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 write_lock_bh(&dev_base_lock);
1123 hlist_del_rcu(&dev->name_hlist);
1124 write_unlock_bh(&dev_base_lock);
1128 write_lock_bh(&dev_base_lock);
1129 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1130 write_unlock_bh(&dev_base_lock);
1132 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1133 ret = notifier_to_errno(ret);
1136 /* err >= 0 after dev_alloc_name() or stores the first errno */
1139 write_seqcount_begin(&devnet_rename_seq);
1140 memcpy(dev->name, oldname, IFNAMSIZ);
1143 pr_err("%s: name change rollback failed: %d\n",
1152 * dev_set_alias - change ifalias of a device
1154 * @alias: name up to IFALIASZ
1155 * @len: limit of bytes to copy from info
1157 * Set ifalias for a device,
1159 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1165 if (len >= IFALIASZ)
1169 kfree(dev->ifalias);
1170 dev->ifalias = NULL;
1174 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1177 dev->ifalias = new_ifalias;
1179 strlcpy(dev->ifalias, alias, len+1);
1185 * netdev_features_change - device changes features
1186 * @dev: device to cause notification
1188 * Called to indicate a device has changed features.
1190 void netdev_features_change(struct net_device *dev)
1192 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1194 EXPORT_SYMBOL(netdev_features_change);
1197 * netdev_state_change - device changes state
1198 * @dev: device to cause notification
1200 * Called to indicate a device has changed state. This function calls
1201 * the notifier chains for netdev_chain and sends a NEWLINK message
1202 * to the routing socket.
1204 void netdev_state_change(struct net_device *dev)
1206 if (dev->flags & IFF_UP) {
1207 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1208 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1211 EXPORT_SYMBOL(netdev_state_change);
1214 * netdev_notify_peers - notify network peers about existence of @dev
1215 * @dev: network device
1217 * Generate traffic such that interested network peers are aware of
1218 * @dev, such as by generating a gratuitous ARP. This may be used when
1219 * a device wants to inform the rest of the network about some sort of
1220 * reconfiguration such as a failover event or virtual machine
1223 void netdev_notify_peers(struct net_device *dev)
1226 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1229 EXPORT_SYMBOL(netdev_notify_peers);
1231 static int __dev_open(struct net_device *dev)
1233 const struct net_device_ops *ops = dev->netdev_ops;
1238 if (!netif_device_present(dev))
1241 /* Block netpoll from trying to do any rx path servicing.
1242 * If we don't do this there is a chance ndo_poll_controller
1243 * or ndo_poll may be running while we open the device
1245 netpoll_rx_disable(dev);
1247 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1248 ret = notifier_to_errno(ret);
1252 set_bit(__LINK_STATE_START, &dev->state);
1254 if (ops->ndo_validate_addr)
1255 ret = ops->ndo_validate_addr(dev);
1257 if (!ret && ops->ndo_open)
1258 ret = ops->ndo_open(dev);
1260 netpoll_rx_enable(dev);
1263 clear_bit(__LINK_STATE_START, &dev->state);
1265 dev->flags |= IFF_UP;
1266 net_dmaengine_get();
1267 dev_set_rx_mode(dev);
1269 add_device_randomness(dev->dev_addr, dev->addr_len);
1276 * dev_open - prepare an interface for use.
1277 * @dev: device to open
1279 * Takes a device from down to up state. The device's private open
1280 * function is invoked and then the multicast lists are loaded. Finally
1281 * the device is moved into the up state and a %NETDEV_UP message is
1282 * sent to the netdev notifier chain.
1284 * Calling this function on an active interface is a nop. On a failure
1285 * a negative errno code is returned.
1287 int dev_open(struct net_device *dev)
1291 if (dev->flags & IFF_UP)
1294 ret = __dev_open(dev);
1298 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1299 call_netdevice_notifiers(NETDEV_UP, dev);
1303 EXPORT_SYMBOL(dev_open);
1305 static int __dev_close_many(struct list_head *head)
1307 struct net_device *dev;
1312 list_for_each_entry(dev, head, close_list) {
1313 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1315 clear_bit(__LINK_STATE_START, &dev->state);
1317 /* Synchronize to scheduled poll. We cannot touch poll list, it
1318 * can be even on different cpu. So just clear netif_running().
1320 * dev->stop() will invoke napi_disable() on all of it's
1321 * napi_struct instances on this device.
1323 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1326 dev_deactivate_many(head);
1328 list_for_each_entry(dev, head, close_list) {
1329 const struct net_device_ops *ops = dev->netdev_ops;
1332 * Call the device specific close. This cannot fail.
1333 * Only if device is UP
1335 * We allow it to be called even after a DETACH hot-plug
1341 dev->flags &= ~IFF_UP;
1342 net_dmaengine_put();
1348 static int __dev_close(struct net_device *dev)
1353 /* Temporarily disable netpoll until the interface is down */
1354 netpoll_rx_disable(dev);
1356 list_add(&dev->close_list, &single);
1357 retval = __dev_close_many(&single);
1360 netpoll_rx_enable(dev);
1364 static int dev_close_many(struct list_head *head)
1366 struct net_device *dev, *tmp;
1368 /* Remove the devices that don't need to be closed */
1369 list_for_each_entry_safe(dev, tmp, head, close_list)
1370 if (!(dev->flags & IFF_UP))
1371 list_del_init(&dev->close_list);
1373 __dev_close_many(head);
1375 list_for_each_entry_safe(dev, tmp, head, close_list) {
1376 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1377 call_netdevice_notifiers(NETDEV_DOWN, dev);
1378 list_del_init(&dev->close_list);
1385 * dev_close - shutdown an interface.
1386 * @dev: device to shutdown
1388 * This function moves an active device into down state. A
1389 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1390 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1393 int dev_close(struct net_device *dev)
1395 if (dev->flags & IFF_UP) {
1398 /* Block netpoll rx while the interface is going down */
1399 netpoll_rx_disable(dev);
1401 list_add(&dev->close_list, &single);
1402 dev_close_many(&single);
1405 netpoll_rx_enable(dev);
1409 EXPORT_SYMBOL(dev_close);
1413 * dev_disable_lro - disable Large Receive Offload on a device
1416 * Disable Large Receive Offload (LRO) on a net device. Must be
1417 * called under RTNL. This is needed if received packets may be
1418 * forwarded to another interface.
1420 void dev_disable_lro(struct net_device *dev)
1423 * If we're trying to disable lro on a vlan device
1424 * use the underlying physical device instead
1426 if (is_vlan_dev(dev))
1427 dev = vlan_dev_real_dev(dev);
1429 /* the same for macvlan devices */
1430 if (netif_is_macvlan(dev))
1431 dev = macvlan_dev_real_dev(dev);
1433 dev->wanted_features &= ~NETIF_F_LRO;
1434 netdev_update_features(dev);
1436 if (unlikely(dev->features & NETIF_F_LRO))
1437 netdev_WARN(dev, "failed to disable LRO!\n");
1439 EXPORT_SYMBOL(dev_disable_lro);
1441 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1442 struct net_device *dev)
1444 struct netdev_notifier_info info;
1446 netdev_notifier_info_init(&info, dev);
1447 return nb->notifier_call(nb, val, &info);
1450 static int dev_boot_phase = 1;
1453 * register_netdevice_notifier - register a network notifier block
1456 * Register a notifier to be called when network device events occur.
1457 * The notifier passed is linked into the kernel structures and must
1458 * not be reused until it has been unregistered. A negative errno code
1459 * is returned on a failure.
1461 * When registered all registration and up events are replayed
1462 * to the new notifier to allow device to have a race free
1463 * view of the network device list.
1466 int register_netdevice_notifier(struct notifier_block *nb)
1468 struct net_device *dev;
1469 struct net_device *last;
1474 err = raw_notifier_chain_register(&netdev_chain, nb);
1480 for_each_netdev(net, dev) {
1481 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1482 err = notifier_to_errno(err);
1486 if (!(dev->flags & IFF_UP))
1489 call_netdevice_notifier(nb, NETDEV_UP, dev);
1500 for_each_netdev(net, dev) {
1504 if (dev->flags & IFF_UP) {
1505 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1507 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1509 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1514 raw_notifier_chain_unregister(&netdev_chain, nb);
1517 EXPORT_SYMBOL(register_netdevice_notifier);
1520 * unregister_netdevice_notifier - unregister a network notifier block
1523 * Unregister a notifier previously registered by
1524 * register_netdevice_notifier(). The notifier is unlinked into the
1525 * kernel structures and may then be reused. A negative errno code
1526 * is returned on a failure.
1528 * After unregistering unregister and down device events are synthesized
1529 * for all devices on the device list to the removed notifier to remove
1530 * the need for special case cleanup code.
1533 int unregister_netdevice_notifier(struct notifier_block *nb)
1535 struct net_device *dev;
1540 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1545 for_each_netdev(net, dev) {
1546 if (dev->flags & IFF_UP) {
1547 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1549 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1551 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1558 EXPORT_SYMBOL(unregister_netdevice_notifier);
1561 * call_netdevice_notifiers_info - call all network notifier blocks
1562 * @val: value passed unmodified to notifier function
1563 * @dev: net_device pointer passed unmodified to notifier function
1564 * @info: notifier information data
1566 * Call all network notifier blocks. Parameters and return value
1567 * are as for raw_notifier_call_chain().
1570 static int call_netdevice_notifiers_info(unsigned long val,
1571 struct net_device *dev,
1572 struct netdev_notifier_info *info)
1575 netdev_notifier_info_init(info, dev);
1576 return raw_notifier_call_chain(&netdev_chain, val, info);
1580 * call_netdevice_notifiers - call all network notifier blocks
1581 * @val: value passed unmodified to notifier function
1582 * @dev: net_device pointer passed unmodified to notifier function
1584 * Call all network notifier blocks. Parameters and return value
1585 * are as for raw_notifier_call_chain().
1588 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1590 struct netdev_notifier_info info;
1592 return call_netdevice_notifiers_info(val, dev, &info);
1594 EXPORT_SYMBOL(call_netdevice_notifiers);
1596 static struct static_key netstamp_needed __read_mostly;
1597 #ifdef HAVE_JUMP_LABEL
1598 /* We are not allowed to call static_key_slow_dec() from irq context
1599 * If net_disable_timestamp() is called from irq context, defer the
1600 * static_key_slow_dec() calls.
1602 static atomic_t netstamp_needed_deferred;
1605 void net_enable_timestamp(void)
1607 #ifdef HAVE_JUMP_LABEL
1608 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1612 static_key_slow_dec(&netstamp_needed);
1616 static_key_slow_inc(&netstamp_needed);
1618 EXPORT_SYMBOL(net_enable_timestamp);
1620 void net_disable_timestamp(void)
1622 #ifdef HAVE_JUMP_LABEL
1623 if (in_interrupt()) {
1624 atomic_inc(&netstamp_needed_deferred);
1628 static_key_slow_dec(&netstamp_needed);
1630 EXPORT_SYMBOL(net_disable_timestamp);
1632 static inline void net_timestamp_set(struct sk_buff *skb)
1634 skb->tstamp.tv64 = 0;
1635 if (static_key_false(&netstamp_needed))
1636 __net_timestamp(skb);
1639 #define net_timestamp_check(COND, SKB) \
1640 if (static_key_false(&netstamp_needed)) { \
1641 if ((COND) && !(SKB)->tstamp.tv64) \
1642 __net_timestamp(SKB); \
1645 static inline bool is_skb_forwardable(struct net_device *dev,
1646 struct sk_buff *skb)
1650 if (!(dev->flags & IFF_UP))
1653 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1654 if (skb->len <= len)
1657 /* if TSO is enabled, we don't care about the length as the packet
1658 * could be forwarded without being segmented before
1660 if (skb_is_gso(skb))
1667 * dev_forward_skb - loopback an skb to another netif
1669 * @dev: destination network device
1670 * @skb: buffer to forward
1673 * NET_RX_SUCCESS (no congestion)
1674 * NET_RX_DROP (packet was dropped, but freed)
1676 * dev_forward_skb can be used for injecting an skb from the
1677 * start_xmit function of one device into the receive queue
1678 * of another device.
1680 * The receiving device may be in another namespace, so
1681 * we have to clear all information in the skb that could
1682 * impact namespace isolation.
1684 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1686 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1687 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1688 atomic_long_inc(&dev->rx_dropped);
1694 if (unlikely(!is_skb_forwardable(dev, skb))) {
1695 atomic_long_inc(&dev->rx_dropped);
1700 skb_scrub_packet(skb, true);
1701 skb->protocol = eth_type_trans(skb, dev);
1703 return netif_rx_internal(skb);
1705 EXPORT_SYMBOL_GPL(dev_forward_skb);
1707 static inline int deliver_skb(struct sk_buff *skb,
1708 struct packet_type *pt_prev,
1709 struct net_device *orig_dev)
1711 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1713 atomic_inc(&skb->users);
1714 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1717 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1719 if (!ptype->af_packet_priv || !skb->sk)
1722 if (ptype->id_match)
1723 return ptype->id_match(ptype, skb->sk);
1724 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1731 * Support routine. Sends outgoing frames to any network
1732 * taps currently in use.
1735 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1737 struct packet_type *ptype;
1738 struct sk_buff *skb2 = NULL;
1739 struct packet_type *pt_prev = NULL;
1742 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1743 /* Never send packets back to the socket
1744 * they originated from - MvS (miquels@drinkel.ow.org)
1746 if ((ptype->dev == dev || !ptype->dev) &&
1747 (!skb_loop_sk(ptype, skb))) {
1749 deliver_skb(skb2, pt_prev, skb->dev);
1754 skb2 = skb_clone(skb, GFP_ATOMIC);
1758 net_timestamp_set(skb2);
1760 /* skb->nh should be correctly
1761 set by sender, so that the second statement is
1762 just protection against buggy protocols.
1764 skb_reset_mac_header(skb2);
1766 if (skb_network_header(skb2) < skb2->data ||
1767 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1768 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1769 ntohs(skb2->protocol),
1771 skb_reset_network_header(skb2);
1774 skb2->transport_header = skb2->network_header;
1775 skb2->pkt_type = PACKET_OUTGOING;
1780 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1785 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1786 * @dev: Network device
1787 * @txq: number of queues available
1789 * If real_num_tx_queues is changed the tc mappings may no longer be
1790 * valid. To resolve this verify the tc mapping remains valid and if
1791 * not NULL the mapping. With no priorities mapping to this
1792 * offset/count pair it will no longer be used. In the worst case TC0
1793 * is invalid nothing can be done so disable priority mappings. If is
1794 * expected that drivers will fix this mapping if they can before
1795 * calling netif_set_real_num_tx_queues.
