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>
134 #include "net-sysfs.h"
136 /* Instead of increasing this, you should create a hash table. */
137 #define MAX_GRO_SKBS 8
139 /* This should be increased if a protocol with a bigger head is added. */
140 #define GRO_MAX_HEAD (MAX_HEADER + 128)
142 static DEFINE_SPINLOCK(ptype_lock);
143 static DEFINE_SPINLOCK(offload_lock);
144 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
145 struct list_head ptype_all __read_mostly; /* Taps */
146 static struct list_head offload_base __read_mostly;
149 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
152 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
154 * Writers must hold the rtnl semaphore while they loop through the
155 * dev_base_head list, and hold dev_base_lock for writing when they do the
156 * actual updates. This allows pure readers to access the list even
157 * while a writer is preparing to update it.
159 * To put it another way, dev_base_lock is held for writing only to
160 * protect against pure readers; the rtnl semaphore provides the
161 * protection against other writers.
163 * See, for example usages, register_netdevice() and
164 * unregister_netdevice(), which must be called with the rtnl
167 DEFINE_RWLOCK(dev_base_lock);
168 EXPORT_SYMBOL(dev_base_lock);
170 /* protects napi_hash addition/deletion and napi_gen_id */
171 static DEFINE_SPINLOCK(napi_hash_lock);
173 static unsigned int napi_gen_id;
174 static DEFINE_HASHTABLE(napi_hash, 8);
176 seqcount_t devnet_rename_seq;
178 static inline void dev_base_seq_inc(struct net *net)
180 while (++net->dev_base_seq == 0);
183 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
185 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
187 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
190 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
192 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
195 static inline void rps_lock(struct softnet_data *sd)
198 spin_lock(&sd->input_pkt_queue.lock);
202 static inline void rps_unlock(struct softnet_data *sd)
205 spin_unlock(&sd->input_pkt_queue.lock);
209 /* Device list insertion */
210 static void list_netdevice(struct net_device *dev)
212 struct net *net = dev_net(dev);
216 write_lock_bh(&dev_base_lock);
217 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
218 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
219 hlist_add_head_rcu(&dev->index_hlist,
220 dev_index_hash(net, dev->ifindex));
221 write_unlock_bh(&dev_base_lock);
223 dev_base_seq_inc(net);
226 /* Device list removal
227 * caller must respect a RCU grace period before freeing/reusing dev
229 static void unlist_netdevice(struct net_device *dev)
233 /* Unlink dev from the device chain */
234 write_lock_bh(&dev_base_lock);
235 list_del_rcu(&dev->dev_list);
236 hlist_del_rcu(&dev->name_hlist);
237 hlist_del_rcu(&dev->index_hlist);
238 write_unlock_bh(&dev_base_lock);
240 dev_base_seq_inc(dev_net(dev));
247 static RAW_NOTIFIER_HEAD(netdev_chain);
250 * Device drivers call our routines to queue packets here. We empty the
251 * queue in the local softnet handler.
254 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
255 EXPORT_PER_CPU_SYMBOL(softnet_data);
257 #ifdef CONFIG_LOCKDEP
259 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
260 * according to dev->type
262 static const unsigned short netdev_lock_type[] =
263 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
264 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
265 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
266 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
267 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
268 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
269 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
270 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
271 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
272 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
273 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
274 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
275 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
276 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
277 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
279 static const char *const netdev_lock_name[] =
280 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
281 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
282 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
283 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
284 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
285 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
286 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
287 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
288 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
289 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
290 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
291 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
292 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
293 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
294 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
296 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
297 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
299 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
303 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
304 if (netdev_lock_type[i] == dev_type)
306 /* the last key is used by default */
307 return ARRAY_SIZE(netdev_lock_type) - 1;
310 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
311 unsigned short dev_type)
315 i = netdev_lock_pos(dev_type);
316 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
317 netdev_lock_name[i]);
320 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
324 i = netdev_lock_pos(dev->type);
325 lockdep_set_class_and_name(&dev->addr_list_lock,
326 &netdev_addr_lock_key[i],
327 netdev_lock_name[i]);
330 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
331 unsigned short dev_type)
334 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
339 /*******************************************************************************
341 Protocol management and registration routines
343 *******************************************************************************/
346 * Add a protocol ID to the list. Now that the input handler is
347 * smarter we can dispense with all the messy stuff that used to be
350 * BEWARE!!! Protocol handlers, mangling input packets,
351 * MUST BE last in hash buckets and checking protocol handlers
352 * MUST start from promiscuous ptype_all chain in net_bh.
353 * It is true now, do not change it.
354 * Explanation follows: if protocol handler, mangling packet, will
355 * be the first on list, it is not able to sense, that packet
356 * is cloned and should be copied-on-write, so that it will
357 * change it and subsequent readers will get broken packet.
361 static inline struct list_head *ptype_head(const struct packet_type *pt)
363 if (pt->type == htons(ETH_P_ALL))
366 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
370 * dev_add_pack - add packet handler
371 * @pt: packet type declaration
373 * Add a protocol handler to the networking stack. The passed &packet_type
374 * is linked into kernel lists and may not be freed until it has been
375 * removed from the kernel lists.
377 * This call does not sleep therefore it can not
378 * guarantee all CPU's that are in middle of receiving packets
379 * will see the new packet type (until the next received packet).
382 void dev_add_pack(struct packet_type *pt)
384 struct list_head *head = ptype_head(pt);
386 spin_lock(&ptype_lock);
387 list_add_rcu(&pt->list, head);
388 spin_unlock(&ptype_lock);
390 EXPORT_SYMBOL(dev_add_pack);
393 * __dev_remove_pack - remove packet handler
394 * @pt: packet type declaration
396 * Remove a protocol handler that was previously added to the kernel
397 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
398 * from the kernel lists and can be freed or reused once this function
401 * The packet type might still be in use by receivers
402 * and must not be freed until after all the CPU's have gone
403 * through a quiescent state.
405 void __dev_remove_pack(struct packet_type *pt)
407 struct list_head *head = ptype_head(pt);
408 struct packet_type *pt1;
410 spin_lock(&ptype_lock);
412 list_for_each_entry(pt1, head, list) {
414 list_del_rcu(&pt->list);
419 pr_warn("dev_remove_pack: %p not found\n", pt);
421 spin_unlock(&ptype_lock);
423 EXPORT_SYMBOL(__dev_remove_pack);
426 * dev_remove_pack - remove packet handler
427 * @pt: packet type declaration
429 * Remove a protocol handler that was previously added to the kernel
430 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
431 * from the kernel lists and can be freed or reused once this function
434 * This call sleeps to guarantee that no CPU is looking at the packet
437 void dev_remove_pack(struct packet_type *pt)
439 __dev_remove_pack(pt);
443 EXPORT_SYMBOL(dev_remove_pack);
447 * dev_add_offload - register offload handlers
448 * @po: protocol offload declaration
450 * Add protocol offload handlers to the networking stack. The passed
451 * &proto_offload is linked into kernel lists and may not be freed until
452 * it has been removed from the kernel lists.
454 * This call does not sleep therefore it can not
455 * guarantee all CPU's that are in middle of receiving packets
456 * will see the new offload handlers (until the next received packet).
458 void dev_add_offload(struct packet_offload *po)
460 struct list_head *head = &offload_base;
462 spin_lock(&offload_lock);
463 list_add_rcu(&po->list, head);
464 spin_unlock(&offload_lock);
466 EXPORT_SYMBOL(dev_add_offload);
469 * __dev_remove_offload - remove offload handler
470 * @po: packet offload declaration
472 * Remove a protocol offload handler that was previously added to the
473 * kernel offload handlers by dev_add_offload(). The passed &offload_type
474 * is removed from the kernel lists and can be freed or reused once this
477 * The packet type might still be in use by receivers
478 * and must not be freed until after all the CPU's have gone
479 * through a quiescent state.
481 void __dev_remove_offload(struct packet_offload *po)
483 struct list_head *head = &offload_base;
484 struct packet_offload *po1;
486 spin_lock(&offload_lock);
488 list_for_each_entry(po1, head, list) {
490 list_del_rcu(&po->list);
495 pr_warn("dev_remove_offload: %p not found\n", po);
497 spin_unlock(&offload_lock);
499 EXPORT_SYMBOL(__dev_remove_offload);
502 * dev_remove_offload - remove packet offload handler
503 * @po: packet offload declaration
505 * Remove a packet offload handler that was previously added to the kernel
506 * offload handlers by dev_add_offload(). The passed &offload_type is
507 * removed from the kernel lists and can be freed or reused once this
510 * This call sleeps to guarantee that no CPU is looking at the packet
513 void dev_remove_offload(struct packet_offload *po)
515 __dev_remove_offload(po);
519 EXPORT_SYMBOL(dev_remove_offload);
521 /******************************************************************************
523 Device Boot-time Settings Routines
525 *******************************************************************************/
527 /* Boot time configuration table */
528 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
531 * netdev_boot_setup_add - add new setup entry
532 * @name: name of the device
533 * @map: configured settings for the device
535 * Adds new setup entry to the dev_boot_setup list. The function
536 * returns 0 on error and 1 on success. This is a generic routine to
539 static int netdev_boot_setup_add(char *name, struct ifmap *map)
541 struct netdev_boot_setup *s;
545 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
546 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
547 memset(s[i].name, 0, sizeof(s[i].name));
548 strlcpy(s[i].name, name, IFNAMSIZ);
549 memcpy(&s[i].map, map, sizeof(s[i].map));
554 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
558 * netdev_boot_setup_check - check boot time settings
559 * @dev: the netdevice
561 * Check boot time settings for the device.
562 * The found settings are set for the device to be used
563 * later in the device probing.
564 * Returns 0 if no settings found, 1 if they are.
566 int netdev_boot_setup_check(struct net_device *dev)
568 struct netdev_boot_setup *s = dev_boot_setup;
571 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
572 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
573 !strcmp(dev->name, s[i].name)) {
574 dev->irq = s[i].map.irq;
575 dev->base_addr = s[i].map.base_addr;
576 dev->mem_start = s[i].map.mem_start;
577 dev->mem_end = s[i].map.mem_end;
583 EXPORT_SYMBOL(netdev_boot_setup_check);
587 * netdev_boot_base - get address from boot time settings
588 * @prefix: prefix for network device
589 * @unit: id for network device
591 * Check boot time settings for the base address of device.
592 * The found settings are set for the device to be used
593 * later in the device probing.
594 * Returns 0 if no settings found.
596 unsigned long netdev_boot_base(const char *prefix, int unit)
598 const struct netdev_boot_setup *s = dev_boot_setup;
602 sprintf(name, "%s%d", prefix, unit);
605 * If device already registered then return base of 1
606 * to indicate not to probe for this interface
608 if (__dev_get_by_name(&init_net, name))
611 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
612 if (!strcmp(name, s[i].name))
613 return s[i].map.base_addr;
618 * Saves at boot time configured settings for any netdevice.
620 int __init netdev_boot_setup(char *str)
625 str = get_options(str, ARRAY_SIZE(ints), ints);
630 memset(&map, 0, sizeof(map));
634 map.base_addr = ints[2];
636 map.mem_start = ints[3];
638 map.mem_end = ints[4];
640 /* Add new entry to the list */
641 return netdev_boot_setup_add(str, &map);
644 __setup("netdev=", netdev_boot_setup);
646 /*******************************************************************************
648 Device Interface Subroutines
650 *******************************************************************************/
653 * __dev_get_by_name - find a device by its name
654 * @net: the applicable net namespace
655 * @name: name to find
657 * Find an interface by name. Must be called under RTNL semaphore
658 * or @dev_base_lock. If the name is found a pointer to the device
659 * is returned. If the name is not found then %NULL is returned. The
660 * reference counters are not incremented so the caller must be
661 * careful with locks.
664 struct net_device *__dev_get_by_name(struct net *net, const char *name)
666 struct net_device *dev;
667 struct hlist_head *head = dev_name_hash(net, name);
669 hlist_for_each_entry(dev, head, name_hlist)
670 if (!strncmp(dev->name, name, IFNAMSIZ))
675 EXPORT_SYMBOL(__dev_get_by_name);
678 * dev_get_by_name_rcu - find a device by its name
679 * @net: the applicable net namespace
680 * @name: name to find
682 * Find an interface by name.
683 * If the name is found a pointer to the device is returned.
684 * If the name is not found then %NULL is returned.
685 * The reference counters are not incremented so the caller must be
686 * careful with locks. The caller must hold RCU lock.
689 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
691 struct net_device *dev;
692 struct hlist_head *head = dev_name_hash(net, name);
694 hlist_for_each_entry_rcu(dev, head, name_hlist)
695 if (!strncmp(dev->name, name, IFNAMSIZ))
700 EXPORT_SYMBOL(dev_get_by_name_rcu);
703 * dev_get_by_name - find a device by its name
704 * @net: the applicable net namespace
705 * @name: name to find
707 * Find an interface by name. This can be called from any
708 * context and does its own locking. The returned handle has
709 * the usage count incremented and the caller must use dev_put() to
710 * release it when it is no longer needed. %NULL is returned if no
711 * matching device is found.
714 struct net_device *dev_get_by_name(struct net *net, const char *name)
716 struct net_device *dev;
719 dev = dev_get_by_name_rcu(net, name);
725 EXPORT_SYMBOL(dev_get_by_name);
728 * __dev_get_by_index - find a device by its ifindex
729 * @net: the applicable net namespace
730 * @ifindex: index of device
732 * Search for an interface by index. Returns %NULL if the device
733 * is not found or a pointer to the device. The device has not
734 * had its reference counter increased so the caller must be careful
735 * about locking. The caller must hold either the RTNL semaphore
739 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
741 struct net_device *dev;
742 struct hlist_head *head = dev_index_hash(net, ifindex);
744 hlist_for_each_entry(dev, head, index_hlist)
745 if (dev->ifindex == ifindex)
750 EXPORT_SYMBOL(__dev_get_by_index);
753 * dev_get_by_index_rcu - find a device by its ifindex
754 * @net: the applicable net namespace
755 * @ifindex: index of device
757 * Search for an interface by index. Returns %NULL if the device
758 * is not found or a pointer to the device. The device has not
759 * had its reference counter increased so the caller must be careful
760 * about locking. The caller must hold RCU lock.
763 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
765 struct net_device *dev;
766 struct hlist_head *head = dev_index_hash(net, ifindex);
768 hlist_for_each_entry_rcu(dev, head, index_hlist)
769 if (dev->ifindex == ifindex)
774 EXPORT_SYMBOL(dev_get_by_index_rcu);
778 * dev_get_by_index - find a device by its ifindex
779 * @net: the applicable net namespace
780 * @ifindex: index of device
782 * Search for an interface by index. Returns NULL if the device
783 * is not found or a pointer to the device. The device returned has
784 * had a reference added and the pointer is safe until the user calls
785 * dev_put to indicate they have finished with it.
788 struct net_device *dev_get_by_index(struct net *net, int ifindex)
790 struct net_device *dev;
793 dev = dev_get_by_index_rcu(net, ifindex);
799 EXPORT_SYMBOL(dev_get_by_index);
802 * dev_getbyhwaddr_rcu - find a device by its hardware address
803 * @net: the applicable net namespace
804 * @type: media type of device
805 * @ha: hardware address
807 * Search for an interface by MAC address. Returns NULL if the device
808 * is not found or a pointer to the device.
809 * The caller must hold RCU or RTNL.
810 * The returned device has not had its ref count increased
811 * and the caller must therefore be careful about locking
815 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
818 struct net_device *dev;
820 for_each_netdev_rcu(net, dev)
821 if (dev->type == type &&
822 !memcmp(dev->dev_addr, ha, dev->addr_len))
827 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
829 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
831 struct net_device *dev;
834 for_each_netdev(net, dev)
835 if (dev->type == type)
840 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
842 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
844 struct net_device *dev, *ret = NULL;
847 for_each_netdev_rcu(net, dev)
848 if (dev->type == type) {
856 EXPORT_SYMBOL(dev_getfirstbyhwtype);
859 * dev_get_by_flags_rcu - find any device with given flags
860 * @net: the applicable net namespace
861 * @if_flags: IFF_* values
862 * @mask: bitmask of bits in if_flags to check
864 * Search for any interface with the given flags. Returns NULL if a device
865 * is not found or a pointer to the device. Must be called inside
866 * rcu_read_lock(), and result refcount is unchanged.
869 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
872 struct net_device *dev, *ret;
875 for_each_netdev_rcu(net, dev) {
876 if (((dev->flags ^ if_flags) & mask) == 0) {
883 EXPORT_SYMBOL(dev_get_by_flags_rcu);
886 * dev_valid_name - check if name is okay for network device
889 * Network device names need to be valid file names to
890 * to allow sysfs to work. We also disallow any kind of
893 bool dev_valid_name(const char *name)
897 if (strlen(name) >= IFNAMSIZ)
899 if (!strcmp(name, ".") || !strcmp(name, ".."))
903 if (*name == '/' || isspace(*name))
909 EXPORT_SYMBOL(dev_valid_name);
912 * __dev_alloc_name - allocate a name for a device
913 * @net: network namespace to allocate the device name in
914 * @name: name format string
915 * @buf: scratch buffer and result name string
917 * Passed a format string - eg "lt%d" it will try and find a suitable
918 * id. It scans list of devices to build up a free map, then chooses
919 * the first empty slot. The caller must hold the dev_base or rtnl lock
920 * while allocating the name and adding the device in order to avoid
922 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
923 * Returns the number of the unit assigned or a negative errno code.
