk_buff *skb2 = NULL;
1737 struct packet_type *pt_prev = NULL;
1740 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1741 /* Never send packets back to the socket
1742 * they originated from - MvS (miquels@drinkel.ow.org)
1744 if ((ptype->dev == dev || !ptype->dev) &&
1745 (!skb_loop_sk(ptype, skb))) {
1747 deliver_skb(skb2, pt_prev, skb->dev);
1752 skb2 = skb_clone(skb, GFP_ATOMIC);
1756 net_timestamp_set(skb2);
1758 /* skb->nh should be correctly
1759 set by sender, so that the second statement is
1760 just protection against buggy protocols.
1762 skb_reset_mac_header(skb2);
1764 if (skb_network_header(skb2) < skb2->data ||
1765 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1766 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1767 ntohs(skb2->protocol),
1769 skb_reset_network_header(skb2);
1772 skb2->transport_header = skb2->network_header;
1773 skb2->pkt_type = PACKET_OUTGOING;
1778 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1783 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1784 * @dev: Network device
1785 * @txq: number of queues available
1787 * If real_num_tx_queues is changed the tc mappings may no longer be
1788 * valid. To resolve this verify the tc mapping remains valid and if
1789 * not NULL the mapping. With no priorities mapping to this
1790 * offset/count pair it will no longer be used. In the worst case TC0
1791 * is invalid nothing can be done so disable priority mappings. If is
1792 * expected that drivers will fix this mapping if they can before
1793 * calling netif_set_real_num_tx_queues.
1795 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1798 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1800 /* If TC0 is invalidated disable TC mapping */
1801 if (tc->offset + tc->count > txq) {
1802 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1807 /* Invalidated prio to tc mappings set to TC0 */
1808 for (i = 1; i < TC_BITMASK + 1; i++) {
1809 int q = netdev_get_prio_tc_map(dev, i);
1811 tc = &dev->tc_to_txq[q];
1812 if (tc->offset + tc->count > txq) {
1813 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1815 netdev_set_prio_tc_map(dev, i, 0);
1821 static DEFINE_MUTEX(xps_map_mutex);
1822 #define xmap_dereference(P) \
1823 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1825 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1828 struct xps_map *map = NULL;
1832 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1834 for (pos = 0; map && pos < map->len; pos++) {
1835 if (map->queues[pos] == index) {
1837 map->queues[pos] = map->queues[--map->len];
1839 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1840 kfree_rcu(map, rcu);
1850 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1852 struct xps_dev_maps *dev_maps;
1854 bool active = false;
1856 mutex_lock(&xps_map_mutex);
1857 dev_maps = xmap_dereference(dev->xps_maps);
1862 for_each_possible_cpu(cpu) {
1863 for (i = index; i < dev->num_tx_queues; i++) {
1864 if (!remove_xps_queue(dev_maps, cpu, i))
1867 if (i == dev->num_tx_queues)
1872 RCU_INIT_POINTER(dev->xps_maps, NULL);
1873 kfree_rcu(dev_maps, rcu);
1876 for (i = index; i < dev->num_tx_queues; i++)
1877 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1881 mutex_unlock(&xps_map_mutex);
1884 static struct xps_map *expand_xps_map(struct xps_map *map,
1887 struct xps_map *new_map;
1888 int alloc_len = XPS_MIN_MAP_ALLOC;
1891 for (pos = 0; map && pos < map->len; pos++) {
1892 if (map->queues[pos] != index)
1897 /* Need to add queue to this CPU's existing map */
1899 if (pos < map->alloc_len)
1902 alloc_len = map->alloc_len * 2;
1905 /* Need to allocate new map to store queue on this CPU's map */
1906 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1911 for (i = 0; i < pos; i++)
1912 new_map->queues[i] = map->queues[i];
1913 new_map->alloc_len = alloc_len;
1919 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1922 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1923 struct xps_map *map, *new_map;
1924 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1925 int cpu, numa_node_id = -2;
1926 bool active = false;
1928 mutex_lock(&xps_map_mutex);
1930 dev_maps = xmap_dereference(dev->xps_maps);
1932 /* allocate memory for queue storage */
1933 for_each_online_cpu(cpu) {
1934 if (!cpumask_test_cpu(cpu, mask))
1938 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1939 if (!new_dev_maps) {
1940 mutex_unlock(&xps_map_mutex);
1944 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1947 map = expand_xps_map(map, cpu, index);
1951 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1955 goto out_no_new_maps;
1957 for_each_possible_cpu(cpu) {
1958 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1959 /* add queue to CPU maps */
1962 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1963 while ((pos < map->len) && (map->queues[pos] != index))
1966 if (pos == map->len)
1967 map->queues[map->len++] = index;
1969 if (numa_node_id == -2)
1970 numa_node_id = cpu_to_node(cpu);
1971 else if (numa_node_id != cpu_to_node(cpu))
1974 } else if (dev_maps) {
1975 /* fill in the new device map from the old device map */
1976 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1977 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1982 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1984 /* Cleanup old maps */
1986 for_each_possible_cpu(cpu) {
1987 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1988 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1989 if (map && map != new_map)
1990 kfree_rcu(map, rcu);
1993 kfree_rcu(dev_maps, rcu);
1996 dev_maps = new_dev_maps;
2000 /* update Tx queue numa node */
2001 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2002 (numa_node_id >= 0) ? numa_node_id :
2008 /* removes queue from unused CPUs */
2009 for_each_possible_cpu(cpu) {
2010 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2013 if (remove_xps_queue(dev_maps, cpu, index))
2017 /* free map if not active */
2019 RCU_INIT_POINTER(dev->xps_maps, NULL);
2020 kfree_rcu(dev_maps, rcu);
2024 mutex_unlock(&xps_map_mutex);
2028 /* remove any maps that we added */
2029 for_each_possible_cpu(cpu) {
2030 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2031 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2033 if (new_map && new_map != map)
2037 mutex_unlock(&xps_map_mutex);
2039 kfree(new_dev_maps);
2042 EXPORT_SYMBOL(netif_set_xps_queue);
2046 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2047 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2049 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2053 if (txq < 1 || txq > dev->num_tx_queues)
2056 if (dev->reg_state == NETREG_REGISTERED ||
2057 dev->reg_state == NETREG_UNREGISTERING) {
2060 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2066 netif_setup_tc(dev, txq);
2068 if (txq < dev->real_num_tx_queues) {
2069 qdisc_reset_all_tx_gt(dev, txq);
2071 netif_reset_xps_queues_gt(dev, txq);
2076 dev->real_num_tx_queues = txq;
2079 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2083 * netif_set_real_num_rx_queues - set actual number of RX queues used
2084 * @dev: Network device
2085 * @rxq: Actual number of RX queues
2087 * This must be called either with the rtnl_lock held or before
2088 * registration of the net device. Returns 0 on success, or a
2089 * negative error code. If called before registration, it always
2092 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2096 if (rxq < 1 || rxq > dev->num_rx_queues)
2099 if (dev->reg_state == NETREG_REGISTERED) {
2102 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2108 dev->real_num_rx_queues = rxq;
2111 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2115 * netif_get_num_default_rss_queues - default number of RSS queues
2117 * This routine should set an upper limit on the number of RSS queues
2118 * used by default by multiqueue devices.
2120 int netif_get_num_default_rss_queues(void)
2122 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2124 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2126 static inline void __netif_reschedule(struct Qdisc *q)
2128 struct softnet_data *sd;
2129 unsigned long flags;
2131 local_irq_save(flags);
2132 sd = &__get_cpu_var(softnet_data);
2133 q->next_sched = NULL;
2134 *sd->output_queue_tailp = q;
2135 sd->output_queue_tailp = &q->next_sched;
2136 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2137 local_irq_restore(flags);
2140 void __netif_schedule(struct Qdisc *q)
2142 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2143 __netif_reschedule(q);
2145 EXPORT_SYMBOL(__netif_schedule);
2147 struct dev_kfree_skb_cb {
2148 enum skb_free_reason reason;
2151 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2153 return (struct dev_kfree_skb_cb *)skb->cb;
2156 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2158 unsigned long flags;
2160 if (likely(atomic_read(&skb->users) == 1)) {
2162 atomic_set(&skb->users, 0);
2163 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2166 get_kfree_skb_cb(skb)->reason = reason;
2167 local_irq_save(flags);
2168 skb->next = __this_cpu_read(softnet_data.completion_queue);
2169 __this_cpu_write(softnet_data.completion_queue, skb);
2170 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2171 local_irq_restore(flags);
2173 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2175 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2177 if (in_irq() || irqs_disabled())
2178 __dev_kfree_skb_irq(skb, reason);
2182 EXPORT_SYMBOL(__dev_kfree_skb_any);
2186 * netif_device_detach - mark device as removed
2187 * @dev: network device
2189 * Mark device as removed from system and therefore no longer available.
2191 void netif_device_detach(struct net_device *dev)
2193 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2194 netif_running(dev)) {
2195 netif_tx_stop_all_queues(dev);
2198 EXPORT_SYMBOL(netif_device_detach);
2201 * netif_device_attach - mark device as attached
2202 * @dev: network device
2204 * Mark device as attached from system and restart if needed.
2206 void netif_device_attach(struct net_device *dev)
2208 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2209 netif_running(dev)) {
2210 netif_tx_wake_all_queues(dev);
2211 __netdev_watchdog_up(dev);
2214 EXPORT_SYMBOL(netif_device_attach);
2216 static void skb_warn_bad_offload(const struct sk_buff *skb)
2218 static const netdev_features_t null_features = 0;
2219 struct net_device *dev = skb->dev;
2220 const char *driver = "";
2222 if (!net_ratelimit())
2225 if (dev && dev->dev.parent)
2226 driver = dev_driver_string(dev->dev.parent);
2228 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2229 "gso_type=%d ip_summed=%d\n",
2230 driver, dev ? &dev->features : &null_features,
2231 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2232 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2233 skb_shinfo(skb)->gso_type, skb->ip_summed);
2237 * Invalidate hardware checksum when packet is to be mangled, and
2238 * complete checksum manually on outgoing path.
2240 int skb_checksum_help(struct sk_buff *skb)
2243 int ret = 0, offset;
2245 if (skb->ip_summed == CHECKSUM_COMPLETE)
2246 goto out_set_summed;
2248 if (unlikely(skb_shinfo(skb)->gso_size)) {
2249 skb_warn_bad_offload(skb);
2253 /* Before computing a checksum, we should make sure no frag could
2254 * be modified by an external entity : checksum could be wrong.
2256 if (skb_has_shared_frag(skb)) {
2257 ret = __skb_linearize(skb);
2262 offset = skb_checksum_start_offset(skb);
2263 BUG_ON(offset >= skb_headlen(skb));
2264 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2266 offset += skb->csum_offset;
2267 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2269 if (skb_cloned(skb) &&
2270 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2271 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2276 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2278 skb->ip_summed = CHECKSUM_NONE;
2282 EXPORT_SYMBOL(skb_checksum_help);
2284 __be16 skb_network_protocol(struct sk_buff *skb)
2286 __be16 type = skb->protocol;
2287 int vlan_depth = ETH_HLEN;
2289 /* Tunnel gso handlers can set protocol to ethernet. */
2290 if (type == htons(ETH_P_TEB)) {
2293 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2296 eth = (struct ethhdr *)skb_mac_header(skb);
2297 type = eth->h_proto;
2300 while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2301 struct vlan_hdr *vh;
2303 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2306 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2307 type = vh->h_vlan_encapsulated_proto;
2308 vlan_depth += VLAN_HLEN;
2315 * skb_mac_gso_segment - mac layer segmentation handler.
2316 * @skb: buffer to segment
2317 * @features: features for the output path (see dev->features)
2319 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2320 netdev_features_t features)
2322 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2323 struct packet_offload *ptype;
2324 __be16 type = skb_network_protocol(skb);
2326 if (unlikely(!type))
2327 return ERR_PTR(-EINVAL);
2329 __skb_pull(skb, skb->mac_len);
2332 list_for_each_entry_rcu(ptype, &offload_base, list) {
2333 if (ptype->type == type && ptype->callbacks.gso_segment) {
2334 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2337 err = ptype->callbacks.gso_send_check(skb);
2338 segs = ERR_PTR(err);
2339 if (err || skb_gso_ok(skb, features))
2341 __skb_push(skb, (skb->data -
2342 skb_network_header(skb)));
2344 segs = ptype->callbacks.gso_segment(skb, features);
2350 __skb_push(skb, skb->data - skb_mac_header(skb));
2354 EXPORT_SYMBOL(skb_mac_gso_segment);
2357 /* openvswitch calls this on rx path, so we need a different check.
2359 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2362 return skb->ip_summed != CHECKSUM_PARTIAL;
2364 return skb->ip_summed == CHECKSUM_NONE;
2368 * __skb_gso_segment - Perform segmentation on skb.
2369 * @skb: buffer to segment
2370 * @features: features for the output path (see dev->features)
2371 * @tx_path: whether it is called in TX path
2373 * This function segments the given skb and returns a list of segments.
2375 * It may return NULL if the skb requires no segmentation. This is
2376 * only possible when GSO is used for verifying header integrity.
2378 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2379 netdev_features_t features, bool tx_path)
2381 if (unlikely(skb_needs_check(skb, tx_path))) {
2384 skb_warn_bad_offload(skb);
2386 if (skb_header_cloned(skb) &&
2387 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2388 return ERR_PTR(err);
2391 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2392 SKB_GSO_CB(skb)->encap_level = 0;
2394 skb_reset_mac_header(skb);
2395 skb_reset_mac_len(skb);
2397 return skb_mac_gso_segment(skb, features);
2399 EXPORT_SYMBOL(__skb_gso_segment);
2401 /* Take action when hardware reception checksum errors are detected. */
2403 void netdev_rx_csum_fault(struct net_device *dev)
2405 if (net_ratelimit()) {
2406 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2410 EXPORT_SYMBOL(netdev_rx_csum_fault);
2413 /* Actually, we should eliminate this check as soon as we know, that:
2414 * 1. IOMMU is present and allows to map all the memory.
