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/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <linux/stat.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/net_tstamp.h>
136 #include <linux/static_key.h>
137 #include <net/flow_keys.h>
139 #include "net-sysfs.h"
141 /* Instead of increasing this, you should create a hash table. */
142 #define MAX_GRO_SKBS 8
144 /* This should be increased if a protocol with a bigger head is added. */
145 #define GRO_MAX_HEAD (MAX_HEADER + 128)
148 * The list of packet types we will receive (as opposed to discard)
149 * and the routines to invoke.
151 * Why 16. Because with 16 the only overlap we get on a hash of the
152 * low nibble of the protocol value is RARP/SNAP/X.25.
154 * NOTE: That is no longer true with the addition of VLAN tags. Not
155 * sure which should go first, but I bet it won't make much
156 * difference if we are running VLANs. The good news is that
157 * this protocol won't be in the list unless compiled in, so
158 * the average user (w/out VLANs) will not be adversely affected.
175 #define PTYPE_HASH_SIZE (16)
176 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
178 static DEFINE_SPINLOCK(ptype_lock);
179 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
180 static struct list_head ptype_all __read_mostly; /* Taps */
183 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
186 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
188 * Writers must hold the rtnl semaphore while they loop through the
189 * dev_base_head list, and hold dev_base_lock for writing when they do the
190 * actual updates. This allows pure readers to access the list even
191 * while a writer is preparing to update it.
193 * To put it another way, dev_base_lock is held for writing only to
194 * protect against pure readers; the rtnl semaphore provides the
195 * protection against other writers.
197 * See, for example usages, register_netdevice() and
198 * unregister_netdevice(), which must be called with the rtnl
201 DEFINE_RWLOCK(dev_base_lock);
202 EXPORT_SYMBOL(dev_base_lock);
204 static inline void dev_base_seq_inc(struct net *net)
206 while (++net->dev_base_seq == 0);
209 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
211 unsigned int hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
213 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
216 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
218 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
221 static inline void rps_lock(struct softnet_data *sd)
224 spin_lock(&sd->input_pkt_queue.lock);
228 static inline void rps_unlock(struct softnet_data *sd)
231 spin_unlock(&sd->input_pkt_queue.lock);
235 /* Device list insertion */
236 static int list_netdevice(struct net_device *dev)
238 struct net *net = dev_net(dev);
242 write_lock_bh(&dev_base_lock);
243 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
244 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
245 hlist_add_head_rcu(&dev->index_hlist,
246 dev_index_hash(net, dev->ifindex));
247 write_unlock_bh(&dev_base_lock);
249 dev_base_seq_inc(net);
254 /* Device list removal
255 * caller must respect a RCU grace period before freeing/reusing dev
257 static void unlist_netdevice(struct net_device *dev)
261 /* Unlink dev from the device chain */
262 write_lock_bh(&dev_base_lock);
263 list_del_rcu(&dev->dev_list);
264 hlist_del_rcu(&dev->name_hlist);
265 hlist_del_rcu(&dev->index_hlist);
266 write_unlock_bh(&dev_base_lock);
268 dev_base_seq_inc(dev_net(dev));
275 static RAW_NOTIFIER_HEAD(netdev_chain);
278 * Device drivers call our routines to queue packets here. We empty the
279 * queue in the local softnet handler.
282 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
283 EXPORT_PER_CPU_SYMBOL(softnet_data);
285 #ifdef CONFIG_LOCKDEP
287 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
288 * according to dev->type
290 static const unsigned short netdev_lock_type[] =
291 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
292 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
293 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
294 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
295 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
296 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
297 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
298 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
299 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
300 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
301 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
302 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
303 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
304 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
305 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
307 static const char *const netdev_lock_name[] =
308 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
309 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
310 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
311 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
312 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
313 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
314 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
315 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
316 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
317 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
318 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
319 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
320 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
321 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
322 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
324 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
325 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
327 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
331 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
332 if (netdev_lock_type[i] == dev_type)
334 /* the last key is used by default */
335 return ARRAY_SIZE(netdev_lock_type) - 1;
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
339 unsigned short dev_type)
343 i = netdev_lock_pos(dev_type);
344 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
345 netdev_lock_name[i]);
348 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
352 i = netdev_lock_pos(dev->type);
353 lockdep_set_class_and_name(&dev->addr_list_lock,
354 &netdev_addr_lock_key[i],
355 netdev_lock_name[i]);
358 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
359 unsigned short dev_type)
362 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
367 /*******************************************************************************
369 Protocol management and registration routines
371 *******************************************************************************/
374 * Add a protocol ID to the list. Now that the input handler is
375 * smarter we can dispense with all the messy stuff that used to be
378 * BEWARE!!! Protocol handlers, mangling input packets,
379 * MUST BE last in hash buckets and checking protocol handlers
380 * MUST start from promiscuous ptype_all chain in net_bh.
381 * It is true now, do not change it.
382 * Explanation follows: if protocol handler, mangling packet, will
383 * be the first on list, it is not able to sense, that packet
384 * is cloned and should be copied-on-write, so that it will
385 * change it and subsequent readers will get broken packet.
389 static inline struct list_head *ptype_head(const struct packet_type *pt)
391 if (pt->type == htons(ETH_P_ALL))
394 return &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
398 * dev_add_pack - add packet handler
399 * @pt: packet type declaration
401 * Add a protocol handler to the networking stack. The passed &packet_type
402 * is linked into kernel lists and may not be freed until it has been
403 * removed from the kernel lists.
405 * This call does not sleep therefore it can not
406 * guarantee all CPU's that are in middle of receiving packets
407 * will see the new packet type (until the next received packet).
410 void dev_add_pack(struct packet_type *pt)
412 struct list_head *head = ptype_head(pt);
414 spin_lock(&ptype_lock);
415 list_add_rcu(&pt->list, head);
416 spin_unlock(&ptype_lock);
418 EXPORT_SYMBOL(dev_add_pack);
421 * __dev_remove_pack - remove packet handler
422 * @pt: packet type declaration
424 * Remove a protocol handler that was previously added to the kernel
425 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
426 * from the kernel lists and can be freed or reused once this function
429 * The packet type might still be in use by receivers
430 * and must not be freed until after all the CPU's have gone
431 * through a quiescent state.
433 void __dev_remove_pack(struct packet_type *pt)
435 struct list_head *head = ptype_head(pt);
436 struct packet_type *pt1;
438 spin_lock(&ptype_lock);
440 list_for_each_entry(pt1, head, list) {
442 list_del_rcu(&pt->list);
447 pr_warn("dev_remove_pack: %p not found\n", pt);
449 spin_unlock(&ptype_lock);
451 EXPORT_SYMBOL(__dev_remove_pack);
454 * dev_remove_pack - remove packet handler
455 * @pt: packet type declaration
457 * Remove a protocol handler that was previously added to the kernel
458 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
459 * from the kernel lists and can be freed or reused once this function
462 * This call sleeps to guarantee that no CPU is looking at the packet
465 void dev_remove_pack(struct packet_type *pt)
467 __dev_remove_pack(pt);
471 EXPORT_SYMBOL(dev_remove_pack);
473 /******************************************************************************
475 Device Boot-time Settings Routines
477 *******************************************************************************/
479 /* Boot time configuration table */
480 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
483 * netdev_boot_setup_add - add new setup entry
484 * @name: name of the device
485 * @map: configured settings for the device
487 * Adds new setup entry to the dev_boot_setup list. The function
488 * returns 0 on error and 1 on success. This is a generic routine to
491 static int netdev_boot_setup_add(char *name, struct ifmap *map)
493 struct netdev_boot_setup *s;
497 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
498 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
499 memset(s[i].name, 0, sizeof(s[i].name));
500 strlcpy(s[i].name, name, IFNAMSIZ);
501 memcpy(&s[i].map, map, sizeof(s[i].map));
506 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
510 * netdev_boot_setup_check - check boot time settings
511 * @dev: the netdevice
513 * Check boot time settings for the device.
514 * The found settings are set for the device to be used
515 * later in the device probing.
516 * Returns 0 if no settings found, 1 if they are.
518 int netdev_boot_setup_check(struct net_device *dev)
520 struct netdev_boot_setup *s = dev_boot_setup;
523 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
524 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
525 !strcmp(dev->name, s[i].name)) {
526 dev->irq = s[i].map.irq;
527 dev->base_addr = s[i].map.base_addr;
528 dev->mem_start = s[i].map.mem_start;
529 dev->mem_end = s[i].map.mem_end;
535 EXPORT_SYMBOL(netdev_boot_setup_check);
539 * netdev_boot_base - get address from boot time settings
540 * @prefix: prefix for network device
541 * @unit: id for network device
543 * Check boot time settings for the base address of device.
544 * The found settings are set for the device to be used
545 * later in the device probing.
546 * Returns 0 if no settings found.
548 unsigned long netdev_boot_base(const char *prefix, int unit)
550 const struct netdev_boot_setup *s = dev_boot_setup;
554 sprintf(name, "%s%d", prefix, unit);
557 * If device already registered then return base of 1
558 * to indicate not to probe for this interface
560 if (__dev_get_by_name(&init_net, name))
563 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
564 if (!strcmp(name, s[i].name))
565 return s[i].map.base_addr;
570 * Saves at boot time configured settings for any netdevice.
572 int __init netdev_boot_setup(char *str)
577 str = get_options(str, ARRAY_SIZE(ints), ints);
582 memset(&map, 0, sizeof(map));
586 map.base_addr = ints[2];
588 map.mem_start = ints[3];
590 map.mem_end = ints[4];
592 /* Add new entry to the list */
593 return netdev_boot_setup_add(str, &map);
596 __setup("netdev=", netdev_boot_setup);
598 /*******************************************************************************
600 Device Interface Subroutines
602 *******************************************************************************/
605 * __dev_get_by_name - find a device by its name
606 * @net: the applicable net namespace
607 * @name: name to find
609 * Find an interface by name. Must be called under RTNL semaphore
610 * or @dev_base_lock. If the name is found a pointer to the device
611 * is returned. If the name is not found then %NULL is returned. The
612 * reference counters are not incremented so the caller must be
613 * careful with locks.
616 struct net_device *__dev_get_by_name(struct net *net, const char *name)
618 struct hlist_node *p;
619 struct net_device *dev;
620 struct hlist_head *head = dev_name_hash(net, name);
622 hlist_for_each_entry(dev, p, head, name_hlist)
623 if (!strncmp(dev->name, name, IFNAMSIZ))
628 EXPORT_SYMBOL(__dev_get_by_name);
631 * dev_get_by_name_rcu - find a device by its name
632 * @net: the applicable net namespace
633 * @name: name to find
635 * Find an interface by name.
636 * If the name is found a pointer to the device is returned.
637 * If the name is not found then %NULL is returned.
638 * The reference counters are not incremented so the caller must be
639 * careful with locks. The caller must hold RCU lock.
642 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
644 struct hlist_node *p;
645 struct net_device *dev;
646 struct hlist_head *head = dev_name_hash(net, name);
648 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
649 if (!strncmp(dev->name, name, IFNAMSIZ))
654 EXPORT_SYMBOL(dev_get_by_name_rcu);
657 * dev_get_by_name - find a device by its name
658 * @net: the applicable net namespace
659 * @name: name to find
661 * Find an interface by name. This can be called from any
662 * context and does its own locking. The returned handle has
663 * the usage count incremented and the caller must use dev_put() to
664 * release it when it is no longer needed. %NULL is returned if no
665 * matching device is found.
668 struct net_device *dev_get_by_name(struct net *net, const char *name)
670 struct net_device *dev;
673 dev = dev_get_by_name_rcu(net, name);
679 EXPORT_SYMBOL(dev_get_by_name);
682 * __dev_get_by_index - find a device by its ifindex
683 * @net: the applicable net namespace
684 * @ifindex: index of device
686 * Search for an interface by index. Returns %NULL if the device
687 * is not found or a pointer to the device. The device has not
688 * had its reference counter increased so the caller must be careful
689 * about locking. The caller must hold either the RTNL semaphore
693 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
695 struct hlist_node *p;
696 struct net_device *dev;
697 struct hlist_head *head = dev_index_hash(net, ifindex);
699 hlist_for_each_entry(dev, p, head, index_hlist)
700 if (dev->ifindex == ifindex)
705 EXPORT_SYMBOL(__dev_get_by_index);
708 * dev_get_by_index_rcu - find a device by its ifindex
709 * @net: the applicable net namespace
710 * @ifindex: index of device
712 * Search for an interface by index. Returns %NULL if the device
713 * is not found or a pointer to the device. The device has not
714 * had its reference counter increased so the caller must be careful
715 * about locking. The caller must hold RCU lock.
718 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
720 struct hlist_node *p;
721 struct net_device *dev;
722 struct hlist_head *head = dev_index_hash(net, ifindex);
724 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
725 if (dev->ifindex == ifindex)
730 EXPORT_SYMBOL(dev_get_by_index_rcu);
734 * dev_get_by_index - find a device by its ifindex
735 * @net: the applicable net namespace
736 * @ifindex: index of device
738 * Search for an interface by index. Returns NULL if the device
739 * is not found or a pointer to the device. The device returned has
740 * had a reference added and the pointer is safe until the user calls
741 * dev_put to indicate they have finished with it.
744 struct net_device *dev_get_by_index(struct net *net, int ifindex)
746 struct net_device *dev;
749 dev = dev_get_by_index_rcu(net, ifindex);
755 EXPORT_SYMBOL(dev_get_by_index);
758 * dev_getbyhwaddr_rcu - find a device by its hardware address
759 * @net: the applicable net namespace
760 * @type: media type of device
761 * @ha: hardware address
763 * Search for an interface by MAC address. Returns NULL if the device
764 * is not found or a pointer to the device.
765 * The caller must hold RCU or RTNL.
766 * The returned device has not had its ref count increased
767 * and the caller must therefore be careful about locking
771 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
774 struct net_device *dev;
776 for_each_netdev_rcu(net, dev)
777 if (dev->type == type &&
778 !memcmp(dev->dev_addr, ha, dev->addr_len))
783 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
785 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
787 struct net_device *dev;
790 for_each_netdev(net, dev)
791 if (dev->type == type)
796 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
798 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
800 struct net_device *dev, *ret = NULL;
803 for_each_netdev_rcu(net, dev)
804 if (dev->type == type) {
812 EXPORT_SYMBOL(dev_getfirstbyhwtype);
815 * dev_get_by_flags_rcu - find any device with given flags
816 * @net: the applicable net namespace
817 * @if_flags: IFF_* values
818 * @mask: bitmask of bits in if_flags to check
820 * Search for any interface with the given flags. Returns NULL if a device
821 * is not found or a pointer to the device. Must be called inside
822 * rcu_read_lock(), and result refcount is unchanged.
825 struct net_device *dev_get_by_flags_rcu(struct net *net, unsigned short if_flags,
828 struct net_device *dev, *ret;
831 for_each_netdev_rcu(net, dev) {
832 if (((dev->flags ^ if_flags) & mask) == 0) {
839 EXPORT_SYMBOL(dev_get_by_flags_rcu);
842 * dev_valid_name - check if name is okay for network device
845 * Network device names need to be valid file names to
846 * to allow sysfs to work. We also disallow any kind of
849 bool dev_valid_name(const char *name)
853 if (strlen(name) >= IFNAMSIZ)
855 if (!strcmp(name, ".") || !strcmp(name, ".."))
859 if (*name == '/' || isspace(*name))
865 EXPORT_SYMBOL(dev_valid_name);
868 * __dev_alloc_name - allocate a name for a device
869 * @net: network namespace to allocate the device name in
870 * @name: name format string
871 * @buf: scratch buffer and result name string
873 * Passed a format string - eg "lt%d" it will try and find a suitable
874 * id. It scans list of devices to build up a free map, then chooses
875 * the first empty slot. The caller must hold the dev_base or rtnl lock
876 * while allocating the name and adding the device in order to avoid
878 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
879 * Returns the number of the unit assigned or a negative errno code.
882 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
886 const int max_netdevices = 8*PAGE_SIZE;
887 unsigned long *inuse;
888 struct net_device *d;
890 p = strnchr(name, IFNAMSIZ-1, '%');
893 * Verify the string as this thing may have come from
894 * the user. There must be either one "%d" and no other "%"
897 if (p[1] != 'd' || strchr(p + 2, '%'))
900 /* Use one page as a bit array of possible slots */
901 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
905 for_each_netdev(net, d) {
906 if (!sscanf(d->name, name, &i))
908 if (i < 0 || i >= max_netdevices)
911 /* avoid cases where sscanf is not exact inverse of printf */
912 snprintf(buf, IFNAMSIZ, name, i);
913 if (!strncmp(buf, d->name, IFNAMSIZ))
917 i = find_first_zero_bit(inuse, max_netdevices);
918 free_page((unsigned long) inuse);
922 snprintf(buf, IFNAMSIZ, name, i);
923 if (!__dev_get_by_name(net, buf))
926 /* It is possible to run out of possible slots
927 * when the name is long and there isn't enough space left
928 * for the digits, or if all bits are used.
934 * dev_alloc_name - allocate a name for a device
936 * @name: name format string
938 * Passed a format string - eg "lt%d" it will try and find a suitable
939 * id. It scans list of devices to build up a free map, then chooses
940 * the first empty slot. The caller must hold the dev_base or rtnl lock
941 * while allocating the name and adding the device in order to avoid
943 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
944 * Returns the number of the unit assigned or a negative errno code.