926 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
930 const int max_netdevices = 8*PAGE_SIZE;
931 unsigned long *inuse;
932 struct net_device *d;
934 p = strnchr(name, IFNAMSIZ-1, '%');
937 * Verify the string as this thing may have come from
938 * the user. There must be either one "%d" and no other "%"
941 if (p[1] != 'd' || strchr(p + 2, '%'))
944 /* Use one page as a bit array of possible slots */
945 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
949 for_each_netdev(net, d) {
950 if (!sscanf(d->name, name, &i))
952 if (i < 0 || i >= max_netdevices)
955 /* avoid cases where sscanf is not exact inverse of printf */
956 snprintf(buf, IFNAMSIZ, name, i);
957 if (!strncmp(buf, d->name, IFNAMSIZ))
961 i = find_first_zero_bit(inuse, max_netdevices);
962 free_page((unsigned long) inuse);
966 snprintf(buf, IFNAMSIZ, name, i);
967 if (!__dev_get_by_name(net, buf))
970 /* It is possible to run out of possible slots
971 * when the name is long and there isn't enough space left
972 * for the digits, or if all bits are used.
978 * dev_alloc_name - allocate a name for a device
980 * @name: name format string
982 * Passed a format string - eg "lt%d" it will try and find a suitable
983 * id. It scans list of devices to build up a free map, then chooses
984 * the first empty slot. The caller must hold the dev_base or rtnl lock
985 * while allocating the name and adding the device in order to avoid
987 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
988 * Returns the number of the unit assigned or a negative errno code.
991 int dev_alloc_name(struct net_device *dev, const char *name)
997 BUG_ON(!dev_net(dev));
999 ret = __dev_alloc_name(net, name, buf);
1001 strlcpy(dev->name, buf, IFNAMSIZ);
1004 EXPORT_SYMBOL(dev_alloc_name);
1006 static int dev_alloc_name_ns(struct net *net,
1007 struct net_device *dev,
1013 ret = __dev_alloc_name(net, name, buf);
1015 strlcpy(dev->name, buf, IFNAMSIZ);
1019 static int dev_get_valid_name(struct net *net,
1020 struct net_device *dev,
1025 if (!dev_valid_name(name))
1028 if (strchr(name, '%'))
1029 return dev_alloc_name_ns(net, dev, name);
1030 else if (__dev_get_by_name(net, name))
1032 else if (dev->name != name)
1033 strlcpy(dev->name, name, IFNAMSIZ);
1039 * dev_change_name - change name of a device
1041 * @newname: name (or format string) must be at least IFNAMSIZ
1043 * Change name of a device, can pass format strings "eth%d".
1046 int dev_change_name(struct net_device *dev, const char *newname)
1048 char oldname[IFNAMSIZ];
1054 BUG_ON(!dev_net(dev));
1057 if (dev->flags & IFF_UP)
1060 write_seqcount_begin(&devnet_rename_seq);
1062 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1063 write_seqcount_end(&devnet_rename_seq);
1067 memcpy(oldname, dev->name, IFNAMSIZ);
1069 err = dev_get_valid_name(net, dev, newname);
1071 write_seqcount_end(&devnet_rename_seq);
1076 ret = device_rename(&dev->dev, dev->name);
1078 memcpy(dev->name, oldname, IFNAMSIZ);
1079 write_seqcount_end(&devnet_rename_seq);
1083 write_seqcount_end(&devnet_rename_seq);
1085 write_lock_bh(&dev_base_lock);
1086 hlist_del_rcu(&dev->name_hlist);
1087 write_unlock_bh(&dev_base_lock);
1091 write_lock_bh(&dev_base_lock);
1092 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1093 write_unlock_bh(&dev_base_lock);
1095 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1096 ret = notifier_to_errno(ret);
1099 /* err >= 0 after dev_alloc_name() or stores the first errno */
1102 write_seqcount_begin(&devnet_rename_seq);
1103 memcpy(dev->name, oldname, IFNAMSIZ);
1106 pr_err("%s: name change rollback failed: %d\n",
1115 * dev_set_alias - change ifalias of a device
1117 * @alias: name up to IFALIASZ
1118 * @len: limit of bytes to copy from info
1120 * Set ifalias for a device,
1122 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1128 if (len >= IFALIASZ)
1132 kfree(dev->ifalias);
1133 dev->ifalias = NULL;
1137 new_ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1140 dev->ifalias = new_ifalias;
1142 strlcpy(dev->ifalias, alias, len+1);
1148 * netdev_features_change - device changes features
1149 * @dev: device to cause notification
1151 * Called to indicate a device has changed features.
1153 void netdev_features_change(struct net_device *dev)
1155 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1157 EXPORT_SYMBOL(netdev_features_change);
1160 * netdev_state_change - device changes state
1161 * @dev: device to cause notification
1163 * Called to indicate a device has changed state. This function calls
1164 * the notifier chains for netdev_chain and sends a NEWLINK message
1165 * to the routing socket.
1167 void netdev_state_change(struct net_device *dev)
1169 if (dev->flags & IFF_UP) {
1170 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1171 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1174 EXPORT_SYMBOL(netdev_state_change);
1177 * netdev_notify_peers - notify network peers about existence of @dev
1178 * @dev: network device
1180 * Generate traffic such that interested network peers are aware of
1181 * @dev, such as by generating a gratuitous ARP. This may be used when
1182 * a device wants to inform the rest of the network about some sort of
1183 * reconfiguration such as a failover event or virtual machine
1186 void netdev_notify_peers(struct net_device *dev)
1189 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1192 EXPORT_SYMBOL(netdev_notify_peers);
1194 static int __dev_open(struct net_device *dev)
1196 const struct net_device_ops *ops = dev->netdev_ops;
1201 if (!netif_device_present(dev))
1204 /* Block netpoll from trying to do any rx path servicing.
1205 * If we don't do this there is a chance ndo_poll_controller
1206 * or ndo_poll may be running while we open the device
1208 netpoll_rx_disable(dev);
1210 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1211 ret = notifier_to_errno(ret);
1215 set_bit(__LINK_STATE_START, &dev->state);
1217 if (ops->ndo_validate_addr)
1218 ret = ops->ndo_validate_addr(dev);
1220 if (!ret && ops->ndo_open)
1221 ret = ops->ndo_open(dev);
1223 netpoll_rx_enable(dev);
1226 clear_bit(__LINK_STATE_START, &dev->state);
1228 dev->flags |= IFF_UP;
1229 net_dmaengine_get();
1230 dev_set_rx_mode(dev);
1232 add_device_randomness(dev->dev_addr, dev->addr_len);
1239 * dev_open - prepare an interface for use.
1240 * @dev: device to open
1242 * Takes a device from down to up state. The device's private open
1243 * function is invoked and then the multicast lists are loaded. Finally
1244 * the device is moved into the up state and a %NETDEV_UP message is
1245 * sent to the netdev notifier chain.
1247 * Calling this function on an active interface is a nop. On a failure
1248 * a negative errno code is returned.
1250 int dev_open(struct net_device *dev)
1254 if (dev->flags & IFF_UP)
1257 ret = __dev_open(dev);
1261 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1262 call_netdevice_notifiers(NETDEV_UP, dev);
1266 EXPORT_SYMBOL(dev_open);
1268 static int __dev_close_many(struct list_head *head)
1270 struct net_device *dev;
1275 list_for_each_entry(dev, head, unreg_list) {
1276 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1278 clear_bit(__LINK_STATE_START, &dev->state);
1280 /* Synchronize to scheduled poll. We cannot touch poll list, it
1281 * can be even on different cpu. So just clear netif_running().
1283 * dev->stop() will invoke napi_disable() on all of it's
1284 * napi_struct instances on this device.
1286 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1289 dev_deactivate_many(head);
1291 list_for_each_entry(dev, head, unreg_list) {
1292 const struct net_device_ops *ops = dev->netdev_ops;
1295 * Call the device specific close. This cannot fail.
1296 * Only if device is UP
1298 * We allow it to be called even after a DETACH hot-plug
1304 dev->flags &= ~IFF_UP;
1305 net_dmaengine_put();
1311 static int __dev_close(struct net_device *dev)
1316 /* Temporarily disable netpoll until the interface is down */
1317 netpoll_rx_disable(dev);
1319 list_add(&dev->unreg_list, &single);
1320 retval = __dev_close_many(&single);
1323 netpoll_rx_enable(dev);
1327 static int dev_close_many(struct list_head *head)
1329 struct net_device *dev, *tmp;
1330 LIST_HEAD(tmp_list);
1332 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1333 if (!(dev->flags & IFF_UP))
1334 list_move(&dev->unreg_list, &tmp_list);
1336 __dev_close_many(head);
1338 list_for_each_entry(dev, head, unreg_list) {
1339 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1340 call_netdevice_notifiers(NETDEV_DOWN, dev);
1343 /* rollback_registered_many needs the complete original list */
1344 list_splice(&tmp_list, head);
1349 * dev_close - shutdown an interface.
1350 * @dev: device to shutdown
1352 * This function moves an active device into down state. A
1353 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1354 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1357 int dev_close(struct net_device *dev)
1359 if (dev->flags & IFF_UP) {
1362 /* Block netpoll rx while the interface is going down */
1363 netpoll_rx_disable(dev);
1365 list_add(&dev->unreg_list, &single);
1366 dev_close_many(&single);
1369 netpoll_rx_enable(dev);
1373 EXPORT_SYMBOL(dev_close);
1377 * dev_disable_lro - disable Large Receive Offload on a device
1380 * Disable Large Receive Offload (LRO) on a net device. Must be
1381 * called under RTNL. This is needed if received packets may be
1382 * forwarded to another interface.
1384 void dev_disable_lro(struct net_device *dev)
1387 * If we're trying to disable lro on a vlan device
1388 * use the underlying physical device instead
1390 if (is_vlan_dev(dev))
1391 dev = vlan_dev_real_dev(dev);
1393 dev->wanted_features &= ~NETIF_F_LRO;
1394 netdev_update_features(dev);
1396 if (unlikely(dev->features & NETIF_F_LRO))
1397 netdev_WARN(dev, "failed to disable LRO!\n");
1399 EXPORT_SYMBOL(dev_disable_lro);
1401 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1402 struct net_device *dev)
1404 struct netdev_notifier_info info;
1406 netdev_notifier_info_init(&info, dev);
1407 return nb->notifier_call(nb, val, &info);
1410 static int dev_boot_phase = 1;
1413 * register_netdevice_notifier - register a network notifier block
1416 * Register a notifier to be called when network device events occur.
1417 * The notifier passed is linked into the kernel structures and must
1418 * not be reused until it has been unregistered. A negative errno code
1419 * is returned on a failure.
1421 * When registered all registration and up events are replayed
1422 * to the new notifier to allow device to have a race free
1423 * view of the network device list.
1426 int register_netdevice_notifier(struct notifier_block *nb)
1428 struct net_device *dev;
1429 struct net_device *last;
1434 err = raw_notifier_chain_register(&netdev_chain, nb);
1440 for_each_netdev(net, dev) {
1441 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1442 err = notifier_to_errno(err);
1446 if (!(dev->flags & IFF_UP))
1449 call_netdevice_notifier(nb, NETDEV_UP, dev);
1460 for_each_netdev(net, dev) {
1464 if (dev->flags & IFF_UP) {
1465 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1467 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1469 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1474 raw_notifier_chain_unregister(&netdev_chain, nb);
1477 EXPORT_SYMBOL(register_netdevice_notifier);
1480 * unregister_netdevice_notifier - unregister a network notifier block
1483 * Unregister a notifier previously registered by
1484 * register_netdevice_notifier(). The notifier is unlinked into the
1485 * kernel structures and may then be reused. A negative errno code
1486 * is returned on a failure.
1488 * After unregistering unregister and down device events are synthesized
1489 * for all devices on the device list to the removed notifier to remove
1490 * the need for special case cleanup code.
1493 int unregister_netdevice_notifier(struct notifier_block *nb)
1495 struct net_device *dev;
1500 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1505 for_each_netdev(net, dev) {
1506 if (dev->flags & IFF_UP) {
1507 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1509 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1511 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1518 EXPORT_SYMBOL(unregister_netdevice_notifier);
1521 * call_netdevice_notifiers_info - call all network notifier blocks
1522 * @val: value passed unmodified to notifier function
1523 * @dev: net_device pointer passed unmodified to notifier function
1524 * @info: notifier information data
1526 * Call all network notifier blocks. Parameters and return value
1527 * are as for raw_notifier_call_chain().
1530 int call_netdevice_notifiers_info(unsigned long val, struct net_device *dev,
1531 struct netdev_notifier_info *info)
1534 netdev_notifier_info_init(info, dev);
1535 return raw_notifier_call_chain(&netdev_chain, val, info);
1537 EXPORT_SYMBOL(call_netdevice_notifiers_info);
1540 * call_netdevice_notifiers - call all network notifier blocks
1541 * @val: value passed unmodified to notifier function
1542 * @dev: net_device pointer passed unmodified to notifier function
1544 * Call all network notifier blocks. Parameters and return value
1545 * are as for raw_notifier_call_chain().
1548 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1550 struct netdev_notifier_info info;
1552 return call_netdevice_notifiers_info(val, dev, &info);
1554 EXPORT_SYMBOL(call_netdevice_notifiers);
1556 static struct static_key netstamp_needed __read_mostly;
1557 #ifdef HAVE_JUMP_LABEL
1558 /* We are not allowed to call static_key_slow_dec() from irq context
1559 * If net_disable_timestamp() is called from irq context, defer the
1560 * static_key_slow_dec() calls.
1562 static atomic_t netstamp_needed_deferred;
1565 void net_enable_timestamp(void)
1567 #ifdef HAVE_JUMP_LABEL
1568 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1572 static_key_slow_dec(&netstamp_needed);
1576 static_key_slow_inc(&netstamp_needed);
1578 EXPORT_SYMBOL(net_enable_timestamp);
1580 void net_disable_timestamp(void)
1582 #ifdef HAVE_JUMP_LABEL
1583 if (in_interrupt()) {
1584 atomic_inc(&netstamp_needed_deferred);
1588 static_key_slow_dec(&netstamp_needed);
1590 EXPORT_SYMBOL(net_disable_timestamp);
1592 static inline void net_timestamp_set(struct sk_buff *skb)
1594 skb->tstamp.tv64 = 0;
1595 if (static_key_false(&netstamp_needed))
1596 __net_timestamp(skb);
1599 #define net_timestamp_check(COND, SKB) \
1600 if (static_key_false(&netstamp_needed)) { \
1601 if ((COND) && !(SKB)->tstamp.tv64) \
1602 __net_timestamp(SKB); \
1605 static inline bool is_skb_forwardable(struct net_device *dev,
1606 struct sk_buff *skb)
1610 if (!(dev->flags & IFF_UP))
1613 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1614 if (skb->len <= len)
1617 /* if TSO is enabled, we don't care about the length as the packet
1618 * could be forwarded without being segmented before
1620 if (skb_is_gso(skb))
1627 * dev_forward_skb - loopback an skb to another netif
1629 * @dev: destination network device
1630 * @skb: buffer to forward
1633 * NET_RX_SUCCESS (no congestion)
1634 * NET_RX_DROP (packet was dropped, but freed)
1636 * dev_forward_skb can be used for injecting an skb from the
1637 * start_xmit function of one device into the receive queue
1638 * of another device.
1640 * The receiving device may be in another namespace, so
1641 * we have to clear all information in the skb that could
1642 * impact namespace isolation.
1644 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1646 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1647 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1648 atomic_long_inc(&dev->rx_dropped);
1656 if (unlikely(!is_skb_forwardable(dev, skb))) {
1657 atomic_long_inc(&dev->rx_dropped);
1663 skb->tstamp.tv64 = 0;
1664 skb->pkt_type = PACKET_HOST;
1665 skb->protocol = eth_type_trans(skb, dev);
1669 nf_reset_trace(skb);
1670 return netif_rx(skb);
1672 EXPORT_SYMBOL_GPL(dev_forward_skb);
1674 static inline int deliver_skb(struct sk_buff *skb,
1675 struct packet_type *pt_prev,
1676 struct net_device *orig_dev)
1678 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
1680 atomic_inc(&skb->users);
1681 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1684 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1686 if (!ptype->af_packet_priv || !skb->sk)
1689 if (ptype->id_match)
1690 return ptype->id_match(ptype, skb->sk);
1691 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1698 * Support routine. Sends outgoing frames to any network
1699 * taps currently in use.
1702 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1704 struct packet_type *ptype;
1705 struct sk_buff *skb2 = NULL;
1706 struct packet_type *pt_prev = NULL;
1709 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1710 /* Never send packets back to the socket
1711 * they originated from - MvS (miquels@drinkel.ow.org)
1713 if ((ptype->dev == dev || !ptype->dev) &&
1714 (!skb_loop_sk(ptype, skb))) {
1716 deliver_skb(skb2, pt_prev, skb->dev);
1721 skb2 = skb_clone(skb, GFP_ATOMIC);
1725 net_timestamp_set(skb2);
1727 /* skb->nh should be correctly
1728 set by sender, so that the second statement is
1729 just protection against buggy protocols.