1797 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1800 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1802 /* If TC0 is invalidated disable TC mapping */
1803 if (tc->offset + tc->count > txq) {
1804 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1809 /* Invalidated prio to tc mappings set to TC0 */
1810 for (i = 1; i < TC_BITMASK + 1; i++) {
1811 int q = netdev_get_prio_tc_map(dev, i);
1813 tc = &dev->tc_to_txq[q];
1814 if (tc->offset + tc->count > txq) {
1815 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1817 netdev_set_prio_tc_map(dev, i, 0);
1823 static DEFINE_MUTEX(xps_map_mutex);
1824 #define xmap_dereference(P) \
1825 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1827 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1830 struct xps_map *map = NULL;
1834 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1836 for (pos = 0; map && pos < map->len; pos++) {
1837 if (map->queues[pos] == index) {
1839 map->queues[pos] = map->queues[--map->len];
1841 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1842 kfree_rcu(map, rcu);
1852 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1854 struct xps_dev_maps *dev_maps;
1856 bool active = false;
1858 mutex_lock(&xps_map_mutex);
1859 dev_maps = xmap_dereference(dev->xps_maps);
1864 for_each_possible_cpu(cpu) {
1865 for (i = index; i < dev->num_tx_queues; i++) {
1866 if (!remove_xps_queue(dev_maps, cpu, i))
1869 if (i == dev->num_tx_queues)
1874 RCU_INIT_POINTER(dev->xps_maps, NULL);
1875 kfree_rcu(dev_maps, rcu);
1878 for (i = index; i < dev->num_tx_queues; i++)
1879 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1883 mutex_unlock(&xps_map_mutex);
1886 static struct xps_map *expand_xps_map(struct xps_map *map,
1889 struct xps_map *new_map;
1890 int alloc_len = XPS_MIN_MAP_ALLOC;
1893 for (pos = 0; map && pos < map->len; pos++) {
1894 if (map->queues[pos] != index)
1899 /* Need to add queue to this CPU's existing map */
1901 if (pos < map->alloc_len)
1904 alloc_len = map->alloc_len * 2;
1907 /* Need to allocate new map to store queue on this CPU's map */
1908 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1913 for (i = 0; i < pos; i++)
1914 new_map->queues[i] = map->queues[i];
1915 new_map->alloc_len = alloc_len;
1921 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1924 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1925 struct xps_map *map, *new_map;
1926 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1927 int cpu, numa_node_id = -2;
1928 bool active = false;
1930 mutex_lock(&xps_map_mutex);
1932 dev_maps = xmap_dereference(dev->xps_maps);
1934 /* allocate memory for queue storage */
1935 for_each_online_cpu(cpu) {
1936 if (!cpumask_test_cpu(cpu, mask))
1940 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1941 if (!new_dev_maps) {
1942 mutex_unlock(&xps_map_mutex);
1946 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1949 map = expand_xps_map(map, cpu, index);
1953 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1957 goto out_no_new_maps;
1959 for_each_possible_cpu(cpu) {
1960 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1961 /* add queue to CPU maps */
1964 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1965 while ((pos < map->len) && (map->queues[pos] != index))
1968 if (pos == map->len)
1969 map->queues[map->len++] = index;
1971 if (numa_node_id == -2)
1972 numa_node_id = cpu_to_node(cpu);
1973 else if (numa_node_id != cpu_to_node(cpu))
1976 } else if (dev_maps) {
1977 /* fill in the new device map from the old device map */
1978 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1979 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1984 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1986 /* Cleanup old maps */
1988 for_each_possible_cpu(cpu) {
1989 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1990 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1991 if (map && map != new_map)
1992 kfree_rcu(map, rcu);
1995 kfree_rcu(dev_maps, rcu);
1998 dev_maps = new_dev_maps;
2002 /* update Tx queue numa node */
2003 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2004 (numa_node_id >= 0) ? numa_node_id :
2010 /* removes queue from unused CPUs */
2011 for_each_possible_cpu(cpu) {
2012 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2015 if (remove_xps_queue(dev_maps, cpu, index))
2019 /* free map if not active */
2021 RCU_INIT_POINTER(dev->xps_maps, NULL);
2022 kfree_rcu(dev_maps, rcu);
2026 mutex_unlock(&xps_map_mutex);
2030 /* remove any maps that we added */
2031 for_each_possible_cpu(cpu) {
2032 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2033 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2035 if (new_map && new_map != map)
2039 mutex_unlock(&xps_map_mutex);
2041 kfree(new_dev_maps);
2044 EXPORT_SYMBOL(netif_set_xps_queue);
2048 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2049 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2051 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2055 if (txq < 1 || txq > dev->num_tx_queues)
2058 if (dev->reg_state == NETREG_REGISTERED ||
2059 dev->reg_state == NETREG_UNREGISTERING) {
2062 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2068 netif_setup_tc(dev, txq);
2070 if (txq < dev->real_num_tx_queues) {
2071 qdisc_reset_all_tx_gt(dev, txq);
2073 netif_reset_xps_queues_gt(dev, txq);
2078 dev->real_num_tx_queues = txq;
2081 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2085 * netif_set_real_num_rx_queues - set actual number of RX queues used
2086 * @dev: Network device
2087 * @rxq: Actual number of RX queues
2089 * This must be called either with the rtnl_lock held or before
2090 * registration of the net device. Returns 0 on success, or a
2091 * negative error code. If called before registration, it always
2094 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2098 if (rxq < 1 || rxq > dev->num_rx_queues)
2101 if (dev->reg_state == NETREG_REGISTERED) {
2104 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2110 dev->real_num_rx_queues = rxq;
2113 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2117 * netif_get_num_default_rss_queues - default number of RSS queues
2119 * This routine should set an upper limit on the number of RSS queues
2120 * used by default by multiqueue devices.
2122 int netif_get_num_default_rss_queues(void)
2124 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2126 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2128 static inline void __netif_reschedule(struct Qdisc *q)
2130 struct softnet_data *sd;
2131 unsigned long flags;
2133 local_irq_save(flags);
2134 sd = &__get_cpu_var(softnet_data);
2135 q->next_sched = NULL;
2136 *sd->output_queue_tailp = q;
2137 sd->output_queue_tailp = &q->next_sched;
2138 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2139 local_irq_restore(flags);
2142 void __netif_schedule(struct Qdisc *q)
2144 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2145 __netif_reschedule(q);
2147 EXPORT_SYMBOL(__netif_schedule);
2149 struct dev_kfree_skb_cb {
2150 enum skb_free_reason reason;
2153 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2155 return (struct dev_kfree_skb_cb *)skb->cb;
2158 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2160 unsigned long flags;
2162 if (likely(atomic_read(&skb->users) == 1)) {
2164 atomic_set(&skb->users, 0);
2165 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2168 get_kfree_skb_cb(skb)->reason = reason;
2169 local_irq_save(flags);
2170 skb->next = __this_cpu_read(softnet_data.completion_queue);
2171 __this_cpu_write(softnet_data.completion_queue, skb);
2172 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2173 local_irq_restore(flags);
2175 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2177 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2179 if (in_irq() || irqs_disabled())
2180 __dev_kfree_skb_irq(skb, reason);
2184 EXPORT_SYMBOL(__dev_kfree_skb_any);
2188 * netif_device_detach - mark device as removed
2189 * @dev: network device
2191 * Mark device as removed from system and therefore no longer available.
2193 void netif_device_detach(struct net_device *dev)
2195 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2196 netif_running(dev)) {
2197 netif_tx_stop_all_queues(dev);
2200 EXPORT_SYMBOL(netif_device_detach);
2203 * netif_device_attach - mark device as attached
2204 * @dev: network device
2206 * Mark device as attached from system and restart if needed.
2208 void netif_device_attach(struct net_device *dev)
2210 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2211 netif_running(dev)) {
2212 netif_tx_wake_all_queues(dev);
2213 __netdev_watchdog_up(dev);
2216 EXPORT_SYMBOL(netif_device_attach);
2218 static void skb_warn_bad_offload(const struct sk_buff *skb)
2220 static const netdev_features_t null_features = 0;
2221 struct net_device *dev = skb->dev;
2222 const char *driver = "";
2224 if (!net_ratelimit())
2227 if (dev && dev->dev.parent)
2228 driver = dev_driver_string(dev->dev.parent);
2230 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2231 "gso_type=%d ip_summed=%d\n",
2232 driver, dev ? &dev->features : &null_features,
2233 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2234 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2235 skb_shinfo(skb)->gso_type, skb->ip_summed);
2239 * Invalidate hardware checksum when packet is to be mangled, and
2240 * complete checksum manually on outgoing path.
2242 int skb_checksum_help(struct sk_buff *skb)
2245 int ret = 0, offset;
2247 if (skb->ip_summed == CHECKSUM_COMPLETE)
2248 goto out_set_summed;
2250 if (unlikely(skb_shinfo(skb)->gso_size)) {
2251 skb_warn_bad_offload(skb);
2255 /* Before computing a checksum, we should make sure no frag could
2256 * be modified by an external entity : checksum could be wrong.
2258 if (skb_has_shared_frag(skb)) {
2259 ret = __skb_linearize(skb);
2264 offset = skb_checksum_start_offset(skb);
2265 BUG_ON(offset >= skb_headlen(skb));
2266 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2268 offset += skb->csum_offset;
2269 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2271 if (skb_cloned(skb) &&
2272 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2273 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2278 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2280 skb->ip_summed = CHECKSUM_NONE;
2284 EXPORT_SYMBOL(skb_checksum_help);
2286 __be16 skb_network_protocol(struct sk_buff *skb)
2288 __be16 type = skb->protocol;
2289 int vlan_depth = ETH_HLEN;
2291 /* Tunnel gso handlers can set protocol to ethernet. */
2292 if (type == htons(ETH_P_TEB)) {
2295 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2298 eth = (struct ethhdr *)skb_mac_header(skb);
2299 type = eth->h_proto;
2302 while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2303 struct vlan_hdr *vh;
2305 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2308 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2309 type = vh->h_vlan_encapsulated_proto;
2310 vlan_depth += VLAN_HLEN;
2317 * skb_mac_gso_segment - mac layer segmentation handler.
2318 * @skb: buffer to segment
2319 * @features: features for the output path (see dev->features)
2321 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2322 netdev_features_t features)
2324 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2325 struct packet_offload *ptype;
2326 __be16 type = skb_network_protocol(skb);
2328 if (unlikely(!type))
2329 return ERR_PTR(-EINVAL);
2331 __skb_pull(skb, skb->mac_len);
2334 list_for_each_entry_rcu(ptype, &offload_base, list) {
2335 if (ptype->type == type && ptype->callbacks.gso_segment) {
2336 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2339 err = ptype->callbacks.gso_send_check(skb);
2340 segs = ERR_PTR(err);
2341 if (err || skb_gso_ok(skb, features))
2343 __skb_push(skb, (skb->data -
2344 skb_network_header(skb)));
2346 segs = ptype->callbacks.gso_segment(skb, features);
2352 __skb_push(skb, skb->data - skb_mac_header(skb));
2356 EXPORT_SYMBOL(skb_mac_gso_segment);
2359 /* openvswitch calls this on rx path, so we need a different check.
2361 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2364 return skb->ip_summed != CHECKSUM_PARTIAL;
2366 return skb->ip_summed == CHECKSUM_NONE;
2370 * __skb_gso_segment - Perform segmentation on skb.
2371 * @skb: buffer to segment
2372 * @features: features for the output path (see dev->features)
2373 * @tx_path: whether it is called in TX path
2375 * This function segments the given skb and returns a list of segments.
2377 * It may return NULL if the skb requires no segmentation. This is
2378 * only possible when GSO is used for verifying header integrity.
2380 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2381 netdev_features_t features, bool tx_path)
2383 if (unlikely(skb_needs_check(skb, tx_path))) {
2386 skb_warn_bad_offload(skb);
2388 if (skb_header_cloned(skb) &&
2389 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2390 return ERR_PTR(err);
2393 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2394 SKB_GSO_CB(skb)->encap_level = 0;
2396 skb_reset_mac_header(skb);
2397 skb_reset_mac_len(skb);
2399 return skb_mac_gso_segment(skb, features);
2401 EXPORT_SYMBOL(__skb_gso_segment);
2403 /* Take action when hardware reception checksum errors are detected. */
2405 void netdev_rx_csum_fault(struct net_device *dev)
2407 if (net_ratelimit()) {
2408 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2412 EXPORT_SYMBOL(netdev_rx_csum_fault);
2415 /* Actually, we should eliminate this check as soon as we know, that:
2416 * 1. IOMMU is present and allows to map all the memory.
2417 * 2. No high memory really exists on this machine.
2420 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2422 #ifdef CONFIG_HIGHMEM
2424 if (!(dev->features & NETIF_F_HIGHDMA)) {
2425 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2426 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2427 if (PageHighMem(skb_frag_page(frag)))
2432 if (PCI_DMA_BUS_IS_PHYS) {
2433 struct device *pdev = dev->dev.parent;
2437 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2438 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2439 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2440 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2449 void (*destructor)(struct sk_buff *skb);
2452 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2454 static void dev_gso_skb_destructor(struct sk_buff *skb)
2456 struct dev_gso_cb *cb;
2458 kfree_skb_list(skb->next);
2461 cb = DEV_GSO_CB(skb);
2463 cb->destructor(skb);
2467 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2468 * @skb: buffer to segment
2469 * @features: device features as applicable to this skb
2471 * This function segments the given skb and stores the list of segments
2474 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2476 struct sk_buff *segs;
2478 segs = skb_gso_segment(skb, features);
2480 /* Verifying header integrity only. */
2485 return PTR_ERR(segs);
2488 DEV_GSO_CB(skb)->destructor = skb->destructor;
2489 skb->destructor = dev_gso_skb_destructor;
2494 static netdev_features_t harmonize_features(struct sk_buff *skb,
2495 netdev_features_t features)
2497 if (skb->ip_summed != CHECKSUM_NONE &&
2498 !can_checksum_protocol(features, skb_network_protocol(skb))) {
2499 features &= ~NETIF_F_ALL_CSUM;
2500 } else if (illegal_highdma(skb->dev, skb)) {
2501 features &= ~NETIF_F_SG;
2507 netdev_features_t netif_skb_features(struct sk_buff *skb)
2509 __be16 protocol = skb->protocol;
2510 netdev_features_t features = skb->dev->features;
2512 if (skb_shinfo(skb)->gso_segs > skb->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, features);
2522 features &= (skb->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, features);
2532 EXPORT_SYMBOL(netif_skb_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
2781 * Queue a buffer for transmission to a network device. The caller must
2782 * have set the device and priority and built the buffer before calling
2783 * this function. The function can be called from an interrupt.
2785 * A negative errno code is returned on a failure. A success does not
2786 * guarantee the frame will be transmitted as it may be dropped due
2787 * to congestion or traffic shaping.