947 int dev_alloc_name(struct net_device *dev, const char *name)
953 BUG_ON(!dev_net(dev));
955 ret = __dev_alloc_name(net, name, buf);
957 strlcpy(dev->name, buf, IFNAMSIZ);
960 EXPORT_SYMBOL(dev_alloc_name);
962 static int dev_get_valid_name(struct net_device *dev, const char *name)
966 BUG_ON(!dev_net(dev));
969 if (!dev_valid_name(name))
972 if (strchr(name, '%'))
973 return dev_alloc_name(dev, name);
974 else if (__dev_get_by_name(net, name))
976 else if (dev->name != name)
977 strlcpy(dev->name, name, IFNAMSIZ);
983 * dev_change_name - change name of a device
985 * @newname: name (or format string) must be at least IFNAMSIZ
987 * Change name of a device, can pass format strings "eth%d".
990 int dev_change_name(struct net_device *dev, const char *newname)
992 char oldname[IFNAMSIZ];
998 BUG_ON(!dev_net(dev));
1001 if (dev->flags & IFF_UP)
1004 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1007 memcpy(oldname, dev->name, IFNAMSIZ);
1009 err = dev_get_valid_name(dev, newname);
1014 ret = device_rename(&dev->dev, dev->name);
1016 memcpy(dev->name, oldname, IFNAMSIZ);
1020 write_lock_bh(&dev_base_lock);
1021 hlist_del_rcu(&dev->name_hlist);
1022 write_unlock_bh(&dev_base_lock);
1026 write_lock_bh(&dev_base_lock);
1027 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1028 write_unlock_bh(&dev_base_lock);
1030 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1031 ret = notifier_to_errno(ret);
1034 /* err >= 0 after dev_alloc_name() or stores the first errno */
1037 memcpy(dev->name, oldname, IFNAMSIZ);
1040 pr_err("%s: name change rollback failed: %d\n",
1049 * dev_set_alias - change ifalias of a device
1051 * @alias: name up to IFALIASZ
1052 * @len: limit of bytes to copy from info
1054 * Set ifalias for a device,
1056 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1060 if (len >= IFALIASZ)
1065 kfree(dev->ifalias);
1066 dev->ifalias = NULL;
1071 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1075 strlcpy(dev->ifalias, alias, len+1);
1081 * netdev_features_change - device changes features
1082 * @dev: device to cause notification
1084 * Called to indicate a device has changed features.
1086 void netdev_features_change(struct net_device *dev)
1088 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1090 EXPORT_SYMBOL(netdev_features_change);
1093 * netdev_state_change - device changes state
1094 * @dev: device to cause notification
1096 * Called to indicate a device has changed state. This function calls
1097 * the notifier chains for netdev_chain and sends a NEWLINK message
1098 * to the routing socket.
1100 void netdev_state_change(struct net_device *dev)
1102 if (dev->flags & IFF_UP) {
1103 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1104 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1107 EXPORT_SYMBOL(netdev_state_change);
1109 int netdev_bonding_change(struct net_device *dev, unsigned long event)
1111 return call_netdevice_notifiers(event, dev);
1113 EXPORT_SYMBOL(netdev_bonding_change);
1116 * dev_load - load a network module
1117 * @net: the applicable net namespace
1118 * @name: name of interface
1120 * If a network interface is not present and the process has suitable
1121 * privileges this function loads the module. If module loading is not
1122 * available in this kernel then it becomes a nop.
1125 void dev_load(struct net *net, const char *name)
1127 struct net_device *dev;
1131 dev = dev_get_by_name_rcu(net, name);
1135 if (no_module && capable(CAP_NET_ADMIN))
1136 no_module = request_module("netdev-%s", name);
1137 if (no_module && capable(CAP_SYS_MODULE)) {
1138 if (!request_module("%s", name))
1139 pr_err("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1143 EXPORT_SYMBOL(dev_load);
1145 static int __dev_open(struct net_device *dev)
1147 const struct net_device_ops *ops = dev->netdev_ops;
1152 if (!netif_device_present(dev))
1155 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1156 ret = notifier_to_errno(ret);
1160 set_bit(__LINK_STATE_START, &dev->state);
1162 if (ops->ndo_validate_addr)
1163 ret = ops->ndo_validate_addr(dev);
1165 if (!ret && ops->ndo_open)
1166 ret = ops->ndo_open(dev);
1169 clear_bit(__LINK_STATE_START, &dev->state);
1171 dev->flags |= IFF_UP;
1172 net_dmaengine_get();
1173 dev_set_rx_mode(dev);
1181 * dev_open - prepare an interface for use.
1182 * @dev: device to open
1184 * Takes a device from down to up state. The device's private open
1185 * function is invoked and then the multicast lists are loaded. Finally
1186 * the device is moved into the up state and a %NETDEV_UP message is
1187 * sent to the netdev notifier chain.
1189 * Calling this function on an active interface is a nop. On a failure
1190 * a negative errno code is returned.
1192 int dev_open(struct net_device *dev)
1196 if (dev->flags & IFF_UP)
1199 ret = __dev_open(dev);
1203 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1204 call_netdevice_notifiers(NETDEV_UP, dev);
1208 EXPORT_SYMBOL(dev_open);
1210 static int __dev_close_many(struct list_head *head)
1212 struct net_device *dev;
1217 list_for_each_entry(dev, head, unreg_list) {
1218 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1220 clear_bit(__LINK_STATE_START, &dev->state);
1222 /* Synchronize to scheduled poll. We cannot touch poll list, it
1223 * can be even on different cpu. So just clear netif_running().
1225 * dev->stop() will invoke napi_disable() on all of it's
1226 * napi_struct instances on this device.
1228 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1231 dev_deactivate_many(head);
1233 list_for_each_entry(dev, head, unreg_list) {
1234 const struct net_device_ops *ops = dev->netdev_ops;
1237 * Call the device specific close. This cannot fail.
1238 * Only if device is UP
1240 * We allow it to be called even after a DETACH hot-plug
1246 dev->flags &= ~IFF_UP;
1247 net_dmaengine_put();
1253 static int __dev_close(struct net_device *dev)
1258 list_add(&dev->unreg_list, &single);
1259 retval = __dev_close_many(&single);
1264 static int dev_close_many(struct list_head *head)
1266 struct net_device *dev, *tmp;
1267 LIST_HEAD(tmp_list);
1269 list_for_each_entry_safe(dev, tmp, head, unreg_list)
1270 if (!(dev->flags & IFF_UP))
1271 list_move(&dev->unreg_list, &tmp_list);
1273 __dev_close_many(head);
1275 list_for_each_entry(dev, head, unreg_list) {
1276 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1277 call_netdevice_notifiers(NETDEV_DOWN, dev);
1280 /* rollback_registered_many needs the complete original list */
1281 list_splice(&tmp_list, head);
1286 * dev_close - shutdown an interface.
1287 * @dev: device to shutdown
1289 * This function moves an active device into down state. A
1290 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1291 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1294 int dev_close(struct net_device *dev)
1296 if (dev->flags & IFF_UP) {
1299 list_add(&dev->unreg_list, &single);
1300 dev_close_many(&single);
1305 EXPORT_SYMBOL(dev_close);
1309 * dev_disable_lro - disable Large Receive Offload on a device
1312 * Disable Large Receive Offload (LRO) on a net device. Must be
1313 * called under RTNL. This is needed if received packets may be
1314 * forwarded to another interface.
1316 void dev_disable_lro(struct net_device *dev)
1319 * If we're trying to disable lro on a vlan device
1320 * use the underlying physical device instead
1322 if (is_vlan_dev(dev))
1323 dev = vlan_dev_real_dev(dev);
1325 dev->wanted_features &= ~NETIF_F_LRO;
1326 netdev_update_features(dev);
1328 if (unlikely(dev->features & NETIF_F_LRO))
1329 netdev_WARN(dev, "failed to disable LRO!\n");
1331 EXPORT_SYMBOL(dev_disable_lro);
1334 static int dev_boot_phase = 1;
1337 * register_netdevice_notifier - register a network notifier block
1340 * Register a notifier to be called when network device events occur.
1341 * The notifier passed is linked into the kernel structures and must
1342 * not be reused until it has been unregistered. A negative errno code
1343 * is returned on a failure.
1345 * When registered all registration and up events are replayed
1346 * to the new notifier to allow device to have a race free
1347 * view of the network device list.
1350 int register_netdevice_notifier(struct notifier_block *nb)
1352 struct net_device *dev;
1353 struct net_device *last;
1358 err = raw_notifier_chain_register(&netdev_chain, nb);
1364 for_each_netdev(net, dev) {
1365 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1366 err = notifier_to_errno(err);
1370 if (!(dev->flags & IFF_UP))
1373 nb->notifier_call(nb, NETDEV_UP, dev);
1384 for_each_netdev(net, dev) {
1388 if (dev->flags & IFF_UP) {
1389 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1390 nb->notifier_call(nb, NETDEV_DOWN, dev);
1392 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1393 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1398 raw_notifier_chain_unregister(&netdev_chain, nb);
1401 EXPORT_SYMBOL(register_netdevice_notifier);
1404 * unregister_netdevice_notifier - unregister a network notifier block
1407 * Unregister a notifier previously registered by
1408 * register_netdevice_notifier(). The notifier is unlinked into the
1409 * kernel structures and may then be reused. A negative errno code
1410 * is returned on a failure.
1412 * After unregistering unregister and down device events are synthesized
1413 * for all devices on the device list to the removed notifier to remove
1414 * the need for special case cleanup code.
1417 int unregister_netdevice_notifier(struct notifier_block *nb)
1419 struct net_device *dev;
1424 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1429 for_each_netdev(net, dev) {
1430 if (dev->flags & IFF_UP) {
1431 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1432 nb->notifier_call(nb, NETDEV_DOWN, dev);
1434 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1435 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1442 EXPORT_SYMBOL(unregister_netdevice_notifier);
1445 * call_netdevice_notifiers - call all network notifier blocks
1446 * @val: value passed unmodified to notifier function
1447 * @dev: net_device pointer passed unmodified to notifier function
1449 * Call all network notifier blocks. Parameters and return value
1450 * are as for raw_notifier_call_chain().
1453 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1456 return raw_notifier_call_chain(&netdev_chain, val, dev);
1458 EXPORT_SYMBOL(call_netdevice_notifiers);
1460 static struct static_key netstamp_needed __read_mostly;
1461 #ifdef HAVE_JUMP_LABEL
1462 /* We are not allowed to call static_key_slow_dec() from irq context
1463 * If net_disable_timestamp() is called from irq context, defer the
1464 * static_key_slow_dec() calls.
1466 static atomic_t netstamp_needed_deferred;
1469 void net_enable_timestamp(void)
1471 #ifdef HAVE_JUMP_LABEL
1472 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1476 static_key_slow_dec(&netstamp_needed);
1480 WARN_ON(in_interrupt());
1481 static_key_slow_inc(&netstamp_needed);
1483 EXPORT_SYMBOL(net_enable_timestamp);
1485 void net_disable_timestamp(void)
1487 #ifdef HAVE_JUMP_LABEL
1488 if (in_interrupt()) {
1489 atomic_inc(&netstamp_needed_deferred);
1493 static_key_slow_dec(&netstamp_needed);
1495 EXPORT_SYMBOL(net_disable_timestamp);
1497 static inline void net_timestamp_set(struct sk_buff *skb)
1499 skb->tstamp.tv64 = 0;
1500 if (static_key_false(&netstamp_needed))
1501 __net_timestamp(skb);
1504 #define net_timestamp_check(COND, SKB) \
1505 if (static_key_false(&netstamp_needed)) { \
1506 if ((COND) && !(SKB)->tstamp.tv64) \
1507 __net_timestamp(SKB); \
1510 static int net_hwtstamp_validate(struct ifreq *ifr)
1512 struct hwtstamp_config cfg;
1513 enum hwtstamp_tx_types tx_type;
1514 enum hwtstamp_rx_filters rx_filter;
1515 int tx_type_valid = 0;
1516 int rx_filter_valid = 0;
1518 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
1521 if (cfg.flags) /* reserved for future extensions */
1524 tx_type = cfg.tx_type;
1525 rx_filter = cfg.rx_filter;
1528 case HWTSTAMP_TX_OFF:
1529 case HWTSTAMP_TX_ON:
1530 case HWTSTAMP_TX_ONESTEP_SYNC:
1535 switch (rx_filter) {
1536 case HWTSTAMP_FILTER_NONE:
1537 case HWTSTAMP_FILTER_ALL:
1538 case HWTSTAMP_FILTER_SOME:
1539 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
1540 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
1541 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
1542 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
1543 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
1544 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
1545 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
1546 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
1547 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
1548 case HWTSTAMP_FILTER_PTP_V2_EVENT:
1549 case HWTSTAMP_FILTER_PTP_V2_SYNC:
1550 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
1551 rx_filter_valid = 1;
1555 if (!tx_type_valid || !rx_filter_valid)
1561 static inline bool is_skb_forwardable(struct net_device *dev,
1562 struct sk_buff *skb)
1566 if (!(dev->flags & IFF_UP))
1569 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1570 if (skb->len <= len)
1573 /* if TSO is enabled, we don't care about the length as the packet
1574 * could be forwarded without being segmented before
1576 if (skb_is_gso(skb))
1583 * dev_forward_skb - loopback an skb to another netif
1585 * @dev: destination network device
1586 * @skb: buffer to forward
1589 * NET_RX_SUCCESS (no congestion)
1590 * NET_RX_DROP (packet was dropped, but freed)
1592 * dev_forward_skb can be used for injecting an skb from the
1593 * start_xmit function of one device into the receive queue
1594 * of another device.
1596 * The receiving device may be in another namespace, so
1597 * we have to clear all information in the skb that could
1598 * impact namespace isolation.
1600 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1602 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
1603 if (skb_copy_ubufs(skb, GFP_ATOMIC)) {
1604 atomic_long_inc(&dev->rx_dropped);
1613 if (unlikely(!is_skb_forwardable(dev, skb))) {
1614 atomic_long_inc(&dev->rx_dropped);
1621 skb->tstamp.tv64 = 0;
1622 skb->pkt_type = PACKET_HOST;
1623 skb->protocol = eth_type_trans(skb, dev);
1627 return netif_rx(skb);
1629 EXPORT_SYMBOL_GPL(dev_forward_skb);
1631 static inline int deliver_skb(struct sk_buff *skb,
1632 struct packet_type *pt_prev,
1633 struct net_device *orig_dev)
1635 atomic_inc(&skb->users);
1636 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1640 * Support routine. Sends outgoing frames to any network
1641 * taps currently in use.
1644 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1646 struct packet_type *ptype;
1647 struct sk_buff *skb2 = NULL;
1648 struct packet_type *pt_prev = NULL;
1651 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1652 /* Never send packets back to the socket
1653 * they originated from - MvS (miquels@drinkel.ow.org)
1655 if ((ptype->dev == dev || !ptype->dev) &&
1656 (ptype->af_packet_priv == NULL ||
1657 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1659 deliver_skb(skb2, pt_prev, skb->dev);
1664 skb2 = skb_clone(skb, GFP_ATOMIC);
1668 net_timestamp_set(skb2);
1670 /* skb->nh should be correctly
1671 set by sender, so that the second statement is
1672 just protection against buggy protocols.
1674 skb_reset_mac_header(skb2);
1676 if (skb_network_header(skb2) < skb2->data ||
1677 skb2->network_header > skb2->tail) {
1678 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1679 ntohs(skb2->protocol),
1681 skb_reset_network_header(skb2);
1684 skb2->transport_header = skb2->network_header;
1685 skb2->pkt_type = PACKET_OUTGOING;
1690 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
1694 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1695 * @dev: Network device
1696 * @txq: number of queues available
1698 * If real_num_tx_queues is changed the tc mappings may no longer be
1699 * valid. To resolve this verify the tc mapping remains valid and if
1700 * not NULL the mapping. With no priorities mapping to this
1701 * offset/count pair it will no longer be used. In the worst case TC0
1702 * is invalid nothing can be done so disable priority mappings. If is
1703 * expected that drivers will fix this mapping if they can before
1704 * calling netif_set_real_num_tx_queues.
1706 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
1709 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
1711 /* If TC0 is invalidated disable TC mapping */
1712 if (tc->offset + tc->count > txq) {
1713 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1718 /* Invalidated prio to tc mappings set to TC0 */
1719 for (i = 1; i < TC_BITMASK + 1; i++) {
1720 int q = netdev_get_prio_tc_map(dev, i);
1722 tc = &dev->tc_to_txq[q];
1723 if (tc->offset + tc->count > txq) {
1724 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1726 netdev_set_prio_tc_map(dev, i, 0);
1732 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1733 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1735 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
1739 if (txq < 1 || txq > dev->num_tx_queues)
1742 if (dev->reg_state == NETREG_REGISTERED ||
1743 dev->reg_state == NETREG_UNREGISTERING) {
1746 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
1752 netif_setup_tc(dev, txq);
1754 if (txq < dev->real_num_tx_queues)
1755 qdisc_reset_all_tx_gt(dev, txq);
1758 dev->real_num_tx_queues = txq;
1761 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
1765 * netif_set_real_num_rx_queues - set actual number of RX queues used
1766 * @dev: Network device
1767 * @rxq: Actual number of RX queues
1769 * This must be called either with the rtnl_lock held or before
1770 * registration of the net device. Returns 0 on success, or a
1771 * negative error code. If called before registration, it always
1774 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
1778 if (rxq < 1 || rxq > dev->num_rx_queues)
1781 if (dev->reg_state == NETREG_REGISTERED) {
1784 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
1790 dev->real_num_rx_queues = rxq;
1793 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
1796 static inline void __netif_reschedule(struct Qdisc *q)
1798 struct softnet_data *sd;
1799 unsigned long flags;
1801 local_irq_save(flags);
1802 sd = &__get_cpu_var(softnet_data);
1803 q->next_sched = NULL;
1804 *sd->output_queue_tailp = q;
1805 sd->output_queue_tailp = &q->next_sched;
1806 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1807 local_irq_restore(flags);
1810 void __netif_schedule(struct Qdisc *q)
1812 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1813 __netif_reschedule(q);
1815 EXPORT_SYMBOL(__netif_schedule);
1817 void dev_kfree_skb_irq(struct sk_buff *skb)
1819 if (atomic_dec_and_test(&skb->users)) {
1820 struct softnet_data *sd;
1821 unsigned long flags;
1823 local_irq_save(flags);
1824 sd = &__get_cpu_var(softnet_data);
1825 skb->next = sd->completion_queue;
1826 sd->completion_queue = skb;
1827 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1828 local_irq_restore(flags);
1831 EXPORT_SYMBOL(dev_kfree_skb_irq);
1833 void dev_kfree_skb_any(struct sk_buff *skb)
1835 if (in_irq() || irqs_disabled())
1836 dev_kfree_skb_irq(skb);
1840 EXPORT_SYMBOL(dev_kfree_skb_any);
1844 * netif_device_detach - mark device as removed
1845 * @dev: network device
1847 * Mark device as removed from system and therefore no longer available.