1731 skb_reset_mac_header(skb2);
1733 if (skb_network_header(skb2) < skb2->data ||
1734 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
1735 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1736 ntohs(skb2->protocol),
1738 skb_reset_network_header(skb2);
1741 skb2->transport_header = skb2->network_header;
1742 skb2->pkt_type = PACKET_OUTGOING;
1747 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1752 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1753 * @dev: Network device
1754 * @txq: number of queues available
1756 * If real_num_tx_queues is changed the tc mappings may no longer be
1757 * valid. To resolve this verify the tc mapping remains valid and if
1758 * not NULL the mapping. With no priorities mapping to this
1759 * offset/count pair it will no longer be used. In the worst case TC0
1760 * is invalid nothing can be done so disable priority mappings. If is
1761 * expected that drivers will fix this mapping if they can before
1762 * calling netif_set_real_num_tx_queues.
1764 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1767 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1769 /* If TC0 is invalidated disable TC mapping */
1770 if (tc->offset + tc->count > txq) {
1771 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1776 /* Invalidated prio to tc mappings set to TC0 */
1777 for (i = 1; i < TC_BITMASK + 1; i++) {
1778 int q = netdev_get_prio_tc_map(dev, i);
1780 tc = &dev->tc_to_txq[q];
1781 if (tc->offset + tc->count > txq) {
1782 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1784 netdev_set_prio_tc_map(dev, i, 0);
1790 static DEFINE_MUTEX(xps_map_mutex);
1791 #define xmap_dereference(P) \
1792 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1794 static struct xps_map *remove_xps_queue(struct xps_dev_maps *dev_maps,
1797 struct xps_map *map = NULL;
1801 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1803 for (pos = 0; map && pos < map->len; pos++) {
1804 if (map->queues[pos] == index) {
1806 map->queues[pos] = map->queues[--map->len];
1808 RCU_INIT_POINTER(dev_maps->cpu_map[cpu], NULL);
1809 kfree_rcu(map, rcu);
1819 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
1821 struct xps_dev_maps *dev_maps;
1823 bool active = false;
1825 mutex_lock(&xps_map_mutex);
1826 dev_maps = xmap_dereference(dev->xps_maps);
1831 for_each_possible_cpu(cpu) {
1832 for (i = index; i < dev->num_tx_queues; i++) {
1833 if (!remove_xps_queue(dev_maps, cpu, i))
1836 if (i == dev->num_tx_queues)
1841 RCU_INIT_POINTER(dev->xps_maps, NULL);
1842 kfree_rcu(dev_maps, rcu);
1845 for (i = index; i < dev->num_tx_queues; i++)
1846 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
1850 mutex_unlock(&xps_map_mutex);
1853 static struct xps_map *expand_xps_map(struct xps_map *map,
1856 struct xps_map *new_map;
1857 int alloc_len = XPS_MIN_MAP_ALLOC;
1860 for (pos = 0; map && pos < map->len; pos++) {
1861 if (map->queues[pos] != index)
1866 /* Need to add queue to this CPU's existing map */
1868 if (pos < map->alloc_len)
1871 alloc_len = map->alloc_len * 2;
1874 /* Need to allocate new map to store queue on this CPU's map */
1875 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
1880 for (i = 0; i < pos; i++)
1881 new_map->queues[i] = map->queues[i];
1882 new_map->alloc_len = alloc_len;
1888 int netif_set_xps_queue(struct net_device *dev, struct cpumask *mask, u16 index)
1890 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
1891 struct xps_map *map, *new_map;
1892 int maps_sz = max_t(unsigned int, XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES);
1893 int cpu, numa_node_id = -2;
1894 bool active = false;
1896 mutex_lock(&xps_map_mutex);
1898 dev_maps = xmap_dereference(dev->xps_maps);
1900 /* allocate memory for queue storage */
1901 for_each_online_cpu(cpu) {
1902 if (!cpumask_test_cpu(cpu, mask))
1906 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
1907 if (!new_dev_maps) {
1908 mutex_unlock(&xps_map_mutex);
1912 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
1915 map = expand_xps_map(map, cpu, index);
1919 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1923 goto out_no_new_maps;
1925 for_each_possible_cpu(cpu) {
1926 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
1927 /* add queue to CPU maps */
1930 map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1931 while ((pos < map->len) && (map->queues[pos] != index))
1934 if (pos == map->len)
1935 map->queues[map->len++] = index;
1937 if (numa_node_id == -2)
1938 numa_node_id = cpu_to_node(cpu);
1939 else if (numa_node_id != cpu_to_node(cpu))
1942 } else if (dev_maps) {
1943 /* fill in the new device map from the old device map */
1944 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1945 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], map);
1950 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1952 /* Cleanup old maps */
1954 for_each_possible_cpu(cpu) {
1955 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1956 map = xmap_dereference(dev_maps->cpu_map[cpu]);
1957 if (map && map != new_map)
1958 kfree_rcu(map, rcu);
1961 kfree_rcu(dev_maps, rcu);
1964 dev_maps = new_dev_maps;
1968 /* update Tx queue numa node */
1969 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
1970 (numa_node_id >= 0) ? numa_node_id :
1976 /* removes queue from unused CPUs */
1977 for_each_possible_cpu(cpu) {
1978 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu))
1981 if (remove_xps_queue(dev_maps, cpu, index))
1985 /* free map if not active */
1987 RCU_INIT_POINTER(dev->xps_maps, NULL);
1988 kfree_rcu(dev_maps, rcu);
1992 mutex_unlock(&xps_map_mutex);
1996 /* remove any maps that we added */
1997 for_each_possible_cpu(cpu) {
1998 new_map = xmap_dereference(new_dev_maps->cpu_map[cpu]);
1999 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[cpu]) :
2001 if (new_map && new_map != map)
2005 mutex_unlock(&xps_map_mutex);
2007 kfree(new_dev_maps);
2010 EXPORT_SYMBOL(netif_set_xps_queue);
2014 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2015 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
2017 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2021 if (txq < 1 || txq > dev->num_tx_queues)
2024 if (dev->reg_state == NETREG_REGISTERED ||
2025 dev->reg_state == NETREG_UNREGISTERING) {
2028 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2034 netif_setup_tc(dev, txq);
2036 if (txq < dev->real_num_tx_queues) {
2037 qdisc_reset_all_tx_gt(dev, txq);
2039 netif_reset_xps_queues_gt(dev, txq);
2044 dev->real_num_tx_queues = txq;
2047 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2051 * netif_set_real_num_rx_queues - set actual number of RX queues used
2052 * @dev: Network device
2053 * @rxq: Actual number of RX queues
2055 * This must be called either with the rtnl_lock held or before
2056 * registration of the net device. Returns 0 on success, or a
2057 * negative error code. If called before registration, it always
2060 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2064 if (rxq < 1 || rxq > dev->num_rx_queues)
2067 if (dev->reg_state == NETREG_REGISTERED) {
2070 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2076 dev->real_num_rx_queues = rxq;
2079 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2083 * netif_get_num_default_rss_queues - default number of RSS queues
2085 * This routine should set an upper limit on the number of RSS queues
2086 * used by default by multiqueue devices.
2088 int netif_get_num_default_rss_queues(void)
2090 return min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2092 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2094 static inline void __netif_reschedule(struct Qdisc *q)
2096 struct softnet_data *sd;
2097 unsigned long flags;
2099 local_irq_save(flags);
2100 sd = &__get_cpu_var(softnet_data);
2101 q->next_sched = NULL;
2102 *sd->output_queue_tailp = q;
2103 sd->output_queue_tailp = &q->next_sched;
2104 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2105 local_irq_restore(flags);
2108 void __netif_schedule(struct Qdisc *q)
2110 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2111 __netif_reschedule(q);
2113 EXPORT_SYMBOL(__netif_schedule);
2115 void dev_kfree_skb_irq(struct sk_buff *skb)
2117 if (atomic_dec_and_test(&skb->users)) {
2118 struct softnet_data *sd;
2119 unsigned long flags;
2121 local_irq_save(flags);
2122 sd = &__get_cpu_var(softnet_data);
2123 skb->next = sd->completion_queue;
2124 sd->completion_queue = skb;
2125 raise_softirq_irqoff(NET_TX_SOFTIRQ);
2126 local_irq_restore(flags);
2129 EXPORT_SYMBOL(dev_kfree_skb_irq);
2131 void dev_kfree_skb_any(struct sk_buff *skb)
2133 if (in_irq() || irqs_disabled())
2134 dev_kfree_skb_irq(skb);
2138 EXPORT_SYMBOL(dev_kfree_skb_any);
2142 * netif_device_detach - mark device as removed
2143 * @dev: network device
2145 * Mark device as removed from system and therefore no longer available.
2147 void netif_device_detach(struct net_device *dev)
2149 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2150 netif_running(dev)) {
2151 netif_tx_stop_all_queues(dev);
2154 EXPORT_SYMBOL(netif_device_detach);
2157 * netif_device_attach - mark device as attached
2158 * @dev: network device
2160 * Mark device as attached from system and restart if needed.
2162 void netif_device_attach(struct net_device *dev)
2164 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2165 netif_running(dev)) {
2166 netif_tx_wake_all_queues(dev);
2167 __netdev_watchdog_up(dev);
2170 EXPORT_SYMBOL(netif_device_attach);
2172 static void skb_warn_bad_offload(const struct sk_buff *skb)
2174 static const netdev_features_t null_features = 0;
2175 struct net_device *dev = skb->dev;
2176 const char *driver = "";
2178 if (!net_ratelimit())
2181 if (dev && dev->dev.parent)
2182 driver = dev_driver_string(dev->dev.parent);
2184 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2185 "gso_type=%d ip_summed=%d\n",
2186 driver, dev ? &dev->features : &null_features,
2187 skb->sk ? &skb->sk->sk_route_caps : &null_features,
2188 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2189 skb_shinfo(skb)->gso_type, skb->ip_summed);
2193 * Invalidate hardware checksum when packet is to be mangled, and
2194 * complete checksum manually on outgoing path.
2196 int skb_checksum_help(struct sk_buff *skb)
2199 int ret = 0, offset;
2201 if (skb->ip_summed == CHECKSUM_COMPLETE)
2202 goto out_set_summed;
2204 if (unlikely(skb_shinfo(skb)->gso_size)) {
2205 skb_warn_bad_offload(skb);
2209 /* Before computing a checksum, we should make sure no frag could
2210 * be modified by an external entity : checksum could be wrong.
2212 if (skb_has_shared_frag(skb)) {
2213 ret = __skb_linearize(skb);
2218 offset = skb_checksum_start_offset(skb);
2219 BUG_ON(offset >= skb_headlen(skb));
2220 csum = skb_checksum(skb, offset, skb->len - offset, 0);
2222 offset += skb->csum_offset;
2223 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2225 if (skb_cloned(skb) &&
2226 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2227 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2232 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2234 skb->ip_summed = CHECKSUM_NONE;
2238 EXPORT_SYMBOL(skb_checksum_help);
2240 __be16 skb_network_protocol(struct sk_buff *skb)
2242 __be16 type = skb->protocol;
2243 int vlan_depth = ETH_HLEN;
2245 /* Tunnel gso handlers can set protocol to ethernet. */
2246 if (type == htons(ETH_P_TEB)) {
2249 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2252 eth = (struct ethhdr *)skb_mac_header(skb);
2253 type = eth->h_proto;
2256 while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
2257 struct vlan_hdr *vh;
2259 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
2262 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
2263 type = vh->h_vlan_encapsulated_proto;
2264 vlan_depth += VLAN_HLEN;
2271 * skb_mac_gso_segment - mac layer segmentation handler.
2272 * @skb: buffer to segment
2273 * @features: features for the output path (see dev->features)
2275 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2276 netdev_features_t features)
2278 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2279 struct packet_offload *ptype;
2280 __be16 type = skb_network_protocol(skb);
2282 if (unlikely(!type))
2283 return ERR_PTR(-EINVAL);
2285 __skb_pull(skb, skb->mac_len);
2288 list_for_each_entry_rcu(ptype, &offload_base, list) {
2289 if (ptype->type == type && ptype->callbacks.gso_segment) {
2290 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
2293 err = ptype->callbacks.gso_send_check(skb);
2294 segs = ERR_PTR(err);
2295 if (err || skb_gso_ok(skb, features))
2297 __skb_push(skb, (skb->data -
2298 skb_network_header(skb)));
2300 segs = ptype->callbacks.gso_segment(skb, features);
2306 __skb_push(skb, skb->data - skb_mac_header(skb));
2310 EXPORT_SYMBOL(skb_mac_gso_segment);
2313 /* openvswitch calls this on rx path, so we need a different check.
2315 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2318 return skb->ip_summed != CHECKSUM_PARTIAL;
2320 return skb->ip_summed == CHECKSUM_NONE;
2324 * __skb_gso_segment - Perform segmentation on skb.
2325 * @skb: buffer to segment
2326 * @features: features for the output path (see dev->features)
2327 * @tx_path: whether it is called in TX path
2329 * This function segments the given skb and returns a list of segments.
2331 * It may return NULL if the skb requires no segmentation. This is
2332 * only possible when GSO is used for verifying header integrity.
2334 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2335 netdev_features_t features, bool tx_path)
2337 if (unlikely(skb_needs_check(skb, tx_path))) {
2340 skb_warn_bad_offload(skb);
2342 if (skb_header_cloned(skb) &&
2343 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
2344 return ERR_PTR(err);
2347 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2348 skb_reset_mac_header(skb);
2349 skb_reset_mac_len(skb);
2351 return skb_mac_gso_segment(skb, features);
2353 EXPORT_SYMBOL(__skb_gso_segment);
2355 /* Take action when hardware reception checksum errors are detected. */
2357 void netdev_rx_csum_fault(struct net_device *dev)
2359 if (net_ratelimit()) {
2360 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2364 EXPORT_SYMBOL(netdev_rx_csum_fault);
2367 /* Actually, we should eliminate this check as soon as we know, that:
2368 * 1. IOMMU is present and allows to map all the memory.
2369 * 2. No high memory really exists on this machine.
2372 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2374 #ifdef CONFIG_HIGHMEM
2376 if (!(dev->features & NETIF_F_HIGHDMA)) {
2377 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2378 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2379 if (PageHighMem(skb_frag_page(frag)))
2384 if (PCI_DMA_BUS_IS_PHYS) {
2385 struct device *pdev = dev->dev.parent;
2389 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2390 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2391 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2392 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2401 void (*destructor)(struct sk_buff *skb);
2404 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2406 static void dev_gso_skb_destructor(struct sk_buff *skb)
2408 struct dev_gso_cb *cb;
2411 struct sk_buff *nskb = skb->next;
2413 skb->next = nskb->next;
2416 } while (skb->next);
2418 cb = DEV_GSO_CB(skb);
2420 cb->destructor(skb);
2424 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2425 * @skb: buffer to segment
2426 * @features: device features as applicable to this skb
2428 * This function segments the given skb and stores the list of segments
2431 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2433 struct sk_buff *segs;
2435 segs = skb_gso_segment(skb, features);
2437 /* Verifying header integrity only. */
2442 return PTR_ERR(segs);
2445 DEV_GSO_CB(skb)->destructor = skb->destructor;
2446 skb->destructor = dev_gso_skb_destructor;
2451 static netdev_features_t harmonize_features(struct sk_buff *skb,
2452 __be16 protocol, netdev_features_t features)
2454 if (skb->ip_summed != CHECKSUM_NONE &&
2455 !can_checksum_protocol(features, protocol)) {
2456 features &= ~NETIF_F_ALL_CSUM;
2457 } else if (illegal_highdma(skb->dev, skb)) {
2458 features &= ~NETIF_F_SG;
2464 netdev_features_t netif_skb_features(struct sk_buff *skb)
2466 __be16 protocol = skb->protocol;
2467 netdev_features_t features = skb->dev->features;
2469 if (skb_shinfo(skb)->gso_segs > skb->dev->gso_max_segs)
2470 features &= ~NETIF_F_GSO_MASK;
2472 if (protocol == htons(ETH_P_8021Q) || protocol == htons(ETH_P_8021AD)) {
2473 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2474 protocol = veh->h_vlan_encapsulated_proto;
2475 } else if (!vlan_tx_tag_present(skb)) {
2476 return harmonize_features(skb, protocol, features);
2479 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_CTAG_TX |
2480 NETIF_F_HW_VLAN_STAG_TX);
2482 if (protocol != htons(ETH_P_8021Q) && protocol != htons(ETH_P_8021AD)) {
2483 return harmonize_features(skb, protocol, features);
2485 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2486 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2487 NETIF_F_HW_VLAN_STAG_TX;
2488 return harmonize_features(skb, protocol, features);
2491 EXPORT_SYMBOL(netif_skb_features);
2494 * Returns true if either:
2495 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2496 * 2. skb is fragmented and the device does not support SG.
