2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/pkt_sched.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include <linux/highmem.h>
106 #include <linux/init.h>
107 #include <linux/module.h>
108 #include <linux/netpoll.h>
109 #include <linux/rcupdate.h>
110 #include <linux/delay.h>
111 #include <net/iw_handler.h>
112 #include <asm/current.h>
113 #include <linux/audit.h>
114 #include <linux/dmaengine.h>
115 #include <linux/err.h>
116 #include <linux/ctype.h>
117 #include <linux/if_arp.h>
118 #include <linux/if_vlan.h>
119 #include <linux/ip.h>
121 #include <linux/ipv6.h>
122 #include <linux/in.h>
123 #include <linux/jhash.h>
124 #include <linux/random.h>
125 #include <trace/events/napi.h>
126 #include <trace/events/net.h>
127 #include <trace/events/skb.h>
128 #include <linux/pci.h>
129 #include <linux/inetdevice.h>
130 #include <linux/cpu_rmap.h>
131 #include <linux/static_key.h>
132 #include <linux/hashtable.h>
133 #include <linux/vmalloc.h>
134 #include <linux/if_macvlan.h>
136 #include "net-sysfs.h"
138 /* Instead of increasing this, you should create a hash table. */
139 #define MAX_GRO_SKBS 8
141 /* This should be increased if a protocol with a bigger head is added. */
142 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 static DEFINE_SPINLOCK(ptype_lock);
145 static DEFINE_SPINLOCK(offload_lock);
146 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
147 struct list_head ptype_all __read_mostly; /* Taps */
148 static struct list_head offload_base __read_mostly;
150 static int netif_rx_internal(struct sk_buff *skb);
153 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
156 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
158 * Writers must hold the rtnl semaphore while they loop through the
159 * dev_base_head list, and hold dev_base_lock for writing when they do the
160 * actual updates. This allows pure readers to access the list even
161 * while a writer is preparing to update it.
163 * To put it another way, dev_base_lock is held for writing only to
164 * protect against pure readers; the rtnl semaphore provides the
165 * protection against other writers.
167 * See, for example usages, register_netdevice() and
168 * unregister_netdevice(), which must be called with the rtnl
171 DEFINE_RWLOCK(dev_base_lock);
172 EXPORT_SYMBOL(dev_base_lock);
174 /* protects napi_hash addition/deletion and napi_gen_id */
175 static DEFINE_SPINLOCK(napi_hash_lock);
177 static unsigned int napi_gen_id;
178 static DEFINE_HASHTABLE(napi_hash, 8);
180 static seqcount_t devnet_rename_seq;
182 static inline void dev_base_seq_inc(struct net *net)
184 while (++net->dev_base_seq == 0);
187 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
189 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
191 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
194 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
196 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
199 static inline void rps_lock(struct softnet_data *sd)
202 spin_lock(&sd->input_pkt_queue.lock);
206 static inline void rps_unlock(struct softnet_data *sd)
209 spin_unlock(&sd->input_pkt_queue.lock);
213 /* Device list insertion */
214 static void list_netdevice(struct net_device *dev)
216 struct net *net = dev_net(dev);
220 write_lock_bh(&dev_base_lock);
221 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
222 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
223 hlist_add_head_rcu(&dev->index_hlist,
224 dev_index_hash(net, dev->ifindex));
225 write_unlock_bh(&dev_base_lock);
227 dev_base_seq_inc(net);
230 /* Device list removal
231 * caller must respect a RCU grace period before freeing/reusing dev
233 static void unlist_netdevice(struct net_device *dev)
237 /* Unlink dev from the device chain */
238 write_lock_bh(&dev_base_lock);
239 list_del_rcu(&dev->dev_list);
240 hlist_del_rcu(&dev->name_hlist);
241 hlist_del_rcu(&dev->index_hlist);
242 write_unlock_bh(&dev_base_lock);
244 dev_base_seq_inc(dev_net(dev));
251 static RAW_NOTIFIER_HEAD(netdev_chain);
254 * Device drivers call our routines to queue packets here. We empty the
255 * queue in the local softnet handler.
258 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
259 EXPORT_PER_CPU_SYMBOL(softnet_data);
261 #ifdef CONFIG_LOCKDEP
263 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
264 * according to dev->type
266 static const unsigned short netdev_lock_type[] =
267 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
268 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
269 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
270 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
271 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
272 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
273 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
274 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
275 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
276 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
277 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
278 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
279 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
280 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
281 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
283 static const char *const netdev_lock_name[] =
284 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
285 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
286 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
287 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
288 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
289 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
290 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
291 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
292 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
293 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
294 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
295 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
296 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
297 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
298 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
300 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
301 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
303 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
307 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
308 if (netdev_lock_type[i] == dev_type)
310 /* the last key is used by default */
311 return ARRAY_SIZE(netdev_lock_type) - 1;
314 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
315 unsigned short dev_type)
319 i = netdev_lock_pos(dev_type);
320 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
321 netdev_lock_name[i]);
324 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
328 i = netdev_lock_pos(dev->type);
329 lockdep_set_class_and_name(&dev->addr_list_lock,
330 &netdev_addr_lock_key[i],
331 netdev_lock_name[i]);
334 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
335 unsigned short dev_type)
338 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
343 /*******************************************************************************
345 Protocol management and registration routines
347 *******************************************************************************/
350 * Add a protocol ID to the list. Now that the input handler is
351 * smarter we can dispense with all the messy stuff that used to be
354 * BEWARE!!! Protocol handlers, mangling input packets,
355 * MUST BE last in hash buckets and checking protocol handlers
356 * MUST start from promiscuous ptype_all chain in net_bh.
357 * It is true now, do not change it.
358 * Explanation follows: if protocol handler, mangling packet, will
359 * be the first on list, it is not able to sense, that packet
360 * is cloned and should be copied-on-write, so that it will
361 * change it and subsequent readers will get broken packet.
365 static inline struct list_head *ptype_head(const struct packet_type *pt)
367 if (pt->type == htons(ETH_P_ALL))
370 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
374 * dev_add_pack - add packet handler
375 * @pt: packet type declaration
377 * Add a protocol handler to the networking stack. The passed &packet_type
378 * is linked into kernel lists and may not be freed until it has been
379 * removed from the kernel lists.
381 * This call does not sleep therefore it can not
382 * guarantee all CPU's that are in middle of receiving packets
383 * will see the new packet type (until the next received packet).
386 void dev_add_pack(struct packet_type *pt)
388 struct list_head *head = ptype_head(pt);
390 spin_lock(&ptype_lock);
391 list_add_rcu(&pt->list, head);
392 spin_unlock(&ptype_lock);
394 EXPORT_SYMBOL(dev_add_pack);
397 * __dev_remove_pack - remove packet handler
398 * @pt: packet type declaration
400 * Remove a protocol handler that was previously added to the kernel
401 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
402 * from the kernel lists and can be freed or reused once this function
405 * The packet type might still be in use by receivers
406 * and must not be freed until after all the CPU's have gone
407 * through a quiescent state.
409 void __dev_remove_pack(struct packet_type *pt)
411 struct list_head *head = ptype_head(pt);
412 struct packet_type *pt1;
414 spin_lock(&ptype_lock);
416 list_for_each_entry(pt1, head, list) {
418 list_del_rcu(&pt->list);
423 pr_warn("dev_remove_pack: %p not found\n", pt);
425 spin_unlock(&ptype_lock);
427 EXPORT_SYMBOL(__dev_remove_pack);
430 * dev_remove_pack - remove packet handler
431 * @pt: packet type declaration
433 * Remove a protocol handler that was previously added to the kernel
434 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
435 * from the kernel lists and can be freed or reused once this function
438 * This call sleeps to guarantee that no CPU is looking at the packet
441 void dev_remove_pack(struct packet_type *pt)
443 __dev_remove_pack(pt);
447 EXPORT_SYMBOL(dev_remove_pack);
451 * dev_add_offload - register offload handlers
452 * @po: protocol offload declaration
454 * Add protocol offload handlers to the networking stack. The passed
455 * &proto_offload is linked into kernel lists and may not be freed until
456 * it has been removed from the kernel lists.
458 * This call does not sleep therefore it can not
459 * guarantee all CPU's that are in middle of receiving packets
460 * will see the new offload handlers (until the next received packet).
462 void dev_add_offload(struct packet_offload *po)
464 struct list_head *head = &offload_base;
466 spin_lock(&offload_lock);
467 list_add_rcu(&po->list, head);
468 spin_unlock(&offload_lock);
470 EXPORT_SYMBOL(dev_add_offload);
473 * __dev_remove_offload - remove offload handler
474 * @po: packet offload declaration
476 * Remove a protocol offload handler that was previously added to the
477 * kernel offload handlers by dev_add_offload(). The passed &offload_type
478 * is removed from the kernel lists and can be freed or reused once this
481 * The packet type might still be in use by receivers
482 * and must not be freed until after all the CPU's have gone
483 * through a quiescent state.
485 static void __dev_remove_offload(struct packet_offload *po)
487 struct list_head *head = &offload_base;
488 struct packet_offload *po1;
490 spin_lock(&offload_lock);
492 list_for_each_entry(po1, head, list) {
494 list_del_rcu(&po->list);
499 pr_warn("dev_remove_offload: %p not found\n", po);
501 spin_unlock(&offload_lock);
505 * dev_remove_offload - remove packet offload handler
506 * @po: packet offload declaration
508 * Remove a packet offload handler that was previously added to the kernel
509 * offload handlers by dev_add_offload(). The passed &offload_type is
510 * removed from the kernel lists and can be freed or reused once this
513 * This call sleeps to guarantee that no CPU is looking at the packet
516 void dev_remove_offload(struct packet_offload *po)
518 __dev_remove_offload(po);
522 EXPORT_SYMBOL(dev_remove_offload);
524 /******************************************************************************
526 Device Boot-time Settings Routines
528 *******************************************************************************/
530 /* Boot time configuration table */
531 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
534 * netdev_boot_setup_add - add new setup entry
535 * @name: name of the device
536 * @map: configured settings for the device
538 * Adds new setup entry to the dev_boot_setup list. The function
539 * returns 0 on error and 1 on success. This is a generic routine to
542 static int netdev_boot_setup_add(char *name, struct ifmap *map)
544 struct netdev_boot_setup *s;
548 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
549 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
550 memset(s[i].name, 0, sizeof(s[i].name));
551 strlcpy(s[i].name, name, IFNAMSIZ);
552 memcpy(&s[i].map, map, sizeof(s[i].map));
557 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
561 * netdev_boot_setup_check - check boot time settings
562 * @dev: the netdevice
564 * Check boot time settings for the device.
565 * The found settings are set for the device to be used
566 * later in the device probing.
567 * Returns 0 if no settings found, 1 if they are.
569 int netdev_boot_setup_check(struct net_device *dev)
571 struct netdev_boot_setup *s = dev_boot_setup;
574 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
575 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
576 !strcmp(dev->name, s[i].name)) {
577 dev->irq = s[i].map.irq;
578 dev->base_addr = s[i].map.base_addr;
579 dev->mem_start = s[i].map.mem_start;
580 dev->mem_end = s[i].map.mem_end;
586 EXPORT_SYMBOL(netdev_boot_setup_check);
590 * netdev_boot_base - get address from boot time settings
591 * @prefix: prefix for network device
592 * @unit: id for network device
594 * Check boot time settings for the base address of device.
595 * The found settings are set for the device to be used
596 * later in the device probing.
597 * Returns 0 if no settings found.
599 unsigned long netdev_boot_base(const char *prefix, int unit)
601 const struct netdev_boot_setup *s = dev_boot_setup;
605 sprintf(name, "%s%d", prefix, unit);
608 * If device already registered then return base of 1
609 * to indicate not to probe for this interface
611 if (__dev_get_by_name(&init_net, name))
614 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
615 if (!strcmp(name, s[i].name))
616 return s[i].map.base_addr;
621 * Saves at boot time configured settings for any netdevice.
623 int __init netdev_boot_setup(char *str)
628 str = get_options(str, ARRAY_SIZE(ints), ints);
633 memset(&map, 0, sizeof(map));
637 map.base_addr = ints[2];
639 map.mem_start = ints[3];
641 map.mem_end = ints[4];
643 /* Add new entry to the list */
644 return netdev_boot_setup_add(str, &map);
647 __setup("netdev=", netdev_boot_setup);
649 /*******************************************************************************
651 Device Interface Subroutines
653 *******************************************************************************/
656 * __dev_get_by_name - find a device by its name
657 * @net: the applicable net namespace
658 * @name: name to find
660 * Find an interface by name. Must be called under RTNL semaphore
661 * or @dev_base_lock. If the name is found a pointer to the device
662 * is returned. If the name is not found then %NULL is returned. The
663 * reference counters are not incremented so the caller must be
664 * careful with locks.
667 struct net_device *__dev_get_by_name(struct net *net, const char *name)
669 struct net_device *dev;
670 struct hlist_head *head = dev_name_hash(net, name);
672 hlist_for_each_entry(dev, head, name_hlist)
673 if (!strncmp(dev->name, name, IFNAMSIZ))
678 EXPORT_SYMBOL(__dev_get_by_name);
681 * dev_get_by_name_rcu - find a device by its name
682 * @net: the applicable net namespace
683 * @name: name to find
685 * Find an interface by name.
686 * If the name is found a pointer to the device is returned.
687 * If the name is not found then %NULL is returned.
688 * The reference counters are not incremented so the caller must be
689 * careful with locks. The caller must hold RCU lock.
692 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
694 struct net_device *dev;
695 struct hlist_head *head = dev_name_hash(net, name);
697 hlist_for_each_entry_rcu(dev, head, name_hlist)
698 if (!strncmp(dev->name, name, IFNAMSIZ))
703 EXPORT_SYMBOL(dev_get_by_name_rcu);
706 * dev_get_by_name - find a device by its name
707 * @net: the applicable net namespace
708 * @name: name to find
710 * Find an interface by name. This can be called from any
711 * context and does its own locking. The returned handle has
712 * the usage count incremented and the caller must use dev_put() to
713 * release it when it is no longer needed. %NULL is returned if no
714 * matching device is found.
717 struct net_device *dev_get_by_name(struct net *net, const char *name)
719 struct net_device *dev;
722 dev = dev_get_by_name_rcu(net, name);
728 EXPORT_SYMBOL(dev_get_by_name);
731 * __dev_get_by_index - find a device by its ifindex
732 * @net: the applicable net namespace
733 * @ifindex: index of device
735 * Search for an interface by index. Returns %NULL if the device
736 * is not found or a pointer to the device. The device has not
737 * had its reference counter increased so the caller must be careful
738 * about locking. The caller must hold either the RTNL semaphore
742 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
744 struct net_device *dev;
745 struct hlist_head *head = dev_index_hash(net, ifindex);
747 hlist_for_each_entry(dev, head, index_hlist)
748 if (dev->ifindex == ifindex)
753 EXPORT_SYMBOL(__dev_get_by_index);
756 * dev_get_by_index_rcu - find a device by its ifindex
757 * @net: the applicable net namespace
758 * @ifindex: index of device
760 * Search for an interface by index. Returns %NULL if the device
761 * is not found or a pointer to the device. The device has not
762 * had its reference counter increased so the caller must be careful
763 * about locking. The caller must hold RCU lock.
766 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
768 struct net_device *dev;
769 struct hlist_head *head = dev_index_hash(net, ifindex);
771 hlist_for_each_entry_rcu(dev, head, index_hlist)
772 if (dev->ifindex == ifindex)
777 EXPORT_SYMBOL(dev_get_by_index_rcu);
781 * dev_get_by_index - find a device by its ifindex
782 * @net: the applicable net namespace
783 * @ifindex: index of device
785 * Search for an interface by index. Returns NULL if the device
786 * is not found or a pointer to the device. The device returned has
787 * had a reference added and the pointer is safe until the user calls
788 * dev_put to indicate they have finished with it.
791 struct net_device *dev_get_by_index(struct net *net, int ifindex)
793 struct net_device *dev;
796 dev = dev_get_by_index_rcu(net, ifindex);
802 EXPORT_SYMBOL(dev_get_by_index);
805 * netdev_get_name - get a netdevice name, knowing its ifindex.
806 * @net: network namespace
807 * @name: a pointer to the buffer where the name will be stored.
808 * @ifindex: the ifindex of the interface to get the name from.
810 * The use of raw_seqcount_begin() and cond_resched() before
811 * retrying is required as we want to give the writers a chance
812 * to complete when CONFIG_PREEMPT is not set.
814 int netdev_get_name(struct net *net, char *name, int ifindex)
816 struct net_device *dev;
820 seq = raw_seqcount_begin(&devnet_rename_seq);
822 dev = dev_get_by_index_rcu(net, ifindex);
828 strcpy(name, dev->name);
830 if (read_seqcount_retry(&devnet_rename_seq, seq)) {
839 * dev_getbyhwaddr_rcu - find a device by its hardware address
840 * @net: the applicable net namespace
841 * @type: media type of device
842 * @ha: hardware address
844 * Search for an interface by MAC address. Returns NULL if the device
845 * is not found or a pointer to the device.
846 * The caller must hold RCU or RTNL.
847 * The returned device has not had its ref count increased
848 * and the caller must therefore be careful about locking
852 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
855 struct net_device *dev;
857 for_each_netdev_rcu(net, dev)
858 if (dev->type == type &&
859 !memcmp(dev->dev_addr, ha, dev->addr_len))
864 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
866 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
868 struct net_device *dev;
871 for_each_netdev(net, dev)
872 if (dev->type == type)
877 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
879 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
881 struct net_device *dev, *ret = NULL;
884 for_each_netdev_rcu(net, dev)
885 if (dev->type == type) {
893 EXPORT_SYMBOL(dev_getfirstbyhwtype);
896 * dev_get_by_flags_rcu - find any device with given flags
897 * @net: the applicable net namespace
898 * @if_flags: IFF_* values
899 * @mask: bitmask of bits in if_flags to check
901 * Search for any interface with the given flags. Returns NULL if a device
902 * is not found or a pointer to the device. Must be called inside
903 * rcu_read_lock(), and result refcount is unchanged.
906 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
909 struct net_device *dev, *ret;
912 for_each_netdev_rcu(net, dev) {
913 if (((dev->flags ^ if_flags) & mask) == 0) {
920 EXPORT_SYMBOL(dev_get_by_flags_rcu);
923 * dev_valid_name - check if name is okay for network device
926 * Network device names need to be valid file names to
927 * to allow sysfs to work. We also disallow any kind of
930 bool dev_valid_name(const char *name)
934 if (strlen(name) >= IFNAMSIZ)
936 if (!strcmp(name, ".") || !strcmp(name, ".."))
940 if (*name == '/' || isspace(*name))
946 EXPORT_SYMBOL(dev_valid_name);
949 * __dev_alloc_name - allocate a name for a device
950 * @net: network namespace to allocate the device name in
951 * @name: name format string
952 * @buf: scratch buffer and result name string
954 * Passed a format string - eg "lt%d" it will try and find a suitable
955 * id. It scans list of devices to build up a free map, then chooses
956 * the first empty slot. The caller must hold the dev_base or rtnl lock
957 * while allocating the name and adding the device in order to avoid
959 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
960 * Returns the number of the unit assigned or a negative errno code.