1849 void netif_device_detach(struct net_device *dev)
1851 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1852 netif_running(dev)) {
1853 netif_tx_stop_all_queues(dev);
1856 EXPORT_SYMBOL(netif_device_detach);
1859 * netif_device_attach - mark device as attached
1860 * @dev: network device
1862 * Mark device as attached from system and restart if needed.
1864 void netif_device_attach(struct net_device *dev)
1866 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1867 netif_running(dev)) {
1868 netif_tx_wake_all_queues(dev);
1869 __netdev_watchdog_up(dev);
1872 EXPORT_SYMBOL(netif_device_attach);
1874 static void skb_warn_bad_offload(const struct sk_buff *skb)
1876 static const netdev_features_t null_features = 0;
1877 struct net_device *dev = skb->dev;
1878 const char *driver = "";
1880 if (dev && dev->dev.parent)
1881 driver = dev_driver_string(dev->dev.parent);
1883 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1884 "gso_type=%d ip_summed=%d\n",
1885 driver, dev ? &dev->features : &null_features,
1886 skb->sk ? &skb->sk->sk_route_caps : &null_features,
1887 skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
1888 skb_shinfo(skb)->gso_type, skb->ip_summed);
1892 * Invalidate hardware checksum when packet is to be mangled, and
1893 * complete checksum manually on outgoing path.
1895 int skb_checksum_help(struct sk_buff *skb)
1898 int ret = 0, offset;
1900 if (skb->ip_summed == CHECKSUM_COMPLETE)
1901 goto out_set_summed;
1903 if (unlikely(skb_shinfo(skb)->gso_size)) {
1904 skb_warn_bad_offload(skb);
1908 offset = skb_checksum_start_offset(skb);
1909 BUG_ON(offset >= skb_headlen(skb));
1910 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1912 offset += skb->csum_offset;
1913 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1915 if (skb_cloned(skb) &&
1916 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1917 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1922 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1924 skb->ip_summed = CHECKSUM_NONE;
1928 EXPORT_SYMBOL(skb_checksum_help);
1931 * skb_gso_segment - Perform segmentation on skb.
1932 * @skb: buffer to segment
1933 * @features: features for the output path (see dev->features)
1935 * This function segments the given skb and returns a list of segments.
1937 * It may return NULL if the skb requires no segmentation. This is
1938 * only possible when GSO is used for verifying header integrity.
1940 struct sk_buff *skb_gso_segment(struct sk_buff *skb,
1941 netdev_features_t features)
1943 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1944 struct packet_type *ptype;
1945 __be16 type = skb->protocol;
1946 int vlan_depth = ETH_HLEN;
1949 while (type == htons(ETH_P_8021Q)) {
1950 struct vlan_hdr *vh;
1952 if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
1953 return ERR_PTR(-EINVAL);
1955 vh = (struct vlan_hdr *)(skb->data + vlan_depth);
1956 type = vh->h_vlan_encapsulated_proto;
1957 vlan_depth += VLAN_HLEN;
1960 skb_reset_mac_header(skb);
1961 skb->mac_len = skb->network_header - skb->mac_header;
1962 __skb_pull(skb, skb->mac_len);
1964 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1965 skb_warn_bad_offload(skb);
1967 if (skb_header_cloned(skb) &&
1968 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1969 return ERR_PTR(err);
1973 list_for_each_entry_rcu(ptype,
1974 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1975 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1976 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1977 err = ptype->gso_send_check(skb);
1978 segs = ERR_PTR(err);
1979 if (err || skb_gso_ok(skb, features))
1981 __skb_push(skb, (skb->data -
1982 skb_network_header(skb)));
1984 segs = ptype->gso_segment(skb, features);
1990 __skb_push(skb, skb->data - skb_mac_header(skb));
1994 EXPORT_SYMBOL(skb_gso_segment);
1996 /* Take action when hardware reception checksum errors are detected. */
1998 void netdev_rx_csum_fault(struct net_device *dev)
2000 if (net_ratelimit()) {
2001 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2005 EXPORT_SYMBOL(netdev_rx_csum_fault);
2008 /* Actually, we should eliminate this check as soon as we know, that:
2009 * 1. IOMMU is present and allows to map all the memory.
2010 * 2. No high memory really exists on this machine.
2013 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2015 #ifdef CONFIG_HIGHMEM
2017 if (!(dev->features & NETIF_F_HIGHDMA)) {
2018 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2019 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2020 if (PageHighMem(skb_frag_page(frag)))
2025 if (PCI_DMA_BUS_IS_PHYS) {
2026 struct device *pdev = dev->dev.parent;
2030 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2031 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2032 dma_addr_t addr = page_to_phys(skb_frag_page(frag));
2033 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
2042 void (*destructor)(struct sk_buff *skb);
2045 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2047 static void dev_gso_skb_destructor(struct sk_buff *skb)
2049 struct dev_gso_cb *cb;
2052 struct sk_buff *nskb = skb->next;
2054 skb->next = nskb->next;
2057 } while (skb->next);
2059 cb = DEV_GSO_CB(skb);
2061 cb->destructor(skb);
2065 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2066 * @skb: buffer to segment
2067 * @features: device features as applicable to this skb
2069 * This function segments the given skb and stores the list of segments
2072 static int dev_gso_segment(struct sk_buff *skb, netdev_features_t features)
2074 struct sk_buff *segs;
2076 segs = skb_gso_segment(skb, features);
2078 /* Verifying header integrity only. */
2083 return PTR_ERR(segs);
2086 DEV_GSO_CB(skb)->destructor = skb->destructor;
2087 skb->destructor = dev_gso_skb_destructor;
2093 * Try to orphan skb early, right before transmission by the device.
2094 * We cannot orphan skb if tx timestamp is requested or the sk-reference
2095 * is needed on driver level for other reasons, e.g. see net/can/raw.c
2097 static inline void skb_orphan_try(struct sk_buff *skb)
2099 struct sock *sk = skb->sk;
2101 if (sk && !skb_shinfo(skb)->tx_flags) {
2102 /* skb_tx_hash() wont be able to get sk.
2103 * We copy sk_hash into skb->rxhash
2106 skb->rxhash = sk->sk_hash;
2111 static bool can_checksum_protocol(netdev_features_t features, __be16 protocol)
2113 return ((features & NETIF_F_GEN_CSUM) ||
2114 ((features & NETIF_F_V4_CSUM) &&
2115 protocol == htons(ETH_P_IP)) ||
2116 ((features & NETIF_F_V6_CSUM) &&
2117 protocol == htons(ETH_P_IPV6)) ||
2118 ((features & NETIF_F_FCOE_CRC) &&
2119 protocol == htons(ETH_P_FCOE)));
2122 static netdev_features_t harmonize_features(struct sk_buff *skb,
2123 __be16 protocol, netdev_features_t features)
2125 if (!can_checksum_protocol(features, protocol)) {
2126 features &= ~NETIF_F_ALL_CSUM;
2127 features &= ~NETIF_F_SG;
2128 } else if (illegal_highdma(skb->dev, skb)) {
2129 features &= ~NETIF_F_SG;
2135 netdev_features_t netif_skb_features(struct sk_buff *skb)
2137 __be16 protocol = skb->protocol;
2138 netdev_features_t features = skb->dev->features;
2140 if (protocol == htons(ETH_P_8021Q)) {
2141 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
2142 protocol = veh->h_vlan_encapsulated_proto;
2143 } else if (!vlan_tx_tag_present(skb)) {
2144 return harmonize_features(skb, protocol, features);
2147 features &= (skb->dev->vlan_features | NETIF_F_HW_VLAN_TX);
2149 if (protocol != htons(ETH_P_8021Q)) {
2150 return harmonize_features(skb, protocol, features);
2152 features &= NETIF_F_SG | NETIF_F_HIGHDMA | NETIF_F_FRAGLIST |
2153 NETIF_F_GEN_CSUM | NETIF_F_HW_VLAN_TX;
2154 return harmonize_features(skb, protocol, features);
2157 EXPORT_SYMBOL(netif_skb_features);
2160 * Returns true if either:
2161 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2162 * 2. skb is fragmented and the device does not support SG, or if
2163 * at least one of fragments is in highmem and device does not
2164 * support DMA from it.
2166 static inline int skb_needs_linearize(struct sk_buff *skb,
2169 return skb_is_nonlinear(skb) &&
2170 ((skb_has_frag_list(skb) &&
2171 !(features & NETIF_F_FRAGLIST)) ||
2172 (skb_shinfo(skb)->nr_frags &&
2173 !(features & NETIF_F_SG)));
2176 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
2177 struct netdev_queue *txq)
2179 const struct net_device_ops *ops = dev->netdev_ops;
2180 int rc = NETDEV_TX_OK;
2181 unsigned int skb_len;
2183 if (likely(!skb->next)) {
2184 netdev_features_t features;
2187 * If device doesn't need skb->dst, release it right now while
2188 * its hot in this cpu cache
2190 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2193 if (!list_empty(&ptype_all))
2194 dev_queue_xmit_nit(skb, dev);
2196 skb_orphan_try(skb);
2198 features = netif_skb_features(skb);
2200 if (vlan_tx_tag_present(skb) &&
2201 !(features & NETIF_F_HW_VLAN_TX)) {
2202 skb = __vlan_put_tag(skb, vlan_tx_tag_get(skb));
2209 if (netif_needs_gso(skb, features)) {
2210 if (unlikely(dev_gso_segment(skb, features)))
2215 if (skb_needs_linearize(skb, features) &&
2216 __skb_linearize(skb))
2219 /* If packet is not checksummed and device does not
2220 * support checksumming for this protocol, complete
2221 * checksumming here.
2223 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2224 skb_set_transport_header(skb,
2225 skb_checksum_start_offset(skb));
2226 if (!(features & NETIF_F_ALL_CSUM) &&
2227 skb_checksum_help(skb))
2233 rc = ops->ndo_start_xmit(skb, dev);
2234 trace_net_dev_xmit(skb, rc, dev, skb_len);
2235 if (rc == NETDEV_TX_OK)
2236 txq_trans_update(txq);
2242 struct sk_buff *nskb = skb->next;
2244 skb->next = nskb->next;
2248 * If device doesn't need nskb->dst, release it right now while
2249 * its hot in this cpu cache
2251 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
2254 skb_len = nskb->len;
2255 rc = ops->ndo_start_xmit(nskb, dev);
2256 trace_net_dev_xmit(nskb, rc, dev, skb_len);
2257 if (unlikely(rc != NETDEV_TX_OK)) {
2258 if (rc & ~NETDEV_TX_MASK)
2259 goto out_kfree_gso_skb;
2260 nskb->next = skb->next;
2264 txq_trans_update(txq);
2265 if (unlikely(netif_xmit_stopped(txq) && skb->next))
2266 return NETDEV_TX_BUSY;
2267 } while (skb->next);
2270 if (likely(skb->next == NULL))
2271 skb->destructor = DEV_GSO_CB(skb)->destructor;
2278 static u32 hashrnd __read_mostly;
2281 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2282 * to be used as a distribution range.
2284 u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2285 unsigned int num_tx_queues)
2289 u16 qcount = num_tx_queues;
2291 if (skb_rx_queue_recorded(skb)) {
2292 hash = skb_get_rx_queue(skb);
2293 while (unlikely(hash >= num_tx_queues))
2294 hash -= num_tx_queues;
2299 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2300 qoffset = dev->tc_to_txq[tc].offset;
2301 qcount = dev->tc_to_txq[tc].count;
2304 if (skb->sk && skb->sk->sk_hash)
2305 hash = skb->sk->sk_hash;
2307 hash = (__force u16) skb->protocol ^ skb->rxhash;
2308 hash = jhash_1word(hash, hashrnd);
2310 return (u16) (((u64) hash * qcount) >> 32) + qoffset;
2312 EXPORT_SYMBOL(__skb_tx_hash);
2314 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
2316 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
2317 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2318 dev->name, queue_index,
2319 dev->real_num_tx_queues);
2325 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
2328 struct xps_dev_maps *dev_maps;
2329 struct xps_map *map;
2330 int queue_index = -1;
2333 dev_maps = rcu_dereference(dev->xps_maps);
2335 map = rcu_dereference(
2336 dev_maps->cpu_map[raw_smp_processor_id()]);
2339 queue_index = map->queues[0];
2342 if (skb->sk && skb->sk->sk_hash)
2343 hash = skb->sk->sk_hash;
2345 hash = (__force u16) skb->protocol ^
2347 hash = jhash_1word(hash, hashrnd);
2348 queue_index = map->queues[
2349 ((u64)hash * map->len) >> 32];
2351 if (unlikely(queue_index >= dev->real_num_tx_queues))
2363 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2364 struct sk_buff *skb)
2367 const struct net_device_ops *ops = dev->netdev_ops;
2369 if (dev->real_num_tx_queues == 1)
2371 else if (ops->ndo_select_queue) {
2372 queue_index = ops->ndo_select_queue(dev, skb);
2373 queue_index = dev_cap_txqueue(dev, queue_index);
2375 struct sock *sk = skb->sk;
2376 queue_index = sk_tx_queue_get(sk);
2378 if (queue_index < 0 || skb->ooo_okay ||
2379 queue_index >= dev->real_num_tx_queues) {
2380 int old_index = queue_index;
2382 queue_index = get_xps_queue(dev, skb);
2383 if (queue_index < 0)
2384 queue_index = skb_tx_hash(dev, skb);
2386 if (queue_index != old_index && sk) {
2387 struct dst_entry *dst =
2388 rcu_dereference_check(sk->sk_dst_cache, 1);
2390 if (dst && skb_dst(skb) == dst)
2391 sk_tx_queue_set(sk, queue_index);
2396 skb_set_queue_mapping(skb, queue_index);
2397 return netdev_get_tx_queue(dev, queue_index);
2400 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2401 struct net_device *dev,
2402 struct netdev_queue *txq)
2404 spinlock_t *root_lock = qdisc_lock(q);
2408 qdisc_skb_cb(skb)->pkt_len = skb->len;
2409 qdisc_calculate_pkt_len(skb, q);
2411 * Heuristic to force contended enqueues to serialize on a
2412 * separate lock before trying to get qdisc main lock.
2413 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2414 * and dequeue packets faster.
2416 contended = qdisc_is_running(q);
2417 if (unlikely(contended))
2418 spin_lock(&q->busylock);
2420 spin_lock(root_lock);
2421 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2424 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2425 qdisc_run_begin(q)) {
2427 * This is a work-conserving queue; there are no old skbs
2428 * waiting to be sent out; and the qdisc is not running -
2429 * xmit the skb directly.
2431 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2434 qdisc_bstats_update(q, skb);
2436 if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
2437 if (unlikely(contended)) {
2438 spin_unlock(&q->busylock);
2445 rc = NET_XMIT_SUCCESS;
2448 rc = q->enqueue(skb, q) & NET_XMIT_MASK;
2449 if (qdisc_run_begin(q)) {
2450 if (unlikely(contended)) {
2451 spin_unlock(&q->busylock);
2457 spin_unlock(root_lock);
2458 if (unlikely(contended))
2459 spin_unlock(&q->busylock);
2463 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2464 static void skb_update_prio(struct sk_buff *skb)
2466 struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap);
2468 if ((!skb->priority) && (skb->sk) && map)
2469 skb->priority = map->priomap[skb->sk->sk_cgrp_prioidx];
2472 #define skb_update_prio(skb)
2475 static DEFINE_PER_CPU(int, xmit_recursion);
2476 #define RECURSION_LIMIT 10
2479 * dev_queue_xmit - transmit a buffer
2480 * @skb: buffer to transmit
2482 * Queue a buffer for transmission to a network device. The caller must
2483 * have set the device and priority and built the buffer before calling
2484 * this function. The function can be called from an interrupt.
2486 * A negative errno code is returned on a failure. A success does not
2487 * guarantee the frame will be transmitted as it may be dropped due
2488 * to congestion or traffic shaping.
2490 * -----------------------------------------------------------------------------------
2491 * I notice this method can also return errors from the queue disciplines,
2492 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2495 * Regardless of the return value, the skb is consumed, so it is currently
2496 * difficult to retry a send to this method. (You can bump the ref count
2497 * before sending to hold a reference for retry if you are careful.)