2789 * -----------------------------------------------------------------------------------
2790 * I notice this method can also return errors from the queue disciplines,
2791 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2794 * Regardless of the return value, the skb is consumed, so it is currently
2795 * difficult to retry a send to this method. (You can bump the ref count
2796 * before sending to hold a reference for retry if you are careful.)
2798 * When calling this method, interrupts MUST be enabled. This is because
2799 * the BH enable code must have IRQs enabled so that it will not deadlock.
2802 int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2804 struct net_device *dev = skb->dev;
2805 struct netdev_queue *txq;
2809 skb_reset_mac_header(skb);
2811 /* Disable soft irqs for various locks below. Also
2812 * stops preemption for RCU.
2816 skb_update_prio(skb);
2818 txq = netdev_pick_tx(dev, skb, accel_priv);
2819 q = rcu_dereference_bh(txq->qdisc);
2821 #ifdef CONFIG_NET_CLS_ACT
2822 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2824 trace_net_dev_queue(skb);
2826 rc = __dev_xmit_skb(skb, q, dev, txq);
2830 /* The device has no queue. Common case for software devices:
2831 loopback, all the sorts of tunnels...
2833 Really, it is unlikely that netif_tx_lock protection is necessary
2834 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2836 However, it is possible, that they rely on protection
2839 Check this and shot the lock. It is not prone from deadlocks.
2840 Either shot noqueue qdisc, it is even simpler 8)
2842 if (dev->flags & IFF_UP) {
2843 int cpu = smp_processor_id(); /* ok because BHs are off */
2845 if (txq->xmit_lock_owner != cpu) {
2847 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2848 goto recursion_alert;
2850 HARD_TX_LOCK(dev, txq, cpu);
2852 if (!netif_xmit_stopped(txq)) {
2853 __this_cpu_inc(xmit_recursion);
2854 rc = dev_hard_start_xmit(skb, dev, txq);
2855 __this_cpu_dec(xmit_recursion);
2856 if (dev_xmit_complete(rc)) {
2857 HARD_TX_UNLOCK(dev, txq);
2861 HARD_TX_UNLOCK(dev, txq);
2862 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2865 /* Recursion is detected! It is possible,
2869 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2875 rcu_read_unlock_bh();
2880 rcu_read_unlock_bh();
2884 int dev_queue_xmit(struct sk_buff *skb)
2886 return __dev_queue_xmit(skb, NULL);
2888 EXPORT_SYMBOL(dev_queue_xmit);
2890 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
2892 return __dev_queue_xmit(skb, accel_priv);
2894 EXPORT_SYMBOL(dev_queue_xmit_accel);
2897 /*=======================================================================
2899 =======================================================================*/
2901 int netdev_max_backlog __read_mostly = 1000;
2902 EXPORT_SYMBOL(netdev_max_backlog);
2904 int netdev_tstamp_prequeue __read_mostly = 1;
2905 int netdev_budget __read_mostly = 300;
2906 int weight_p __read_mostly = 64; /* old backlog weight */
2908 /* Called with irq disabled */
2909 static inline void ____napi_schedule(struct softnet_data *sd,
2910 struct napi_struct *napi)
2912 list_add_tail(&napi->poll_list, &sd->poll_list);
2913 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2918 /* One global table that all flow-based protocols share. */
2919 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2920 EXPORT_SYMBOL(rps_sock_flow_table);
2922 struct static_key rps_needed __read_mostly;
2924 static struct rps_dev_flow *
2925 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2926 struct rps_dev_flow *rflow, u16 next_cpu)
2928 if (next_cpu != RPS_NO_CPU) {
2929 #ifdef CONFIG_RFS_ACCEL
2930 struct netdev_rx_queue *rxqueue;
2931 struct rps_dev_flow_table *flow_table;
2932 struct rps_dev_flow *old_rflow;
2937 /* Should we steer this flow to a different hardware queue? */
2938 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2939 !(dev->features & NETIF_F_NTUPLE))
2941 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2942 if (rxq_index == skb_get_rx_queue(skb))
2945 rxqueue = dev->_rx + rxq_index;
2946 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2949 flow_id = skb->rxhash & flow_table->mask;
2950 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2951 rxq_index, flow_id);
2955 rflow = &flow_table->flows[flow_id];
2957 if (old_rflow->filter == rflow->filter)
2958 old_rflow->filter = RPS_NO_FILTER;
2962 per_cpu(softnet_data, next_cpu).input_queue_head;
2965 rflow->cpu = next_cpu;
2970 * get_rps_cpu is called from netif_receive_skb and returns the target
2971 * CPU from the RPS map of the receiving queue for a given skb.
2972 * rcu_read_lock must be held on entry.
2974 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2975 struct rps_dev_flow **rflowp)
2977 struct netdev_rx_queue *rxqueue;
2978 struct rps_map *map;
2979 struct rps_dev_flow_table *flow_table;
2980 struct rps_sock_flow_table *sock_flow_table;
2984 if (skb_rx_queue_recorded(skb)) {
2985 u16 index = skb_get_rx_queue(skb);
2986 if (unlikely(index >= dev->real_num_rx_queues)) {
2987 WARN_ONCE(dev->real_num_rx_queues > 1,
2988 "%s received packet on queue %u, but number "
2989 "of RX queues is %u\n",
2990 dev->name, index, dev->real_num_rx_queues);
2993 rxqueue = dev->_rx + index;
2997 map = rcu_dereference(rxqueue->rps_map);
2999 if (map->len == 1 &&
3000 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3001 tcpu = map->cpus[0];
3002 if (cpu_online(tcpu))
3006 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3010 skb_reset_network_header(skb);
3011 if (!skb_get_hash(skb))
3014 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3015 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3016 if (flow_table && sock_flow_table) {
3018 struct rps_dev_flow *rflow;
3020 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
3023 next_cpu = sock_flow_table->ents[skb->rxhash &
3024 sock_flow_table->mask];
3027 * If the desired CPU (where last recvmsg was done) is
3028 * different from current CPU (one in the rx-queue flow
3029 * table entry), switch if one of the following holds:
3030 * - Current CPU is unset (equal to RPS_NO_CPU).
3031 * - Current CPU is offline.
3032 * - The current CPU's queue tail has advanced beyond the
3033 * last packet that was enqueued using this table entry.
3034 * This guarantees that all previous packets for the flow
3035 * have been dequeued, thus preserving in order delivery.
3037 if (unlikely(tcpu != next_cpu) &&
3038 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3039 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3040 rflow->last_qtail)) >= 0)) {
3042 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3045 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3053 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
3055 if (cpu_online(tcpu)) {
3065 #ifdef CONFIG_RFS_ACCEL
3068 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3069 * @dev: Device on which the filter was set
3070 * @rxq_index: RX queue index
3071 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3072 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3074 * Drivers that implement ndo_rx_flow_steer() should periodically call
3075 * this function for each installed filter and remove the filters for
3076 * which it returns %true.
3078 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3079 u32 flow_id, u16 filter_id)
3081 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3082 struct rps_dev_flow_table *flow_table;
3083 struct rps_dev_flow *rflow;
3088 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3089 if (flow_table && flow_id <= flow_table->mask) {
3090 rflow = &flow_table->flows[flow_id];
3091 cpu = ACCESS_ONCE(rflow->cpu);
3092 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3093 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3094 rflow->last_qtail) <
3095 (int)(10 * flow_table->mask)))
3101 EXPORT_SYMBOL(rps_may_expire_flow);
3103 #endif /* CONFIG_RFS_ACCEL */
3105 /* Called from hardirq (IPI) context */
3106 static void rps_trigger_softirq(void *data)
3108 struct softnet_data *sd = data;
3110 ____napi_schedule(sd, &sd->backlog);
3114 #endif /* CONFIG_RPS */
3117 * Check if this softnet_data structure is another cpu one
3118 * If yes, queue it to our IPI list and return 1
3121 static int rps_ipi_queued(struct softnet_data *sd)
3124 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3127 sd->rps_ipi_next = mysd->rps_ipi_list;
3128 mysd->rps_ipi_list = sd;
3130 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3133 #endif /* CONFIG_RPS */
3137 #ifdef CONFIG_NET_FLOW_LIMIT
3138 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3141 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3143 #ifdef CONFIG_NET_FLOW_LIMIT
3144 struct sd_flow_limit *fl;
3145 struct softnet_data *sd;
3146 unsigned int old_flow, new_flow;
3148 if (qlen < (netdev_max_backlog >> 1))
3151 sd = &__get_cpu_var(softnet_data);
3154 fl = rcu_dereference(sd->flow_limit);
3156 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3157 old_flow = fl->history[fl->history_head];
3158 fl->history[fl->history_head] = new_flow;
3161 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3163 if (likely(fl->buckets[old_flow]))
3164 fl->buckets[old_flow]--;
3166 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3178 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3179 * queue (may be a remote CPU queue).
3181 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3182 unsigned int *qtail)
3184 struct softnet_data *sd;
3185 unsigned long flags;
3188 sd = &per_cpu(softnet_data, cpu);
3190 local_irq_save(flags);
3193 qlen = skb_queue_len(&sd->input_pkt_queue);
3194 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3195 if (skb_queue_len(&sd->input_pkt_queue)) {
3197 __skb_queue_tail(&sd->input_pkt_queue, skb);
3198 input_queue_tail_incr_save(sd, qtail);
3200 local_irq_restore(flags);
3201 return NET_RX_SUCCESS;
3204 /* Schedule NAPI for backlog device
3205 * We can use non atomic operation since we own the queue lock
3207 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3208 if (!rps_ipi_queued(sd))
3209 ____napi_schedule(sd, &sd->backlog);
3217 local_irq_restore(flags);
3219 atomic_long_inc(&skb->dev->rx_dropped);
3224 static int netif_rx_internal(struct sk_buff *skb)
3228 /* if netpoll wants it, pretend we never saw it */
3229 if (netpoll_rx(skb))
3232 net_timestamp_check(netdev_tstamp_prequeue, skb);
3234 trace_netif_rx(skb);
3236 if (static_key_false(&rps_needed)) {
3237 struct rps_dev_flow voidflow, *rflow = &voidflow;
3243 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3245 cpu = smp_processor_id();
3247 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3255 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3262 * netif_rx - post buffer to the network code
3263 * @skb: buffer to post
3265 * This function receives a packet from a device driver and queues it for
3266 * the upper (protocol) levels to process. It always succeeds. The buffer
3267 * may be dropped during processing for congestion control or by the
3271 * NET_RX_SUCCESS (no congestion)
3272 * NET_RX_DROP (packet was dropped)
3276 int netif_rx(struct sk_buff *skb)
3278 trace_netif_rx_entry(skb);
3280 return netif_rx_internal(skb);
3282 EXPORT_SYMBOL(netif_rx);
3284 int netif_rx_ni(struct sk_buff *skb)
3288 trace_netif_rx_ni_entry(skb);
3291 err = netif_rx_internal(skb);
3292 if (local_softirq_pending())
3298 EXPORT_SYMBOL(netif_rx_ni);
3300 static void net_tx_action(struct softirq_action *h)
3302 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3304 if (sd->completion_queue) {
3305 struct sk_buff *clist;
3307 local_irq_disable();
3308 clist = sd->completion_queue;
3309 sd->completion_queue = NULL;
3313 struct sk_buff *skb = clist;
3314 clist = clist->next;
3316 WARN_ON(atomic_read(&skb->users));
3317 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3318 trace_consume_skb(skb);
3320 trace_kfree_skb(skb, net_tx_action);
3325 if (sd->output_queue) {
3328 local_irq_disable();
3329 head = sd->output_queue;
3330 sd->output_queue = NULL;
3331 sd->output_queue_tailp = &sd->output_queue;
3335 struct Qdisc *q = head;
3336 spinlock_t *root_lock;
3338 head = head->next_sched;
3340 root_lock = qdisc_lock(q);
3341 if (spin_trylock(root_lock)) {
3342 smp_mb__before_clear_bit();
3343 clear_bit(__QDISC_STATE_SCHED,
3346 spin_unlock(root_lock);
3348 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3350 __netif_reschedule(q);
3352 smp_mb__before_clear_bit();
3353 clear_bit(__QDISC_STATE_SCHED,
3361 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3362 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3363 /* This hook is defined here for ATM LANE */
3364 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3365 unsigned char *addr) __read_mostly;
3366 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3369 #ifdef CONFIG_NET_CLS_ACT
3370 /* TODO: Maybe we should just force sch_ingress to be compiled in
3371 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3372 * a compare and 2 stores extra right now if we dont have it on
3373 * but have CONFIG_NET_CLS_ACT
3374 * NOTE: This doesn't stop any functionality; if you dont have
3375 * the ingress scheduler, you just can't add policies on ingress.
3378 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3380 struct net_device *dev = skb->dev;
3381 u32 ttl = G_TC_RTTL(skb->tc_verd);
3382 int result = TC_ACT_OK;
3385 if (unlikely(MAX_RED_LOOP < ttl++)) {
3386 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3387 skb->skb_iif, dev->ifindex);
3391 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3392 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3395 if (q != &noop_qdisc) {
3396 spin_lock(qdisc_lock(q));
3397 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3398 result = qdisc_enqueue_root(skb, q);
3399 spin_unlock(qdisc_lock(q));
3405 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3406 struct packet_type **pt_prev,
3407 int *ret, struct net_device *orig_dev)
3409 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3411 if (!rxq || rxq->qdisc == &noop_qdisc)
3415 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3419 switch (ing_filter(skb, rxq)) {
3433 * netdev_rx_handler_register - register receive handler
3434 * @dev: device to register a handler for
3435 * @rx_handler: receive handler to register
3436 * @rx_handler_data: data pointer that is used by rx handler
3438 * Register a receive hander for a device. This handler will then be
3439 * called from __netif_receive_skb. A negative errno code is returned
3442 * The caller must hold the rtnl_mutex.
3444 * For a general description of rx_handler, see enum rx_handler_result.
3446 int netdev_rx_handler_register(struct net_device *dev,
3447 rx_handler_func_t *rx_handler,
3448 void *rx_handler_data)
3452 if (dev->rx_handler)
3455 /* Note: rx_handler_data must be set before rx_handler */
3456 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3457 rcu_assign_pointer(dev->rx_handler, rx_handler);
3461 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3464 * netdev_rx_handler_unregister - unregister receive handler
3465 * @dev: device to unregister a handler from
3467 * Unregister a receive handler from a device.
3469 * The caller must hold the rtnl_mutex.
3471 void netdev_rx_handler_unregister(struct net_device *dev)
3475 RCU_INIT_POINTER(dev->rx_handler, NULL);
3476 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3477 * section has a guarantee to see a non NULL rx_handler_data
3481 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3483 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3486 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3487 * the special handling of PFMEMALLOC skbs.