963 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
967 const int max_netdevices = 8*PAGE_SIZE;
968 unsigned long *inuse;
969 struct net_device *d;
971 p = strnchr(name, IFNAMSIZ-1, '%');
974 * Verify the string as this thing may have come from
975 * the user. There must be either one "%d" and no other "%"
978 if (p[1] != 'd' || strchr(p + 2, '%'))
981 /* Use one page as a bit array of possible slots */
982 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
986 for_each_netdev(net, d) {
987 if (!sscanf(d->name, name, &i))
989 if (i < 0 || i >= max_netdevices)
992 /* avoid cases where sscanf is not exact inverse of printf */
993 snprintf(buf, IFNAMSIZ, name, i);
994 if (!strncmp(buf, d->name, IFNAMSIZ))
998 i = find_first_zero_bit(inuse, max_netdevices);
999 free_page((unsigned long) inuse);
1003 snprintf(buf, IFNAMSIZ, name, i);
1004 if (!__dev_get_by_name(net, buf))
1007 /* It is possible to run out of possible slots
1008 * when the name is long and there isn't enough space left
1009 * for the digits, or if all bits are used.
1015 * dev_alloc_name - allocate a name for a device
1017 * @name: name format string
1019 * Passed a format string - eg "lt%d" it will try and find a suitable
1020 * id. It scans list of devices to build up a free map, then chooses
1021 * the first empty slot. The caller must hold the dev_base or rtnl lock
1022 * while allocating the name and adding the device in order to avoid
1024 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1025 * Returns the number of the unit assigned or a negative errno code.
1028 int dev_alloc_name(struct net_device *dev, const char *name)
1034 BUG_ON(!dev_net(dev));
1036 ret = __dev_alloc_name(net, name, buf);
1038 strlcpy(dev->name, buf, IFNAMSIZ);
1041 EXPORT_SYMBOL(dev_alloc_name);
1043 static int dev_alloc_name_ns(struct net *net,
1044 struct net_device *dev,
1050 ret = __dev_alloc_name(net, name, buf);
1052 strlcpy(dev->name, buf, IFNAMSIZ);
1056 static int dev_get_valid_name(struct net *net,
1057 struct net_device *dev,
1062 if (!dev_valid_name(name))
1065 if (strchr(name, '%'))
1066 return dev_alloc_name_ns(net, dev, name);
1067 else if (__dev_get_by_name(net, name))
1069 else if (dev->name != name)
1070 strlcpy(dev->name, name, IFNAMSIZ);
1076 * dev_change_name - change name of a device
1078 * @newname: name (or format string) must be at least IFNAMSIZ
1080 * Change name of a device, can pass format strings "eth%d".
1083 int dev_change_name(struct net_device *dev, const char *newname)
1085 char oldname[IFNAMSIZ];
1091 BUG_ON(!dev_net(dev));
1094 if (dev->flags & IFF_UP)
1097 write_seqcount_begin(&devnet_rename_seq);
1099 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1100 write_seqcount_end(&devnet_rename_seq);
1104 memcpy(oldname, dev->name, IFNAMSIZ);
1106 err = dev_get_valid_name(net, dev, newname);
1108 write_seqcount_end(&devnet_rename_seq);
1113 ret = device_rename(&dev->dev, dev->name);
1115 memcpy(dev->name, oldname, IFNAMSIZ);
1116 write_seqcount_end(&devnet_rename_seq);
1120 write_seqcount_end(&devnet_rename_seq);
1122 netdev_adjacent_rename_links(dev, oldname);
1124 write_lock_bh(&dev_base_lock);
1125 hlist_del_rcu(&dev->name_hlist);
1126 write_unlock_bh(&dev_base_lock);
1130 write_lock_bh(&dev_base_lock);
1131 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1132 write_unlock_bh(&dev_base_lock);
1134 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1135 ret = notifier_to_errno(ret);
1138 /* err >= 0 after dev_alloc_name() or stores the first errno */
1141 write_seqcount_begin(&devnet_rename_seq);
1142 memcpy(dev->name, oldname, IFNAMSIZ);
1143 memcpy(oldname, newname, IFNAMSIZ);
1146 pr_err("%s: name change rollback failed: %d\n",
1155 * dev_set_alias - change ifalias of a device
1157 * @alias: name up to IFALIASZ
1158 * @len: limit of bytes to copy from info
1160 * Set ifalias for a device,
1162 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1168 if (len >= IFALIASZ)
1172 kfree(dev->ifalias);
1173 dev->ifalias = NULL;
1177 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1180 dev->ifalias = new_ifalias;
1182 strlcpy(dev->ifalias, alias, len+1);
1188 * netdev_features_change - device changes features
1189 * @dev: device to cause notification
1191 * Called to indicate a device has changed features.
1193 void netdev_features_change(struct net_device *dev)
1195 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1197 EXPORT_SYMBOL(netdev_features_change);
1200 * netdev_state_change - device changes state
1201 * @dev: device to cause notification
1203 * Called to indicate a device has changed state. This function calls
1204 * the notifier chains for netdev_chain and sends a NEWLINK message
1205 * to the routing socket.
1207 void netdev_state_change(struct net_device *dev)
1209 if (dev->flags & IFF_UP) {
1210 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1211 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1214 EXPORT_SYMBOL(netdev_state_change);
1217 * netdev_notify_peers - notify network peers about existence of @dev
1218 * @dev: network device
1220 * Generate traffic such that interested network peers are aware of
1221 * @dev, such as by generating a gratuitous ARP. This may be used when
1222 * a device wants to inform the rest of the network about some sort of
1223 * reconfiguration such as a failover event or virtual machine
1226 void netdev_notify_peers(struct net_device *dev)
1229 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1232 EXPORT_SYMBOL(netdev_notify_peers);
1234 static int __dev_open(struct net_device *dev)
1236 const struct net_device_ops *ops = dev->netdev_ops;
1241 if (!netif_device_present(dev))
1244 /* Block netpoll from trying to do any rx path servicing.
1245 * If we don't do this there is a chance ndo_poll_controller
1246 * or ndo_poll may be running while we open the device
1248 netpoll_rx_disable(dev);
1250 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1251 ret = notifier_to_errno(ret);
1255 set_bit(__LINK_STATE_START, &dev->state);
1257 if (ops->ndo_validate_addr)
1258 ret = ops->ndo_validate_addr(dev);
1260 if (!ret && ops->ndo_open)
1261 ret = ops->ndo_open(dev);
1263 netpoll_rx_enable(dev);
1266 clear_bit(__LINK_STATE_START, &dev->state);
1268 dev->flags |= IFF_UP;
1269 net_dmaengine_get();
1270 dev_set_rx_mode(dev);
1272 add_device_randomness(dev->dev_addr, dev->addr_len);
1279 * dev_open - prepare an interface for use.
1280 * @dev: device to open
1282 * Takes a device from down to up state. The device's private open
1283 * function is invoked and then the multicast lists are loaded. Finally
1284 * the device is moved into the up state and a %NETDEV_UP message is
1285 * sent to the netdev notifier chain.
1287 * Calling this function on an active interface is a nop. On a failure
1288 * a negative errno code is returned.
1290 int dev_open(struct net_device *dev)
1294 if (dev->flags & IFF_UP)
1297 ret = __dev_open(dev);
1301 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1302 call_netdevice_notifiers(NETDEV_UP, dev);
1306 EXPORT_SYMBOL(dev_open);
1308 static int __dev_close_many(struct list_head *head)
1310 struct net_device *dev;
1315 list_for_each_entry(dev, head, close_list) {
1316 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1318 clear_bit(__LINK_STATE_START, &dev->state);
1320 /* Synchronize to scheduled poll. We cannot touch poll list, it
1321 * can be even on different cpu. So just clear netif_running().
1323 * dev->stop() will invoke napi_disable() on all of it's
1324 * napi_struct instances on this device.
1326 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1329 dev_deactivate_many(head);
1331 list_for_each_entry(dev, head, close_list) {
1332 const struct net_device_ops *ops = dev->netdev_ops;
1335 * Call the device specific close. This cannot fail.
1336 * Only if device is UP
1338 * We allow it to be called even after a DETACH hot-plug
1344 dev->flags &= ~IFF_UP;
1345 net_dmaengine_put();
1351 static int __dev_close(struct net_device *dev)
1356 /* Temporarily disable netpoll until the interface is down */
1357 netpoll_rx_disable(dev);
1359 list_add(&dev->close_list, &single);
1360 retval = __dev_close_many(&single);
1363 netpoll_rx_enable(dev);
1367 static int dev_close_many(struct list_head *head)
1369 struct net_device *dev, *tmp;
1371 /* Remove the devices that don't need to be closed */
1372 list_for_each_entry_safe(dev, tmp, head, close_list)
1373 if (!(dev->flags & IFF_UP))
1374 list_del_init(&dev->close_list);
1376 __dev_close_many(head);
1378 list_for_each_entry_safe(dev, tmp, head, close_list) {
1379 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1380 call_netdevice_notifiers(NETDEV_DOWN, dev);
1381 list_del_init(&dev->close_list);
1388 * dev_close - shutdown an interface.
1389 * @dev: device to shutdown
1391 * This function moves an active device into down state. A
1392 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1393 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1396 int dev_close(struct net_device *dev)
1398 if (dev->flags & IFF_UP) {
1401 /* Block netpoll rx while the interface is going down */
1402 netpoll_rx_disable(dev);
1404 list_add(&dev->close_list, &single);
1405 dev_close_many(&single);
1408 netpoll_rx_enable(dev);
1412 EXPORT_SYMBOL(dev_close);
1416 * dev_disable_lro - disable Large Receive Offload on a device
1419 * Disable Large Receive Offload (LRO) on a net device. Must be
1420 * called under RTNL. This is needed if received packets may be
1421 * forwarded to another interface.
1423 void dev_disable_lro(struct net_device *dev)
1426 * If we're trying to disable lro on a vlan device
1427 * use the underlying physical device instead
1429 if (is_vlan_dev(dev))
1430 dev = vlan_dev_real_dev(dev);
1432 /* the same for macvlan devices */
1433 if (netif_is_macvlan(dev))
1434 dev = macvlan_dev_real_dev(dev);
1436 dev->wanted_features &= ~NETIF_F_LRO;
1437 netdev_update_features(dev);
1439 if (unlikely(dev->features & NETIF_F_LRO))
1440 netdev_WARN(dev, "failed to disable LRO!\n");
1442 EXPORT_SYMBOL(dev_disable_lro);
1444 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1445 struct net_device *dev)
1447 struct netdev_notifier_info info;
1449 netdev_notifier_info_init(&info, dev);
1450 return nb->notifier_call(nb, val, &info);
1453 static int dev_boot_phase = 1;
1456 * register_netdevice_notifier - register a network notifier block
1459 * Register a notifier to be called when network device events occur.
1460 * The notifier passed is linked into the kernel structures and must
1461 * not be reused until it has been unregistered. A negative errno code
1462 * is returned on a failure.
1464 * When registered all registration and up events are replayed
1465 * to the new notifier to allow device to have a race free
1466 * view of the network device list.
1469 int register_netdevice_notifier(struct notifier_block *nb)
1471 struct net_device *dev;
1472 struct net_device *last;
1477 err = raw_notifier_chain_register(&netdev_chain, nb);
1483 for_each_netdev(net, dev) {
1484 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1485 err = notifier_to_errno(err);
1489 if (!(dev->flags & IFF_UP))
1492 call_netdevice_notifier(nb, NETDEV_UP, dev);
1503 for_each_netdev(net, dev) {
1507 if (dev->flags & IFF_UP) {
1508 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1510 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1512 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1517 raw_notifier_chain_unregister(&netdev_chain, nb);
1520 EXPORT_SYMBOL(register_netdevice_notifier);
1523 * unregister_netdevice_notifier - unregister a network notifier block
1526 * Unregister a notifier previously registered by
1527 * register_netdevice_notifier(). The notifier is unlinked into the
1528 * kernel structures and may then be reused. A negative errno code
1529 * is returned on a failure.
1531 * After unregistering unregister and down device events are synthesized
1532 * for all devices on the device list to the removed notifier to remove
1533 * the need for special case cleanup code.
1536 int unregister_netdevice_notifier(struct notifier_block *nb)
1538 struct net_device *dev;
1543 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1548 for_each_netdev(net, dev) {
1549 if (dev->flags & IFF_UP) {
1550 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1552 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1554 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1561 EXPORT_SYMBOL(unregister_netdevice_notifier);
1564 * call_netdevice_notifiers_info - call all network notifier blocks
1565 * @val: value passed unmodified to notifier function
1566 * @dev: net_device pointer passed unmodified to notifier function
1567 * @info: notifier information data
1569 * Call all network notifier blocks. Parameters and return value
1570 * are as for raw_notifier_call_chain().
1573 static int call_netdevice_notifiers_info(unsigned long val,
1574 struct net_device *dev,
1575 struct netdev_notifier_info *info)
1578 netdev_notifier_info_init(info, dev);
1579 return raw_notifier_call_chain(&netdev_chain, val, info);
1583 * call_netdevice_notifiers - call all network notifier blocks
1584 * @val: value passed unmodified to notifier function
1585 * @dev: net_device pointer passed unmodified to notifier function
1587 * Call all network notifier blocks. Parameters and return value
1588 * are as for raw_notifier_call_chain().
1591 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1593 struct netdev_notifier_info info;
1595 return call_netdevice_notifiers_info(val, dev, &info);
1597 EXPORT_SYMBOL(call_netdevice_notifiers);
1599 static struct static_key netstamp_needed __read_mostly;
1600 #ifdef HAVE_JUMP_LABEL
1601 /* We are not allowed to call static_key_slow_dec() from irq context
1602 * If net_disable_timestamp() is called from irq context, defer the
1603 * static_key_slow_dec() calls.
1605 static atomic_t netstamp_needed_deferred;
1608 void net_enable_timestamp(void)
1610 #ifdef HAVE_JUMP_LABEL
1611 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1615 static_key_slow_dec(&netstamp_needed);
1619 static_key_slow_inc(&netstamp_needed);
1621 EXPORT_SYMBOL(net_enable_timestamp);
1623 void net_disable_timestamp(void)
1625 #ifdef HAVE_JUMP_LABEL
1626 if (in_interrupt()) {
1627 atomic_inc(&netstamp_needed_deferred);
1631 static_key_slow_dec(&netstamp_needed);
1633 EXPORT_SYMBOL(net_disable_timestamp);
1635 static inline void net_timestamp_set(struct sk_buff *skb)
1637 skb->tstamp.tv64 = 0;
1638 if (static_key_false(&netstamp_needed))
1639 __net_timestamp(skb);
1642 #define net_timestamp_check(COND, SKB) \
1643 if (static_key_false(&netstamp_needed)) { \
1644 if ((COND) && !(SKB)->tstamp.tv64) \
1645 __net_timestamp(SKB); \
1648 static inline bool is_skb_forwardable(struct net_device *dev,
1649 struct sk_buff *skb)
1653 if (!(dev->flags & IFF_UP))
1656 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1657 if (skb->len <= len)
1660 /* if TSO is enabled, we don't care about the length as the packet
1661 * could be forwarded without being segmented before
1663 if (skb_is_gso(skb))
1670 * dev_forward_skb - loopback an skb to another netif
1672 * @dev: destination network device
1673 * @skb: buffer to forward
1676 * NET_RX_SUCCESS (no congestion)
1677 * NET_RX_DROP (packet was dropped, but freed)
1679 * dev_forward_skb can be used for injecting an skb from the
1680 * start_xmit function of one device into the receive queue
1681 * of another device.
1683 * The receiving device may be in another namespace, so
1684 * we have to clear all information in the skb that could
1685 * impact namespace isolation.
1687 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1689 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1690 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1691 atomic_long_inc(&dev->rx_dropped);
1697 if (unlikely(!is_skb_forwardable(dev, skb))) {
1698 atomic_long_inc(&dev->rx_dropped);
1703 skb_scrub_packet(skb, true);
1704 skb->protocol = eth_type_trans(skb, dev);
1706 return netif_rx_internal(skb);
1708 EXPORT_SYMBOL_GPL(dev_forward_skb);
1710 static inline int deliver_skb(struct sk_buff *skb,
1711 struct packet_type *pt_prev,
1712 struct net_device *orig_dev)
1714 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1716 atomic_inc(&skb->users);
1717 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1720 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1722 if (!ptype->af_packet_priv || !skb->sk)
1725 if (ptype->id_match)
1726 return ptype->id_match(ptype, skb->sk);
1727 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1734 * Support routine. Sends outgoing frames to any network
1735 * taps currently in use.
1738 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1740 struct packet_type *ptype;
1741 struct sk_buff *skb2 = NULL;
1742 struct packet_type *pt_prev = NULL;
1745 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1746 /* Never send packets back to the socket
1747 * they originated from - MvS (miquels@drinkel.ow.org)
1749 if ((ptype->dev == dev || !ptype->dev) &&
1750 (!skb_loop_sk(ptype, skb))) {
1752 deliver_skb(skb2, pt_prev, skb->dev);
1757 skb2 = skb_clone(skb, GFP_ATOMIC);
1761 net_timestamp_set(skb2);
1763 /* skb->nh should be correctly
1764 set by sender, so that the second statement is
1765 just protection against buggy protocols.
1767 skb_reset_mac_header(skb2);
1769 if (skb_network_header(skb2) < skb2->data ||
1770 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1771 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1772 ntohs(skb2->protocol),
1774 skb_reset_network_header(skb2);
1777 skb2->transport_header = skb2->network_header;
1778 skb2->pkt_type = PACKET_OUTGOING;
1783 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1788 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1789 * @dev: Network device
1790 * @txq: number of queues available
1792 * If real_num_tx_queues is changed the tc mappings may no longer be
1793 * valid. To resolve this verify the tc mapping remains valid and if
1794 * not NULL the mapping. With no priorities mapping to this
1795 * offset/count pair it will no longer be used. In the worst case TC0
1796 * is invalid nothing can be done so disable priority mappings. If is
1797 * expected that drivers will fix this mapping if they can before
1798 * calling netif_set_real_num_tx_queues.
