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 <asm/system.h>
77 #include <linux/bitops.h>
78 #include <linux/capability.h>
79 #include <linux/cpu.h>
80 #include <linux/types.h>
81 #include <linux/kernel.h>
82 #include <linux/hash.h>
83 #include <linux/slab.h>
84 #include <linux/sched.h>
85 #include <linux/mutex.h>
86 #include <linux/string.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/errno.h>
91 #include <linux/interrupt.h>
92 #include <linux/if_ether.h>
93 #include <linux/netdevice.h>
94 #include <linux/etherdevice.h>
95 #include <linux/ethtool.h>
96 #include <linux/notifier.h>
97 #include <linux/skbuff.h>
98 #include <net/net_namespace.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/stat.h>
104 #include <linux/if_bridge.h>
105 #include <linux/if_macvlan.h>
107 #include <net/pkt_sched.h>
108 #include <net/checksum.h>
109 #include <net/xfrm.h>
110 #include <linux/highmem.h>
111 #include <linux/init.h>
112 #include <linux/kmod.h>
113 #include <linux/module.h>
114 #include <linux/netpoll.h>
115 #include <linux/rcupdate.h>
116 #include <linux/delay.h>
117 #include <net/wext.h>
118 #include <net/iw_handler.h>
119 #include <asm/current.h>
120 #include <linux/audit.h>
121 #include <linux/dmaengine.h>
122 #include <linux/err.h>
123 #include <linux/ctype.h>
124 #include <linux/if_arp.h>
125 #include <linux/if_vlan.h>
126 #include <linux/ip.h>
128 #include <linux/ipv6.h>
129 #include <linux/in.h>
130 #include <linux/jhash.h>
131 #include <linux/random.h>
132 #include <trace/events/napi.h>
133 #include <linux/pci.h>
135 #include "net-sysfs.h"
137 /* Instead of increasing this, you should create a hash table. */
138 #define MAX_GRO_SKBS 8
140 /* This should be increased if a protocol with a bigger head is added. */
141 #define GRO_MAX_HEAD (MAX_HEADER + 128)
144 * The list of packet types we will receive (as opposed to discard)
145 * and the routines to invoke.
147 * Why 16. Because with 16 the only overlap we get on a hash of the
148 * low nibble of the protocol value is RARP/SNAP/X.25.
150 * NOTE: That is no longer true with the addition of VLAN tags. Not
151 * sure which should go first, but I bet it won't make much
152 * difference if we are running VLANs. The good news is that
153 * this protocol won't be in the list unless compiled in, so
154 * the average user (w/out VLANs) will not be adversely affected.
171 #define PTYPE_HASH_SIZE (16)
172 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
174 static DEFINE_SPINLOCK(ptype_lock);
175 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
176 static struct list_head ptype_all __read_mostly; /* Taps */
179 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
182 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
184 * Writers must hold the rtnl semaphore while they loop through the
185 * dev_base_head list, and hold dev_base_lock for writing when they do the
186 * actual updates. This allows pure readers to access the list even
187 * while a writer is preparing to update it.
189 * To put it another way, dev_base_lock is held for writing only to
190 * protect against pure readers; the rtnl semaphore provides the
191 * protection against other writers.
193 * See, for example usages, register_netdevice() and
194 * unregister_netdevice(), which must be called with the rtnl
197 DEFINE_RWLOCK(dev_base_lock);
198 EXPORT_SYMBOL(dev_base_lock);
200 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
202 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
203 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
206 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
208 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
211 static inline void rps_lock(struct softnet_data *sd)
214 spin_lock(&sd->input_pkt_queue.lock);
218 static inline void rps_unlock(struct softnet_data *sd)
221 spin_unlock(&sd->input_pkt_queue.lock);
225 /* Device list insertion */
226 static int list_netdevice(struct net_device *dev)
228 struct net *net = dev_net(dev);
232 write_lock_bh(&dev_base_lock);
233 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
234 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
235 hlist_add_head_rcu(&dev->index_hlist,
236 dev_index_hash(net, dev->ifindex));
237 write_unlock_bh(&dev_base_lock);
241 /* Device list removal
242 * caller must respect a RCU grace period before freeing/reusing dev
244 static void unlist_netdevice(struct net_device *dev)
248 /* Unlink dev from the device chain */
249 write_lock_bh(&dev_base_lock);
250 list_del_rcu(&dev->dev_list);
251 hlist_del_rcu(&dev->name_hlist);
252 hlist_del_rcu(&dev->index_hlist);
253 write_unlock_bh(&dev_base_lock);
260 static RAW_NOTIFIER_HEAD(netdev_chain);
263 * Device drivers call our routines to queue packets here. We empty the
264 * queue in the local softnet handler.
267 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
268 EXPORT_PER_CPU_SYMBOL(softnet_data);
270 #ifdef CONFIG_LOCKDEP
272 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
273 * according to dev->type
275 static const unsigned short netdev_lock_type[] =
276 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
277 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
278 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
279 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
280 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
281 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
282 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
283 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
284 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
285 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
286 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
287 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
288 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
289 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
290 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
291 ARPHRD_VOID, ARPHRD_NONE};
293 static const char *const netdev_lock_name[] =
294 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
295 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
296 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
297 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
298 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
299 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
300 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
301 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
302 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
303 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
304 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
305 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
306 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
307 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
308 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
309 "_xmit_VOID", "_xmit_NONE"};
311 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
312 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
314 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
318 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
319 if (netdev_lock_type[i] == dev_type)
321 /* the last key is used by default */
322 return ARRAY_SIZE(netdev_lock_type) - 1;
325 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
326 unsigned short dev_type)
330 i = netdev_lock_pos(dev_type);
331 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
332 netdev_lock_name[i]);
335 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
339 i = netdev_lock_pos(dev->type);
340 lockdep_set_class_and_name(&dev->addr_list_lock,
341 &netdev_addr_lock_key[i],
342 netdev_lock_name[i]);
345 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
346 unsigned short dev_type)
349 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
354 /*******************************************************************************
356 Protocol management and registration routines
358 *******************************************************************************/
361 * Add a protocol ID to the list. Now that the input handler is
362 * smarter we can dispense with all the messy stuff that used to be
365 * BEWARE!!! Protocol handlers, mangling input packets,
366 * MUST BE last in hash buckets and checking protocol handlers
367 * MUST start from promiscuous ptype_all chain in net_bh.
368 * It is true now, do not change it.
369 * Explanation follows: if protocol handler, mangling packet, will
370 * be the first on list, it is not able to sense, that packet
371 * is cloned and should be copied-on-write, so that it will
372 * change it and subsequent readers will get broken packet.
377 * dev_add_pack - add packet handler
378 * @pt: packet type declaration
380 * Add a protocol handler to the networking stack. The passed &packet_type
381 * is linked into kernel lists and may not be freed until it has been
382 * removed from the kernel lists.
384 * This call does not sleep therefore it can not
385 * guarantee all CPU's that are in middle of receiving packets
386 * will see the new packet type (until the next received packet).
389 void dev_add_pack(struct packet_type *pt)
393 spin_lock_bh(&ptype_lock);
394 if (pt->type == htons(ETH_P_ALL))
395 list_add_rcu(&pt->list, &ptype_all);
397 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
398 list_add_rcu(&pt->list, &ptype_base[hash]);
400 spin_unlock_bh(&ptype_lock);
402 EXPORT_SYMBOL(dev_add_pack);
405 * __dev_remove_pack - remove packet handler
406 * @pt: packet type declaration
408 * Remove a protocol handler that was previously added to the kernel
409 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
410 * from the kernel lists and can be freed or reused once this function
413 * The packet type might still be in use by receivers
414 * and must not be freed until after all the CPU's have gone
415 * through a quiescent state.
417 void __dev_remove_pack(struct packet_type *pt)
419 struct list_head *head;
420 struct packet_type *pt1;
422 spin_lock_bh(&ptype_lock);
424 if (pt->type == htons(ETH_P_ALL))
427 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
429 list_for_each_entry(pt1, head, list) {
431 list_del_rcu(&pt->list);
436 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
438 spin_unlock_bh(&ptype_lock);
440 EXPORT_SYMBOL(__dev_remove_pack);
443 * dev_remove_pack - remove packet handler
444 * @pt: packet type declaration
446 * Remove a protocol handler that was previously added to the kernel
447 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
448 * from the kernel lists and can be freed or reused once this function
451 * This call sleeps to guarantee that no CPU is looking at the packet
454 void dev_remove_pack(struct packet_type *pt)
456 __dev_remove_pack(pt);
460 EXPORT_SYMBOL(dev_remove_pack);
462 /******************************************************************************
464 Device Boot-time Settings Routines
466 *******************************************************************************/
468 /* Boot time configuration table */
469 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
472 * netdev_boot_setup_add - add new setup entry
473 * @name: name of the device
474 * @map: configured settings for the device
476 * Adds new setup entry to the dev_boot_setup list. The function
477 * returns 0 on error and 1 on success. This is a generic routine to
480 static int netdev_boot_setup_add(char *name, struct ifmap *map)
482 struct netdev_boot_setup *s;
486 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
487 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
488 memset(s[i].name, 0, sizeof(s[i].name));
489 strlcpy(s[i].name, name, IFNAMSIZ);
490 memcpy(&s[i].map, map, sizeof(s[i].map));
495 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
499 * netdev_boot_setup_check - check boot time settings
500 * @dev: the netdevice
502 * Check boot time settings for the device.
503 * The found settings are set for the device to be used
504 * later in the device probing.
505 * Returns 0 if no settings found, 1 if they are.
507 int netdev_boot_setup_check(struct net_device *dev)
509 struct netdev_boot_setup *s = dev_boot_setup;
512 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
513 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
514 !strcmp(dev->name, s[i].name)) {
515 dev->irq = s[i].map.irq;
516 dev->base_addr = s[i].map.base_addr;
517 dev->mem_start = s[i].map.mem_start;
518 dev->mem_end = s[i].map.mem_end;
524 EXPORT_SYMBOL(netdev_boot_setup_check);
528 * netdev_boot_base - get address from boot time settings
529 * @prefix: prefix for network device
530 * @unit: id for network device
532 * Check boot time settings for the base address of device.
533 * The found settings are set for the device to be used
534 * later in the device probing.
535 * Returns 0 if no settings found.
537 unsigned long netdev_boot_base(const char *prefix, int unit)
539 const struct netdev_boot_setup *s = dev_boot_setup;
543 sprintf(name, "%s%d", prefix, unit);
546 * If device already registered then return base of 1
547 * to indicate not to probe for this interface
549 if (__dev_get_by_name(&init_net, name))
552 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
553 if (!strcmp(name, s[i].name))
554 return s[i].map.base_addr;
559 * Saves at boot time configured settings for any netdevice.
561 int __init netdev_boot_setup(char *str)
566 str = get_options(str, ARRAY_SIZE(ints), ints);
571 memset(&map, 0, sizeof(map));
575 map.base_addr = ints[2];
577 map.mem_start = ints[3];
579 map.mem_end = ints[4];
581 /* Add new entry to the list */
582 return netdev_boot_setup_add(str, &map);
585 __setup("netdev=", netdev_boot_setup);
587 /*******************************************************************************
589 Device Interface Subroutines
591 *******************************************************************************/
594 * __dev_get_by_name - find a device by its name
595 * @net: the applicable net namespace
596 * @name: name to find
598 * Find an interface by name. Must be called under RTNL semaphore
599 * or @dev_base_lock. If the name is found a pointer to the device
600 * is returned. If the name is not found then %NULL is returned. The
601 * reference counters are not incremented so the caller must be
602 * careful with locks.