2415 * 2. No high memory really exists on this machine.
2418 static int illegal_highdma(const struct net_device *dev, struct sk_buff *skb)
2420 #ifdef CONFIG_HIGHMEM
2422 if (!(dev->features & NETIF_F_HIGHDMA)) {
2423 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2424 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2425 if (PageHighMem(skb_frag_page(frag)))
2430 if (PCI_DMA_BUS_IS_PHYS) {
2431 struct device *pdev = dev->dev.parent;
2435 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2436 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2437 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2438 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2447 void (*destructor)(struct sk_buff *skb);
2450 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2452 static void dev_gso_skb_destructor(struct sk_buff *skb)
2454 struct dev_gso_cb *cb;
2456 kfree_skb_list(skb->next);
2459 cb = DEV_GSO_CB(skb);
2461 cb->destructor(skb);
2465 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2466 * @skb: buffer to segment
2467 * @features: device features as applicable to this skb
2469 * This function segments the given skb and stores the list of segments
2472 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2474 struct sk_buff *segs;
2476 segs = skb_gso_segment(skb, features);
2478 /* Verifying header integrity only. */
2483 return PTR_ERR(segs);
2486 DEV_GSO_CB(skb)->destructor = skb->destructor;
2487 skb->destructor = dev_gso_skb_destructor;
2492 static netdev_features_t harmonize_features(struct sk_buff *skb,
2493 const struct net_device *dev,
2494 netdev_features_t features)
2496 if (skb->ip_summed != CHECKSUM_NONE &&
2497 !can_checksum_protocol(features, skb_network_protocol(skb))) {
2498 features &= ~NETIF_F_ALL_CSUM;
2499 } else if (illegal_highdma(dev, skb)) {
2500 features &= ~NETIF_F_SG;
2506 netdev_features_t netif_skb_dev_features(struct sk_buff *skb,
2507 const struct net_device *dev)
2509 __be16 protocol = skb->protocol;
2510 netdev_features_t features = dev->features;
2512 if (skb_shinfo(skb)->gso_segs > dev->gso_max_segs)
2513 features &= ~NETIF_F_GSO_MASK;
2515 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2516 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2517 protocol = veh->h_vlan_encapsulated_proto;
2518 } else if (!vlan_tx_tag_present(skb)) {
2519 return harmonize_features(skb, dev, features);
2522 features &= (dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2523 NETIF_F_HW_VLAN_STAG_TX);
2525 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2526 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2527 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2528 NETIF_F_HW_VLAN_STAG_TX;
2530 return harmonize_features(skb, dev, features);
2532 EXPORT_SYMBOL(netif_skb_dev_features);
2534 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2535 struct netdev_queue *txq)
2537 const struct net_device_ops *ops = dev->netdev_ops;
2538 int rc = NETDEV_TX_OK;
2539 unsigned int skb_len;
2541 if (likely(!skb->next)) {
2542 netdev_features_t features;
2545 * If device doesn't need skb->dst, release it right now while
2546 * its hot in this cpu cache
2548 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2551 features = netif_skb_features(skb);
2553 if (vlan_tx_tag_present(skb) &&
2554 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2555 skb = __vlan_put_tag(skb, skb->vlan_proto,
2556 vlan_tx_tag_get(skb));
2563 /* If encapsulation offload request, verify we are testing
2564 * hardware encapsulation features instead of standard
2565 * features for the netdev
2567 if (skb->encapsulation)
2568 features &= dev->hw_enc_features;
2570 if (netif_needs_gso(skb, features)) {
2571 if (unlikely(dev_gso_segment(skb, features)))
2576 if (skb_needs_linearize(skb, features) &&
2577 __skb_linearize(skb))
2580 /* If packet is not checksummed and device does not
2581 * support checksumming for this protocol, complete
2582 * checksumming here.
2584 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2585 if (skb->encapsulation)
2586 skb_set_inner_transport_header(skb,
2587 skb_checksum_start_offset(skb));
2589 skb_set_transport_header(skb,
2590 skb_checksum_start_offset(skb));
2591 if (!(features & NETIF_F_ALL_CSUM) &&
2592 skb_checksum_help(skb))
2597 if (!list_empty(&ptype_all))
2598 dev_queue_xmit_nit(skb, dev);
2601 trace_net_dev_start_xmit(skb, dev);
2602 rc = ops->ndo_start_xmit(skb, dev);
2603 trace_net_dev_xmit(skb, rc, dev, skb_len);
2604 if (rc == NETDEV_TX_OK)
2605 txq_trans_update(txq);
2611 struct sk_buff *nskb = skb->next;
2613 skb->next = nskb->next;
2616 if (!list_empty(&ptype_all))
2617 dev_queue_xmit_nit(nskb, dev);
2619 skb_len = nskb->len;
2620 trace_net_dev_start_xmit(nskb, dev);
2621 rc = ops->ndo_start_xmit(nskb, dev);
2622 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2623 if (unlikely(rc != NETDEV_TX_OK)) {
2624 if (rc & ~NETDEV_TX_MASK)
2625 goto out_kfree_gso_skb;
2626 nskb->next = skb->next;
2630 txq_trans_update(txq);
2631 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2632 return NETDEV_TX_BUSY;
2633 } while (skb->next);
2636 if (likely(skb->next == NULL)) {
2637 skb->destructor = DEV_GSO_CB(skb)->destructor;
2646 EXPORT_SYMBOL_GPL(dev_hard_start_xmit);
2648 static void qdisc_pkt_len_init(struct sk_buff *skb)
2650 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2652 qdisc_skb_cb(skb)->pkt_len = skb->len;
2654 /* To get more precise estimation of bytes sent on wire,
2655 * we add to pkt_len the headers size of all segments
2657 if (shinfo->gso_size) {
2658 unsigned int hdr_len;
2659 u16 gso_segs = shinfo->gso_segs;
2661 /* mac layer + network layer */
2662 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2664 /* + transport layer */
2665 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2666 hdr_len += tcp_hdrlen(skb);
2668 hdr_len += sizeof(struct udphdr);
2670 if (shinfo->gso_type & SKB_GSO_DODGY)
2671 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2674 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2678 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2679 struct net_device *dev,
2680 struct netdev_queue *txq)
2682 spinlock_t *root_lock = qdisc_lock(q);
2686 qdisc_pkt_len_init(skb);
2687 qdisc_calculate_pkt_len(skb, q);
2689 * Heuristic to force contended enqueues to serialize on a
2690 * separate lock before trying to get qdisc main lock.
2691 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2692 * and dequeue packets faster.
2694 contended = qdisc_is_running(q);
2695 if (unlikely(contended))
2696 spin_lock(&q->busylock);
2698 spin_lock(root_lock);
2699 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2702 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2703 qdisc_run_begin(q)) {
2705 * This is a work-conserving queue; there are no old skbs
2706 * waiting to be sent out; and the qdisc is not running -
2707 * xmit the skb directly.
2709 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2712 qdisc_bstats_update(q, skb);
2714 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2715 if (unlikely(contended)) {
2716 spin_unlock(&q->busylock);
2723 rc = NET_XMIT_SUCCESS;
2726 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2727 if (qdisc_run_begin(q)) {
2728 if (unlikely(contended)) {
2729 spin_unlock(&q->busylock);
2735 spin_unlock(root_lock);
2736 if (unlikely(contended))
2737 spin_unlock(&q->busylock);
2741 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2742 static void skb_update_prio(struct sk_buff *skb)
2744 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2746 if (!skb->priority && skb->sk && map) {
2747 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2749 if (prioidx < map->priomap_len)
2750 skb->priority = map->priomap[prioidx];
2754 #define skb_update_prio(skb)
2757 static DEFINE_PER_CPU(int, xmit_recursion);
2758 #define RECURSION_LIMIT 10
2761 * dev_loopback_xmit - loop back @skb
2762 * @skb: buffer to transmit
2764 int dev_loopback_xmit(struct sk_buff *skb)
2766 skb_reset_mac_header(skb);
2767 __skb_pull(skb, skb_network_offset(skb));
2768 skb->pkt_type = PACKET_LOOPBACK;
2769 skb->ip_summed = CHECKSUM_UNNECESSARY;
2770 WARN_ON(!skb_dst(skb));
2775 EXPORT_SYMBOL(dev_loopback_xmit);
2778 * __dev_queue_xmit - transmit a buffer
2779 * @skb: buffer to transmit
2780 * @accel_priv: private data used for L2 forwarding offload
2782 * Queue a buffer for transmission to a network device. The caller must
2783 * have set the device and priority and built the buffer before calling
2784 * this function. The function can be called from an interrupt.
2786 * A negative errno code is returned on a failure. A success does not
2787 * guarantee the frame will be transmitted as it may be dropped due
2788 * to congestion or traffic shaping.
2790 * -----------------------------------------------------------------------------------
2791 * I notice this method can also return errors from the queue disciplines,
2792 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2795 * Regardless of the return value, the skb is consumed, so it is currently
2796 * difficult to retry a send to this method. (You can bump the ref count
2797 * before sending to hold a reference for retry if you are careful.)
2799 * When calling this method, interrupts MUST be enabled. This is because
2800 * the BH enable code must have IRQs enabled so that it will not deadlock.
2803 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2805 struct net_device *dev = skb->dev;
2806 struct netdev_queue *txq;
2810 skb_reset_mac_header(skb);
2812 /* Disable soft irqs for various locks below. Also
2813 * stops preemption for RCU.
2817 skb_update_prio(skb);
2819 txq = netdev_pick_tx(dev, skb, accel_priv);
2820 q = rcu_dereference_bh(txq->qdisc);
2822 #ifdef CONFIG_NET_CLS_ACT
2823 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2825 trace_net_dev_queue(skb);
2827 rc = __dev_xmit_skb(skb, q, dev, txq);
2831 /* The device has no queue. Common case for software devices:
2832 loopback, all the sorts of tunnels...
2834 Really, it is unlikely that netif_tx_lock protection is necessary
2835 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2837 However, it is possible, that they rely on protection
2840 Check this and shot the lock. It is not prone from deadlocks.
2841 Either shot noqueue qdisc, it is even simpler 8)
2843 if (dev->flags & IFF_UP) {
2844 int cpu = smp_processor_id(); /* ok because BHs are off */
2846 if (txq->xmit_lock_owner != cpu) {
2848 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2849 goto recursion_alert;
2851 HARD_TX_LOCK(dev, txq, cpu);
2853 if (!netif_xmit_stopped(txq)) {
2854 __this_cpu_inc(xmit_recursion);
2855 rc = dev_hard_start_xmit(skb, dev, txq);
2856 __this_cpu_dec(xmit_recursion);
2857 if (dev_xmit_complete(rc)) {
2858 HARD_TX_UNLOCK(dev, txq);
2862 HARD_TX_UNLOCK(dev, txq);
2863 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2866 /* Recursion is detected! It is possible,
2870 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2876 rcu_read_unlock_bh();
2878 atomic_long_inc(&dev->tx_dropped);
2882 rcu_read_unlock_bh();
2886 int dev_queue_xmit(struct sk_buff *skb)
2888 return __dev_queue_xmit(skb, NULL);
2890 EXPORT_SYMBOL(dev_queue_xmit);
2892 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
2894 return __dev_queue_xmit(skb, accel_priv);
2896 EXPORT_SYMBOL(dev_queue_xmit_accel);
2899 /*=======================================================================
2901 =======================================================================*/
2903 int netdev_max_backlog __read_mostly = 1000;
2904 EXPORT_SYMBOL(netdev_max_backlog);
2906 int netdev_tstamp_prequeue __read_mostly = 1;
2907 int netdev_budget __read_mostly = 300;
2908 int weight_p __read_mostly = 64; /* old backlog weight */
2910 /* Called with irq disabled */
2911 static inline void ____napi_schedule(struct softnet_data *sd,
2912 struct napi_struct *napi)
2914 list_add_tail(&napi->poll_list, &sd->poll_list);
2915 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2920 /* One global table that all flow-based protocols share. */
2921 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2922 EXPORT_SYMBOL(rps_sock_flow_table);
2924 struct static_key rps_needed __read_mostly;
2926 static struct rps_dev_flow *
2927 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2928 struct rps_dev_flow *rflow, u16 next_cpu)
2930 if (next_cpu != RPS_NO_CPU) {
2931 #ifdef CONFIG_RFS_ACCEL
2932 struct netdev_rx_queue *rxqueue;
2933 struct rps_dev_flow_table *flow_table;
2934 struct rps_dev_flow *old_rflow;
2939 /* Should we steer this flow to a different hardware queue? */
2940 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2941 !(dev->features & NETIF_F_NTUPLE))
2943 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2944 if (rxq_index == skb_get_rx_queue(skb))
2947 rxqueue = dev->_rx + rxq_index;
2948 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2951 flow_id = skb_get_hash(skb) & flow_table->mask;
2952 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2953 rxq_index, flow_id);
2957 rflow = &flow_table->flows[flow_id];
2959 if (old_rflow->filter == rflow->filter)
2960 old_rflow->filter = RPS_NO_FILTER;
2964 per_cpu(softnet_data, next_cpu).input_queue_head;
2967 rflow->cpu = next_cpu;
2972 * get_rps_cpu is called from netif_receive_skb and returns the target
2973 * CPU from the RPS map of the receiving queue for a given skb.
2974 * rcu_read_lock must be held on entry.