2499 * When calling this method, interrupts MUST be enabled. This is because
2500 * the BH enable code must have IRQs enabled so that it will not deadlock.
2503 int dev_queue_xmit(struct sk_buff *skb)
2505 struct net_device *dev = skb->dev;
2506 struct netdev_queue *txq;
2510 /* Disable soft irqs for various locks below. Also
2511 * stops preemption for RCU.
2515 skb_update_prio(skb);
2517 txq = dev_pick_tx(dev, skb);
2518 q = rcu_dereference_bh(txq->qdisc);
2520 #ifdef CONFIG_NET_CLS_ACT
2521 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2523 trace_net_dev_queue(skb);
2525 rc = __dev_xmit_skb(skb, q, dev, txq);
2529 /* The device has no queue. Common case for software devices:
2530 loopback, all the sorts of tunnels...
2532 Really, it is unlikely that netif_tx_lock protection is necessary
2533 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2535 However, it is possible, that they rely on protection
2538 Check this and shot the lock. It is not prone from deadlocks.
2539 Either shot noqueue qdisc, it is even simpler 8)
2541 if (dev->flags & IFF_UP) {
2542 int cpu = smp_processor_id(); /* ok because BHs are off */
2544 if (txq->xmit_lock_owner != cpu) {
2546 if (__this_cpu_read(xmit_recursion) > RECURSION_LIMIT)
2547 goto recursion_alert;
2549 HARD_TX_LOCK(dev, txq, cpu);
2551 if (!netif_xmit_stopped(txq)) {
2552 __this_cpu_inc(xmit_recursion);
2553 rc = dev_hard_start_xmit(skb, dev, txq);
2554 __this_cpu_dec(xmit_recursion);
2555 if (dev_xmit_complete(rc)) {
2556 HARD_TX_UNLOCK(dev, txq);
2560 HARD_TX_UNLOCK(dev, txq);
2561 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2564 /* Recursion is detected! It is possible,
2568 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2574 rcu_read_unlock_bh();
2579 rcu_read_unlock_bh();
2582 EXPORT_SYMBOL(dev_queue_xmit);
2585 /*=======================================================================
2587 =======================================================================*/
2589 int netdev_max_backlog __read_mostly = 1000;
2590 int netdev_tstamp_prequeue __read_mostly = 1;
2591 int netdev_budget __read_mostly = 300;
2592 int weight_p __read_mostly = 64; /* old backlog weight */
2594 /* Called with irq disabled */
2595 static inline void ____napi_schedule(struct softnet_data *sd,
2596 struct napi_struct *napi)
2598 list_add_tail(&napi->poll_list, &sd->poll_list);
2599 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2603 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2604 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2605 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2606 * if hash is a canonical 4-tuple hash over transport ports.
2608 void __skb_get_rxhash(struct sk_buff *skb)
2610 struct flow_keys keys;
2613 if (!skb_flow_dissect(skb, &keys))
2617 if ((__force u16)keys.port16[1] < (__force u16)keys.port16[0])
2618 swap(keys.port16[0], keys.port16[1]);
2622 /* get a consistent hash (same value on both flow directions) */
2623 if ((__force u32)keys.dst < (__force u32)keys.src)
2624 swap(keys.dst, keys.src);
2626 hash = jhash_3words((__force u32)keys.dst,
2627 (__force u32)keys.src,
2628 (__force u32)keys.ports, hashrnd);
2634 EXPORT_SYMBOL(__skb_get_rxhash);
2638 /* One global table that all flow-based protocols share. */
2639 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
2640 EXPORT_SYMBOL(rps_sock_flow_table);
2642 struct static_key rps_needed __read_mostly;
2644 static struct rps_dev_flow *
2645 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2646 struct rps_dev_flow *rflow, u16 next_cpu)
2648 if (next_cpu != RPS_NO_CPU) {
2649 #ifdef CONFIG_RFS_ACCEL
2650 struct netdev_rx_queue *rxqueue;
2651 struct rps_dev_flow_table *flow_table;
2652 struct rps_dev_flow *old_rflow;
2657 /* Should we steer this flow to a different hardware queue? */
2658 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
2659 !(dev->features & NETIF_F_NTUPLE))
2661 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
2662 if (rxq_index == skb_get_rx_queue(skb))
2665 rxqueue = dev->_rx + rxq_index;
2666 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2669 flow_id = skb->rxhash & flow_table->mask;
2670 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
2671 rxq_index, flow_id);
2675 rflow = &flow_table->flows[flow_id];
2677 if (old_rflow->filter == rflow->filter)
2678 old_rflow->filter = RPS_NO_FILTER;
2682 per_cpu(softnet_data, next_cpu).input_queue_head;
2685 rflow->cpu = next_cpu;
2690 * get_rps_cpu is called from netif_receive_skb and returns the target
2691 * CPU from the RPS map of the receiving queue for a given skb.
2692 * rcu_read_lock must be held on entry.
2694 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2695 struct rps_dev_flow **rflowp)
2697 struct netdev_rx_queue *rxqueue;
2698 struct rps_map *map;
2699 struct rps_dev_flow_table *flow_table;
2700 struct rps_sock_flow_table *sock_flow_table;
2704 if (skb_rx_queue_recorded(skb)) {
2705 u16 index = skb_get_rx_queue(skb);
2706 if (unlikely(index >= dev->real_num_rx_queues)) {
2707 WARN_ONCE(dev->real_num_rx_queues > 1,
2708 "%s received packet on queue %u, but number "
2709 "of RX queues is %u\n",
2710 dev->name, index, dev->real_num_rx_queues);
2713 rxqueue = dev->_rx + index;
2717 map = rcu_dereference(rxqueue->rps_map);
2719 if (map->len == 1 &&
2720 !rcu_access_pointer(rxqueue->rps_flow_table)) {
2721 tcpu = map->cpus[0];
2722 if (cpu_online(tcpu))
2726 } else if (!rcu_access_pointer(rxqueue->rps_flow_table)) {
2730 skb_reset_network_header(skb);
2731 if (!skb_get_rxhash(skb))
2734 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2735 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2736 if (flow_table && sock_flow_table) {
2738 struct rps_dev_flow *rflow;
2740 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2743 next_cpu = sock_flow_table->ents[skb->rxhash &
2744 sock_flow_table->mask];
2747 * If the desired CPU (where last recvmsg was done) is
2748 * different from current CPU (one in the rx-queue flow
2749 * table entry), switch if one of the following holds:
2750 * - Current CPU is unset (equal to RPS_NO_CPU).
2751 * - Current CPU is offline.
2752 * - The current CPU's queue tail has advanced beyond the
2753 * last packet that was enqueued using this table entry.
2754 * This guarantees that all previous packets for the flow
2755 * have been dequeued, thus preserving in order delivery.
2757 if (unlikely(tcpu != next_cpu) &&
2758 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2759 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2760 rflow->last_qtail)) >= 0))
2761 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
2763 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2771 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2773 if (cpu_online(tcpu)) {
2783 #ifdef CONFIG_RFS_ACCEL
2786 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2787 * @dev: Device on which the filter was set
2788 * @rxq_index: RX queue index
2789 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2790 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2792 * Drivers that implement ndo_rx_flow_steer() should periodically call
2793 * this function for each installed filter and remove the filters for
2794 * which it returns %true.
2796 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
2797 u32 flow_id, u16 filter_id)
2799 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
2800 struct rps_dev_flow_table *flow_table;
2801 struct rps_dev_flow *rflow;
2806 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2807 if (flow_table && flow_id <= flow_table->mask) {
2808 rflow = &flow_table->flows[flow_id];
2809 cpu = ACCESS_ONCE(rflow->cpu);
2810 if (rflow->filter == filter_id && cpu != RPS_NO_CPU &&
2811 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
2812 rflow->last_qtail) <
2813 (int)(10 * flow_table->mask)))
2819 EXPORT_SYMBOL(rps_may_expire_flow);
2821 #endif /* CONFIG_RFS_ACCEL */
2823 /* Called from hardirq (IPI) context */
2824 static void rps_trigger_softirq(void *data)
2826 struct softnet_data *sd = data;
2828 ____napi_schedule(sd, &sd->backlog);
2832 #endif /* CONFIG_RPS */
2835 * Check if this softnet_data structure is another cpu one
2836 * If yes, queue it to our IPI list and return 1
2839 static int rps_ipi_queued(struct softnet_data *sd)
2842 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2845 sd->rps_ipi_next = mysd->rps_ipi_list;
2846 mysd->rps_ipi_list = sd;
2848 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2851 #endif /* CONFIG_RPS */
2856 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2857 * queue (may be a remote CPU queue).
2859 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2860 unsigned int *qtail)
2862 struct softnet_data *sd;
2863 unsigned long flags;
2865 sd = &per_cpu(softnet_data, cpu);
2867 local_irq_save(flags);
2870 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2871 if (skb_queue_len(&sd->input_pkt_queue)) {
2873 __skb_queue_tail(&sd->input_pkt_queue, skb);
2874 input_queue_tail_incr_save(sd, qtail);
2876 local_irq_restore(flags);
2877 return NET_RX_SUCCESS;
2880 /* Schedule NAPI for backlog device
2881 * We can use non atomic operation since we own the queue lock
2883 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2884 if (!rps_ipi_queued(sd))
2885 ____napi_schedule(sd, &sd->backlog);
2893 local_irq_restore(flags);
2895 atomic_long_inc(&skb->dev->rx_dropped);
2901 * netif_rx - post buffer to the network code
2902 * @skb: buffer to post
2904 * This function receives a packet from a device driver and queues it for
2905 * the upper (protocol) levels to process. It always succeeds. The buffer
2906 * may be dropped during processing for congestion control or by the
2910 * NET_RX_SUCCESS (no congestion)
2911 * NET_RX_DROP (packet was dropped)
2915 int netif_rx(struct sk_buff *skb)
2919 /* if netpoll wants it, pretend we never saw it */
2920 if (netpoll_rx(skb))
2923 net_timestamp_check(netdev_tstamp_prequeue, skb);
2925 trace_netif_rx(skb);
2927 if (static_key_false(&rps_needed)) {
2928 struct rps_dev_flow voidflow, *rflow = &voidflow;
2934 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2936 cpu = smp_processor_id();
2938 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2946 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2951 EXPORT_SYMBOL(netif_rx);
2953 int netif_rx_ni(struct sk_buff *skb)
2958 err = netif_rx(skb);
2959 if (local_softirq_pending())
2965 EXPORT_SYMBOL(netif_rx_ni);
2967 static void net_tx_action(struct softirq_action *h)
2969 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2971 if (sd->completion_queue) {
2972 struct sk_buff *clist;
2974 local_irq_disable();
2975 clist = sd->completion_queue;
2976 sd->completion_queue = NULL;
2980 struct sk_buff *skb = clist;
2981 clist = clist->next;
2983 WARN_ON(atomic_read(&skb->users));
2984 trace_kfree_skb(skb, net_tx_action);
2989 if (sd->output_queue) {
2992 local_irq_disable();
2993 head = sd->output_queue;
2994 sd->output_queue = NULL;
2995 sd->output_queue_tailp = &sd->output_queue;
2999 struct Qdisc *q = head;
3000 spinlock_t *root_lock;
3002 head = head->next_sched;
3004 root_lock = qdisc_lock(q);
3005 if (spin_trylock(root_lock)) {
3006 smp_mb__before_clear_bit();
3007 clear_bit(__QDISC_STATE_SCHED,
3010 spin_unlock(root_lock);
3012 if (!test_bit(__QDISC_STATE_DEACTIVATED,
3014 __netif_reschedule(q);
3016 smp_mb__before_clear_bit();
3017 clear_bit(__QDISC_STATE_SCHED,
3025 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3026 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3027 /* This hook is defined here for ATM LANE */
3028 int (*br_fdb_test_addr_hook)(struct net_device *dev,
3029 unsigned char *addr) __read_mostly;
3030 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
3033 #ifdef CONFIG_NET_CLS_ACT
3034 /* TODO: Maybe we should just force sch_ingress to be compiled in
3035 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3036 * a compare and 2 stores extra right now if we dont have it on
3037 * but have CONFIG_NET_CLS_ACT
3038 * NOTE: This doesn't stop any functionality; if you dont have
3039 * the ingress scheduler, you just can't add policies on ingress.
3042 static int ing_filter(struct sk_buff *skb, struct netdev_queue *rxq)
3044 struct net_device *dev = skb->dev;
3045 u32 ttl = G_TC_RTTL(skb->tc_verd);
3046 int result = TC_ACT_OK;
3049 if (unlikely(MAX_RED_LOOP < ttl++)) {
3050 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3051 skb->skb_iif, dev->ifindex);
3055 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
3056 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
3059 if (q != &noop_qdisc) {
3060 spin_lock(qdisc_lock(q));
3061 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
3062 result = qdisc_enqueue_root(skb, q);
3063 spin_unlock(qdisc_lock(q));
3069 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
3070 struct packet_type **pt_prev,
3071 int *ret, struct net_device *orig_dev)
3073 struct netdev_queue *rxq = rcu_dereference(skb->dev->ingress_queue);
3075 if (!rxq || rxq->qdisc == &noop_qdisc)
3079 *ret = deliver_skb(skb, *pt_prev, orig_dev);
3083 switch (ing_filter(skb, rxq)) {
3097 * netdev_rx_handler_register - register receive handler
3098 * @dev: device to register a handler for
3099 * @rx_handler: receive handler to register
3100 * @rx_handler_data: data pointer that is used by rx handler
3102 * Register a receive hander for a device. This handler will then be
3103 * called from __netif_receive_skb. A negative errno code is returned
3106 * The caller must hold the rtnl_mutex.
3108 * For a general description of rx_handler, see enum rx_handler_result.
3110 int netdev_rx_handler_register(struct net_device *dev,
3111 rx_handler_func_t *rx_handler,
3112 void *rx_handler_data)
3116 if (dev->rx_handler)
3119 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
3120 rcu_assign_pointer(dev->rx_handler, rx_handler);
3124 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
3127 * netdev_rx_handler_unregister - unregister receive handler
3128 * @dev: device to unregister a handler from
3130 * Unregister a receive hander from a device.
3132 * The caller must hold the rtnl_mutex.
3134 void netdev_rx_handler_unregister(struct net_device *dev)
3138 RCU_INIT_POINTER(dev->rx_handler, NULL);
3139 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
3141 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
3143 static int __netif_receive_skb(struct sk_buff *skb)
3145 struct packet_type *ptype, *pt_prev;
3146 rx_handler_func_t *rx_handler;
3147 struct net_device *orig_dev;
3148 struct net_device *null_or_dev;
3149 bool deliver_exact = false;
3150 int ret = NET_RX_DROP;
3153 net_timestamp_check(!netdev_tstamp_prequeue, skb);
3155 trace_netif_receive_skb(skb);
3157 /* if we've gotten here through NAPI, check netpoll */
3158 if (netpoll_receive_skb(skb))
3162 skb->skb_iif = skb->dev->ifindex;
3163 orig_dev = skb->dev;
3165 skb_reset_network_header(skb);
3166 skb_reset_transport_header(skb);
3167 skb_reset_mac_len(skb);
3175 __this_cpu_inc(softnet_data.processed);
3177 if (skb->protocol == cpu_to_be16(ETH_P_8021Q)) {
3178 skb = vlan_untag(skb);
3183 #ifdef CONFIG_NET_CLS_ACT
3184 if (skb->tc_verd & TC_NCLS) {
3185 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
3190 list_for_each_entry_rcu(ptype, &ptype_all, list) {
3191 if (!ptype->dev || ptype->dev == skb->dev) {
3193 ret = deliver_skb(skb, pt_prev, orig_dev);
3198 #ifdef CONFIG_NET_CLS_ACT
3199 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
3205 rx_handler = rcu_dereference(skb->dev->rx_handler);
3206 if (vlan_tx_tag_present(skb)) {
3208 ret = deliver_skb(skb, pt_prev, orig_dev);
3211 if (vlan_do_receive(&skb, !rx_handler))
3213 else if (unlikely(!skb))
3219 ret = deliver_skb(skb, pt_prev, orig_dev);
3222 switch (rx_handler(&skb)) {
3223 case RX_HANDLER_CONSUMED:
3225 case RX_HANDLER_ANOTHER:
3227 case RX_HANDLER_EXACT:
3228 deliver_exact = true;
3229 case RX_HANDLER_PASS:
3236 /* deliver only exact match when indicated */
3237 null_or_dev = deliver_exact ? skb->dev : NULL;
3239 type = skb->protocol;
3240 list_for_each_entry_rcu(ptype,
3241 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
3242 if (ptype->type == type &&
3243 (ptype->dev == null_or_dev || ptype->dev == skb->dev ||
3244 ptype->dev == orig_dev)) {
3246 ret = deliver_skb(skb, pt_prev, orig_dev);
3252 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
3254 atomic_long_inc(&skb->dev->rx_dropped);
3256 /* Jamal, now you will not able to escape explaining
3257 * me how you were going to use this. :-)
3268 * netif_receive_skb - process receive buffer from network
3269 * @skb: buffer to process
3271 * netif_receive_skb() is the main receive data processing function.
3272 * It always succeeds. The buffer may be dropped during processing
3273 * for congestion control or by the protocol layers.
3275 * This function may only be called from softirq context and interrupts
3276 * should be enabled.