3489 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3491 switch (skb->protocol) {
3492 case __constant_htons(ETH_P_ARP):
3493 case __constant_htons(ETH_P_IP):
3494 case __constant_htons(ETH_P_IPV6):
3495 case __constant_htons(ETH_P_8021Q):
3496 case __constant_htons(ETH_P_8021AD):
3503 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3505 struct packet_type *ptype, *pt_prev;
3506 rx_handler_func_t *rx_handler;
3507 struct net_device *orig_dev;
3508 struct net_device *null_or_dev;
3509 bool deliver_exact = false;
3510 int ret = NET_RX_DROP;
3513 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3515 trace_netif_receive_skb(skb);
3517 /* if we've gotten here through NAPI, check netpoll */
3518 if (netpoll_receive_skb(skb))
3521 orig_dev = skb->dev;
3523 skb_reset_network_header(skb);
3524 if (!skb_transport_header_was_set(skb))
3525 skb_reset_transport_header(skb);
3526 skb_reset_mac_len(skb);
3533 skb->skb_iif = skb->dev->ifindex;
3535 __this_cpu_inc(softnet_data.processed);
3537 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3538 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3539 skb = vlan_untag(skb);
3544 #ifdef CONFIG_NET_CLS_ACT
3545 if (skb->tc_verd & TC_NCLS) {
3546 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3554 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3555 if (!ptype->dev || ptype->dev == skb->dev) {
3557 ret = deliver_skb(skb, pt_prev, orig_dev);
3563 #ifdef CONFIG_NET_CLS_ACT
3564 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3570 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3573 if (vlan_tx_tag_present(skb)) {
3575 ret = deliver_skb(skb, pt_prev, orig_dev);
3578 if (vlan_do_receive(&skb))
3580 else if (unlikely(!skb))
3584 rx_handler = rcu_dereference(skb->dev->rx_handler);
3587 ret = deliver_skb(skb, pt_prev, orig_dev);
3590 switch (rx_handler(&skb)) {
3591 case RX_HANDLER_CONSUMED:
3592 ret = NET_RX_SUCCESS;
3594 case RX_HANDLER_ANOTHER:
3596 case RX_HANDLER_EXACT:
3597 deliver_exact = true;
3598 case RX_HANDLER_PASS:
3605 if (unlikely(vlan_tx_tag_present(skb))) {
3606 if (vlan_tx_tag_get_id(skb))
3607 skb->pkt_type = PACKET_OTHERHOST;
3608 /* Note: we might in the future use prio bits
3609 * and set skb->priority like in vlan_do_receive()
3610 * For the time being, just ignore Priority Code Point
3615 /* deliver only exact match when indicated */
3616 null_or_dev = deliver_exact ? skb->dev : NULL;
3618 type = skb->protocol;
3619 list_for_each_entry_rcu(ptype,
3620 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3621 if (ptype->type == type &&
3622 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3623 ptype->dev == orig_dev)) {
3625 ret = deliver_skb(skb, pt_prev, orig_dev);
3631 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3634 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3637 atomic_long_inc(&skb->dev->rx_dropped);
3639 /* Jamal, now you will not able to escape explaining
3640 * me how you were going to use this. :-)
3651 static int __netif_receive_skb(struct sk_buff *skb)
3655 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3656 unsigned long pflags = current->flags;
3659 * PFMEMALLOC skbs are special, they should
3660 * - be delivered to SOCK_MEMALLOC sockets only
3661 * - stay away from userspace
3662 * - have bounded memory usage
3664 * Use PF_MEMALLOC as this saves us from propagating the allocation
3665 * context down to all allocation sites.
3667 current->flags |= PF_MEMALLOC;
3668 ret = __netif_receive_skb_core(skb, true);
3669 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3671 ret = __netif_receive_skb_core(skb, false);
3676 static int netif_receive_skb_internal(struct sk_buff *skb)
3678 net_timestamp_check(netdev_tstamp_prequeue, skb);
3680 if (skb_defer_rx_timestamp(skb))
3681 return NET_RX_SUCCESS;
3684 if (static_key_false(&rps_needed)) {
3685 struct rps_dev_flow voidflow, *rflow = &voidflow;
3690 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3693 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3700 return __netif_receive_skb(skb);
3704 * netif_receive_skb - process receive buffer from network
3705 * @skb: buffer to process
3707 * netif_receive_skb() is the main receive data processing function.
3708 * It always succeeds. The buffer may be dropped during processing
3709 * for congestion control or by the protocol layers.
3711 * This function may only be called from softirq context and interrupts
3712 * should be enabled.
3714 * Return values (usually ignored):
3715 * NET_RX_SUCCESS: no congestion
3716 * NET_RX_DROP: packet was dropped
3718 int netif_receive_skb(struct sk_buff *skb)
3720 trace_netif_receive_skb_entry(skb);
3722 return netif_receive_skb_internal(skb);
3724 EXPORT_SYMBOL(netif_receive_skb);
3726 /* Network device is going away, flush any packets still pending
3727 * Called with irqs disabled.
3729 static void flush_backlog(void *arg)
3731 struct net_device *dev = arg;
3732 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3733 struct sk_buff *skb, *tmp;
3736 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3737 if (skb->dev == dev) {
3738 __skb_unlink(skb, &sd->input_pkt_queue);
3740 input_queue_head_incr(sd);
3745 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3746 if (skb->dev == dev) {
3747 __skb_unlink(skb, &sd->process_queue);
3749 input_queue_head_incr(sd);
3754 static int napi_gro_complete(struct sk_buff *skb)
3756 struct packet_offload *ptype;
3757 __be16 type = skb->protocol;
3758 struct list_head *head = &offload_base;
3761 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3763 if (NAPI_GRO_CB(skb)->count == 1) {
3764 skb_shinfo(skb)->gso_size = 0;
3769 list_for_each_entry_rcu(ptype, head, list) {
3770 if (ptype->type != type || !ptype->callbacks.gro_complete)
3773 err = ptype->callbacks.gro_complete(skb, 0);
3779 WARN_ON(&ptype->list == head);
3781 return NET_RX_SUCCESS;
3785 return netif_receive_skb_internal(skb);
3788 /* napi->gro_list contains packets ordered by age.
3789 * youngest packets at the head of it.
3790 * Complete skbs in reverse order to reduce latencies.
3792 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3794 struct sk_buff *skb, *prev = NULL;
3796 /* scan list and build reverse chain */
3797 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3802 for (skb = prev; skb; skb = prev) {
3805 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3809 napi_gro_complete(skb);
3813 napi->gro_list = NULL;
3815 EXPORT_SYMBOL(napi_gro_flush);
3817 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3820 unsigned int maclen = skb->dev->hard_header_len;
3822 for (p = napi->gro_list; p; p = p->next) {
3823 unsigned long diffs;
3825 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3826 diffs |= p->vlan_tci ^ skb->vlan_tci;
3827 if (maclen == ETH_HLEN)
3828 diffs |= compare_ether_header(skb_mac_header(p),
3829 skb_gro_mac_header(skb));
3831 diffs = memcmp(skb_mac_header(p),
3832 skb_gro_mac_header(skb),
3834 NAPI_GRO_CB(p)->same_flow = !diffs;
3835 NAPI_GRO_CB(p)->flush = 0;
3839 static void skb_gro_reset_offset(struct sk_buff *skb)
3841 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3842 const skb_frag_t *frag0 = &pinfo->frags[0];
3844 NAPI_GRO_CB(skb)->data_offset = 0;
3845 NAPI_GRO_CB(skb)->frag0 = NULL;
3846 NAPI_GRO_CB(skb)->frag0_len = 0;
3848 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3850 !PageHighMem(skb_frag_page(frag0))) {
3851 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3852 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3856 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3858 struct sk_buff **pp = NULL;
3859 struct packet_offload *ptype;
3860 __be16 type = skb->protocol;
3861 struct list_head *head = &offload_base;
3863 enum gro_result ret;
3865 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3868 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3871 skb_gro_reset_offset(skb);
3872 gro_list_prepare(napi, skb);
3873 NAPI_GRO_CB(skb)->csum = skb->csum; /* Needed for CHECKSUM_COMPLETE */
3876 list_for_each_entry_rcu(ptype, head, list) {
3877 if (ptype->type != type || !ptype->callbacks.gro_receive)
3880 skb_set_network_header(skb, skb_gro_offset(skb));
3881 skb_reset_mac_len(skb);
3882 NAPI_GRO_CB(skb)->same_flow = 0;
3883 NAPI_GRO_CB(skb)->flush = 0;
3884 NAPI_GRO_CB(skb)->free = 0;
3886 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3891 if (&ptype->list == head)
3894 same_flow = NAPI_GRO_CB(skb)->same_flow;
3895 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3898 struct sk_buff *nskb = *pp;
3902 napi_gro_complete(nskb);
3909 if (NAPI_GRO_CB(skb)->flush)
3912 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
3913 struct sk_buff *nskb = napi->gro_list;
3915 /* locate the end of the list to select the 'oldest' flow */
3916 while (nskb->next) {
3922 napi_gro_complete(nskb);
3926 NAPI_GRO_CB(skb)->count = 1;
3927 NAPI_GRO_CB(skb)->age = jiffies;
3928 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3929 skb->next = napi->gro_list;
3930 napi->gro_list = skb;
3934 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3935 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3937 BUG_ON(skb->end - skb->tail < grow);
3939 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3942 skb->data_len -= grow;
3944 skb_shinfo(skb)->frags[0].page_offset += grow;
3945 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3947 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3948 skb_frag_unref(skb, 0);
3949 memmove(skb_shinfo(skb)->frags,
3950 skb_shinfo(skb)->frags + 1,
3951 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3963 struct packet_offload *gro_find_receive_by_type(__be16 type)
3965 struct list_head *offload_head = &offload_base;
3966 struct packet_offload *ptype;
3968 list_for_each_entry_rcu(ptype, offload_head, list) {
3969 if (ptype->type != type || !ptype->callbacks.gro_receive)
3976 struct packet_offload *gro_find_complete_by_type(__be16 type)
3978 struct list_head *offload_head = &offload_base;
3979 struct packet_offload *ptype;
3981 list_for_each_entry_rcu(ptype, offload_head, list) {
3982 if (ptype->type != type || !ptype->callbacks.gro_complete)
3989 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3993 if (netif_receive_skb_internal(skb))
4001 case GRO_MERGED_FREE:
4002 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4003 kmem_cache_free(skbuff_head_cache, skb);
4016 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4018 trace_napi_gro_receive_entry(skb);
4020 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4022 EXPORT_SYMBOL(napi_gro_receive);
4024 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4026 __skb_pull(skb, skb_headlen(skb));
4027 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4028 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4030 skb->dev = napi->dev;
4036 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4038 struct sk_buff *skb = napi->skb;
4041 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4046 EXPORT_SYMBOL(napi_get_frags);
4048 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
4053 if (netif_receive_skb_internal(skb))
4058 case GRO_MERGED_FREE:
4059 napi_reuse_skb(napi, skb);
4070 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4072 struct sk_buff *skb = napi->skb;
4076 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) {
4077 napi_reuse_skb(napi, skb);
4080 skb->protocol = eth_type_trans(skb, skb->dev);
4085 gro_result_t napi_gro_frags(struct napi_struct *napi)
4087 struct sk_buff *skb = napi_frags_skb(napi);
4092 trace_napi_gro_frags_entry(skb);
4094 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4096 EXPORT_SYMBOL(napi_gro_frags);
4099 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4100 * Note: called with local irq disabled, but exits with local irq enabled.
4102 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4105 struct softnet_data *remsd = sd->rps_ipi_list;
4108 sd->rps_ipi_list = NULL;
4112 /* Send pending IPI's to kick RPS processing on remote cpus. */
4114 struct softnet_data *next = remsd->rps_ipi_next;
4116 if (cpu_online(remsd->cpu))
4117 __smp_call_function_single(remsd->cpu,
4126 static int process_backlog(struct napi_struct *napi, int quota)
4129 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4132 /* Check if we have pending ipi, its better to send them now,
4133 * not waiting net_rx_action() end.
4135 if (sd->rps_ipi_list) {
4136 local_irq_disable();
4137 net_rps_action_and_irq_enable(sd);
4140 napi->weight = weight_p;
4141 local_irq_disable();
4142 while (work < quota) {
4143 struct sk_buff *skb;
4146 while ((skb = __skb_dequeue(&sd->process_queue))) {
4148 __netif_receive_skb(skb);
4149 local_irq_disable();
4150 input_queue_head_incr(sd);
4151 if (++work >= quota) {
4158 qlen = skb_queue_len(&sd->input_pkt_queue);
4160 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4161 &sd->process_queue);
4163 if (qlen < quota - work) {
4165 * Inline a custom version of __napi_complete().
4166 * only current cpu owns and manipulates this napi,
4167 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4168 * we can use a plain write instead of clear_bit(),
4169 * and we dont need an smp_mb() memory barrier.
4171 list_del(&napi->poll_list);
4174 quota = work + qlen;
4184 * __napi_schedule - schedule for receive
4185 * @n: entry to schedule
4187 * The entry's receive function will be scheduled to run
4189 void __napi_schedule(struct napi_struct *n)
4191 unsigned long flags;
4193 local_irq_save(flags);
4194 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4195 local_irq_restore(flags);
4197 EXPORT_SYMBOL(__napi_schedule);
4199 void __napi_complete(struct napi_struct *n)
4201 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4202 BUG_ON(n->gro_list);
4204 list_del(&n->poll_list);
4205 smp_mb__before_clear_bit();
4206 clear_bit(NAPI_STATE_SCHED, &n->state);
4208 EXPORT_SYMBOL(__napi_complete);
4210 void napi_complete(struct napi_struct *n)
4212 unsigned long flags;
4215 * don't let napi dequeue from the cpu poll list
4216 * just in case its running on a different cpu
4218 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4221 napi_gro_flush(n, false);
4222 local_irq_save(flags);
4224 local_irq_restore(flags);
4226 EXPORT_SYMBOL(napi_complete);
4228 /* must be called under rcu_read_lock(), as we dont take a reference */
4229 struct napi_struct *napi_by_id(unsigned int napi_id)
4231 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4232 struct napi_struct *napi;
4234 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4235 if (napi->napi_id == napi_id)
4240 EXPORT_SYMBOL_GPL(napi_by_id);
4242 void napi_hash_add(struct napi_struct *napi)
4244 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4246 spin_lock(&napi_hash_lock);
4248 /* 0 is not a valid id, we also skip an id that is taken
4249 * we expect both events to be extremely rare
4252 while (!napi->napi_id) {
4253 napi->napi_id = ++napi_gen_id;
4254 if (napi_by_id(napi->napi_id))
4258 hlist_add_head_rcu(&napi->napi_hash_node,
4259 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4261 spin_unlock(&napi_hash_lock);
4264 EXPORT_SYMBOL_GPL(napi_hash_add);
4266 /* Warning : caller is responsible to make sure rcu grace period
4267 * is respected before freeing memory containing @napi
4269 void napi_hash_del(struct napi_struct *napi)
4271 spin_lock(&napi_hash_lock);
4273 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4274 hlist_del_rcu(&napi->napi_hash_node);
4276 spin_unlock(&napi_hash_lock);
4278 EXPORT_SYMBOL_GPL(napi_hash_del);
4280 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4281 int (*poll)(struct napi_struct *, int), int weight)
4283 INIT_LIST_HEAD(&napi->poll_list);
4284 napi->gro_count = 0;
4285 napi->gro_list = NULL;
4288 if (weight > NAPI_POLL_WEIGHT)
4289 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4291 napi->weight = weight;
4292 list_add(&napi->dev_list, &dev->napi_list);
4294 #ifdef CONFIG_NETPOLL
4295 spin_lock_init(&napi->poll_lock);
4296 napi->poll_owner = -1;
4298 set_bit(NAPI_STATE_SCHED, &napi->state);
4300 EXPORT_SYMBOL(netif_napi_add);
4302 void netif_napi_del(struct napi_struct *napi)
4304 list_del_init(&napi->dev_list);
4305 napi_free_frags(napi);
4307 kfree_skb_list(napi->gro_list);
4308 napi->gro_list = NULL;
4309 napi->gro_count = 0;
4311 EXPORT_SYMBOL(netif_napi_del);
4313 static void net_rx_action(struct softirq_action *h)
4315 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4316 unsigned long time_limit = jiffies + 2;
4317 int budget = netdev_budget;
4320 local_irq_disable();
4322 while (!list_empty(&sd->poll_list)) {
4323 struct napi_struct *n;
4326 /* If softirq window is exhuasted then punt.