2976 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2977 struct rps_dev_flow **rflowp)
2979 struct netdev_rx_queue *rxqueue;
2980 struct rps_map *map;
2981 struct rps_dev_flow_table *flow_table;
2982 struct rps_sock_flow_table *sock_flow_table;
2987 if (skb_rx_queue_recorded(skb)) {
2988 u16 index = skb_get_rx_queue(skb);
2989 if (unlikely(index >= dev->real_num_rx_queues)) {
2990 WARN_ONCE(dev->real_num_rx_queues > 1,
2991 "%s received packet on queue %u, but number "
2992 "of RX queues is %u\n",
2993 dev->name, index, dev->real_num_rx_queues);
2996 rxqueue = dev->_rx + index;
3000 map = rcu_dereference(rxqueue->rps_map);
3002 if (map->len == 1 &&
3003 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3004 tcpu = map->cpus[0];
3005 if (cpu_online(tcpu))
3009 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3013 skb_reset_network_header(skb);
3014 hash = skb_get_hash(skb);
3018 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3019 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3020 if (flow_table && sock_flow_table) {
3022 struct rps_dev_flow *rflow;
3024 rflow = &flow_table->flows[hash & flow_table->mask];
3027 next_cpu = sock_flow_table->ents[hash & sock_flow_table->mask];
3030 * If the desired CPU (where last recvmsg was done) is
3031 * different from current CPU (one in the rx-queue flow
3032 * table entry), switch if one of the following holds:
3033 * - Current CPU is unset (equal to RPS_NO_CPU).
3034 * - Current CPU is offline.
3035 * - The current CPU's queue tail has advanced beyond the
3036 * last packet that was enqueued using this table entry.
3037 * This guarantees that all previous packets for the flow
3038 * have been dequeued, thus preserving in order delivery.
3040 if (unlikely(tcpu != next_cpu) &&
3041 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3042 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3043 rflow->last_qtail)) >= 0)) {
3045 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3048 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3056 tcpu = map->cpus[((u64) hash * map->len) >> 32];
3058 if (cpu_online(tcpu)) {
3068 #ifdef CONFIG_RFS_ACCEL
3071 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3072 * @dev: Device on which the filter was set
3073 * @rxq_index: RX queue index
3074 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3075 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3077 * Drivers that implement ndo_rx_flow_steer() should periodically call
3078 * this function for each installed filter and remove the filters for
3079 * which it returns %true.
3081 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3082 u32 flow_id, u16 filter_id)
3084 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3085 struct rps_dev_flow_table *flow_table;
3086 struct rps_dev_flow *rflow;
3091 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3092 if (flow_table && flow_id <= flow_table->mask) {
3093 rflow = &flow_table->flows[flow_id];
3094 cpu = ACCESS_ONCE(rflow->cpu);
3095 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3096 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3097 rflow->last_qtail) <
3098 (int)(10 * flow_table->mask)))
3104 EXPORT_SYMBOL(rps_may_expire_flow);
3106 #endif /* CONFIG_RFS_ACCEL */
3108 /* Called from hardirq (IPI) context */
3109 static void rps_trigger_softirq(void *data)
3111 struct softnet_data *sd = data;
3113 ____napi_schedule(sd, &sd->backlog);
3117 #endif /* CONFIG_RPS */
3120 * Check if this softnet_data structure is another cpu one
3121 * If yes, queue it to our IPI list and return 1
3124 static int rps_ipi_queued(struct softnet_data *sd)
3127 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3130 sd->rps_ipi_next = mysd->rps_ipi_list;
3131 mysd->rps_ipi_list = sd;
3133 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3136 #endif /* CONFIG_RPS */
3140 #ifdef CONFIG_NET_FLOW_LIMIT
3141 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3144 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3146 #ifdef CONFIG_NET_FLOW_LIMIT
3147 struct sd_flow_limit *fl;
3148 struct softnet_data *sd;
3149 unsigned int old_flow, new_flow;
3151 if (qlen < (netdev_max_backlog >> 1))
3154 sd = &__get_cpu_var(softnet_data);
3157 fl = rcu_dereference(sd->flow_limit);
3159 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3160 old_flow = fl->history[fl->history_head];
3161 fl->history[fl->history_head] = new_flow;
3164 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3166 if (likely(fl->buckets[old_flow]))
3167 fl->buckets[old_flow]--;
3169 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3181 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3182 * queue (may be a remote CPU queue).
3184 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3185 unsigned int *qtail)
3187 struct softnet_data *sd;
3188 unsigned long flags;
3191 sd = &per_cpu(softnet_data, cpu);
3193 local_irq_save(flags);
3196 qlen = skb_queue_len(&sd->input_pkt_queue);
3197 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3198 if (skb_queue_len(&sd->input_pkt_queue)) {
3200 __skb_queue_tail(&sd->input_pkt_queue, skb);
3201 input_queue_tail_incr_save(sd, qtail);
3203 local_irq_restore(flags);
3204 return NET_RX_SUCCESS;
3207 /* Schedule NAPI for backlog device
3208 * We can use non atomic operation since we own the queue lock
3210 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3211 if (!rps_ipi_queued(sd))
3212 ____napi_schedule(sd, &sd->backlog);
3220 local_irq_restore(flags);
3222 atomic_long_inc(&skb->dev->rx_dropped);
3227 static int netif_rx_internal(struct sk_buff *skb)
3231 net_timestamp_check(netdev_tstamp_prequeue, skb);
3233 trace_netif_rx(skb);
3235 if (static_key_false(&rps_needed)) {
3236 struct rps_dev_flow voidflow, *rflow = &voidflow;
3242 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3244 cpu = smp_processor_id();
3246 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3254 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3261 * netif_rx - post buffer to the network code
3262 * @skb: buffer to post
3264 * This function receives a packet from a device driver and queues it for
3265 * the upper (protocol) levels to process. It always succeeds. The buffer
3266 * may be dropped during processing for congestion control or by the
3270 * NET_RX_SUCCESS (no congestion)
3271 * NET_RX_DROP (packet was dropped)
3275 int netif_rx(struct sk_buff *skb)
3277 trace_netif_rx_entry(skb);
3279 return netif_rx_internal(skb);
3281 EXPORT_SYMBOL(netif_rx);
3283 int netif_rx_ni(struct sk_buff *skb)
3287 trace_netif_rx_ni_entry(skb);
3290 err = netif_rx_internal(skb);
3291 if (local_softirq_pending())
3297 EXPORT_SYMBOL(netif_rx_ni);
3299 static void net_tx_action(struct softirq_action *h)
3301 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3303 if (sd->completion_queue) {
3304 struct sk_buff *clist;
3306 local_irq_disable();
3307 clist = sd->completion_queue;
3308 sd->completion_queue = NULL;
3312 struct sk_buff *skb = clist;
3313 clist = clist->next;
3315 WARN_ON(atomic_read(&skb->users));
3316 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3317 trace_consume_skb(skb);
3319 trace_kfree_skb(skb, net_tx_action);
3324 if (sd->output_queue) {
3327 local_irq_disable();
3328 head = sd->output_queue;
3329 sd->output_queue = NULL;
3330 sd->output_queue_tailp = &sd->output_queue;
3334 struct Qdisc *q = head;
3335 spinlock_t *root_lock;
3337 head = head->next_sched;
3339 root_lock = qdisc_lock(q);
3340 if (spin_trylock(root_lock)) {
3341 smp_mb__before_clear_bit();
3342 clear_bit(__QDISC_STATE_SCHED,
3345 spin_unlock(root_lock);
3347 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3349 __netif_reschedule(q);
3351 smp_mb__before_clear_bit();
3352 clear_bit(__QDISC_STATE_SCHED,
3360 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3361 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3362 /* This hook is defined here for ATM LANE */
3363 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3364 unsigned char *addr) __read_mostly;
3365 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3368 #ifdef CONFIG_NET_CLS_ACT
3369 /* TODO: Maybe we should just force sch_ingress to be compiled in
3370 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3371 * a compare and 2 stores extra right now if we dont have it on
3372 * but have CONFIG_NET_CLS_ACT
3373 * NOTE: This doesn't stop any functionality; if you dont have
3374 * the ingress scheduler, you just can't add policies on ingress.
3377 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3379 struct net_device *dev = skb->dev;
3380 u32 ttl = G_TC_RTTL(skb->tc_verd);
3381 int result = TC_ACT_OK;
3384 if (unlikely(MAX_RED_LOOP < ttl++)) {
3385 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3386 skb->skb_iif, dev->ifindex);
3390 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3391 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3394 if (q != &noop_qdisc) {
3395 spin_lock(qdisc_lock(q));
3396 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3397 result = qdisc_enqueue_root(skb, q);
3398 spin_unlock(qdisc_lock(q));
3404 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3405 struct packet_type **pt_prev,
3406 int *ret, struct net_device *orig_dev)
3408 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3410 if (!rxq || rxq->qdisc == &noop_qdisc)
3414 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3418 switch (ing_filter(skb, rxq)) {
3432 * netdev_rx_handler_register - register receive handler
3433 * @dev: device to register a handler for
3434 * @rx_handler: receive handler to register
3435 * @rx_handler_data: data pointer that is used by rx handler
3437 * Register a receive hander for a device. This handler will then be
3438 * called from __netif_receive_skb. A negative errno code is returned
3441 * The caller must hold the rtnl_mutex.
3443 * For a general description of rx_handler, see enum rx_handler_result.
3445 int netdev_rx_handler_register(struct net_device *dev,
3446 rx_handler_func_t *rx_handler,
3447 void *rx_handler_data)
3451 if (dev->rx_handler)
3454 /* Note: rx_handler_data must be set before rx_handler */
3455 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3456 rcu_assign_pointer(dev->rx_handler, rx_handler);
3460 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3463 * netdev_rx_handler_unregister - unregister receive handler
3464 * @dev: device to unregister a handler from
3466 * Unregister a receive handler from a device.
3468 * The caller must hold the rtnl_mutex.
3470 void netdev_rx_handler_unregister(struct net_device *dev)
3474 RCU_INIT_POINTER(dev->rx_handler, NULL);
3475 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3476 * section has a guarantee to see a non NULL rx_handler_data
3480 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3482 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3485 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3486 * the special handling of PFMEMALLOC skbs.
3488 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3490 switch (skb->protocol) {
3491 case htons(ETH_P_ARP):
3492 case htons(ETH_P_IP):
3493 case htons(ETH_P_IPV6):
3494 case htons(ETH_P_8021Q):
3495 case htons(ETH_P_8021AD):
3502 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3504 struct packet_type *ptype, *pt_prev;
3505 rx_handler_func_t *rx_handler;
3506 struct net_device *orig_dev;
3507 struct net_device *null_or_dev;
3508 bool deliver_exact = false;
3509 int ret = NET_RX_DROP;
3512 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3514 trace_netif_receive_skb(skb);
3516 orig_dev = skb->dev;
3518 skb_reset_network_header(skb);
3519 if (!skb_transport_header_was_set(skb))
3520 skb_reset_transport_header(skb);
3521 skb_reset_mac_len(skb);
3528 skb->skb_iif = skb->dev->ifindex;
3530 __this_cpu_inc(softnet_data.processed);
3532 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3533 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3534 skb = vlan_untag(skb);
3539 #ifdef CONFIG_NET_CLS_ACT
3540 if (skb->tc_verd & TC_NCLS) {
3541 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3549 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3550 if (!ptype->dev || ptype->dev == skb->dev) {
3552 ret = deliver_skb(skb, pt_prev, orig_dev);
3558 #ifdef CONFIG_NET_CLS_ACT
3559 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3565 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3568 if (vlan_tx_tag_present(skb)) {
3570 ret = deliver_skb(skb, pt_prev, orig_dev);
3573 if (vlan_do_receive(&skb))
3575 else if (unlikely(!skb))
3579 rx_handler = rcu_dereference(skb->dev->rx_handler);
3582 ret = deliver_skb(skb, pt_prev, orig_dev);
3585 switch (rx_handler(&skb)) {
3586 case RX_HANDLER_CONSUMED:
3587 ret = NET_RX_SUCCESS;
3589 case RX_HANDLER_ANOTHER:
3591 case RX_HANDLER_EXACT:
3592 deliver_exact = true;
3593 case RX_HANDLER_PASS:
3600 if (unlikely(vlan_tx_tag_present(skb))) {
3601 if (vlan_tx_tag_get_id(skb))
3602 skb->pkt_type = PACKET_OTHERHOST;
3603 /* Note: we might in the future use prio bits
3604 * and set skb->priority like in vlan_do_receive()
3605 * For the time being, just ignore Priority Code Point
3610 /* deliver only exact match when indicated */
3611 null_or_dev = deliver_exact ? skb->dev : NULL;
3613 type = skb->protocol;
3614 list_for_each_entry_rcu(ptype,
3615 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3616 if (ptype->type == type &&
3617 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3618 ptype->dev == orig_dev)) {
3620 ret = deliver_skb(skb, pt_prev, orig_dev);
3626 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3629 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3632 atomic_long_inc(&skb->dev->rx_dropped);
3634 /* Jamal, now you will not able to escape explaining
3635 * me how you were going to use this. :-)
3645 static int __netif_receive_skb(struct sk_buff *skb)
3649 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3650 unsigned long pflags = current->flags;
3653 * PFMEMALLOC skbs are special, they should
3654 * - be delivered to SOCK_MEMALLOC sockets only
3655 * - stay away from userspace
3656 * - have bounded memory usage
3658 * Use PF_MEMALLOC as this saves us from propagating the allocation
3659 * context down to all allocation sites.
3661 current->flags |= PF_MEMALLOC;
3662 ret = __netif_receive_skb_core(skb, true);
3663 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3665 ret = __netif_receive_skb_core(skb, false);
3670 static int netif_receive_skb_internal(struct sk_buff *skb)
3672 net_timestamp_check(netdev_tstamp_prequeue, skb);
3674 if (skb_defer_rx_timestamp(skb))
3675 return NET_RX_SUCCESS;
3678 if (static_key_false(&rps_needed)) {
3679 struct rps_dev_flow voidflow, *rflow = &voidflow;
3684 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3687 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3694 return __netif_receive_skb(skb);
3698 * netif_receive_skb - process receive buffer from network
3699 * @skb: buffer to process
3701 * netif_receive_skb() is the main receive data processing function.
3702 * It always succeeds. The buffer may be dropped during processing
3703 * for congestion control or by the protocol layers.
3705 * This function may only be called from softirq context and interrupts
3706 * should be enabled.