3278 * Return values (usually ignored):
3279 * NET_RX_SUCCESS: no congestion
3280 * NET_RX_DROP: packet was dropped
3282 int netif_receive_skb(struct sk_buff *skb)
3284 net_timestamp_check(netdev_tstamp_prequeue, skb);
3286 if (skb_defer_rx_timestamp(skb))
3287 return NET_RX_SUCCESS;
3290 if (static_key_false(&rps_needed)) {
3291 struct rps_dev_flow voidflow, *rflow = &voidflow;
3296 cpu = get_rps_cpu(skb->dev, skb, &rflow);
3299 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
3306 return __netif_receive_skb(skb);
3308 EXPORT_SYMBOL(netif_receive_skb);
3310 /* Network device is going away, flush any packets still pending
3311 * Called with irqs disabled.
3313 static void flush_backlog(void *arg)
3315 struct net_device *dev = arg;
3316 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3317 struct sk_buff *skb, *tmp;
3320 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3321 if (skb->dev == dev) {
3322 __skb_unlink(skb, &sd->input_pkt_queue);
3324 input_queue_head_incr(sd);
3329 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3330 if (skb->dev == dev) {
3331 __skb_unlink(skb, &sd->process_queue);
3333 input_queue_head_incr(sd);
3338 static int napi_gro_complete(struct sk_buff *skb)
3340 struct packet_type *ptype;
3341 __be16 type = skb->protocol;
3342 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3345 if (NAPI_GRO_CB(skb)->count == 1) {
3346 skb_shinfo(skb)->gso_size = 0;
3351 list_for_each_entry_rcu(ptype, head, list) {
3352 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3355 err = ptype->gro_complete(skb);
3361 WARN_ON(&ptype->list == head);
3363 return NET_RX_SUCCESS;
3367 return netif_receive_skb(skb);
3370 inline void napi_gro_flush(struct napi_struct *napi)
3372 struct sk_buff *skb, *next;
3374 for (skb = napi->gro_list; skb; skb = next) {
3377 napi_gro_complete(skb);
3380 napi->gro_count = 0;
3381 napi->gro_list = NULL;
3383 EXPORT_SYMBOL(napi_gro_flush);
3385 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3387 struct sk_buff **pp = NULL;
3388 struct packet_type *ptype;
3389 __be16 type = skb->protocol;
3390 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3393 enum gro_result ret;
3395 if (!(skb->dev->features & NETIF_F_GRO) || netpoll_rx_on(skb))
3398 if (skb_is_gso(skb) || skb_has_frag_list(skb))
3402 list_for_each_entry_rcu(ptype, head, list) {
3403 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3406 skb_set_network_header(skb, skb_gro_offset(skb));
3407 mac_len = skb->network_header - skb->mac_header;
3408 skb->mac_len = mac_len;
3409 NAPI_GRO_CB(skb)->same_flow = 0;
3410 NAPI_GRO_CB(skb)->flush = 0;
3411 NAPI_GRO_CB(skb)->free = 0;
3413 pp = ptype->gro_receive(&napi->gro_list, skb);
3418 if (&ptype->list == head)
3421 same_flow = NAPI_GRO_CB(skb)->same_flow;
3422 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3425 struct sk_buff *nskb = *pp;
3429 napi_gro_complete(nskb);
3436 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3440 NAPI_GRO_CB(skb)->count = 1;
3441 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3442 skb->next = napi->gro_list;
3443 napi->gro_list = skb;
3447 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3448 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3450 BUG_ON(skb->end - skb->tail < grow);
3452 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3455 skb->data_len -= grow;
3457 skb_shinfo(skb)->frags[0].page_offset += grow;
3458 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], grow);
3460 if (unlikely(!skb_frag_size(&skb_shinfo(skb)->frags[0]))) {
3461 skb_frag_unref(skb, 0);
3462 memmove(skb_shinfo(skb)->frags,
3463 skb_shinfo(skb)->frags + 1,
3464 --skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t));
3475 EXPORT_SYMBOL(dev_gro_receive);
3477 static inline gro_result_t
3478 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3481 unsigned int maclen = skb->dev->hard_header_len;
3483 for (p = napi->gro_list; p; p = p->next) {
3484 unsigned long diffs;
3486 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
3487 diffs |= p->vlan_tci ^ skb->vlan_tci;
3488 if (maclen == ETH_HLEN)
3489 diffs |= compare_ether_header(skb_mac_header(p),
3490 skb_gro_mac_header(skb));
3492 diffs = memcmp(skb_mac_header(p),
3493 skb_gro_mac_header(skb),
3495 NAPI_GRO_CB(p)->same_flow = !diffs;
3496 NAPI_GRO_CB(p)->flush = 0;
3499 return dev_gro_receive(napi, skb);
3502 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3506 if (netif_receive_skb(skb))
3514 case GRO_MERGED_FREE:
3515 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
3516 kmem_cache_free(skbuff_head_cache, skb);
3528 EXPORT_SYMBOL(napi_skb_finish);
3530 void skb_gro_reset_offset(struct sk_buff *skb)
3532 NAPI_GRO_CB(skb)->data_offset = 0;
3533 NAPI_GRO_CB(skb)->frag0 = NULL;
3534 NAPI_GRO_CB(skb)->frag0_len = 0;
3536 if (skb->mac_header == skb->tail &&
3537 !PageHighMem(skb_frag_page(&skb_shinfo(skb)->frags[0]))) {
3538 NAPI_GRO_CB(skb)->frag0 =
3539 skb_frag_address(&skb_shinfo(skb)->frags[0]);
3540 NAPI_GRO_CB(skb)->frag0_len = skb_frag_size(&skb_shinfo(skb)->frags[0]);
3543 EXPORT_SYMBOL(skb_gro_reset_offset);
3545 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3547 skb_gro_reset_offset(skb);
3549 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3551 EXPORT_SYMBOL(napi_gro_receive);
3553 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3555 __skb_pull(skb, skb_headlen(skb));
3556 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3557 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
3559 skb->dev = napi->dev;
3565 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3567 struct sk_buff *skb = napi->skb;
3570 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3576 EXPORT_SYMBOL(napi_get_frags);
3578 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3584 skb->protocol = eth_type_trans(skb, skb->dev);
3586 if (ret == GRO_HELD)
3587 skb_gro_pull(skb, -ETH_HLEN);
3588 else if (netif_receive_skb(skb))
3593 case GRO_MERGED_FREE:
3594 napi_reuse_skb(napi, skb);
3603 EXPORT_SYMBOL(napi_frags_finish);
3605 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3607 struct sk_buff *skb = napi->skb;
3614 skb_reset_mac_header(skb);
3615 skb_gro_reset_offset(skb);
3617 off = skb_gro_offset(skb);
3618 hlen = off + sizeof(*eth);
3619 eth = skb_gro_header_fast(skb, off);
3620 if (skb_gro_header_hard(skb, hlen)) {
3621 eth = skb_gro_header_slow(skb, hlen, off);
3622 if (unlikely(!eth)) {
3623 napi_reuse_skb(napi, skb);
3629 skb_gro_pull(skb, sizeof(*eth));
3632 * This works because the only protocols we care about don't require
3633 * special handling. We'll fix it up properly at the end.
3635 skb->protocol = eth->h_proto;
3640 EXPORT_SYMBOL(napi_frags_skb);
3642 gro_result_t napi_gro_frags(struct napi_struct *napi)
3644 struct sk_buff *skb = napi_frags_skb(napi);
3649 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3651 EXPORT_SYMBOL(napi_gro_frags);
3654 * net_rps_action sends any pending IPI's for rps.
3655 * Note: called with local irq disabled, but exits with local irq enabled.
3657 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3660 struct softnet_data *remsd = sd->rps_ipi_list;
3663 sd->rps_ipi_list = NULL;
3667 /* Send pending IPI's to kick RPS processing on remote cpus. */
3669 struct softnet_data *next = remsd->rps_ipi_next;
3671 if (cpu_online(remsd->cpu))
3672 __smp_call_function_single(remsd->cpu,
3681 static int process_backlog(struct napi_struct *napi, int quota)
3684 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3687 /* Check if we have pending ipi, its better to send them now,
3688 * not waiting net_rx_action() end.
3690 if (sd->rps_ipi_list) {
3691 local_irq_disable();
3692 net_rps_action_and_irq_enable(sd);
3695 napi->weight = weight_p;
3696 local_irq_disable();
3697 while (work < quota) {
3698 struct sk_buff *skb;
3701 while ((skb = __skb_dequeue(&sd->process_queue))) {
3703 __netif_receive_skb(skb);
3704 local_irq_disable();
3705 input_queue_head_incr(sd);
3706 if (++work >= quota) {
3713 qlen = skb_queue_len(&sd->input_pkt_queue);
3715 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3716 &sd->process_queue);
3718 if (qlen < quota - work) {
3720 * Inline a custom version of __napi_complete().
3721 * only current cpu owns and manipulates this napi,
3722 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3723 * we can use a plain write instead of clear_bit(),
3724 * and we dont need an smp_mb() memory barrier.
3726 list_del(&napi->poll_list);
3729 quota = work + qlen;
3739 * __napi_schedule - schedule for receive
3740 * @n: entry to schedule
3742 * The entry's receive function will be scheduled to run
3744 void __napi_schedule(struct napi_struct *n)
3746 unsigned long flags;
3748 local_irq_save(flags);
3749 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3750 local_irq_restore(flags);
3752 EXPORT_SYMBOL(__napi_schedule);
3754 void __napi_complete(struct napi_struct *n)
3756 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3757 BUG_ON(n->gro_list);
3759 list_del(&n->poll_list);
3760 smp_mb__before_clear_bit();
3761 clear_bit(NAPI_STATE_SCHED, &n->state);
3763 EXPORT_SYMBOL(__napi_complete);
3765 void napi_complete(struct napi_struct *n)
3767 unsigned long flags;
3770 * don't let napi dequeue from the cpu poll list
3771 * just in case its running on a different cpu
3773 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3777 local_irq_save(flags);
3779 local_irq_restore(flags);
3781 EXPORT_SYMBOL(napi_complete);
3783 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3784 int (*poll)(struct napi_struct *, int), int weight)
3786 INIT_LIST_HEAD(&napi->poll_list);
3787 napi->gro_count = 0;
3788 napi->gro_list = NULL;
3791 napi->weight = weight;
3792 list_add(&napi->dev_list, &dev->napi_list);
3794 #ifdef CONFIG_NETPOLL
3795 spin_lock_init(&napi->poll_lock);
3796 napi->poll_owner = -1;
3798 set_bit(NAPI_STATE_SCHED, &napi->state);
3800 EXPORT_SYMBOL(netif_napi_add);
3802 void netif_napi_del(struct napi_struct *napi)
3804 struct sk_buff *skb, *next;
3806 list_del_init(&napi->dev_list);
3807 napi_free_frags(napi);
3809 for (skb = napi->gro_list; skb; skb = next) {
3815 napi->gro_list = NULL;
3816 napi->gro_count = 0;
3818 EXPORT_SYMBOL(netif_napi_del);
3820 static void net_rx_action(struct softirq_action *h)
3822 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3823 unsigned long time_limit = jiffies + 2;
3824 int budget = netdev_budget;
3827 local_irq_disable();
3829 while (!list_empty(&sd->poll_list)) {
3830 struct napi_struct *n;
3833 /* If softirq window is exhuasted then punt.
3834 * Allow this to run for 2 jiffies since which will allow
3835 * an average latency of 1.5/HZ.
3837 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3842 /* Even though interrupts have been re-enabled, this
3843 * access is safe because interrupts can only add new
3844 * entries to the tail of this list, and only ->poll()
3845 * calls can remove this head entry from the list.
3847 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3849 have = netpoll_poll_lock(n);
3853 /* This NAPI_STATE_SCHED test is for avoiding a race
3854 * with netpoll's poll_napi(). Only the entity which
3855 * obtains the lock and sees NAPI_STATE_SCHED set will
3856 * actually make the ->poll() call. Therefore we avoid
3857 * accidentally calling ->poll() when NAPI is not scheduled.
3860 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3861 work = n->poll(n, weight);
3865 WARN_ON_ONCE(work > weight);
3869 local_irq_disable();
3871 /* Drivers must not modify the NAPI state if they
3872 * consume the entire weight. In such cases this code
3873 * still "owns" the NAPI instance and therefore can
3874 * move the instance around on the list at-will.
3876 if (unlikely(work == weight)) {
3877 if (unlikely(napi_disable_pending(n))) {
3880 local_irq_disable();
3882 list_move_tail(&n->poll_list, &sd->poll_list);
3885 netpoll_poll_unlock(have);
3888 net_rps_action_and_irq_enable(sd);
3890 #ifdef CONFIG_NET_DMA
3892 * There may not be any more sk_buffs coming right now, so push
3893 * any pending DMA copies to hardware
3895 dma_issue_pending_all();
3902 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3906 static gifconf_func_t *gifconf_list[NPROTO];
3909 * register_gifconf - register a SIOCGIF handler
3910 * @family: Address family
3911 * @gifconf: Function handler
3913 * Register protocol dependent address dumping routines. The handler
3914 * that is passed must not be freed or reused until it has been replaced
3915 * by another handler.
3917 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3919 if (family >= NPROTO)
3921 gifconf_list[family] = gifconf;
3924 EXPORT_SYMBOL(register_gifconf);
3928 * Map an interface index to its name (SIOCGIFNAME)
3932 * We need this ioctl for efficient implementation of the
3933 * if_indextoname() function required by the IPv6 API. Without
3934 * it, we would have to search all the interfaces to find a
3938 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3940 struct net_device *dev;
3944 * Fetch the caller's info block.
3947 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3951 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3957 strcpy(ifr.ifr_name, dev->name);
3960 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3966 * Perform a SIOCGIFCONF call. This structure will change
3967 * size eventually, and there is nothing I can do about it.
3968 * Thus we will need a 'compatibility mode'.
3971 static int dev_ifconf(struct net *net, char __user *arg)
3974 struct net_device *dev;
3981 * Fetch the caller's info block.
3984 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3991 * Loop over the interfaces, and write an info block for each.
3995 for_each_netdev(net, dev) {
3996 for (i = 0; i < NPROTO; i++) {
3997 if (gifconf_list[i]) {
4000 done = gifconf_list[i](dev, NULL, 0);
4002 done = gifconf_list[i](dev, pos + total,
4012 * All done. Write the updated control block back to the caller.
4014 ifc.ifc_len = total;
4017 * Both BSD and Solaris return 0 here, so we do too.