1800 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1803 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1805 /* If TC0 is invalidated disable TC mapping */
1806 if (tc->offset + tc->count > txq) {
1807 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1812 /* Invalidated prio to tc mappings set to TC0 */
1813 for (i = 1; i < TC_BITMASK + 1; i++) {
1814 int q = netdev_get_prio_tc_map(dev, i);
1816 tc = &dev->tc_to_txq[q];
1817 if (tc->offset + tc->count > txq) {
1818 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1820 netdev_set_prio_tc_map(dev, i, 0);
1826 static DEFINE_MUTEX(xps_map_mutex);
1827 #define xmap_dereference(P) \
1828 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1830 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1833 struct xps_map *map = NULL;
1837 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1839 for (pos = 0; map && pos < map->len; pos++) {
1840 if (map->queues[pos] == index) {
1842 map->queues[pos] = map->queues[--map->len];
1844 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1845 kfree_rcu(map, rcu);
1855 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1857 struct xps_dev_maps *dev_maps;
1859 bool active = false;
1861 mutex_lock(&xps_map_mutex);
1862 dev_maps = xmap_dereference(dev->xps_maps);
1867 for_each_possible_cpu(cpu) {
1868 for (i = index; i < dev->num_tx_queues; i++) {
1869 if (!remove_xps_queue(dev_maps, cpu, i))
1872 if (i == dev->num_tx_queues)
1877 RCU_INIT_POINTER(dev->xps_maps, NULL);
1878 kfree_rcu(dev_maps, rcu);
1881 for (i = index; i < dev->num_tx_queues; i++)
1882 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1886 mutex_unlock(&xps_map_mutex);
1889 static struct xps_map *expand_xps_map(struct xps_map *map,
1892 struct xps_map *new_map;
1893 int alloc_len = XPS_MIN_MAP_ALLOC;
1896 for (pos = 0; map && pos < map->len; pos++) {
1897 if (map->queues[pos] != index)
1902 /* Need to add queue to this CPU's existing map */
1904 if (pos < map->alloc_len)
1907 alloc_len = map->alloc_len * 2;
1910 /* Need to allocate new map to store queue on this CPU's map */
1911 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1916 for (i = 0; i < pos; i++)
1917 new_map->queues[i] = map->queues[i];
1918 new_map->alloc_len = alloc_len;
1924 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
1927 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1928 struct xps_map *map, *new_map;
1929 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1930 int cpu, numa_node_id = -2;
1931 bool active = false;
1933 mutex_lock(&xps_map_mutex);
1935 dev_maps = xmap_dereference(dev->xps_maps);
1937 /* allocate memory for queue storage */
1938 for_each_online_cpu(cpu) {
1939 if (!cpumask_test_cpu(cpu, mask))
1943 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1944 if (!new_dev_maps) {
1945 mutex_unlock(&xps_map_mutex);
1949 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1952 map = expand_xps_map(map, cpu, index);
1956 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1960 goto out_no_new_maps;
1962 for_each_possible_cpu(cpu) {
1963 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1964 /* add queue to CPU maps */
1967 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1968 while ((pos < map->len) && (map->queues[pos] != index))
1971 if (pos == map->len)
1972 map->queues[map->len++] = index;
1974 if (numa_node_id == -2)
1975 numa_node_id = cpu_to_node(cpu);
1976 else if (numa_node_id != cpu_to_node(cpu))
1979 } else if (dev_maps) {
1980 /* fill in the new device map from the old device map */
1981 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1982 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1987 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1989 /* Cleanup old maps */
1991 for_each_possible_cpu(cpu) {
1992 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1993 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1994 if (map && map != new_map)
1995 kfree_rcu(map, rcu);
1998 kfree_rcu(dev_maps, rcu);
2001 dev_maps = new_dev_maps;
2005 /* update Tx queue numa node */
2006 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2007 (numa_node_id >= 0) ? numa_node_id :
2013 /* removes queue from unused CPUs */
2014 for_each_possible_cpu(cpu) {
2015 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
2018 if (remove_xps_queue(dev_maps, cpu, index))
2022 /* free map if not active */
2024 RCU_INIT_POINTER(dev->xps_maps, NULL);
2025 kfree_rcu(dev_maps, rcu);
2029 mutex_unlock(&xps_map_mutex);
2033 /* remove any maps that we added */
2034 for_each_possible_cpu(cpu) {
2035 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
2036 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2038 if (new_map && new_map != map)
2042 mutex_unlock(&xps_map_mutex);
2044 kfree(new_dev_maps);
2047 EXPORT_SYMBOL(netif_set_xps_queue);
2051 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2052 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2054 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2058 if (txq < 1 || txq > dev->num_tx_queues)
2061 if (dev->reg_state == NETREG_REGISTERED ||
2062 dev->reg_state == NETREG_UNREGISTERING) {
2065 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2071 netif_setup_tc(dev, txq);
2073 if (txq < dev->real_num_tx_queues) {
2074 qdisc_reset_all_tx_gt(dev, txq);
2076 netif_reset_xps_queues_gt(dev, txq);
2081 dev->real_num_tx_queues = txq;
2084 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2088 * netif_set_real_num_rx_queues - set actual number of RX queues used
2089 * @dev: Network device
2090 * @rxq: Actual number of RX queues
2092 * This must be called either with the rtnl_lock held or before
2093 * registration of the net device. Returns 0 on success, or a
2094 * negative error code. If called before registration, it always
2097 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2101 if (rxq < 1 || rxq > dev->num_rx_queues)
2104 if (dev->reg_state == NETREG_REGISTERED) {
2107 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2113 dev->real_num_rx_queues = rxq;
2116 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2120 * netif_get_num_default_rss_queues - default number of RSS queues
2122 * This routine should set an upper limit on the number of RSS queues
2123 * used by default by multiqueue devices.
2125 int netif_get_num_default_rss_queues(void)
2127 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2129 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2131 static inline void __netif_reschedule(struct Qdisc *q)
2133 struct softnet_data *sd;
2134 unsigned long flags;
2136 local_irq_save(flags);
2137 sd = &__get_cpu_var(softnet_data);
2138 q->next_sched = NULL;
2139 *sd->output_queue_tailp = q;
2140 sd->output_queue_tailp = &q->next_sched;
2141 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2142 local_irq_restore(flags);
2145 void __netif_schedule(struct Qdisc *q)
2147 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2148 __netif_reschedule(q);
2150 EXPORT_SYMBOL(__netif_schedule);
2152 struct dev_kfree_skb_cb {
2153 enum skb_free_reason reason;
2156 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2158 return (struct dev_kfree_skb_cb *)skb->cb;
2161 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2163 unsigned long flags;
2165 if (likely(atomic_read(&skb->users) == 1)) {
2167 atomic_set(&skb->users, 0);
2168 } else if (likely(!atomic_dec_and_test(&skb->users))) {
2171 get_kfree_skb_cb(skb)->reason = reason;
2172 local_irq_save(flags);
2173 skb->next = __this_cpu_read(softnet_data.completion_queue);
2174 __this_cpu_write(softnet_data.completion_queue, skb);
2175 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2176 local_irq_restore(flags);
2178 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2180 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2182 if (in_irq() || irqs_disabled())
2183 __dev_kfree_skb_irq(skb, reason);
2187 EXPORT_SYMBOL(__dev_kfree_skb_any);
2191 * netif_device_detach - mark device as removed
2192 * @dev: network device
2194 * Mark device as removed from system and therefore no longer available.
2196 void netif_device_detach(struct net_device *dev)
2198 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2199 netif_running(dev)) {
2200 netif_tx_stop_all_queues(dev);
2203 EXPORT_SYMBOL(netif_device_detach);
2206 * netif_device_attach - mark device as attached
2207 * @dev: network device
2209 * Mark device as attached from system and restart if needed.
2211 void netif_device_attach(struct net_device *dev)
2213 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2214 netif_running(dev)) {
2215 netif_tx_wake_all_queues(dev);
2216 __netdev_watchdog_up(dev);
2219 EXPORT_SYMBOL(netif_device_attach);
2221 static void skb_warn_bad_offload(const struct sk_buff *skb)
2223 static const netdev_features_t null_features = 0;
2224 struct net_device *dev = skb->dev;
2225 const char *driver = "";
2227 if (!net_ratelimit())
2230 if (dev && dev->dev.parent)
2231 driver = dev_driver_string(dev->dev.parent);
2233 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2234 "gso_type=%d ip_summed=%d\n",
2235 driver, dev ? &dev->features : &null_features,
2236 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2237 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2238 skb_shinfo(skb)->gso_type, skb->ip_summed);
2242 * Invalidate hardware checksum when packet is to be mangled, and
2243 * complete checksum manually on outgoing path.
2245 int skb_checksum_help(struct sk_buff *skb)
2248 int ret = 0, offset;
2250 if (skb->ip_summed == CHECKSUM_COMPLETE)
2251 goto out_set_summed;
2253 if (unlikely(skb_shinfo(skb)->gso_size)) {
2254 skb_warn_bad_offload(skb);
2258 /* Before computing a checksum, we should make sure no frag could
2259 * be modified by an external entity : checksum could be wrong.
2261 if (skb_has_shared_frag(skb)) {
2262 ret = __skb_linearize(skb);
2267 offset = skb_checksum_start_offset(skb);
2268 BUG_ON(offset >= skb_headlen(skb));
2269 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2271 offset += skb->csum_offset;
2272 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2274 if (skb_cloned(skb) &&
2275 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2276 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2281 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2283 skb->ip_summed = CHECKSUM_NONE;
2287 EXPORT_SYMBOL(skb_checksum_help);
2289 __be16 skb_network_protocol(struct sk_buff *skb)
2291 __be16 type = skb->protocol;
2292 int vlan_depth = ETH_HLEN;
2294 /* Tunnel gso handlers can set protocol to ethernet. */
2295 if (type == htons(ETH_P_TEB)) {
2298 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2301 eth = (struct ethhdr *)skb_mac_header(skb);
2302 type = eth->h_proto;
2305 while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2306 struct vlan_hdr *vh;
2308 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2311 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2312 type = vh->h_vlan_encapsulated_proto;
2313 vlan_depth += VLAN_HLEN;
2320 * skb_mac_gso_segment - mac layer segmentation handler.
2321 * @skb: buffer to segment
2322 * @features: features for the output path (see dev->features)
2324 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2325 netdev_features_t features)
2327 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2328 struct packet_offload *ptype;
2329 __be16 type = skb_network_protocol(skb);
2331 if (unlikely(!type))
2332 return ERR_PTR(-EINVAL);
2334 __skb_pull(skb, skb->mac_len);
2337 list_for_each_entry_rcu(ptype, &offload_base, list) {
2338 if (ptype->type == type && ptype->callbacks.gso_segment) {
2339 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2342 err = ptype->callbacks.gso_send_check(skb);
2343 segs = ERR_PTR(err);
2344 if (err || skb_gso_ok(skb, features))
2346 __skb_push(skb, (skb->data -
2347 skb_network_header(skb)));
2349 segs = ptype->callbacks.gso_segment(skb, features);
2355 __skb_push(skb, skb->data - skb_mac_header(skb));
2359 EXPORT_SYMBOL(skb_mac_gso_segment);
2362 /* openvswitch calls this on rx path, so we need a different check.
2364 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2367 return skb->ip_summed != CHECKSUM_PARTIAL;
2369 return skb->ip_summed == CHECKSUM_NONE;
2373 * __skb_gso_segment - Perform segmentation on skb.
2374 * @skb: buffer to segment
2375 * @features: features for the output path (see dev->features)
2376 * @tx_path: whether it is called in TX path
2378 * This function segments the given skb and returns a list of segments.
2380 * It may return NULL if the skb requires no segmentation. This is
2381 * only possible when GSO is used for verifying header integrity.
2383 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2384 netdev_features_t features, bool tx_path)
2386 if (unlikely(skb_needs_check(skb, tx_path))) {
2389 skb_warn_bad_offload(skb);
2391 if (skb_header_cloned(skb) &&
2392 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2393 return ERR_PTR(err);
2396 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2397 SKB_GSO_CB(skb)->encap_level = 0;
2399 skb_reset_mac_header(skb);
2400 skb_reset_mac_len(skb);
2402 return skb_mac_gso_segment(skb, features);
2404 EXPORT_SYMBOL(__skb_gso_segment);
2406 /* Take action when hardware reception checksum errors are detected. */
2408 void netdev_rx_csum_fault(struct net_device *dev)
2410 if (net_ratelimit()) {
2411 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2415 EXPORT_SYMBOL(netdev_rx_csum_fault);
2418 /* Actually, we should eliminate this check as soon as we know, that:
2419 * 1. IOMMU is present and allows to map all the memory.
2420 * 2. No high memory really exists on this machine.
2423 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2425 #ifdef CONFIG_HIGHMEM
2427 if (!(dev->features & NETIF_F_HIGHDMA)) {
2428 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2429 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2430 if (PageHighMem(skb_frag_page(frag)))
2435 if (PCI_DMA_BUS_IS_PHYS) {
2436 struct device *pdev = dev->dev.parent;
2440 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2441 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2442 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2443 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2452 void (*destructor)(struct sk_buff *skb);
2455 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2457 static void dev_gso_skb_destructor(struct sk_buff *skb)
2459 struct dev_gso_cb *cb;
2461 kfree_skb_list(skb->next);
2464 cb = DEV_GSO_CB(skb);
2466 cb->destructor(skb);
2470 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2471 * @skb: buffer to segment
2472 * @features: device features as applicable to this skb
2474 * This function segments the given skb and stores the list of segments
2477 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2479 struct sk_buff *segs;
2481 segs = skb_gso_segment(skb, features);
2483 /* Verifying header integrity only. */
2488 return PTR_ERR(segs);
2491 DEV_GSO_CB(skb)->destructor = skb->destructor;
2492 skb->destructor = dev_gso_skb_destructor;
2497 static netdev_features_t harmonize_features(struct sk_buff *skb,
2498 netdev_features_t features)
2500 if (skb->ip_summed != CHECKSUM_NONE &&
2501 !can_checksum_protocol(features, skb_network_protocol(skb))) {
2502 features &= ~NETIF_F_ALL_CSUM;
2503 } else if (illegal_highdma(skb->dev, skb)) {
2504 features &= ~NETIF_F_SG;
2510 netdev_features_t netif_skb_features(struct sk_buff *skb)
2512 __be16 protocol = skb->protocol;
2513 netdev_features_t features = skb->dev->features;
2515 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2516 features &= ~NETIF_F_GSO_MASK;
2518 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2519 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2520 protocol = veh->h_vlan_encapsulated_proto;
2521 } else if (!vlan_tx_tag_present(skb)) {
2522 return harmonize_features(skb, features);
2525 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2526 NETIF_F_HW_VLAN_STAG_TX);
2528 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD))
2529 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2530 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2531 NETIF_F_HW_VLAN_STAG_TX;
2533 return harmonize_features(skb, features);
2535 EXPORT_SYMBOL(netif_skb_features);
2537 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2538 struct netdev_queue *txq)
2540 const struct net_device_ops *ops = dev->netdev_ops;
2541 int rc = NETDEV_TX_OK;
2542 unsigned int skb_len;
2544 if (likely(!skb->next)) {
2545 netdev_features_t features;
2548 * If device doesn't need skb->dst, release it right now while
2549 * its hot in this cpu cache
2551 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2554 features = netif_skb_features(skb);
2556 if (vlan_tx_tag_present(skb) &&
2557 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2558 skb = __vlan_put_tag(skb, skb->vlan_proto,
2559 vlan_tx_tag_get(skb));
2566 /* If encapsulation offload request, verify we are testing
2567 * hardware encapsulation features instead of standard
2568 * features for the netdev
2570 if (skb->encapsulation)
2571 features &= dev->hw_enc_features;
2573 if (netif_needs_gso(skb, features)) {
2574 if (unlikely(dev_gso_segment(skb, features)))
2579 if (skb_needs_linearize(skb, features) &&
2580 __skb_linearize(skb))
2583 /* If packet is not checksummed and device does not
2584 * support checksumming for this protocol, complete
2585 * checksumming here.
2587 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2588 if (skb->encapsulation)
2589 skb_set_inner_transport_header(skb,
2590 skb_checksum_start_offset(skb));
2592 skb_set_transport_header(skb,
2593 skb_checksum_start_offset(skb));
2594 if (!(features & NETIF_F_ALL_CSUM) &&
2595 skb_checksum_help(skb))
2600 if (!list_empty(&ptype_all))
2601 dev_queue_xmit_nit(skb, dev);
2604 trace_net_dev_start_xmit(skb, dev);
2605 rc = ops->ndo_start_xmit(skb, dev);
2606 trace_net_dev_xmit(skb, rc, dev, skb_len);
2607 if (rc == NETDEV_TX_OK)
2608 txq_trans_update(txq);
2614 struct sk_buff *nskb = skb->next;
2616 skb->next = nskb->next;
2619 if (!list_empty(&ptype_all))
2620 dev_queue_xmit_nit(nskb, dev);
2622 skb_len = nskb->len;
2623 trace_net_dev_start_xmit(nskb, dev);
2624 rc = ops->ndo_start_xmit(nskb, dev);
2625 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2626 if (unlikely(rc != NETDEV_TX_OK)) {
2627 if (rc & ~NETDEV_TX_MASK)
2628 goto out_kfree_gso_skb;
2629 nskb->next = skb->next;
2633 txq_trans_update(txq);
2634 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2635 return NETDEV_TX_BUSY;
2636 } while (skb->next);
2639 if (likely(skb->next == NULL)) {
2640 skb->destructor = DEV_GSO_CB(skb)->destructor;
2649 EXPORT_SYMBOL_GPL(dev_hard_start_xmit);
2651 static void qdisc_pkt_len_init(struct sk_buff *skb)
2653 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2655 qdisc_skb_cb(skb)->pkt_len = skb->len;
2657 /* To get more precise estimation of bytes sent on wire,
2658 * we add to pkt_len the headers size of all segments
2660 if (shinfo->gso_size) {
2661 unsigned int hdr_len;
2662 u16 gso_segs = shinfo->gso_segs;
2664 /* mac layer + network layer */
2665 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2667 /* + transport layer */
2668 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2669 hdr_len += tcp_hdrlen(skb);
2671 hdr_len += sizeof(struct udphdr);
2673 if (shinfo->gso_type & SKB_GSO_DODGY)
2674 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2677 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2681 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2682 struct net_device *dev,
2683 struct netdev_queue *txq)
2685 spinlock_t *root_lock = qdisc_lock(q);
2689 qdisc_pkt_len_init(skb);
2690 qdisc_calculate_pkt_len(skb, q);
2692 * Heuristic to force contended enqueues to serialize on a
2693 * separate lock before trying to get qdisc main lock.
2694 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2695 * and dequeue packets faster.
2697 contended = qdisc_is_running(q);
2698 if (unlikely(contended))
2699 spin_lock(&q->busylock);
2701 spin_lock(root_lock);
2702 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2705 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2706 qdisc_run_begin(q)) {
2708 * This is a work-conserving queue; there are no old skbs
2709 * waiting to be sent out; and the qdisc is not running -
2710 * xmit the skb directly.