4327 * Allow this to run for 2 jiffies since which will allow
4328 * an average latency of 1.5/HZ.
4330 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4335 /* Even though interrupts have been re-enabled, this
4336 * access is safe because interrupts can only add new
4337 * entries to the tail of this list, and only ->poll()
4338 * calls can remove this head entry from the list.
4340 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4342 have = netpoll_poll_lock(n);
4346 /* This NAPI_STATE_SCHED test is for avoiding a race
4347 * with netpoll's poll_napi(). Only the entity which
4348 * obtains the lock and sees NAPI_STATE_SCHED set will
4349 * actually make the ->poll() call. Therefore we avoid
4350 * accidentally calling ->poll() when NAPI is not scheduled.
4353 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4354 work = n->poll(n, weight);
4358 WARN_ON_ONCE(work > weight);
4362 local_irq_disable();
4364 /* Drivers must not modify the NAPI state if they
4365 * consume the entire weight. In such cases this code
4366 * still "owns" the NAPI instance and therefore can
4367 * move the instance around on the list at-will.
4369 if (unlikely(work == weight)) {
4370 if (unlikely(napi_disable_pending(n))) {
4373 local_irq_disable();
4376 /* flush too old packets
4377 * If HZ < 1000, flush all packets.
4380 napi_gro_flush(n, HZ >= 1000);
4381 local_irq_disable();
4383 list_move_tail(&n->poll_list, &sd->poll_list);
4387 netpoll_poll_unlock(have);
4390 net_rps_action_and_irq_enable(sd);
4392 #ifdef CONFIG_NET_DMA
4394 * There may not be any more sk_buffs coming right now, so push
4395 * any pending DMA copies to hardware
4397 dma_issue_pending_all();
4404 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4408 struct netdev_adjacent {
4409 struct net_device *dev;
4411 /* upper master flag, there can only be one master device per list */
4414 /* counter for the number of times this device was added to us */
4417 /* private field for the users */
4420 struct list_head list;
4421 struct rcu_head rcu;
4424 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4425 struct net_device *adj_dev,
4426 struct list_head *adj_list)
4428 struct netdev_adjacent *adj;
4430 list_for_each_entry(adj, adj_list, list) {
4431 if (adj->dev == adj_dev)
4438 * netdev_has_upper_dev - Check if device is linked to an upper device
4440 * @upper_dev: upper device to check
4442 * Find out if a device is linked to specified upper device and return true
4443 * in case it is. Note that this checks only immediate upper device,
4444 * not through a complete stack of devices. The caller must hold the RTNL lock.
4446 bool netdev_has_upper_dev(struct net_device *dev,
4447 struct net_device *upper_dev)
4451 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4453 EXPORT_SYMBOL(netdev_has_upper_dev);
4456 * netdev_has_any_upper_dev - Check if device is linked to some device
4459 * Find out if a device is linked to an upper device and return true in case
4460 * it is. The caller must hold the RTNL lock.
4462 static bool netdev_has_any_upper_dev(struct net_device *dev)
4466 return !list_empty(&dev->all_adj_list.upper);
4470 * netdev_master_upper_dev_get - Get master upper device
4473 * Find a master upper device and return pointer to it or NULL in case
4474 * it's not there. The caller must hold the RTNL lock.
4476 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4478 struct netdev_adjacent *upper;
4482 if (list_empty(&dev->adj_list.upper))
4485 upper = list_first_entry(&dev->adj_list.upper,
4486 struct netdev_adjacent, list);
4487 if (likely(upper->master))
4491 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4493 void *netdev_adjacent_get_private(struct list_head *adj_list)
4495 struct netdev_adjacent *adj;
4497 adj = list_entry(adj_list, struct netdev_adjacent, list);
4499 return adj->private;
4501 EXPORT_SYMBOL(netdev_adjacent_get_private);
4504 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4506 * @iter: list_head ** of the current position
4508 * Gets the next device from the dev's upper list, starting from iter
4509 * position. The caller must hold RCU read lock.
4511 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4512 struct list_head **iter)
4514 struct netdev_adjacent *upper;
4516 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4518 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4520 if (&upper->list == &dev->all_adj_list.upper)
4523 *iter = &upper->list;
4527 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4530 * netdev_lower_get_next_private - Get the next ->private from the
4531 * lower neighbour list
4533 * @iter: list_head ** of the current position
4535 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4536 * list, starting from iter position. The caller must hold either hold the
4537 * RTNL lock or its own locking that guarantees that the neighbour lower
4538 * list will remain unchainged.
4540 void *netdev_lower_get_next_private(struct net_device *dev,
4541 struct list_head **iter)
4543 struct netdev_adjacent *lower;
4545 lower = list_entry(*iter, struct netdev_adjacent, list);
4547 if (&lower->list == &dev->adj_list.lower)
4551 *iter = lower->list.next;
4553 return lower->private;
4555 EXPORT_SYMBOL(netdev_lower_get_next_private);
4558 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4559 * lower neighbour list, RCU
4562 * @iter: list_head ** of the current position
4564 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4565 * list, starting from iter position. The caller must hold RCU read lock.
4567 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4568 struct list_head **iter)
4570 struct netdev_adjacent *lower;
4572 WARN_ON_ONCE(!rcu_read_lock_held());
4574 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4576 if (&lower->list == &dev->adj_list.lower)
4580 *iter = &lower->list;
4582 return lower->private;
4584 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4587 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4588 * lower neighbour list, RCU
4592 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4593 * list. The caller must hold RCU read lock.
4595 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4597 struct netdev_adjacent *lower;
4599 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4600 struct netdev_adjacent, list);
4602 return lower->private;
4605 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4608 * netdev_master_upper_dev_get_rcu - Get master upper device
4611 * Find a master upper device and return pointer to it or NULL in case
4612 * it's not there. The caller must hold the RCU read lock.
4614 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4616 struct netdev_adjacent *upper;
4618 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4619 struct netdev_adjacent, list);
4620 if (upper && likely(upper->master))
4624 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4626 int netdev_adjacent_sysfs_add(struct net_device *dev,
4627 struct net_device *adj_dev,
4628 struct list_head *dev_list)
4630 char linkname[IFNAMSIZ+7];
4631 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4632 "upper_%s" : "lower_%s", adj_dev->name);
4633 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4636 void netdev_adjacent_sysfs_del(struct net_device *dev,
4638 struct list_head *dev_list)
4640 char linkname[IFNAMSIZ+7];
4641 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4642 "upper_%s" : "lower_%s", name);
4643 sysfs_remove_link(&(dev->dev.kobj), linkname);
4646 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4647 (dev_list == &dev->adj_list.upper || \
4648 dev_list == &dev->adj_list.lower)
4650 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4651 struct net_device *adj_dev,
4652 struct list_head *dev_list,
4653 void *private, bool master)
4655 struct netdev_adjacent *adj;
4658 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4665 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4670 adj->master = master;
4672 adj->private = private;
4675 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4676 adj_dev->name, dev->name, adj_dev->name);
4678 if (netdev_adjacent_is_neigh_list(dev, dev_list)) {
4679 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
4684 /* Ensure that master link is always the first item in list. */
4686 ret = sysfs_create_link(&(dev->dev.kobj),
4687 &(adj_dev->dev.kobj), "master");
4689 goto remove_symlinks;
4691 list_add_rcu(&adj->list, dev_list);
4693 list_add_tail_rcu(&adj->list, dev_list);
4699 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4700 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4708 static void __netdev_adjacent_dev_remove(struct net_device *dev,
4709 struct net_device *adj_dev,
4710 struct list_head *dev_list)
4712 struct netdev_adjacent *adj;
4714 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4717 pr_err("tried to remove device %s from %s\n",
4718 dev->name, adj_dev->name);
4722 if (adj->ref_nr > 1) {
4723 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4730 sysfs_remove_link(&(dev->dev.kobj), "master");
4732 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4733 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4735 list_del_rcu(&adj->list);
4736 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4737 adj_dev->name, dev->name, adj_dev->name);
4739 kfree_rcu(adj, rcu);
4742 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4743 struct net_device *upper_dev,
4744 struct list_head *up_list,
4745 struct list_head *down_list,
4746 void *private, bool master)
4750 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4755 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4758 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4765 static int __netdev_adjacent_dev_link(struct net_device *dev,
4766 struct net_device *upper_dev)
4768 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4769 &dev->all_adj_list.upper,
4770 &upper_dev->all_adj_list.lower,
4774 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4775 struct net_device *upper_dev,
4776 struct list_head *up_list,
4777 struct list_head *down_list)
4779 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4780 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4783 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
4784 struct net_device *upper_dev)
4786 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4787 &dev->all_adj_list.upper,
4788 &upper_dev->all_adj_list.lower);
4791 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4792 struct net_device *upper_dev,
4793 void *private, bool master)
4795 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4800 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4801 &dev->adj_list.upper,
4802 &upper_dev->adj_list.lower,
4805 __netdev_adjacent_dev_unlink(dev, upper_dev);
4812 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4813 struct net_device *upper_dev)
4815 __netdev_adjacent_dev_unlink(dev, upper_dev);
4816 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4817 &dev->adj_list.upper,
4818 &upper_dev->adj_list.lower);
4821 static int __netdev_upper_dev_link(struct net_device *dev,
4822 struct net_device *upper_dev, bool master,
4825 struct netdev_adjacent *i, *j, *to_i, *to_j;
4830 if (dev == upper_dev)
4833 /* To prevent loops, check if dev is not upper device to upper_dev. */
4834 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
4837 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
4840 if (master && netdev_master_upper_dev_get(dev))
4843 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
4848 /* Now that we linked these devs, make all the upper_dev's
4849 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4850 * versa, and don't forget the devices itself. All of these
4851 * links are non-neighbours.
4853 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4854 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4855 pr_debug("Interlinking %s with %s, non-neighbour\n",
4856 i->dev->name, j->dev->name);
4857 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
4863 /* add dev to every upper_dev's upper device */
4864 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4865 pr_debug("linking %s's upper device %s with %s\n",
4866 upper_dev->name, i->dev->name, dev->name);
4867 ret = __netdev_adjacent_dev_link(dev, i->dev);
4869 goto rollback_upper_mesh;
4872 /* add upper_dev to every dev's lower device */
4873 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4874 pr_debug("linking %s's lower device %s with %s\n", dev->name,
4875 i->dev->name, upper_dev->name);
4876 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
4878 goto rollback_lower_mesh;
4881 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4884 rollback_lower_mesh:
4886 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4889 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4894 rollback_upper_mesh:
4896 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4899 __netdev_adjacent_dev_unlink(dev, i->dev);
4907 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4908 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4909 if (i == to_i && j == to_j)
4911 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4917 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4923 * netdev_upper_dev_link - Add a link to the upper device
4925 * @upper_dev: new upper device
4927 * Adds a link to device which is upper to this one. The caller must hold
4928 * the RTNL lock. On a failure a negative errno code is returned.
4929 * On success the reference counts are adjusted and the function
4932 int netdev_upper_dev_link(struct net_device *dev,
4933 struct net_device *upper_dev)
4935 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
4937 EXPORT_SYMBOL(netdev_upper_dev_link);
4940 * netdev_master_upper_dev_link - Add a master link to the upper device
4942 * @upper_dev: new upper device
4944 * Adds a link to device which is upper to this one. In this case, only
4945 * one master upper device can be linked, although other non-master devices
4946 * might be linked as well. The caller must hold the RTNL lock.
4947 * On a failure a negative errno code is returned. On success the reference
4948 * counts are adjusted and the function returns zero.
4950 int netdev_master_upper_dev_link(struct net_device *dev,
4951 struct net_device *upper_dev)
4953 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
4955 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4957 int netdev_master_upper_dev_link_private(struct net_device *dev,
4958 struct net_device *upper_dev,
4961 return __netdev_upper_dev_link(dev, upper_dev, true, private);
4963 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
4966 * netdev_upper_dev_unlink - Removes a link to upper device
4968 * @upper_dev: new upper device
4970 * Removes a link to device which is upper to this one. The caller must hold
4973 void netdev_upper_dev_unlink(struct net_device *dev,
4974 struct net_device *upper_dev)
4976 struct netdev_adjacent *i, *j;
4979 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4981 /* Here is the tricky part. We must remove all dev's lower
4982 * devices from all upper_dev's upper devices and vice
4983 * versa, to maintain the graph relationship.
4985 list_for_each_entry(i, &dev->all_adj_list.lower, list)
4986 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
4987 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4989 /* remove also the devices itself from lower/upper device
4992 list_for_each_entry(i, &dev->all_adj_list.lower, list)
4993 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4995 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
4996 __netdev_adjacent_dev_unlink(dev, i->dev);
4998 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5000 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5002 void *netdev_lower_dev_get_private(struct net_device *dev,
5003 struct net_device *lower_dev)
5005 struct netdev_adjacent *lower;
5009 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5013 return lower->private;
5015 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5017 static void dev_change_rx_flags(struct net_device *dev, int flags)
5019 const struct net_device_ops *ops = dev->netdev_ops;
5021 if (ops->ndo_change_rx_flags)
5022 ops->ndo_change_rx_flags(dev, flags);
5025 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5027 unsigned int old_flags = dev->flags;
5033 dev->flags |= IFF_PROMISC;
5034 dev->promiscuity += inc;
5035 if (dev->promiscuity == 0) {
5038 * If inc causes overflow, untouch promisc and return error.