3708 * Return values (usually ignored):
3709 * NET_RX_SUCCESS: no congestion
3710 * NET_RX_DROP: packet was dropped
3712 int netif_receive_skb(struct sk_buff *skb)
3714 trace_netif_receive_skb_entry(skb);
3716 return netif_receive_skb_internal(skb);
3718 EXPORT_SYMBOL(netif_receive_skb);
3720 /* Network device is going away, flush any packets still pending
3721 * Called with irqs disabled.
3723 static void flush_backlog(void *arg)
3725 struct net_device *dev = arg;
3726 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3727 struct sk_buff *skb, *tmp;
3730 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3731 if (skb->dev == dev) {
3732 __skb_unlink(skb, &sd->input_pkt_queue);
3734 input_queue_head_incr(sd);
3739 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3740 if (skb->dev == dev) {
3741 __skb_unlink(skb, &sd->process_queue);
3743 input_queue_head_incr(sd);
3748 static int napi_gro_complete(struct sk_buff *skb)
3750 struct packet_offload *ptype;
3751 __be16 type = skb->protocol;
3752 struct list_head *head = &offload_base;
3755 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3757 if (NAPI_GRO_CB(skb)->count == 1) {
3758 skb_shinfo(skb)->gso_size = 0;
3763 list_for_each_entry_rcu(ptype, head, list) {
3764 if (ptype->type != type || !ptype->callbacks.gro_complete)
3767 err = ptype->callbacks.gro_complete(skb, 0);
3773 WARN_ON(&ptype->list == head);
3775 return NET_RX_SUCCESS;
3779 return netif_receive_skb_internal(skb);
3782 /* napi->gro_list contains packets ordered by age.
3783 * youngest packets at the head of it.
3784 * Complete skbs in reverse order to reduce latencies.
3786 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3788 struct sk_buff *skb, *prev = NULL;
3790 /* scan list and build reverse chain */
3791 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3796 for (skb = prev; skb; skb = prev) {
3799 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3803 napi_gro_complete(skb);
3807 napi->gro_list = NULL;
3809 EXPORT_SYMBOL(napi_gro_flush);
3811 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3814 unsigned int maclen = skb->dev->hard_header_len;
3815 u32 hash = skb_get_hash_raw(skb);
3817 for (p = napi->gro_list; p; p = p->next) {
3818 unsigned long diffs;
3820 NAPI_GRO_CB(p)->flush = 0;
3822 if (hash != skb_get_hash_raw(p)) {
3823 NAPI_GRO_CB(p)->same_flow = 0;
3827 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3828 diffs |= p->vlan_tci ^ skb->vlan_tci;
3829 if (maclen == ETH_HLEN)
3830 diffs |= compare_ether_header(skb_mac_header(p),
3831 skb_gro_mac_header(skb));
3833 diffs = memcmp(skb_mac_header(p),
3834 skb_gro_mac_header(skb),
3836 NAPI_GRO_CB(p)->same_flow = !diffs;
3840 static void skb_gro_reset_offset(struct sk_buff *skb)
3842 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3843 const skb_frag_t *frag0 = &pinfo->frags[0];
3845 NAPI_GRO_CB(skb)->data_offset = 0;
3846 NAPI_GRO_CB(skb)->frag0 = NULL;
3847 NAPI_GRO_CB(skb)->frag0_len = 0;
3849 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3851 !PageHighMem(skb_frag_page(frag0))) {
3852 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3853 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3857 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3859 struct sk_buff **pp = NULL;
3860 struct packet_offload *ptype;
3861 __be16 type = skb->protocol;
3862 struct list_head *head = &offload_base;
3864 enum gro_result ret;
3866 if (!(skb->dev->features & NETIF_F_GRO))
3869 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3872 skb_gro_reset_offset(skb);
3873 gro_list_prepare(napi, skb);
3874 NAPI_GRO_CB(skb)->csum = skb->csum; /* Needed for CHECKSUM_COMPLETE */
3877 list_for_each_entry_rcu(ptype, head, list) {
3878 if (ptype->type != type || !ptype->callbacks.gro_receive)
3881 skb_set_network_header(skb, skb_gro_offset(skb));
3882 skb_reset_mac_len(skb);
3883 NAPI_GRO_CB(skb)->same_flow = 0;
3884 NAPI_GRO_CB(skb)->flush = 0;
3885 NAPI_GRO_CB(skb)->free = 0;
3886 NAPI_GRO_CB(skb)->udp_mark = 0;
3888 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3893 if (&ptype->list == head)
3896 same_flow = NAPI_GRO_CB(skb)->same_flow;
3897 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3900 struct sk_buff *nskb = *pp;
3904 napi_gro_complete(nskb);
3911 if (NAPI_GRO_CB(skb)->flush)
3914 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
3915 struct sk_buff *nskb = napi->gro_list;
3917 /* locate the end of the list to select the 'oldest' flow */
3918 while (nskb->next) {
3924 napi_gro_complete(nskb);
3928 NAPI_GRO_CB(skb)->count = 1;
3929 NAPI_GRO_CB(skb)->age = jiffies;
3930 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3931 skb->next = napi->gro_list;
3932 napi->gro_list = skb;
3936 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3937 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3939 BUG_ON(skb->end - skb->tail < grow);
3941 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3944 skb->data_len -= grow;
3946 skb_shinfo(skb)->frags[0].page_offset += grow;
3947 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3949 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3950 skb_frag_unref(skb, 0);
3951 memmove(skb_shinfo(skb)->frags,
3952 skb_shinfo(skb)->frags + 1,
3953 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3965 struct packet_offload *gro_find_receive_by_type(__be16 type)
3967 struct list_head *offload_head = &offload_base;
3968 struct packet_offload *ptype;
3970 list_for_each_entry_rcu(ptype, offload_head, list) {
3971 if (ptype->type != type || !ptype->callbacks.gro_receive)
3977 EXPORT_SYMBOL(gro_find_receive_by_type);
3979 struct packet_offload *gro_find_complete_by_type(__be16 type)
3981 struct list_head *offload_head = &offload_base;
3982 struct packet_offload *ptype;
3984 list_for_each_entry_rcu(ptype, offload_head, list) {
3985 if (ptype->type != type || !ptype->callbacks.gro_complete)
3991 EXPORT_SYMBOL(gro_find_complete_by_type);
3993 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3997 if (netif_receive_skb_internal(skb))
4005 case GRO_MERGED_FREE:
4006 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4007 kmem_cache_free(skbuff_head_cache, skb);
4020 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4022 trace_napi_gro_receive_entry(skb);
4024 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4026 EXPORT_SYMBOL(napi_gro_receive);
4028 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4030 __skb_pull(skb, skb_headlen(skb));
4031 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4032 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4034 skb->dev = napi->dev;
4040 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4042 struct sk_buff *skb = napi->skb;
4045 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4050 EXPORT_SYMBOL(napi_get_frags);
4052 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
4057 if (netif_receive_skb_internal(skb))
4062 case GRO_MERGED_FREE:
4063 napi_reuse_skb(napi, skb);
4074 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4076 struct sk_buff *skb = napi->skb;
4080 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) {
4081 napi_reuse_skb(napi, skb);
4084 skb->protocol = eth_type_trans(skb, skb->dev);
4089 gro_result_t napi_gro_frags(struct napi_struct *napi)
4091 struct sk_buff *skb = napi_frags_skb(napi);
4096 trace_napi_gro_frags_entry(skb);
4098 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4100 EXPORT_SYMBOL(napi_gro_frags);
4103 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4104 * Note: called with local irq disabled, but exits with local irq enabled.
4106 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4109 struct softnet_data *remsd = sd->rps_ipi_list;
4112 sd->rps_ipi_list = NULL;
4116 /* Send pending IPI's to kick RPS processing on remote cpus. */
4118 struct softnet_data *next = remsd->rps_ipi_next;
4120 if (cpu_online(remsd->cpu))
4121 __smp_call_function_single(remsd->cpu,
4130 static int process_backlog(struct napi_struct *napi, int quota)
4133 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4136 /* Check if we have pending ipi, its better to send them now,
4137 * not waiting net_rx_action() end.
4139 if (sd->rps_ipi_list) {
4140 local_irq_disable();
4141 net_rps_action_and_irq_enable(sd);
4144 napi->weight = weight_p;
4145 local_irq_disable();
4146 while (work < quota) {
4147 struct sk_buff *skb;
4150 while ((skb = __skb_dequeue(&sd->process_queue))) {
4152 __netif_receive_skb(skb);
4153 local_irq_disable();
4154 input_queue_head_incr(sd);
4155 if (++work >= quota) {
4162 qlen = skb_queue_len(&sd->input_pkt_queue);
4164 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4165 &sd->process_queue);
4167 if (qlen < quota - work) {
4169 * Inline a custom version of __napi_complete().
4170 * only current cpu owns and manipulates this napi,
4171 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4172 * we can use a plain write instead of clear_bit(),
4173 * and we dont need an smp_mb() memory barrier.
4175 list_del(&napi->poll_list);
4178 quota = work + qlen;
4188 * __napi_schedule - schedule for receive
4189 * @n: entry to schedule
4191 * The entry's receive function will be scheduled to run
4193 void __napi_schedule(struct napi_struct *n)
4195 unsigned long flags;
4197 local_irq_save(flags);
4198 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4199 local_irq_restore(flags);
4201 EXPORT_SYMBOL(__napi_schedule);
4203 void __napi_complete(struct napi_struct *n)
4205 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4206 BUG_ON(n->gro_list);
4208 list_del(&n->poll_list);
4209 smp_mb__before_clear_bit();
4210 clear_bit(NAPI_STATE_SCHED, &n->state);
4212 EXPORT_SYMBOL(__napi_complete);
4214 void napi_complete(struct napi_struct *n)
4216 unsigned long flags;
4219 * don't let napi dequeue from the cpu poll list
4220 * just in case its running on a different cpu
4222 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4225 napi_gro_flush(n, false);
4226 local_irq_save(flags);
4228 local_irq_restore(flags);
4230 EXPORT_SYMBOL(napi_complete);
4232 /* must be called under rcu_read_lock(), as we dont take a reference */
4233 struct napi_struct *napi_by_id(unsigned int napi_id)
4235 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4236 struct napi_struct *napi;
4238 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4239 if (napi->napi_id == napi_id)
4244 EXPORT_SYMBOL_GPL(napi_by_id);
4246 void napi_hash_add(struct napi_struct *napi)
4248 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4250 spin_lock(&napi_hash_lock);
4252 /* 0 is not a valid id, we also skip an id that is taken
4253 * we expect both events to be extremely rare
4256 while (!napi->napi_id) {
4257 napi->napi_id = ++napi_gen_id;
4258 if (napi_by_id(napi->napi_id))
4262 hlist_add_head_rcu(&napi->napi_hash_node,
4263 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4265 spin_unlock(&napi_hash_lock);
4268 EXPORT_SYMBOL_GPL(napi_hash_add);
4270 /* Warning : caller is responsible to make sure rcu grace period
4271 * is respected before freeing memory containing @napi
4273 void napi_hash_del(struct napi_struct *napi)
4275 spin_lock(&napi_hash_lock);
4277 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4278 hlist_del_rcu(&napi->napi_hash_node);
4280 spin_unlock(&napi_hash_lock);
4282 EXPORT_SYMBOL_GPL(napi_hash_del);
4284 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4285 int (*poll)(struct napi_struct *, int), int weight)
4287 INIT_LIST_HEAD(&napi->poll_list);
4288 napi->gro_count = 0;
4289 napi->gro_list = NULL;
4292 if (weight > NAPI_POLL_WEIGHT)
4293 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4295 napi->weight = weight;
4296 list_add(&napi->dev_list, &dev->napi_list);
4298 #ifdef CONFIG_NETPOLL
4299 spin_lock_init(&napi->poll_lock);
4300 napi->poll_owner = -1;
4302 set_bit(NAPI_STATE_SCHED, &napi->state);
4304 EXPORT_SYMBOL(netif_napi_add);
4306 void netif_napi_del(struct napi_struct *napi)
4308 list_del_init(&napi->dev_list);
4309 napi_free_frags(napi);
4311 kfree_skb_list(napi->gro_list);
4312 napi->gro_list = NULL;
4313 napi->gro_count = 0;
4315 EXPORT_SYMBOL(netif_napi_del);
4317 static void net_rx_action(struct softirq_action *h)
4319 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4320 unsigned long time_limit = jiffies + 2;
4321 int budget = netdev_budget;
4324 local_irq_disable();
4326 while (!list_empty(&sd->poll_list)) {
4327 struct napi_struct *n;
4330 /* If softirq window is exhuasted then punt.
4331 * Allow this to run for 2 jiffies since which will allow
4332 * an average latency of 1.5/HZ.
4334 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4339 /* Even though interrupts have been re-enabled, this
4340 * access is safe because interrupts can only add new
4341 * entries to the tail of this list, and only ->poll()
4342 * calls can remove this head entry from the list.
4344 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4346 have = netpoll_poll_lock(n);
4350 /* This NAPI_STATE_SCHED test is for avoiding a race
4351 * with netpoll's poll_napi(). Only the entity which
4352 * obtains the lock and sees NAPI_STATE_SCHED set will
4353 * actually make the ->poll() call. Therefore we avoid
4354 * accidentally calling ->poll() when NAPI is not scheduled.
4357 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4358 work = n->poll(n, weight);
4362 WARN_ON_ONCE(work > weight);
4366 local_irq_disable();
4368 /* Drivers must not modify the NAPI state if they
4369 * consume the entire weight. In such cases this code
4370 * still "owns" the NAPI instance and therefore can
4371 * move the instance around on the list at-will.
4373 if (unlikely(work == weight)) {
4374 if (unlikely(napi_disable_pending(n))) {
4377 local_irq_disable();
4380 /* flush too old packets
4381 * If HZ < 1000, flush all packets.