2712 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2715 qdisc_bstats_update(q, skb);
2717 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2718 if (unlikely(contended)) {
2719 spin_unlock(&q->busylock);
2726 rc = NET_XMIT_SUCCESS;
2729 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2730 if (qdisc_run_begin(q)) {
2731 if (unlikely(contended)) {
2732 spin_unlock(&q->busylock);
2738 spin_unlock(root_lock);
2739 if (unlikely(contended))
2740 spin_unlock(&q->busylock);
2744 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
2745 static void skb_update_prio(struct sk_buff *skb)
2747 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2749 if (!skb->priority && skb->sk && map) {
2750 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2752 if (prioidx < map->priomap_len)
2753 skb->priority = map->priomap[prioidx];
2757 #define skb_update_prio(skb)
2760 static DEFINE_PER_CPU(int, xmit_recursion);
2761 #define RECURSION_LIMIT 10
2764 * dev_loopback_xmit - loop back @skb
2765 * @skb: buffer to transmit
2767 int dev_loopback_xmit(struct sk_buff *skb)
2769 skb_reset_mac_header(skb);
2770 __skb_pull(skb, skb_network_offset(skb));
2771 skb->pkt_type = PACKET_LOOPBACK;
2772 skb->ip_summed = CHECKSUM_UNNECESSARY;
2773 WARN_ON(!skb_dst(skb));
2778 EXPORT_SYMBOL(dev_loopback_xmit);
2781 * dev_queue_xmit - transmit a buffer
2782 * @skb: buffer to transmit
2784 * Queue a buffer for transmission to a network device. The caller must
2785 * have set the device and priority and built the buffer before calling
2786 * this function. The function can be called from an interrupt.
2788 * A negative errno code is returned on a failure. A success does not
2789 * guarantee the frame will be transmitted as it may be dropped due
2790 * to congestion or traffic shaping.
2792 * -----------------------------------------------------------------------------------
2793 * I notice this method can also return errors from the queue disciplines,
2794 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2797 * Regardless of the return value, the skb is consumed, so it is currently
2798 * difficult to retry a send to this method. (You can bump the ref count
2799 * before sending to hold a reference for retry if you are careful.)
2801 * When calling this method, interrupts MUST be enabled. This is because
2802 * the BH enable code must have IRQs enabled so that it will not deadlock.
2805 int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
2807 struct net_device *dev = skb->dev;
2808 struct netdev_queue *txq;
2812 skb_reset_mac_header(skb);
2814 /* Disable soft irqs for various locks below. Also
2815 * stops preemption for RCU.
2819 skb_update_prio(skb);
2821 txq = netdev_pick_tx(dev, skb, accel_priv);
2822 q = rcu_dereference_bh(txq->qdisc);
2824 #ifdef CONFIG_NET_CLS_ACT
2825 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2827 trace_net_dev_queue(skb);
2829 rc = __dev_xmit_skb(skb, q, dev, txq);
2833 /* The device has no queue. Common case for software devices:
2834 loopback, all the sorts of tunnels...
2836 Really, it is unlikely that netif_tx_lock protection is necessary
2837 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2839 However, it is possible, that they rely on protection
2842 Check this and shot the lock. It is not prone from deadlocks.
2843 Either shot noqueue qdisc, it is even simpler 8)
2845 if (dev->flags & IFF_UP) {
2846 int cpu = smp_processor_id(); /* ok because BHs are off */
2848 if (txq->xmit_lock_owner != cpu) {
2850 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2851 goto recursion_alert;
2853 HARD_TX_LOCK(dev, txq, cpu);
2855 if (!netif_xmit_stopped(txq)) {
2856 __this_cpu_inc(xmit_recursion);
2857 rc = dev_hard_start_xmit(skb, dev, txq);
2858 __this_cpu_dec(xmit_recursion);
2859 if (dev_xmit_complete(rc)) {
2860 HARD_TX_UNLOCK(dev, txq);
2864 HARD_TX_UNLOCK(dev, txq);
2865 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2868 /* Recursion is detected! It is possible,
2872 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2878 rcu_read_unlock_bh();
2883 rcu_read_unlock_bh();
2887 int dev_queue_xmit(struct sk_buff *skb)
2889 return __dev_queue_xmit(skb, NULL);
2891 EXPORT_SYMBOL(dev_queue_xmit);
2893 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
2895 return __dev_queue_xmit(skb, accel_priv);
2897 EXPORT_SYMBOL(dev_queue_xmit_accel);
2900 /*=======================================================================
2902 =======================================================================*/
2904 int netdev_max_backlog __read_mostly = 1000;
2905 EXPORT_SYMBOL(netdev_max_backlog);
2907 int netdev_tstamp_prequeue __read_mostly = 1;
2908 int netdev_budget __read_mostly = 300;
2909 int weight_p __read_mostly = 64; /* old backlog weight */
2911 /* Called with irq disabled */
2912 static inline void ____napi_schedule(struct softnet_data *sd,
2913 struct napi_struct *napi)
2915 list_add_tail(&napi->poll_list, &sd->poll_list);
2916 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2921 /* One global table that all flow-based protocols share. */
2922 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2923 EXPORT_SYMBOL(rps_sock_flow_table);
2925 struct static_key rps_needed __read_mostly;
2927 static struct rps_dev_flow *
2928 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2929 struct rps_dev_flow *rflow, u16 next_cpu)
2931 if (next_cpu != RPS_NO_CPU) {
2932 #ifdef CONFIG_RFS_ACCEL
2933 struct netdev_rx_queue *rxqueue;
2934 struct rps_dev_flow_table *flow_table;
2935 struct rps_dev_flow *old_rflow;
2940 /* Should we steer this flow to a different hardware queue? */
2941 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2942 !(dev->features & NETIF_F_NTUPLE))
2944 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2945 if (rxq_index == skb_get_rx_queue(skb))
2948 rxqueue = dev->_rx + rxq_index;
2949 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2952 flow_id = skb->rxhash & flow_table->mask;
2953 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2954 rxq_index, flow_id);
2958 rflow = &flow_table->flows[flow_id];
2960 if (old_rflow->filter == rflow->filter)
2961 old_rflow->filter = RPS_NO_FILTER;
2965 per_cpu(softnet_data, next_cpu).input_queue_head;
2968 rflow->cpu = next_cpu;
2973 * get_rps_cpu is called from netif_receive_skb and returns the target
2974 * CPU from the RPS map of the receiving queue for a given skb.
2975 * rcu_read_lock must be held on entry.
2977 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2978 struct rps_dev_flow **rflowp)
2980 struct netdev_rx_queue *rxqueue;
2981 struct rps_map *map;
2982 struct rps_dev_flow_table *flow_table;
2983 struct rps_sock_flow_table *sock_flow_table;
2987 if (skb_rx_queue_recorded(skb)) {
2988 u16 index = skb_get_rx_queue(skb);
2989 if (unlikely(index >= dev->real_num_rx_queues)) {
2990 WARN_ONCE(dev->real_num_rx_queues > 1,
2991 "%s received packet on queue %u, but number "
2992 "of RX queues is %u\n",
2993 dev->name, index, dev->real_num_rx_queues);
2996 rxqueue = dev->_rx + index;
3000 map = rcu_dereference(rxqueue->rps_map);
3002 if (map->len == 1 &&
3003 !rcu_access_pointer(rxqueue->rps_flow_table)) {
3004 tcpu = map->cpus[0];
3005 if (cpu_online(tcpu))
3009 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
3013 skb_reset_network_header(skb);
3014 if (!skb_get_hash(skb))
3017 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3018 sock_flow_table = rcu_dereference(rps_sock_flow_table);
3019 if (flow_table && sock_flow_table) {
3021 struct rps_dev_flow *rflow;
3023 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
3026 next_cpu = sock_flow_table->ents[skb->rxhash &
3027 sock_flow_table->mask];
3030 * If the desired CPU (where last recvmsg was done) is
3031 * different from current CPU (one in the rx-queue flow
3032 * table entry), switch if one of the following holds:
3033 * - Current CPU is unset (equal to RPS_NO_CPU).
3034 * - Current CPU is offline.
3035 * - The current CPU's queue tail has advanced beyond the
3036 * last packet that was enqueued using this table entry.
3037 * This guarantees that all previous packets for the flow
3038 * have been dequeued, thus preserving in order delivery.
3040 if (unlikely(tcpu != next_cpu) &&
3041 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
3042 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3043 rflow->last_qtail)) >= 0)) {
3045 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3048 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3056 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
3058 if (cpu_online(tcpu)) {
3068 #ifdef CONFIG_RFS_ACCEL
3071 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3072 * @dev: Device on which the filter was set
3073 * @rxq_index: RX queue index
3074 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3075 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3077 * Drivers that implement ndo_rx_flow_steer() should periodically call
3078 * this function for each installed filter and remove the filters for
3079 * which it returns %true.
3081 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3082 u32 flow_id, u16 filter_id)
3084 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3085 struct rps_dev_flow_table *flow_table;
3086 struct rps_dev_flow *rflow;
3091 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3092 if (flow_table && flow_id <= flow_table->mask) {
3093 rflow = &flow_table->flows[flow_id];
3094 cpu = ACCESS_ONCE(rflow->cpu);
3095 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3096 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3097 rflow->last_qtail) <
3098 (int)(10 * flow_table->mask)))
3104 EXPORT_SYMBOL(rps_may_expire_flow);
3106 #endif /* CONFIG_RFS_ACCEL */
3108 /* Called from hardirq (IPI) context */
3109 static void rps_trigger_softirq(void *data)
3111 struct softnet_data *sd = data;
3113 ____napi_schedule(sd, &sd->backlog);
3117 #endif /* CONFIG_RPS */
3120 * Check if this softnet_data structure is another cpu one
3121 * If yes, queue it to our IPI list and return 1
3124 static int rps_ipi_queued(struct softnet_data *sd)
3127 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3130 sd->rps_ipi_next = mysd->rps_ipi_list;
3131 mysd->rps_ipi_list = sd;
3133 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3136 #endif /* CONFIG_RPS */
3140 #ifdef CONFIG_NET_FLOW_LIMIT
3141 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3144 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3146 #ifdef CONFIG_NET_FLOW_LIMIT
3147 struct sd_flow_limit *fl;
3148 struct softnet_data *sd;
3149 unsigned int old_flow, new_flow;
3151 if (qlen < (netdev_max_backlog >> 1))
3154 sd = &__get_cpu_var(softnet_data);
3157 fl = rcu_dereference(sd->flow_limit);
3159 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3160 old_flow = fl->history[fl->history_head];
3161 fl->history[fl->history_head] = new_flow;
3164 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3166 if (likely(fl->buckets[old_flow]))
3167 fl->buckets[old_flow]--;
3169 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3181 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3182 * queue (may be a remote CPU queue).
3184 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3185 unsigned int *qtail)
3187 struct softnet_data *sd;
3188 unsigned long flags;
3191 sd = &per_cpu(softnet_data, cpu);
3193 local_irq_save(flags);
3196 qlen = skb_queue_len(&sd->input_pkt_queue);
3197 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3198 if (skb_queue_len(&sd->input_pkt_queue)) {
3200 __skb_queue_tail(&sd->input_pkt_queue, skb);
3201 input_queue_tail_incr_save(sd, qtail);
3203 local_irq_restore(flags);
3204 return NET_RX_SUCCESS;
3207 /* Schedule NAPI for backlog device
3208 * We can use non atomic operation since we own the queue lock
3210 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3211 if (!rps_ipi_queued(sd))
3212 ____napi_schedule(sd, &sd->backlog);
3220 local_irq_restore(flags);
3222 atomic_long_inc(&skb->dev->rx_dropped);
3227 static int netif_rx_internal(struct sk_buff *skb)
3231 /* if netpoll wants it, pretend we never saw it */
3232 if (netpoll_rx(skb))
3235 net_timestamp_check(netdev_tstamp_prequeue, skb);
3237 trace_netif_rx(skb);
3239 if (static_key_false(&rps_needed)) {
3240 struct rps_dev_flow voidflow, *rflow = &voidflow;
3246 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3248 cpu = smp_processor_id();
3250 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3258 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3265 * netif_rx - post buffer to the network code
3266 * @skb: buffer to post
3268 * This function receives a packet from a device driver and queues it for
3269 * the upper (protocol) levels to process. It always succeeds. The buffer
3270 * may be dropped during processing for congestion control or by the
3274 * NET_RX_SUCCESS (no congestion)
3275 * NET_RX_DROP (packet was dropped)
3279 int netif_rx(struct sk_buff *skb)
3281 trace_netif_rx_entry(skb);
3283 return netif_rx_internal(skb);
3285 EXPORT_SYMBOL(netif_rx);
3287 int netif_rx_ni(struct sk_buff *skb)
3291 trace_netif_rx_ni_entry(skb);
3294 err = netif_rx_internal(skb);
3295 if (local_softirq_pending())
3301 EXPORT_SYMBOL(netif_rx_ni);
3303 static void net_tx_action(struct softirq_action *h)
3305 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3307 if (sd->completion_queue) {
3308 struct sk_buff *clist;
3310 local_irq_disable();
3311 clist = sd->completion_queue;
3312 sd->completion_queue = NULL;
3316 struct sk_buff *skb = clist;
3317 clist = clist->next;
3319 WARN_ON(atomic_read(&skb->users));
3320 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
3321 trace_consume_skb(skb);
3323 trace_kfree_skb(skb, net_tx_action);
3328 if (sd->output_queue) {
3331 local_irq_disable();
3332 head = sd->output_queue;
3333 sd->output_queue = NULL;
3334 sd->output_queue_tailp = &sd->output_queue;
3338 struct Qdisc *q = head;
3339 spinlock_t *root_lock;
3341 head = head->next_sched;
3343 root_lock = qdisc_lock(q);
3344 if (spin_trylock(root_lock)) {
3345 smp_mb__before_clear_bit();
3346 clear_bit(__QDISC_STATE_SCHED,
3349 spin_unlock(root_lock);
3351 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3353 __netif_reschedule(q);
3355 smp_mb__before_clear_bit();
3356 clear_bit(__QDISC_STATE_SCHED,
3364 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3365 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3366 /* This hook is defined here for ATM LANE */
3367 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3368 unsigned char *addr) __read_mostly;
3369 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3372 #ifdef CONFIG_NET_CLS_ACT
3373 /* TODO: Maybe we should just force sch_ingress to be compiled in
3374 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3375 * a compare and 2 stores extra right now if we dont have it on
3376 * but have CONFIG_NET_CLS_ACT
3377 * NOTE: This doesn't stop any functionality; if you dont have
3378 * the ingress scheduler, you just can't add policies on ingress.
3381 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3383 struct net_device *dev = skb->dev;
3384 u32 ttl = G_TC_RTTL(skb->tc_verd);
3385 int result = TC_ACT_OK;
3388 if (unlikely(MAX_RED_LOOP < ttl++)) {
3389 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3390 skb->skb_iif, dev->ifindex);
3394 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3395 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3398 if (q != &noop_qdisc) {
3399 spin_lock(qdisc_lock(q));
3400 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3401 result = qdisc_enqueue_root(skb, q);
3402 spin_unlock(qdisc_lock(q));
3408 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3409 struct packet_type **pt_prev,
3410 int *ret, struct net_device *orig_dev)
3412 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3414 if (!rxq || rxq->qdisc == &noop_qdisc)
3418 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3422 switch (ing_filter(skb, rxq)) {
3436 * netdev_rx_handler_register - register receive handler
3437 * @dev: device to register a handler for
3438 * @rx_handler: receive handler to register
3439 * @rx_handler_data: data pointer that is used by rx handler
3441 * Register a receive hander for a device. This handler will then be
3442 * called from __netif_receive_skb. A negative errno code is returned
3445 * The caller must hold the rtnl_mutex.
3447 * For a general description of rx_handler, see enum rx_handler_result.
3449 int netdev_rx_handler_register(struct net_device *dev,
3450 rx_handler_func_t *rx_handler,
3451 void *rx_handler_data)
3455 if (dev->rx_handler)
3458 /* Note: rx_handler_data must be set before rx_handler */
3459 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3460 rcu_assign_pointer(dev->rx_handler, rx_handler);
3464 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3467 * netdev_rx_handler_unregister - unregister receive handler
3468 * @dev: device to unregister a handler from
3470 * Unregister a receive handler from a device.
3472 * The caller must hold the rtnl_mutex.
3474 void netdev_rx_handler_unregister(struct net_device *dev)
3478 RCU_INIT_POINTER(dev->rx_handler, NULL);
3479 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3480 * section has a guarantee to see a non NULL rx_handler_data
3484 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3486 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3489 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3490 * the special handling of PFMEMALLOC skbs.
3492 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3494 switch (skb->protocol) {
3495 case __constant_htons(ETH_P_ARP):
3496 case __constant_htons(ETH_P_IP):
3497 case __constant_htons(ETH_P_IPV6):
3498 case __constant_htons(ETH_P_8021Q):
3499 case __constant_htons(ETH_P_8021AD):
3506 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3508 struct packet_type *ptype, *pt_prev;
3509 rx_handler_func_t *rx_handler;
3510 struct net_device *orig_dev;
3511 struct net_device *null_or_dev;
3512 bool deliver_exact = false;
3513 int ret = NET_RX_DROP;
3516 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3518 trace_netif_receive_skb(skb);
3520 /* if we've gotten here through NAPI, check netpoll */
3521 if (netpoll_receive_skb(skb))
3524 orig_dev = skb->dev;
3526 skb_reset_network_header(skb);
3527 if (!skb_transport_header_was_set(skb))
3528 skb_reset_transport_header(skb);
3529 skb_reset_mac_len(skb);
3536 skb->skb_iif = skb->dev->ifindex;
3538 __this_cpu_inc(softnet_data.processed);
3540 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3541 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3542 skb = vlan_untag(skb);
3547 #ifdef CONFIG_NET_CLS_ACT
3548 if (skb->tc_verd & TC_NCLS) {
3549 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3557 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3558 if (!ptype->dev || ptype->dev == skb->dev) {
3560 ret = deliver_skb(skb, pt_prev, orig_dev);
3566 #ifdef CONFIG_NET_CLS_ACT
3567 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3573 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3576 if (vlan_tx_tag_present(skb)) {
3578 ret = deliver_skb(skb, pt_prev, orig_dev);
3581 if (vlan_do_receive(&skb))
3583 else if (unlikely(!skb))
3587 rx_handler = rcu_dereference(skb->dev->rx_handler);
3590 ret = deliver_skb(skb, pt_prev, orig_dev);
3593 switch (rx_handler(&skb)) {
3594 case RX_HANDLER_CONSUMED:
3595 ret = NET_RX_SUCCESS;
3597 case RX_HANDLER_ANOTHER:
3599 case RX_HANDLER_EXACT:
3600 deliver_exact = true;
3601 case RX_HANDLER_PASS:
3608 if (unlikely(vlan_tx_tag_present(skb))) {
3609 if (vlan_tx_tag_get_id(skb))
3610 skb->pkt_type = PACKET_OTHERHOST;
3611 /* Note: we might in the future use prio bits
3612 * and set skb->priority like in vlan_do_receive()
3613 * For the time being, just ignore Priority Code Point
3618 /* deliver only exact match when indicated */
3619 null_or_dev = deliver_exact ? skb->dev : NULL;
3621 type = skb->protocol;
3622 list_for_each_entry_rcu(ptype,
3623 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3624 if (ptype->type == type &&
3625 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3626 ptype->dev == orig_dev)) {
3628 ret = deliver_skb(skb, pt_prev, orig_dev);
3634 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3637 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3640 atomic_long_inc(&skb->dev->rx_dropped);
3642 /* Jamal, now you will not able to escape explaining
3643 * me how you were going to use this. :-)
3654 static int __netif_receive_skb(struct sk_buff *skb)
3658 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3659 unsigned long pflags = current->flags;
3662 * PFMEMALLOC skbs are special, they should
3663 * - be delivered to SOCK_MEMALLOC sockets only
3664 * - stay away from userspace
3665 * - have bounded memory usage
3667 * Use PF_MEMALLOC as this saves us from propagating the allocation
3668 * context down to all allocation sites.