605 struct net_device *__dev_get_by_name(struct net *net, const char *name)
607 struct hlist_node *p;
608 struct net_device *dev;
609 struct hlist_head *head = dev_name_hash(net, name);
611 hlist_for_each_entry(dev, p, head, name_hlist)
612 if (!strncmp(dev->name, name, IFNAMSIZ))
617 EXPORT_SYMBOL(__dev_get_by_name);
620 * dev_get_by_name_rcu - find a device by its name
621 * @net: the applicable net namespace
622 * @name: name to find
624 * Find an interface by name.
625 * If the name is found a pointer to the device is returned.
626 * If the name is not found then %NULL is returned.
627 * The reference counters are not incremented so the caller must be
628 * careful with locks. The caller must hold RCU lock.
631 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
633 struct hlist_node *p;
634 struct net_device *dev;
635 struct hlist_head *head = dev_name_hash(net, name);
637 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
638 if (!strncmp(dev->name, name, IFNAMSIZ))
643 EXPORT_SYMBOL(dev_get_by_name_rcu);
646 * dev_get_by_name - find a device by its name
647 * @net: the applicable net namespace
648 * @name: name to find
650 * Find an interface by name. This can be called from any
651 * context and does its own locking. The returned handle has
652 * the usage count incremented and the caller must use dev_put() to
653 * release it when it is no longer needed. %NULL is returned if no
654 * matching device is found.
657 struct net_device *dev_get_by_name(struct net *net, const char *name)
659 struct net_device *dev;
662 dev = dev_get_by_name_rcu(net, name);
668 EXPORT_SYMBOL(dev_get_by_name);
671 * __dev_get_by_index - find a device by its ifindex
672 * @net: the applicable net namespace
673 * @ifindex: index of device
675 * Search for an interface by index. Returns %NULL if the device
676 * is not found or a pointer to the device. The device has not
677 * had its reference counter increased so the caller must be careful
678 * about locking. The caller must hold either the RTNL semaphore
682 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
684 struct hlist_node *p;
685 struct net_device *dev;
686 struct hlist_head *head = dev_index_hash(net, ifindex);
688 hlist_for_each_entry(dev, p, head, index_hlist)
689 if (dev->ifindex == ifindex)
694 EXPORT_SYMBOL(__dev_get_by_index);
697 * dev_get_by_index_rcu - find a device by its ifindex
698 * @net: the applicable net namespace
699 * @ifindex: index of device
701 * Search for an interface by index. Returns %NULL if the device
702 * is not found or a pointer to the device. The device has not
703 * had its reference counter increased so the caller must be careful
704 * about locking. The caller must hold RCU lock.
707 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
709 struct hlist_node *p;
710 struct net_device *dev;
711 struct hlist_head *head = dev_index_hash(net, ifindex);
713 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
714 if (dev->ifindex == ifindex)
719 EXPORT_SYMBOL(dev_get_by_index_rcu);
723 * dev_get_by_index - find a device by its ifindex
724 * @net: the applicable net namespace
725 * @ifindex: index of device
727 * Search for an interface by index. Returns NULL if the device
728 * is not found or a pointer to the device. The device returned has
729 * had a reference added and the pointer is safe until the user calls
730 * dev_put to indicate they have finished with it.
733 struct net_device *dev_get_by_index(struct net *net, int ifindex)
735 struct net_device *dev;
738 dev = dev_get_by_index_rcu(net, ifindex);
744 EXPORT_SYMBOL(dev_get_by_index);
747 * dev_getbyhwaddr - find a device by its hardware address
748 * @net: the applicable net namespace
749 * @type: media type of device
750 * @ha: hardware address
752 * Search for an interface by MAC address. Returns NULL if the device
753 * is not found or a pointer to the device. The caller must hold the
754 * rtnl semaphore. The returned device has not had its ref count increased
755 * and the caller must therefore be careful about locking
758 * If the API was consistent this would be __dev_get_by_hwaddr
761 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
763 struct net_device *dev;
767 for_each_netdev(net, dev)
768 if (dev->type == type &&
769 !memcmp(dev->dev_addr, ha, dev->addr_len))
774 EXPORT_SYMBOL(dev_getbyhwaddr);
776 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
778 struct net_device *dev;
781 for_each_netdev(net, dev)
782 if (dev->type == type)
787 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
789 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
791 struct net_device *dev, *ret = NULL;
794 for_each_netdev_rcu(net, dev)
795 if (dev->type == type) {
803 EXPORT_SYMBOL(dev_getfirstbyhwtype);
806 * dev_get_by_flags - find any device with given flags
807 * @net: the applicable net namespace
808 * @if_flags: IFF_* values
809 * @mask: bitmask of bits in if_flags to check
811 * Search for any interface with the given flags. Returns NULL if a device
812 * is not found or a pointer to the device. The device returned has
813 * had a reference added and the pointer is safe until the user calls
814 * dev_put to indicate they have finished with it.
817 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
820 struct net_device *dev, *ret;
824 for_each_netdev_rcu(net, dev) {
825 if (((dev->flags ^ if_flags) & mask) == 0) {
834 EXPORT_SYMBOL(dev_get_by_flags);
837 * dev_valid_name - check if name is okay for network device
840 * Network device names need to be valid file names to
841 * to allow sysfs to work. We also disallow any kind of
844 int dev_valid_name(const char *name)
848 if (strlen(name) >= IFNAMSIZ)
850 if (!strcmp(name, ".") || !strcmp(name, ".."))
854 if (*name == '/' || isspace(*name))
860 EXPORT_SYMBOL(dev_valid_name);
863 * __dev_alloc_name - allocate a name for a device
864 * @net: network namespace to allocate the device name in
865 * @name: name format string
866 * @buf: scratch buffer and result name string
868 * Passed a format string - eg "lt%d" it will try and find a suitable
869 * id. It scans list of devices to build up a free map, then chooses
870 * the first empty slot. The caller must hold the dev_base or rtnl lock
871 * while allocating the name and adding the device in order to avoid
873 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
874 * Returns the number of the unit assigned or a negative errno code.
877 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
881 const int max_netdevices = 8*PAGE_SIZE;
882 unsigned long *inuse;
883 struct net_device *d;
885 p = strnchr(name, IFNAMSIZ-1, '%');
888 * Verify the string as this thing may have come from
889 * the user. There must be either one "%d" and no other "%"
892 if (p[1] != 'd' || strchr(p + 2, '%'))
895 /* Use one page as a bit array of possible slots */
896 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
900 for_each_netdev(net, d) {
901 if (!sscanf(d->name, name, &i))
903 if (i < 0 || i >= max_netdevices)
906 /* avoid cases where sscanf is not exact inverse of printf */
907 snprintf(buf, IFNAMSIZ, name, i);
908 if (!strncmp(buf, d->name, IFNAMSIZ))
912 i = find_first_zero_bit(inuse, max_netdevices);
913 free_page((unsigned long) inuse);
917 snprintf(buf, IFNAMSIZ, name, i);
918 if (!__dev_get_by_name(net, buf))
921 /* It is possible to run out of possible slots
922 * when the name is long and there isn't enough space left
923 * for the digits, or if all bits are used.
929 * dev_alloc_name - allocate a name for a device
931 * @name: name format string
933 * Passed a format string - eg "lt%d" it will try and find a suitable
934 * id. It scans list of devices to build up a free map, then chooses
935 * the first empty slot. The caller must hold the dev_base or rtnl lock
936 * while allocating the name and adding the device in order to avoid
938 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
939 * Returns the number of the unit assigned or a negative errno code.
942 int dev_alloc_name(struct net_device *dev, const char *name)
948 BUG_ON(!dev_net(dev));
950 ret = __dev_alloc_name(net, name, buf);
952 strlcpy(dev->name, buf, IFNAMSIZ);
955 EXPORT_SYMBOL(dev_alloc_name);
957 static int dev_get_valid_name(struct net_device *dev, const char *name, bool fmt)
961 BUG_ON(!dev_net(dev));
964 if (!dev_valid_name(name))
967 if (fmt && strchr(name, '%'))
968 return dev_alloc_name(dev, name);
969 else if (__dev_get_by_name(net, name))
971 else if (dev->name != name) {
972 if (strncmp(name, dev->name, IFNAMSIZ))
973 printk(KERN_INFO "%s renamed to %s by %s [%u]\n",
974 dev->name, name, current->comm, current->pid);
975 strlcpy(dev->name, name, IFNAMSIZ);
982 * dev_change_name - change name of a device
984 * @newname: name (or format string) must be at least IFNAMSIZ
986 * Change name of a device, can pass format strings "eth%d".
989 int dev_change_name(struct net_device *dev, const char *newname)
991 char oldname[IFNAMSIZ];
997 BUG_ON(!dev_net(dev));
1000 if (dev->flags & IFF_UP)
1003 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
1006 memcpy(oldname, dev->name, IFNAMSIZ);
1008 err = dev_get_valid_name(dev, newname, 1);
1013 ret = device_rename(&dev->dev, dev->name);
1015 memcpy(dev->name, oldname, IFNAMSIZ);
1019 write_lock_bh(&dev_base_lock);
1020 hlist_del(&dev->name_hlist);
1021 write_unlock_bh(&dev_base_lock);
1025 write_lock_bh(&dev_base_lock);
1026 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1027 write_unlock_bh(&dev_base_lock);
1029 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1030 ret = notifier_to_errno(ret);
1033 /* err >= 0 after dev_alloc_name() or stores the first errno */
1036 memcpy(dev->name, oldname, IFNAMSIZ);
1040 "%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;
1130 dev = dev_get_by_name_rcu(net, name);
1133 if (!dev && capable(CAP_NET_ADMIN))
1134 request_module("%s", name);
1136 EXPORT_SYMBOL(dev_load);
1138 static int __dev_open(struct net_device *dev)
1140 const struct net_device_ops *ops = dev->netdev_ops;
1146 * Is it even present?
1148 if (!netif_device_present(dev))
1151 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1152 ret = notifier_to_errno(ret);
1157 * Call device private open method
1159 set_bit(__LINK_STATE_START, &dev->state);
1161 if (ops->ndo_validate_addr)
1162 ret = ops->ndo_validate_addr(dev);
1164 if (!ret && ops->ndo_open)
1165 ret = ops->ndo_open(dev);
1168 * If it went open OK then:
1172 clear_bit(__LINK_STATE_START, &dev->state);
1177 dev->flags |= IFF_UP;
1182 net_dmaengine_get();
1185 * Initialize multicasting status
1187 dev_set_rx_mode(dev);
1190 * Wakeup transmit queue engine
1199 * dev_open - prepare an interface for use.
1200 * @dev: device to open
1202 * Takes a device from down to up state. The device's private open
1203 * function is invoked and then the multicast lists are loaded. Finally
1204 * the device is moved into the up state and a %NETDEV_UP message is
1205 * sent to the netdev notifier chain.
1207 * Calling this function on an active interface is a nop. On a failure
1208 * a negative errno code is returned.
1210 int dev_open(struct net_device *dev)
1217 if (dev->flags & IFF_UP)
1223 ret = __dev_open(dev);
1228 * ... and announce new interface.
1230 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1231 call_netdevice_notifiers(NETDEV_UP, dev);
1235 EXPORT_SYMBOL(dev_open);
1237 static int __dev_close(struct net_device *dev)
1239 const struct net_device_ops *ops = dev->netdev_ops;
1245 * Tell people we are going down, so that they can
1246 * prepare to death, when device is still operating.
1248 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1250 clear_bit(__LINK_STATE_START, &dev->state);
1252 /* Synchronize to scheduled poll. We cannot touch poll list,
1253 * it can be even on different cpu. So just clear netif_running().
1255 * dev->stop() will invoke napi_disable() on all of it's
1256 * napi_struct instances on this device.
1258 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1260 dev_deactivate(dev);
1263 * Call the device specific close. This cannot fail.
1264 * Only if device is UP
1266 * We allow it to be called even after a DETACH hot-plug
1273 * Device is now down.