4019 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
4022 #ifdef CONFIG_PROC_FS
4024 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4026 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4027 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4028 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4030 static inline struct net_device *dev_from_same_bucket(struct seq_file *seq, loff_t *pos)
4032 struct net *net = seq_file_net(seq);
4033 struct net_device *dev;
4034 struct hlist_node *p;
4035 struct hlist_head *h;
4036 unsigned int count = 0, offset = get_offset(*pos);
4038 h = &net->dev_name_head[get_bucket(*pos)];
4039 hlist_for_each_entry_rcu(dev, p, h, name_hlist) {
4040 if (++count == offset)
4047 static inline struct net_device *dev_from_bucket(struct seq_file *seq, loff_t *pos)
4049 struct net_device *dev;
4050 unsigned int bucket;
4053 dev = dev_from_same_bucket(seq, pos);
4057 bucket = get_bucket(*pos) + 1;
4058 *pos = set_bucket_offset(bucket, 1);
4059 } while (bucket < NETDEV_HASHENTRIES);
4065 * This is invoked by the /proc filesystem handler to display a device
4068 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
4073 return SEQ_START_TOKEN;
4075 if (get_bucket(*pos) >= NETDEV_HASHENTRIES)
4078 return dev_from_bucket(seq, pos);
4081 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4084 return dev_from_bucket(seq, pos);
4087 void dev_seq_stop(struct seq_file *seq, void *v)
4093 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
4095 struct rtnl_link_stats64 temp;
4096 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
4098 seq_printf(seq, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4099 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4100 dev->name, stats->rx_bytes, stats->rx_packets,
4102 stats->rx_dropped + stats->rx_missed_errors,
4103 stats->rx_fifo_errors,
4104 stats->rx_length_errors + stats->rx_over_errors +
4105 stats->rx_crc_errors + stats->rx_frame_errors,
4106 stats->rx_compressed, stats->multicast,
4107 stats->tx_bytes, stats->tx_packets,
4108 stats->tx_errors, stats->tx_dropped,
4109 stats->tx_fifo_errors, stats->collisions,
4110 stats->tx_carrier_errors +
4111 stats->tx_aborted_errors +
4112 stats->tx_window_errors +
4113 stats->tx_heartbeat_errors,
4114 stats->tx_compressed);
4118 * Called from the PROCfs module. This now uses the new arbitrary sized
4119 * /proc/net interface to create /proc/net/dev
4121 static int dev_seq_show(struct seq_file *seq, void *v)
4123 if (v == SEQ_START_TOKEN)
4124 seq_puts(seq, "Inter-| Receive "
4126 " face |bytes packets errs drop fifo frame "
4127 "compressed multicast|bytes packets errs "
4128 "drop fifo colls carrier compressed\n");
4130 dev_seq_printf_stats(seq, v);
4134 static struct softnet_data *softnet_get_online(loff_t *pos)
4136 struct softnet_data *sd = NULL;
4138 while (*pos < nr_cpu_ids)
4139 if (cpu_online(*pos)) {
4140 sd = &per_cpu(softnet_data, *pos);
4147 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
4149 return softnet_get_online(pos);
4152 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4155 return softnet_get_online(pos);
4158 static void softnet_seq_stop(struct seq_file *seq, void *v)
4162 static int softnet_seq_show(struct seq_file *seq, void *v)
4164 struct softnet_data *sd = v;
4166 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4167 sd->processed, sd->dropped, sd->time_squeeze, 0,
4168 0, 0, 0, 0, /* was fastroute */
4169 sd->cpu_collision, sd->received_rps);
4173 static const struct seq_operations dev_seq_ops = {
4174 .start = dev_seq_start,
4175 .next = dev_seq_next,
4176 .stop = dev_seq_stop,
4177 .show = dev_seq_show,
4180 static int dev_seq_open(struct inode *inode, struct file *file)
4182 return seq_open_net(inode, file, &dev_seq_ops,
4183 sizeof(struct seq_net_private));
4186 static const struct file_operations dev_seq_fops = {
4187 .owner = THIS_MODULE,
4188 .open = dev_seq_open,
4190 .llseek = seq_lseek,
4191 .release = seq_release_net,
4194 static const struct seq_operations softnet_seq_ops = {
4195 .start = softnet_seq_start,
4196 .next = softnet_seq_next,
4197 .stop = softnet_seq_stop,
4198 .show = softnet_seq_show,
4201 static int softnet_seq_open(struct inode *inode, struct file *file)
4203 return seq_open(file, &softnet_seq_ops);
4206 static const struct file_operations softnet_seq_fops = {
4207 .owner = THIS_MODULE,
4208 .open = softnet_seq_open,
4210 .llseek = seq_lseek,
4211 .release = seq_release,
4214 static void *ptype_get_idx(loff_t pos)
4216 struct packet_type *pt = NULL;
4220 list_for_each_entry_rcu(pt, &ptype_all, list) {
4226 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
4227 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
4236 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
4240 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
4243 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4245 struct packet_type *pt;
4246 struct list_head *nxt;
4250 if (v == SEQ_START_TOKEN)
4251 return ptype_get_idx(0);
4254 nxt = pt->list.next;
4255 if (pt->type == htons(ETH_P_ALL)) {
4256 if (nxt != &ptype_all)
4259 nxt = ptype_base[0].next;
4261 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
4263 while (nxt == &ptype_base[hash]) {
4264 if (++hash >= PTYPE_HASH_SIZE)
4266 nxt = ptype_base[hash].next;
4269 return list_entry(nxt, struct packet_type, list);
4272 static void ptype_seq_stop(struct seq_file *seq, void *v)
4278 static int ptype_seq_show(struct seq_file *seq, void *v)
4280 struct packet_type *pt = v;
4282 if (v == SEQ_START_TOKEN)
4283 seq_puts(seq, "Type Device Function\n");
4284 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
4285 if (pt->type == htons(ETH_P_ALL))
4286 seq_puts(seq, "ALL ");
4288 seq_printf(seq, "%04x", ntohs(pt->type));
4290 seq_printf(seq, " %-8s %pF\n",
4291 pt->dev ? pt->dev->name : "", pt->func);
4297 static const struct seq_operations ptype_seq_ops = {
4298 .start = ptype_seq_start,
4299 .next = ptype_seq_next,
4300 .stop = ptype_seq_stop,
4301 .show = ptype_seq_show,
4304 static int ptype_seq_open(struct inode *inode, struct file *file)
4306 return seq_open_net(inode, file, &ptype_seq_ops,
4307 sizeof(struct seq_net_private));
4310 static const struct file_operations ptype_seq_fops = {
4311 .owner = THIS_MODULE,
4312 .open = ptype_seq_open,
4314 .llseek = seq_lseek,
4315 .release = seq_release_net,
4319 static int __net_init dev_proc_net_init(struct net *net)
4323 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
4325 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
4327 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
4330 if (wext_proc_init(net))
4336 proc_net_remove(net, "ptype");
4338 proc_net_remove(net, "softnet_stat");
4340 proc_net_remove(net, "dev");
4344 static void __net_exit dev_proc_net_exit(struct net *net)
4346 wext_proc_exit(net);
4348 proc_net_remove(net, "ptype");
4349 proc_net_remove(net, "softnet_stat");
4350 proc_net_remove(net, "dev");
4353 static struct pernet_operations __net_initdata dev_proc_ops = {
4354 .init = dev_proc_net_init,
4355 .exit = dev_proc_net_exit,
4358 static int __init dev_proc_init(void)
4360 return register_pernet_subsys(&dev_proc_ops);
4363 #define dev_proc_init() 0
4364 #endif /* CONFIG_PROC_FS */
4368 * netdev_set_master - set up master pointer
4369 * @slave: slave device
4370 * @master: new master device
4372 * Changes the master device of the slave. Pass %NULL to break the
4373 * bonding. The caller must hold the RTNL semaphore. On a failure
4374 * a negative errno code is returned. On success the reference counts
4375 * are adjusted and the function returns zero.
4377 int netdev_set_master(struct net_device *slave, struct net_device *master)
4379 struct net_device *old = slave->master;
4389 slave->master = master;
4395 EXPORT_SYMBOL(netdev_set_master);
4398 * netdev_set_bond_master - set up bonding master/slave pair
4399 * @slave: slave device
4400 * @master: new master device
4402 * Changes the master device of the slave. Pass %NULL to break the
4403 * bonding. The caller must hold the RTNL semaphore. On a failure
4404 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4405 * to the routing socket and the function returns zero.
4407 int netdev_set_bond_master(struct net_device *slave, struct net_device *master)
4413 err = netdev_set_master(slave, master);
4417 slave->flags |= IFF_SLAVE;
4419 slave->flags &= ~IFF_SLAVE;
4421 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4424 EXPORT_SYMBOL(netdev_set_bond_master);
4426 static void dev_change_rx_flags(struct net_device *dev, int flags)
4428 const struct net_device_ops *ops = dev->netdev_ops;
4430 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4431 ops->ndo_change_rx_flags(dev, flags);
4434 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4436 unsigned int old_flags = dev->flags;
4442 dev->flags |= IFF_PROMISC;
4443 dev->promiscuity += inc;
4444 if (dev->promiscuity == 0) {
4447 * If inc causes overflow, untouch promisc and return error.
4450 dev->flags &= ~IFF_PROMISC;
4452 dev->promiscuity -= inc;
4453 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4458 if (dev->flags != old_flags) {
4459 pr_info("device %s %s promiscuous mode\n",
4461 dev->flags & IFF_PROMISC ? "entered" : "left");
4462 if (audit_enabled) {
4463 current_uid_gid(&uid, &gid);
4464 audit_log(current->audit_context, GFP_ATOMIC,
4465 AUDIT_ANOM_PROMISCUOUS,
4466 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4467 dev->name, (dev->flags & IFF_PROMISC),
4468 (old_flags & IFF_PROMISC),
4469 audit_get_loginuid(current),
4471 audit_get_sessionid(current));
4474 dev_change_rx_flags(dev, IFF_PROMISC);
4480 * dev_set_promiscuity - update promiscuity count on a device
4484 * Add or remove promiscuity from a device. While the count in the device
4485 * remains above zero the interface remains promiscuous. Once it hits zero
4486 * the device reverts back to normal filtering operation. A negative inc
4487 * value is used to drop promiscuity on the device.
4488 * Return 0 if successful or a negative errno code on error.
4490 int dev_set_promiscuity(struct net_device *dev, int inc)
4492 unsigned int old_flags = dev->flags;
4495 err = __dev_set_promiscuity(dev, inc);
4498 if (dev->flags != old_flags)
4499 dev_set_rx_mode(dev);
4502 EXPORT_SYMBOL(dev_set_promiscuity);
4505 * dev_set_allmulti - update allmulti count on a device
4509 * Add or remove reception of all multicast frames to a device. While the
4510 * count in the device remains above zero the interface remains listening
4511 * to all interfaces. Once it hits zero the device reverts back to normal
4512 * filtering operation. A negative @inc value is used to drop the counter
4513 * when releasing a resource needing all multicasts.
4514 * Return 0 if successful or a negative errno code on error.
4517 int dev_set_allmulti(struct net_device *dev, int inc)
4519 unsigned int old_flags = dev->flags;
4523 dev->flags |= IFF_ALLMULTI;
4524 dev->allmulti += inc;
4525 if (dev->allmulti == 0) {
4528 * If inc causes overflow, untouch allmulti and return error.
4531 dev->flags &= ~IFF_ALLMULTI;
4533 dev->allmulti -= inc;
4534 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4539 if (dev->flags ^ old_flags) {
4540 dev_change_rx_flags(dev, IFF_ALLMULTI);
4541 dev_set_rx_mode(dev);
4545 EXPORT_SYMBOL(dev_set_allmulti);
4548 * Upload unicast and multicast address lists to device and
4549 * configure RX filtering. When the device doesn't support unicast
4550 * filtering it is put in promiscuous mode while unicast addresses
4553 void __dev_set_rx_mode(struct net_device *dev)
4555 const struct net_device_ops *ops = dev->netdev_ops;
4557 /* dev_open will call this function so the list will stay sane. */
4558 if (!(dev->flags&IFF_UP))
4561 if (!netif_device_present(dev))
4564 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
4565 /* Unicast addresses changes may only happen under the rtnl,
4566 * therefore calling __dev_set_promiscuity here is safe.
4568 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4569 __dev_set_promiscuity(dev, 1);
4570 dev->uc_promisc = true;
4571 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4572 __dev_set_promiscuity(dev, -1);
4573 dev->uc_promisc = false;
4577 if (ops->ndo_set_rx_mode)
4578 ops->ndo_set_rx_mode(dev);
4581 void dev_set_rx_mode(struct net_device *dev)
4583 netif_addr_lock_bh(dev);
4584 __dev_set_rx_mode(dev);
4585 netif_addr_unlock_bh(dev);
4589 * dev_get_flags - get flags reported to userspace
4592 * Get the combination of flag bits exported through APIs to userspace.
4594 unsigned int dev_get_flags(const struct net_device *dev)
4598 flags = (dev->flags & ~(IFF_PROMISC |
4603 (dev->gflags & (IFF_PROMISC |
4606 if (netif_running(dev)) {
4607 if (netif_oper_up(dev))
4608 flags |= IFF_RUNNING;
4609 if (netif_carrier_ok(dev))
4610 flags |= IFF_LOWER_UP;
4611 if (netif_dormant(dev))
4612 flags |= IFF_DORMANT;
4617 EXPORT_SYMBOL(dev_get_flags);
4619 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4621 unsigned int old_flags = dev->flags;
4627 * Set the flags on our device.
4630 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4631 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4633 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4637 * Load in the correct multicast list now the flags have changed.
4640 if ((old_flags ^ flags) & IFF_MULTICAST)
4641 dev_change_rx_flags(dev, IFF_MULTICAST);
4643 dev_set_rx_mode(dev);
4646 * Have we downed the interface. We handle IFF_UP ourselves
4647 * according to user attempts to set it, rather than blindly
4652 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4653 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4656 dev_set_rx_mode(dev);
4659 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4660 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4662 dev->gflags ^= IFF_PROMISC;
4663 dev_set_promiscuity(dev, inc);
4666 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4667 is important. Some (broken) drivers set IFF_PROMISC, when
4668 IFF_ALLMULTI is requested not asking us and not reporting.
4670 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4671 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4673 dev->gflags ^= IFF_ALLMULTI;
4674 dev_set_allmulti(dev, inc);
4680 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4682 unsigned int changes = dev->flags ^ old_flags;
4684 if (changes & IFF_UP) {
4685 if (dev->flags & IFF_UP)
4686 call_netdevice_notifiers(NETDEV_UP, dev);
4688 call_netdevice_notifiers(NETDEV_DOWN, dev);
4691 if (dev->flags & IFF_UP &&
4692 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4693 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4697 * dev_change_flags - change device settings
4699 * @flags: device state flags
4701 * Change settings on device based state flags. The flags are
4702 * in the userspace exported format.
4704 int dev_change_flags(struct net_device *dev, unsigned int flags)
4707 unsigned int changes, old_flags = dev->flags;
4709 ret = __dev_change_flags(dev, flags);
4713 changes = old_flags ^ dev->flags;
4715 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4717 __dev_notify_flags(dev, old_flags);
4720 EXPORT_SYMBOL(dev_change_flags);
4723 * dev_set_mtu - Change maximum transfer unit
4725 * @new_mtu: new transfer unit
4727 * Change the maximum transfer size of the network device.
4729 int dev_set_mtu(struct net_device *dev, int new_mtu)
4731 const struct net_device_ops *ops = dev->netdev_ops;
4734 if (new_mtu == dev->mtu)
4737 /* MTU must be positive. */
4741 if (!netif_device_present(dev))
4745 if (ops->ndo_change_mtu)
4746 err = ops->ndo_change_mtu(dev, new_mtu);
4750 if (!err && dev->flags & IFF_UP)
4751 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4754 EXPORT_SYMBOL(dev_set_mtu);
4757 * dev_set_group - Change group this device belongs to
4759 * @new_group: group this device should belong to
4761 void dev_set_group(struct net_device *dev, int new_group)
4763 dev->group = new_group;
4765 EXPORT_SYMBOL(dev_set_group);
4768 * dev_set_mac_address - Change Media Access Control Address
4772 * Change the hardware (MAC) address of the device
4774 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4776 const struct net_device_ops *ops = dev->netdev_ops;
4779 if (!ops->ndo_set_mac_address)
4781 if (sa->sa_family != dev->type)
4783 if (!netif_device_present(dev))
4785 err = ops->ndo_set_mac_address(dev, sa);
4787 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4790 EXPORT_SYMBOL(dev_set_mac_address);
4793 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4795 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4798 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4804 case SIOCGIFFLAGS: /* Get interface flags */
4805 ifr->ifr_flags = (short) dev_get_flags(dev);
4808 case SIOCGIFMETRIC: /* Get the metric on the interface
4809 (currently unused) */
4810 ifr->ifr_metric = 0;
4813 case SIOCGIFMTU: /* Get the MTU of a device */
4814 ifr->ifr_mtu = dev->mtu;
4819 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4821 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4822 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4823 ifr->ifr_hwaddr.sa_family = dev->type;
4831 ifr->ifr_map.mem_start = dev->mem_start;
4832 ifr->ifr_map.mem_end = dev->mem_end;
4833 ifr->ifr_map.base_addr = dev->base_addr;
4834 ifr->ifr_map.irq = dev->irq;
4835 ifr->ifr_map.dma = dev->dma;
4836 ifr->ifr_map.port = dev->if_port;
4840 ifr->ifr_ifindex = dev->ifindex;
4844 ifr->ifr_qlen = dev->tx_queue_len;
4848 /* dev_ioctl() should ensure this case
4860 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4862 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4865 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4866 const struct net_device_ops *ops;
4871 ops = dev->netdev_ops;
4874 case SIOCSIFFLAGS: /* Set interface flags */
4875 return dev_change_flags(dev, ifr->ifr_flags);
4877 case SIOCSIFMETRIC: /* Set the metric on the interface
4878 (currently unused) */
4881 case SIOCSIFMTU: /* Set the MTU of a device */
4882 return dev_set_mtu(dev, ifr->ifr_mtu);
4885 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4887 case SIOCSIFHWBROADCAST:
4888 if (ifr->ifr_hwaddr.sa_family != dev->type)
4890 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4891 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4892 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4896 if (ops->ndo_set_config) {
4897 if (!netif_device_present(dev))
4899 return ops->ndo_set_config(dev, &ifr->ifr_map);
4904 if (!ops->ndo_set_rx_mode ||
4905 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4907 if (!netif_device_present(dev))
4909 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4912 if (!ops->ndo_set_rx_mode ||
4913 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4915 if (!netif_device_present(dev))
4917 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4920 if (ifr->ifr_qlen < 0)
4922 dev->tx_queue_len = ifr->ifr_qlen;
4926 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4927 return dev_change_name(dev, ifr->ifr_newname);
4930 err = net_hwtstamp_validate(ifr);
4936 * Unknown or private ioctl
4939 if ((cmd >= SIOCDEVPRIVATE &&
4940 cmd <= SIOCDEVPRIVATE + 15) ||
4941 cmd == SIOCBONDENSLAVE ||
4942 cmd == SIOCBONDRELEASE ||
4943 cmd == SIOCBONDSETHWADDR ||
4944 cmd == SIOCBONDSLAVEINFOQUERY ||
4945 cmd == SIOCBONDINFOQUERY ||
4946 cmd == SIOCBONDCHANGEACTIVE ||
4947 cmd == SIOCGMIIPHY ||
4948 cmd == SIOCGMIIREG ||
4949 cmd == SIOCSMIIREG ||
4950 cmd == SIOCBRADDIF ||
4951 cmd == SIOCBRDELIF ||
4952 cmd == SIOCSHWTSTAMP ||
4953 cmd == SIOCWANDEV) {
4955 if (ops->ndo_do_ioctl) {
4956 if (netif_device_present(dev))
4957 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4969 * This function handles all "interface"-type I/O control requests. The actual
4970 * 'doing' part of this is dev_ifsioc above.
4974 * dev_ioctl - network device ioctl
4975 * @net: the applicable net namespace
4976 * @cmd: command to issue
4977 * @arg: pointer to a struct ifreq in user space
4979 * Issue ioctl functions to devices. This is normally called by the
4980 * user space syscall interfaces but can sometimes be useful for
4981 * other purposes. The return value is the return from the syscall if
4982 * positive or a negative errno code on error.
4985 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4991 /* One special case: SIOCGIFCONF takes ifconf argument
4992 and requires shared lock, because it sleeps writing
4996 if (cmd == SIOCGIFCONF) {
4998 ret = dev_ifconf(net, (char __user *) arg);
5002 if (cmd == SIOCGIFNAME)
5003 return dev_ifname(net, (struct ifreq __user *)arg);
5005 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
5008 ifr.ifr_name[IFNAMSIZ-1] = 0;
5010 colon = strchr(ifr.ifr_name, ':');
5015 * See which interface the caller is talking about.