3670 current->flags |= PF_MEMALLOC;
3671 ret = __netif_receive_skb_core(skb, true);
3672 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3674 ret = __netif_receive_skb_core(skb, false);
3679 static int netif_receive_skb_internal(struct sk_buff *skb)
3681 net_timestamp_check(netdev_tstamp_prequeue, skb);
3683 if (skb_defer_rx_timestamp(skb))
3684 return NET_RX_SUCCESS;
3687 if (static_key_false(&rps_needed)) {
3688 struct rps_dev_flow voidflow, *rflow = &voidflow;
3693 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3696 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3703 return __netif_receive_skb(skb);
3707 * netif_receive_skb - process receive buffer from network
3708 * @skb: buffer to process
3710 * netif_receive_skb() is the main receive data processing function.
3711 * It always succeeds. The buffer may be dropped during processing
3712 * for congestion control or by the protocol layers.
3714 * This function may only be called from softirq context and interrupts
3715 * should be enabled.
3717 * Return values (usually ignored):
3718 * NET_RX_SUCCESS: no congestion
3719 * NET_RX_DROP: packet was dropped
3721 int netif_receive_skb(struct sk_buff *skb)
3723 trace_netif_receive_skb_entry(skb);
3725 return netif_receive_skb_internal(skb);
3727 EXPORT_SYMBOL(netif_receive_skb);
3729 /* Network device is going away, flush any packets still pending
3730 * Called with irqs disabled.
3732 static void flush_backlog(void *arg)
3734 struct net_device *dev = arg;
3735 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3736 struct sk_buff *skb, *tmp;
3739 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3740 if (skb->dev == dev) {
3741 __skb_unlink(skb, &sd->input_pkt_queue);
3743 input_queue_head_incr(sd);
3748 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3749 if (skb->dev == dev) {
3750 __skb_unlink(skb, &sd->process_queue);
3752 input_queue_head_incr(sd);
3757 static int napi_gro_complete(struct sk_buff *skb)
3759 struct packet_offload *ptype;
3760 __be16 type = skb->protocol;
3761 struct list_head *head = &offload_base;
3764 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3766 if (NAPI_GRO_CB(skb)->count == 1) {
3767 skb_shinfo(skb)->gso_size = 0;
3772 list_for_each_entry_rcu(ptype, head, list) {
3773 if (ptype->type != type || !ptype->callbacks.gro_complete)
3776 err = ptype->callbacks.gro_complete(skb, 0);
3782 WARN_ON(&ptype->list == head);
3784 return NET_RX_SUCCESS;
3788 return netif_receive_skb_internal(skb);
3791 /* napi->gro_list contains packets ordered by age.
3792 * youngest packets at the head of it.
3793 * Complete skbs in reverse order to reduce latencies.
3795 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3797 struct sk_buff *skb, *prev = NULL;
3799 /* scan list and build reverse chain */
3800 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3805 for (skb = prev; skb; skb = prev) {
3808 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3812 napi_gro_complete(skb);
3816 napi->gro_list = NULL;
3818 EXPORT_SYMBOL(napi_gro_flush);
3820 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3823 unsigned int maclen = skb->dev->hard_header_len;
3824 u32 hash = skb_get_hash_raw(skb);
3826 for (p = napi->gro_list; p; p = p->next) {
3827 unsigned long diffs;
3829 NAPI_GRO_CB(p)->flush = 0;
3831 if (hash != skb_get_hash_raw(p)) {
3832 NAPI_GRO_CB(p)->same_flow = 0;
3836 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3837 diffs |= p->vlan_tci ^ skb->vlan_tci;
3838 if (maclen == ETH_HLEN)
3839 diffs |= compare_ether_header(skb_mac_header(p),
3840 skb_gro_mac_header(skb));
3842 diffs = memcmp(skb_mac_header(p),
3843 skb_gro_mac_header(skb),
3845 NAPI_GRO_CB(p)->same_flow = !diffs;
3849 static void skb_gro_reset_offset(struct sk_buff *skb)
3851 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3852 const skb_frag_t *frag0 = &pinfo->frags[0];
3854 NAPI_GRO_CB(skb)->data_offset = 0;
3855 NAPI_GRO_CB(skb)->frag0 = NULL;
3856 NAPI_GRO_CB(skb)->frag0_len = 0;
3858 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3860 !PageHighMem(skb_frag_page(frag0))) {
3861 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3862 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3866 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3868 struct sk_buff **pp = NULL;
3869 struct packet_offload *ptype;
3870 __be16 type = skb->protocol;
3871 struct list_head *head = &offload_base;
3873 enum gro_result ret;
3875 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3878 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3881 skb_gro_reset_offset(skb);
3882 gro_list_prepare(napi, skb);
3883 NAPI_GRO_CB(skb)->csum = skb->csum; /* Needed for CHECKSUM_COMPLETE */
3886 list_for_each_entry_rcu(ptype, head, list) {
3887 if (ptype->type != type || !ptype->callbacks.gro_receive)
3890 skb_set_network_header(skb, skb_gro_offset(skb));
3891 skb_reset_mac_len(skb);
3892 NAPI_GRO_CB(skb)->same_flow = 0;
3893 NAPI_GRO_CB(skb)->flush = 0;
3894 NAPI_GRO_CB(skb)->free = 0;
3896 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3901 if (&ptype->list == head)
3904 same_flow = NAPI_GRO_CB(skb)->same_flow;
3905 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3908 struct sk_buff *nskb = *pp;
3912 napi_gro_complete(nskb);
3919 if (NAPI_GRO_CB(skb)->flush)
3922 if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
3923 struct sk_buff *nskb = napi->gro_list;
3925 /* locate the end of the list to select the 'oldest' flow */
3926 while (nskb->next) {
3932 napi_gro_complete(nskb);
3936 NAPI_GRO_CB(skb)->count = 1;
3937 NAPI_GRO_CB(skb)->age = jiffies;
3938 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3939 skb->next = napi->gro_list;
3940 napi->gro_list = skb;
3944 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3945 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3947 BUG_ON(skb->end - skb->tail < grow);
3949 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3952 skb->data_len -= grow;
3954 skb_shinfo(skb)->frags[0].page_offset += grow;
3955 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3957 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3958 skb_frag_unref(skb, 0);
3959 memmove(skb_shinfo(skb)->frags,
3960 skb_shinfo(skb)->frags + 1,
3961 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3973 struct packet_offload *gro_find_receive_by_type(__be16 type)
3975 struct list_head *offload_head = &offload_base;
3976 struct packet_offload *ptype;
3978 list_for_each_entry_rcu(ptype, offload_head, list) {
3979 if (ptype->type != type || !ptype->callbacks.gro_receive)
3986 struct packet_offload *gro_find_complete_by_type(__be16 type)
3988 struct list_head *offload_head = &offload_base;
3989 struct packet_offload *ptype;
3991 list_for_each_entry_rcu(ptype, offload_head, list) {
3992 if (ptype->type != type || !ptype->callbacks.gro_complete)
3999 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
4003 if (netif_receive_skb_internal(skb))
4011 case GRO_MERGED_FREE:
4012 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
4013 kmem_cache_free(skbuff_head_cache, skb);
4026 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4028 trace_napi_gro_receive_entry(skb);
4030 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
4032 EXPORT_SYMBOL(napi_gro_receive);
4034 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
4036 __skb_pull(skb, skb_headlen(skb));
4037 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
4038 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
4040 skb->dev = napi->dev;
4046 struct sk_buff *napi_get_frags(struct napi_struct *napi)
4048 struct sk_buff *skb = napi->skb;
4051 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
4056 EXPORT_SYMBOL(napi_get_frags);
4058 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
4063 if (netif_receive_skb_internal(skb))
4068 case GRO_MERGED_FREE:
4069 napi_reuse_skb(napi, skb);
4080 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
4082 struct sk_buff *skb = napi->skb;
4086 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) {
4087 napi_reuse_skb(napi, skb);
4090 skb->protocol = eth_type_trans(skb, skb->dev);
4095 gro_result_t napi_gro_frags(struct napi_struct *napi)
4097 struct sk_buff *skb = napi_frags_skb(napi);
4102 trace_napi_gro_frags_entry(skb);
4104 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4106 EXPORT_SYMBOL(napi_gro_frags);
4109 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
4110 * Note: called with local irq disabled, but exits with local irq enabled.
4112 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4115 struct softnet_data *remsd = sd->rps_ipi_list;
4118 sd->rps_ipi_list = NULL;
4122 /* Send pending IPI's to kick RPS processing on remote cpus. */
4124 struct softnet_data *next = remsd->rps_ipi_next;
4126 if (cpu_online(remsd->cpu))
4127 __smp_call_function_single(remsd->cpu,
4136 static int process_backlog(struct napi_struct *napi, int quota)
4139 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4142 /* Check if we have pending ipi, its better to send them now,
4143 * not waiting net_rx_action() end.
4145 if (sd->rps_ipi_list) {
4146 local_irq_disable();
4147 net_rps_action_and_irq_enable(sd);
4150 napi->weight = weight_p;
4151 local_irq_disable();
4152 while (work < quota) {
4153 struct sk_buff *skb;
4156 while ((skb = __skb_dequeue(&sd->process_queue))) {
4158 __netif_receive_skb(skb);
4159 local_irq_disable();
4160 input_queue_head_incr(sd);
4161 if (++work >= quota) {
4168 qlen = skb_queue_len(&sd->input_pkt_queue);
4170 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4171 &sd->process_queue);
4173 if (qlen < quota - work) {
4175 * Inline a custom version of __napi_complete().
4176 * only current cpu owns and manipulates this napi,
4177 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4178 * we can use a plain write instead of clear_bit(),
4179 * and we dont need an smp_mb() memory barrier.
4181 list_del(&napi->poll_list);
4184 quota = work + qlen;
4194 * __napi_schedule - schedule for receive
4195 * @n: entry to schedule
4197 * The entry's receive function will be scheduled to run
4199 void __napi_schedule(struct napi_struct *n)
4201 unsigned long flags;
4203 local_irq_save(flags);
4204 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4205 local_irq_restore(flags);
4207 EXPORT_SYMBOL(__napi_schedule);
4209 void __napi_complete(struct napi_struct *n)
4211 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4212 BUG_ON(n->gro_list);
4214 list_del(&n->poll_list);
4215 smp_mb__before_clear_bit();
4216 clear_bit(NAPI_STATE_SCHED, &n->state);
4218 EXPORT_SYMBOL(__napi_complete);
4220 void napi_complete(struct napi_struct *n)
4222 unsigned long flags;
4225 * don't let napi dequeue from the cpu poll list
4226 * just in case its running on a different cpu
4228 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4231 napi_gro_flush(n, false);
4232 local_irq_save(flags);
4234 local_irq_restore(flags);
4236 EXPORT_SYMBOL(napi_complete);
4238 /* must be called under rcu_read_lock(), as we dont take a reference */
4239 struct napi_struct *napi_by_id(unsigned int napi_id)
4241 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4242 struct napi_struct *napi;
4244 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4245 if (napi->napi_id == napi_id)
4250 EXPORT_SYMBOL_GPL(napi_by_id);
4252 void napi_hash_add(struct napi_struct *napi)
4254 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4256 spin_lock(&napi_hash_lock);
4258 /* 0 is not a valid id, we also skip an id that is taken
4259 * we expect both events to be extremely rare
4262 while (!napi->napi_id) {
4263 napi->napi_id = ++napi_gen_id;
4264 if (napi_by_id(napi->napi_id))
4268 hlist_add_head_rcu(&napi->napi_hash_node,
4269 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4271 spin_unlock(&napi_hash_lock);
4274 EXPORT_SYMBOL_GPL(napi_hash_add);
4276 /* Warning : caller is responsible to make sure rcu grace period
4277 * is respected before freeing memory containing @napi
4279 void napi_hash_del(struct napi_struct *napi)
4281 spin_lock(&napi_hash_lock);
4283 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4284 hlist_del_rcu(&napi->napi_hash_node);
4286 spin_unlock(&napi_hash_lock);
4288 EXPORT_SYMBOL_GPL(napi_hash_del);
4290 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4291 int (*poll)(struct napi_struct *, int), int weight)
4293 INIT_LIST_HEAD(&napi->poll_list);
4294 napi->gro_count = 0;
4295 napi->gro_list = NULL;
4298 if (weight > NAPI_POLL_WEIGHT)
4299 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4301 napi->weight = weight;
4302 list_add(&napi->dev_list, &dev->napi_list);
4304 #ifdef CONFIG_NETPOLL
4305 spin_lock_init(&napi->poll_lock);
4306 napi->poll_owner = -1;
4308 set_bit(NAPI_STATE_SCHED, &napi->state);
4310 EXPORT_SYMBOL(netif_napi_add);
4312 void netif_napi_del(struct napi_struct *napi)
4314 list_del_init(&napi->dev_list);
4315 napi_free_frags(napi);
4317 kfree_skb_list(napi->gro_list);
4318 napi->gro_list = NULL;
4319 napi->gro_count = 0;
4321 EXPORT_SYMBOL(netif_napi_del);
4323 static void net_rx_action(struct softirq_action *h)
4325 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4326 unsigned long time_limit = jiffies + 2;
4327 int budget = netdev_budget;
4330 local_irq_disable();
4332 while (!list_empty(&sd->poll_list)) {
4333 struct napi_struct *n;
4336 /* If softirq window is exhuasted then punt.
4337 * Allow this to run for 2 jiffies since which will allow
4338 * an average latency of 1.5/HZ.
4340 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4345 /* Even though interrupts have been re-enabled, this
4346 * access is safe because interrupts can only add new
4347 * entries to the tail of this list, and only ->poll()
4348 * calls can remove this head entry from the list.
4350 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4352 have = netpoll_poll_lock(n);
4356 /* This NAPI_STATE_SCHED test is for avoiding a race
4357 * with netpoll's poll_napi(). Only the entity which
4358 * obtains the lock and sees NAPI_STATE_SCHED set will
4359 * actually make the ->poll() call. Therefore we avoid
4360 * accidentally calling ->poll() when NAPI is not scheduled.
4363 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4364 work = n->poll(n, weight);
4368 WARN_ON_ONCE(work > weight);
4372 local_irq_disable();
4374 /* Drivers must not modify the NAPI state if they
4375 * consume the entire weight. In such cases this code
4376 * still "owns" the NAPI instance and therefore can
4377 * move the instance around on the list at-will.
4379 if (unlikely(work == weight)) {
4380 if (unlikely(napi_disable_pending(n))) {
4383 local_irq_disable();
4386 /* flush too old packets
4387 * If HZ < 1000, flush all packets.
4390 napi_gro_flush(n, HZ >= 1000);
4391 local_irq_disable();
4393 list_move_tail(&n->poll_list, &sd->poll_list);
4397 netpoll_poll_unlock(have);
4400 net_rps_action_and_irq_enable(sd);
4402 #ifdef CONFIG_NET_DMA
4404 * There may not be any more sk_buffs coming right now, so push
4405 * any pending DMA copies to hardware
4407 dma_issue_pending_all();
4414 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4418 struct netdev_adjacent {
4419 struct net_device *dev;
4421 /* upper master flag, there can only be one master device per list */
4424 /* counter for the number of times this device was added to us */
4427 /* private field for the users */
4430 struct list_head list;
4431 struct rcu_head rcu;
4434 static struct netdev_adjacent *__netdev_find_adj(struct net_device *dev,
4435 struct net_device *adj_dev,
4436 struct list_head *adj_list)
4438 struct netdev_adjacent *adj;
4440 list_for_each_entry(adj, adj_list, list) {
4441 if (adj->dev == adj_dev)
4448 * netdev_has_upper_dev - Check if device is linked to an upper device
4450 * @upper_dev: upper device to check
4452 * Find out if a device is linked to specified upper device and return true
4453 * in case it is. Note that this checks only immediate upper device,
4454 * not through a complete stack of devices. The caller must hold the RTNL lock.
4456 bool netdev_has_upper_dev(struct net_device *dev,
4457 struct net_device *upper_dev)
4461 return __netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper);
4463 EXPORT_SYMBOL(netdev_has_upper_dev);
4466 * netdev_has_any_upper_dev - Check if device is linked to some device
4469 * Find out if a device is linked to an upper device and return true in case
4470 * it is. The caller must hold the RTNL lock.
4472 static bool netdev_has_any_upper_dev(struct net_device *dev)
4476 return !list_empty(&dev->all_adj_list.upper);
4480 * netdev_master_upper_dev_get - Get master upper device
4483 * Find a master upper device and return pointer to it or NULL in case
4484 * it's not there. The caller must hold the RTNL lock.
4486 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4488 struct netdev_adjacent *upper;
4492 if (list_empty(&dev->adj_list.upper))
4495 upper = list_first_entry(&dev->adj_list.upper,
4496 struct netdev_adjacent, list);
4497 if (likely(upper->master))
4501 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4503 void *netdev_adjacent_get_private(struct list_head *adj_list)
4505 struct netdev_adjacent *adj;
4507 adj = list_entry(adj_list, struct netdev_adjacent, list);
4509 return adj->private;
4511 EXPORT_SYMBOL(netdev_adjacent_get_private);
4514 * netdev_all_upper_get_next_dev_rcu - Get the next dev from upper list
4516 * @iter: list_head ** of the current position
4518 * Gets the next device from the dev's upper list, starting from iter
4519 * position. The caller must hold RCU read lock.
4521 struct net_device *netdev_all_upper_get_next_dev_rcu(struct net_device *dev,
4522 struct list_head **iter)
4524 struct netdev_adjacent *upper;
4526 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
4528 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4530 if (&upper->list == &dev->all_adj_list.upper)
4533 *iter = &upper->list;
4537 EXPORT_SYMBOL(netdev_all_upper_get_next_dev_rcu);
4540 * netdev_lower_get_next_private - Get the next ->private from the
4541 * lower neighbour list
4543 * @iter: list_head ** of the current position
4545 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4546 * list, starting from iter position. The caller must hold either hold the
4547 * RTNL lock or its own locking that guarantees that the neighbour lower
4548 * list will remain unchainged.