1276 dev->flags &= ~IFF_UP;
1281 net_dmaengine_put();
1287 * dev_close - shutdown an interface.
1288 * @dev: device to shutdown
1290 * This function moves an active device into down state. A
1291 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1292 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1295 int dev_close(struct net_device *dev)
1297 if (!(dev->flags & IFF_UP))
1303 * Tell people we are down
1305 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING);
1306 call_netdevice_notifiers(NETDEV_DOWN, dev);
1310 EXPORT_SYMBOL(dev_close);
1314 * dev_disable_lro - disable Large Receive Offload on a device
1317 * Disable Large Receive Offload (LRO) on a net device. Must be
1318 * called under RTNL. This is needed if received packets may be
1319 * forwarded to another interface.
1321 void dev_disable_lro(struct net_device *dev)
1323 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1324 dev->ethtool_ops->set_flags) {
1325 u32 flags = dev->ethtool_ops->get_flags(dev);
1326 if (flags & ETH_FLAG_LRO) {
1327 flags &= ~ETH_FLAG_LRO;
1328 dev->ethtool_ops->set_flags(dev, flags);
1331 WARN_ON(dev->features & NETIF_F_LRO);
1333 EXPORT_SYMBOL(dev_disable_lro);
1336 static int dev_boot_phase = 1;
1339 * Device change register/unregister. These are not inline or static
1340 * as we export them to the world.
1344 * register_netdevice_notifier - register a network notifier block
1347 * Register a notifier to be called when network device events occur.
1348 * The notifier passed is linked into the kernel structures and must
1349 * not be reused until it has been unregistered. A negative errno code
1350 * is returned on a failure.
1352 * When registered all registration and up events are replayed
1353 * to the new notifier to allow device to have a race free
1354 * view of the network device list.
1357 int register_netdevice_notifier(struct notifier_block *nb)
1359 struct net_device *dev;
1360 struct net_device *last;
1365 err = raw_notifier_chain_register(&netdev_chain, nb);
1371 for_each_netdev(net, dev) {
1372 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1373 err = notifier_to_errno(err);
1377 if (!(dev->flags & IFF_UP))
1380 nb->notifier_call(nb, NETDEV_UP, dev);
1391 for_each_netdev(net, dev) {
1395 if (dev->flags & IFF_UP) {
1396 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1397 nb->notifier_call(nb, NETDEV_DOWN, dev);
1399 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1400 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1404 raw_notifier_chain_unregister(&netdev_chain, nb);
1407 EXPORT_SYMBOL(register_netdevice_notifier);
1410 * unregister_netdevice_notifier - unregister a network notifier block
1413 * Unregister a notifier previously registered by
1414 * register_netdevice_notifier(). The notifier is unlinked into the
1415 * kernel structures and may then be reused. A negative errno code
1416 * is returned on a failure.
1419 int unregister_netdevice_notifier(struct notifier_block *nb)
1424 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1428 EXPORT_SYMBOL(unregister_netdevice_notifier);
1431 * call_netdevice_notifiers - call all network notifier blocks
1432 * @val: value passed unmodified to notifier function
1433 * @dev: net_device pointer passed unmodified to notifier function
1435 * Call all network notifier blocks. Parameters and return value
1436 * are as for raw_notifier_call_chain().
1439 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1442 return raw_notifier_call_chain(&netdev_chain, val, dev);
1445 /* When > 0 there are consumers of rx skb time stamps */
1446 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1448 void net_enable_timestamp(void)
1450 atomic_inc(&netstamp_needed);
1452 EXPORT_SYMBOL(net_enable_timestamp);
1454 void net_disable_timestamp(void)
1456 atomic_dec(&netstamp_needed);
1458 EXPORT_SYMBOL(net_disable_timestamp);
1460 static inline void net_timestamp_set(struct sk_buff *skb)
1462 if (atomic_read(&netstamp_needed))
1463 __net_timestamp(skb);
1465 skb->tstamp.tv64 = 0;
1468 static inline void net_timestamp_check(struct sk_buff *skb)
1470 if (!skb->tstamp.tv64 && atomic_read(&netstamp_needed))
1471 __net_timestamp(skb);
1475 * dev_forward_skb - loopback an skb to another netif
1477 * @dev: destination network device
1478 * @skb: buffer to forward
1481 * NET_RX_SUCCESS (no congestion)
1482 * NET_RX_DROP (packet was dropped, but freed)
1484 * dev_forward_skb can be used for injecting an skb from the
1485 * start_xmit function of one device into the receive queue
1486 * of another device.
1488 * The receiving device may be in another namespace, so
1489 * we have to clear all information in the skb that could
1490 * impact namespace isolation.
1492 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1496 if (!(dev->flags & IFF_UP) ||
1497 (skb->len > (dev->mtu + dev->hard_header_len))) {
1501 skb_set_dev(skb, dev);
1502 skb->tstamp.tv64 = 0;
1503 skb->pkt_type = PACKET_HOST;
1504 skb->protocol = eth_type_trans(skb, dev);
1505 return netif_rx(skb);
1507 EXPORT_SYMBOL_GPL(dev_forward_skb);
1510 * Support routine. Sends outgoing frames to any network
1511 * taps currently in use.
1514 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1516 struct packet_type *ptype;
1518 #ifdef CONFIG_NET_CLS_ACT
1519 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1520 net_timestamp_set(skb);
1522 net_timestamp_set(skb);
1526 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1527 /* Never send packets back to the socket
1528 * they originated from - MvS (miquels@drinkel.ow.org)
1530 if ((ptype->dev == dev || !ptype->dev) &&
1531 (ptype->af_packet_priv == NULL ||
1532 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1533 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1537 /* skb->nh should be correctly
1538 set by sender, so that the second statement is
1539 just protection against buggy protocols.
1541 skb_reset_mac_header(skb2);
1543 if (skb_network_header(skb2) < skb2->data ||
1544 skb2->network_header > skb2->tail) {
1545 if (net_ratelimit())
1546 printk(KERN_CRIT "protocol %04x is "
1548 skb2->protocol, dev->name);
1549 skb_reset_network_header(skb2);
1552 skb2->transport_header = skb2->network_header;
1553 skb2->pkt_type = PACKET_OUTGOING;
1554 ptype->func(skb2, skb->dev, ptype, skb->dev);
1561 static inline void __netif_reschedule(struct Qdisc *q)
1563 struct softnet_data *sd;
1564 unsigned long flags;
1566 local_irq_save(flags);
1567 sd = &__get_cpu_var(softnet_data);
1568 q->next_sched = NULL;
1569 *sd->output_queue_tailp = q;
1570 sd->output_queue_tailp = &q->next_sched;
1571 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1572 local_irq_restore(flags);
1575 void __netif_schedule(struct Qdisc *q)
1577 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1578 __netif_reschedule(q);
1580 EXPORT_SYMBOL(__netif_schedule);
1582 void dev_kfree_skb_irq(struct sk_buff *skb)
1584 if (atomic_dec_and_test(&skb->users)) {
1585 struct softnet_data *sd;
1586 unsigned long flags;
1588 local_irq_save(flags);
1589 sd = &__get_cpu_var(softnet_data);
1590 skb->next = sd->completion_queue;
1591 sd->completion_queue = skb;
1592 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1593 local_irq_restore(flags);
1596 EXPORT_SYMBOL(dev_kfree_skb_irq);
1598 void dev_kfree_skb_any(struct sk_buff *skb)
1600 if (in_irq() || irqs_disabled())
1601 dev_kfree_skb_irq(skb);
1605 EXPORT_SYMBOL(dev_kfree_skb_any);
1609 * netif_device_detach - mark device as removed
1610 * @dev: network device
1612 * Mark device as removed from system and therefore no longer available.
1614 void netif_device_detach(struct net_device *dev)
1616 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1617 netif_running(dev)) {
1618 netif_tx_stop_all_queues(dev);
1621 EXPORT_SYMBOL(netif_device_detach);
1624 * netif_device_attach - mark device as attached
1625 * @dev: network device
1627 * Mark device as attached from system and restart if needed.
1629 void netif_device_attach(struct net_device *dev)
1631 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1632 netif_running(dev)) {
1633 netif_tx_wake_all_queues(dev);
1634 __netdev_watchdog_up(dev);
1637 EXPORT_SYMBOL(netif_device_attach);
1639 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1641 return ((features & NETIF_F_GEN_CSUM) ||
1642 ((features & NETIF_F_IP_CSUM) &&
1643 protocol == htons(ETH_P_IP)) ||
1644 ((features & NETIF_F_IPV6_CSUM) &&
1645 protocol == htons(ETH_P_IPV6)) ||
1646 ((features & NETIF_F_FCOE_CRC) &&
1647 protocol == htons(ETH_P_FCOE)));
1650 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1652 if (can_checksum_protocol(dev->features, skb->protocol))
1655 if (skb->protocol == htons(ETH_P_8021Q)) {
1656 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1657 if (can_checksum_protocol(dev->features & dev->vlan_features,
1658 veh->h_vlan_encapsulated_proto))
1666 * skb_dev_set -- assign a new device to a buffer
1667 * @skb: buffer for the new device
1668 * @dev: network device
1670 * If an skb is owned by a device already, we have to reset
1671 * all data private to the namespace a device belongs to
1672 * before assigning it a new device.
1674 #ifdef CONFIG_NET_NS
1675 void skb_set_dev(struct sk_buff *skb, struct net_device *dev)
1678 if (skb->dev && !net_eq(dev_net(skb->dev), dev_net(dev))) {
1681 skb_init_secmark(skb);
1685 skb->ipvs_property = 0;
1686 #ifdef CONFIG_NET_SCHED
1692 EXPORT_SYMBOL(skb_set_dev);
1693 #endif /* CONFIG_NET_NS */
1696 * Invalidate hardware checksum when packet is to be mangled, and
1697 * complete checksum manually on outgoing path.
1699 int skb_checksum_help(struct sk_buff *skb)
1702 int ret = 0, offset;
1704 if (skb->ip_summed == CHECKSUM_COMPLETE)
1705 goto out_set_summed;
1707 if (unlikely(skb_shinfo(skb)->gso_size)) {
1708 /* Let GSO fix up the checksum. */
1709 goto out_set_summed;
1712 offset = skb->csum_start - skb_headroom(skb);
1713 BUG_ON(offset >= skb_headlen(skb));
1714 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1716 offset += skb->csum_offset;
1717 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1719 if (skb_cloned(skb) &&
1720 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1721 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1726 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1728 skb->ip_summed = CHECKSUM_NONE;
1732 EXPORT_SYMBOL(skb_checksum_help);
1735 * skb_gso_segment - Perform segmentation on skb.
1736 * @skb: buffer to segment
1737 * @features: features for the output path (see dev->features)
1739 * This function segments the given skb and returns a list of segments.
1741 * It may return NULL if the skb requires no segmentation. This is
1742 * only possible when GSO is used for verifying header integrity.
1744 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1746 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1747 struct packet_type *ptype;
1748 __be16 type = skb->protocol;
1751 skb_reset_mac_header(skb);
1752 skb->mac_len = skb->network_header - skb->mac_header;
1753 __skb_pull(skb, skb->mac_len);
1755 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1756 struct net_device *dev = skb->dev;
1757 struct ethtool_drvinfo info = {};
1759 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1760 dev->ethtool_ops->get_drvinfo(dev, &info);
1762 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1764 info.driver, dev ? dev->features : 0L,
1765 skb->sk ? skb->sk->sk_route_caps : 0L,
1766 skb->len, skb->data_len, skb->ip_summed);
1768 if (skb_header_cloned(skb) &&
1769 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1770 return ERR_PTR(err);
1774 list_for_each_entry_rcu(ptype,
1775 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1776 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1777 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1778 err = ptype->gso_send_check(skb);
1779 segs = ERR_PTR(err);
1780 if (err || skb_gso_ok(skb, features))
1782 __skb_push(skb, (skb->data -
1783 skb_network_header(skb)));
1785 segs = ptype->gso_segment(skb, features);
1791 __skb_push(skb, skb->data - skb_mac_header(skb));
1795 EXPORT_SYMBOL(skb_gso_segment);
1797 /* Take action when hardware reception checksum errors are detected. */
1799 void netdev_rx_csum_fault(struct net_device *dev)
1801 if (net_ratelimit()) {
1802 printk(KERN_ERR "%s: hw csum failure.\n",
1803 dev ? dev->name : "<unknown>");
1807 EXPORT_SYMBOL(netdev_rx_csum_fault);
1810 /* Actually, we should eliminate this check as soon as we know, that:
1811 * 1. IOMMU is present and allows to map all the memory.