4384 napi_gro_flush(n, HZ >= 1000);
4385 local_irq_disable();
4387 list_move_tail(&n->poll_list, &sd->poll_list);
4391 netpoll_poll_unlock(have);
4394 net_rps_action_and_irq_enable(sd);
4396 #ifdef CONFIG_NET_DMA
4398 * There may not be any more sk_buffs coming right now, so push
4399 * any pending DMA copies to hardware
4401 dma_issue_pending_all();
4408 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4412 struct netdev_adjacent {
4413 struct net_device *dev;
4415 /* upper master flag, there can only be one master device per list */
4418 /* counter for the number of times this device was added to us */
4421 /* private field for the users */
4424 struct list_head list;
4425 struct rcu_head rcu;
4428 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4429 struct net_device *adj_dev,
4430 struct list_head *adj_list)
4432 struct netdev_adjacent *adj;
4434 list_for_each_entry(adj, adj_list, list) {
4435 if (adj->dev == adj_dev)
4442 * netdev_has_upper_dev - Check if device is linked to an upper device
4444 * @upper_dev: upper device to check
4446 * Find out if a device is linked to specified upper device and return true
4447 * in case it is. Note that this checks only immediate upper device,
4448 * not through a complete stack of devices. The caller must hold the RTNL lock.
4450 bool netdev_has_upper_dev(struct net_device *dev,
4451 struct net_device *upper_dev)
4455 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4457 EXPORT_SYMBOL(netdev_has_upper_dev);
4460 * netdev_has_any_upper_dev - Check if device is linked to some device
4463 * Find out if a device is linked to an upper device and return true in case
4464 * it is. The caller must hold the RTNL lock.
4466 static bool netdev_has_any_upper_dev(struct net_device *dev)
4470 return !list_empty(&dev->all_adj_list.upper);
4474 * netdev_master_upper_dev_get - Get master upper device
4477 * Find a master upper device and return pointer to it or NULL in case
4478 * it's not there. The caller must hold the RTNL lock.
4480 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4482 struct netdev_adjacent *upper;
4486 if (list_empty(&dev->adj_list.upper))
4489 upper = list_first_entry(&dev->adj_list.upper,
4490 struct netdev_adjacent, list);
4491 if (likely(upper->master))
4495 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4497 void *netdev_adjacent_get_private(struct list_head *adj_list)
4499 struct netdev_adjacent *adj;
4501 adj = list_entry(adj_list, struct netdev_adjacent, list);
4503 return adj->private;
4505 EXPORT_SYMBOL(netdev_adjacent_get_private);
4508 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4510 * @iter: list_head ** of the current position
4512 * Gets the next device from the dev's upper list, starting from iter
4513 * position. The caller must hold RCU read lock.
4515 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4516 struct list_head **iter)
4518 struct netdev_adjacent *upper;
4520 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4522 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4524 if (&upper->list == &dev->all_adj_list.upper)
4527 *iter = &upper->list;
4531 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4534 * netdev_lower_get_next_private - Get the next ->private from the
4535 * lower neighbour list
4537 * @iter: list_head ** of the current position
4539 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4540 * list, starting from iter position. The caller must hold either hold the
4541 * RTNL lock or its own locking that guarantees that the neighbour lower
4542 * list will remain unchainged.
4544 void *netdev_lower_get_next_private(struct net_device *dev,
4545 struct list_head **iter)
4547 struct netdev_adjacent *lower;
4549 lower = list_entry(*iter, struct netdev_adjacent, list);
4551 if (&lower->list == &dev->adj_list.lower)
4555 *iter = lower->list.next;
4557 return lower->private;
4559 EXPORT_SYMBOL(netdev_lower_get_next_private);
4562 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4563 * lower neighbour list, RCU
4566 * @iter: list_head ** of the current position
4568 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4569 * list, starting from iter position. The caller must hold RCU read lock.
4571 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4572 struct list_head **iter)
4574 struct netdev_adjacent *lower;
4576 WARN_ON_ONCE(!rcu_read_lock_held());
4578 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4580 if (&lower->list == &dev->adj_list.lower)
4584 *iter = &lower->list;
4586 return lower->private;
4588 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4591 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4592 * lower neighbour list, RCU
4596 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4597 * list. The caller must hold RCU read lock.
4599 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4601 struct netdev_adjacent *lower;
4603 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4604 struct netdev_adjacent, list);
4606 return lower->private;
4609 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4612 * netdev_master_upper_dev_get_rcu - Get master upper device
4615 * Find a master upper device and return pointer to it or NULL in case
4616 * it's not there. The caller must hold the RCU read lock.
4618 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4620 struct netdev_adjacent *upper;
4622 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4623 struct netdev_adjacent, list);
4624 if (upper && likely(upper->master))
4628 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4630 static int netdev_adjacent_sysfs_add(struct net_device *dev,
4631 struct net_device *adj_dev,
4632 struct list_head *dev_list)
4634 char linkname[IFNAMSIZ+7];
4635 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4636 "upper_%s" : "lower_%s", adj_dev->name);
4637 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4640 static void netdev_adjacent_sysfs_del(struct net_device *dev,
4642 struct list_head *dev_list)
4644 char linkname[IFNAMSIZ+7];
4645 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4646 "upper_%s" : "lower_%s", name);
4647 sysfs_remove_link(&(dev->dev.kobj), linkname);
4650 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4651 (dev_list == &dev->adj_list.upper || \
4652 dev_list == &dev->adj_list.lower)
4654 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4655 struct net_device *adj_dev,
4656 struct list_head *dev_list,
4657 void *private, bool master)
4659 struct netdev_adjacent *adj;
4662 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4669 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4674 adj->master = master;
4676 adj->private = private;
4679 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4680 adj_dev->name, dev->name, adj_dev->name);
4682 if (netdev_adjacent_is_neigh_list(dev, dev_list)) {
4683 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
4688 /* Ensure that master link is always the first item in list. */
4690 ret = sysfs_create_link(&(dev->dev.kobj),
4691 &(adj_dev->dev.kobj), "master");
4693 goto remove_symlinks;
4695 list_add_rcu(&adj->list, dev_list);
4697 list_add_tail_rcu(&adj->list, dev_list);
4703 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4704 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4712 static void __netdev_adjacent_dev_remove(struct net_device *dev,
4713 struct net_device *adj_dev,
4714 struct list_head *dev_list)
4716 struct netdev_adjacent *adj;
4718 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4721 pr_err("tried to remove device %s from %s\n",
4722 dev->name, adj_dev->name);
4726 if (adj->ref_nr > 1) {
4727 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4734 sysfs_remove_link(&(dev->dev.kobj), "master");
4736 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4737 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4739 list_del_rcu(&adj->list);
4740 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4741 adj_dev->name, dev->name, adj_dev->name);
4743 kfree_rcu(adj, rcu);
4746 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4747 struct net_device *upper_dev,
4748 struct list_head *up_list,
4749 struct list_head *down_list,
4750 void *private, bool master)
4754 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4759 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4762 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4769 static int __netdev_adjacent_dev_link(struct net_device *dev,
4770 struct net_device *upper_dev)
4772 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4773 &dev->all_adj_list.upper,
4774 &upper_dev->all_adj_list.lower,
4778 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4779 struct net_device *upper_dev,
4780 struct list_head *up_list,
4781 struct list_head *down_list)
4783 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4784 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4787 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
4788 struct net_device *upper_dev)
4790 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4791 &dev->all_adj_list.upper,
4792 &upper_dev->all_adj_list.lower);
4795 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4796 struct net_device *upper_dev,
4797 void *private, bool master)
4799 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4804 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4805 &dev->adj_list.upper,
4806 &upper_dev->adj_list.lower,
4809 __netdev_adjacent_dev_unlink(dev, upper_dev);
4816 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4817 struct net_device *upper_dev)
4819 __netdev_adjacent_dev_unlink(dev, upper_dev);
4820 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4821 &dev->adj_list.upper,
4822 &upper_dev->adj_list.lower);
4825 static int __netdev_upper_dev_link(struct net_device *dev,
4826 struct net_device *upper_dev, bool master,
4829 struct netdev_adjacent *i, *j, *to_i, *to_j;
4834 if (dev == upper_dev)
4837 /* To prevent loops, check if dev is not upper device to upper_dev. */
4838 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
4841 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
4844 if (master && netdev_master_upper_dev_get(dev))
4847 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
4852 /* Now that we linked these devs, make all the upper_dev's
4853 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4854 * versa, and don't forget the devices itself. All of these
4855 * links are non-neighbours.
4857 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4858 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4859 pr_debug("Interlinking %s with %s, non-neighbour\n",
4860 i->dev->name, j->dev->name);
4861 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
4867 /* add dev to every upper_dev's upper device */
4868 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4869 pr_debug("linking %s's upper device %s with %s\n",
4870 upper_dev->name, i->dev->name, dev->name);
4871 ret = __netdev_adjacent_dev_link(dev, i->dev);
4873 goto rollback_upper_mesh;
4876 /* add upper_dev to every dev's lower device */
4877 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4878 pr_debug("linking %s's lower device %s with %s\n", dev->name,
4879 i->dev->name, upper_dev->name);
4880 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
4882 goto rollback_lower_mesh;
4885 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4888 rollback_lower_mesh:
4890 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4893 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4898 rollback_upper_mesh:
4900 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4903 __netdev_adjacent_dev_unlink(dev, i->dev);
4911 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4912 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4913 if (i == to_i && j == to_j)
4915 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4921 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4927 * netdev_upper_dev_link - Add a link to the upper device
4929 * @upper_dev: new upper device
4931 * Adds a link to device which is upper to this one. The caller must hold
4932 * the RTNL lock. On a failure a negative errno code is returned.
4933 * On success the reference counts are adjusted and the function
4936 int netdev_upper_dev_link(struct net_device *dev,
4937 struct net_device *upper_dev)
4939 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
4941 EXPORT_SYMBOL(netdev_upper_dev_link);
4944 * netdev_master_upper_dev_link - Add a master link to the upper device
4946 * @upper_dev: new upper device
4948 * Adds a link to device which is upper to this one. In this case, only
4949 * one master upper device can be linked, although other non-master devices
4950 * might be linked as well. The caller must hold the RTNL lock.
4951 * On a failure a negative errno code is returned. On success the reference
4952 * counts are adjusted and the function returns zero.
4954 int netdev_master_upper_dev_link(struct net_device *dev,
4955 struct net_device *upper_dev)
4957 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
4959 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4961 int netdev_master_upper_dev_link_private(struct net_device *dev,
4962 struct net_device *upper_dev,
4965 return __netdev_upper_dev_link(dev, upper_dev, true, private);
4967 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
4970 * netdev_upper_dev_unlink - Removes a link to upper device
4972 * @upper_dev: new upper device
4974 * Removes a link to device which is upper to this one. The caller must hold
4977 void netdev_upper_dev_unlink(struct net_device *dev,
4978 struct net_device *upper_dev)
4980 struct netdev_adjacent *i, *j;
4983 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4985 /* Here is the tricky part. We must remove all dev's lower
4986 * devices from all upper_dev's upper devices and vice
4987 * versa, to maintain the graph relationship.
4989 list_for_each_entry(i, &dev->all_adj_list.lower, list)
4990 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
4991 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4993 /* remove also the devices itself from lower/upper device
4996 list_for_each_entry(i, &dev->all_adj_list.lower, list)
4997 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4999 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5000 __netdev_adjacent_dev_unlink(dev, i->dev);
5002 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5004 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5006 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5008 struct netdev_adjacent *iter;
5010 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5011 netdev_adjacent_sysfs_del(iter->dev, oldname,
5012 &iter->dev->adj_list.lower);
5013 netdev_adjacent_sysfs_add(iter->dev, dev,
5014 &iter->dev->adj_list.lower);
5017 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5018 netdev_adjacent_sysfs_del(iter->dev, oldname,
5019 &iter->dev->adj_list.upper);
5020 netdev_adjacent_sysfs_add(iter->dev, dev,
5021 &iter->dev->adj_list.upper);
5025 void *netdev_lower_dev_get_private(struct net_device *dev,
5026 struct net_device *lower_dev)
5028 struct netdev_adjacent *lower;
5032 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5036 return lower->private;
5038 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5040 static void dev_change_rx_flags(struct net_device *dev, int flags)
5042 const struct net_device_ops *ops = dev->netdev_ops;
5044 if (ops->ndo_change_rx_flags)
5045 ops->ndo_change_rx_flags(dev, flags);
5048 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5050 unsigned int old_flags = dev->flags;
5056 dev->flags |= IFF_PROMISC;
5057 dev->promiscuity += inc;
5058 if (dev->promiscuity == 0) {
5061 * If inc causes overflow, untouch promisc and return error.
5064 dev->flags &= ~IFF_PROMISC;
5066 dev->promiscuity -= inc;
5067 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5072 if (dev->flags != old_flags) {
5073 pr_info("device %s %s promiscuous mode\n",
5075 dev->flags & IFF_PROMISC ? "entered" : "left");
5076 if (audit_enabled) {
5077 current_uid_gid(&uid, &gid);
5078 audit_log(current->audit_context, GFP_ATOMIC,
5079 AUDIT_ANOM_PROMISCUOUS,
5080 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5081 dev->name, (dev->flags & IFF_PROMISC),
5082 (old_flags & IFF_PROMISC),
5083 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5084 from_kuid(&init_user_ns, uid),
5085 from_kgid(&init_user_ns, gid),
5086 audit_get_sessionid(current));
5089 dev_change_rx_flags(dev, IFF_PROMISC);
5092 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5097 * dev_set_promiscuity - update promiscuity count on a device
5101 * Add or remove promiscuity from a device. While the count in the device
5102 * remains above zero the interface remains promiscuous. Once it hits zero
5103 * the device reverts back to normal filtering operation. A negative inc
5104 * value is used to drop promiscuity on the device.
5105 * Return 0 if successful or a negative errno code on error.
5107 int dev_set_promiscuity(struct net_device *dev, int inc)
5109 unsigned int old_flags = dev->flags;
5112 err = __dev_set_promiscuity(dev, inc, true);
5115 if (dev->flags != old_flags)
5116 dev_set_rx_mode(dev);
5119 EXPORT_SYMBOL(dev_set_promiscuity);
5121 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5123 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5127 dev->flags |= IFF_ALLMULTI;
5128 dev->allmulti += inc;
5129 if (dev->allmulti == 0) {
5132 * If inc causes overflow, untouch allmulti and return error.