4550 void *netdev_lower_get_next_private(struct net_device *dev,
4551 struct list_head **iter)
4553 struct netdev_adjacent *lower;
4555 lower = list_entry(*iter, struct netdev_adjacent, list);
4557 if (&lower->list == &dev->adj_list.lower)
4561 *iter = lower->list.next;
4563 return lower->private;
4565 EXPORT_SYMBOL(netdev_lower_get_next_private);
4568 * netdev_lower_get_next_private_rcu - Get the next ->private from the
4569 * lower neighbour list, RCU
4572 * @iter: list_head ** of the current position
4574 * Gets the next netdev_adjacent->private from the dev's lower neighbour
4575 * list, starting from iter position. The caller must hold RCU read lock.
4577 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
4578 struct list_head **iter)
4580 struct netdev_adjacent *lower;
4582 WARN_ON_ONCE(!rcu_read_lock_held());
4584 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
4586 if (&lower->list == &dev->adj_list.lower)
4590 *iter = &lower->list;
4592 return lower->private;
4594 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
4597 * netdev_lower_get_first_private_rcu - Get the first ->private from the
4598 * lower neighbour list, RCU
4602 * Gets the first netdev_adjacent->private from the dev's lower neighbour
4603 * list. The caller must hold RCU read lock.
4605 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
4607 struct netdev_adjacent *lower;
4609 lower = list_first_or_null_rcu(&dev->adj_list.lower,
4610 struct netdev_adjacent, list);
4612 return lower->private;
4615 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
4618 * netdev_master_upper_dev_get_rcu - Get master upper device
4621 * Find a master upper device and return pointer to it or NULL in case
4622 * it's not there. The caller must hold the RCU read lock.
4624 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4626 struct netdev_adjacent *upper;
4628 upper = list_first_or_null_rcu(&dev->adj_list.upper,
4629 struct netdev_adjacent, list);
4630 if (upper && likely(upper->master))
4634 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4636 int netdev_adjacent_sysfs_add(struct net_device *dev,
4637 struct net_device *adj_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", adj_dev->name);
4643 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
4646 void netdev_adjacent_sysfs_del(struct net_device *dev,
4648 struct list_head *dev_list)
4650 char linkname[IFNAMSIZ+7];
4651 sprintf(linkname, dev_list == &dev->adj_list.upper ?
4652 "upper_%s" : "lower_%s", name);
4653 sysfs_remove_link(&(dev->dev.kobj), linkname);
4656 #define netdev_adjacent_is_neigh_list(dev, dev_list) \
4657 (dev_list == &dev->adj_list.upper || \
4658 dev_list == &dev->adj_list.lower)
4660 static int __netdev_adjacent_dev_insert(struct net_device *dev,
4661 struct net_device *adj_dev,
4662 struct list_head *dev_list,
4663 void *private, bool master)
4665 struct netdev_adjacent *adj;
4668 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4675 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
4680 adj->master = master;
4682 adj->private = private;
4685 pr_debug("dev_hold for %s, because of link added from %s to %s\n",
4686 adj_dev->name, dev->name, adj_dev->name);
4688 if (netdev_adjacent_is_neigh_list(dev, dev_list)) {
4689 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
4694 /* Ensure that master link is always the first item in list. */
4696 ret = sysfs_create_link(&(dev->dev.kobj),
4697 &(adj_dev->dev.kobj), "master");
4699 goto remove_symlinks;
4701 list_add_rcu(&adj->list, dev_list);
4703 list_add_tail_rcu(&adj->list, dev_list);
4709 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4710 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4718 static void __netdev_adjacent_dev_remove(struct net_device *dev,
4719 struct net_device *adj_dev,
4720 struct list_head *dev_list)
4722 struct netdev_adjacent *adj;
4724 adj = __netdev_find_adj(dev, adj_dev, dev_list);
4727 pr_err("tried to remove device %s from %s\n",
4728 dev->name, adj_dev->name);
4732 if (adj->ref_nr > 1) {
4733 pr_debug("%s to %s ref_nr-- = %d\n", dev->name, adj_dev->name,
4740 sysfs_remove_link(&(dev->dev.kobj), "master");
4742 if (netdev_adjacent_is_neigh_list(dev, dev_list))
4743 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
4745 list_del_rcu(&adj->list);
4746 pr_debug("dev_put for %s, because link removed from %s to %s\n",
4747 adj_dev->name, dev->name, adj_dev->name);
4749 kfree_rcu(adj, rcu);
4752 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
4753 struct net_device *upper_dev,
4754 struct list_head *up_list,
4755 struct list_head *down_list,
4756 void *private, bool master)
4760 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, private,
4765 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, private,
4768 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4775 static int __netdev_adjacent_dev_link(struct net_device *dev,
4776 struct net_device *upper_dev)
4778 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
4779 &dev->all_adj_list.upper,
4780 &upper_dev->all_adj_list.lower,
4784 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
4785 struct net_device *upper_dev,
4786 struct list_head *up_list,
4787 struct list_head *down_list)
4789 __netdev_adjacent_dev_remove(dev, upper_dev, up_list);
4790 __netdev_adjacent_dev_remove(upper_dev, dev, down_list);
4793 static void __netdev_adjacent_dev_unlink(struct net_device *dev,
4794 struct net_device *upper_dev)
4796 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4797 &dev->all_adj_list.upper,
4798 &upper_dev->all_adj_list.lower);
4801 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
4802 struct net_device *upper_dev,
4803 void *private, bool master)
4805 int ret = __netdev_adjacent_dev_link(dev, upper_dev);
4810 ret = __netdev_adjacent_dev_link_lists(dev, upper_dev,
4811 &dev->adj_list.upper,
4812 &upper_dev->adj_list.lower,
4815 __netdev_adjacent_dev_unlink(dev, upper_dev);
4822 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
4823 struct net_device *upper_dev)
4825 __netdev_adjacent_dev_unlink(dev, upper_dev);
4826 __netdev_adjacent_dev_unlink_lists(dev, upper_dev,
4827 &dev->adj_list.upper,
4828 &upper_dev->adj_list.lower);
4831 static int __netdev_upper_dev_link(struct net_device *dev,
4832 struct net_device *upper_dev, bool master,
4835 struct netdev_adjacent *i, *j, *to_i, *to_j;
4840 if (dev == upper_dev)
4843 /* To prevent loops, check if dev is not upper device to upper_dev. */
4844 if (__netdev_find_adj(upper_dev, dev, &upper_dev->all_adj_list.upper))
4847 if (__netdev_find_adj(dev, upper_dev, &dev->all_adj_list.upper))
4850 if (master && netdev_master_upper_dev_get(dev))
4853 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, private,
4858 /* Now that we linked these devs, make all the upper_dev's
4859 * all_adj_list.upper visible to every dev's all_adj_list.lower an
4860 * versa, and don't forget the devices itself. All of these
4861 * links are non-neighbours.
4863 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4864 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4865 pr_debug("Interlinking %s with %s, non-neighbour\n",
4866 i->dev->name, j->dev->name);
4867 ret = __netdev_adjacent_dev_link(i->dev, j->dev);
4873 /* add dev to every upper_dev's upper device */
4874 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4875 pr_debug("linking %s's upper device %s with %s\n",
4876 upper_dev->name, i->dev->name, dev->name);
4877 ret = __netdev_adjacent_dev_link(dev, i->dev);
4879 goto rollback_upper_mesh;
4882 /* add upper_dev to every dev's lower device */
4883 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4884 pr_debug("linking %s's lower device %s with %s\n", dev->name,
4885 i->dev->name, upper_dev->name);
4886 ret = __netdev_adjacent_dev_link(i->dev, upper_dev);
4888 goto rollback_lower_mesh;
4891 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4894 rollback_lower_mesh:
4896 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4899 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
4904 rollback_upper_mesh:
4906 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list) {
4909 __netdev_adjacent_dev_unlink(dev, i->dev);
4917 list_for_each_entry(i, &dev->all_adj_list.lower, list) {
4918 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list) {
4919 if (i == to_i && j == to_j)
4921 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4927 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4933 * netdev_upper_dev_link - Add a link to the upper device
4935 * @upper_dev: new upper device
4937 * Adds a link to device which is upper to this one. The caller must hold
4938 * the RTNL lock. On a failure a negative errno code is returned.
4939 * On success the reference counts are adjusted and the function
4942 int netdev_upper_dev_link(struct net_device *dev,
4943 struct net_device *upper_dev)
4945 return __netdev_upper_dev_link(dev, upper_dev, false, NULL);
4947 EXPORT_SYMBOL(netdev_upper_dev_link);
4950 * netdev_master_upper_dev_link - Add a master link to the upper device
4952 * @upper_dev: new upper device
4954 * Adds a link to device which is upper to this one. In this case, only
4955 * one master upper device can be linked, although other non-master devices
4956 * might be linked as well. The caller must hold the RTNL lock.
4957 * On a failure a negative errno code is returned. On success the reference
4958 * counts are adjusted and the function returns zero.
4960 int netdev_master_upper_dev_link(struct net_device *dev,
4961 struct net_device *upper_dev)
4963 return __netdev_upper_dev_link(dev, upper_dev, true, NULL);
4965 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4967 int netdev_master_upper_dev_link_private(struct net_device *dev,
4968 struct net_device *upper_dev,
4971 return __netdev_upper_dev_link(dev, upper_dev, true, private);
4973 EXPORT_SYMBOL(netdev_master_upper_dev_link_private);
4976 * netdev_upper_dev_unlink - Removes a link to upper device
4978 * @upper_dev: new upper device
4980 * Removes a link to device which is upper to this one. The caller must hold
4983 void netdev_upper_dev_unlink(struct net_device *dev,
4984 struct net_device *upper_dev)
4986 struct netdev_adjacent *i, *j;
4989 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
4991 /* Here is the tricky part. We must remove all dev's lower
4992 * devices from all upper_dev's upper devices and vice
4993 * versa, to maintain the graph relationship.
4995 list_for_each_entry(i, &dev->all_adj_list.lower, list)
4996 list_for_each_entry(j, &upper_dev->all_adj_list.upper, list)
4997 __netdev_adjacent_dev_unlink(i->dev, j->dev);
4999 /* remove also the devices itself from lower/upper device
5002 list_for_each_entry(i, &dev->all_adj_list.lower, list)
5003 __netdev_adjacent_dev_unlink(i->dev, upper_dev);
5005 list_for_each_entry(i, &upper_dev->all_adj_list.upper, list)
5006 __netdev_adjacent_dev_unlink(dev, i->dev);
5008 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
5010 EXPORT_SYMBOL(netdev_upper_dev_unlink);
5012 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
5014 struct netdev_adjacent *iter;
5016 list_for_each_entry(iter, &dev->adj_list.upper, list) {
5017 netdev_adjacent_sysfs_del(iter->dev, oldname,
5018 &iter->dev->adj_list.lower);
5019 netdev_adjacent_sysfs_add(iter->dev, dev,
5020 &iter->dev->adj_list.lower);
5023 list_for_each_entry(iter, &dev->adj_list.lower, list) {
5024 netdev_adjacent_sysfs_del(iter->dev, oldname,
5025 &iter->dev->adj_list.upper);
5026 netdev_adjacent_sysfs_add(iter->dev, dev,
5027 &iter->dev->adj_list.upper);
5031 void *netdev_lower_dev_get_private(struct net_device *dev,
5032 struct net_device *lower_dev)
5034 struct netdev_adjacent *lower;
5038 lower = __netdev_find_adj(dev, lower_dev, &dev->adj_list.lower);
5042 return lower->private;
5044 EXPORT_SYMBOL(netdev_lower_dev_get_private);
5046 static void dev_change_rx_flags(struct net_device *dev, int flags)
5048 const struct net_device_ops *ops = dev->netdev_ops;
5050 if (ops->ndo_change_rx_flags)
5051 ops->ndo_change_rx_flags(dev, flags);
5054 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
5056 unsigned int old_flags = dev->flags;
5062 dev->flags |= IFF_PROMISC;
5063 dev->promiscuity += inc;
5064 if (dev->promiscuity == 0) {
5067 * If inc causes overflow, untouch promisc and return error.
5070 dev->flags &= ~IFF_PROMISC;
5072 dev->promiscuity -= inc;
5073 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
5078 if (dev->flags != old_flags) {
5079 pr_info("device %s %s promiscuous mode\n",
5081 dev->flags & IFF_PROMISC ? "entered" : "left");
5082 if (audit_enabled) {
5083 current_uid_gid(&uid, &gid);
5084 audit_log(current->audit_context, GFP_ATOMIC,
5085 AUDIT_ANOM_PROMISCUOUS,
5086 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
5087 dev->name, (dev->flags & IFF_PROMISC),
5088 (old_flags & IFF_PROMISC),
5089 from_kuid(&init_user_ns, audit_get_loginuid(current)),
5090 from_kuid(&init_user_ns, uid),
5091 from_kgid(&init_user_ns, gid),
5092 audit_get_sessionid(current));
5095 dev_change_rx_flags(dev, IFF_PROMISC);
5098 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
5103 * dev_set_promiscuity - update promiscuity count on a device
5107 * Add or remove promiscuity from a device. While the count in the device
5108 * remains above zero the interface remains promiscuous. Once it hits zero
5109 * the device reverts back to normal filtering operation. A negative inc
5110 * value is used to drop promiscuity on the device.
5111 * Return 0 if successful or a negative errno code on error.
5113 int dev_set_promiscuity(struct net_device *dev, int inc)
5115 unsigned int old_flags = dev->flags;
5118 err = __dev_set_promiscuity(dev, inc, true);
5121 if (dev->flags != old_flags)
5122 dev_set_rx_mode(dev);
5125 EXPORT_SYMBOL(dev_set_promiscuity);
5127 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
5129 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
5133 dev->flags |= IFF_ALLMULTI;
5134 dev->allmulti += inc;
5135 if (dev->allmulti == 0) {
5138 * If inc causes overflow, untouch allmulti and return error.
5141 dev->flags &= ~IFF_ALLMULTI;
5143 dev->allmulti -= inc;
5144 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
5149 if (dev->flags ^ old_flags) {
5150 dev_change_rx_flags(dev, IFF_ALLMULTI);
5151 dev_set_rx_mode(dev);
5153 __dev_notify_flags(dev, old_flags,
5154 dev->gflags ^ old_gflags);
5160 * dev_set_allmulti - update allmulti count on a device
5164 * Add or remove reception of all multicast frames to a device. While the
5165 * count in the device remains above zero the interface remains listening
5166 * to all interfaces. Once it hits zero the device reverts back to normal
5167 * filtering operation. A negative @inc value is used to drop the counter
5168 * when releasing a resource needing all multicasts.
5169 * Return 0 if successful or a negative errno code on error.
5172 int dev_set_allmulti(struct net_device *dev, int inc)
5174 return __dev_set_allmulti(dev, inc, true);
5176 EXPORT_SYMBOL(dev_set_allmulti);
5179 * Upload unicast and multicast address lists to device and
5180 * configure RX filtering. When the device doesn't support unicast
5181 * filtering it is put in promiscuous mode while unicast addresses
5184 void __dev_set_rx_mode(struct net_device *dev)
5186 const struct net_device_ops *ops = dev->netdev_ops;
5188 /* dev_open will call this function so the list will stay sane. */
5189 if (!(dev->flags&IFF_UP))
5192 if (!netif_device_present(dev))
5195 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
5196 /* Unicast addresses changes may only happen under the rtnl,
5197 * therefore calling __dev_set_promiscuity here is safe.
5199 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
5200 __dev_set_promiscuity(dev, 1, false);
5201 dev->uc_promisc = true;
5202 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
5203 __dev_set_promiscuity(dev, -1, false);
5204 dev->uc_promisc = false;
5208 if (ops->ndo_set_rx_mode)
5209 ops->ndo_set_rx_mode(dev);
5212 void dev_set_rx_mode(struct net_device *dev)
5214 netif_addr_lock_bh(dev);
5215 __dev_set_rx_mode(dev);
5216 netif_addr_unlock_bh(dev);
5220 * dev_get_flags - get flags reported to userspace
5223 * Get the combination of flag bits exported through APIs to userspace.
5225 unsigned int dev_get_flags(const struct net_device *dev)
5229 flags = (dev->flags & ~(IFF_PROMISC |
5234 (dev->gflags & (IFF_PROMISC |
5237 if (netif_running(dev)) {
5238 if (netif_oper_up(dev))
5239 flags |= IFF_RUNNING;
5240 if (netif_carrier_ok(dev))
5241 flags |= IFF_LOWER_UP;
5242 if (netif_dormant(dev))
5243 flags |= IFF_DORMANT;
5248 EXPORT_SYMBOL(dev_get_flags);
5250 int __dev_change_flags(struct net_device *dev, unsigned int flags)
5252 unsigned int old_flags = dev->flags;
5258 * Set the flags on our device.
5261 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
5262 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
5264 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
5268 * Load in the correct multicast list now the flags have changed.
5271 if ((old_flags ^ flags) & IFF_MULTICAST)
5272 dev_change_rx_flags(dev, IFF_MULTICAST);
5274 dev_set_rx_mode(dev);
5277 * Have we downed the interface. We handle IFF_UP ourselves
5278 * according to user attempts to set it, rather than blindly
5283 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
5284 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
5287 dev_set_rx_mode(dev);
5290 if ((flags ^ dev->gflags) & IFF_PROMISC) {
5291 int inc = (flags & IFF_PROMISC) ? 1 : -1;
5292 unsigned int old_flags = dev->flags;
5294 dev->gflags ^= IFF_PROMISC;
5296 if (__dev_set_promiscuity(dev, inc, false) >= 0)
5297 if (dev->flags != old_flags)
5298 dev_set_rx_mode(dev);
5301 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
5302 is important. Some (broken) drivers set IFF_PROMISC, when
5303 IFF_ALLMULTI is requested not asking us and not reporting.
5305 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
5306 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
5308 dev->gflags ^= IFF_ALLMULTI;
5309 __dev_set_allmulti(dev, inc, false);
5315 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
5316 unsigned int gchanges)
5318 unsigned int changes = dev->flags ^ old_flags;
5321 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
5323 if (changes & IFF_UP) {
5324 if (dev->flags & IFF_UP)
5325 call_netdevice_notifiers(NETDEV_UP, dev);
5327 call_netdevice_notifiers(NETDEV_DOWN, dev);
5330 if (dev->flags & IFF_UP &&
5331 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
5332 struct netdev_notifier_change_info change_info;
5334 change_info.flags_changed = changes;
5335 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
5341 * dev_change_flags - change device settings
5343 * @flags: device state flags
5345 * Change settings on device based state flags. The flags are
5346 * in the userspace exported format.
5348 int dev_change_flags(struct net_device *dev, unsigned int flags)
5351 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
5353 ret = __dev_change_flags(dev, flags);
5357 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
5358 __dev_notify_flags(dev, old_flags, changes);
5361 EXPORT_SYMBOL(dev_change_flags);
5363 static int __dev_set_mtu(struct net_device *dev, int new_mtu)
5365 const struct net_device_ops *ops = dev->netdev_ops;
5367 if (ops->ndo_change_mtu)
5368 return ops->ndo_change_mtu(dev, new_mtu);
5375 * dev_set_mtu - Change maximum transfer unit
5377 * @new_mtu: new transfer unit
5379 * Change the maximum transfer size of the network device.