1812 * 2. No high memory really exists on this machine.
1815 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1817 #ifdef CONFIG_HIGHMEM
1819 if (!(dev->features & NETIF_F_HIGHDMA)) {
1820 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1821 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1825 if (PCI_DMA_BUS_IS_PHYS) {
1826 struct device *pdev = dev->dev.parent;
1830 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1831 dma_addr_t addr = page_to_phys(skb_shinfo(skb)->frags[i].page);
1832 if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask)
1841 void (*destructor)(struct sk_buff *skb);
1844 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1846 static void dev_gso_skb_destructor(struct sk_buff *skb)
1848 struct dev_gso_cb *cb;
1851 struct sk_buff *nskb = skb->next;
1853 skb->next = nskb->next;
1856 } while (skb->next);
1858 cb = DEV_GSO_CB(skb);
1860 cb->destructor(skb);
1864 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1865 * @skb: buffer to segment
1867 * This function segments the given skb and stores the list of segments
1870 static int dev_gso_segment(struct sk_buff *skb)
1872 struct net_device *dev = skb->dev;
1873 struct sk_buff *segs;
1874 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1877 segs = skb_gso_segment(skb, features);
1879 /* Verifying header integrity only. */
1884 return PTR_ERR(segs);
1887 DEV_GSO_CB(skb)->destructor = skb->destructor;
1888 skb->destructor = dev_gso_skb_destructor;
1894 * Try to orphan skb early, right before transmission by the device.
1895 * We cannot orphan skb if tx timestamp is requested, since
1896 * drivers need to call skb_tstamp_tx() to send the timestamp.
1898 static inline void skb_orphan_try(struct sk_buff *skb)
1900 if (!skb_tx(skb)->flags)
1904 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1905 struct netdev_queue *txq)
1907 const struct net_device_ops *ops = dev->netdev_ops;
1908 int rc = NETDEV_TX_OK;
1910 if (likely(!skb->next)) {
1911 if (!list_empty(&ptype_all))
1912 dev_queue_xmit_nit(skb, dev);
1915 * If device doesnt need skb->dst, release it right now while
1916 * its hot in this cpu cache
1918 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1921 skb_orphan_try(skb);
1923 if (netif_needs_gso(dev, skb)) {
1924 if (unlikely(dev_gso_segment(skb)))
1930 rc = ops->ndo_start_xmit(skb, dev);
1931 if (rc == NETDEV_TX_OK)
1932 txq_trans_update(txq);
1938 struct sk_buff *nskb = skb->next;
1940 skb->next = nskb->next;
1944 * If device doesnt need nskb->dst, release it right now while
1945 * its hot in this cpu cache
1947 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1950 rc = ops->ndo_start_xmit(nskb, dev);
1951 if (unlikely(rc != NETDEV_TX_OK)) {
1952 if (rc & ~NETDEV_TX_MASK)
1953 goto out_kfree_gso_skb;
1954 nskb->next = skb->next;
1958 txq_trans_update(txq);
1959 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1960 return NETDEV_TX_BUSY;
1961 } while (skb->next);
1964 if (likely(skb->next == NULL))
1965 skb->destructor = DEV_GSO_CB(skb)->destructor;
1971 static u32 hashrnd __read_mostly;
1973 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1977 if (skb_rx_queue_recorded(skb)) {
1978 hash = skb_get_rx_queue(skb);
1979 while (unlikely(hash >= dev->real_num_tx_queues))
1980 hash -= dev->real_num_tx_queues;
1984 if (skb->sk && skb->sk->sk_hash)
1985 hash = skb->sk->sk_hash;
1987 hash = (__force u16) skb->protocol;
1989 hash = jhash_1word(hash, hashrnd);
1991 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1993 EXPORT_SYMBOL(skb_tx_hash);
1995 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1997 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1998 if (net_ratelimit()) {
1999 pr_warning("%s selects TX queue %d, but "
2000 "real number of TX queues is %d\n",
2001 dev->name, queue_index, dev->real_num_tx_queues);
2008 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
2009 struct sk_buff *skb)
2012 struct sock *sk = skb->sk;
2014 if (sk_tx_queue_recorded(sk)) {
2015 queue_index = sk_tx_queue_get(sk);
2017 const struct net_device_ops *ops = dev->netdev_ops;
2019 if (ops->ndo_select_queue) {
2020 queue_index = ops->ndo_select_queue(dev, skb);
2021 queue_index = dev_cap_txqueue(dev, queue_index);
2024 if (dev->real_num_tx_queues > 1)
2025 queue_index = skb_tx_hash(dev, skb);
2028 struct dst_entry *dst = rcu_dereference_check(sk->sk_dst_cache, 1);
2030 if (dst && skb_dst(skb) == dst)
2031 sk_tx_queue_set(sk, queue_index);
2036 skb_set_queue_mapping(skb, queue_index);
2037 return netdev_get_tx_queue(dev, queue_index);
2040 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
2041 struct net_device *dev,
2042 struct netdev_queue *txq)
2044 spinlock_t *root_lock = qdisc_lock(q);
2047 spin_lock(root_lock);
2048 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
2051 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
2052 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
2054 * This is a work-conserving queue; there are no old skbs
2055 * waiting to be sent out; and the qdisc is not running -
2056 * xmit the skb directly.
2058 if (!(dev->priv_flags & IFF_XMIT_DST_RELEASE))
2060 __qdisc_update_bstats(q, skb->len);
2061 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
2064 clear_bit(__QDISC_STATE_RUNNING, &q->state);
2066 rc = NET_XMIT_SUCCESS;
2069 rc = qdisc_enqueue_root(skb, q);
2072 spin_unlock(root_lock);
2078 * Returns true if either:
2079 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2080 * 2. skb is fragmented and the device does not support SG, or if
2081 * at least one of fragments is in highmem and device does not
2082 * support DMA from it.
2084 static inline int skb_needs_linearize(struct sk_buff *skb,
2085 struct net_device *dev)
2087 return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
2088 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
2089 illegal_highdma(dev, skb)));
2093 * dev_queue_xmit - transmit a buffer
2094 * @skb: buffer to transmit
2096 * Queue a buffer for transmission to a network device. The caller must
2097 * have set the device and priority and built the buffer before calling
2098 * this function. The function can be called from an interrupt.
2100 * A negative errno code is returned on a failure. A success does not
2101 * guarantee the frame will be transmitted as it may be dropped due
2102 * to congestion or traffic shaping.
2104 * -----------------------------------------------------------------------------------
2105 * I notice this method can also return errors from the queue disciplines,
2106 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2109 * Regardless of the return value, the skb is consumed, so it is currently
2110 * difficult to retry a send to this method. (You can bump the ref count
2111 * before sending to hold a reference for retry if you are careful.)
2113 * When calling this method, interrupts MUST be enabled. This is because
2114 * the BH enable code must have IRQs enabled so that it will not deadlock.
2117 int dev_queue_xmit(struct sk_buff *skb)
2119 struct net_device *dev = skb->dev;
2120 struct netdev_queue *txq;
2124 /* GSO will handle the following emulations directly. */
2125 if (netif_needs_gso(dev, skb))
2128 /* Convert a paged skb to linear, if required */
2129 if (skb_needs_linearize(skb, dev) && __skb_linearize(skb))
2132 /* If packet is not checksummed and device does not support
2133 * checksumming for this protocol, complete checksumming here.
2135 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2136 skb_set_transport_header(skb, skb->csum_start -
2138 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2143 /* Disable soft irqs for various locks below. Also
2144 * stops preemption for RCU.
2148 txq = dev_pick_tx(dev, skb);
2149 q = rcu_dereference_bh(txq->qdisc);
2151 #ifdef CONFIG_NET_CLS_ACT
2152 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2155 rc = __dev_xmit_skb(skb, q, dev, txq);
2159 /* The device has no queue. Common case for software devices:
2160 loopback, all the sorts of tunnels...
2162 Really, it is unlikely that netif_tx_lock protection is necessary
2163 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2165 However, it is possible, that they rely on protection
2168 Check this and shot the lock. It is not prone from deadlocks.
2169 Either shot noqueue qdisc, it is even simpler 8)
2171 if (dev->flags & IFF_UP) {
2172 int cpu = smp_processor_id(); /* ok because BHs are off */
2174 if (txq->xmit_lock_owner != cpu) {
2176 HARD_TX_LOCK(dev, txq, cpu);
2178 if (!netif_tx_queue_stopped(txq)) {
2179 rc = dev_hard_start_xmit(skb, dev, txq);
2180 if (dev_xmit_complete(rc)) {
2181 HARD_TX_UNLOCK(dev, txq);
2185 HARD_TX_UNLOCK(dev, txq);
2186 if (net_ratelimit())
2187 printk(KERN_CRIT "Virtual device %s asks to "
2188 "queue packet!\n", dev->name);
2190 /* Recursion is detected! It is possible,
2192 if (net_ratelimit())
2193 printk(KERN_CRIT "Dead loop on virtual device "
2194 "%s, fix it urgently!\n", dev->name);
2199 rcu_read_unlock_bh();
2205 rcu_read_unlock_bh();
2208 EXPORT_SYMBOL(dev_queue_xmit);
2211 /*=======================================================================
2213 =======================================================================*/
2215 int netdev_max_backlog __read_mostly = 1000;
2216 int netdev_tstamp_prequeue __read_mostly = 1;
2217 int netdev_budget __read_mostly = 300;
2218 int weight_p __read_mostly = 64; /* old backlog weight */
2220 /* Called with irq disabled */
2221 static inline void ____napi_schedule(struct softnet_data *sd,
2222 struct napi_struct *napi)
2224 list_add_tail(&napi->poll_list, &sd->poll_list);
2225 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2230 /* One global table that all flow-based protocols share. */
2231 struct rps_sock_flow_table *rps_sock_flow_table __read_mostly;
2232 EXPORT_SYMBOL(rps_sock_flow_table);
2235 * get_rps_cpu is called from netif_receive_skb and returns the target
2236 * CPU from the RPS map of the receiving queue for a given skb.
2237 * rcu_read_lock must be held on entry.