5135 dev->flags &= ~IFF_ALLMULTI;
5137 dev->allmulti -= inc;
5138 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5143 if (dev->flags ^ old_flags) {
5144 dev_change_rx_flags(dev, IFF_ALLMULTI);
5145 dev_set_rx_mode(dev);
5147 __dev_notify_flags(dev, old_flags,
5148 dev->gflags ^ old_gflags);
5154 * dev_set_allmulti - update allmulti count on a device
5158 * Add or remove reception of all multicast frames to a device. While the
5159 * count in the device remains above zero the interface remains listening
5160 * to all interfaces. Once it hits zero the device reverts back to normal
5161 * filtering operation. A negative @inc value is used to drop the counter
5162 * when releasing a resource needing all multicasts.
5163 * Return 0 if successful or a negative errno code on error.
5166 int dev_set_allmulti(struct net_device *dev, int inc)
5168 return __dev_set_allmulti(dev, inc, true);
5170 EXPORT_SYMBOL(dev_set_allmulti);
5173 * Upload unicast and multicast address lists to device and
5174 * configure RX filtering. When the device doesn't support unicast
5175 * filtering it is put in promiscuous mode while unicast addresses
5178 void __dev_set_rx_mode(struct net_device *dev)
5180 const struct net_device_ops *ops = dev->netdev_ops;
5182 /* dev_open will call this function so the list will stay sane. */
5183 if (!(dev->flags&IFF_UP))
5186 if (!netif_device_present(dev))
5189 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5190 /* Unicast addresses changes may only happen under the rtnl,
5191 * therefore calling __dev_set_promiscuity here is safe.
5193 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5194 __dev_set_promiscuity(dev, 1, false);
5195 dev->uc_promisc = true;
5196 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5197 __dev_set_promiscuity(dev, -1, false);
5198 dev->uc_promisc = false;
5202 if (ops->ndo_set_rx_mode)
5203 ops->ndo_set_rx_mode(dev);
5206 void dev_set_rx_mode(struct net_device *dev)
5208 netif_addr_lock_bh(dev);
5209 __dev_set_rx_mode(dev);
5210 netif_addr_unlock_bh(dev);
5214 * dev_get_flags - get flags reported to userspace
5217 * Get the combination of flag bits exported through APIs to userspace.
5219 unsigned int dev_get_flags(const struct net_device *dev)
5223 flags = (dev->flags & ~(IFF_PROMISC |
5228 (dev->gflags & (IFF_PROMISC |
5231 if (netif_running(dev)) {
5232 if (netif_oper_up(dev))
5233 flags |= IFF_RUNNING;
5234 if (netif_carrier_ok(dev))
5235 flags |= IFF_LOWER_UP;
5236 if (netif_dormant(dev))
5237 flags |= IFF_DORMANT;
5242 EXPORT_SYMBOL(dev_get_flags);
5244 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5246 unsigned int old_flags = dev->flags;
5252 * Set the flags on our device.
5255 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5256 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5258 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5262 * Load in the correct multicast list now the flags have changed.
5265 if ((old_flags ^ flags) & IFF_MULTICAST)
5266 dev_change_rx_flags(dev, IFF_MULTICAST);
5268 dev_set_rx_mode(dev);
5271 * Have we downed the interface. We handle IFF_UP ourselves
5272 * according to user attempts to set it, rather than blindly
5277 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
5278 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5281 dev_set_rx_mode(dev);
5284 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5285 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5286 unsigned int old_flags = dev->flags;
5288 dev->gflags ^= IFF_PROMISC;
5290 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5291 if (dev->flags != old_flags)
5292 dev_set_rx_mode(dev);
5295 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5296 is important. Some (broken) drivers set IFF_PROMISC, when
5297 IFF_ALLMULTI is requested not asking us and not reporting.
5299 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5300 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5302 dev->gflags ^= IFF_ALLMULTI;
5303 __dev_set_allmulti(dev, inc, false);
5309 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5310 unsigned int gchanges)
5312 unsigned int changes = dev->flags ^ old_flags;
5315 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5317 if (changes & IFF_UP) {
5318 if (dev->flags & IFF_UP)
5319 call_netdevice_notifiers(NETDEV_UP, dev);
5321 call_netdevice_notifiers(NETDEV_DOWN, dev);
5324 if (dev->flags & IFF_UP &&
5325 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5326 struct netdev_notifier_change_info change_info;
5328 change_info.flags_changed = changes;
5329 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5335 * dev_change_flags - change device settings
5337 * @flags: device state flags
5339 * Change settings on device based state flags. The flags are
5340 * in the userspace exported format.
5342 int dev_change_flags(struct net_device *dev, unsigned int flags)
5345 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5347 ret = __dev_change_flags(dev, flags);
5351 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5352 __dev_notify_flags(dev, old_flags, changes);
5355 EXPORT_SYMBOL(dev_change_flags);
5357 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5359 const struct net_device_ops *ops = dev->netdev_ops;
5361 if (ops->ndo_change_mtu)
5362 return ops->ndo_change_mtu(dev, new_mtu);
5369 * dev_set_mtu - Change maximum transfer unit
5371 * @new_mtu: new transfer unit
5373 * Change the maximum transfer size of the network device.
5375 int dev_set_mtu(struct net_device *dev, int new_mtu)
5379 if (new_mtu == dev->mtu)
5382 /* MTU must be positive. */
5386 if (!netif_device_present(dev))
5389 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5390 err = notifier_to_errno(err);
5394 orig_mtu = dev->mtu;
5395 err = __dev_set_mtu(dev, new_mtu);
5398 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5399 err = notifier_to_errno(err);
5401 /* setting mtu back and notifying everyone again,
5402 * so that they have a chance to revert changes.
5404 __dev_set_mtu(dev, orig_mtu);
5405 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5410 EXPORT_SYMBOL(dev_set_mtu);
5413 * dev_set_group - Change group this device belongs to
5415 * @new_group: group this device should belong to
5417 void dev_set_group(struct net_device *dev, int new_group)
5419 dev->group = new_group;
5421 EXPORT_SYMBOL(dev_set_group);
5424 * dev_set_mac_address - Change Media Access Control Address
5428 * Change the hardware (MAC) address of the device
5430 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5432 const struct net_device_ops *ops = dev->netdev_ops;
5435 if (!ops->ndo_set_mac_address)
5437 if (sa->sa_family != dev->type)
5439 if (!netif_device_present(dev))
5441 err = ops->ndo_set_mac_address(dev, sa);
5444 dev->addr_assign_type = NET_ADDR_SET;
5445 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5446 add_device_randomness(dev->dev_addr, dev->addr_len);
5449 EXPORT_SYMBOL(dev_set_mac_address);
5452 * dev_change_carrier - Change device carrier
5454 * @new_carrier: new value
5456 * Change device carrier
5458 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5460 const struct net_device_ops *ops = dev->netdev_ops;
5462 if (!ops->ndo_change_carrier)
5464 if (!netif_device_present(dev))
5466 return ops->ndo_change_carrier(dev, new_carrier);
5468 EXPORT_SYMBOL(dev_change_carrier);
5471 * dev_get_phys_port_id - Get device physical port ID
5475 * Get device physical port ID
5477 int dev_get_phys_port_id(struct net_device *dev,
5478 struct netdev_phys_port_id *ppid)
5480 const struct net_device_ops *ops = dev->netdev_ops;
5482 if (!ops->ndo_get_phys_port_id)
5484 return ops->ndo_get_phys_port_id(dev, ppid);
5486 EXPORT_SYMBOL(dev_get_phys_port_id);
5489 * dev_new_index - allocate an ifindex
5490 * @net: the applicable net namespace
5492 * Returns a suitable unique value for a new device interface
5493 * number. The caller must hold the rtnl semaphore or the
5494 * dev_base_lock to be sure it remains unique.
5496 static int dev_new_index(struct net *net)
5498 int ifindex = net->ifindex;
5502 if (!__dev_get_by_index(net, ifindex))
5503 return net->ifindex = ifindex;
5507 /* Delayed registration/unregisteration */
5508 static LIST_HEAD(net_todo_list);
5509 static DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5511 static void net_set_todo(struct net_device *dev)
5513 list_add_tail(&dev->todo_list, &net_todo_list);
5514 dev_net(dev)->dev_unreg_count++;
5517 static void rollback_registered_many(struct list_head *head)
5519 struct net_device *dev, *tmp;
5520 LIST_HEAD(close_head);
5522 BUG_ON(dev_boot_phase);
5525 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5526 /* Some devices call without registering
5527 * for initialization unwind. Remove those
5528 * devices and proceed with the remaining.
5530 if (dev->reg_state == NETREG_UNINITIALIZED) {
5531 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5535 list_del(&dev->unreg_list);
5538 dev->dismantle = true;
5539 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5542 /* If device is running, close it first. */
5543 list_for_each_entry(dev, head, unreg_list)
5544 list_add_tail(&dev->close_list, &close_head);
5545 dev_close_many(&close_head);
5547 list_for_each_entry(dev, head, unreg_list) {
5548 /* And unlink it from device chain. */
5549 unlist_netdevice(dev);
5551 dev->reg_state = NETREG_UNREGISTERING;
5556 list_for_each_entry(dev, head, unreg_list) {
5557 /* Shutdown queueing discipline. */
5561 /* Notify protocols, that we are about to destroy
5562 this device. They should clean all the things.
5564 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5566 if (!dev->rtnl_link_ops ||
5567 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5568 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5571 * Flush the unicast and multicast chains
5576 if (dev->netdev_ops->ndo_uninit)
5577 dev->netdev_ops->ndo_uninit(dev);
5579 /* Notifier chain MUST detach us all upper devices. */
5580 WARN_ON(netdev_has_any_upper_dev(dev));
5582 /* Remove entries from kobject tree */
5583 netdev_unregister_kobject(dev);
5585 /* Remove XPS queueing entries */
5586 netif_reset_xps_queues_gt(dev, 0);
5592 list_for_each_entry(dev, head, unreg_list)
5596 static void rollback_registered(struct net_device *dev)
5600 list_add(&dev->unreg_list, &single);
5601 rollback_registered_many(&single);
5605 static netdev_features_t netdev_fix_features(struct net_device *dev,
5606 netdev_features_t features)
5608 /* Fix illegal checksum combinations */
5609 if ((features & NETIF_F_HW_CSUM) &&
5610 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5611 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5612 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5615 /* TSO requires that SG is present as well. */
5616 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5617 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5618 features &= ~NETIF_F_ALL_TSO;
5621 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5622 !(features & NETIF_F_IP_CSUM)) {
5623 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5624 features &= ~NETIF_F_TSO;
5625 features &= ~NETIF_F_TSO_ECN;
5628 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5629 !(features & NETIF_F_IPV6_CSUM)) {
5630 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5631 features &= ~NETIF_F_TSO6;
5634 /* TSO ECN requires that TSO is present as well. */
5635 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5636 features &= ~NETIF_F_TSO_ECN;
5638 /* Software GSO depends on SG. */
5639 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5640 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5641 features &= ~NETIF_F_GSO;
5644 /* UFO needs SG and checksumming */
5645 if (features & NETIF_F_UFO) {
5646 /* maybe split UFO into V4 and V6? */
5647 if (!((features & NETIF_F_GEN_CSUM) ||
5648 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5649 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5651 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5652 features &= ~NETIF_F_UFO;
5655 if (!(features & NETIF_F_SG)) {
5657 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5658 features &= ~NETIF_F_UFO;
5665 int __netdev_update_features(struct net_device *dev)
5667 netdev_features_t features;
5672 features = netdev_get_wanted_features(dev);
5674 if (dev->netdev_ops->ndo_fix_features)
5675 features = dev->netdev_ops->ndo_fix_features(dev, features);
5677 /* driver might be less strict about feature dependencies */
5678 features = netdev_fix_features(dev, features);
5680 if (dev->features == features)
5683 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5684 &dev->features, &features);
5686 if (dev->netdev_ops->ndo_set_features)
5687 err = dev->netdev_ops->ndo_set_features(dev, features);
5689 if (unlikely(err < 0)) {
5691 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5692 err, &features, &dev->features);
5697 dev->features = features;
5703 * netdev_update_features - recalculate device features
5704 * @dev: the device to check
5706 * Recalculate dev->features set and send notifications if it
5707 * has changed. Should be called after driver or hardware dependent
5708 * conditions might have changed that influence the features.
5710 void netdev_update_features(struct net_device *dev)
5712 if (__netdev_update_features(dev))
5713 netdev_features_change(dev);
5715 EXPORT_SYMBOL(netdev_update_features);
5718 * netdev_change_features - recalculate device features
5719 * @dev: the device to check
5721 * Recalculate dev->features set and send notifications even
5722 * if they have not changed. Should be called instead of
5723 * netdev_update_features() if also dev->vlan_features might
5724 * have changed to allow the changes to be propagated to stacked
5727 void netdev_change_features(struct net_device *dev)
5729 __netdev_update_features(dev);
5730 netdev_features_change(dev);
5732 EXPORT_SYMBOL(netdev_change_features);
5735 * netif_stacked_transfer_operstate - transfer operstate
5736 * @rootdev: the root or lower level device to transfer state from
5737 * @dev: the device to transfer operstate to
5739 * Transfer operational state from root to device. This is normally
5740 * called when a stacking relationship exists between the root
5741 * device and the device(a leaf device).