5381 int dev_set_mtu(struct net_device *dev, int new_mtu)
5385 if (new_mtu == dev->mtu)
5388 /* MTU must be positive. */
5392 if (!netif_device_present(dev))
5395 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
5396 err = notifier_to_errno(err);
5400 orig_mtu = dev->mtu;
5401 err = __dev_set_mtu(dev, new_mtu);
5404 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5405 err = notifier_to_errno(err);
5407 /* setting mtu back and notifying everyone again,
5408 * so that they have a chance to revert changes.
5410 __dev_set_mtu(dev, orig_mtu);
5411 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
5416 EXPORT_SYMBOL(dev_set_mtu);
5419 * dev_set_group - Change group this device belongs to
5421 * @new_group: group this device should belong to
5423 void dev_set_group(struct net_device *dev, int new_group)
5425 dev->group = new_group;
5427 EXPORT_SYMBOL(dev_set_group);
5430 * dev_set_mac_address - Change Media Access Control Address
5434 * Change the hardware (MAC) address of the device
5436 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
5438 const struct net_device_ops *ops = dev->netdev_ops;
5441 if (!ops->ndo_set_mac_address)
5443 if (sa->sa_family != dev->type)
5445 if (!netif_device_present(dev))
5447 err = ops->ndo_set_mac_address(dev, sa);
5450 dev->addr_assign_type = NET_ADDR_SET;
5451 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
5452 add_device_randomness(dev->dev_addr, dev->addr_len);
5455 EXPORT_SYMBOL(dev_set_mac_address);
5458 * dev_change_carrier - Change device carrier
5460 * @new_carrier: new value
5462 * Change device carrier
5464 int dev_change_carrier(struct net_device *dev, bool new_carrier)
5466 const struct net_device_ops *ops = dev->netdev_ops;
5468 if (!ops->ndo_change_carrier)
5470 if (!netif_device_present(dev))
5472 return ops->ndo_change_carrier(dev, new_carrier);
5474 EXPORT_SYMBOL(dev_change_carrier);
5477 * dev_get_phys_port_id - Get device physical port ID
5481 * Get device physical port ID
5483 int dev_get_phys_port_id(struct net_device *dev,
5484 struct netdev_phys_port_id *ppid)
5486 const struct net_device_ops *ops = dev->netdev_ops;
5488 if (!ops->ndo_get_phys_port_id)
5490 return ops->ndo_get_phys_port_id(dev, ppid);
5492 EXPORT_SYMBOL(dev_get_phys_port_id);
5495 * dev_new_index - allocate an ifindex
5496 * @net: the applicable net namespace
5498 * Returns a suitable unique value for a new device interface
5499 * number. The caller must hold the rtnl semaphore or the
5500 * dev_base_lock to be sure it remains unique.
5502 static int dev_new_index(struct net *net)
5504 int ifindex = net->ifindex;
5508 if (!__dev_get_by_index(net, ifindex))
5509 return net->ifindex = ifindex;
5513 /* Delayed registration/unregisteration */
5514 static LIST_HEAD(net_todo_list);
5515 static DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
5517 static void net_set_todo(struct net_device *dev)
5519 list_add_tail(&dev->todo_list, &net_todo_list);
5520 dev_net(dev)->dev_unreg_count++;
5523 static void rollback_registered_many(struct list_head *head)
5525 struct net_device *dev, *tmp;
5526 LIST_HEAD(close_head);
5528 BUG_ON(dev_boot_phase);
5531 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5532 /* Some devices call without registering
5533 * for initialization unwind. Remove those
5534 * devices and proceed with the remaining.
5536 if (dev->reg_state == NETREG_UNINITIALIZED) {
5537 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5541 list_del(&dev->unreg_list);
5544 dev->dismantle = true;
5545 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5548 /* If device is running, close it first. */
5549 list_for_each_entry(dev, head, unreg_list)
5550 list_add_tail(&dev->close_list, &close_head);
5551 dev_close_many(&close_head);
5553 list_for_each_entry(dev, head, unreg_list) {
5554 /* And unlink it from device chain. */
5555 unlist_netdevice(dev);
5557 dev->reg_state = NETREG_UNREGISTERING;
5562 list_for_each_entry(dev, head, unreg_list) {
5563 /* Shutdown queueing discipline. */
5567 /* Notify protocols, that we are about to destroy
5568 this device. They should clean all the things.
5570 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5572 if (!dev->rtnl_link_ops ||
5573 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5574 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
5577 * Flush the unicast and multicast chains
5582 if (dev->netdev_ops->ndo_uninit)
5583 dev->netdev_ops->ndo_uninit(dev);
5585 /* Notifier chain MUST detach us all upper devices. */
5586 WARN_ON(netdev_has_any_upper_dev(dev));
5588 /* Remove entries from kobject tree */
5589 netdev_unregister_kobject(dev);
5591 /* Remove XPS queueing entries */
5592 netif_reset_xps_queues_gt(dev, 0);
5598 list_for_each_entry(dev, head, unreg_list)
5602 static void rollback_registered(struct net_device *dev)
5606 list_add(&dev->unreg_list, &single);
5607 rollback_registered_many(&single);
5611 static netdev_features_t netdev_fix_features(struct net_device *dev,
5612 netdev_features_t features)
5614 /* Fix illegal checksum combinations */
5615 if ((features & NETIF_F_HW_CSUM) &&
5616 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5617 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5618 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5621 /* TSO requires that SG is present as well. */
5622 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5623 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5624 features &= ~NETIF_F_ALL_TSO;
5627 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5628 !(features & NETIF_F_IP_CSUM)) {
5629 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5630 features &= ~NETIF_F_TSO;
5631 features &= ~NETIF_F_TSO_ECN;
5634 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5635 !(features & NETIF_F_IPV6_CSUM)) {
5636 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5637 features &= ~NETIF_F_TSO6;
5640 /* TSO ECN requires that TSO is present as well. */
5641 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5642 features &= ~NETIF_F_TSO_ECN;
5644 /* Software GSO depends on SG. */
5645 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5646 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5647 features &= ~NETIF_F_GSO;
5650 /* UFO needs SG and checksumming */
5651 if (features & NETIF_F_UFO) {
5652 /* maybe split UFO into V4 and V6? */
5653 if (!((features & NETIF_F_GEN_CSUM) ||
5654 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5655 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5657 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5658 features &= ~NETIF_F_UFO;
5661 if (!(features & NETIF_F_SG)) {
5663 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5664 features &= ~NETIF_F_UFO;
5671 int __netdev_update_features(struct net_device *dev)
5673 netdev_features_t features;
5678 features = netdev_get_wanted_features(dev);
5680 if (dev->netdev_ops->ndo_fix_features)
5681 features = dev->netdev_ops->ndo_fix_features(dev, features);
5683 /* driver might be less strict about feature dependencies */
5684 features = netdev_fix_features(dev, features);
5686 if (dev->features == features)
5689 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5690 &dev->features, &features);
5692 if (dev->netdev_ops->ndo_set_features)
5693 err = dev->netdev_ops->ndo_set_features(dev, features);
5695 if (unlikely(err < 0)) {
5697 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5698 err, &features, &dev->features);
5703 dev->features = features;
5709 * netdev_update_features - recalculate device features
5710 * @dev: the device to check
5712 * Recalculate dev->features set and send notifications if it
5713 * has changed. Should be called after driver or hardware dependent
5714 * conditions might have changed that influence the features.
5716 void netdev_update_features(struct net_device *dev)
5718 if (__netdev_update_features(dev))
5719 netdev_features_change(dev);
5721 EXPORT_SYMBOL(netdev_update_features);
5724 * netdev_change_features - recalculate device features
5725 * @dev: the device to check
5727 * Recalculate dev->features set and send notifications even
5728 * if they have not changed. Should be called instead of
5729 * netdev_update_features() if also dev->vlan_features might
5730 * have changed to allow the changes to be propagated to stacked
5733 void netdev_change_features(struct net_device *dev)
5735 __netdev_update_features(dev);
5736 netdev_features_change(dev);
5738 EXPORT_SYMBOL(netdev_change_features);
5741 * netif_stacked_transfer_operstate - transfer operstate
5742 * @rootdev: the root or lower level device to transfer state from
5743 * @dev: the device to transfer operstate to
5745 * Transfer operational state from root to device. This is normally
5746 * called when a stacking relationship exists between the root
5747 * device and the device(a leaf device).
5749 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5750 struct net_device *dev)
5752 if (rootdev->operstate == IF_OPER_DORMANT)
5753 netif_dormant_on(dev);
5755 netif_dormant_off(dev);
5757 if (netif_carrier_ok(rootdev)) {
5758 if (!netif_carrier_ok(dev))
5759 netif_carrier_on(dev);
5761 if (netif_carrier_ok(dev))
5762 netif_carrier_off(dev);
5765 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5768 static int netif_alloc_rx_queues(struct net_device *dev)
5770 unsigned int i, count = dev->num_rx_queues;
5771 struct netdev_rx_queue *rx;
5775 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5781 for (i = 0; i < count; i++)
5787 static void netdev_init_one_queue(struct net_device *dev,
5788 struct netdev_queue *queue, void *_unused)
5790 /* Initialize queue lock */
5791 spin_lock_init(&queue->_xmit_lock);
5792 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5793 queue->xmit_lock_owner = -1;
5794 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5797 dql_init(&queue->dql, HZ);
5801 static void netif_free_tx_queues(struct net_device *dev)
5803 if (is_vmalloc_addr(dev->_tx))
5809 static int netif_alloc_netdev_queues(struct net_device *dev)
5811 unsigned int count = dev->num_tx_queues;
5812 struct netdev_queue *tx;
5813 size_t sz = count * sizeof(*tx);
5815 BUG_ON(count < 1 || count > 0xffff);
5817 tx = kzalloc(sz, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
5825 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5826 spin_lock_init(&dev->tx_global_lock);
5832 * register_netdevice - register a network device
5833 * @dev: device to register
5835 * Take a completed network device structure and add it to the kernel
5836 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5837 * chain. 0 is returned on success. A negative errno code is returned
5838 * on a failure to set up the device, or if the name is a duplicate.
5840 * Callers must hold the rtnl semaphore. You may want
5841 * register_netdev() instead of this.
5844 * The locking appears insufficient to guarantee two parallel registers
5845 * will not get the same name.
5848 int register_netdevice(struct net_device *dev)
5851 struct net *net = dev_net(dev);
5853 BUG_ON(dev_boot_phase);
5858 /* When net_device's are persistent, this will be fatal. */
5859 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5862 spin_lock_init(&dev->addr_list_lock);
5863 netdev_set_addr_lockdep_class(dev);
5867 ret = dev_get_valid_name(net, dev, dev->name);
5871 /* Init, if this function is available */
5872 if (dev->netdev_ops->ndo_init) {
5873 ret = dev->netdev_ops->ndo_init(dev);
5881 if (((dev->hw_features | dev->features) &
5882 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5883 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5884 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5885 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5892 dev->ifindex = dev_new_index(net);
5893 else if (__dev_get_by_index(net, dev->ifindex))
5896 if (dev->iflink == -1)
5897 dev->iflink = dev->ifindex;
5899 /* Transfer changeable features to wanted_features and enable
5900 * software offloads (GSO and GRO).
5902 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5903 dev->features |= NETIF_F_SOFT_FEATURES;
5904 dev->wanted_features = dev->features & dev->hw_features;
5906 if (!(dev->flags & IFF_LOOPBACK)) {
5907 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5910 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5912 dev->vlan_features |= NETIF_F_HIGHDMA;
5914 /* Make NETIF_F_SG inheritable to tunnel devices.
5916 dev->hw_enc_features |= NETIF_F_SG;
5918 /* Make NETIF_F_SG inheritable to MPLS.
5920 dev->mpls_features |= NETIF_F_SG;
5922 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5923 ret = notifier_to_errno(ret);
5927 ret = netdev_register_kobject(dev);
5930 dev->reg_state = NETREG_REGISTERED;
5932 __netdev_update_features(dev);
5935 * Default initial state at registry is that the
5936 * device is present.
5939 set_bit(__LINK_STATE_PRESENT, &dev->state);
5941 linkwatch_init_dev(dev);
5943 dev_init_scheduler(dev);
5945 list_netdevice(dev);
5946 add_device_randomness(dev->dev_addr, dev->addr_len);
5948 /* If the device has permanent device address, driver should
5949 * set dev_addr and also addr_assign_type should be set to
5950 * NET_ADDR_PERM (default value).
5952 if (dev->addr_assign_type == NET_ADDR_PERM)
5953 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5955 /* Notify protocols, that a new device appeared. */
5956 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5957 ret = notifier_to_errno(ret);
5959 rollback_registered(dev);
5960 dev->reg_state = NETREG_UNREGISTERED;
5963 * Prevent userspace races by waiting until the network
5964 * device is fully setup before sending notifications.
5966 if (!dev->rtnl_link_ops ||
5967 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5968 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
5974 if (dev->netdev_ops->ndo_uninit)
5975 dev->netdev_ops->ndo_uninit(dev);
5978 EXPORT_SYMBOL(register_netdevice);
5981 * init_dummy_netdev - init a dummy network device for NAPI
5982 * @dev: device to init
5984 * This takes a network device structure and initialize the minimum
5985 * amount of fields so it can be used to schedule NAPI polls without
5986 * registering a full blown interface. This is to be used by drivers
5987 * that need to tie several hardware interfaces to a single NAPI
5988 * poll scheduler due to HW limitations.
5990 int init_dummy_netdev(struct net_device *dev)
5992 /* Clear everything. Note we don't initialize spinlocks
5993 * are they aren't supposed to be taken by any of the
5994 * NAPI code and this dummy netdev is supposed to be
5995 * only ever used for NAPI polls
5997 memset(dev, 0, sizeof(struct net_device));
5999 /* make sure we BUG if trying to hit standard
6000 * register/unregister code path
6002 dev->reg_state = NETREG_DUMMY;
6004 /* NAPI wants this */
6005 INIT_LIST_HEAD(&dev->napi_list);
6007 /* a dummy interface is started by default */
6008 set_bit(__LINK_STATE_PRESENT, &dev->state);
6009 set_bit(__LINK_STATE_START, &dev->state);
6011 /* Note : We dont allocate pcpu_refcnt for dummy devices,
6012 * because users of this 'device' dont need to change
6018 EXPORT_SYMBOL_GPL(init_dummy_netdev);
6022 * register_netdev - register a network device
6023 * @dev: device to register
6025 * Take a completed network device structure and add it to the kernel
6026 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
6027 * chain. 0 is returned on success. A negative errno code is returned
6028 * on a failure to set up the device, or if the name is a duplicate.
6030 * This is a wrapper around register_netdevice that takes the rtnl semaphore
6031 * and expands the device name if you passed a format string to
6034 int register_netdev(struct net_device *dev)
6039 err = register_netdevice(dev);
6043 EXPORT_SYMBOL(register_netdev);
6045 int netdev_refcnt_read(const struct net_device *dev)
6049 for_each_possible_cpu(i)
6050 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
6053 EXPORT_SYMBOL(netdev_refcnt_read);
6056 * netdev_wait_allrefs - wait until all references are gone.
6057 * @dev: target net_device
6059 * This is called when unregistering network devices.
6061 * Any protocol or device that holds a reference should register
6062 * for netdevice notification, and cleanup and put back the
6063 * reference if they receive an UNREGISTER event.
6064 * We can get stuck here if buggy protocols don't correctly
6067 static void netdev_wait_allrefs(struct net_device *dev)
6069 unsigned long rebroadcast_time, warning_time;
6072 linkwatch_forget_dev(dev);
6074 rebroadcast_time = warning_time = jiffies;
6075 refcnt = netdev_refcnt_read(dev);
6077 while (refcnt != 0) {
6078 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
6081 /* Rebroadcast unregister notification */
6082 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6088 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6089 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
6091 /* We must not have linkwatch events
6092 * pending on unregister. If this
6093 * happens, we simply run the queue
6094 * unscheduled, resulting in a noop
6097 linkwatch_run_queue();
6102 rebroadcast_time = jiffies;
6107 refcnt = netdev_refcnt_read(dev);
6109 if (time_after(jiffies, warning_time + 10 * HZ)) {
6110 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
6112 warning_time = jiffies;
6121 * register_netdevice(x1);
6122 * register_netdevice(x2);
6124 * unregister_netdevice(y1);
6125 * unregister_netdevice(y2);
6131 * We are invoked by rtnl_unlock().
6132 * This allows us to deal with problems:
6133 * 1) We can delete sysfs objects which invoke hotplug
6134 * without deadlocking with linkwatch via keventd.
6135 * 2) Since we run with the RTNL semaphore not held, we can sleep
6136 * safely in order to wait for the netdev refcnt to drop to zero.
6138 * We must not return until all unregister events added during
6139 * the interval the lock was held have been completed.
6141 void netdev_run_todo(void)
6143 struct list_head list;
6145 /* Snapshot list, allow later requests */
6146 list_replace_init(&net_todo_list, &list);
6151 /* Wait for rcu callbacks to finish before next phase */
6152 if (!list_empty(&list))
6155 while (!list_empty(&list)) {
6156 struct net_device *dev
6157 = list_first_entry(&list, struct net_device, todo_list);
6158 list_del(&dev->todo_list);
6161 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6164 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
6165 pr_err("network todo '%s' but state %d\n",
6166 dev->name, dev->reg_state);
6171 dev->reg_state = NETREG_UNREGISTERED;
6173 on_each_cpu(flush_backlog, dev, 1);
6175 netdev_wait_allrefs(dev);
6178 BUG_ON(netdev_refcnt_read(dev));
6179 WARN_ON(rcu_access_pointer(dev->ip_ptr));
6180 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
6181 WARN_ON(dev->dn_ptr);
6183 if (dev->destructor)
6184 dev->destructor(dev);
6186 /* Report a network device has been unregistered */
6188 dev_net(dev)->dev_unreg_count--;
6190 wake_up(&netdev_unregistering_wq);
6192 /* Free network device */
6193 kobject_put(&dev->dev.kobj);
6197 /* Convert net_device_stats to rtnl_link_stats64. They have the same
6198 * fields in the same order, with only the type differing.
6200 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
6201 const struct net_device_stats *netdev_stats)
6203 #if BITS_PER_LONG == 64
6204 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
6205 memcpy(stats64, netdev_stats, sizeof(*stats64));
6207 size_t i, n = sizeof(*stats64) / sizeof(u64);
6208 const unsigned long *src = (const unsigned long *)netdev_stats;
6209 u64 *dst = (u64 *)stats64;
6211 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
6212 sizeof(*stats64) / sizeof(u64));
6213 for (i = 0; i < n; i++)
6217 EXPORT_SYMBOL(netdev_stats_to_stats64);
6220 * dev_get_stats - get network device statistics
6221 * @dev: device to get statistics from
6222 * @storage: place to store stats
6224 * Get network statistics from device. Return @storage.