2498 static inline int skb_needs_linearize(struct sk_buff *skb,
2499 netdev_features_t features)
2501 return skb_is_nonlinear(skb) &&
2502 ((skb_has_frag_list(skb) &&
2503 !(features & NETIF_F_FRAGLIST)) ||
2504 (skb_shinfo(skb)->nr_frags &&
2505 !(features & NETIF_F_SG)));
2508 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2509 struct netdev_queue *txq)
2511 const struct net_device_ops *ops = dev->netdev_ops;
2512 int rc = NETDEV_TX_OK;
2513 unsigned int skb_len;
2515 if (likely(!skb->next)) {
2516 netdev_features_t features;
2519 * If device doesn't need skb->dst, release it right now while
2520 * its hot in this cpu cache
2522 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2525 features = netif_skb_features(skb);
2527 if (vlan_tx_tag_present(skb) &&
2528 !vlan_hw_offload_capable(features, skb->vlan_proto)) {
2529 skb = __vlan_put_tag(skb, skb->vlan_proto,
2530 vlan_tx_tag_get(skb));
2537 /* If encapsulation offload request, verify we are testing
2538 * hardware encapsulation features instead of standard
2539 * features for the netdev
2541 if (skb->encapsulation)
2542 features &= dev->hw_enc_features;
2544 if (netif_needs_gso(skb, features)) {
2545 if (unlikely(dev_gso_segment(skb, features)))
2550 if (skb_needs_linearize(skb, features) &&
2551 __skb_linearize(skb))
2554 /* If packet is not checksummed and device does not
2555 * support checksumming for this protocol, complete
2556 * checksumming here.
2558 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2559 if (skb->encapsulation)
2560 skb_set_inner_transport_header(skb,
2561 skb_checksum_start_offset(skb));
2563 skb_set_transport_header(skb,
2564 skb_checksum_start_offset(skb));
2565 if (!(features & NETIF_F_ALL_CSUM) &&
2566 skb_checksum_help(skb))
2571 if (!list_empty(&ptype_all))
2572 dev_queue_xmit_nit(skb, dev);
2575 rc = ops->ndo_start_xmit(skb, dev);
2576 trace_net_dev_xmit(skb, rc, dev, skb_len);
2577 if (rc == NETDEV_TX_OK)
2578 txq_trans_update(txq);
2584 struct sk_buff *nskb = skb->next;
2586 skb->next = nskb->next;
2589 if (!list_empty(&ptype_all))
2590 dev_queue_xmit_nit(nskb, dev);
2592 skb_len = nskb->len;
2593 rc = ops->ndo_start_xmit(nskb, dev);
2594 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2595 if (unlikely(rc != NETDEV_TX_OK)) {
2596 if (rc & ~NETDEV_TX_MASK)
2597 goto out_kfree_gso_skb;
2598 nskb->next = skb->next;
2602 txq_trans_update(txq);
2603 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2604 return NETDEV_TX_BUSY;
2605 } while (skb->next);
2608 if (likely(skb->next == NULL)) {
2609 skb->destructor = DEV_GSO_CB(skb)->destructor;
2619 static void qdisc_pkt_len_init(struct sk_buff *skb)
2621 const struct skb_shared_info *shinfo = skb_shinfo(skb);
2623 qdisc_skb_cb(skb)->pkt_len = skb->len;
2625 /* To get more precise estimation of bytes sent on wire,
2626 * we add to pkt_len the headers size of all segments
2628 if (shinfo->gso_size) {
2629 unsigned int hdr_len;
2630 u16 gso_segs = shinfo->gso_segs;
2632 /* mac layer + network layer */
2633 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
2635 /* + transport layer */
2636 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
2637 hdr_len += tcp_hdrlen(skb);
2639 hdr_len += sizeof(struct udphdr);
2641 if (shinfo->gso_type & SKB_GSO_DODGY)
2642 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
2645 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
2649 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2650 struct net_device *dev,
2651 struct netdev_queue *txq)
2653 spinlock_t *root_lock = qdisc_lock(q);
2657 qdisc_pkt_len_init(skb);
2658 qdisc_calculate_pkt_len(skb, q);
2660 * Heuristic to force contended enqueues to serialize on a
2661 * separate lock before trying to get qdisc main lock.
2662 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2663 * and dequeue packets faster.
2665 contended = qdisc_is_running(q);
2666 if (unlikely(contended))
2667 spin_lock(&q->busylock);
2669 spin_lock(root_lock);
2670 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2673 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2674 qdisc_run_begin(q)) {
2676 * This is a work-conserving queue; there are no old skbs
2677 * waiting to be sent out; and the qdisc is not running -
2678 * xmit the skb directly.
2680 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2683 qdisc_bstats_update(q, skb);
2685 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2686 if (unlikely(contended)) {
2687 spin_unlock(&q->busylock);
2694 rc = NET_XMIT_SUCCESS;
2697 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2698 if (qdisc_run_begin(q)) {
2699 if (unlikely(contended)) {
2700 spin_unlock(&q->busylock);
2706 spin_unlock(root_lock);
2707 if (unlikely(contended))
2708 spin_unlock(&q->busylock);
2712 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2713 static void skb_update_prio(struct sk_buff *skb)
2715 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2717 if (!skb->priority && skb->sk && map) {
2718 unsigned int prioidx = skb->sk->sk_cgrp_prioidx;
2720 if (prioidx < map->priomap_len)
2721 skb->priority = map->priomap[prioidx];
2725 #define skb_update_prio(skb)
2728 static DEFINE_PER_CPU(int, xmit_recursion);
2729 #define RECURSION_LIMIT 10
2732 * dev_loopback_xmit - loop back @skb
2733 * @skb: buffer to transmit
2735 int dev_loopback_xmit(struct sk_buff *skb)
2737 skb_reset_mac_header(skb);
2738 __skb_pull(skb, skb_network_offset(skb));
2739 skb->pkt_type = PACKET_LOOPBACK;
2740 skb->ip_summed = CHECKSUM_UNNECESSARY;
2741 WARN_ON(!skb_dst(skb));
2746 EXPORT_SYMBOL(dev_loopback_xmit);
2749 * dev_queue_xmit - transmit a buffer
2750 * @skb: buffer to transmit
2752 * Queue a buffer for transmission to a network device. The caller must
2753 * have set the device and priority and built the buffer before calling
2754 * this function. The function can be called from an interrupt.
2756 * A negative errno code is returned on a failure. A success does not
2757 * guarantee the frame will be transmitted as it may be dropped due
2758 * to congestion or traffic shaping.
2760 * -----------------------------------------------------------------------------------
2761 * I notice this method can also return errors from the queue disciplines,
2762 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2765 * Regardless of the return value, the skb is consumed, so it is currently
2766 * difficult to retry a send to this method. (You can bump the ref count
2767 * before sending to hold a reference for retry if you are careful.)
2769 * When calling this method, interrupts MUST be enabled. This is because
2770 * the BH enable code must have IRQs enabled so that it will not deadlock.
2773 int dev_queue_xmit(struct sk_buff *skb)
2775 struct net_device *dev = skb->dev;
2776 struct netdev_queue *txq;
2780 skb_reset_mac_header(skb);
2782 /* Disable soft irqs for various locks below. Also
2783 * stops preemption for RCU.
2787 skb_update_prio(skb);
2789 txq = netdev_pick_tx(dev, skb);
2790 q = rcu_dereference_bh(txq->qdisc);
2792 #ifdef CONFIG_NET_CLS_ACT
2793 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2795 trace_net_dev_queue(skb);
2797 rc = __dev_xmit_skb(skb, q, dev, txq);
2801 /* The device has no queue. Common case for software devices:
2802 loopback, all the sorts of tunnels...
2804 Really, it is unlikely that netif_tx_lock protection is necessary
2805 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2807 However, it is possible, that they rely on protection
2810 Check this and shot the lock. It is not prone from deadlocks.
2811 Either shot noqueue qdisc, it is even simpler 8)
2813 if (dev->flags & IFF_UP) {
2814 int cpu = smp_processor_id(); /* ok because BHs are off */
2816 if (txq->xmit_lock_owner != cpu) {
2818 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2819 goto recursion_alert;
2821 HARD_TX_LOCK(dev, txq, cpu);
2823 if (!netif_xmit_stopped(txq)) {
2824 __this_cpu_inc(xmit_recursion);
2825 rc = dev_hard_start_xmit(skb, dev, txq);
2826 __this_cpu_dec(xmit_recursion);
2827 if (dev_xmit_complete(rc)) {
2828 HARD_TX_UNLOCK(dev, txq);
2832 HARD_TX_UNLOCK(dev, txq);
2833 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2836 /* Recursion is detected! It is possible,
2840 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2846 rcu_read_unlock_bh();
2851 rcu_read_unlock_bh();
2854 EXPORT_SYMBOL(dev_queue_xmit);
2857 /*=======================================================================
2859 =======================================================================*/
2861 int netdev_max_backlog __read_mostly = 1000;
2862 EXPORT_SYMBOL(netdev_max_backlog);
2864 int netdev_tstamp_prequeue __read_mostly = 1;
2865 int netdev_budget __read_mostly = 300;
2866 int weight_p __read_mostly = 64; /* old backlog weight */
2868 /* Called with irq disabled */
2869 static inline void ____napi_schedule(struct softnet_data *sd,
2870 struct napi_struct *napi)
2872 list_add_tail(&napi->poll_list, &sd->poll_list);
2873 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2878 /* One global table that all flow-based protocols share. */
2879 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2880 EXPORT_SYMBOL(rps_sock_flow_table);
2882 struct static_key rps_needed __read_mostly;
2884 static struct rps_dev_flow *
2885 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2886 struct rps_dev_flow *rflow, u16 next_cpu)
2888 if (next_cpu != RPS_NO_CPU) {
2889 #ifdef CONFIG_RFS_ACCEL
2890 struct netdev_rx_queue *rxqueue;
2891 struct rps_dev_flow_table *flow_table;
2892 struct rps_dev_flow *old_rflow;
2897 /* Should we steer this flow to a different hardware queue? */
2898 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2899 !(dev->features & NETIF_F_NTUPLE))
2901 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2902 if (rxq_index == skb_get_rx_queue(skb))
2905 rxqueue = dev->_rx + rxq_index;
2906 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2909 flow_id = skb->rxhash & flow_table->mask;
2910 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2911 rxq_index, flow_id);
2915 rflow = &flow_table->flows[flow_id];
2917 if (old_rflow->filter == rflow->filter)
2918 old_rflow->filter = RPS_NO_FILTER;
2922 per_cpu(softnet_data, next_cpu).input_queue_head;
2925 rflow->cpu = next_cpu;
2930 * get_rps_cpu is called from netif_receive_skb and returns the target
2931 * CPU from the RPS map of the receiving queue for a given skb.
2932 * rcu_read_lock must be held on entry.
2934 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2935 struct rps_dev_flow **rflowp)
2937 struct netdev_rx_queue *rxqueue;
2938 struct rps_map *map;
2939 struct rps_dev_flow_table *flow_table;
2940 struct rps_sock_flow_table *sock_flow_table;
2944 if (skb_rx_queue_recorded(skb)) {
2945 u16 index = skb_get_rx_queue(skb);
2946 if (unlikely(index >= dev->real_num_rx_queues)) {
2947 WARN_ONCE(dev->real_num_rx_queues > 1,
2948 "%s received packet on queue %u, but number "
2949 "of RX queues is %u\n",
2950 dev->name, index, dev->real_num_rx_queues);
2953 rxqueue = dev->_rx + index;
2957 map = rcu_dereference(rxqueue->rps_map);
2959 if (map->len == 1 &&
2960 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2961 tcpu = map->cpus[0];
2962 if (cpu_online(tcpu))
2966 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2970 skb_reset_network_header(skb);
2971 if (!skb_get_rxhash(skb))
2974 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2975 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2976 if (flow_table && sock_flow_table) {
2978 struct rps_dev_flow *rflow;
2980 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2983 next_cpu = sock_flow_table->ents[skb->rxhash &
2984 sock_flow_table->mask];
2987 * If the desired CPU (where last recvmsg was done) is
2988 * different from current CPU (one in the rx-queue flow
2989 * table entry), switch if one of the following holds:
2990 * - Current CPU is unset (equal to RPS_NO_CPU).
2991 * - Current CPU is offline.
2992 * - The current CPU's queue tail has advanced beyond the
2993 * last packet that was enqueued using this table entry.
2994 * This guarantees that all previous packets for the flow
2995 * have been dequeued, thus preserving in order delivery.
2997 if (unlikely(tcpu != next_cpu) &&
2998 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2999 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3000 rflow->last_qtail)) >= 0)) {
3002 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3005 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
3013 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
3015 if (cpu_online(tcpu)) {
3025 #ifdef CONFIG_RFS_ACCEL
3028 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3029 * @dev: Device on which the filter was set
3030 * @rxq_index: RX queue index
3031 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3032 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3034 * Drivers that implement ndo_rx_flow_steer() should periodically call
3035 * this function for each installed filter and remove the filters for
3036 * which it returns %true.
3038 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3039 u32 flow_id, u16 filter_id)
3041 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3042 struct rps_dev_flow_table *flow_table;
3043 struct rps_dev_flow *rflow;
3048 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3049 if (flow_table && flow_id <= flow_table->mask) {
3050 rflow = &flow_table->flows[flow_id];
3051 cpu = ACCESS_ONCE(rflow->cpu);
3052 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
3053 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3054 rflow->last_qtail) <
3055 (int)(10 * flow_table->mask)))
3061 EXPORT_SYMBOL(rps_may_expire_flow);
3063 #endif /* CONFIG_RFS_ACCEL */
3065 /* Called from hardirq (IPI) context */
3066 static void rps_trigger_softirq(void *data)
3068 struct softnet_data *sd = data;
3070 ____napi_schedule(sd, &sd->backlog);
3074 #endif /* CONFIG_RPS */
3077 * Check if this softnet_data structure is another cpu one
3078 * If yes, queue it to our IPI list and return 1
3081 static int rps_ipi_queued(struct softnet_data *sd)
3084 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
3087 sd->rps_ipi_next = mysd->rps_ipi_list;
3088 mysd->rps_ipi_list = sd;
3090 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3093 #endif /* CONFIG_RPS */
3097 #ifdef CONFIG_NET_FLOW_LIMIT
3098 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3101 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3103 #ifdef CONFIG_NET_FLOW_LIMIT
3104 struct sd_flow_limit *fl;
3105 struct softnet_data *sd;
3106 unsigned int old_flow, new_flow;
3108 if (qlen < (netdev_max_backlog >> 1))
3111 sd = &__get_cpu_var(softnet_data);
3114 fl = rcu_dereference(sd->flow_limit);
3116 new_flow = skb_get_rxhash(skb) & (fl->num_buckets - 1);
3117 old_flow = fl->history[fl->history_head];
3118 fl->history[fl->history_head] = new_flow;
3121 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3123 if (likely(fl->buckets[old_flow]))
3124 fl->buckets[old_flow]--;
3126 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3138 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3139 * queue (may be a remote CPU queue).
3141 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3142 unsigned int *qtail)
3144 struct softnet_data *sd;
3145 unsigned long flags;
3148 sd = &per_cpu(softnet_data, cpu);
3150 local_irq_save(flags);
3153 qlen = skb_queue_len(&sd->input_pkt_queue);
3154 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3155 if (skb_queue_len(&sd->input_pkt_queue)) {
3157 __skb_queue_tail(&sd->input_pkt_queue, skb);
3158 input_queue_tail_incr_save(sd, qtail);
3160 local_irq_restore(flags);
3161 return NET_RX_SUCCESS;
3164 /* Schedule NAPI for backlog device
3165 * We can use non atomic operation since we own the queue lock
3167 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3168 if (!rps_ipi_queued(sd))
3169 ____napi_schedule(sd, &sd->backlog);
3177 local_irq_restore(flags);
3179 atomic_long_inc(&skb->dev->rx_dropped);
3185 * netif_rx - post buffer to the network code
3186 * @skb: buffer to post
3188 * This function receives a packet from a device driver and queues it for
3189 * the upper (protocol) levels to process. It always succeeds. The buffer
3190 * may be dropped during processing for congestion control or by the
3194 * NET_RX_SUCCESS (no congestion)
3195 * NET_RX_DROP (packet was dropped)
3199 int netif_rx(struct sk_buff *skb)
3203 /* if netpoll wants it, pretend we never saw it */
3204 if (netpoll_rx(skb))
3207 net_timestamp_check(netdev_tstamp_prequeue, skb);
3209 trace_netif_rx(skb);
3211 if (static_key_false(&rps_needed)) {
3212 struct rps_dev_flow voidflow, *rflow = &voidflow;
3218 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3220 cpu = smp_processor_id();
3222 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3230 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
3235 EXPORT_SYMBOL(netif_rx);
3237 int netif_rx_ni(struct sk_buff *skb)
3242 err = netif_rx(skb);
3243 if (local_softirq_pending())
3249 EXPORT_SYMBOL(netif_rx_ni);
3251 static void net_tx_action(struct softirq_action *h)
3253 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3255 if (sd->completion_queue) {
3256 struct sk_buff *clist;
3258 local_irq_disable();
3259 clist = sd->completion_queue;
3260 sd->completion_queue = NULL;
3264 struct sk_buff *skb = clist;
3265 clist = clist->next;
3267 WARN_ON(atomic_read(&skb->users));
3268 trace_kfree_skb(skb, net_tx_action);
3273 if (sd->output_queue) {
3276 local_irq_disable();
3277 head = sd->output_queue;
3278 sd->output_queue = NULL;
3279 sd->output_queue_tailp = &sd->output_queue;
3283 struct Qdisc *q = head;
3284 spinlock_t *root_lock;
3286 head = head->next_sched;
3288 root_lock = qdisc_lock(q);
3289 if (spin_trylock(root_lock)) {
3290 smp_mb__before_clear_bit();
3291 clear_bit(__QDISC_STATE_SCHED,
3294 spin_unlock(root_lock);
3296 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3298 __netif_reschedule(q);
3300 smp_mb__before_clear_bit();
3301 clear_bit(__QDISC_STATE_SCHED,
3309 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3310 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3311 /* This hook is defined here for ATM LANE */
3312 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3313 unsigned char *addr) __read_mostly;
3314 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3317 #ifdef CONFIG_NET_CLS_ACT
3318 /* TODO: Maybe we should just force sch_ingress to be compiled in
3319 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3320 * a compare and 2 stores extra right now if we dont have it on
3321 * but have CONFIG_NET_CLS_ACT
3322 * NOTE: This doesn't stop any functionality; if you dont have
3323 * the ingress scheduler, you just can't add policies on ingress.