2239 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
2240 struct rps_dev_flow **rflowp)
2242 struct ipv6hdr *ip6;
2244 struct netdev_rx_queue *rxqueue;
2245 struct rps_map *map;
2246 struct rps_dev_flow_table *flow_table;
2247 struct rps_sock_flow_table *sock_flow_table;
2251 u32 addr1, addr2, ihl;
2257 if (skb_rx_queue_recorded(skb)) {
2258 u16 index = skb_get_rx_queue(skb);
2259 if (unlikely(index >= dev->num_rx_queues)) {
2260 if (net_ratelimit()) {
2261 pr_warning("%s received packet on queue "
2262 "%u, but number of RX queues is %u\n",
2263 dev->name, index, dev->num_rx_queues);
2267 rxqueue = dev->_rx + index;
2271 if (!rxqueue->rps_map && !rxqueue->rps_flow_table)
2275 goto got_hash; /* Skip hash computation on packet header */
2277 switch (skb->protocol) {
2278 case __constant_htons(ETH_P_IP):
2279 if (!pskb_may_pull(skb, sizeof(*ip)))
2282 ip = (struct iphdr *) skb->data;
2283 ip_proto = ip->protocol;
2284 addr1 = (__force u32) ip->saddr;
2285 addr2 = (__force u32) ip->daddr;
2288 case __constant_htons(ETH_P_IPV6):
2289 if (!pskb_may_pull(skb, sizeof(*ip6)))
2292 ip6 = (struct ipv6hdr *) skb->data;
2293 ip_proto = ip6->nexthdr;
2294 addr1 = (__force u32) ip6->saddr.s6_addr32[3];
2295 addr2 = (__force u32) ip6->daddr.s6_addr32[3];
2308 case IPPROTO_UDPLITE:
2309 if (pskb_may_pull(skb, (ihl * 4) + 4)) {
2310 ports.v32 = * (__force u32 *) (skb->data + (ihl * 4));
2311 if (ports.v16[1] < ports.v16[0])
2312 swap(ports.v16[0], ports.v16[1]);
2320 /* get a consistent hash (same value on both flow directions) */
2323 skb->rxhash = jhash_3words(addr1, addr2, ports.v32, hashrnd);
2328 flow_table = rcu_dereference(rxqueue->rps_flow_table);
2329 sock_flow_table = rcu_dereference(rps_sock_flow_table);
2330 if (flow_table && sock_flow_table) {
2332 struct rps_dev_flow *rflow;
2334 rflow = &flow_table->flows[skb->rxhash & flow_table->mask];
2337 next_cpu = sock_flow_table->ents[skb->rxhash &
2338 sock_flow_table->mask];
2341 * If the desired CPU (where last recvmsg was done) is
2342 * different from current CPU (one in the rx-queue flow
2343 * table entry), switch if one of the following holds:
2344 * - Current CPU is unset (equal to RPS_NO_CPU).
2345 * - Current CPU is offline.
2346 * - The current CPU's queue tail has advanced beyond the
2347 * last packet that was enqueued using this table entry.
2348 * This guarantees that all previous packets for the flow
2349 * have been dequeued, thus preserving in order delivery.
2351 if (unlikely(tcpu != next_cpu) &&
2352 (tcpu == RPS_NO_CPU || !cpu_online(tcpu) ||
2353 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
2354 rflow->last_qtail)) >= 0)) {
2355 tcpu = rflow->cpu = next_cpu;
2356 if (tcpu != RPS_NO_CPU)
2357 rflow->last_qtail = per_cpu(softnet_data,
2358 tcpu).input_queue_head;
2360 if (tcpu != RPS_NO_CPU && cpu_online(tcpu)) {
2367 map = rcu_dereference(rxqueue->rps_map);
2369 tcpu = map->cpus[((u64) skb->rxhash * map->len) >> 32];
2371 if (cpu_online(tcpu)) {
2381 /* Called from hardirq (IPI) context */
2382 static void rps_trigger_softirq(void *data)
2384 struct softnet_data *sd = data;
2386 ____napi_schedule(sd, &sd->backlog);
2390 #endif /* CONFIG_RPS */
2393 * Check if this softnet_data structure is another cpu one
2394 * If yes, queue it to our IPI list and return 1
2397 static int rps_ipi_queued(struct softnet_data *sd)
2400 struct softnet_data *mysd = &__get_cpu_var(softnet_data);
2403 sd->rps_ipi_next = mysd->rps_ipi_list;
2404 mysd->rps_ipi_list = sd;
2406 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2409 #endif /* CONFIG_RPS */
2414 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2415 * queue (may be a remote CPU queue).
2417 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
2418 unsigned int *qtail)
2420 struct softnet_data *sd;
2421 unsigned long flags;
2423 sd = &per_cpu(softnet_data, cpu);
2425 local_irq_save(flags);
2428 if (skb_queue_len(&sd->input_pkt_queue) <= netdev_max_backlog) {
2429 if (skb_queue_len(&sd->input_pkt_queue)) {
2431 __skb_queue_tail(&sd->input_pkt_queue, skb);
2432 input_queue_tail_incr_save(sd, qtail);
2434 local_irq_restore(flags);
2435 return NET_RX_SUCCESS;
2438 /* Schedule NAPI for backlog device
2439 * We can use non atomic operation since we own the queue lock
2441 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
2442 if (!rps_ipi_queued(sd))
2443 ____napi_schedule(sd, &sd->backlog);
2451 local_irq_restore(flags);
2458 * netif_rx - post buffer to the network code
2459 * @skb: buffer to post
2461 * This function receives a packet from a device driver and queues it for
2462 * the upper (protocol) levels to process. It always succeeds. The buffer
2463 * may be dropped during processing for congestion control or by the
2467 * NET_RX_SUCCESS (no congestion)
2468 * NET_RX_DROP (packet was dropped)
2472 int netif_rx(struct sk_buff *skb)
2476 /* if netpoll wants it, pretend we never saw it */
2477 if (netpoll_rx(skb))
2480 if (netdev_tstamp_prequeue)
2481 net_timestamp_check(skb);
2485 struct rps_dev_flow voidflow, *rflow = &voidflow;
2490 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2492 cpu = smp_processor_id();
2494 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2501 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
2507 EXPORT_SYMBOL(netif_rx);
2509 int netif_rx_ni(struct sk_buff *skb)
2514 err = netif_rx(skb);
2515 if (local_softirq_pending())
2521 EXPORT_SYMBOL(netif_rx_ni);
2523 static void net_tx_action(struct softirq_action *h)
2525 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2527 if (sd->completion_queue) {
2528 struct sk_buff *clist;
2530 local_irq_disable();
2531 clist = sd->completion_queue;
2532 sd->completion_queue = NULL;
2536 struct sk_buff *skb = clist;
2537 clist = clist->next;
2539 WARN_ON(atomic_read(&skb->users));
2544 if (sd->output_queue) {
2547 local_irq_disable();
2548 head = sd->output_queue;
2549 sd->output_queue = NULL;
2550 sd->output_queue_tailp = &sd->output_queue;
2554 struct Qdisc *q = head;
2555 spinlock_t *root_lock;
2557 head = head->next_sched;
2559 root_lock = qdisc_lock(q);
2560 if (spin_trylock(root_lock)) {
2561 smp_mb__before_clear_bit();
2562 clear_bit(__QDISC_STATE_SCHED,
2565 spin_unlock(root_lock);
2567 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2569 __netif_reschedule(q);
2571 smp_mb__before_clear_bit();
2572 clear_bit(__QDISC_STATE_SCHED,
2580 static inline int deliver_skb(struct sk_buff *skb,
2581 struct packet_type *pt_prev,
2582 struct net_device *orig_dev)
2584 atomic_inc(&skb->users);
2585 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2588 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2590 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2591 /* This hook is defined here for ATM LANE */
2592 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2593 unsigned char *addr) __read_mostly;
2594 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2598 * If bridge module is loaded call bridging hook.
2599 * returns NULL if packet was consumed.
2601 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2602 struct sk_buff *skb) __read_mostly;
2603 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2605 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2606 struct packet_type **pt_prev, int *ret,
2607 struct net_device *orig_dev)
2609 struct net_bridge_port *port;
2611 if (skb->pkt_type == PACKET_LOOPBACK ||
2612 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2616 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2620 return br_handle_frame_hook(port, skb);
2623 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2626 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2627 struct sk_buff *(*macvlan_handle_frame_hook)(struct macvlan_port *p,
2628 struct sk_buff *skb) __read_mostly;
2629 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2631 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2632 struct packet_type **pt_prev,
2634 struct net_device *orig_dev)
2636 struct macvlan_port *port;
2638 port = rcu_dereference(skb->dev->macvlan_port);
2643 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2646 return macvlan_handle_frame_hook(port, skb);
2649 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2652 #ifdef CONFIG_NET_CLS_ACT
2653 /* TODO: Maybe we should just force sch_ingress to be compiled in
2654 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2655 * a compare and 2 stores extra right now if we dont have it on
2656 * but have CONFIG_NET_CLS_ACT
2657 * NOTE: This doesnt stop any functionality; if you dont have
2658 * the ingress scheduler, you just cant add policies on ingress.
2661 static int ing_filter(struct sk_buff *skb)
2663 struct net_device *dev = skb->dev;
2664 u32 ttl = G_TC_RTTL(skb->tc_verd);
2665 struct netdev_queue *rxq;
2666 int result = TC_ACT_OK;
2669 if (MAX_RED_LOOP < ttl++) {
2671 "Redir loop detected Dropping packet (%d->%d)\n",
2672 skb->skb_iif, dev->ifindex);
2676 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2677 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2679 rxq = &dev->rx_queue;
2682 if (q != &noop_qdisc) {
2683 spin_lock(qdisc_lock(q));
2684 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2685 result = qdisc_enqueue_root(skb, q);
2686 spin_unlock(qdisc_lock(q));
2692 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2693 struct packet_type **pt_prev,
2694 int *ret, struct net_device *orig_dev)
2696 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2700 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2703 /* Huh? Why does turning on AF_PACKET affect this? */
2704 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2707 switch (ing_filter(skb)) {
2721 * netif_nit_deliver - deliver received packets to network taps
2724 * This function is used to deliver incoming packets to network
2725 * taps. It should be used when the normal netif_receive_skb path
2726 * is bypassed, for example because of VLAN acceleration.
2728 void netif_nit_deliver(struct sk_buff *skb)
2730 struct packet_type *ptype;
2732 if (list_empty(&ptype_all))
2735 skb_reset_network_header(skb);
2736 skb_reset_transport_header(skb);
2737 skb->mac_len = skb->network_header - skb->mac_header;
2740 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2741 if (!ptype->dev || ptype->dev == skb->dev)
2742 deliver_skb(skb, ptype, skb->dev);
2747 static inline void skb_bond_set_mac_by_master(struct sk_buff *skb,
2748 struct net_device *master)
2750 if (skb->pkt_type == PACKET_HOST) {
2751 u16 *dest = (u16 *) eth_hdr(skb)->h_dest;
2753 memcpy(dest, master->dev_addr, ETH_ALEN);
2757 /* On bonding slaves other than the currently active slave, suppress
2758 * duplicates except for 802.3ad ETH_P_SLOW, alb non-mcast/bcast, and
2759 * ARP on active-backup slaves with arp_validate enabled.
2761 int __skb_bond_should_drop(struct sk_buff *skb, struct net_device *master)
2763 struct net_device *dev = skb->dev;
2765 if (master->priv_flags & IFF_MASTER_ARPMON)
2766 dev->last_rx = jiffies;
2768 if ((master->priv_flags & IFF_MASTER_ALB) && master->br_port) {
2769 /* Do address unmangle. The local destination address
2770 * will be always the one master has. Provides the right
2771 * functionality in a bridge.
2773 skb_bond_set_mac_by_master(skb, master);
2776 if (dev->priv_flags & IFF_SLAVE_INACTIVE) {
2777 if ((dev->priv_flags & IFF_SLAVE_NEEDARP) &&
2778 skb->protocol == __cpu_to_be16(ETH_P_ARP))
2781 if (master->priv_flags & IFF_MASTER_ALB) {
2782 if (skb->pkt_type != PACKET_BROADCAST &&
2783 skb->pkt_type != PACKET_MULTICAST)
2786 if (master->priv_flags & IFF_MASTER_8023AD &&
2787 skb->protocol == __cpu_to_be16(ETH_P_SLOW))
2794 EXPORT_SYMBOL(__skb_bond_should_drop);
2797 * Filter the protocols for which the reserves are adequate.
2799 * Before adding a protocol make sure that it is either covered by the existing
2800 * reserves, or add reserves covering the memory need of the new protocol's
2801 * packet processing.