6225 * The device driver may provide its own method by setting
6226 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
6227 * otherwise the internal statistics structure is used.
6229 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
6230 struct rtnl_link_stats64 *storage)
6232 const struct net_device_ops *ops = dev->netdev_ops;
6234 if (ops->ndo_get_stats64) {
6235 memset(storage, 0, sizeof(*storage));
6236 ops->ndo_get_stats64(dev, storage);
6237 } else if (ops->ndo_get_stats) {
6238 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
6240 netdev_stats_to_stats64(storage, &dev->stats);
6242 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
6245 EXPORT_SYMBOL(dev_get_stats);
6247 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
6249 struct netdev_queue *queue = dev_ingress_queue(dev);
6251 #ifdef CONFIG_NET_CLS_ACT
6254 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
6257 netdev_init_one_queue(dev, queue, NULL);
6258 queue->qdisc = &noop_qdisc;
6259 queue->qdisc_sleeping = &noop_qdisc;
6260 rcu_assign_pointer(dev->ingress_queue, queue);
6265 static const struct ethtool_ops default_ethtool_ops;
6267 void netdev_set_default_ethtool_ops(struct net_device *dev,
6268 const struct ethtool_ops *ops)
6270 if (dev->ethtool_ops == &default_ethtool_ops)
6271 dev->ethtool_ops = ops;
6273 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
6275 void netdev_freemem(struct net_device *dev)
6277 char *addr = (char *)dev - dev->padded;
6279 if (is_vmalloc_addr(addr))
6286 * alloc_netdev_mqs - allocate network device
6287 * @sizeof_priv: size of private data to allocate space for
6288 * @name: device name format string
6289 * @setup: callback to initialize device
6290 * @txqs: the number of TX subqueues to allocate
6291 * @rxqs: the number of RX subqueues to allocate
6293 * Allocates a struct net_device with private data area for driver use
6294 * and performs basic initialization. Also allocates subquue structs
6295 * for each queue on the device.
6297 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
6298 void (*setup)(struct net_device *),
6299 unsigned int txqs, unsigned int rxqs)
6301 struct net_device *dev;
6303 struct net_device *p;
6305 BUG_ON(strlen(name) >= sizeof(dev->name));
6308 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
6314 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
6319 alloc_size = sizeof(struct net_device);
6321 /* ensure 32-byte alignment of private area */
6322 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
6323 alloc_size += sizeof_priv;
6325 /* ensure 32-byte alignment of whole construct */
6326 alloc_size += NETDEV_ALIGN - 1;
6328 p = kzalloc(alloc_size, GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
6330 p = vzalloc(alloc_size);
6334 dev = PTR_ALIGN(p, NETDEV_ALIGN);
6335 dev->padded = (char *)dev - (char *)p;
6337 dev->pcpu_refcnt = alloc_percpu(int);
6338 if (!dev->pcpu_refcnt)
6341 if (dev_addr_init(dev))
6347 dev_net_set(dev, &init_net);
6349 dev->gso_max_size = GSO_MAX_SIZE;
6350 dev->gso_max_segs = GSO_MAX_SEGS;
6352 INIT_LIST_HEAD(&dev->napi_list);
6353 INIT_LIST_HEAD(&dev->unreg_list);
6354 INIT_LIST_HEAD(&dev->close_list);
6355 INIT_LIST_HEAD(&dev->link_watch_list);
6356 INIT_LIST_HEAD(&dev->adj_list.upper);
6357 INIT_LIST_HEAD(&dev->adj_list.lower);
6358 INIT_LIST_HEAD(&dev->all_adj_list.upper);
6359 INIT_LIST_HEAD(&dev->all_adj_list.lower);
6360 dev->priv_flags = IFF_XMIT_DST_RELEASE;
6363 dev->num_tx_queues = txqs;
6364 dev->real_num_tx_queues = txqs;
6365 if (netif_alloc_netdev_queues(dev))
6369 dev->num_rx_queues = rxqs;
6370 dev->real_num_rx_queues = rxqs;
6371 if (netif_alloc_rx_queues(dev))
6375 strcpy(dev->name, name);
6376 dev->group = INIT_NETDEV_GROUP;
6377 if (!dev->ethtool_ops)
6378 dev->ethtool_ops = &default_ethtool_ops;
6386 free_percpu(dev->pcpu_refcnt);
6387 netif_free_tx_queues(dev);
6393 netdev_freemem(dev);
6396 EXPORT_SYMBOL(alloc_netdev_mqs);
6399 * free_netdev - free network device
6402 * This function does the last stage of destroying an allocated device
6403 * interface. The reference to the device object is released.
6404 * If this is the last reference then it will be freed.
6406 void free_netdev(struct net_device *dev)
6408 struct napi_struct *p, *n;
6410 release_net(dev_net(dev));
6412 netif_free_tx_queues(dev);
6417 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
6419 /* Flush device addresses */
6420 dev_addr_flush(dev);
6422 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
6425 free_percpu(dev->pcpu_refcnt);
6426 dev->pcpu_refcnt = NULL;
6428 /* Compatibility with error handling in drivers */
6429 if (dev->reg_state == NETREG_UNINITIALIZED) {
6430 netdev_freemem(dev);
6434 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6435 dev->reg_state = NETREG_RELEASED;
6437 /* will free via device release */
6438 put_device(&dev->dev);
6440 EXPORT_SYMBOL(free_netdev);
6443 * synchronize_net - Synchronize with packet receive processing
6445 * Wait for packets currently being received to be done.
6446 * Does not block later packets from starting.
6448 void synchronize_net(void)
6451 if (rtnl_is_locked())
6452 synchronize_rcu_expedited();
6456 EXPORT_SYMBOL(synchronize_net);
6459 * unregister_netdevice_queue - remove device from the kernel
6463 * This function shuts down a device interface and removes it
6464 * from the kernel tables.
6465 * If head not NULL, device is queued to be unregistered later.
6467 * Callers must hold the rtnl semaphore. You may want
6468 * unregister_netdev() instead of this.
6471 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6476 list_move_tail(&dev->unreg_list, head);
6478 rollback_registered(dev);
6479 /* Finish processing unregister after unlock */
6483 EXPORT_SYMBOL(unregister_netdevice_queue);
6486 * unregister_netdevice_many - unregister many devices
6487 * @head: list of devices
6489 void unregister_netdevice_many(struct list_head *head)
6491 struct net_device *dev;
6493 if (!list_empty(head)) {
6494 rollback_registered_many(head);
6495 list_for_each_entry(dev, head, unreg_list)
6499 EXPORT_SYMBOL(unregister_netdevice_many);
6502 * unregister_netdev - remove device from the kernel
6505 * This function shuts down a device interface and removes it
6506 * from the kernel tables.
6508 * This is just a wrapper for unregister_netdevice that takes
6509 * the rtnl semaphore. In general you want to use this and not
6510 * unregister_netdevice.
6512 void unregister_netdev(struct net_device *dev)
6515 unregister_netdevice(dev);
6518 EXPORT_SYMBOL(unregister_netdev);
6521 * dev_change_net_namespace - move device to different nethost namespace
6523 * @net: network namespace
6524 * @pat: If not NULL name pattern to try if the current device name
6525 * is already taken in the destination network namespace.
6527 * This function shuts down a device interface and moves it
6528 * to a new network namespace. On success 0 is returned, on
6529 * a failure a netagive errno code is returned.
6531 * Callers must hold the rtnl semaphore.
6534 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6540 /* Don't allow namespace local devices to be moved. */
6542 if (dev->features & NETIF_F_NETNS_LOCAL)
6545 /* Ensure the device has been registrered */
6546 if (dev->reg_state != NETREG_REGISTERED)
6549 /* Get out if there is nothing todo */
6551 if (net_eq(dev_net(dev), net))
6554 /* Pick the destination device name, and ensure
6555 * we can use it in the destination network namespace.
6558 if (__dev_get_by_name(net, dev->name)) {
6559 /* We get here if we can't use the current device name */
6562 if (dev_get_valid_name(net, dev, pat) < 0)
6567 * And now a mini version of register_netdevice unregister_netdevice.
6570 /* If device is running close it first. */
6573 /* And unlink it from device chain */
6575 unlist_netdevice(dev);
6579 /* Shutdown queueing discipline. */
6582 /* Notify protocols, that we are about to destroy
6583 this device. They should clean all the things.
6585 Note that dev->reg_state stays at NETREG_REGISTERED.
6586 This is wanted because this way 8021q and macvlan know
6587 the device is just moving and can keep their slaves up.
6589 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6591 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6592 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL);
6595 * Flush the unicast and multicast chains
6600 /* Send a netdev-removed uevent to the old namespace */
6601 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6603 /* Actually switch the network namespace */
6604 dev_net_set(dev, net);
6606 /* If there is an ifindex conflict assign a new one */
6607 if (__dev_get_by_index(net, dev->ifindex)) {
6608 int iflink = (dev->iflink == dev->ifindex);
6609 dev->ifindex = dev_new_index(net);
6611 dev->iflink = dev->ifindex;
6614 /* Send a netdev-add uevent to the new namespace */
6615 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6617 /* Fixup kobjects */
6618 err = device_rename(&dev->dev, dev->name);
6621 /* Add the device back in the hashes */
6622 list_netdevice(dev);
6624 /* Notify protocols, that a new device appeared. */
6625 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6628 * Prevent userspace races by waiting until the network
6629 * device is fully setup before sending notifications.
6631 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
6638 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6640 static int dev_cpu_callback(struct notifier_block *nfb,
6641 unsigned long action,
6644 struct sk_buff **list_skb;
6645 struct sk_buff *skb;
6646 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6647 struct softnet_data *sd, *oldsd;
6649 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6652 local_irq_disable();
6653 cpu = smp_processor_id();
6654 sd = &per_cpu(softnet_data, cpu);
6655 oldsd = &per_cpu(softnet_data, oldcpu);
6657 /* Find end of our completion_queue. */
6658 list_skb = &sd->completion_queue;
6660 list_skb = &(*list_skb)->next;
6661 /* Append completion queue from offline CPU. */
6662 *list_skb = oldsd->completion_queue;
6663 oldsd->completion_queue = NULL;
6665 /* Append output queue from offline CPU. */
6666 if (oldsd->output_queue) {
6667 *sd->output_queue_tailp = oldsd->output_queue;
6668 sd->output_queue_tailp = oldsd->output_queue_tailp;
6669 oldsd->output_queue = NULL;
6670 oldsd->output_queue_tailp = &oldsd->output_queue;
6672 /* Append NAPI poll list from offline CPU. */
6673 if (!list_empty(&oldsd->poll_list)) {
6674 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6675 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6678 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6681 /* Process offline CPU's input_pkt_queue */
6682 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6683 netif_rx_internal(skb);
6684 input_queue_head_incr(oldsd);
6686 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6687 netif_rx_internal(skb);
6688 input_queue_head_incr(oldsd);
6696 * netdev_increment_features - increment feature set by one
6697 * @all: current feature set
6698 * @one: new feature set
6699 * @mask: mask feature set
6701 * Computes a new feature set after adding a device with feature set
6702 * @one to the master device with current feature set @all. Will not
6703 * enable anything that is off in @mask. Returns the new feature set.
6705 netdev_features_t netdev_increment_features(netdev_features_t all,
6706 netdev_features_t one, netdev_features_t mask)
6708 if (mask & NETIF_F_GEN_CSUM)
6709 mask |= NETIF_F_ALL_CSUM;
6710 mask |= NETIF_F_VLAN_CHALLENGED;
6712 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6713 all &= one | ~NETIF_F_ALL_FOR_ALL;
6715 /* If one device supports hw checksumming, set for all. */
6716 if (all & NETIF_F_GEN_CSUM)
6717 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6721 EXPORT_SYMBOL(netdev_increment_features);
6723 static struct hlist_head * __net_init netdev_create_hash(void)
6726 struct hlist_head *hash;
6728 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6730 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6731 INIT_HLIST_HEAD(&hash[i]);
6736 /* Initialize per network namespace state */
6737 static int __net_init netdev_init(struct net *net)
6739 if (net != &init_net)
6740 INIT_LIST_HEAD(&net->dev_base_head);
6742 net->dev_name_head = netdev_create_hash();
6743 if (net->dev_name_head == NULL)
6746 net->dev_index_head = netdev_create_hash();
6747 if (net->dev_index_head == NULL)
6753 kfree(net->dev_name_head);
6759 * netdev_drivername - network driver for the device
6760 * @dev: network device
6762 * Determine network driver for device.
6764 const char *netdev_drivername(const struct net_device *dev)
6766 const struct device_driver *driver;
6767 const struct device *parent;
6768 const char *empty = "";
6770 parent = dev->dev.parent;
6774 driver = parent->driver;
6775 if (driver && driver->name)
6776 return driver->name;
6780 static int __netdev_printk(const char *level, const struct net_device *dev,
6781 struct va_format *vaf)
6785 if (dev && dev->dev.parent) {
6786 r = dev_printk_emit(level[1] - '0',
6789 dev_driver_string(dev->dev.parent),
6790 dev_name(dev->dev.parent),
6791 netdev_name(dev), vaf);
6793 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6795 r = printk("%s(NULL net_device): %pV", level, vaf);
6801 int netdev_printk(const char *level, const struct net_device *dev,
6802 const char *format, ...)
6804 struct va_format vaf;
6808 va_start(args, format);
6813 r = __netdev_printk(level, dev, &vaf);
6819 EXPORT_SYMBOL(netdev_printk);
6821 #define define_netdev_printk_level(func, level) \
6822 int func(const struct net_device *dev, const char *fmt, ...) \
6825 struct va_format vaf; \
6828 va_start(args, fmt); \
6833 r = __netdev_printk(level, dev, &vaf); \
6839 EXPORT_SYMBOL(func);
6841 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6842 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6843 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6844 define_netdev_printk_level(netdev_err, KERN_ERR);
6845 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6846 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6847 define_netdev_printk_level(netdev_info, KERN_INFO);
6849 static void __net_exit netdev_exit(struct net *net)
6851 kfree(net->dev_name_head);
6852 kfree(net->dev_index_head);
6855 static struct pernet_operations __net_initdata netdev_net_ops = {
6856 .init = netdev_init,
6857 .exit = netdev_exit,
6860 static void __net_exit default_device_exit(struct net *net)
6862 struct net_device *dev, *aux;
6864 * Push all migratable network devices back to the
6865 * initial network namespace
6868 for_each_netdev_safe(net, dev, aux) {
6870 char fb_name[IFNAMSIZ];
6872 /* Ignore unmoveable devices (i.e. loopback) */
6873 if (dev->features & NETIF_F_NETNS_LOCAL)
6876 /* Leave virtual devices for the generic cleanup */
6877 if (dev->rtnl_link_ops)
6880 /* Push remaining network devices to init_net */
6881 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6882 err = dev_change_net_namespace(dev, &init_net, fb_name);
6884 pr_emerg("%s: failed to move %s to init_net: %d\n",
6885 __func__, dev->name, err);
6892 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
6894 /* Return with the rtnl_lock held when there are no network
6895 * devices unregistering in any network namespace in net_list.
6902 prepare_to_wait(&netdev_unregistering_wq, &wait,
6903 TASK_UNINTERRUPTIBLE);
6904 unregistering = false;
6906 list_for_each_entry(net, net_list, exit_list) {
6907 if (net->dev_unreg_count > 0) {
6908 unregistering = true;
6917 finish_wait(&netdev_unregistering_wq, &wait);
6920 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6922 /* At exit all network devices most be removed from a network
6923 * namespace. Do this in the reverse order of registration.
6924 * Do this across as many network namespaces as possible to
6925 * improve batching efficiency.
6927 struct net_device *dev;
6929 LIST_HEAD(dev_kill_list);
6931 /* To prevent network device cleanup code from dereferencing
6932 * loopback devices or network devices that have been freed
6933 * wait here for all pending unregistrations to complete,
6934 * before unregistring the loopback device and allowing the
6935 * network namespace be freed.
6937 * The netdev todo list containing all network devices
6938 * unregistrations that happen in default_device_exit_batch
6939 * will run in the rtnl_unlock() at the end of
6940 * default_device_exit_batch.
6942 rtnl_lock_unregistering(net_list);
6943 list_for_each_entry(net, net_list, exit_list) {
6944 for_each_netdev_reverse(net, dev) {
6945 if (dev->rtnl_link_ops)
6946 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6948 unregister_netdevice_queue(dev, &dev_kill_list);
6951 unregister_netdevice_many(&dev_kill_list);
6952 list_del(&dev_kill_list);
6956 static struct pernet_operations __net_initdata default_device_ops = {
6957 .exit = default_device_exit,
6958 .exit_batch = default_device_exit_batch,
6962 * Initialize the DEV module. At boot time this walks the device list and
6963 * unhooks any devices that fail to initialise (normally hardware not
6964 * present) and leaves us with a valid list of present and active devices.
6969 * This is called single threaded during boot, so no need
6970 * to take the rtnl semaphore.
6972 static int __init net_dev_init(void)
6974 int i, rc = -ENOMEM;
6976 BUG_ON(!dev_boot_phase);
6978 if (dev_proc_init())
6981 if (netdev_kobject_init())
6984 INIT_LIST_HEAD(&ptype_all);
6985 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6986 INIT_LIST_HEAD(&ptype_base[i]);
6988 INIT_LIST_HEAD(&offload_base);
6990 if (register_pernet_subsys(&netdev_net_ops))
6994 * Initialise the packet receive queues.
6997 for_each_possible_cpu(i) {
6998 struct softnet_data *sd = &per_cpu(softnet_data, i);
7000 memset(sd, 0, sizeof(*sd));
7001 skb_queue_head_init(&sd->input_pkt_queue);
7002 skb_queue_head_init(&sd->process_queue);
7003 sd->completion_queue = NULL;
7004 INIT_LIST_HEAD(&sd->poll_list);
7005 sd->output_queue = NULL;
7006 sd->output_queue_tailp = &sd->output_queue;
7008 sd->csd.func = rps_trigger_softirq;
7014 sd->backlog.poll = process_backlog;
7015 sd->backlog.weight = weight_p;
7016 sd->backlog.gro_list = NULL;
7017 sd->backlog.gro_count = 0;
7019 #ifdef CONFIG_NET_FLOW_LIMIT
7020 sd->flow_limit = NULL;
7026 /* The loopback device is special if any other network devices
7027 * is present in a network namespace the loopback device must
7028 * be present. Since we now dynamically allocate and free the
7029 * loopback device ensure this invariant is maintained by
7030 * keeping the loopback device as the first device on the
7031 * list of network devices. Ensuring the loopback devices
7032 * is the first device that appears and the last network device
7035 if (register_pernet_device(&loopback_net_ops))
7038 if (register_pernet_device(&default_device_ops))
7041 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
7042 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
7044 hotcpu_notifier(dev_cpu_callback, 0);
7051 subsys_initcall(net_dev_init);