3326 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3328 struct net_device *dev = skb->dev;
3329 u32 ttl = G_TC_RTTL(skb->tc_verd);
3330 int result = TC_ACT_OK;
3333 if (unlikely(MAX_RED_LOOP < ttl++)) {
3334 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3335 skb->skb_iif, dev->ifindex);
3339 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3340 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3343 if (q != &noop_qdisc) {
3344 spin_lock(qdisc_lock(q));
3345 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3346 result = qdisc_enqueue_root(skb, q);
3347 spin_unlock(qdisc_lock(q));
3353 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3354 struct packet_type **pt_prev,
3355 int *ret, struct net_device *orig_dev)
3357 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3359 if (!rxq || rxq->qdisc == &noop_qdisc)
3363 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3367 switch (ing_filter(skb, rxq)) {
3381 * netdev_rx_handler_register - register receive handler
3382 * @dev: device to register a handler for
3383 * @rx_handler: receive handler to register
3384 * @rx_handler_data: data pointer that is used by rx handler
3386 * Register a receive hander for a device. This handler will then be
3387 * called from __netif_receive_skb. A negative errno code is returned
3390 * The caller must hold the rtnl_mutex.
3392 * For a general description of rx_handler, see enum rx_handler_result.
3394 int netdev_rx_handler_register(struct net_device *dev,
3395 rx_handler_func_t *rx_handler,
3396 void *rx_handler_data)
3400 if (dev->rx_handler)
3403 /* Note: rx_handler_data must be set before rx_handler */
3404 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3405 rcu_assign_pointer(dev->rx_handler, rx_handler);
3409 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3412 * netdev_rx_handler_unregister - unregister receive handler
3413 * @dev: device to unregister a handler from
3415 * Unregister a receive handler from a device.
3417 * The caller must hold the rtnl_mutex.
3419 void netdev_rx_handler_unregister(struct net_device *dev)
3423 RCU_INIT_POINTER(dev->rx_handler, NULL);
3424 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
3425 * section has a guarantee to see a non NULL rx_handler_data
3429 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3431 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3434 * Limit the use of PFMEMALLOC reserves to those protocols that implement
3435 * the special handling of PFMEMALLOC skbs.
3437 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
3439 switch (skb->protocol) {
3440 case __constant_htons(ETH_P_ARP):
3441 case __constant_htons(ETH_P_IP):
3442 case __constant_htons(ETH_P_IPV6):
3443 case __constant_htons(ETH_P_8021Q):
3444 case __constant_htons(ETH_P_8021AD):
3451 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
3453 struct packet_type *ptype, *pt_prev;
3454 rx_handler_func_t *rx_handler;
3455 struct net_device *orig_dev;
3456 struct net_device *null_or_dev;
3457 bool deliver_exact = false;
3458 int ret = NET_RX_DROP;
3461 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3463 trace_netif_receive_skb(skb);
3465 /* if we've gotten here through NAPI, check netpoll */
3466 if (netpoll_receive_skb(skb))
3469 orig_dev = skb->dev;
3471 skb_reset_network_header(skb);
3472 if (!skb_transport_header_was_set(skb))
3473 skb_reset_transport_header(skb);
3474 skb_reset_mac_len(skb);
3481 skb->skb_iif = skb->dev->ifindex;
3483 __this_cpu_inc(softnet_data.processed);
3485 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
3486 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
3487 skb = vlan_untag(skb);
3492 #ifdef CONFIG_NET_CLS_ACT
3493 if (skb->tc_verd & TC_NCLS) {
3494 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3502 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3503 if (!ptype->dev || ptype->dev == skb->dev) {
3505 ret = deliver_skb(skb, pt_prev, orig_dev);
3511 #ifdef CONFIG_NET_CLS_ACT
3512 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3518 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
3521 if (vlan_tx_tag_present(skb)) {
3523 ret = deliver_skb(skb, pt_prev, orig_dev);
3526 if (vlan_do_receive(&skb))
3528 else if (unlikely(!skb))
3532 rx_handler = rcu_dereference(skb->dev->rx_handler);
3535 ret = deliver_skb(skb, pt_prev, orig_dev);
3538 switch (rx_handler(&skb)) {
3539 case RX_HANDLER_CONSUMED:
3540 ret = NET_RX_SUCCESS;
3542 case RX_HANDLER_ANOTHER:
3544 case RX_HANDLER_EXACT:
3545 deliver_exact = true;
3546 case RX_HANDLER_PASS:
3553 if (vlan_tx_nonzero_tag_present(skb))
3554 skb->pkt_type = PACKET_OTHERHOST;
3556 /* deliver only exact match when indicated */
3557 null_or_dev = deliver_exact ? skb->dev : NULL;
3559 type = skb->protocol;
3560 list_for_each_entry_rcu(ptype,
3561 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3562 if (ptype->type == type &&
3563 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3564 ptype->dev == orig_dev)) {
3566 ret = deliver_skb(skb, pt_prev, orig_dev);
3572 if (unlikely(skb_orphan_frags(skb, GFP_ATOMIC)))
3575 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3578 atomic_long_inc(&skb->dev->rx_dropped);
3580 /* Jamal, now you will not able to escape explaining
3581 * me how you were going to use this. :-)
3592 static int __netif_receive_skb(struct sk_buff *skb)
3596 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
3597 unsigned long pflags = current->flags;
3600 * PFMEMALLOC skbs are special, they should
3601 * - be delivered to SOCK_MEMALLOC sockets only
3602 * - stay away from userspace
3603 * - have bounded memory usage
3605 * Use PF_MEMALLOC as this saves us from propagating the allocation
3606 * context down to all allocation sites.
3608 current->flags |= PF_MEMALLOC;
3609 ret = __netif_receive_skb_core(skb, true);
3610 tsk_restore_flags(current, pflags, PF_MEMALLOC);
3612 ret = __netif_receive_skb_core(skb, false);
3618 * netif_receive_skb - process receive buffer from network
3619 * @skb: buffer to process
3621 * netif_receive_skb() is the main receive data processing function.
3622 * It always succeeds. The buffer may be dropped during processing
3623 * for congestion control or by the protocol layers.
3625 * This function may only be called from softirq context and interrupts
3626 * should be enabled.
3628 * Return values (usually ignored):
3629 * NET_RX_SUCCESS: no congestion
3630 * NET_RX_DROP: packet was dropped
3632 int netif_receive_skb(struct sk_buff *skb)
3634 net_timestamp_check(netdev_tstamp_prequeue, skb);
3636 if (skb_defer_rx_timestamp(skb))
3637 return NET_RX_SUCCESS;
3640 if (static_key_false(&rps_needed)) {
3641 struct rps_dev_flow voidflow, *rflow = &voidflow;
3646 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3649 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3656 return __netif_receive_skb(skb);
3658 EXPORT_SYMBOL(netif_receive_skb);
3660 /* Network device is going away, flush any packets still pending
3661 * Called with irqs disabled.
3663 static void flush_backlog(void *arg)
3665 struct net_device *dev = arg;
3666 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3667 struct sk_buff *skb, *tmp;
3670 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3671 if (skb->dev == dev) {
3672 __skb_unlink(skb, &sd->input_pkt_queue);
3674 input_queue_head_incr(sd);
3679 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3680 if (skb->dev == dev) {
3681 __skb_unlink(skb, &sd->process_queue);
3683 input_queue_head_incr(sd);
3688 static int napi_gro_complete(struct sk_buff *skb)
3690 struct packet_offload *ptype;
3691 __be16 type = skb->protocol;
3692 struct list_head *head = &offload_base;
3695 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
3697 if (NAPI_GRO_CB(skb)->count == 1) {
3698 skb_shinfo(skb)->gso_size = 0;
3703 list_for_each_entry_rcu(ptype, head, list) {
3704 if (ptype->type != type || !ptype->callbacks.gro_complete)
3707 err = ptype->callbacks.gro_complete(skb);
3713 WARN_ON(&ptype->list == head);
3715 return NET_RX_SUCCESS;
3719 return netif_receive_skb(skb);
3722 /* napi->gro_list contains packets ordered by age.
3723 * youngest packets at the head of it.
3724 * Complete skbs in reverse order to reduce latencies.
3726 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
3728 struct sk_buff *skb, *prev = NULL;
3730 /* scan list and build reverse chain */
3731 for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
3736 for (skb = prev; skb; skb = prev) {
3739 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
3743 napi_gro_complete(skb);
3747 napi->gro_list = NULL;
3749 EXPORT_SYMBOL(napi_gro_flush);
3751 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
3754 unsigned int maclen = skb->dev->hard_header_len;
3756 for (p = napi->gro_list; p; p = p->next) {
3757 unsigned long diffs;
3759 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3760 diffs |= p->vlan_tci ^ skb->vlan_tci;
3761 if (maclen == ETH_HLEN)
3762 diffs |= compare_ether_header(skb_mac_header(p),
3763 skb_gro_mac_header(skb));
3765 diffs = memcmp(skb_mac_header(p),
3766 skb_gro_mac_header(skb),
3768 NAPI_GRO_CB(p)->same_flow = !diffs;
3769 NAPI_GRO_CB(p)->flush = 0;
3773 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3775 struct sk_buff **pp = NULL;
3776 struct packet_offload *ptype;
3777 __be16 type = skb->protocol;
3778 struct list_head *head = &offload_base;
3780 enum gro_result ret;
3782 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3785 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3788 gro_list_prepare(napi, skb);
3791 list_for_each_entry_rcu(ptype, head, list) {
3792 if (ptype->type != type || !ptype->callbacks.gro_receive)
3795 skb_set_network_header(skb, skb_gro_offset(skb));
3796 skb_reset_mac_len(skb);
3797 NAPI_GRO_CB(skb)->same_flow = 0;
3798 NAPI_GRO_CB(skb)->flush = 0;
3799 NAPI_GRO_CB(skb)->free = 0;
3801 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
3806 if (&ptype->list == head)
3809 same_flow = NAPI_GRO_CB(skb)->same_flow;
3810 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3813 struct sk_buff *nskb = *pp;
3817 napi_gro_complete(nskb);
3824 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3828 NAPI_GRO_CB(skb)->count = 1;
3829 NAPI_GRO_CB(skb)->age = jiffies;
3830 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3831 skb->next = napi->gro_list;
3832 napi->gro_list = skb;
3836 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3837 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3839 BUG_ON(skb->end - skb->tail < grow);
3841 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3844 skb->data_len -= grow;
3846 skb_shinfo(skb)->frags[0].page_offset += grow;
3847 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3849 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3850 skb_frag_unref(skb, 0);
3851 memmove(skb_shinfo(skb)->frags,
3852 skb_shinfo(skb)->frags + 1,
3853 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3866 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3870 if (netif_receive_skb(skb))
3878 case GRO_MERGED_FREE:
3879 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3880 kmem_cache_free(skbuff_head_cache, skb);
3893 static void skb_gro_reset_offset(struct sk_buff *skb)
3895 const struct skb_shared_info *pinfo = skb_shinfo(skb);
3896 const skb_frag_t *frag0 = &pinfo->frags[0];
3898 NAPI_GRO_CB(skb)->data_offset = 0;
3899 NAPI_GRO_CB(skb)->frag0 = NULL;
3900 NAPI_GRO_CB(skb)->frag0_len = 0;
3902 if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
3904 !PageHighMem(skb_frag_page(frag0))) {
3905 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
3906 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(frag0);
3910 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3912 skb_gro_reset_offset(skb);
3914 return napi_skb_finish(dev_gro_receive(napi, skb), skb);
3916 EXPORT_SYMBOL(napi_gro_receive);
3918 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3920 __skb_pull(skb, skb_headlen(skb));
3921 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3922 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3924 skb->dev = napi->dev;
3930 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3932 struct sk_buff *skb = napi->skb;
3935 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3941 EXPORT_SYMBOL(napi_get_frags);
3943 static gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3949 skb->protocol = eth_type_trans(skb, skb->dev);
3951 if (ret == GRO_HELD)
3952 skb_gro_pull(skb, -ETH_HLEN);
3953 else if (netif_receive_skb(skb))
3958 case GRO_MERGED_FREE:
3959 napi_reuse_skb(napi, skb);
3969 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3971 struct sk_buff *skb = napi->skb;
3978 skb_reset_mac_header(skb);
3979 skb_gro_reset_offset(skb);
3981 off = skb_gro_offset(skb);
3982 hlen = off + sizeof(*eth);
3983 eth = skb_gro_header_fast(skb, off);
3984 if (skb_gro_header_hard(skb, hlen)) {
3985 eth = skb_gro_header_slow(skb, hlen, off);
3986 if (unlikely(!eth)) {
3987 napi_reuse_skb(napi, skb);
3993 skb_gro_pull(skb, sizeof(*eth));
3996 * This works because the only protocols we care about don't require
3997 * special handling. We'll fix it up properly at the end.
3999 skb->protocol = eth->h_proto;
4005 gro_result_t napi_gro_frags(struct napi_struct *napi)
4007 struct sk_buff *skb = napi_frags_skb(napi);
4012 return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
4014 EXPORT_SYMBOL(napi_gro_frags);
4017 * net_rps_action sends any pending IPI's for rps.
4018 * Note: called with local irq disabled, but exits with local irq enabled.
4020 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
4023 struct softnet_data *remsd = sd->rps_ipi_list;
4026 sd->rps_ipi_list = NULL;
4030 /* Send pending IPI's to kick RPS processing on remote cpus. */
4032 struct softnet_data *next = remsd->rps_ipi_next;
4034 if (cpu_online(remsd->cpu))
4035 __smp_call_function_single(remsd->cpu,
4044 static int process_backlog(struct napi_struct *napi, int quota)
4047 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
4050 /* Check if we have pending ipi, its better to send them now,
4051 * not waiting net_rx_action() end.
4053 if (sd->rps_ipi_list) {
4054 local_irq_disable();
4055 net_rps_action_and_irq_enable(sd);
4058 napi->weight = weight_p;
4059 local_irq_disable();
4060 while (work < quota) {
4061 struct sk_buff *skb;
4064 while ((skb = __skb_dequeue(&sd->process_queue))) {
4066 __netif_receive_skb(skb);
4067 local_irq_disable();
4068 input_queue_head_incr(sd);
4069 if (++work >= quota) {
4076 qlen = skb_queue_len(&sd->input_pkt_queue);
4078 skb_queue_splice_tail_init(&sd->input_pkt_queue,
4079 &sd->process_queue);
4081 if (qlen < quota - work) {
4083 * Inline a custom version of __napi_complete().
4084 * only current cpu owns and manipulates this napi,
4085 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
4086 * we can use a plain write instead of clear_bit(),
4087 * and we dont need an smp_mb() memory barrier.