5743 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5744 struct net_device *dev)
5746 if (rootdev->operstate == IF_OPER_DORMANT)
5747 netif_dormant_on(dev);
5749 netif_dormant_off(dev);
5751 if (netif_carrier_ok(rootdev)) {
5752 if (!netif_carrier_ok(dev))
5753 netif_carrier_on(dev);
5755 if (netif_carrier_ok(dev))
5756 netif_carrier_off(dev);
5759 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5762 static int netif_alloc_rx_queues(struct net_device *dev)
5764 unsigned int i, count = dev->num_rx_queues;
5765 struct netdev_rx_queue *rx;
5769 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5775 for (i = 0; i < count; i++)
5781 static void netdev_init_one_queue(struct net_device *dev,
5782 struct netdev_queue *queue, void *_unused)
5784 /* Initialize queue lock */
5785 spin_lock_init(&queue->_xmit_lock);
5786 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5787 queue->xmit_lock_owner = -1;
5788 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5791 dql_init(&queue->dql, HZ);
5795 static void netif_free_tx_queues(struct net_device *dev)
5797 if (is_vmalloc_addr(dev->_tx))
5803 static int netif_alloc_netdev_queues(struct net_device *dev)
5805 unsigned int count = dev->num_tx_queues;
5806 struct netdev_queue *tx;
5807 size_t sz = count * sizeof(*tx);
5809 BUG_ON(count < 1 || count > 0xffff);
5811 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
5819 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5820 spin_lock_init(&dev->tx_global_lock);
5826 * register_netdevice - register a network device
5827 * @dev: device to register
5829 * Take a completed network device structure and add it to the kernel
5830 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5831 * chain. 0 is returned on success. A negative errno code is returned
5832 * on a failure to set up the device, or if the name is a duplicate.
5834 * Callers must hold the rtnl semaphore. You may want
5835 * register_netdev() instead of this.
5838 * The locking appears insufficient to guarantee two parallel registers
5839 * will not get the same name.
5842 int register_netdevice(struct net_device *dev)
5845 struct net *net = dev_net(dev);
5847 BUG_ON(dev_boot_phase);
5852 /* When net_device's are persistent, this will be fatal. */
5853 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5856 spin_lock_init(&dev->addr_list_lock);
5857 netdev_set_addr_lockdep_class(dev);
5861 ret = dev_get_valid_name(net, dev, dev->name);
5865 /* Init, if this function is available */
5866 if (dev->netdev_ops->ndo_init) {
5867 ret = dev->netdev_ops->ndo_init(dev);
5875 if (((dev->hw_features | dev->features) &
5876 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5877 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5878 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5879 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5886 dev->ifindex = dev_new_index(net);
5887 else if (__dev_get_by_index(net, dev->ifindex))
5890 if (dev->iflink == -1)
5891 dev->iflink = dev->ifindex;
5893 /* Transfer changeable features to wanted_features and enable
5894 * software offloads (GSO and GRO).
5896 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5897 dev->features |= NETIF_F_SOFT_FEATURES;
5898 dev->wanted_features = dev->features & dev->hw_features;
5900 if (!(dev->flags & IFF_LOOPBACK)) {
5901 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5904 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5906 dev->vlan_features |= NETIF_F_HIGHDMA;
5908 /* Make NETIF_F_SG inheritable to tunnel devices.
5910 dev->hw_enc_features |= NETIF_F_SG;
5912 /* Make NETIF_F_SG inheritable to MPLS.
5914 dev->mpls_features |= NETIF_F_SG;
5916 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5917 ret = notifier_to_errno(ret);
5921 ret = netdev_register_kobject(dev);
5924 dev->reg_state = NETREG_REGISTERED;
5926 __netdev_update_features(dev);
5929 * Default initial state at registry is that the
5930 * device is present.
5933 set_bit(__LINK_STATE_PRESENT, &dev->state);
5935 linkwatch_init_dev(dev);
5937 dev_init_scheduler(dev);
5939 list_netdevice(dev);
5940 add_device_randomness(dev->dev_addr, dev->addr_len);
5942 /* If the device has permanent device address, driver should
5943 * set dev_addr and also addr_assign_type should be set to
5944 * NET_ADDR_PERM (default value).
5946 if (dev->addr_assign_type == NET_ADDR_PERM)
5947 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5949 /* Notify protocols, that a new device appeared. */
5950 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5951 ret = notifier_to_errno(ret);
5953 rollback_registered(dev);
5954 dev->reg_state = NETREG_UNREGISTERED;
5957 * Prevent userspace races by waiting until the network
5958 * device is fully setup before sending notifications.
5960 if (!dev->rtnl_link_ops ||
5961 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5962 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
5968 if (dev->netdev_ops->ndo_uninit)
5969 dev->netdev_ops->ndo_uninit(dev);
5972 EXPORT_SYMBOL(register_netdevice);
5975 * init_dummy_netdev - init a dummy network device for NAPI
5976 * @dev: device to init
5978 * This takes a network device structure and initialize the minimum
5979 * amount of fields so it can be used to schedule NAPI polls without
5980 * registering a full blown interface. This is to be used by drivers
5981 * that need to tie several hardware interfaces to a single NAPI
5982 * poll scheduler due to HW limitations.
5984 int init_dummy_netdev(struct net_device *dev)
5986 /* Clear everything. Note we don't initialize spinlocks
5987 * are they aren't supposed to be taken by any of the
5988 * NAPI code and this dummy netdev is supposed to be
5989 * only ever used for NAPI polls
5991 memset(dev, 0, sizeof(struct net_device));
5993 /* make sure we BUG if trying to hit standard
5994 * register/unregister code path
5996 dev->reg_state = NETREG_DUMMY;
5998 /* NAPI wants this */
5999 INIT_LIST_HEAD(&dev->napi_list);
6001 /* a dummy interface is started by default */
6002 set_bit(__LINK_STATE_PRESENT, &dev->state);
6003 set_bit(__LINK_STATE_START, &dev->state);
6005 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6006 * because users of this 'device' dont need to change
6012 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6016 * register_netdev - register a network device
6017 * @dev: device to register
6019 * Take a completed network device structure and add it to the kernel
6020 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6021 * chain. 0 is returned on success. A negative errno code is returned
6022 * on a failure to set up the device, or if the name is a duplicate.
6024 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6025 * and expands the device name if you passed a format string to
6028 int register_netdev(struct net_device *dev)
6033 err = register_netdevice(dev);
6037 EXPORT_SYMBOL(register_netdev);
6039 int netdev_refcnt_read(const struct net_device *dev)
6043 for_each_possible_cpu(i)
6044 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6047 EXPORT_SYMBOL(netdev_refcnt_read);
6050 * netdev_wait_allrefs - wait until all references are gone.
6051 * @dev: target net_device
6053 * This is called when unregistering network devices.
6055 * Any protocol or device that holds a reference should register
6056 * for netdevice notification, and cleanup and put back the
6057 * reference if they receive an UNREGISTER event.
6058 * We can get stuck here if buggy protocols don't correctly
6061 static void netdev_wait_allrefs(struct net_device *dev)
6063 unsigned long rebroadcast_time, warning_time;
6066 linkwatch_forget_dev(dev);
6068 rebroadcast_time = warning_time = jiffies;
6069 refcnt = netdev_refcnt_read(dev);
6071 while (refcnt != 0) {
6072 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6075 /* Rebroadcast unregister notification */
6076 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6082 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6083 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6085 /* We must not have linkwatch events
6086 * pending on unregister. If this
6087 * happens, we simply run the queue
6088 * unscheduled, resulting in a noop
6091 linkwatch_run_queue();
6096 rebroadcast_time = jiffies;
6101 refcnt = netdev_refcnt_read(dev);
6103 if (time_after(jiffies, warning_time + 10 * HZ)) {
6104 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6106 warning_time = jiffies;
6115 * register_netdevice(x1);
6116 * register_netdevice(x2);
6118 * unregister_netdevice(y1);
6119 * unregister_netdevice(y2);
6125 * We are invoked by rtnl_unlock().
6126 * This allows us to deal with problems:
6127 * 1) We can delete sysfs objects which invoke hotplug
6128 * without deadlocking with linkwatch via keventd.
6129 * 2) Since we run with the RTNL semaphore not held, we can sleep
6130 * safely in order to wait for the netdev refcnt to drop to zero.
6132 * We must not return until all unregister events added during
6133 * the interval the lock was held have been completed.
6135 void netdev_run_todo(void)
6137 struct list_head list;
6139 /* Snapshot list, allow later requests */
6140 list_replace_init(&net_todo_list, &list);
6145 /* Wait for rcu callbacks to finish before next phase */
6146 if (!list_empty(&list))
6149 while (!list_empty(&list)) {
6150 struct net_device *dev
6151 = list_first_entry(&list, struct net_device, todo_list);
6152 list_del(&dev->todo_list);
6155 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6158 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6159 pr_err("network todo '%s' but state %d\n",
6160 dev->name, dev->reg_state);
6165 dev->reg_state = NETREG_UNREGISTERED;
6167 on_each_cpu(flush_backlog, dev, 1);
6169 netdev_wait_allrefs(dev);
6172 BUG_ON(netdev_refcnt_read(dev));
6173 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6174 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6175 WARN_ON(dev->dn_ptr);
6177 if (dev->destructor)
6178 dev->destructor(dev);
6180 /* Report a network device has been unregistered */
6182 dev_net(dev)->dev_unreg_count--;
6184 wake_up(&netdev_unregistering_wq);
6186 /* Free network device */
6187 kobject_put(&dev->dev.kobj);
6191 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6192 * fields in the same order, with only the type differing.
6194 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6195 const struct net_device_stats *netdev_stats)
6197 #if BITS_PER_LONG == 64
6198 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6199 memcpy(stats64, netdev_stats, sizeof(*stats64));
6201 size_t i, n = sizeof(*stats64) / sizeof(u64);
6202 const unsigned long *src = (const unsigned long *)netdev_stats;
6203 u64 *dst = (u64 *)stats64;
6205 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6206 sizeof(*stats64) / sizeof(u64));
6207 for (i = 0; i < n; i++)
6211 EXPORT_SYMBOL(netdev_stats_to_stats64);
6214 * dev_get_stats - get network device statistics
6215 * @dev: device to get statistics from
6216 * @storage: place to store stats
6218 * Get network statistics from device. Return @storage.
6219 * The device driver may provide its own method by setting
6220 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6221 * otherwise the internal statistics structure is used.
6223 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6224 struct rtnl_link_stats64 *storage)
6226 const struct net_device_ops *ops = dev->netdev_ops;
6228 if (ops->ndo_get_stats64) {
6229 memset(storage, 0, sizeof(*storage));
6230 ops->ndo_get_stats64(dev, storage);
6231 } else if (ops->ndo_get_stats) {
6232 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6234 netdev_stats_to_stats64(storage, &dev->stats);
6236 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6237 storage->tx_dropped += atomic_long_read(&dev->tx_dropped);
6240 EXPORT_SYMBOL(dev_get_stats);
6242 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6244 struct netdev_queue *queue = dev_ingress_queue(dev);
6246 #ifdef CONFIG_NET_CLS_ACT
6249 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6252 netdev_init_one_queue(dev, queue, NULL);
6253 queue->qdisc = &noop_qdisc;
6254 queue->qdisc_sleeping = &noop_qdisc;
6255 rcu_assign_pointer(dev->ingress_queue, queue);
6260 static const struct ethtool_ops default_ethtool_ops;
6262 void netdev_set_default_ethtool_ops(struct net_device *dev,
6263 const struct ethtool_ops *ops)
6265 if (dev->ethtool_ops == &default_ethtool_ops)
6266 dev->ethtool_ops = ops;
6268 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6270 void netdev_freemem(struct net_device *dev)
6272 char *addr = (char *)dev - dev->padded;
6274 if (is_vmalloc_addr(addr))
6281 * alloc_netdev_mqs - allocate network device
6282 * @sizeof_priv: size of private data to allocate space for
6283 * @name: device name format string
6284 * @setup: callback to initialize device
6285 * @txqs: the number of TX subqueues to allocate
6286 * @rxqs: the number of RX subqueues to allocate
6288 * Allocates a struct net_device with private data area for driver use
6289 * and performs basic initialization. Also allocates subqueue structs
6290 * for each queue on the device.
6292 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6293 void (*setup)(struct net_device *),
6294 unsigned int txqs, unsigned int rxqs)
6296 struct net_device *dev;
6298 struct net_device *p;
6300 BUG_ON(strlen(name) >= sizeof(dev->name));
6303 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6309 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6314 alloc_size = sizeof(struct net_device);
6316 /* ensure 32-byte alignment of private area */
6317 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6318 alloc_size += sizeof_priv;
6320 /* ensure 32-byte alignment of whole construct */
6321 alloc_size += NETDEV_ALIGN - 1;
6323 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6325 p = vzalloc(alloc_size);
6329 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6330 dev->padded = (char *)dev - (char *)p;
6332 dev->pcpu_refcnt = alloc_percpu(int);
6333 if (!dev->pcpu_refcnt)
6336 if (dev_addr_init(dev))
6342 dev_net_set(dev, &init_net);
6344 dev->gso_max_size = GSO_MAX_SIZE;
6345 dev->gso_max_segs = GSO_MAX_SEGS;
6347 INIT_LIST_HEAD(&dev->napi_list);
6348 INIT_LIST_HEAD(&dev->unreg_list);
6349 INIT_LIST_HEAD(&dev->close_list);
6350 INIT_LIST_HEAD(&dev->link_watch_list);
6351 INIT_LIST_HEAD(&dev->adj_list.upper);
6352 INIT_LIST_HEAD(&dev->adj_list.lower);
6353 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6354 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6355 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6358 dev->num_tx_queues = txqs;
6359 dev->real_num_tx_queues = txqs;
6360 if (netif_alloc_netdev_queues(dev))
6364 dev->num_rx_queues = rxqs;
6365 dev->real_num_rx_queues = rxqs;
6366 if (netif_alloc_rx_queues(dev))
6370 strcpy(dev->name, name);
6371 dev->group = INIT_NETDEV_GROUP;
6372 if (!dev->ethtool_ops)
6373 dev->ethtool_ops = &default_ethtool_ops;
6381 free_percpu(dev->pcpu_refcnt);
6382 netif_free_tx_queues(dev);
6388 netdev_freemem(dev);
6391 EXPORT_SYMBOL(alloc_netdev_mqs);
6394 * free_netdev - free network device
6397 * This function does the last stage of destroying an allocated device
6398 * interface. The reference to the device object is released.
6399 * If this is the last reference then it will be freed.