5041 dev->flags &= ~IFF_PROMISC;
5043 dev->promiscuity -= inc;
5044 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5049 if (dev->flags != old_flags) {
5050 pr_info("device %s %s promiscuous mode\n",
5052 dev->flags & IFF_PROMISC ? "entered" : "left");
5053 if (audit_enabled) {
5054 current_uid_gid(&uid, &gid);
5055 audit_log(current->audit_context, GFP_ATOMIC,
5056 AUDIT_ANOM_PROMISCUOUS,
5057 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5058 dev->name, (dev->flags & IFF_PROMISC),
5059 (old_flags & IFF_PROMISC),
5060 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5061 from_kuid(&init_user_ns, uid),
5062 from_kgid(&init_user_ns, gid),
5063 audit_get_sessionid(current));
5066 dev_change_rx_flags(dev, IFF_PROMISC);
5069 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5074 * dev_set_promiscuity - update promiscuity count on a device
5078 * Add or remove promiscuity from a device. While the count in the device
5079 * remains above zero the interface remains promiscuous. Once it hits zero
5080 * the device reverts back to normal filtering operation. A negative inc
5081 * value is used to drop promiscuity on the device.
5082 * Return 0 if successful or a negative errno code on error.
5084 int dev_set_promiscuity(struct net_device *dev, int inc)
5086 unsigned int old_flags = dev->flags;
5089 err = __dev_set_promiscuity(dev, inc, true);
5092 if (dev->flags != old_flags)
5093 dev_set_rx_mode(dev);
5096 EXPORT_SYMBOL(dev_set_promiscuity);
5098 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5100 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5104 dev->flags |= IFF_ALLMULTI;
5105 dev->allmulti += inc;
5106 if (dev->allmulti == 0) {
5109 * If inc causes overflow, untouch allmulti and return error.
5112 dev->flags &= ~IFF_ALLMULTI;
5114 dev->allmulti -= inc;
5115 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5120 if (dev->flags ^ old_flags) {
5121 dev_change_rx_flags(dev, IFF_ALLMULTI);
5122 dev_set_rx_mode(dev);
5124 __dev_notify_flags(dev, old_flags,
5125 dev->gflags ^ old_gflags);
5131 * dev_set_allmulti - update allmulti count on a device
5135 * Add or remove reception of all multicast frames to a device. While the
5136 * count in the device remains above zero the interface remains listening
5137 * to all interfaces. Once it hits zero the device reverts back to normal
5138 * filtering operation. A negative @inc value is used to drop the counter
5139 * when releasing a resource needing all multicasts.
5140 * Return 0 if successful or a negative errno code on error.
5143 int dev_set_allmulti(struct net_device *dev, int inc)
5145 return __dev_set_allmulti(dev, inc, true);
5147 EXPORT_SYMBOL(dev_set_allmulti);
5150 * Upload unicast and multicast address lists to device and
5151 * configure RX filtering. When the device doesn't support unicast
5152 * filtering it is put in promiscuous mode while unicast addresses
5155 void __dev_set_rx_mode(struct net_device *dev)
5157 const struct net_device_ops *ops = dev->netdev_ops;
5159 /* dev_open will call this function so the list will stay sane. */
5160 if (!(dev->flags&IFF_UP))
5163 if (!netif_device_present(dev))
5166 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5167 /* Unicast addresses changes may only happen under the rtnl,
5168 * therefore calling __dev_set_promiscuity here is safe.
5170 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5171 __dev_set_promiscuity(dev, 1, false);
5172 dev->uc_promisc = true;
5173 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5174 __dev_set_promiscuity(dev, -1, false);
5175 dev->uc_promisc = false;
5179 if (ops->ndo_set_rx_mode)
5180 ops->ndo_set_rx_mode(dev);
5183 void dev_set_rx_mode(struct net_device *dev)
5185 netif_addr_lock_bh(dev);
5186 __dev_set_rx_mode(dev);
5187 netif_addr_unlock_bh(dev);
5191 * dev_get_flags - get flags reported to userspace
5194 * Get the combination of flag bits exported through APIs to userspace.
5196 unsigned int dev_get_flags(const struct net_device *dev)
5200 flags = (dev->flags & ~(IFF_PROMISC |
5205 (dev->gflags & (IFF_PROMISC |
5208 if (netif_running(dev)) {
5209 if (netif_oper_up(dev))
5210 flags |= IFF_RUNNING;
5211 if (netif_carrier_ok(dev))
5212 flags |= IFF_LOWER_UP;
5213 if (netif_dormant(dev))
5214 flags |= IFF_DORMANT;
5219 EXPORT_SYMBOL(dev_get_flags);
5221 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5223 unsigned int old_flags = dev->flags;
5229 * Set the flags on our device.
5232 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5233 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5235 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5239 * Load in the correct multicast list now the flags have changed.
5242 if ((old_flags ^ flags) & IFF_MULTICAST)
5243 dev_change_rx_flags(dev, IFF_MULTICAST);
5245 dev_set_rx_mode(dev);
5248 * Have we downed the interface. We handle IFF_UP ourselves
5249 * according to user attempts to set it, rather than blindly
5254 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
5255 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5258 dev_set_rx_mode(dev);
5261 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5262 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5263 unsigned int old_flags = dev->flags;
5265 dev->gflags ^= IFF_PROMISC;
5267 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5268 if (dev->flags != old_flags)
5269 dev_set_rx_mode(dev);
5272 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5273 is important. Some (broken) drivers set IFF_PROMISC, when
5274 IFF_ALLMULTI is requested not asking us and not reporting.
5276 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5277 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5279 dev->gflags ^= IFF_ALLMULTI;
5280 __dev_set_allmulti(dev, inc, false);
5286 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5287 unsigned int gchanges)
5289 unsigned int changes = dev->flags ^ old_flags;
5292 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5294 if (changes & IFF_UP) {
5295 if (dev->flags & IFF_UP)
5296 call_netdevice_notifiers(NETDEV_UP, dev);
5298 call_netdevice_notifiers(NETDEV_DOWN, dev);
5301 if (dev->flags & IFF_UP &&
5302 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5303 struct netdev_notifier_change_info change_info;
5305 change_info.flags_changed = changes;
5306 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5312 * dev_change_flags - change device settings
5314 * @flags: device state flags
5316 * Change settings on device based state flags. The flags are
5317 * in the userspace exported format.
5319 int dev_change_flags(struct net_device *dev, unsigned int flags)
5322 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5324 ret = __dev_change_flags(dev, flags);
5328 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5329 __dev_notify_flags(dev, old_flags, changes);
5332 EXPORT_SYMBOL(dev_change_flags);
5334 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5336 const struct net_device_ops *ops = dev->netdev_ops;
5338 if (ops->ndo_change_mtu)
5339 return ops->ndo_change_mtu(dev, new_mtu);
5346 * dev_set_mtu - Change maximum transfer unit
5348 * @new_mtu: new transfer unit
5350 * Change the maximum transfer size of the network device.
5352 int dev_set_mtu(struct net_device *dev, int new_mtu)
5356 if (new_mtu == dev->mtu)
5359 /* MTU must be positive. */
5363 if (!netif_device_present(dev))
5366 orig_mtu = dev->mtu;
5367 err = __dev_set_mtu(dev, new_mtu);
5370 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5371 err = notifier_to_errno(err);
5373 /* setting mtu back and notifying everyone again,
5374 * so that they have a chance to revert changes.
5376 __dev_set_mtu(dev, orig_mtu);
5377 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5382 EXPORT_SYMBOL(dev_set_mtu);
5385 * dev_set_group - Change group this device belongs to
5387 * @new_group: group this device should belong to
5389 void dev_set_group(struct net_device *dev, int new_group)
5391 dev->group = new_group;
5393 EXPORT_SYMBOL(dev_set_group);
5396 * dev_set_mac_address - Change Media Access Control Address
5400 * Change the hardware (MAC) address of the device
5402 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5404 const struct net_device_ops *ops = dev->netdev_ops;
5407 if (!ops->ndo_set_mac_address)
5409 if (sa->sa_family != dev->type)
5411 if (!netif_device_present(dev))
5413 err = ops->ndo_set_mac_address(dev, sa);
5416 dev->addr_assign_type = NET_ADDR_SET;
5417 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5418 add_device_randomness(dev->dev_addr, dev->addr_len);
5421 EXPORT_SYMBOL(dev_set_mac_address);
5424 * dev_change_carrier - Change device carrier
5426 * @new_carrier: new value
5428 * Change device carrier
5430 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5432 const struct net_device_ops *ops = dev->netdev_ops;
5434 if (!ops->ndo_change_carrier)
5436 if (!netif_device_present(dev))
5438 return ops->ndo_change_carrier(dev, new_carrier);
5440 EXPORT_SYMBOL(dev_change_carrier);
5443 * dev_get_phys_port_id - Get device physical port ID
5447 * Get device physical port ID
5449 int dev_get_phys_port_id(struct net_device *dev,
5450 struct netdev_phys_port_id *ppid)
5452 const struct net_device_ops *ops = dev->netdev_ops;
5454 if (!ops->ndo_get_phys_port_id)
5456 return ops->ndo_get_phys_port_id(dev, ppid);
5458 EXPORT_SYMBOL(dev_get_phys_port_id);
5461 * dev_new_index - allocate an ifindex
5462 * @net: the applicable net namespace
5464 * Returns a suitable unique value for a new device interface
5465 * number. The caller must hold the rtnl semaphore or the
5466 * dev_base_lock to be sure it remains unique.
5468 static int dev_new_index(struct net *net)
5470 int ifindex = net->ifindex;
5474 if (!__dev_get_by_index(net, ifindex))
5475 return net->ifindex = ifindex;
5479 /* Delayed registration/unregisteration */
5480 static LIST_HEAD(net_todo_list);
5481 static DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5483 static void net_set_todo(struct net_device *dev)
5485 list_add_tail(&dev->todo_list, &net_todo_list);
5486 dev_net(dev)->dev_unreg_count++;
5489 static void rollback_registered_many(struct list_head *head)
5491 struct net_device *dev, *tmp;
5492 LIST_HEAD(close_head);
5494 BUG_ON(dev_boot_phase);
5497 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5498 /* Some devices call without registering
5499 * for initialization unwind. Remove those
5500 * devices and proceed with the remaining.
5502 if (dev->reg_state == NETREG_UNINITIALIZED) {
5503 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5507 list_del(&dev->unreg_list);
5510 dev->dismantle = true;
5511 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5514 /* If device is running, close it first. */
5515 list_for_each_entry(dev, head, unreg_list)
5516 list_add_tail(&dev->close_list, &close_head);
5517 dev_close_many(&close_head);
5519 list_for_each_entry(dev, head, unreg_list) {
5520 /* And unlink it from device chain. */
5521 unlist_netdevice(dev);
5523 dev->reg_state = NETREG_UNREGISTERING;
5528 list_for_each_entry(dev, head, unreg_list) {
5529 /* Shutdown queueing discipline. */
5533 /* Notify protocols, that we are about to destroy
5534 this device. They should clean all the things.
5536 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5538 if (!dev->rtnl_link_ops ||
5539 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5540 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5543 * Flush the unicast and multicast chains
5548 if (dev->netdev_ops->ndo_uninit)
5549 dev->netdev_ops->ndo_uninit(dev);
5551 /* Notifier chain MUST detach us all upper devices. */
5552 WARN_ON(netdev_has_any_upper_dev(dev));
5554 /* Remove entries from kobject tree */
5555 netdev_unregister_kobject(dev);
5557 /* Remove XPS queueing entries */
5558 netif_reset_xps_queues_gt(dev, 0);
5564 list_for_each_entry(dev, head, unreg_list)
5568 static void rollback_registered(struct net_device *dev)
5572 list_add(&dev->unreg_list, &single);
5573 rollback_registered_many(&single);
5577 static netdev_features_t netdev_fix_features(struct net_device *dev,
5578 netdev_features_t features)
5580 /* Fix illegal checksum combinations */
5581 if ((features & NETIF_F_HW_CSUM) &&
5582 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5583 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5584 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5587 /* TSO requires that SG is present as well. */
5588 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5589 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5590 features &= ~NETIF_F_ALL_TSO;
5593 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5594 !(features & NETIF_F_IP_CSUM)) {
5595 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5596 features &= ~NETIF_F_TSO;
5597 features &= ~NETIF_F_TSO_ECN;
5600 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5601 !(features & NETIF_F_IPV6_CSUM)) {
5602 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5603 features &= ~NETIF_F_TSO6;
5606 /* TSO ECN requires that TSO is present as well. */
5607 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5608 features &= ~NETIF_F_TSO_ECN;
5610 /* Software GSO depends on SG. */
5611 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5612 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5613 features &= ~NETIF_F_GSO;
5616 /* UFO needs SG and checksumming */
5617 if (features & NETIF_F_UFO) {
5618 /* maybe split UFO into V4 and V6? */
5619 if (!((features & NETIF_F_GEN_CSUM) ||
5620 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5621 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5623 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5624 features &= ~NETIF_F_UFO;
5627 if (!(features & NETIF_F_SG)) {
5629 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5630 features &= ~NETIF_F_UFO;
5637 int __netdev_update_features(struct net_device *dev)
5639 netdev_features_t features;
5644 features = netdev_get_wanted_features(dev);
5646 if (dev->netdev_ops->ndo_fix_features)
5647 features = dev->netdev_ops->ndo_fix_features(dev, features);
5649 /* driver might be less strict about feature dependencies */
5650 features = netdev_fix_features(dev, features);
5652 if (dev->features == features)
5655 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5656 &dev->features, &features);
5658 if (dev->netdev_ops->ndo_set_features)
5659 err = dev->netdev_ops->ndo_set_features(dev, features);
5661 if (unlikely(err < 0)) {
5663 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5664 err, &features, &dev->features);
5669 dev->features = features;
5675 * netdev_update_features - recalculate device features
5676 * @dev: the device to check
5678 * Recalculate dev->features set and send notifications if it
5679 * has changed. Should be called after driver or hardware dependent
5680 * conditions might have changed that influence the features.