4089 list_del(&napi->poll_list);
4092 quota = work + qlen;
4102 * __napi_schedule - schedule for receive
4103 * @n: entry to schedule
4105 * The entry's receive function will be scheduled to run
4107 void __napi_schedule(struct napi_struct *n)
4109 unsigned long flags;
4111 local_irq_save(flags);
4112 ____napi_schedule(&__get_cpu_var(softnet_data), n);
4113 local_irq_restore(flags);
4115 EXPORT_SYMBOL(__napi_schedule);
4117 void __napi_complete(struct napi_struct *n)
4119 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
4120 BUG_ON(n->gro_list);
4122 list_del(&n->poll_list);
4123 smp_mb__before_clear_bit();
4124 clear_bit(NAPI_STATE_SCHED, &n->state);
4126 EXPORT_SYMBOL(__napi_complete);
4128 void napi_complete(struct napi_struct *n)
4130 unsigned long flags;
4133 * don't let napi dequeue from the cpu poll list
4134 * just in case its running on a different cpu
4136 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
4139 napi_gro_flush(n, false);
4140 local_irq_save(flags);
4142 local_irq_restore(flags);
4144 EXPORT_SYMBOL(napi_complete);
4146 /* must be called under rcu_read_lock(), as we dont take a reference */
4147 struct napi_struct *napi_by_id(unsigned int napi_id)
4149 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
4150 struct napi_struct *napi;
4152 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
4153 if (napi->napi_id == napi_id)
4158 EXPORT_SYMBOL_GPL(napi_by_id);
4160 void napi_hash_add(struct napi_struct *napi)
4162 if (!test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) {
4164 spin_lock(&napi_hash_lock);
4166 /* 0 is not a valid id, we also skip an id that is taken
4167 * we expect both events to be extremely rare
4170 while (!napi->napi_id) {
4171 napi->napi_id = ++napi_gen_id;
4172 if (napi_by_id(napi->napi_id))
4176 hlist_add_head_rcu(&napi->napi_hash_node,
4177 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
4179 spin_unlock(&napi_hash_lock);
4182 EXPORT_SYMBOL_GPL(napi_hash_add);
4184 /* Warning : caller is responsible to make sure rcu grace period
4185 * is respected before freeing memory containing @napi
4187 void napi_hash_del(struct napi_struct *napi)
4189 spin_lock(&napi_hash_lock);
4191 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state))
4192 hlist_del_rcu(&napi->napi_hash_node);
4194 spin_unlock(&napi_hash_lock);
4196 EXPORT_SYMBOL_GPL(napi_hash_del);
4198 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
4199 int (*poll)(struct napi_struct *, int), int weight)
4201 INIT_LIST_HEAD(&napi->poll_list);
4202 napi->gro_count = 0;
4203 napi->gro_list = NULL;
4206 if (weight > NAPI_POLL_WEIGHT)
4207 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
4209 napi->weight = weight;
4210 list_add(&napi->dev_list, &dev->napi_list);
4212 #ifdef CONFIG_NETPOLL
4213 spin_lock_init(&napi->poll_lock);
4214 napi->poll_owner = -1;
4216 set_bit(NAPI_STATE_SCHED, &napi->state);
4218 EXPORT_SYMBOL(netif_napi_add);
4220 void netif_napi_del(struct napi_struct *napi)
4222 struct sk_buff *skb, *next;
4224 list_del_init(&napi->dev_list);
4225 napi_free_frags(napi);
4227 for (skb = napi->gro_list; skb; skb = next) {
4233 napi->gro_list = NULL;
4234 napi->gro_count = 0;
4236 EXPORT_SYMBOL(netif_napi_del);
4238 static void net_rx_action(struct softirq_action *h)
4240 struct softnet_data *sd = &__get_cpu_var(softnet_data);
4241 unsigned long time_limit = jiffies + 2;
4242 int budget = netdev_budget;
4245 local_irq_disable();
4247 while (!list_empty(&sd->poll_list)) {
4248 struct napi_struct *n;
4251 /* If softirq window is exhuasted then punt.
4252 * Allow this to run for 2 jiffies since which will allow
4253 * an average latency of 1.5/HZ.
4255 if (unlikely(budget <= 0 || time_after_eq(jiffies, time_limit)))
4260 /* Even though interrupts have been re-enabled, this
4261 * access is safe because interrupts can only add new
4262 * entries to the tail of this list, and only ->poll()
4263 * calls can remove this head entry from the list.
4265 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
4267 have = netpoll_poll_lock(n);
4271 /* This NAPI_STATE_SCHED test is for avoiding a race
4272 * with netpoll's poll_napi(). Only the entity which
4273 * obtains the lock and sees NAPI_STATE_SCHED set will
4274 * actually make the ->poll() call. Therefore we avoid
4275 * accidentally calling ->poll() when NAPI is not scheduled.
4278 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
4279 work = n->poll(n, weight);
4283 WARN_ON_ONCE(work > weight);
4287 local_irq_disable();
4289 /* Drivers must not modify the NAPI state if they
4290 * consume the entire weight. In such cases this code
4291 * still "owns" the NAPI instance and therefore can
4292 * move the instance around on the list at-will.
4294 if (unlikely(work == weight)) {
4295 if (unlikely(napi_disable_pending(n))) {
4298 local_irq_disable();
4301 /* flush too old packets
4302 * If HZ < 1000, flush all packets.
4305 napi_gro_flush(n, HZ >= 1000);
4306 local_irq_disable();
4308 list_move_tail(&n->poll_list, &sd->poll_list);
4312 netpoll_poll_unlock(have);
4315 net_rps_action_and_irq_enable(sd);
4317 #ifdef CONFIG_NET_DMA
4319 * There may not be any more sk_buffs coming right now, so push
4320 * any pending DMA copies to hardware
4322 dma_issue_pending_all();
4329 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4333 struct netdev_upper {
4334 struct net_device *dev;
4336 struct list_head list;
4337 struct rcu_head rcu;
4338 struct list_head search_list;
4341 static void __append_search_uppers(struct list_head *search_list,
4342 struct net_device *dev)
4344 struct netdev_upper *upper;
4346 list_for_each_entry(upper, &dev->upper_dev_list, list) {
4347 /* check if this upper is not already in search list */
4348 if (list_empty(&upper->search_list))
4349 list_add_tail(&upper->search_list, search_list);
4353 static bool __netdev_search_upper_dev(struct net_device *dev,
4354 struct net_device *upper_dev)
4356 LIST_HEAD(search_list);
4357 struct netdev_upper *upper;
4358 struct netdev_upper *tmp;
4361 __append_search_uppers(&search_list, dev);
4362 list_for_each_entry(upper, &search_list, search_list) {
4363 if (upper->dev == upper_dev) {
4367 __append_search_uppers(&search_list, upper->dev);
4369 list_for_each_entry_safe(upper, tmp, &search_list, search_list)
4370 INIT_LIST_HEAD(&upper->search_list);
4374 static struct netdev_upper *__netdev_find_upper(struct net_device *dev,
4375 struct net_device *upper_dev)
4377 struct netdev_upper *upper;
4379 list_for_each_entry(upper, &dev->upper_dev_list, list) {
4380 if (upper->dev == upper_dev)
4387 * netdev_has_upper_dev - Check if device is linked to an upper device
4389 * @upper_dev: upper device to check
4391 * Find out if a device is linked to specified upper device and return true
4392 * in case it is. Note that this checks only immediate upper device,
4393 * not through a complete stack of devices. The caller must hold the RTNL lock.
4395 bool netdev_has_upper_dev(struct net_device *dev,
4396 struct net_device *upper_dev)
4400 return __netdev_find_upper(dev, upper_dev);
4402 EXPORT_SYMBOL(netdev_has_upper_dev);
4405 * netdev_has_any_upper_dev - Check if device is linked to some device
4408 * Find out if a device is linked to an upper device and return true in case
4409 * it is. The caller must hold the RTNL lock.
4411 bool netdev_has_any_upper_dev(struct net_device *dev)
4415 return !list_empty(&dev->upper_dev_list);
4417 EXPORT_SYMBOL(netdev_has_any_upper_dev);
4420 * netdev_master_upper_dev_get - Get master upper device
4423 * Find a master upper device and return pointer to it or NULL in case
4424 * it's not there. The caller must hold the RTNL lock.
4426 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
4428 struct netdev_upper *upper;
4432 if (list_empty(&dev->upper_dev_list))
4435 upper = list_first_entry(&dev->upper_dev_list,
4436 struct netdev_upper, list);
4437 if (likely(upper->master))
4441 EXPORT_SYMBOL(netdev_master_upper_dev_get);
4444 * netdev_master_upper_dev_get_rcu - Get master upper device
4447 * Find a master upper device and return pointer to it or NULL in case
4448 * it's not there. The caller must hold the RCU read lock.
4450 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
4452 struct netdev_upper *upper;
4454 upper = list_first_or_null_rcu(&dev->upper_dev_list,
4455 struct netdev_upper, list);
4456 if (upper && likely(upper->master))
4460 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
4462 static int __netdev_upper_dev_link(struct net_device *dev,
4463 struct net_device *upper_dev, bool master)
4465 struct netdev_upper *upper;
4469 if (dev == upper_dev)
4472 /* To prevent loops, check if dev is not upper device to upper_dev. */
4473 if (__netdev_search_upper_dev(upper_dev, dev))
4476 if (__netdev_find_upper(dev, upper_dev))
4479 if (master && netdev_master_upper_dev_get(dev))
4482 upper = kmalloc(sizeof(*upper), GFP_KERNEL);
4486 upper->dev = upper_dev;
4487 upper->master = master;
4488 INIT_LIST_HEAD(&upper->search_list);
4490 /* Ensure that master upper link is always the first item in list. */
4492 list_add_rcu(&upper->list, &dev->upper_dev_list);
4494 list_add_tail_rcu(&upper->list, &dev->upper_dev_list);
4495 dev_hold(upper_dev);
4496 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4501 * netdev_upper_dev_link - Add a link to the upper device
4503 * @upper_dev: new upper device
4505 * Adds a link to device which is upper to this one. The caller must hold
4506 * the RTNL lock. On a failure a negative errno code is returned.
4507 * On success the reference counts are adjusted and the function
4510 int netdev_upper_dev_link(struct net_device *dev,
4511 struct net_device *upper_dev)
4513 return __netdev_upper_dev_link(dev, upper_dev, false);
4515 EXPORT_SYMBOL(netdev_upper_dev_link);
4518 * netdev_master_upper_dev_link - Add a master link to the upper device
4520 * @upper_dev: new upper device
4522 * Adds a link to device which is upper to this one. In this case, only
4523 * one master upper device can be linked, although other non-master devices
4524 * might be linked as well. The caller must hold the RTNL lock.
4525 * On a failure a negative errno code is returned. On success the reference
4526 * counts are adjusted and the function returns zero.
4528 int netdev_master_upper_dev_link(struct net_device *dev,
4529 struct net_device *upper_dev)
4531 return __netdev_upper_dev_link(dev, upper_dev, true);
4533 EXPORT_SYMBOL(netdev_master_upper_dev_link);
4536 * netdev_upper_dev_unlink - Removes a link to upper device
4538 * @upper_dev: new upper device
4540 * Removes a link to device which is upper to this one. The caller must hold
4543 void netdev_upper_dev_unlink(struct net_device *dev,
4544 struct net_device *upper_dev)
4546 struct netdev_upper *upper;
4550 upper = __netdev_find_upper(dev, upper_dev);
4553 list_del_rcu(&upper->list);
4555 kfree_rcu(upper, rcu);
4556 call_netdevice_notifiers(NETDEV_CHANGEUPPER, dev);
4558 EXPORT_SYMBOL(netdev_upper_dev_unlink);
4560 static void dev_change_rx_flags(struct net_device *dev, int flags)
4562 const struct net_device_ops *ops = dev->netdev_ops;
4564 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4565 ops->ndo_change_rx_flags(dev, flags);
4568 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4570 unsigned int old_flags = dev->flags;
4576 dev->flags |= IFF_PROMISC;
4577 dev->promiscuity += inc;
4578 if (dev->promiscuity == 0) {
4581 * If inc causes overflow, untouch promisc and return error.
4584 dev->flags &= ~IFF_PROMISC;
4586 dev->promiscuity -= inc;
4587 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4592 if (dev->flags != old_flags) {
4593 pr_info("device %s %s promiscuous mode\n",
4595 dev->flags & IFF_PROMISC ? "entered" : "left");
4596 if (audit_enabled) {
4597 current_uid_gid(&uid, &gid);
4598 audit_log(current->audit_context, GFP_ATOMIC,
4599 AUDIT_ANOM_PROMISCUOUS,
4600 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4601 dev->name, (dev->flags & IFF_PROMISC),
4602 (old_flags & IFF_PROMISC),
4603 from_kuid(&init_user_ns, audit_get_loginuid(current)),
4604 from_kuid(&init_user_ns, uid),
4605 from_kgid(&init_user_ns, gid),
4606 audit_get_sessionid(current));
4609 dev_change_rx_flags(dev, IFF_PROMISC);
4615 * dev_set_promiscuity - update promiscuity count on a device
4619 * Add or remove promiscuity from a device. While the count in the device
4620 * remains above zero the interface remains promiscuous. Once it hits zero
4621 * the device reverts back to normal filtering operation. A negative inc
4622 * value is used to drop promiscuity on the device.
4623 * Return 0 if successful or a negative errno code on error.
4625 int dev_set_promiscuity(struct net_device *dev, int inc)
4627 unsigned int old_flags = dev->flags;
4630 err = __dev_set_promiscuity(dev, inc);
4633 if (dev->flags != old_flags)
4634 dev_set_rx_mode(dev);
4637 EXPORT_SYMBOL(dev_set_promiscuity);
4640 * dev_set_allmulti - update allmulti count on a device
4644 * Add or remove reception of all multicast frames to a device. While the
4645 * count in the device remains above zero the interface remains listening
4646 * to all interfaces. Once it hits zero the device reverts back to normal
4647 * filtering operation. A negative @inc value is used to drop the counter
4648 * when releasing a resource needing all multicasts.
4649 * Return 0 if successful or a negative errno code on error.
4652 int dev_set_allmulti(struct net_device *dev, int inc)
4654 unsigned int old_flags = dev->flags;
4658 dev->flags |= IFF_ALLMULTI;
4659 dev->allmulti += inc;
4660 if (dev->allmulti == 0) {
4663 * If inc causes overflow, untouch allmulti and return error.
4666 dev->flags &= ~IFF_ALLMULTI;
4668 dev->allmulti -= inc;
4669 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4674 if (dev->flags ^ old_flags) {
4675 dev_change_rx_flags(dev, IFF_ALLMULTI);
4676 dev_set_rx_mode(dev);
4680 EXPORT_SYMBOL(dev_set_allmulti);
4683 * Upload unicast and multicast address lists to device and
4684 * configure RX filtering. When the device doesn't support unicast
4685 * filtering it is put in promiscuous mode while unicast addresses
4688 void __dev_set_rx_mode(struct net_device *dev)
4690 const struct net_device_ops *ops = dev->netdev_ops;
4692 /* dev_open will call this function so the list will stay sane. */
4693 if (!(dev->flags&IFF_UP))
4696 if (!netif_device_present(dev))
4699 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4700 /* Unicast addresses changes may only happen under the rtnl,
4701 * therefore calling __dev_set_promiscuity here is safe.
4703 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4704 __dev_set_promiscuity(dev, 1);
4705 dev->uc_promisc = true;
4706 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4707 __dev_set_promiscuity(dev, -1);
4708 dev->uc_promisc = false;
4712 if (ops->ndo_set_rx_mode)
4713 ops->ndo_set_rx_mode(dev);
4716 void dev_set_rx_mode(struct net_device *dev)
4718 netif_addr_lock_bh(dev);
4719 __dev_set_rx_mode(dev);
4720 netif_addr_unlock_bh(dev);
4724 * dev_get_flags - get flags reported to userspace
4727 * Get the combination of flag bits exported through APIs to userspace.
4729 unsigned int dev_get_flags(const struct net_device *dev)
4733 flags = (dev->flags & ~(IFF_PROMISC |
4738 (dev->gflags & (IFF_PROMISC |
4741 if (netif_running(dev)) {
4742 if (netif_oper_up(dev))
4743 flags |= IFF_RUNNING;
4744 if (netif_carrier_ok(dev))
4745 flags |= IFF_LOWER_UP;
4746 if (netif_dormant(dev))
4747 flags |= IFF_DORMANT;
4752 EXPORT_SYMBOL(dev_get_flags);
4754 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4756 unsigned int old_flags = dev->flags;
4762 * Set the flags on our device.
4765 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4766 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4768 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4772 * Load in the correct multicast list now the flags have changed.
4775 if ((old_flags ^ flags) & IFF_MULTICAST)
4776 dev_change_rx_flags(dev, IFF_MULTICAST);
4778 dev_set_rx_mode(dev);
4781 * Have we downed the interface. We handle IFF_UP ourselves
4782 * according to user attempts to set it, rather than blindly
4787 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4788 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4791 dev_set_rx_mode(dev);
4794 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4795 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4797 dev->gflags ^= IFF_PROMISC;
4798 dev_set_promiscuity(dev, inc);
4801 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4802 is important. Some (broken) drivers set IFF_PROMISC, when
4803 IFF_ALLMULTI is requested not asking us and not reporting.
4805 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4806 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4808 dev->gflags ^= IFF_ALLMULTI;
4809 dev_set_allmulti(dev, inc);
4815 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4817 unsigned int changes = dev->flags ^ old_flags;
4819 if (changes & IFF_UP) {
4820 if (dev->flags & IFF_UP)
4821 call_netdevice_notifiers(NETDEV_UP, dev);
4823 call_netdevice_notifiers(NETDEV_DOWN, dev);
4826 if (dev->flags & IFF_UP &&
4827 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
4828 struct netdev_notifier_change_info change_info;
4830 change_info.flags_changed = changes;
4831 call_netdevice_notifiers_info(NETDEV_CHANGE, dev,
4837 * dev_change_flags - change device settings
4839 * @flags: device state flags
4841 * Change settings on device based state flags. The flags are
4842 * in the userspace exported format.
4844 int dev_change_flags(struct net_device *dev, unsigned int flags)
4847 unsigned int changes, old_flags = dev->flags;
4849 ret = __dev_change_flags(dev, flags);
4853 changes = old_flags ^ dev->flags;
4855 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4857 __dev_notify_flags(dev, old_flags);
4860 EXPORT_SYMBOL(dev_change_flags);
4863 * dev_set_mtu - Change maximum transfer unit
4865 * @new_mtu: new transfer unit
4867 * Change the maximum transfer size of the network device.