6401 void free_netdev(struct net_device *dev)
6403 struct napi_struct *p, *n;
6405 release_net(dev_net(dev));
6407 netif_free_tx_queues(dev);
6412 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6414 /* Flush device addresses */
6415 dev_addr_flush(dev);
6417 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6420 free_percpu(dev->pcpu_refcnt);
6421 dev->pcpu_refcnt = NULL;
6423 /* Compatibility with error handling in drivers */
6424 if (dev->reg_state == NETREG_UNINITIALIZED) {
6425 netdev_freemem(dev);
6429 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6430 dev->reg_state = NETREG_RELEASED;
6432 /* will free via device release */
6433 put_device(&dev->dev);
6435 EXPORT_SYMBOL(free_netdev);
6438 * synchronize_net - Synchronize with packet receive processing
6440 * Wait for packets currently being received to be done.
6441 * Does not block later packets from starting.
6443 void synchronize_net(void)
6446 if (rtnl_is_locked())
6447 synchronize_rcu_expedited();
6451 EXPORT_SYMBOL(synchronize_net);
6454 * unregister_netdevice_queue - remove device from the kernel
6458 * This function shuts down a device interface and removes it
6459 * from the kernel tables.
6460 * If head not NULL, device is queued to be unregistered later.
6462 * Callers must hold the rtnl semaphore. You may want
6463 * unregister_netdev() instead of this.
6466 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6471 list_move_tail(&dev->unreg_list, head);
6473 rollback_registered(dev);
6474 /* Finish processing unregister after unlock */
6478 EXPORT_SYMBOL(unregister_netdevice_queue);
6481 * unregister_netdevice_many - unregister many devices
6482 * @head: list of devices
6484 void unregister_netdevice_many(struct list_head *head)
6486 struct net_device *dev;
6488 if (!list_empty(head)) {
6489 rollback_registered_many(head);
6490 list_for_each_entry(dev, head, unreg_list)
6494 EXPORT_SYMBOL(unregister_netdevice_many);
6497 * unregister_netdev - remove device from the kernel
6500 * This function shuts down a device interface and removes it
6501 * from the kernel tables.
6503 * This is just a wrapper for unregister_netdevice that takes
6504 * the rtnl semaphore. In general you want to use this and not
6505 * unregister_netdevice.
6507 void unregister_netdev(struct net_device *dev)
6510 unregister_netdevice(dev);
6513 EXPORT_SYMBOL(unregister_netdev);
6516 * dev_change_net_namespace - move device to different nethost namespace
6518 * @net: network namespace
6519 * @pat: If not NULL name pattern to try if the current device name
6520 * is already taken in the destination network namespace.
6522 * This function shuts down a device interface and moves it
6523 * to a new network namespace. On success 0 is returned, on
6524 * a failure a netagive errno code is returned.
6526 * Callers must hold the rtnl semaphore.
6529 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6535 /* Don't allow namespace local devices to be moved. */
6537 if (dev->features & NETIF_F_NETNS_LOCAL)
6540 /* Ensure the device has been registrered */
6541 if (dev->reg_state != NETREG_REGISTERED)
6544 /* Get out if there is nothing todo */
6546 if (net_eq(dev_net(dev), net))
6549 /* Pick the destination device name, and ensure
6550 * we can use it in the destination network namespace.
6553 if (__dev_get_by_name(net, dev->name)) {
6554 /* We get here if we can't use the current device name */
6557 if (dev_get_valid_name(net, dev, pat) < 0)
6562 * And now a mini version of register_netdevice unregister_netdevice.
6565 /* If device is running close it first. */
6568 /* And unlink it from device chain */
6570 unlist_netdevice(dev);
6574 /* Shutdown queueing discipline. */
6577 /* Notify protocols, that we are about to destroy
6578 this device. They should clean all the things.
6580 Note that dev->reg_state stays at NETREG_REGISTERED.
6581 This is wanted because this way 8021q and macvlan know
6582 the device is just moving and can keep their slaves up.
6584 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6586 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6587 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6590 * Flush the unicast and multicast chains
6595 /* Send a netdev-removed uevent to the old namespace */
6596 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6598 /* Actually switch the network namespace */
6599 dev_net_set(dev, net);
6601 /* If there is an ifindex conflict assign a new one */
6602 if (__dev_get_by_index(net, dev->ifindex)) {
6603 int iflink = (dev->iflink == dev->ifindex);
6604 dev->ifindex = dev_new_index(net);
6606 dev->iflink = dev->ifindex;
6609 /* Send a netdev-add uevent to the new namespace */
6610 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6612 /* Fixup kobjects */
6613 err = device_rename(&dev->dev, dev->name);
6616 /* Add the device back in the hashes */
6617 list_netdevice(dev);
6619 /* Notify protocols, that a new device appeared. */
6620 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6623 * Prevent userspace races by waiting until the network
6624 * device is fully setup before sending notifications.
6626 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6633 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6635 static int dev_cpu_callback(struct notifier_block *nfb,
6636 unsigned long action,
6639 struct sk_buff **list_skb;
6640 struct sk_buff *skb;
6641 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6642 struct softnet_data *sd, *oldsd;
6644 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6647 local_irq_disable();
6648 cpu = smp_processor_id();
6649 sd = &per_cpu(softnet_data, cpu);
6650 oldsd = &per_cpu(softnet_data, oldcpu);
6652 /* Find end of our completion_queue. */
6653 list_skb = &sd->completion_queue;
6655 list_skb = &(*list_skb)->next;
6656 /* Append completion queue from offline CPU. */
6657 *list_skb = oldsd->completion_queue;
6658 oldsd->completion_queue = NULL;
6660 /* Append output queue from offline CPU. */
6661 if (oldsd->output_queue) {
6662 *sd->output_queue_tailp = oldsd->output_queue;
6663 sd->output_queue_tailp = oldsd->output_queue_tailp;
6664 oldsd->output_queue = NULL;
6665 oldsd->output_queue_tailp = &oldsd->output_queue;
6667 /* Append NAPI poll list from offline CPU. */
6668 if (!list_empty(&oldsd->poll_list)) {
6669 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6670 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6673 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6676 /* Process offline CPU's input_pkt_queue */
6677 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6678 netif_rx_internal(skb);
6679 input_queue_head_incr(oldsd);
6681 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6682 netif_rx_internal(skb);
6683 input_queue_head_incr(oldsd);
6691 * netdev_increment_features - increment feature set by one
6692 * @all: current feature set
6693 * @one: new feature set
6694 * @mask: mask feature set
6696 * Computes a new feature set after adding a device with feature set
6697 * @one to the master device with current feature set @all. Will not
6698 * enable anything that is off in @mask. Returns the new feature set.
6700 netdev_features_t netdev_increment_features(netdev_features_t all,
6701 netdev_features_t one, netdev_features_t mask)
6703 if (mask & NETIF_F_GEN_CSUM)
6704 mask |= NETIF_F_ALL_CSUM;
6705 mask |= NETIF_F_VLAN_CHALLENGED;
6707 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6708 all &= one | ~NETIF_F_ALL_FOR_ALL;
6710 /* If one device supports hw checksumming, set for all. */
6711 if (all & NETIF_F_GEN_CSUM)
6712 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6716 EXPORT_SYMBOL(netdev_increment_features);
6718 static struct hlist_head * __net_init netdev_create_hash(void)
6721 struct hlist_head *hash;
6723 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6725 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6726 INIT_HLIST_HEAD(&hash[i]);
6731 /* Initialize per network namespace state */
6732 static int __net_init netdev_init(struct net *net)
6734 if (net != &init_net)
6735 INIT_LIST_HEAD(&net->dev_base_head);
6737 net->dev_name_head = netdev_create_hash();
6738 if (net->dev_name_head == NULL)
6741 net->dev_index_head = netdev_create_hash();
6742 if (net->dev_index_head == NULL)
6748 kfree(net->dev_name_head);
6754 * netdev_drivername - network driver for the device
6755 * @dev: network device
6757 * Determine network driver for device.
6759 const char *netdev_drivername(const struct net_device *dev)
6761 const struct device_driver *driver;
6762 const struct device *parent;
6763 const char *empty = "";
6765 parent = dev->dev.parent;
6769 driver = parent->driver;
6770 if (driver && driver->name)
6771 return driver->name;
6775 static int __netdev_printk(const char *level, const struct net_device *dev,
6776 struct va_format *vaf)
6780 if (dev && dev->dev.parent) {
6781 r = dev_printk_emit(level[1] - '0',
6784 dev_driver_string(dev->dev.parent),
6785 dev_name(dev->dev.parent),
6786 netdev_name(dev), vaf);
6788 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6790 r = printk("%s(NULL net_device): %pV", level, vaf);
6796 int netdev_printk(const char *level, const struct net_device *dev,
6797 const char *format, ...)
6799 struct va_format vaf;
6803 va_start(args, format);
6808 r = __netdev_printk(level, dev, &vaf);
6814 EXPORT_SYMBOL(netdev_printk);
6816 #define define_netdev_printk_level(func, level) \
6817 int func(const struct net_device *dev, const char *fmt, ...) \
6820 struct va_format vaf; \
6823 va_start(args, fmt); \
6828 r = __netdev_printk(level, dev, &vaf); \
6834 EXPORT_SYMBOL(func);
6836 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6837 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6838 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6839 define_netdev_printk_level(netdev_err, KERN_ERR);
6840 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6841 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6842 define_netdev_printk_level(netdev_info, KERN_INFO);
6844 static void __net_exit netdev_exit(struct net *net)
6846 kfree(net->dev_name_head);
6847 kfree(net->dev_index_head);
6850 static struct pernet_operations __net_initdata netdev_net_ops = {
6851 .init = netdev_init,
6852 .exit = netdev_exit,
6855 static void __net_exit default_device_exit(struct net *net)
6857 struct net_device *dev, *aux;
6859 * Push all migratable network devices back to the
6860 * initial network namespace
6863 for_each_netdev_safe(net, dev, aux) {
6865 char fb_name[IFNAMSIZ];
6867 /* Ignore unmoveable devices (i.e. loopback) */
6868 if (dev->features & NETIF_F_NETNS_LOCAL)
6871 /* Leave virtual devices for the generic cleanup */
6872 if (dev->rtnl_link_ops)
6875 /* Push remaining network devices to init_net */
6876 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6877 err = dev_change_net_namespace(dev, &init_net, fb_name);
6879 pr_emerg("%s: failed to move %s to init_net: %d\n",
6880 __func__, dev->name, err);
6887 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
6889 /* Return with the rtnl_lock held when there are no network
6890 * devices unregistering in any network namespace in net_list.
6897 prepare_to_wait(&netdev_unregistering_wq, &wait,
6898 TASK_UNINTERRUPTIBLE);
6899 unregistering = false;
6901 list_for_each_entry(net, net_list, exit_list) {
6902 if (net->dev_unreg_count > 0) {
6903 unregistering = true;
6912 finish_wait(&netdev_unregistering_wq, &wait);
6915 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6917 /* At exit all network devices most be removed from a network
6918 * namespace. Do this in the reverse order of registration.
6919 * Do this across as many network namespaces as possible to
6920 * improve batching efficiency.
6922 struct net_device *dev;
6924 LIST_HEAD(dev_kill_list);
6926 /* To prevent network device cleanup code from dereferencing
6927 * loopback devices or network devices that have been freed
6928 * wait here for all pending unregistrations to complete,
6929 * before unregistring the loopback device and allowing the
6930 * network namespace be freed.
6932 * The netdev todo list containing all network devices
6933 * unregistrations that happen in default_device_exit_batch
6934 * will run in the rtnl_unlock() at the end of
6935 * default_device_exit_batch.
6937 rtnl_lock_unregistering(net_list);
6938 list_for_each_entry(net, net_list, exit_list) {
6939 for_each_netdev_reverse(net, dev) {
6940 if (dev->rtnl_link_ops)
6941 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6943 unregister_netdevice_queue(dev, &dev_kill_list);
6946 unregister_netdevice_many(&dev_kill_list);
6947 list_del(&dev_kill_list);
6951 static struct pernet_operations __net_initdata default_device_ops = {
6952 .exit = default_device_exit,
6953 .exit_batch = default_device_exit_batch,
6957 * Initialize the DEV module. At boot time this walks the device list and
6958 * unhooks any devices that fail to initialise (normally hardware not
6959 * present) and leaves us with a valid list of present and active devices.
6964 * This is called single threaded during boot, so no need
6965 * to take the rtnl semaphore.
6967 static int __init net_dev_init(void)
6969 int i, rc = -ENOMEM;
6971 BUG_ON(!dev_boot_phase);
6973 if (dev_proc_init())
6976 if (netdev_kobject_init())
6979 INIT_LIST_HEAD(&ptype_all);
6980 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6981 INIT_LIST_HEAD(&ptype_base[i]);
6983 INIT_LIST_HEAD(&offload_base);
6985 if (register_pernet_subsys(&netdev_net_ops))
6989 * Initialise the packet receive queues.
6992 for_each_possible_cpu(i) {
6993 struct softnet_data *sd = &per_cpu(softnet_data, i);
6995 skb_queue_head_init(&sd->input_pkt_queue);
6996 skb_queue_head_init(&sd->process_queue);
6997 INIT_LIST_HEAD(&sd->poll_list);
6998 sd->output_queue_tailp = &sd->output_queue;
7000 sd->csd.func = rps_trigger_softirq;
7005 sd->backlog.poll = process_backlog;
7006 sd->backlog.weight = weight_p;
7011 /* The loopback device is special if any other network devices
7012 * is present in a network namespace the loopback device must
7013 * be present. Since we now dynamically allocate and free the
7014 * loopback device ensure this invariant is maintained by
7015 * keeping the loopback device as the first device on the
7016 * list of network devices. Ensuring the loopback devices
7017 * is the first device that appears and the last network device
7020 if (register_pernet_device(&loopback_net_ops))
7023 if (register_pernet_device(&default_device_ops))
7026 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7027 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7029 hotcpu_notifier(dev_cpu_callback, 0);
7036 subsys_initcall(net_dev_init);