5682 void netdev_update_features(struct net_device *dev)
5684 if (__netdev_update_features(dev))
5685 netdev_features_change(dev);
5687 EXPORT_SYMBOL(netdev_update_features);
5690 * netdev_change_features - recalculate device features
5691 * @dev: the device to check
5693 * Recalculate dev->features set and send notifications even
5694 * if they have not changed. Should be called instead of
5695 * netdev_update_features() if also dev->vlan_features might
5696 * have changed to allow the changes to be propagated to stacked
5699 void netdev_change_features(struct net_device *dev)
5701 __netdev_update_features(dev);
5702 netdev_features_change(dev);
5704 EXPORT_SYMBOL(netdev_change_features);
5707 * netif_stacked_transfer_operstate - transfer operstate
5708 * @rootdev: the root or lower level device to transfer state from
5709 * @dev: the device to transfer operstate to
5711 * Transfer operational state from root to device. This is normally
5712 * called when a stacking relationship exists between the root
5713 * device and the device(a leaf device).
5715 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5716 struct net_device *dev)
5718 if (rootdev->operstate == IF_OPER_DORMANT)
5719 netif_dormant_on(dev);
5721 netif_dormant_off(dev);
5723 if (netif_carrier_ok(rootdev)) {
5724 if (!netif_carrier_ok(dev))
5725 netif_carrier_on(dev);
5727 if (netif_carrier_ok(dev))
5728 netif_carrier_off(dev);
5731 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5734 static int netif_alloc_rx_queues(struct net_device *dev)
5736 unsigned int i, count = dev->num_rx_queues;
5737 struct netdev_rx_queue *rx;
5741 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5747 for (i = 0; i < count; i++)
5753 static void netdev_init_one_queue(struct net_device *dev,
5754 struct netdev_queue *queue, void *_unused)
5756 /* Initialize queue lock */
5757 spin_lock_init(&queue->_xmit_lock);
5758 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5759 queue->xmit_lock_owner = -1;
5760 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5763 dql_init(&queue->dql, HZ);
5767 static void netif_free_tx_queues(struct net_device *dev)
5769 if (is_vmalloc_addr(dev->_tx))
5775 static int netif_alloc_netdev_queues(struct net_device *dev)
5777 unsigned int count = dev->num_tx_queues;
5778 struct netdev_queue *tx;
5779 size_t sz = count * sizeof(*tx);
5781 BUG_ON(count < 1 || count > 0xffff);
5783 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
5791 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5792 spin_lock_init(&dev->tx_global_lock);
5798 * register_netdevice - register a network device
5799 * @dev: device to register
5801 * Take a completed network device structure and add it to the kernel
5802 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5803 * chain. 0 is returned on success. A negative errno code is returned
5804 * on a failure to set up the device, or if the name is a duplicate.
5806 * Callers must hold the rtnl semaphore. You may want
5807 * register_netdev() instead of this.
5810 * The locking appears insufficient to guarantee two parallel registers
5811 * will not get the same name.
5814 int register_netdevice(struct net_device *dev)
5817 struct net *net = dev_net(dev);
5819 BUG_ON(dev_boot_phase);
5824 /* When net_device's are persistent, this will be fatal. */
5825 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5828 spin_lock_init(&dev->addr_list_lock);
5829 netdev_set_addr_lockdep_class(dev);
5833 ret = dev_get_valid_name(net, dev, dev->name);
5837 /* Init, if this function is available */
5838 if (dev->netdev_ops->ndo_init) {
5839 ret = dev->netdev_ops->ndo_init(dev);
5847 if (((dev->hw_features | dev->features) &
5848 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5849 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5850 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5851 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5858 dev->ifindex = dev_new_index(net);
5859 else if (__dev_get_by_index(net, dev->ifindex))
5862 if (dev->iflink == -1)
5863 dev->iflink = dev->ifindex;
5865 /* Transfer changeable features to wanted_features and enable
5866 * software offloads (GSO and GRO).
5868 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5869 dev->features |= NETIF_F_SOFT_FEATURES;
5870 dev->wanted_features = dev->features & dev->hw_features;
5872 if (!(dev->flags & IFF_LOOPBACK)) {
5873 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5876 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5878 dev->vlan_features |= NETIF_F_HIGHDMA;
5880 /* Make NETIF_F_SG inheritable to tunnel devices.
5882 dev->hw_enc_features |= NETIF_F_SG;
5884 /* Make NETIF_F_SG inheritable to MPLS.
5886 dev->mpls_features |= NETIF_F_SG;
5888 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5889 ret = notifier_to_errno(ret);
5893 ret = netdev_register_kobject(dev);
5896 dev->reg_state = NETREG_REGISTERED;
5898 __netdev_update_features(dev);
5901 * Default initial state at registry is that the
5902 * device is present.
5905 set_bit(__LINK_STATE_PRESENT, &dev->state);
5907 linkwatch_init_dev(dev);
5909 dev_init_scheduler(dev);
5911 list_netdevice(dev);
5912 add_device_randomness(dev->dev_addr, dev->addr_len);
5914 /* If the device has permanent device address, driver should
5915 * set dev_addr and also addr_assign_type should be set to
5916 * NET_ADDR_PERM (default value).
5918 if (dev->addr_assign_type == NET_ADDR_PERM)
5919 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5921 /* Notify protocols, that a new device appeared. */
5922 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5923 ret = notifier_to_errno(ret);
5925 rollback_registered(dev);
5926 dev->reg_state = NETREG_UNREGISTERED;
5929 * Prevent userspace races by waiting until the network
5930 * device is fully setup before sending notifications.
5932 if (!dev->rtnl_link_ops ||
5933 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5934 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
5940 if (dev->netdev_ops->ndo_uninit)
5941 dev->netdev_ops->ndo_uninit(dev);
5944 EXPORT_SYMBOL(register_netdevice);
5947 * init_dummy_netdev - init a dummy network device for NAPI
5948 * @dev: device to init
5950 * This takes a network device structure and initialize the minimum
5951 * amount of fields so it can be used to schedule NAPI polls without
5952 * registering a full blown interface. This is to be used by drivers
5953 * that need to tie several hardware interfaces to a single NAPI
5954 * poll scheduler due to HW limitations.
5956 int init_dummy_netdev(struct net_device *dev)
5958 /* Clear everything. Note we don't initialize spinlocks
5959 * are they aren't supposed to be taken by any of the
5960 * NAPI code and this dummy netdev is supposed to be
5961 * only ever used for NAPI polls
5963 memset(dev, 0, sizeof(struct net_device));
5965 /* make sure we BUG if trying to hit standard
5966 * register/unregister code path
5968 dev->reg_state = NETREG_DUMMY;
5970 /* NAPI wants this */
5971 INIT_LIST_HEAD(&dev->napi_list);
5973 /* a dummy interface is started by default */
5974 set_bit(__LINK_STATE_PRESENT, &dev->state);
5975 set_bit(__LINK_STATE_START, &dev->state);
5977 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5978 * because users of this 'device' dont need to change
5984 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5988 * register_netdev - register a network device
5989 * @dev: device to register
5991 * Take a completed network device structure and add it to the kernel
5992 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5993 * chain. 0 is returned on success. A negative errno code is returned
5994 * on a failure to set up the device, or if the name is a duplicate.
5996 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5997 * and expands the device name if you passed a format string to
6000 int register_netdev(struct net_device *dev)
6005 err = register_netdevice(dev);
6009 EXPORT_SYMBOL(register_netdev);
6011 int netdev_refcnt_read(const struct net_device *dev)
6015 for_each_possible_cpu(i)
6016 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6019 EXPORT_SYMBOL(netdev_refcnt_read);
6022 * netdev_wait_allrefs - wait until all references are gone.
6023 * @dev: target net_device
6025 * This is called when unregistering network devices.
6027 * Any protocol or device that holds a reference should register
6028 * for netdevice notification, and cleanup and put back the
6029 * reference if they receive an UNREGISTER event.
6030 * We can get stuck here if buggy protocols don't correctly
6033 static void netdev_wait_allrefs(struct net_device *dev)
6035 unsigned long rebroadcast_time, warning_time;
6038 linkwatch_forget_dev(dev);
6040 rebroadcast_time = warning_time = jiffies;
6041 refcnt = netdev_refcnt_read(dev);
6043 while (refcnt != 0) {
6044 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6047 /* Rebroadcast unregister notification */
6048 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6054 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6055 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6057 /* We must not have linkwatch events
6058 * pending on unregister. If this
6059 * happens, we simply run the queue
6060 * unscheduled, resulting in a noop
6063 linkwatch_run_queue();
6068 rebroadcast_time = jiffies;
6073 refcnt = netdev_refcnt_read(dev);
6075 if (time_after(jiffies, warning_time + 10 * HZ)) {
6076 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6078 warning_time = jiffies;
6087 * register_netdevice(x1);
6088 * register_netdevice(x2);
6090 * unregister_netdevice(y1);
6091 * unregister_netdevice(y2);
6097 * We are invoked by rtnl_unlock().
6098 * This allows us to deal with problems:
6099 * 1) We can delete sysfs objects which invoke hotplug
6100 * without deadlocking with linkwatch via keventd.
6101 * 2) Since we run with the RTNL semaphore not held, we can sleep
6102 * safely in order to wait for the netdev refcnt to drop to zero.
6104 * We must not return until all unregister events added during
6105 * the interval the lock was held have been completed.
6107 void netdev_run_todo(void)
6109 struct list_head list;
6111 /* Snapshot list, allow later requests */
6112 list_replace_init(&net_todo_list, &list);
6117 /* Wait for rcu callbacks to finish before next phase */
6118 if (!list_empty(&list))
6121 while (!list_empty(&list)) {
6122 struct net_device *dev
6123 = list_first_entry(&list, struct net_device, todo_list);
6124 list_del(&dev->todo_list);
6127 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6130 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6131 pr_err("network todo '%s' but state %d\n",
6132 dev->name, dev->reg_state);
6137 dev->reg_state = NETREG_UNREGISTERED;
6139 on_each_cpu(flush_backlog, dev, 1);
6141 netdev_wait_allrefs(dev);
6144 BUG_ON(netdev_refcnt_read(dev));
6145 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6146 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6147 WARN_ON(dev->dn_ptr);
6149 if (dev->destructor)
6150 dev->destructor(dev);
6152 /* Report a network device has been unregistered */
6154 dev_net(dev)->dev_unreg_count--;
6156 wake_up(&netdev_unregistering_wq);
6158 /* Free network device */
6159 kobject_put(&dev->dev.kobj);
6163 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6164 * fields in the same order, with only the type differing.
6166 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6167 const struct net_device_stats *netdev_stats)
6169 #if BITS_PER_LONG == 64
6170 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6171 memcpy(stats64, netdev_stats, sizeof(*stats64));
6173 size_t i, n = sizeof(*stats64) / sizeof(u64);
6174 const unsigned long *src = (const unsigned long *)netdev_stats;
6175 u64 *dst = (u64 *)stats64;
6177 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6178 sizeof(*stats64) / sizeof(u64));
6179 for (i = 0; i < n; i++)
6183 EXPORT_SYMBOL(netdev_stats_to_stats64);
6186 * dev_get_stats - get network device statistics
6187 * @dev: device to get statistics from
6188 * @storage: place to store stats
6190 * Get network statistics from device. Return @storage.
6191 * The device driver may provide its own method by setting
6192 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6193 * otherwise the internal statistics structure is used.
6195 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6196 struct rtnl_link_stats64 *storage)
6198 const struct net_device_ops *ops = dev->netdev_ops;
6200 if (ops->ndo_get_stats64) {
6201 memset(storage, 0, sizeof(*storage));
6202 ops->ndo_get_stats64(dev, storage);
6203 } else if (ops->ndo_get_stats) {
6204 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6206 netdev_stats_to_stats64(storage, &dev->stats);
6208 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6211 EXPORT_SYMBOL(dev_get_stats);
6213 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6215 struct netdev_queue *queue = dev_ingress_queue(dev);
6217 #ifdef CONFIG_NET_CLS_ACT
6220 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6223 netdev_init_one_queue(dev, queue, NULL);
6224 queue->qdisc = &noop_qdisc;
6225 queue->qdisc_sleeping = &noop_qdisc;
6226 rcu_assign_pointer(dev->ingress_queue, queue);
6231 static const struct ethtool_ops default_ethtool_ops;
6233 void netdev_set_default_ethtool_ops(struct net_device *dev,
6234 const struct ethtool_ops *ops)
6236 if (dev->ethtool_ops == &default_ethtool_ops)
6237 dev->ethtool_ops = ops;
6239 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6241 void netdev_freemem(struct net_device *dev)
6243 char *addr = (char *)dev - dev->padded;
6245 if (is_vmalloc_addr(addr))
6252 * alloc_netdev_mqs - allocate network device
6253 * @sizeof_priv: size of private data to allocate space for
6254 * @name: device name format string
6255 * @setup: callback to initialize device
6256 * @txqs: the number of TX subqueues to allocate
6257 * @rxqs: the number of RX subqueues to allocate
6259 * Allocates a struct net_device with private data area for driver use
6260 * and performs basic initialization. Also allocates subquue structs
6261 * for each queue on the device.
6263 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6264 void (*setup)(struct net_device *),
6265 unsigned int txqs, unsigned int rxqs)
6267 struct net_device *dev;
6269 struct net_device *p;
6271 BUG_ON(strlen(name) >= sizeof(dev->name));
6274 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6280 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6285 alloc_size = sizeof(struct net_device);
6287 /* ensure 32-byte alignment of private area */
6288 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6289 alloc_size += sizeof_priv;
6291 /* ensure 32-byte alignment of whole construct */
6292 alloc_size += NETDEV_ALIGN - 1;
6294 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6296 p = vzalloc(alloc_size);
6300 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6301 dev->padded = (char *)dev - (char *)p;
6303 dev->pcpu_refcnt = alloc_percpu(int);
6304 if (!dev->pcpu_refcnt)
6307 if (dev_addr_init(dev))
6313 dev_net_set(dev, &init_net);
6315 dev->gso_max_size = GSO_MAX_SIZE;
6316 dev->gso_max_segs = GSO_MAX_SEGS;
6318 INIT_LIST_HEAD(&dev->napi_list);
6319 INIT_LIST_HEAD(&dev->unreg_list);
6320 INIT_LIST_HEAD(&dev->close_list);
6321 INIT_LIST_HEAD(&dev->link_watch_list);
6322 INIT_LIST_HEAD(&dev->adj_list.upper);
6323 INIT_LIST_HEAD(&dev->adj_list.lower);
6324 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6325 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6326 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6329 dev->num_tx_queues = txqs;
6330 dev->real_num_tx_queues = txqs;
6331 if (netif_alloc_netdev_queues(dev))
6335 dev->num_rx_queues = rxqs;
6336 dev->real_num_rx_queues = rxqs;
6337 if (netif_alloc_rx_queues(dev))
6341 strcpy(dev->name, name);
6342 dev->group = INIT_NETDEV_GROUP;
6343 if (!dev->ethtool_ops)
6344 dev->ethtool_ops = &default_ethtool_ops;
6352 free_percpu(dev->pcpu_refcnt);
6353 netif_free_tx_queues(dev);
6359 netdev_freemem(dev);
6362 EXPORT_SYMBOL(alloc_netdev_mqs);
6365 * free_netdev - free network device
6368 * This function does the last stage of destroying an allocated device
6369 * interface. The reference to the device object is released.