4869 int dev_set_mtu(struct net_device *dev, int new_mtu)
4871 const struct net_device_ops *ops = dev->netdev_ops;
4874 if (new_mtu == dev->mtu)
4877 /* MTU must be positive. */
4881 if (!netif_device_present(dev))
4885 if (ops->ndo_change_mtu)
4886 err = ops->ndo_change_mtu(dev, new_mtu);
4891 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4894 EXPORT_SYMBOL(dev_set_mtu);
4897 * dev_set_group - Change group this device belongs to
4899 * @new_group: group this device should belong to
4901 void dev_set_group(struct net_device *dev, int new_group)
4903 dev->group = new_group;
4905 EXPORT_SYMBOL(dev_set_group);
4908 * dev_set_mac_address - Change Media Access Control Address
4912 * Change the hardware (MAC) address of the device
4914 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4916 const struct net_device_ops *ops = dev->netdev_ops;
4919 if (!ops->ndo_set_mac_address)
4921 if (sa->sa_family != dev->type)
4923 if (!netif_device_present(dev))
4925 err = ops->ndo_set_mac_address(dev, sa);
4928 dev->addr_assign_type = NET_ADDR_SET;
4929 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4930 add_device_randomness(dev->dev_addr, dev->addr_len);
4933 EXPORT_SYMBOL(dev_set_mac_address);
4936 * dev_change_carrier - Change device carrier
4938 * @new_carrier: new value
4940 * Change device carrier
4942 int dev_change_carrier(struct net_device *dev, bool new_carrier)
4944 const struct net_device_ops *ops = dev->netdev_ops;
4946 if (!ops->ndo_change_carrier)
4948 if (!netif_device_present(dev))
4950 return ops->ndo_change_carrier(dev, new_carrier);
4952 EXPORT_SYMBOL(dev_change_carrier);
4955 * dev_new_index - allocate an ifindex
4956 * @net: the applicable net namespace
4958 * Returns a suitable unique value for a new device interface
4959 * number. The caller must hold the rtnl semaphore or the
4960 * dev_base_lock to be sure it remains unique.
4962 static int dev_new_index(struct net *net)
4964 int ifindex = net->ifindex;
4968 if (!__dev_get_by_index(net, ifindex))
4969 return net->ifindex = ifindex;
4973 /* Delayed registration/unregisteration */
4974 static LIST_HEAD(net_todo_list);
4976 static void net_set_todo(struct net_device *dev)
4978 list_add_tail(&dev->todo_list, &net_todo_list);
4981 static void rollback_registered_many(struct list_head *head)
4983 struct net_device *dev, *tmp;
4985 BUG_ON(dev_boot_phase);
4988 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4989 /* Some devices call without registering
4990 * for initialization unwind. Remove those
4991 * devices and proceed with the remaining.
4993 if (dev->reg_state == NETREG_UNINITIALIZED) {
4994 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
4998 list_del(&dev->unreg_list);
5001 dev->dismantle = true;
5002 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5005 /* If device is running, close it first. */
5006 dev_close_many(head);
5008 list_for_each_entry(dev, head, unreg_list) {
5009 /* And unlink it from device chain. */
5010 unlist_netdevice(dev);
5012 dev->reg_state = NETREG_UNREGISTERING;
5017 list_for_each_entry(dev, head, unreg_list) {
5018 /* Shutdown queueing discipline. */
5022 /* Notify protocols, that we are about to destroy
5023 this device. They should clean all the things.
5025 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5027 if (!dev->rtnl_link_ops ||
5028 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5029 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5032 * Flush the unicast and multicast chains
5037 if (dev->netdev_ops->ndo_uninit)
5038 dev->netdev_ops->ndo_uninit(dev);
5040 /* Notifier chain MUST detach us all upper devices. */
5041 WARN_ON(netdev_has_any_upper_dev(dev));
5043 /* Remove entries from kobject tree */
5044 netdev_unregister_kobject(dev);
5046 /* Remove XPS queueing entries */
5047 netif_reset_xps_queues_gt(dev, 0);
5053 list_for_each_entry(dev, head, unreg_list)
5057 static void rollback_registered(struct net_device *dev)
5061 list_add(&dev->unreg_list, &single);
5062 rollback_registered_many(&single);
5066 static netdev_features_t netdev_fix_features(struct net_device *dev,
5067 netdev_features_t features)
5069 /* Fix illegal checksum combinations */
5070 if ((features & NETIF_F_HW_CSUM) &&
5071 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5072 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5073 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5076 /* TSO requires that SG is present as well. */
5077 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5078 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5079 features &= ~NETIF_F_ALL_TSO;
5082 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
5083 !(features & NETIF_F_IP_CSUM)) {
5084 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
5085 features &= ~NETIF_F_TSO;
5086 features &= ~NETIF_F_TSO_ECN;
5089 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
5090 !(features & NETIF_F_IPV6_CSUM)) {
5091 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
5092 features &= ~NETIF_F_TSO6;
5095 /* TSO ECN requires that TSO is present as well. */
5096 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5097 features &= ~NETIF_F_TSO_ECN;
5099 /* Software GSO depends on SG. */
5100 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5101 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5102 features &= ~NETIF_F_GSO;
5105 /* UFO needs SG and checksumming */
5106 if (features & NETIF_F_UFO) {
5107 /* maybe split UFO into V4 and V6? */
5108 if (!((features & NETIF_F_GEN_CSUM) ||
5109 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5110 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5112 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5113 features &= ~NETIF_F_UFO;
5116 if (!(features & NETIF_F_SG)) {
5118 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5119 features &= ~NETIF_F_UFO;
5126 int __netdev_update_features(struct net_device *dev)
5128 netdev_features_t features;
5133 features = netdev_get_wanted_features(dev);
5135 if (dev->netdev_ops->ndo_fix_features)
5136 features = dev->netdev_ops->ndo_fix_features(dev, features);
5138 /* driver might be less strict about feature dependencies */
5139 features = netdev_fix_features(dev, features);
5141 if (dev->features == features)
5144 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5145 &dev->features, &features);
5147 if (dev->netdev_ops->ndo_set_features)
5148 err = dev->netdev_ops->ndo_set_features(dev, features);
5150 if (unlikely(err < 0)) {
5152 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5153 err, &features, &dev->features);
5158 dev->features = features;
5164 * netdev_update_features - recalculate device features
5165 * @dev: the device to check
5167 * Recalculate dev->features set and send notifications if it
5168 * has changed. Should be called after driver or hardware dependent
5169 * conditions might have changed that influence the features.
5171 void netdev_update_features(struct net_device *dev)
5173 if (__netdev_update_features(dev))
5174 netdev_features_change(dev);
5176 EXPORT_SYMBOL(netdev_update_features);
5179 * netdev_change_features - recalculate device features
5180 * @dev: the device to check
5182 * Recalculate dev->features set and send notifications even
5183 * if they have not changed. Should be called instead of
5184 * netdev_update_features() if also dev->vlan_features might
5185 * have changed to allow the changes to be propagated to stacked
5188 void netdev_change_features(struct net_device *dev)
5190 __netdev_update_features(dev);
5191 netdev_features_change(dev);
5193 EXPORT_SYMBOL(netdev_change_features);
5196 * netif_stacked_transfer_operstate - transfer operstate
5197 * @rootdev: the root or lower level device to transfer state from
5198 * @dev: the device to transfer operstate to
5200 * Transfer operational state from root to device. This is normally
5201 * called when a stacking relationship exists between the root
5202 * device and the device(a leaf device).
5204 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5205 struct net_device *dev)
5207 if (rootdev->operstate == IF_OPER_DORMANT)
5208 netif_dormant_on(dev);
5210 netif_dormant_off(dev);
5212 if (netif_carrier_ok(rootdev)) {
5213 if (!netif_carrier_ok(dev))
5214 netif_carrier_on(dev);
5216 if (netif_carrier_ok(dev))
5217 netif_carrier_off(dev);
5220 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5223 static int netif_alloc_rx_queues(struct net_device *dev)
5225 unsigned int i, count = dev->num_rx_queues;
5226 struct netdev_rx_queue *rx;
5230 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5236 for (i = 0; i < count; i++)
5242 static void netdev_init_one_queue(struct net_device *dev,
5243 struct netdev_queue *queue, void *_unused)
5245 /* Initialize queue lock */
5246 spin_lock_init(&queue->_xmit_lock);
5247 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5248 queue->xmit_lock_owner = -1;
5249 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5252 dql_init(&queue->dql, HZ);
5256 static int netif_alloc_netdev_queues(struct net_device *dev)
5258 unsigned int count = dev->num_tx_queues;
5259 struct netdev_queue *tx;
5263 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5269 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5270 spin_lock_init(&dev->tx_global_lock);
5276 * register_netdevice - register a network device
5277 * @dev: device to register
5279 * Take a completed network device structure and add it to the kernel
5280 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5281 * chain. 0 is returned on success. A negative errno code is returned
5282 * on a failure to set up the device, or if the name is a duplicate.
5284 * Callers must hold the rtnl semaphore. You may want
5285 * register_netdev() instead of this.
5288 * The locking appears insufficient to guarantee two parallel registers
5289 * will not get the same name.
5292 int register_netdevice(struct net_device *dev)
5295 struct net *net = dev_net(dev);
5297 BUG_ON(dev_boot_phase);
5302 /* When net_device's are persistent, this will be fatal. */
5303 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5306 spin_lock_init(&dev->addr_list_lock);
5307 netdev_set_addr_lockdep_class(dev);
5311 ret = dev_get_valid_name(net, dev, dev->name);
5315 /* Init, if this function is available */
5316 if (dev->netdev_ops->ndo_init) {
5317 ret = dev->netdev_ops->ndo_init(dev);
5325 if (((dev->hw_features | dev->features) &
5326 NETIF_F_HW_VLAN_CTAG_FILTER) &&
5327 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
5328 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
5329 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
5336 dev->ifindex = dev_new_index(net);
5337 else if (__dev_get_by_index(net, dev->ifindex))
5340 if (dev->iflink == -1)
5341 dev->iflink = dev->ifindex;
5343 /* Transfer changeable features to wanted_features and enable
5344 * software offloads (GSO and GRO).
5346 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5347 dev->features |= NETIF_F_SOFT_FEATURES;
5348 dev->wanted_features = dev->features & dev->hw_features;
5350 /* Turn on no cache copy if HW is doing checksum */
5351 if (!(dev->flags & IFF_LOOPBACK)) {
5352 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5353 if (dev->features & NETIF_F_ALL_CSUM) {
5354 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5355 dev->features |= NETIF_F_NOCACHE_COPY;
5359 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5361 dev->vlan_features |= NETIF_F_HIGHDMA;
5363 /* Make NETIF_F_SG inheritable to tunnel devices.
5365 dev->hw_enc_features |= NETIF_F_SG;
5367 /* Make NETIF_F_SG inheritable to MPLS.
5369 dev->mpls_features |= NETIF_F_SG;
5371 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5372 ret = notifier_to_errno(ret);
5376 ret = netdev_register_kobject(dev);
5379 dev->reg_state = NETREG_REGISTERED;
5381 __netdev_update_features(dev);
5384 * Default initial state at registry is that the
5385 * device is present.
5388 set_bit(__LINK_STATE_PRESENT, &dev->state);
5390 linkwatch_init_dev(dev);
5392 dev_init_scheduler(dev);
5394 list_netdevice(dev);
5395 add_device_randomness(dev->dev_addr, dev->addr_len);
5397 /* If the device has permanent device address, driver should
5398 * set dev_addr and also addr_assign_type should be set to
5399 * NET_ADDR_PERM (default value).
5401 if (dev->addr_assign_type == NET_ADDR_PERM)
5402 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
5404 /* Notify protocols, that a new device appeared. */
5405 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5406 ret = notifier_to_errno(ret);
5408 rollback_registered(dev);
5409 dev->reg_state = NETREG_UNREGISTERED;
5412 * Prevent userspace races by waiting until the network
5413 * device is fully setup before sending notifications.
5415 if (!dev->rtnl_link_ops ||
5416 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5417 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5423 if (dev->netdev_ops->ndo_uninit)
5424 dev->netdev_ops->ndo_uninit(dev);
5427 EXPORT_SYMBOL(register_netdevice);
5430 * init_dummy_netdev - init a dummy network device for NAPI
5431 * @dev: device to init
5433 * This takes a network device structure and initialize the minimum
5434 * amount of fields so it can be used to schedule NAPI polls without
5435 * registering a full blown interface. This is to be used by drivers
5436 * that need to tie several hardware interfaces to a single NAPI
5437 * poll scheduler due to HW limitations.
5439 int init_dummy_netdev(struct net_device *dev)
5441 /* Clear everything. Note we don't initialize spinlocks
5442 * are they aren't supposed to be taken by any of the
5443 * NAPI code and this dummy netdev is supposed to be
5444 * only ever used for NAPI polls
5446 memset(dev, 0, sizeof(struct net_device));
5448 /* make sure we BUG if trying to hit standard
5449 * register/unregister code path
5451 dev->reg_state = NETREG_DUMMY;
5453 /* NAPI wants this */
5454 INIT_LIST_HEAD(&dev->napi_list);
5456 /* a dummy interface is started by default */
5457 set_bit(__LINK_STATE_PRESENT, &dev->state);
5458 set_bit(__LINK_STATE_START, &dev->state);
5460 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5461 * because users of this 'device' dont need to change
5467 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5471 * register_netdev - register a network device
5472 * @dev: device to register
5474 * Take a completed network device structure and add it to the kernel
5475 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5476 * chain. 0 is returned on success. A negative errno code is returned
5477 * on a failure to set up the device, or if the name is a duplicate.
5479 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5480 * and expands the device name if you passed a format string to
5483 int register_netdev(struct net_device *dev)
5488 err = register_netdevice(dev);
5492 EXPORT_SYMBOL(register_netdev);
5494 int netdev_refcnt_read(const struct net_device *dev)
5498 for_each_possible_cpu(i)
5499 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5502 EXPORT_SYMBOL(netdev_refcnt_read);
5505 * netdev_wait_allrefs - wait until all references are gone.
5506 * @dev: target net_device
5508 * This is called when unregistering network devices.
5510 * Any protocol or device that holds a reference should register
5511 * for netdevice notification, and cleanup and put back the
5512 * reference if they receive an UNREGISTER event.
5513 * We can get stuck here if buggy protocols don't correctly
5516 static void netdev_wait_allrefs(struct net_device *dev)
5518 unsigned long rebroadcast_time, warning_time;
5521 linkwatch_forget_dev(dev);
5523 rebroadcast_time = warning_time = jiffies;
5524 refcnt = netdev_refcnt_read(dev);
5526 while (refcnt != 0) {
5527 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5530 /* Rebroadcast unregister notification */
5531 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5537 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5538 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5540 /* We must not have linkwatch events
5541 * pending on unregister. If this
5542 * happens, we simply run the queue
5543 * unscheduled, resulting in a noop
5546 linkwatch_run_queue();
5551 rebroadcast_time = jiffies;
5556 refcnt = netdev_refcnt_read(dev);
5558 if (time_after(jiffies, warning_time + 10 * HZ)) {
5559 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5561 warning_time = jiffies;
5570 * register_netdevice(x1);
5571 * register_netdevice(x2);
5573 * unregister_netdevice(y1);
5574 * unregister_netdevice(y2);
5580 * We are invoked by rtnl_unlock().
5581 * This allows us to deal with problems:
5582 * 1) We can delete sysfs objects which invoke hotplug
5583 * without deadlocking with linkwatch via keventd.
5584 * 2) Since we run with the RTNL semaphore not held, we can sleep
5585 * safely in order to wait for the netdev refcnt to drop to zero.
5587 * We must not return until all unregister events added during
5588 * the interval the lock was held have been completed.
5590 void netdev_run_todo(void)
5592 struct list_head list;
5594 /* Snapshot list, allow later requests */
5595 list_replace_init(&net_todo_list, &list);
5600 /* Wait for rcu callbacks to finish before next phase */
5601 if (!list_empty(&list))
5604 while (!list_empty(&list)) {
5605 struct net_device *dev
5606 = list_first_entry(&list, struct net_device, todo_list);
5607 list_del(&dev->todo_list);
5610 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
5613 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5614 pr_err("network todo '%s' but state %d\n",
5615 dev->name, dev->reg_state);
5620 dev->reg_state = NETREG_UNREGISTERED;
5622 on_each_cpu(flush_backlog, dev, 1);
5624 netdev_wait_allrefs(dev);
5627 BUG_ON(netdev_refcnt_read(dev));
5628 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5629 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5630 WARN_ON(dev->dn_ptr);
5632 if (dev->destructor)
5633 dev->destructor(dev);
5635 /* Free network device */
5636 kobject_put(&dev->dev.kobj);
5640 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5641 * fields in the same order, with only the type differing.
5643 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5644 const struct net_device_stats *netdev_stats)
5646 #if BITS_PER_LONG == 64
5647 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5648 memcpy(stats64, netdev_stats, sizeof(*stats64));
5650 size_t i, n = sizeof(*stats64) / sizeof(u64);
5651 const unsigned long *src = (const unsigned long *)netdev_stats;
5652 u64 *dst = (u64 *)stats64;
5654 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5655 sizeof(*stats64) / sizeof(u64));
5656 for (i = 0; i < n; i++)
5660 EXPORT_SYMBOL(netdev_stats_to_stats64);
5663 * dev_get_stats - get network device statistics
5664 * @dev: device to get statistics from
5665 * @storage: place to store stats
5667 * Get network statistics from device. Return @storage.