2803 static int skb_emergency_protocol(struct sk_buff *skb)
2805 if (skb_emergency(skb))
2806 switch (skb->protocol) {
2807 case __constant_htons(ETH_P_ARP):
2808 case __constant_htons(ETH_P_IP):
2809 case __constant_htons(ETH_P_IPV6):
2810 case __constant_htons(ETH_P_8021Q):
2820 static int __netif_receive_skb(struct sk_buff *skb)
2822 struct packet_type *ptype, *pt_prev;
2823 struct net_device *orig_dev;
2824 struct net_device *master;
2825 struct net_device *null_or_orig;
2826 struct net_device *orig_or_bond;
2827 int ret = NET_RX_DROP;
2829 unsigned long pflags = current->flags;
2831 if (!netdev_tstamp_prequeue)
2832 net_timestamp_check(skb);
2834 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2835 return NET_RX_SUCCESS;
2837 /* Emergency skb are special, they should
2838 * - be delivered to SOCK_MEMALLOC sockets only
2839 * - stay away from userspace
2840 * - have bounded memory usage
2842 * Use PF_MEMALLOC as a poor mans memory pool - the grouping kind.
2843 * This saves us from propagating the allocation context down to all
2846 if (skb_emergency(skb))
2847 current->flags |= PF_MEMALLOC;
2849 /* if we've gotten here through NAPI, check netpoll */
2850 if (netpoll_receive_skb(skb))
2854 skb->skb_iif = skb->dev->ifindex;
2856 null_or_orig = NULL;
2857 orig_dev = skb->dev;
2858 master = ACCESS_ONCE(orig_dev->master);
2860 if (skb_bond_should_drop(skb, master))
2861 null_or_orig = orig_dev; /* deliver only exact match */
2866 __get_cpu_var(softnet_data).processed++;
2868 skb_reset_network_header(skb);
2869 skb_reset_transport_header(skb);
2870 skb->mac_len = skb->network_header - skb->mac_header;
2876 #ifdef CONFIG_NET_CLS_ACT
2877 if (skb->tc_verd & TC_NCLS) {
2878 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2883 if (skb_emergency(skb))
2886 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2887 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2888 ptype->dev == orig_dev) {
2890 ret = deliver_skb(skb, pt_prev, orig_dev);
2896 #ifdef CONFIG_NET_CLS_ACT
2897 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2903 if (!skb_emergency_protocol(skb))
2906 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2909 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2914 * Make sure frames received on VLAN interfaces stacked on
2915 * bonding interfaces still make their way to any base bonding
2916 * device that may have registered for a specific ptype. The
2917 * handler may have to adjust skb->dev and orig_dev.
2919 orig_or_bond = orig_dev;
2920 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2921 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2922 orig_or_bond = vlan_dev_real_dev(skb->dev);
2925 type = skb->protocol;
2926 list_for_each_entry_rcu(ptype,
2927 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2928 if (ptype->type == type && (ptype->dev == null_or_orig ||
2929 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2930 ptype->dev == orig_or_bond)) {
2932 ret = deliver_skb(skb, pt_prev, orig_dev);
2938 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2942 /* Jamal, now you will not able to escape explaining
2943 * me how you were going to use this. :-)
2951 tsk_restore_flags(current, pflags, PF_MEMALLOC);
2956 * netif_receive_skb - process receive buffer from network
2957 * @skb: buffer to process
2959 * netif_receive_skb() is the main receive data processing function.
2960 * It always succeeds. The buffer may be dropped during processing
2961 * for congestion control or by the protocol layers.
2963 * This function may only be called from softirq context and interrupts
2964 * should be enabled.
2966 * Return values (usually ignored):
2967 * NET_RX_SUCCESS: no congestion
2968 * NET_RX_DROP: packet was dropped
2970 int netif_receive_skb(struct sk_buff *skb)
2972 if (netdev_tstamp_prequeue)
2973 net_timestamp_check(skb);
2977 struct rps_dev_flow voidflow, *rflow = &voidflow;
2982 cpu = get_rps_cpu(skb->dev, skb, &rflow);
2985 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
2989 ret = __netif_receive_skb(skb);
2995 return __netif_receive_skb(skb);
2998 EXPORT_SYMBOL(netif_receive_skb);
3000 /* Network device is going away, flush any packets still pending
3001 * Called with irqs disabled.
3003 static void flush_backlog(void *arg)
3005 struct net_device *dev = arg;
3006 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3007 struct sk_buff *skb, *tmp;
3010 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
3011 if (skb->dev == dev) {
3012 __skb_unlink(skb, &sd->input_pkt_queue);
3014 input_queue_head_incr(sd);
3019 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
3020 if (skb->dev == dev) {
3021 __skb_unlink(skb, &sd->process_queue);
3023 input_queue_head_incr(sd);
3028 static int napi_gro_complete(struct sk_buff *skb)
3030 struct packet_type *ptype;
3031 __be16 type = skb->protocol;
3032 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3035 if (NAPI_GRO_CB(skb)->count == 1) {
3036 skb_shinfo(skb)->gso_size = 0;
3041 list_for_each_entry_rcu(ptype, head, list) {
3042 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
3045 err = ptype->gro_complete(skb);
3051 WARN_ON(&ptype->list == head);
3053 return NET_RX_SUCCESS;
3057 return netif_receive_skb(skb);
3060 static void napi_gro_flush(struct napi_struct *napi)
3062 struct sk_buff *skb, *next;
3064 for (skb = napi->gro_list; skb; skb = next) {
3067 napi_gro_complete(skb);
3070 napi->gro_count = 0;
3071 napi->gro_list = NULL;
3074 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3076 struct sk_buff **pp = NULL;
3077 struct packet_type *ptype;
3078 __be16 type = skb->protocol;
3079 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
3082 enum gro_result ret;
3084 if (!(skb->dev->features & NETIF_F_GRO))
3087 if (skb_is_gso(skb) || skb_has_frags(skb))
3091 list_for_each_entry_rcu(ptype, head, list) {
3092 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
3095 skb_set_network_header(skb, skb_gro_offset(skb));
3096 mac_len = skb->network_header - skb->mac_header;
3097 skb->mac_len = mac_len;
3098 NAPI_GRO_CB(skb)->same_flow = 0;
3099 NAPI_GRO_CB(skb)->flush = 0;
3100 NAPI_GRO_CB(skb)->free = 0;
3102 pp = ptype->gro_receive(&napi->gro_list, skb);
3107 if (&ptype->list == head)
3110 same_flow = NAPI_GRO_CB(skb)->same_flow;
3111 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
3114 struct sk_buff *nskb = *pp;
3118 napi_gro_complete(nskb);
3125 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
3129 NAPI_GRO_CB(skb)->count = 1;
3130 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
3131 skb->next = napi->gro_list;
3132 napi->gro_list = skb;
3136 if (skb_headlen(skb) < skb_gro_offset(skb)) {
3137 int grow = skb_gro_offset(skb) - skb_headlen(skb);
3139 BUG_ON(skb->end - skb->tail < grow);
3141 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
3144 skb->data_len -= grow;
3146 skb_shinfo(skb)->frags[0].page_offset += grow;
3147 skb_shinfo(skb)->frags[0].size -= grow;
3149 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
3150 put_page(skb_shinfo(skb)->frags[0].page);
3151 memmove(skb_shinfo(skb)->frags,
3152 skb_shinfo(skb)->frags + 1,
3153 --skb_shinfo(skb)->nr_frags);
3164 EXPORT_SYMBOL(dev_gro_receive);
3167 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3171 if (netpoll_rx_on(skb))
3174 for (p = napi->gro_list; p; p = p->next) {
3175 NAPI_GRO_CB(p)->same_flow =
3176 (p->dev == skb->dev) &&
3177 !compare_ether_header(skb_mac_header(p),
3178 skb_gro_mac_header(skb));
3179 NAPI_GRO_CB(p)->flush = 0;
3182 return dev_gro_receive(napi, skb);
3185 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
3189 if (netif_receive_skb(skb))
3194 case GRO_MERGED_FREE:
3205 EXPORT_SYMBOL(napi_skb_finish);
3207 void skb_gro_reset_offset(struct sk_buff *skb)
3209 NAPI_GRO_CB(skb)->data_offset = 0;
3210 NAPI_GRO_CB(skb)->frag0 = NULL;
3211 NAPI_GRO_CB(skb)->frag0_len = 0;
3213 if (skb->mac_header == skb->tail &&
3214 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
3215 NAPI_GRO_CB(skb)->frag0 =
3216 page_address(skb_shinfo(skb)->frags[0].page) +
3217 skb_shinfo(skb)->frags[0].page_offset;
3218 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
3221 EXPORT_SYMBOL(skb_gro_reset_offset);
3223 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
3225 skb_gro_reset_offset(skb);
3227 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
3229 EXPORT_SYMBOL(napi_gro_receive);
3231 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
3233 __skb_pull(skb, skb_headlen(skb));
3234 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
3238 EXPORT_SYMBOL(napi_reuse_skb);
3240 struct sk_buff *napi_get_frags(struct napi_struct *napi)
3242 struct sk_buff *skb = napi->skb;
3245 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
3251 EXPORT_SYMBOL(napi_get_frags);
3253 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
3259 skb->protocol = eth_type_trans(skb, skb->dev);
3261 if (ret == GRO_HELD)
3262 skb_gro_pull(skb, -ETH_HLEN);
3263 else if (netif_receive_skb(skb))
3268 case GRO_MERGED_FREE:
3269 napi_reuse_skb(napi, skb);
3278 EXPORT_SYMBOL(napi_frags_finish);
3280 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
3282 struct sk_buff *skb = napi->skb;
3289 skb_reset_mac_header(skb);
3290 skb_gro_reset_offset(skb);
3292 off = skb_gro_offset(skb);
3293 hlen = off + sizeof(*eth);
3294 eth = skb_gro_header_fast(skb, off);
3295 if (skb_gro_header_hard(skb, hlen)) {
3296 eth = skb_gro_header_slow(skb, hlen, off);
3297 if (unlikely(!eth)) {
3298 napi_reuse_skb(napi, skb);
3304 skb_gro_pull(skb, sizeof(*eth));
3307 * This works because the only protocols we care about don't require
3308 * special handling. We'll fix it up properly at the end.
3310 skb->protocol = eth->h_proto;
3315 EXPORT_SYMBOL(napi_frags_skb);
3317 gro_result_t napi_gro_frags(struct napi_struct *napi)
3319 struct sk_buff *skb = napi_frags_skb(napi);
3324 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
3326 EXPORT_SYMBOL(napi_gro_frags);
3329 * net_rps_action sends any pending IPI's for rps.
3330 * Note: called with local irq disabled, but exits with local irq enabled.
3332 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
3335 struct softnet_data *remsd = sd->rps_ipi_list;
3338 sd->rps_ipi_list = NULL;
3342 /* Send pending IPI's to kick RPS processing on remote cpus. */
3344 struct softnet_data *next = remsd->rps_ipi_next;
3346 if (cpu_online(remsd->cpu))
3347 __smp_call_function_single(remsd->cpu,
3356 static int process_backlog(struct napi_struct *napi, int quota)
3359 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
3362 /* Check if we have pending ipi, its better to send them now,
3363 * not waiting net_rx_action() end.
3365 if (sd->rps_ipi_list) {
3366 local_irq_disable();
3367 net_rps_action_and_irq_enable(sd);
3370 napi->weight = weight_p;
3371 local_irq_disable();
3372 while (work < quota) {
3373 struct sk_buff *skb;
3376 while ((skb = __skb_dequeue(&sd->process_queue))) {
3378 __netif_receive_skb(skb);
3379 local_irq_disable();
3380 input_queue_head_incr(sd);
3381 if (++work >= quota) {
3388 qlen = skb_queue_len(&sd->input_pkt_queue);
3390 skb_queue_splice_tail_init(&sd->input_pkt_queue,
3391 &sd->process_queue);
3393 if (qlen < quota - work) {
3395 * Inline a custom version of __napi_complete().