5020 * These ioctl calls:
5021 * - can be done by all.
5022 * - atomic and do not require locking.
5033 dev_load(net, ifr.ifr_name);
5035 ret = dev_ifsioc_locked(net, &ifr, cmd);
5040 if (copy_to_user(arg, &ifr,
5041 sizeof(struct ifreq)))
5047 dev_load(net, ifr.ifr_name);
5049 ret = dev_ethtool(net, &ifr);
5054 if (copy_to_user(arg, &ifr,
5055 sizeof(struct ifreq)))
5061 * These ioctl calls:
5062 * - require superuser power.
5063 * - require strict serialization.
5069 if (!capable(CAP_NET_ADMIN))
5071 dev_load(net, ifr.ifr_name);
5073 ret = dev_ifsioc(net, &ifr, cmd);
5078 if (copy_to_user(arg, &ifr,
5079 sizeof(struct ifreq)))
5085 * These ioctl calls:
5086 * - require superuser power.
5087 * - require strict serialization.
5088 * - do not return a value
5098 case SIOCSIFHWBROADCAST:
5101 case SIOCBONDENSLAVE:
5102 case SIOCBONDRELEASE:
5103 case SIOCBONDSETHWADDR:
5104 case SIOCBONDCHANGEACTIVE:
5108 if (!capable(CAP_NET_ADMIN))
5111 case SIOCBONDSLAVEINFOQUERY:
5112 case SIOCBONDINFOQUERY:
5113 dev_load(net, ifr.ifr_name);
5115 ret = dev_ifsioc(net, &ifr, cmd);
5120 /* Get the per device memory space. We can add this but
5121 * currently do not support it */
5123 /* Set the per device memory buffer space.
5124 * Not applicable in our case */
5129 * Unknown or private ioctl.
5132 if (cmd == SIOCWANDEV ||
5133 (cmd >= SIOCDEVPRIVATE &&
5134 cmd <= SIOCDEVPRIVATE + 15)) {
5135 dev_load(net, ifr.ifr_name);
5137 ret = dev_ifsioc(net, &ifr, cmd);
5139 if (!ret && copy_to_user(arg, &ifr,
5140 sizeof(struct ifreq)))
5144 /* Take care of Wireless Extensions */
5145 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
5146 return wext_handle_ioctl(net, &ifr, cmd, arg);
5153 * dev_new_index - allocate an ifindex
5154 * @net: the applicable net namespace
5156 * Returns a suitable unique value for a new device interface
5157 * number. The caller must hold the rtnl semaphore or the
5158 * dev_base_lock to be sure it remains unique.
5160 static int dev_new_index(struct net *net)
5166 if (!__dev_get_by_index(net, ifindex))
5171 /* Delayed registration/unregisteration */
5172 static LIST_HEAD(net_todo_list);
5174 static void net_set_todo(struct net_device *dev)
5176 list_add_tail(&dev->todo_list, &net_todo_list);
5179 static void rollback_registered_many(struct list_head *head)
5181 struct net_device *dev, *tmp;
5183 BUG_ON(dev_boot_phase);
5186 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
5187 /* Some devices call without registering
5188 * for initialization unwind. Remove those
5189 * devices and proceed with the remaining.
5191 if (dev->reg_state == NETREG_UNINITIALIZED) {
5192 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5196 list_del(&dev->unreg_list);
5199 dev->dismantle = true;
5200 BUG_ON(dev->reg_state != NETREG_REGISTERED);
5203 /* If device is running, close it first. */
5204 dev_close_many(head);
5206 list_for_each_entry(dev, head, unreg_list) {
5207 /* And unlink it from device chain. */
5208 unlist_netdevice(dev);
5210 dev->reg_state = NETREG_UNREGISTERING;
5215 list_for_each_entry(dev, head, unreg_list) {
5216 /* Shutdown queueing discipline. */
5220 /* Notify protocols, that we are about to destroy
5221 this device. They should clean all the things.
5223 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5225 if (!dev->rtnl_link_ops ||
5226 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5227 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
5230 * Flush the unicast and multicast chains
5235 if (dev->netdev_ops->ndo_uninit)
5236 dev->netdev_ops->ndo_uninit(dev);
5238 /* Notifier chain MUST detach us from master device. */
5239 WARN_ON(dev->master);
5241 /* Remove entries from kobject tree */
5242 netdev_unregister_kobject(dev);
5245 /* Process any work delayed until the end of the batch */
5246 dev = list_first_entry(head, struct net_device, unreg_list);
5247 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5251 list_for_each_entry(dev, head, unreg_list)
5255 static void rollback_registered(struct net_device *dev)
5259 list_add(&dev->unreg_list, &single);
5260 rollback_registered_many(&single);
5264 static netdev_features_t netdev_fix_features(struct net_device *dev,
5265 netdev_features_t features)
5267 /* Fix illegal checksum combinations */
5268 if ((features & NETIF_F_HW_CSUM) &&
5269 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5270 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
5271 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5274 /* Fix illegal SG+CSUM combinations. */
5275 if ((features & NETIF_F_SG) &&
5276 !(features & NETIF_F_ALL_CSUM)) {
5278 "Dropping NETIF_F_SG since no checksum feature.\n");
5279 features &= ~NETIF_F_SG;
5282 /* TSO requires that SG is present as well. */
5283 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
5284 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
5285 features &= ~NETIF_F_ALL_TSO;
5288 /* TSO ECN requires that TSO is present as well. */
5289 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
5290 features &= ~NETIF_F_TSO_ECN;
5292 /* Software GSO depends on SG. */
5293 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
5294 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
5295 features &= ~NETIF_F_GSO;
5298 /* UFO needs SG and checksumming */
5299 if (features & NETIF_F_UFO) {
5300 /* maybe split UFO into V4 and V6? */
5301 if (!((features & NETIF_F_GEN_CSUM) ||
5302 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))
5303 == (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5305 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5306 features &= ~NETIF_F_UFO;
5309 if (!(features & NETIF_F_SG)) {
5311 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5312 features &= ~NETIF_F_UFO;
5319 int __netdev_update_features(struct net_device *dev)
5321 netdev_features_t features;
5326 features = netdev_get_wanted_features(dev);
5328 if (dev->netdev_ops->ndo_fix_features)
5329 features = dev->netdev_ops->ndo_fix_features(dev, features);
5331 /* driver might be less strict about feature dependencies */
5332 features = netdev_fix_features(dev, features);
5334 if (dev->features == features)
5337 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
5338 &dev->features, &features);
5340 if (dev->netdev_ops->ndo_set_features)
5341 err = dev->netdev_ops->ndo_set_features(dev, features);
5343 if (unlikely(err < 0)) {
5345 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5346 err, &features, &dev->features);
5351 dev->features = features;
5357 * netdev_update_features - recalculate device features
5358 * @dev: the device to check
5360 * Recalculate dev->features set and send notifications if it
5361 * has changed. Should be called after driver or hardware dependent
5362 * conditions might have changed that influence the features.
5364 void netdev_update_features(struct net_device *dev)
5366 if (__netdev_update_features(dev))
5367 netdev_features_change(dev);
5369 EXPORT_SYMBOL(netdev_update_features);
5372 * netdev_change_features - recalculate device features
5373 * @dev: the device to check
5375 * Recalculate dev->features set and send notifications even
5376 * if they have not changed. Should be called instead of
5377 * netdev_update_features() if also dev->vlan_features might
5378 * have changed to allow the changes to be propagated to stacked
5381 void netdev_change_features(struct net_device *dev)
5383 __netdev_update_features(dev);
5384 netdev_features_change(dev);
5386 EXPORT_SYMBOL(netdev_change_features);
5389 * netif_stacked_transfer_operstate - transfer operstate
5390 * @rootdev: the root or lower level device to transfer state from
5391 * @dev: the device to transfer operstate to
5393 * Transfer operational state from root to device. This is normally
5394 * called when a stacking relationship exists between the root
5395 * device and the device(a leaf device).
5397 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
5398 struct net_device *dev)
5400 if (rootdev->operstate == IF_OPER_DORMANT)
5401 netif_dormant_on(dev);
5403 netif_dormant_off(dev);
5405 if (netif_carrier_ok(rootdev)) {
5406 if (!netif_carrier_ok(dev))
5407 netif_carrier_on(dev);
5409 if (netif_carrier_ok(dev))
5410 netif_carrier_off(dev);
5413 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
5416 static int netif_alloc_rx_queues(struct net_device *dev)
5418 unsigned int i, count = dev->num_rx_queues;
5419 struct netdev_rx_queue *rx;
5423 rx = kcalloc(count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5425 pr_err("netdev: Unable to allocate %u rx queues\n", count);
5430 for (i = 0; i < count; i++)
5436 static void netdev_init_one_queue(struct net_device *dev,
5437 struct netdev_queue *queue, void *_unused)
5439 /* Initialize queue lock */
5440 spin_lock_init(&queue->_xmit_lock);
5441 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
5442 queue->xmit_lock_owner = -1;
5443 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
5446 dql_init(&queue->dql, HZ);
5450 static int netif_alloc_netdev_queues(struct net_device *dev)
5452 unsigned int count = dev->num_tx_queues;
5453 struct netdev_queue *tx;
5457 tx = kcalloc(count, sizeof(struct netdev_queue), GFP_KERNEL);
5459 pr_err("netdev: Unable to allocate %u tx queues\n", count);
5464 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5465 spin_lock_init(&dev->tx_global_lock);
5471 * register_netdevice - 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 * Callers must hold the rtnl semaphore. You may want
5480 * register_netdev() instead of this.
5483 * The locking appears insufficient to guarantee two parallel registers
5484 * will not get the same name.
5487 int register_netdevice(struct net_device *dev)
5490 struct net *net = dev_net(dev);
5492 BUG_ON(dev_boot_phase);
5497 /* When net_device's are persistent, this will be fatal. */
5498 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
5501 spin_lock_init(&dev->addr_list_lock);
5502 netdev_set_addr_lockdep_class(dev);
5506 ret = dev_get_valid_name(dev, dev->name);
5510 /* Init, if this function is available */
5511 if (dev->netdev_ops->ndo_init) {
5512 ret = dev->netdev_ops->ndo_init(dev);
5520 dev->ifindex = dev_new_index(net);
5521 if (dev->iflink == -1)
5522 dev->iflink = dev->ifindex;
5524 /* Transfer changeable features to wanted_features and enable
5525 * software offloads (GSO and GRO).
5527 dev->hw_features |= NETIF_F_SOFT_FEATURES;
5528 dev->features |= NETIF_F_SOFT_FEATURES;
5529 dev->wanted_features = dev->features & dev->hw_features;
5531 /* Turn on no cache copy if HW is doing checksum */
5532 if (!(dev->flags & IFF_LOOPBACK)) {
5533 dev->hw_features |= NETIF_F_NOCACHE_COPY;
5534 if (dev->features & NETIF_F_ALL_CSUM) {
5535 dev->wanted_features |= NETIF_F_NOCACHE_COPY;
5536 dev->features |= NETIF_F_NOCACHE_COPY;
5540 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5542 dev->vlan_features |= NETIF_F_HIGHDMA;
5544 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5545 ret = notifier_to_errno(ret);
5549 ret = netdev_register_kobject(dev);
5552 dev->reg_state = NETREG_REGISTERED;
5554 __netdev_update_features(dev);
5557 * Default initial state at registry is that the
5558 * device is present.
5561 set_bit(__LINK_STATE_PRESENT, &dev->state);
5563 dev_init_scheduler(dev);
5565 list_netdevice(dev);
5567 /* Notify protocols, that a new device appeared. */
5568 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5569 ret = notifier_to_errno(ret);
5571 rollback_registered(dev);
5572 dev->reg_state = NETREG_UNREGISTERED;
5575 * Prevent userspace races by waiting until the network
5576 * device is fully setup before sending notifications.
5578 if (!dev->rtnl_link_ops ||
5579 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5580 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5586 if (dev->netdev_ops->ndo_uninit)
5587 dev->netdev_ops->ndo_uninit(dev);
5590 EXPORT_SYMBOL(register_netdevice);
5593 * init_dummy_netdev - init a dummy network device for NAPI
5594 * @dev: device to init
5596 * This takes a network device structure and initialize the minimum
5597 * amount of fields so it can be used to schedule NAPI polls without
5598 * registering a full blown interface. This is to be used by drivers
5599 * that need to tie several hardware interfaces to a single NAPI
5600 * poll scheduler due to HW limitations.
5602 int init_dummy_netdev(struct net_device *dev)
5604 /* Clear everything. Note we don't initialize spinlocks
5605 * are they aren't supposed to be taken by any of the
5606 * NAPI code and this dummy netdev is supposed to be
5607 * only ever used for NAPI polls
5609 memset(dev, 0, sizeof(struct net_device));
5611 /* make sure we BUG if trying to hit standard
5612 * register/unregister code path
5614 dev->reg_state = NETREG_DUMMY;
5616 /* NAPI wants this */
5617 INIT_LIST_HEAD(&dev->napi_list);
5619 /* a dummy interface is started by default */
5620 set_bit(__LINK_STATE_PRESENT, &dev->state);
5621 set_bit(__LINK_STATE_START, &dev->state);
5623 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5624 * because users of this 'device' dont need to change
5630 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5634 * register_netdev - register a network device
5635 * @dev: device to register
5637 * Take a completed network device structure and add it to the kernel
5638 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5639 * chain. 0 is returned on success. A negative errno code is returned
5640 * on a failure to set up the device, or if the name is a duplicate.
5642 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5643 * and expands the device name if you passed a format string to
5646 int register_netdev(struct net_device *dev)
5651 err = register_netdevice(dev);
5655 EXPORT_SYMBOL(register_netdev);
5657 int netdev_refcnt_read(const struct net_device *dev)
5661 for_each_possible_cpu(i)
5662 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
5665 EXPORT_SYMBOL(netdev_refcnt_read);
5668 * netdev_wait_allrefs - wait until all references are gone.
5670 * This is called when unregistering network devices.
5672 * Any protocol or device that holds a reference should register
5673 * for netdevice notification, and cleanup and put back the
5674 * reference if they receive an UNREGISTER event.
5675 * We can get stuck here if buggy protocols don't correctly
5678 static void netdev_wait_allrefs(struct net_device *dev)
5680 unsigned long rebroadcast_time, warning_time;
5683 linkwatch_forget_dev(dev);
5685 rebroadcast_time = warning_time = jiffies;
5686 refcnt = netdev_refcnt_read(dev);
5688 while (refcnt != 0) {
5689 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5692 /* Rebroadcast unregister notification */
5693 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5694 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5695 * should have already handle it the first time */
5697 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5699 /* We must not have linkwatch events
5700 * pending on unregister. If this
5701 * happens, we simply run the queue
5702 * unscheduled, resulting in a noop
5705 linkwatch_run_queue();
5710 rebroadcast_time = jiffies;
5715 refcnt = netdev_refcnt_read(dev);
5717 if (time_after(jiffies, warning_time + 10 * HZ)) {
5718 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5720 warning_time = jiffies;
5729 * register_netdevice(x1);
5730 * register_netdevice(x2);
5732 * unregister_netdevice(y1);
5733 * unregister_netdevice(y2);
5739 * We are invoked by rtnl_unlock().
5740 * This allows us to deal with problems:
5741 * 1) We can delete sysfs objects which invoke hotplug
5742 * without deadlocking with linkwatch via keventd.
5743 * 2) Since we run with the RTNL semaphore not held, we can sleep
5744 * safely in order to wait for the netdev refcnt to drop to zero.
5746 * We must not return until all unregister events added during
5747 * the interval the lock was held have been completed.
5749 void netdev_run_todo(void)
5751 struct list_head list;
5753 /* Snapshot list, allow later requests */
5754 list_replace_init(&net_todo_list, &list);
5758 /* Wait for rcu callbacks to finish before attempting to drain
5759 * the device list. This usually avoids a 250ms wait.
5761 if (!list_empty(&list))
5764 while (!list_empty(&list)) {
5765 struct net_device *dev
5766 = list_first_entry(&list, struct net_device, todo_list);
5767 list_del(&dev->todo_list);
5769 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5770 pr_err("network todo '%s' but state %d\n",
5771 dev->name, dev->reg_state);
5776 dev->reg_state = NETREG_UNREGISTERED;
5778 on_each_cpu(flush_backlog, dev, 1);
5780 netdev_wait_allrefs(dev);
5783 BUG_ON(netdev_refcnt_read(dev));
5784 WARN_ON(rcu_access_pointer(dev->ip_ptr));
5785 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
5786 WARN_ON(dev->dn_ptr);
5788 if (dev->destructor)
5789 dev->destructor(dev);
5791 /* Free network device */
5792 kobject_put(&dev->dev.kobj);
5796 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5797 * fields in the same order, with only the type differing.
5799 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
5800 const struct net_device_stats *netdev_stats)
5802 #if BITS_PER_LONG == 64
5803 BUILD_BUG_ON(sizeof(*stats64) != sizeof(*netdev_stats));
5804 memcpy(stats64, netdev_stats, sizeof(*stats64));
5806 size_t i, n = sizeof(*stats64) / sizeof(u64);
5807 const unsigned long *src = (const unsigned long *)netdev_stats;
5808 u64 *dst = (u64 *)stats64;
5810 BUILD_BUG_ON(sizeof(*netdev_stats) / sizeof(unsigned long) !=
5811 sizeof(*stats64) / sizeof(u64));
5812 for (i = 0; i < n; i++)
5816 EXPORT_SYMBOL(netdev_stats_to_stats64);
5819 * dev_get_stats - get network device statistics
5820 * @dev: device to get statistics from
5821 * @storage: place to store stats
5823 * Get network statistics from device. Return @storage.