5668 * The device driver may provide its own method by setting
5669 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5670 * otherwise the internal statistics structure is used.
5672 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5673 struct rtnl_link_stats64 *storage)
5675 const struct net_device_ops *ops = dev->netdev_ops;
5677 if (ops->ndo_get_stats64) {
5678 memset(storage, 0, sizeof(*storage));
5679 ops->ndo_get_stats64(dev, storage);
5680 } else if (ops->ndo_get_stats) {
5681 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5683 netdev_stats_to_stats64(storage, &dev->stats);
5685 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5688 EXPORT_SYMBOL(dev_get_stats);
5690 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5692 struct netdev_queue *queue = dev_ingress_queue(dev);
5694 #ifdef CONFIG_NET_CLS_ACT
5697 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5700 netdev_init_one_queue(dev, queue, NULL);
5701 queue->qdisc = &noop_qdisc;
5702 queue->qdisc_sleeping = &noop_qdisc;
5703 rcu_assign_pointer(dev->ingress_queue, queue);
5708 static const struct ethtool_ops default_ethtool_ops;
5710 void netdev_set_default_ethtool_ops(struct net_device *dev,
5711 const struct ethtool_ops *ops)
5713 if (dev->ethtool_ops == &default_ethtool_ops)
5714 dev->ethtool_ops = ops;
5716 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
5719 * alloc_netdev_mqs - allocate network device
5720 * @sizeof_priv: size of private data to allocate space for
5721 * @name: device name format string
5722 * @setup: callback to initialize device
5723 * @txqs: the number of TX subqueues to allocate
5724 * @rxqs: the number of RX subqueues to allocate
5726 * Allocates a struct net_device with private data area for driver use
5727 * and performs basic initialization. Also allocates subquue structs
5728 * for each queue on the device.
5730 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5731 void (*setup)(struct net_device *),
5732 unsigned int txqs, unsigned int rxqs)
5734 struct net_device *dev;
5736 struct net_device *p;
5738 BUG_ON(strlen(name) >= sizeof(dev->name));
5741 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5747 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5752 alloc_size = sizeof(struct net_device);
5754 /* ensure 32-byte alignment of private area */
5755 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5756 alloc_size += sizeof_priv;
5758 /* ensure 32-byte alignment of whole construct */
5759 alloc_size += NETDEV_ALIGN - 1;
5761 p = kzalloc(alloc_size, GFP_KERNEL);
5765 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5766 dev->padded = (char *)dev - (char *)p;
5768 dev->pcpu_refcnt = alloc_percpu(int);
5769 if (!dev->pcpu_refcnt)
5772 if (dev_addr_init(dev))
5778 dev_net_set(dev, &init_net);
5780 dev->gso_max_size = GSO_MAX_SIZE;
5781 dev->gso_max_segs = GSO_MAX_SEGS;
5783 INIT_LIST_HEAD(&dev->napi_list);
5784 INIT_LIST_HEAD(&dev->unreg_list);
5785 INIT_LIST_HEAD(&dev->link_watch_list);
5786 INIT_LIST_HEAD(&dev->upper_dev_list);
5787 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5790 dev->num_tx_queues = txqs;
5791 dev->real_num_tx_queues = txqs;
5792 if (netif_alloc_netdev_queues(dev))
5796 dev->num_rx_queues = rxqs;
5797 dev->real_num_rx_queues = rxqs;
5798 if (netif_alloc_rx_queues(dev))
5802 strcpy(dev->name, name);
5803 dev->group = INIT_NETDEV_GROUP;
5804 if (!dev->ethtool_ops)
5805 dev->ethtool_ops = &default_ethtool_ops;
5813 free_percpu(dev->pcpu_refcnt);
5823 EXPORT_SYMBOL(alloc_netdev_mqs);
5826 * free_netdev - free network device
5829 * This function does the last stage of destroying an allocated device
5830 * interface. The reference to the device object is released.
5831 * If this is the last reference then it will be freed.
5833 void free_netdev(struct net_device *dev)
5835 struct napi_struct *p, *n;
5837 release_net(dev_net(dev));
5844 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
5846 /* Flush device addresses */
5847 dev_addr_flush(dev);
5849 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5852 free_percpu(dev->pcpu_refcnt);
5853 dev->pcpu_refcnt = NULL;
5855 /* Compatibility with error handling in drivers */
5856 if (dev->reg_state == NETREG_UNINITIALIZED) {
5857 kfree((char *)dev - dev->padded);
5861 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5862 dev->reg_state = NETREG_RELEASED;
5864 /* will free via device release */
5865 put_device(&dev->dev);
5867 EXPORT_SYMBOL(free_netdev);
5870 * synchronize_net - Synchronize with packet receive processing
5872 * Wait for packets currently being received to be done.
5873 * Does not block later packets from starting.
5875 void synchronize_net(void)
5878 if (rtnl_is_locked())
5879 synchronize_rcu_expedited();
5883 EXPORT_SYMBOL(synchronize_net);
5886 * unregister_netdevice_queue - remove device from the kernel
5890 * This function shuts down a device interface and removes it
5891 * from the kernel tables.
5892 * If head not NULL, device is queued to be unregistered later.
5894 * Callers must hold the rtnl semaphore. You may want
5895 * unregister_netdev() instead of this.
5898 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5903 list_move_tail(&dev->unreg_list, head);
5905 rollback_registered(dev);
5906 /* Finish processing unregister after unlock */
5910 EXPORT_SYMBOL(unregister_netdevice_queue);
5913 * unregister_netdevice_many - unregister many devices
5914 * @head: list of devices
5916 void unregister_netdevice_many(struct list_head *head)
5918 struct net_device *dev;
5920 if (!list_empty(head)) {
5921 rollback_registered_many(head);
5922 list_for_each_entry(dev, head, unreg_list)
5926 EXPORT_SYMBOL(unregister_netdevice_many);
5929 * unregister_netdev - remove device from the kernel
5932 * This function shuts down a device interface and removes it
5933 * from the kernel tables.
5935 * This is just a wrapper for unregister_netdevice that takes
5936 * the rtnl semaphore. In general you want to use this and not
5937 * unregister_netdevice.
5939 void unregister_netdev(struct net_device *dev)
5942 unregister_netdevice(dev);
5945 EXPORT_SYMBOL(unregister_netdev);
5948 * dev_change_net_namespace - move device to different nethost namespace
5950 * @net: network namespace
5951 * @pat: If not NULL name pattern to try if the current device name
5952 * is already taken in the destination network namespace.
5954 * This function shuts down a device interface and moves it
5955 * to a new network namespace. On success 0 is returned, on
5956 * a failure a netagive errno code is returned.
5958 * Callers must hold the rtnl semaphore.
5961 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5967 /* Don't allow namespace local devices to be moved. */
5969 if (dev->features & NETIF_F_NETNS_LOCAL)
5972 /* Ensure the device has been registrered */
5973 if (dev->reg_state != NETREG_REGISTERED)
5976 /* Get out if there is nothing todo */
5978 if (net_eq(dev_net(dev), net))
5981 /* Pick the destination device name, and ensure
5982 * we can use it in the destination network namespace.
5985 if (__dev_get_by_name(net, dev->name)) {
5986 /* We get here if we can't use the current device name */
5989 if (dev_get_valid_name(net, dev, pat) < 0)
5994 * And now a mini version of register_netdevice unregister_netdevice.
5997 /* If device is running close it first. */
6000 /* And unlink it from device chain */
6002 unlist_netdevice(dev);
6006 /* Shutdown queueing discipline. */
6009 /* Notify protocols, that we are about to destroy
6010 this device. They should clean all the things.
6012 Note that dev->reg_state stays at NETREG_REGISTERED.
6013 This is wanted because this way 8021q and macvlan know
6014 the device is just moving and can keep their slaves up.
6016 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6018 call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev);
6019 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6022 * Flush the unicast and multicast chains
6027 /* Send a netdev-removed uevent to the old namespace */
6028 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
6030 /* Actually switch the network namespace */
6031 dev_net_set(dev, net);
6033 /* If there is an ifindex conflict assign a new one */
6034 if (__dev_get_by_index(net, dev->ifindex)) {
6035 int iflink = (dev->iflink == dev->ifindex);
6036 dev->ifindex = dev_new_index(net);
6038 dev->iflink = dev->ifindex;
6041 /* Send a netdev-add uevent to the new namespace */
6042 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
6044 /* Fixup kobjects */
6045 err = device_rename(&dev->dev, dev->name);
6048 /* Add the device back in the hashes */
6049 list_netdevice(dev);
6051 /* Notify protocols, that a new device appeared. */
6052 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6055 * Prevent userspace races by waiting until the network
6056 * device is fully setup before sending notifications.
6058 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6065 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6067 static int dev_cpu_callback(struct notifier_block *nfb,
6068 unsigned long action,
6071 struct sk_buff **list_skb;
6072 struct sk_buff *skb;
6073 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6074 struct softnet_data *sd, *oldsd;
6076 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6079 local_irq_disable();
6080 cpu = smp_processor_id();
6081 sd = &per_cpu(softnet_data, cpu);
6082 oldsd = &per_cpu(softnet_data, oldcpu);
6084 /* Find end of our completion_queue. */
6085 list_skb = &sd->completion_queue;
6087 list_skb = &(*list_skb)->next;
6088 /* Append completion queue from offline CPU. */
6089 *list_skb = oldsd->completion_queue;
6090 oldsd->completion_queue = NULL;
6092 /* Append output queue from offline CPU. */
6093 if (oldsd->output_queue) {
6094 *sd->output_queue_tailp = oldsd->output_queue;
6095 sd->output_queue_tailp = oldsd->output_queue_tailp;
6096 oldsd->output_queue = NULL;
6097 oldsd->output_queue_tailp = &oldsd->output_queue;
6099 /* Append NAPI poll list from offline CPU. */
6100 if (!list_empty(&oldsd->poll_list)) {
6101 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6102 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6105 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6108 /* Process offline CPU's input_pkt_queue */
6109 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6111 input_queue_head_incr(oldsd);
6113 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6115 input_queue_head_incr(oldsd);
6123 * netdev_increment_features - increment feature set by one
6124 * @all: current feature set
6125 * @one: new feature set
6126 * @mask: mask feature set
6128 * Computes a new feature set after adding a device with feature set
6129 * @one to the master device with current feature set @all. Will not
6130 * enable anything that is off in @mask. Returns the new feature set.
6132 netdev_features_t netdev_increment_features(netdev_features_t all,
6133 netdev_features_t one, netdev_features_t mask)
6135 if (mask & NETIF_F_GEN_CSUM)
6136 mask |= NETIF_F_ALL_CSUM;
6137 mask |= NETIF_F_VLAN_CHALLENGED;
6139 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6140 all &= one | ~NETIF_F_ALL_FOR_ALL;
6142 /* If one device supports hw checksumming, set for all. */
6143 if (all & NETIF_F_GEN_CSUM)
6144 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6148 EXPORT_SYMBOL(netdev_increment_features);
6150 static struct hlist_head * __net_init netdev_create_hash(void)
6153 struct hlist_head *hash;
6155 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6157 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6158 INIT_HLIST_HEAD(&hash[i]);
6163 /* Initialize per network namespace state */
6164 static int __net_init netdev_init(struct net *net)
6166 if (net != &init_net)
6167 INIT_LIST_HEAD(&net->dev_base_head);
6169 net->dev_name_head = netdev_create_hash();
6170 if (net->dev_name_head == NULL)
6173 net->dev_index_head = netdev_create_hash();
6174 if (net->dev_index_head == NULL)
6180 kfree(net->dev_name_head);
6186 * netdev_drivername - network driver for the device
6187 * @dev: network device
6189 * Determine network driver for device.
6191 const char *netdev_drivername(const struct net_device *dev)
6193 const struct device_driver *driver;
6194 const struct device *parent;
6195 const char *empty = "";
6197 parent = dev->dev.parent;
6201 driver = parent->driver;
6202 if (driver && driver->name)
6203 return driver->name;
6207 static int __netdev_printk(const char *level, const struct net_device *dev,
6208 struct va_format *vaf)
6212 if (dev && dev->dev.parent) {
6213 r = dev_printk_emit(level[1] - '0',
6216 dev_driver_string(dev->dev.parent),
6217 dev_name(dev->dev.parent),
6218 netdev_name(dev), vaf);
6220 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6222 r = printk("%s(NULL net_device): %pV", level, vaf);
6228 int netdev_printk(const char *level, const struct net_device *dev,
6229 const char *format, ...)
6231 struct va_format vaf;
6235 va_start(args, format);
6240 r = __netdev_printk(level, dev, &vaf);
6246 EXPORT_SYMBOL(netdev_printk);
6248 #define define_netdev_printk_level(func, level) \
6249 int func(const struct net_device *dev, const char *fmt, ...) \
6252 struct va_format vaf; \
6255 va_start(args, fmt); \
6260 r = __netdev_printk(level, dev, &vaf); \
6266 EXPORT_SYMBOL(func);
6268 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6269 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6270 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6271 define_netdev_printk_level(netdev_err, KERN_ERR);
6272 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6273 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6274 define_netdev_printk_level(netdev_info, KERN_INFO);
6276 static void __net_exit netdev_exit(struct net *net)
6278 kfree(net->dev_name_head);
6279 kfree(net->dev_index_head);
6282 static struct pernet_operations __net_initdata netdev_net_ops = {
6283 .init = netdev_init,
6284 .exit = netdev_exit,
6287 static void __net_exit default_device_exit(struct net *net)
6289 struct net_device *dev, *aux;
6291 * Push all migratable network devices back to the
6292 * initial network namespace
6295 for_each_netdev_safe(net, dev, aux) {
6297 char fb_name[IFNAMSIZ];
6299 /* Ignore unmoveable devices (i.e. loopback) */
6300 if (dev->features & NETIF_F_NETNS_LOCAL)
6303 /* Leave virtual devices for the generic cleanup */
6304 if (dev->rtnl_link_ops)
6307 /* Push remaining network devices to init_net */
6308 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6309 err = dev_change_net_namespace(dev, &init_net, fb_name);
6311 pr_emerg("%s: failed to move %s to init_net: %d\n",
6312 __func__, dev->name, err);
6319 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6321 /* At exit all network devices most be removed from a network
6322 * namespace. Do this in the reverse order of registration.
6323 * Do this across as many network namespaces as possible to
6324 * improve batching efficiency.
6326 struct net_device *dev;
6328 LIST_HEAD(dev_kill_list);
6331 list_for_each_entry(net, net_list, exit_list) {
6332 for_each_netdev_reverse(net, dev) {
6333 if (dev->rtnl_link_ops)
6334 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6336 unregister_netdevice_queue(dev, &dev_kill_list);
6339 unregister_netdevice_many(&dev_kill_list);
6340 list_del(&dev_kill_list);
6344 static struct pernet_operations __net_initdata default_device_ops = {
6345 .exit = default_device_exit,
6346 .exit_batch = default_device_exit_batch,
6350 * Initialize the DEV module. At boot time this walks the device list and
6351 * unhooks any devices that fail to initialise (normally hardware not
6352 * present) and leaves us with a valid list of present and active devices.
6357 * This is called single threaded during boot, so no need
6358 * to take the rtnl semaphore.
6360 static int __init net_dev_init(void)
6362 int i, rc = -ENOMEM;
6364 BUG_ON(!dev_boot_phase);
6366 if (dev_proc_init())
6369 if (netdev_kobject_init())
6372 INIT_LIST_HEAD(&ptype_all);
6373 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6374 INIT_LIST_HEAD(&ptype_base[i]);
6376 INIT_LIST_HEAD(&offload_base);
6378 if (register_pernet_subsys(&netdev_net_ops))
6382 * Initialise the packet receive queues.
6385 for_each_possible_cpu(i) {
6386 struct softnet_data *sd = &per_cpu(softnet_data, i);
6388 memset(sd, 0, sizeof(*sd));
6389 skb_queue_head_init(&sd->input_pkt_queue);
6390 skb_queue_head_init(&sd->process_queue);
6391 sd->completion_queue = NULL;
6392 INIT_LIST_HEAD(&sd->poll_list);
6393 sd->output_queue = NULL;
6394 sd->output_queue_tailp = &sd->output_queue;
6396 sd->csd.func = rps_trigger_softirq;
6402 sd->backlog.poll = process_backlog;
6403 sd->backlog.weight = weight_p;
6404 sd->backlog.gro_list = NULL;
6405 sd->backlog.gro_count = 0;
6407 #ifdef CONFIG_NET_FLOW_LIMIT
6408 sd->flow_limit = NULL;
6414 /* The loopback device is special if any other network devices
6415 * is present in a network namespace the loopback device must
6416 * be present. Since we now dynamically allocate and free the
6417 * loopback device ensure this invariant is maintained by
6418 * keeping the loopback device as the first device on the
6419 * list of network devices. Ensuring the loopback devices
6420 * is the first device that appears and the last network device
6423 if (register_pernet_device(&loopback_net_ops))
6426 if (register_pernet_device(&default_device_ops))
6429 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6430 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6432 hotcpu_notifier(dev_cpu_callback, 0);
6439 subsys_initcall(net_dev_init);