3396 * only current cpu owns and manipulates this napi,
3397 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3398 * we can use a plain write instead of clear_bit(),
3399 * and we dont need an smp_mb() memory barrier.
3401 list_del(&napi->poll_list);
3404 quota = work + qlen;
3414 * __napi_schedule - schedule for receive
3415 * @n: entry to schedule
3417 * The entry's receive function will be scheduled to run
3419 void __napi_schedule(struct napi_struct *n)
3421 unsigned long flags;
3423 local_irq_save(flags);
3424 ____napi_schedule(&__get_cpu_var(softnet_data), n);
3425 local_irq_restore(flags);
3427 EXPORT_SYMBOL(__napi_schedule);
3429 void __napi_complete(struct napi_struct *n)
3431 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
3432 BUG_ON(n->gro_list);
3434 list_del(&n->poll_list);
3435 smp_mb__before_clear_bit();
3436 clear_bit(NAPI_STATE_SCHED, &n->state);
3438 EXPORT_SYMBOL(__napi_complete);
3440 void napi_complete(struct napi_struct *n)
3442 unsigned long flags;
3445 * don't let napi dequeue from the cpu poll list
3446 * just in case its running on a different cpu
3448 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
3452 local_irq_save(flags);
3454 local_irq_restore(flags);
3456 EXPORT_SYMBOL(napi_complete);
3458 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
3459 int (*poll)(struct napi_struct *, int), int weight)
3461 INIT_LIST_HEAD(&napi->poll_list);
3462 napi->gro_count = 0;
3463 napi->gro_list = NULL;
3466 napi->weight = weight;
3467 list_add(&napi->dev_list, &dev->napi_list);
3469 #ifdef CONFIG_NETPOLL
3470 spin_lock_init(&napi->poll_lock);
3471 napi->poll_owner = -1;
3473 set_bit(NAPI_STATE_SCHED, &napi->state);
3475 EXPORT_SYMBOL(netif_napi_add);
3477 void netif_napi_del(struct napi_struct *napi)
3479 struct sk_buff *skb, *next;
3481 list_del_init(&napi->dev_list);
3482 napi_free_frags(napi);
3484 for (skb = napi->gro_list; skb; skb = next) {
3490 napi->gro_list = NULL;
3491 napi->gro_count = 0;
3493 EXPORT_SYMBOL(netif_napi_del);
3495 static void net_rx_action(struct softirq_action *h)
3497 struct softnet_data *sd = &__get_cpu_var(softnet_data);
3498 unsigned long time_limit = jiffies + 2;
3499 int budget = netdev_budget;
3502 local_irq_disable();
3504 while (!list_empty(&sd->poll_list)) {
3505 struct napi_struct *n;
3508 /* If softirq window is exhuasted then punt.
3509 * Allow this to run for 2 jiffies since which will allow
3510 * an average latency of 1.5/HZ.
3512 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
3517 /* Even though interrupts have been re-enabled, this
3518 * access is safe because interrupts can only add new
3519 * entries to the tail of this list, and only ->poll()
3520 * calls can remove this head entry from the list.
3522 n = list_first_entry(&sd->poll_list, struct napi_struct, poll_list);
3524 have = netpoll_poll_lock(n);
3528 /* This NAPI_STATE_SCHED test is for avoiding a race
3529 * with netpoll's poll_napi(). Only the entity which
3530 * obtains the lock and sees NAPI_STATE_SCHED set will
3531 * actually make the ->poll() call. Therefore we avoid
3532 * accidently calling ->poll() when NAPI is not scheduled.
3535 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3536 work = n->poll(n, weight);
3540 WARN_ON_ONCE(work > weight);
3544 local_irq_disable();
3546 /* Drivers must not modify the NAPI state if they
3547 * consume the entire weight. In such cases this code
3548 * still "owns" the NAPI instance and therefore can
3549 * move the instance around on the list at-will.
3551 if (unlikely(work == weight)) {
3552 if (unlikely(napi_disable_pending(n))) {
3555 local_irq_disable();
3557 list_move_tail(&n->poll_list, &sd->poll_list);
3560 netpoll_poll_unlock(have);
3563 net_rps_action_and_irq_enable(sd);
3565 #ifdef CONFIG_NET_DMA
3567 * There may not be any more sk_buffs coming right now, so push
3568 * any pending DMA copies to hardware
3570 dma_issue_pending_all();
3577 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3581 static gifconf_func_t *gifconf_list[NPROTO];
3584 * register_gifconf - register a SIOCGIF handler
3585 * @family: Address family
3586 * @gifconf: Function handler
3588 * Register protocol dependent address dumping routines. The handler
3589 * that is passed must not be freed or reused until it has been replaced
3590 * by another handler.
3592 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3594 if (family >= NPROTO)
3596 gifconf_list[family] = gifconf;
3599 EXPORT_SYMBOL(register_gifconf);
3603 * Map an interface index to its name (SIOCGIFNAME)
3607 * We need this ioctl for efficient implementation of the
3608 * if_indextoname() function required by the IPv6 API. Without
3609 * it, we would have to search all the interfaces to find a
3613 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3615 struct net_device *dev;
3619 * Fetch the caller's info block.
3622 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3626 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3632 strcpy(ifr.ifr_name, dev->name);
3635 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3641 * Perform a SIOCGIFCONF call. This structure will change
3642 * size eventually, and there is nothing I can do about it.
3643 * Thus we will need a 'compatibility mode'.
3646 static int dev_ifconf(struct net *net, char __user *arg)
3649 struct net_device *dev;
3656 * Fetch the caller's info block.
3659 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3666 * Loop over the interfaces, and write an info block for each.
3670 for_each_netdev(net, dev) {
3671 for (i = 0; i < NPROTO; i++) {
3672 if (gifconf_list[i]) {
3675 done = gifconf_list[i](dev, NULL, 0);
3677 done = gifconf_list[i](dev, pos + total,
3687 * All done. Write the updated control block back to the caller.
3689 ifc.ifc_len = total;
3692 * Both BSD and Solaris return 0 here, so we do too.
3694 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3697 #ifdef CONFIG_PROC_FS
3699 * This is invoked by the /proc filesystem handler to display a device
3702 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3705 struct net *net = seq_file_net(seq);
3707 struct net_device *dev;
3711 return SEQ_START_TOKEN;
3714 for_each_netdev_rcu(net, dev)
3721 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3723 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3724 first_net_device(seq_file_net(seq)) :
3725 next_net_device((struct net_device *)v);
3728 return rcu_dereference(dev);
3731 void dev_seq_stop(struct seq_file *seq, void *v)
3737 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3739 const struct net_device_stats *stats = dev_get_stats(dev);
3741 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3742 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3743 dev->name, stats->rx_bytes, stats->rx_packets,
3745 stats->rx_dropped + stats->rx_missed_errors,
3746 stats->rx_fifo_errors,
3747 stats->rx_length_errors + stats->rx_over_errors +
3748 stats->rx_crc_errors + stats->rx_frame_errors,
3749 stats->rx_compressed, stats->multicast,
3750 stats->tx_bytes, stats->tx_packets,
3751 stats->tx_errors, stats->tx_dropped,
3752 stats->tx_fifo_errors, stats->collisions,
3753 stats->tx_carrier_errors +
3754 stats->tx_aborted_errors +
3755 stats->tx_window_errors +
3756 stats->tx_heartbeat_errors,
3757 stats->tx_compressed);
3761 * Called from the PROCfs module. This now uses the new arbitrary sized
3762 * /proc/net interface to create /proc/net/dev
3764 static int dev_seq_show(struct seq_file *seq, void *v)
3766 if (v == SEQ_START_TOKEN)
3767 seq_puts(seq, "Inter-| Receive "
3769 " face |bytes packets errs drop fifo frame "
3770 "compressed multicast|bytes packets errs "
3771 "drop fifo colls carrier compressed\n");
3773 dev_seq_printf_stats(seq, v);
3777 static struct softnet_data *softnet_get_online(loff_t *pos)
3779 struct softnet_data *sd = NULL;
3781 while (*pos < nr_cpu_ids)
3782 if (cpu_online(*pos)) {
3783 sd = &per_cpu(softnet_data, *pos);
3790 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3792 return softnet_get_online(pos);
3795 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3798 return softnet_get_online(pos);
3801 static void softnet_seq_stop(struct seq_file *seq, void *v)
3805 static int softnet_seq_show(struct seq_file *seq, void *v)
3807 struct softnet_data *sd = v;
3809 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3810 sd->processed, sd->dropped, sd->time_squeeze, 0,
3811 0, 0, 0, 0, /* was fastroute */
3812 sd->cpu_collision, sd->received_rps);
3816 static const struct seq_operations dev_seq_ops = {
3817 .start = dev_seq_start,
3818 .next = dev_seq_next,
3819 .stop = dev_seq_stop,
3820 .show = dev_seq_show,
3823 static int dev_seq_open(struct inode *inode, struct file *file)
3825 return seq_open_net(inode, file, &dev_seq_ops,
3826 sizeof(struct seq_net_private));
3829 static const struct file_operations dev_seq_fops = {
3830 .owner = THIS_MODULE,
3831 .open = dev_seq_open,
3833 .llseek = seq_lseek,
3834 .release = seq_release_net,
3837 static const struct seq_operations softnet_seq_ops = {
3838 .start = softnet_seq_start,
3839 .next = softnet_seq_next,
3840 .stop = softnet_seq_stop,
3841 .show = softnet_seq_show,
3844 static int softnet_seq_open(struct inode *inode, struct file *file)
3846 return seq_open(file, &softnet_seq_ops);
3849 static const struct file_operations softnet_seq_fops = {
3850 .owner = THIS_MODULE,
3851 .open = softnet_seq_open,
3853 .llseek = seq_lseek,
3854 .release = seq_release,
3857 static void *ptype_get_idx(loff_t pos)
3859 struct packet_type *pt = NULL;
3863 list_for_each_entry_rcu(pt, &ptype_all, list) {
3869 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3870 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3879 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3883 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3886 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3888 struct packet_type *pt;
3889 struct list_head *nxt;
3893 if (v == SEQ_START_TOKEN)
3894 return ptype_get_idx(0);
3897 nxt = pt->list.next;
3898 if (pt->type == htons(ETH_P_ALL)) {
3899 if (nxt != &ptype_all)
3902 nxt = ptype_base[0].next;
3904 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3906 while (nxt == &ptype_base[hash]) {
3907 if (++hash >= PTYPE_HASH_SIZE)
3909 nxt = ptype_base[hash].next;
3912 return list_entry(nxt, struct packet_type, list);
3915 static void ptype_seq_stop(struct seq_file *seq, void *v)
3921 static int ptype_seq_show(struct seq_file *seq, void *v)
3923 struct packet_type *pt = v;
3925 if (v == SEQ_START_TOKEN)
3926 seq_puts(seq, "Type Device Function\n");
3927 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3928 if (pt->type == htons(ETH_P_ALL))
3929 seq_puts(seq, "ALL ");
3931 seq_printf(seq, "%04x", ntohs(pt->type));
3933 seq_printf(seq, " %-8s %pF\n",
3934 pt->dev ? pt->dev->name : "", pt->func);
3940 static const struct seq_operations ptype_seq_ops = {
3941 .start = ptype_seq_start,
3942 .next = ptype_seq_next,
3943 .stop = ptype_seq_stop,
3944 .show = ptype_seq_show,
3947 static int ptype_seq_open(struct inode *inode, struct file *file)
3949 return seq_open_net(inode, file, &ptype_seq_ops,
3950 sizeof(struct seq_net_private));
3953 static const struct file_operations ptype_seq_fops = {
3954 .owner = THIS_MODULE,
3955 .open = ptype_seq_open,
3957 .llseek = seq_lseek,
3958 .release = seq_release_net,
3962 static int __net_init dev_proc_net_init(struct net *net)
3966 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3968 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3970 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3973 if (wext_proc_init(net))
3979 proc_net_remove(net, "ptype");
3981 proc_net_remove(net, "softnet_stat");
3983 proc_net_remove(net, "dev");
3987 static void __net_exit dev_proc_net_exit(struct net *net)
3989 wext_proc_exit(net);
3991 proc_net_remove(net, "ptype");
3992 proc_net_remove(net, "softnet_stat");
3993 proc_net_remove(net, "dev");
3996 static struct pernet_operations __net_initdata dev_proc_ops = {
3997 .init = dev_proc_net_init,
3998 .exit = dev_proc_net_exit,
4001 static int __init dev_proc_init(void)
4003 return register_pernet_subsys(&dev_proc_ops);
4006 #define dev_proc_init() 0
4007 #endif /* CONFIG_PROC_FS */
4011 * netdev_set_master - set up master/slave pair
4012 * @slave: slave device
4013 * @master: new master device
4015 * Changes the master device of the slave. Pass %NULL to break the
4016 * bonding. The caller must hold the RTNL semaphore. On a failure
4017 * a negative errno code is returned. On success the reference counts
4018 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
4019 * function returns zero.