6370 * If this is the last reference then it will be freed.
6372 void free_netdev(struct net_device *dev)
6374 struct napi_struct *p, *n;
6376 release_net(dev_net(dev));
6378 netif_free_tx_queues(dev);
6383 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6385 /* Flush device addresses */
6386 dev_addr_flush(dev);
6388 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6391 free_percpu(dev->pcpu_refcnt);
6392 dev->pcpu_refcnt = NULL;
6394 /* Compatibility with error handling in drivers */
6395 if (dev->reg_state == NETREG_UNINITIALIZED) {
6396 netdev_freemem(dev);
6400 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6401 dev->reg_state = NETREG_RELEASED;
6403 /* will free via device release */
6404 put_device(&dev->dev);
6406 EXPORT_SYMBOL(free_netdev);
6409 * synchronize_net - Synchronize with packet receive processing
6411 * Wait for packets currently being received to be done.
6412 * Does not block later packets from starting.
6414 void synchronize_net(void)
6417 if (rtnl_is_locked())
6418 synchronize_rcu_expedited();
6422 EXPORT_SYMBOL(synchronize_net);
6425 * unregister_netdevice_queue - remove device from the kernel
6429 * This function shuts down a device interface and removes it
6430 * from the kernel tables.
6431 * If head not NULL, device is queued to be unregistered later.
6433 * Callers must hold the rtnl semaphore. You may want
6434 * unregister_netdev() instead of this.
6437 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6442 list_move_tail(&dev->unreg_list, head);
6444 rollback_registered(dev);
6445 /* Finish processing unregister after unlock */
6449 EXPORT_SYMBOL(unregister_netdevice_queue);
6452 * unregister_netdevice_many - unregister many devices
6453 * @head: list of devices
6455 void unregister_netdevice_many(struct list_head *head)
6457 struct net_device *dev;
6459 if (!list_empty(head)) {
6460 rollback_registered_many(head);
6461 list_for_each_entry(dev, head, unreg_list)
6465 EXPORT_SYMBOL(unregister_netdevice_many);
6468 * unregister_netdev - remove device from the kernel
6471 * This function shuts down a device interface and removes it
6472 * from the kernel tables.
6474 * This is just a wrapper for unregister_netdevice that takes
6475 * the rtnl semaphore. In general you want to use this and not
6476 * unregister_netdevice.
6478 void unregister_netdev(struct net_device *dev)
6481 unregister_netdevice(dev);
6484 EXPORT_SYMBOL(unregister_netdev);
6487 * dev_change_net_namespace - move device to different nethost namespace
6489 * @net: network namespace
6490 * @pat: If not NULL name pattern to try if the current device name
6491 * is already taken in the destination network namespace.
6493 * This function shuts down a device interface and moves it
6494 * to a new network namespace. On success 0 is returned, on
6495 * a failure a netagive errno code is returned.
6497 * Callers must hold the rtnl semaphore.
6500 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6506 /* Don't allow namespace local devices to be moved. */
6508 if (dev->features & NETIF_F_NETNS_LOCAL)
6511 /* Ensure the device has been registrered */
6512 if (dev->reg_state != NETREG_REGISTERED)
6515 /* Get out if there is nothing todo */
6517 if (net_eq(dev_net(dev), net))
6520 /* Pick the destination device name, and ensure
6521 * we can use it in the destination network namespace.
6524 if (__dev_get_by_name(net, dev->name)) {
6525 /* We get here if we can't use the current device name */
6528 if (dev_get_valid_name(net, dev, pat) < 0)
6533 * And now a mini version of register_netdevice unregister_netdevice.
6536 /* If device is running close it first. */
6539 /* And unlink it from device chain */
6541 unlist_netdevice(dev);
6545 /* Shutdown queueing discipline. */
6548 /* Notify protocols, that we are about to destroy
6549 this device. They should clean all the things.
6551 Note that dev->reg_state stays at NETREG_REGISTERED.
6552 This is wanted because this way 8021q and macvlan know
6553 the device is just moving and can keep their slaves up.
6555 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6557 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6558 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6561 * Flush the unicast and multicast chains
6566 /* Send a netdev-removed uevent to the old namespace */
6567 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6569 /* Actually switch the network namespace */
6570 dev_net_set(dev, net);
6572 /* If there is an ifindex conflict assign a new one */
6573 if (__dev_get_by_index(net, dev->ifindex)) {
6574 int iflink = (dev->iflink == dev->ifindex);
6575 dev->ifindex = dev_new_index(net);
6577 dev->iflink = dev->ifindex;
6580 /* Send a netdev-add uevent to the new namespace */
6581 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6583 /* Fixup kobjects */
6584 err = device_rename(&dev->dev, dev->name);
6587 /* Add the device back in the hashes */
6588 list_netdevice(dev);
6590 /* Notify protocols, that a new device appeared. */
6591 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6594 * Prevent userspace races by waiting until the network
6595 * device is fully setup before sending notifications.
6597 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6604 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6606 static int dev_cpu_callback(struct notifier_block *nfb,
6607 unsigned long action,
6610 struct sk_buff **list_skb;
6611 struct sk_buff *skb;
6612 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6613 struct softnet_data *sd, *oldsd;
6615 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6618 local_irq_disable();
6619 cpu = smp_processor_id();
6620 sd = &per_cpu(softnet_data, cpu);
6621 oldsd = &per_cpu(softnet_data, oldcpu);
6623 /* Find end of our completion_queue. */
6624 list_skb = &sd->completion_queue;
6626 list_skb = &(*list_skb)->next;
6627 /* Append completion queue from offline CPU. */
6628 *list_skb = oldsd->completion_queue;
6629 oldsd->completion_queue = NULL;
6631 /* Append output queue from offline CPU. */
6632 if (oldsd->output_queue) {
6633 *sd->output_queue_tailp = oldsd->output_queue;
6634 sd->output_queue_tailp = oldsd->output_queue_tailp;
6635 oldsd->output_queue = NULL;
6636 oldsd->output_queue_tailp = &oldsd->output_queue;
6638 /* Append NAPI poll list from offline CPU. */
6639 if (!list_empty(&oldsd->poll_list)) {
6640 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6641 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6644 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6647 /* Process offline CPU's input_pkt_queue */
6648 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6649 netif_rx_internal(skb);
6650 input_queue_head_incr(oldsd);
6652 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6653 netif_rx_internal(skb);
6654 input_queue_head_incr(oldsd);
6662 * netdev_increment_features - increment feature set by one
6663 * @all: current feature set
6664 * @one: new feature set
6665 * @mask: mask feature set
6667 * Computes a new feature set after adding a device with feature set
6668 * @one to the master device with current feature set @all. Will not
6669 * enable anything that is off in @mask. Returns the new feature set.
6671 netdev_features_t netdev_increment_features(netdev_features_t all,
6672 netdev_features_t one, netdev_features_t mask)
6674 if (mask & NETIF_F_GEN_CSUM)
6675 mask |= NETIF_F_ALL_CSUM;
6676 mask |= NETIF_F_VLAN_CHALLENGED;
6678 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6679 all &= one | ~NETIF_F_ALL_FOR_ALL;
6681 /* If one device supports hw checksumming, set for all. */
6682 if (all & NETIF_F_GEN_CSUM)
6683 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6687 EXPORT_SYMBOL(netdev_increment_features);
6689 static struct hlist_head * __net_init netdev_create_hash(void)
6692 struct hlist_head *hash;
6694 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6696 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6697 INIT_HLIST_HEAD(&hash[i]);
6702 /* Initialize per network namespace state */
6703 static int __net_init netdev_init(struct net *net)
6705 if (net != &init_net)
6706 INIT_LIST_HEAD(&net->dev_base_head);
6708 net->dev_name_head = netdev_create_hash();
6709 if (net->dev_name_head == NULL)
6712 net->dev_index_head = netdev_create_hash();
6713 if (net->dev_index_head == NULL)
6719 kfree(net->dev_name_head);
6725 * netdev_drivername - network driver for the device
6726 * @dev: network device
6728 * Determine network driver for device.
6730 const char *netdev_drivername(const struct net_device *dev)
6732 const struct device_driver *driver;
6733 const struct device *parent;
6734 const char *empty = "";
6736 parent = dev->dev.parent;
6740 driver = parent->driver;
6741 if (driver && driver->name)
6742 return driver->name;
6746 static int __netdev_printk(const char *level, const struct net_device *dev,
6747 struct va_format *vaf)
6751 if (dev && dev->dev.parent) {
6752 r = dev_printk_emit(level[1] - '0',
6755 dev_driver_string(dev->dev.parent),
6756 dev_name(dev->dev.parent),
6757 netdev_name(dev), vaf);
6759 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6761 r = printk("%s(NULL net_device): %pV", level, vaf);
6767 int netdev_printk(const char *level, const struct net_device *dev,
6768 const char *format, ...)
6770 struct va_format vaf;
6774 va_start(args, format);
6779 r = __netdev_printk(level, dev, &vaf);
6785 EXPORT_SYMBOL(netdev_printk);
6787 #define define_netdev_printk_level(func, level) \
6788 int func(const struct net_device *dev, const char *fmt, ...) \
6791 struct va_format vaf; \
6794 va_start(args, fmt); \
6799 r = __netdev_printk(level, dev, &vaf); \
6805 EXPORT_SYMBOL(func);
6807 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6808 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6809 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6810 define_netdev_printk_level(netdev_err, KERN_ERR);
6811 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6812 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6813 define_netdev_printk_level(netdev_info, KERN_INFO);
6815 static void __net_exit netdev_exit(struct net *net)
6817 kfree(net->dev_name_head);
6818 kfree(net->dev_index_head);
6821 static struct pernet_operations __net_initdata netdev_net_ops = {
6822 .init = netdev_init,
6823 .exit = netdev_exit,
6826 static void __net_exit default_device_exit(struct net *net)
6828 struct net_device *dev, *aux;
6830 * Push all migratable network devices back to the
6831 * initial network namespace
6834 for_each_netdev_safe(net, dev, aux) {
6836 char fb_name[IFNAMSIZ];
6838 /* Ignore unmoveable devices (i.e. loopback) */
6839 if (dev->features & NETIF_F_NETNS_LOCAL)
6842 /* Leave virtual devices for the generic cleanup */
6843 if (dev->rtnl_link_ops)
6846 /* Push remaining network devices to init_net */
6847 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6848 err = dev_change_net_namespace(dev, &init_net, fb_name);
6850 pr_emerg("%s: failed to move %s to init_net: %d\n",
6851 __func__, dev->name, err);
6858 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
6860 /* Return with the rtnl_lock held when there are no network
6861 * devices unregistering in any network namespace in net_list.
6868 prepare_to_wait(&netdev_unregistering_wq, &wait,
6869 TASK_UNINTERRUPTIBLE);
6870 unregistering = false;
6872 list_for_each_entry(net, net_list, exit_list) {
6873 if (net->dev_unreg_count > 0) {
6874 unregistering = true;
6883 finish_wait(&netdev_unregistering_wq, &wait);
6886 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6888 /* At exit all network devices most be removed from a network
6889 * namespace. Do this in the reverse order of registration.
6890 * Do this across as many network namespaces as possible to
6891 * improve batching efficiency.
6893 struct net_device *dev;
6895 LIST_HEAD(dev_kill_list);
6897 /* To prevent network device cleanup code from dereferencing
6898 * loopback devices or network devices that have been freed
6899 * wait here for all pending unregistrations to complete,
6900 * before unregistring the loopback device and allowing the
6901 * network namespace be freed.
6903 * The netdev todo list containing all network devices
6904 * unregistrations that happen in default_device_exit_batch
6905 * will run in the rtnl_unlock() at the end of
6906 * default_device_exit_batch.
6908 rtnl_lock_unregistering(net_list);
6909 list_for_each_entry(net, net_list, exit_list) {
6910 for_each_netdev_reverse(net, dev) {
6911 if (dev->rtnl_link_ops)
6912 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6914 unregister_netdevice_queue(dev, &dev_kill_list);
6917 unregister_netdevice_many(&dev_kill_list);
6918 list_del(&dev_kill_list);
6922 static struct pernet_operations __net_initdata default_device_ops = {
6923 .exit = default_device_exit,
6924 .exit_batch = default_device_exit_batch,
6928 * Initialize the DEV module. At boot time this walks the device list and
6929 * unhooks any devices that fail to initialise (normally hardware not
6930 * present) and leaves us with a valid list of present and active devices.
6935 * This is called single threaded during boot, so no need
6936 * to take the rtnl semaphore.
6938 static int __init net_dev_init(void)
6940 int i, rc = -ENOMEM;
6942 BUG_ON(!dev_boot_phase);
6944 if (dev_proc_init())
6947 if (netdev_kobject_init())
6950 INIT_LIST_HEAD(&ptype_all);
6951 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6952 INIT_LIST_HEAD(&ptype_base[i]);
6954 INIT_LIST_HEAD(&offload_base);
6956 if (register_pernet_subsys(&netdev_net_ops))
6960 * Initialise the packet receive queues.
6963 for_each_possible_cpu(i) {
6964 struct softnet_data *sd = &per_cpu(softnet_data, i);
6966 memset(sd, 0, sizeof(*sd));
6967 skb_queue_head_init(&sd->input_pkt_queue);
6968 skb_queue_head_init(&sd->process_queue);
6969 sd->completion_queue = NULL;
6970 INIT_LIST_HEAD(&sd->poll_list);
6971 sd->output_queue = NULL;
6972 sd->output_queue_tailp = &sd->output_queue;
6974 sd->csd.func = rps_trigger_softirq;
6980 sd->backlog.poll = process_backlog;
6981 sd->backlog.weight = weight_p;
6982 sd->backlog.gro_list = NULL;
6983 sd->backlog.gro_count = 0;
6985 #ifdef CONFIG_NET_FLOW_LIMIT
6986 sd->flow_limit = NULL;
6992 /* The loopback device is special if any other network devices
6993 * is present in a network namespace the loopback device must
6994 * be present. Since we now dynamically allocate and free the
6995 * loopback device ensure this invariant is maintained by
6996 * keeping the loopback device as the first device on the
6997 * list of network devices. Ensuring the loopback devices
6998 * is the first device that appears and the last network device
7001 if (register_pernet_device(&loopback_net_ops))
7004 if (register_pernet_device(&default_device_ops))
7007 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7008 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7010 hotcpu_notifier(dev_cpu_callback, 0);
7017 subsys_initcall(net_dev_init);