5824 * The device driver may provide its own method by setting
5825 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5826 * otherwise the internal statistics structure is used.
5828 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
5829 struct rtnl_link_stats64 *storage)
5831 const struct net_device_ops *ops = dev->netdev_ops;
5833 if (ops->ndo_get_stats64) {
5834 memset(storage, 0, sizeof(*storage));
5835 ops->ndo_get_stats64(dev, storage);
5836 } else if (ops->ndo_get_stats) {
5837 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
5839 netdev_stats_to_stats64(storage, &dev->stats);
5841 storage->rx_dropped += atomic_long_read(&dev->rx_dropped);
5844 EXPORT_SYMBOL(dev_get_stats);
5846 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
5848 struct netdev_queue *queue = dev_ingress_queue(dev);
5850 #ifdef CONFIG_NET_CLS_ACT
5853 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
5856 netdev_init_one_queue(dev, queue, NULL);
5857 queue->qdisc = &noop_qdisc;
5858 queue->qdisc_sleeping = &noop_qdisc;
5859 rcu_assign_pointer(dev->ingress_queue, queue);
5865 * alloc_netdev_mqs - allocate network device
5866 * @sizeof_priv: size of private data to allocate space for
5867 * @name: device name format string
5868 * @setup: callback to initialize device
5869 * @txqs: the number of TX subqueues to allocate
5870 * @rxqs: the number of RX subqueues to allocate
5872 * Allocates a struct net_device with private data area for driver use
5873 * and performs basic initialization. Also allocates subquue structs
5874 * for each queue on the device.
5876 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
5877 void (*setup)(struct net_device *),
5878 unsigned int txqs, unsigned int rxqs)
5880 struct net_device *dev;
5882 struct net_device *p;
5884 BUG_ON(strlen(name) >= sizeof(dev->name));
5887 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5893 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5898 alloc_size = sizeof(struct net_device);
5900 /* ensure 32-byte alignment of private area */
5901 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5902 alloc_size += sizeof_priv;
5904 /* ensure 32-byte alignment of whole construct */
5905 alloc_size += NETDEV_ALIGN - 1;
5907 p = kzalloc(alloc_size, GFP_KERNEL);
5909 pr_err("alloc_netdev: Unable to allocate device\n");
5913 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5914 dev->padded = (char *)dev - (char *)p;
5916 dev->pcpu_refcnt = alloc_percpu(int);
5917 if (!dev->pcpu_refcnt)
5920 if (dev_addr_init(dev))
5926 dev_net_set(dev, &init_net);
5928 dev->gso_max_size = GSO_MAX_SIZE;
5930 INIT_LIST_HEAD(&dev->napi_list);
5931 INIT_LIST_HEAD(&dev->unreg_list);
5932 INIT_LIST_HEAD(&dev->link_watch_list);
5933 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5936 dev->num_tx_queues = txqs;
5937 dev->real_num_tx_queues = txqs;
5938 if (netif_alloc_netdev_queues(dev))
5942 dev->num_rx_queues = rxqs;
5943 dev->real_num_rx_queues = rxqs;
5944 if (netif_alloc_rx_queues(dev))
5948 strcpy(dev->name, name);
5949 dev->group = INIT_NETDEV_GROUP;
5957 free_percpu(dev->pcpu_refcnt);
5967 EXPORT_SYMBOL(alloc_netdev_mqs);
5970 * free_netdev - free network device
5973 * This function does the last stage of destroying an allocated device
5974 * interface. The reference to the device object is released.
5975 * If this is the last reference then it will be freed.
5977 void free_netdev(struct net_device *dev)
5979 struct napi_struct *p, *n;
5981 release_net(dev_net(dev));
5988 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
5990 /* Flush device addresses */
5991 dev_addr_flush(dev);
5993 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5996 free_percpu(dev->pcpu_refcnt);
5997 dev->pcpu_refcnt = NULL;
5999 /* Compatibility with error handling in drivers */
6000 if (dev->reg_state == NETREG_UNINITIALIZED) {
6001 kfree((char *)dev - dev->padded);
6005 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
6006 dev->reg_state = NETREG_RELEASED;
6008 /* will free via device release */
6009 put_device(&dev->dev);
6011 EXPORT_SYMBOL(free_netdev);
6014 * synchronize_net - Synchronize with packet receive processing
6016 * Wait for packets currently being received to be done.
6017 * Does not block later packets from starting.
6019 void synchronize_net(void)
6022 if (rtnl_is_locked())
6023 synchronize_rcu_expedited();
6027 EXPORT_SYMBOL(synchronize_net);
6030 * unregister_netdevice_queue - remove device from the kernel
6034 * This function shuts down a device interface and removes it
6035 * from the kernel tables.
6036 * If head not NULL, device is queued to be unregistered later.
6038 * Callers must hold the rtnl semaphore. You may want
6039 * unregister_netdev() instead of this.
6042 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
6047 list_move_tail(&dev->unreg_list, head);
6049 rollback_registered(dev);
6050 /* Finish processing unregister after unlock */
6054 EXPORT_SYMBOL(unregister_netdevice_queue);
6057 * unregister_netdevice_many - unregister many devices
6058 * @head: list of devices
6060 void unregister_netdevice_many(struct list_head *head)
6062 struct net_device *dev;
6064 if (!list_empty(head)) {
6065 rollback_registered_many(head);
6066 list_for_each_entry(dev, head, unreg_list)
6070 EXPORT_SYMBOL(unregister_netdevice_many);
6073 * unregister_netdev - remove device from the kernel
6076 * This function shuts down a device interface and removes it
6077 * from the kernel tables.
6079 * This is just a wrapper for unregister_netdevice that takes
6080 * the rtnl semaphore. In general you want to use this and not
6081 * unregister_netdevice.
6083 void unregister_netdev(struct net_device *dev)
6086 unregister_netdevice(dev);
6089 EXPORT_SYMBOL(unregister_netdev);
6092 * dev_change_net_namespace - move device to different nethost namespace
6094 * @net: network namespace
6095 * @pat: If not NULL name pattern to try if the current device name
6096 * is already taken in the destination network namespace.
6098 * This function shuts down a device interface and moves it
6099 * to a new network namespace. On success 0 is returned, on
6100 * a failure a netagive errno code is returned.
6102 * Callers must hold the rtnl semaphore.
6105 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
6111 /* Don't allow namespace local devices to be moved. */
6113 if (dev->features & NETIF_F_NETNS_LOCAL)
6116 /* Ensure the device has been registrered */
6118 if (dev->reg_state != NETREG_REGISTERED)
6121 /* Get out if there is nothing todo */
6123 if (net_eq(dev_net(dev), net))
6126 /* Pick the destination device name, and ensure
6127 * we can use it in the destination network namespace.
6130 if (__dev_get_by_name(net, dev->name)) {
6131 /* We get here if we can't use the current device name */
6134 if (dev_get_valid_name(dev, pat) < 0)
6139 * And now a mini version of register_netdevice unregister_netdevice.
6142 /* If device is running close it first. */
6145 /* And unlink it from device chain */
6147 unlist_netdevice(dev);
6151 /* Shutdown queueing discipline. */
6154 /* Notify protocols, that we are about to destroy
6155 this device. They should clean all the things.
6157 Note that dev->reg_state stays at NETREG_REGISTERED.
6158 This is wanted because this way 8021q and macvlan know
6159 the device is just moving and can keep their slaves up.
6161 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
6162 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
6163 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
6166 * Flush the unicast and multicast chains
6171 /* Actually switch the network namespace */
6172 dev_net_set(dev, net);
6174 /* If there is an ifindex conflict assign a new one */
6175 if (__dev_get_by_index(net, dev->ifindex)) {
6176 int iflink = (dev->iflink == dev->ifindex);
6177 dev->ifindex = dev_new_index(net);
6179 dev->iflink = dev->ifindex;
6182 /* Fixup kobjects */
6183 err = device_rename(&dev->dev, dev->name);
6186 /* Add the device back in the hashes */
6187 list_netdevice(dev);
6189 /* Notify protocols, that a new device appeared. */
6190 call_netdevice_notifiers(NETDEV_REGISTER, dev);
6193 * Prevent userspace races by waiting until the network
6194 * device is fully setup before sending notifications.
6196 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
6203 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
6205 static int dev_cpu_callback(struct notifier_block *nfb,
6206 unsigned long action,
6209 struct sk_buff **list_skb;
6210 struct sk_buff *skb;
6211 unsigned int cpu, oldcpu = (unsigned long)ocpu;
6212 struct softnet_data *sd, *oldsd;
6214 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
6217 local_irq_disable();
6218 cpu = smp_processor_id();
6219 sd = &per_cpu(softnet_data, cpu);
6220 oldsd = &per_cpu(softnet_data, oldcpu);
6222 /* Find end of our completion_queue. */
6223 list_skb = &sd->completion_queue;
6225 list_skb = &(*list_skb)->next;
6226 /* Append completion queue from offline CPU. */
6227 *list_skb = oldsd->completion_queue;
6228 oldsd->completion_queue = NULL;
6230 /* Append output queue from offline CPU. */
6231 if (oldsd->output_queue) {
6232 *sd->output_queue_tailp = oldsd->output_queue;
6233 sd->output_queue_tailp = oldsd->output_queue_tailp;
6234 oldsd->output_queue = NULL;
6235 oldsd->output_queue_tailp = &oldsd->output_queue;
6237 /* Append NAPI poll list from offline CPU. */
6238 if (!list_empty(&oldsd->poll_list)) {
6239 list_splice_init(&oldsd->poll_list, &sd->poll_list);
6240 raise_softirq_irqoff(NET_RX_SOFTIRQ);
6243 raise_softirq_irqoff(NET_TX_SOFTIRQ);
6246 /* Process offline CPU's input_pkt_queue */
6247 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
6249 input_queue_head_incr(oldsd);
6251 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
6253 input_queue_head_incr(oldsd);
6261 * netdev_increment_features - increment feature set by one
6262 * @all: current feature set
6263 * @one: new feature set
6264 * @mask: mask feature set
6266 * Computes a new feature set after adding a device with feature set
6267 * @one to the master device with current feature set @all. Will not
6268 * enable anything that is off in @mask. Returns the new feature set.
6270 netdev_features_t netdev_increment_features(netdev_features_t all,
6271 netdev_features_t one, netdev_features_t mask)
6273 if (mask & NETIF_F_GEN_CSUM)
6274 mask |= NETIF_F_ALL_CSUM;
6275 mask |= NETIF_F_VLAN_CHALLENGED;
6277 all |= one & (NETIF_F_ONE_FOR_ALL|NETIF_F_ALL_CSUM) & mask;
6278 all &= one | ~NETIF_F_ALL_FOR_ALL;
6280 /* If one device supports hw checksumming, set for all. */
6281 if (all & NETIF_F_GEN_CSUM)
6282 all &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_GEN_CSUM);
6286 EXPORT_SYMBOL(netdev_increment_features);
6288 static struct hlist_head *netdev_create_hash(void)
6291 struct hlist_head *hash;
6293 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
6295 for (i = 0; i < NETDEV_HASHENTRIES; i++)
6296 INIT_HLIST_HEAD(&hash[i]);
6301 /* Initialize per network namespace state */
6302 static int __net_init netdev_init(struct net *net)
6304 INIT_LIST_HEAD(&net->dev_base_head);
6306 net->dev_name_head = netdev_create_hash();
6307 if (net->dev_name_head == NULL)
6310 net->dev_index_head = netdev_create_hash();
6311 if (net->dev_index_head == NULL)
6317 kfree(net->dev_name_head);
6323 * netdev_drivername - network driver for the device
6324 * @dev: network device
6326 * Determine network driver for device.
6328 const char *netdev_drivername(const struct net_device *dev)
6330 const struct device_driver *driver;
6331 const struct device *parent;
6332 const char *empty = "";
6334 parent = dev->dev.parent;
6338 driver = parent->driver;
6339 if (driver && driver->name)
6340 return driver->name;
6344 int __netdev_printk(const char *level, const struct net_device *dev,
6345 struct va_format *vaf)
6349 if (dev && dev->dev.parent)
6350 r = dev_printk(level, dev->dev.parent, "%s: %pV",
6351 netdev_name(dev), vaf);
6353 r = printk("%s%s: %pV", level, netdev_name(dev), vaf);
6355 r = printk("%s(NULL net_device): %pV", level, vaf);
6359 EXPORT_SYMBOL(__netdev_printk);
6361 int netdev_printk(const char *level, const struct net_device *dev,
6362 const char *format, ...)
6364 struct va_format vaf;
6368 va_start(args, format);
6373 r = __netdev_printk(level, dev, &vaf);
6378 EXPORT_SYMBOL(netdev_printk);
6380 #define define_netdev_printk_level(func, level) \
6381 int func(const struct net_device *dev, const char *fmt, ...) \
6384 struct va_format vaf; \
6387 va_start(args, fmt); \
6392 r = __netdev_printk(level, dev, &vaf); \
6397 EXPORT_SYMBOL(func);
6399 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
6400 define_netdev_printk_level(netdev_alert, KERN_ALERT);
6401 define_netdev_printk_level(netdev_crit, KERN_CRIT);
6402 define_netdev_printk_level(netdev_err, KERN_ERR);
6403 define_netdev_printk_level(netdev_warn, KERN_WARNING);
6404 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
6405 define_netdev_printk_level(netdev_info, KERN_INFO);
6407 static void __net_exit netdev_exit(struct net *net)
6409 kfree(net->dev_name_head);
6410 kfree(net->dev_index_head);
6413 static struct pernet_operations __net_initdata netdev_net_ops = {
6414 .init = netdev_init,
6415 .exit = netdev_exit,
6418 static void __net_exit default_device_exit(struct net *net)
6420 struct net_device *dev, *aux;
6422 * Push all migratable network devices back to the
6423 * initial network namespace
6426 for_each_netdev_safe(net, dev, aux) {
6428 char fb_name[IFNAMSIZ];
6430 /* Ignore unmoveable devices (i.e. loopback) */
6431 if (dev->features & NETIF_F_NETNS_LOCAL)
6434 /* Leave virtual devices for the generic cleanup */
6435 if (dev->rtnl_link_ops)
6438 /* Push remaining network devices to init_net */
6439 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
6440 err = dev_change_net_namespace(dev, &init_net, fb_name);
6442 pr_emerg("%s: failed to move %s to init_net: %d\n",
6443 __func__, dev->name, err);
6450 static void __net_exit default_device_exit_batch(struct list_head *net_list)
6452 /* At exit all network devices most be removed from a network
6453 * namespace. Do this in the reverse order of registration.
6454 * Do this across as many network namespaces as possible to
6455 * improve batching efficiency.
6457 struct net_device *dev;
6459 LIST_HEAD(dev_kill_list);
6462 list_for_each_entry(net, net_list, exit_list) {
6463 for_each_netdev_reverse(net, dev) {
6464 if (dev->rtnl_link_ops)
6465 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
6467 unregister_netdevice_queue(dev, &dev_kill_list);
6470 unregister_netdevice_many(&dev_kill_list);
6471 list_del(&dev_kill_list);
6475 static struct pernet_operations __net_initdata default_device_ops = {
6476 .exit = default_device_exit,
6477 .exit_batch = default_device_exit_batch,
6481 * Initialize the DEV module. At boot time this walks the device list and
6482 * unhooks any devices that fail to initialise (normally hardware not
6483 * present) and leaves us with a valid list of present and active devices.
6488 * This is called single threaded during boot, so no need
6489 * to take the rtnl semaphore.
6491 static int __init net_dev_init(void)
6493 int i, rc = -ENOMEM;
6495 BUG_ON(!dev_boot_phase);
6497 if (dev_proc_init())
6500 if (netdev_kobject_init())
6503 INIT_LIST_HEAD(&ptype_all);
6504 for (i = 0; i < PTYPE_HASH_SIZE; i++)
6505 INIT_LIST_HEAD(&ptype_base[i]);
6507 if (register_pernet_subsys(&netdev_net_ops))
6511 * Initialise the packet receive queues.
6514 for_each_possible_cpu(i) {
6515 struct softnet_data *sd = &per_cpu(softnet_data, i);
6517 memset(sd, 0, sizeof(*sd));
6518 skb_queue_head_init(&sd->input_pkt_queue);
6519 skb_queue_head_init(&sd->process_queue);
6520 sd->completion_queue = NULL;
6521 INIT_LIST_HEAD(&sd->poll_list);
6522 sd->output_queue = NULL;
6523 sd->output_queue_tailp = &sd->output_queue;
6525 sd->csd.func = rps_trigger_softirq;
6531 sd->backlog.poll = process_backlog;
6532 sd->backlog.weight = weight_p;
6533 sd->backlog.gro_list = NULL;
6534 sd->backlog.gro_count = 0;
6539 /* The loopback device is special if any other network devices
6540 * is present in a network namespace the loopback device must
6541 * be present. Since we now dynamically allocate and free the
6542 * loopback device ensure this invariant is maintained by
6543 * keeping the loopback device as the first device on the
6544 * list of network devices. Ensuring the loopback devices
6545 * is the first device that appears and the last network device
6548 if (register_pernet_device(&loopback_net_ops))
6551 if (register_pernet_device(&default_device_ops))
6554 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
6555 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
6557 hotcpu_notifier(dev_cpu_callback, 0);
6565 subsys_initcall(net_dev_init);
6567 static int __init initialize_hashrnd(void)
6569 get_random_bytes(&hashrnd, sizeof(hashrnd));
6573 late_initcall_sync(initialize_hashrnd);