4021 int netdev_set_master(struct net_device *slave, struct net_device *master)
4023 struct net_device *old = slave->master;
4033 slave->master = master;
4040 slave->flags |= IFF_SLAVE;
4042 slave->flags &= ~IFF_SLAVE;
4044 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
4047 EXPORT_SYMBOL(netdev_set_master);
4049 static void dev_change_rx_flags(struct net_device *dev, int flags)
4051 const struct net_device_ops *ops = dev->netdev_ops;
4053 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
4054 ops->ndo_change_rx_flags(dev, flags);
4057 static int __dev_set_promiscuity(struct net_device *dev, int inc)
4059 unsigned short old_flags = dev->flags;
4065 dev->flags |= IFF_PROMISC;
4066 dev->promiscuity += inc;
4067 if (dev->promiscuity == 0) {
4070 * If inc causes overflow, untouch promisc and return error.
4073 dev->flags &= ~IFF_PROMISC;
4075 dev->promiscuity -= inc;
4076 printk(KERN_WARNING "%s: promiscuity touches roof, "
4077 "set promiscuity failed, promiscuity feature "
4078 "of device might be broken.\n", dev->name);
4082 if (dev->flags != old_flags) {
4083 printk(KERN_INFO "device %s %s promiscuous mode\n",
4084 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
4086 if (audit_enabled) {
4087 current_uid_gid(&uid, &gid);
4088 audit_log(current->audit_context, GFP_ATOMIC,
4089 AUDIT_ANOM_PROMISCUOUS,
4090 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4091 dev->name, (dev->flags & IFF_PROMISC),
4092 (old_flags & IFF_PROMISC),
4093 audit_get_loginuid(current),
4095 audit_get_sessionid(current));
4098 dev_change_rx_flags(dev, IFF_PROMISC);
4104 * dev_set_promiscuity - update promiscuity count on a device
4108 * Add or remove promiscuity from a device. While the count in the device
4109 * remains above zero the interface remains promiscuous. Once it hits zero
4110 * the device reverts back to normal filtering operation. A negative inc
4111 * value is used to drop promiscuity on the device.
4112 * Return 0 if successful or a negative errno code on error.
4114 int dev_set_promiscuity(struct net_device *dev, int inc)
4116 unsigned short old_flags = dev->flags;
4119 err = __dev_set_promiscuity(dev, inc);
4122 if (dev->flags != old_flags)
4123 dev_set_rx_mode(dev);
4126 EXPORT_SYMBOL(dev_set_promiscuity);
4129 * dev_set_allmulti - update allmulti count on a device
4133 * Add or remove reception of all multicast frames to a device. While the
4134 * count in the device remains above zero the interface remains listening
4135 * to all interfaces. Once it hits zero the device reverts back to normal
4136 * filtering operation. A negative @inc value is used to drop the counter
4137 * when releasing a resource needing all multicasts.
4138 * Return 0 if successful or a negative errno code on error.
4141 int dev_set_allmulti(struct net_device *dev, int inc)
4143 unsigned short old_flags = dev->flags;
4147 dev->flags |= IFF_ALLMULTI;
4148 dev->allmulti += inc;
4149 if (dev->allmulti == 0) {
4152 * If inc causes overflow, untouch allmulti and return error.
4155 dev->flags &= ~IFF_ALLMULTI;
4157 dev->allmulti -= inc;
4158 printk(KERN_WARNING "%s: allmulti touches roof, "
4159 "set allmulti failed, allmulti feature of "
4160 "device might be broken.\n", dev->name);
4164 if (dev->flags ^ old_flags) {
4165 dev_change_rx_flags(dev, IFF_ALLMULTI);
4166 dev_set_rx_mode(dev);
4170 EXPORT_SYMBOL(dev_set_allmulti);
4173 * Upload unicast and multicast address lists to device and
4174 * configure RX filtering. When the device doesn't support unicast
4175 * filtering it is put in promiscuous mode while unicast addresses
4178 void __dev_set_rx_mode(struct net_device *dev)
4180 const struct net_device_ops *ops = dev->netdev_ops;
4182 /* dev_open will call this function so the list will stay sane. */
4183 if (!(dev->flags&IFF_UP))
4186 if (!netif_device_present(dev))
4189 if (ops->ndo_set_rx_mode)
4190 ops->ndo_set_rx_mode(dev);
4192 /* Unicast addresses changes may only happen under the rtnl,
4193 * therefore calling __dev_set_promiscuity here is safe.
4195 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
4196 __dev_set_promiscuity(dev, 1);
4197 dev->uc_promisc = 1;
4198 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
4199 __dev_set_promiscuity(dev, -1);
4200 dev->uc_promisc = 0;
4203 if (ops->ndo_set_multicast_list)
4204 ops->ndo_set_multicast_list(dev);
4208 void dev_set_rx_mode(struct net_device *dev)
4210 netif_addr_lock_bh(dev);
4211 __dev_set_rx_mode(dev);
4212 netif_addr_unlock_bh(dev);
4216 * dev_get_flags - get flags reported to userspace
4219 * Get the combination of flag bits exported through APIs to userspace.
4221 unsigned dev_get_flags(const struct net_device *dev)
4225 flags = (dev->flags & ~(IFF_PROMISC |
4230 (dev->gflags & (IFF_PROMISC |
4233 if (netif_running(dev)) {
4234 if (netif_oper_up(dev))
4235 flags |= IFF_RUNNING;
4236 if (netif_carrier_ok(dev))
4237 flags |= IFF_LOWER_UP;
4238 if (netif_dormant(dev))
4239 flags |= IFF_DORMANT;
4244 EXPORT_SYMBOL(dev_get_flags);
4246 int __dev_change_flags(struct net_device *dev, unsigned int flags)
4248 int old_flags = dev->flags;
4254 * Set the flags on our device.
4257 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4258 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4260 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4264 * Load in the correct multicast list now the flags have changed.
4267 if ((old_flags ^ flags) & IFF_MULTICAST)
4268 dev_change_rx_flags(dev, IFF_MULTICAST);
4270 dev_set_rx_mode(dev);
4273 * Have we downed the interface. We handle IFF_UP ourselves
4274 * according to user attempts to set it, rather than blindly
4279 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4280 ret = ((old_flags & IFF_UP) ? __dev_close : __dev_open)(dev);
4283 dev_set_rx_mode(dev);
4286 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4287 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4289 dev->gflags ^= IFF_PROMISC;
4290 dev_set_promiscuity(dev, inc);
4293 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4294 is important. Some (broken) drivers set IFF_PROMISC, when
4295 IFF_ALLMULTI is requested not asking us and not reporting.
4297 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4298 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4300 dev->gflags ^= IFF_ALLMULTI;
4301 dev_set_allmulti(dev, inc);
4307 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags)
4309 unsigned int changes = dev->flags ^ old_flags;
4311 if (changes & IFF_UP) {
4312 if (dev->flags & IFF_UP)
4313 call_netdevice_notifiers(NETDEV_UP, dev);
4315 call_netdevice_notifiers(NETDEV_DOWN, dev);
4318 if (dev->flags & IFF_UP &&
4319 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE)))
4320 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4324 * dev_change_flags - change device settings
4326 * @flags: device state flags
4328 * Change settings on device based state flags. The flags are
4329 * in the userspace exported format.
4331 int dev_change_flags(struct net_device *dev, unsigned flags)
4334 int old_flags = dev->flags;
4336 ret = __dev_change_flags(dev, flags);
4340 changes = old_flags ^ dev->flags;
4342 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4344 __dev_notify_flags(dev, old_flags);
4347 EXPORT_SYMBOL(dev_change_flags);
4350 * dev_set_mtu - Change maximum transfer unit
4352 * @new_mtu: new transfer unit
4354 * Change the maximum transfer size of the network device.
4356 int dev_set_mtu(struct net_device *dev, int new_mtu)
4358 const struct net_device_ops *ops = dev->netdev_ops;
4361 if (new_mtu == dev->mtu)
4364 /* MTU must be positive. */
4368 if (!netif_device_present(dev))
4372 if (ops->ndo_change_mtu)
4373 err = ops->ndo_change_mtu(dev, new_mtu);
4377 if (!err && dev->flags & IFF_UP)
4378 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4381 EXPORT_SYMBOL(dev_set_mtu);
4384 * dev_set_mac_address - Change Media Access Control Address
4388 * Change the hardware (MAC) address of the device
4390 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4392 const struct net_device_ops *ops = dev->netdev_ops;
4395 if (!ops->ndo_set_mac_address)
4397 if (sa->sa_family != dev->type)
4399 if (!netif_device_present(dev))
4401 err = ops->ndo_set_mac_address(dev, sa);
4403 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4406 EXPORT_SYMBOL(dev_set_mac_address);
4409 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4411 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4414 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4420 case SIOCGIFFLAGS: /* Get interface flags */
4421 ifr->ifr_flags = (short) dev_get_flags(dev);
4424 case SIOCGIFMETRIC: /* Get the metric on the interface
4425 (currently unused) */
4426 ifr->ifr_metric = 0;
4429 case SIOCGIFMTU: /* Get the MTU of a device */
4430 ifr->ifr_mtu = dev->mtu;
4435 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4437 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4438 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4439 ifr->ifr_hwaddr.sa_family = dev->type;
4447 ifr->ifr_map.mem_start = dev->mem_start;
4448 ifr->ifr_map.mem_end = dev->mem_end;
4449 ifr->ifr_map.base_addr = dev->base_addr;
4450 ifr->ifr_map.irq = dev->irq;
4451 ifr->ifr_map.dma = dev->dma;
4452 ifr->ifr_map.port = dev->if_port;
4456 ifr->ifr_ifindex = dev->ifindex;
4460 ifr->ifr_qlen = dev->tx_queue_len;
4464 /* dev_ioctl() should ensure this case
4476 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4478 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4481 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4482 const struct net_device_ops *ops;
4487 ops = dev->netdev_ops;
4490 case SIOCSIFFLAGS: /* Set interface flags */
4491 return dev_change_flags(dev, ifr->ifr_flags);
4493 case SIOCSIFMETRIC: /* Set the metric on the interface
4494 (currently unused) */
4497 case SIOCSIFMTU: /* Set the MTU of a device */
4498 return dev_set_mtu(dev, ifr->ifr_mtu);
4501 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4503 case SIOCSIFHWBROADCAST:
4504 if (ifr->ifr_hwaddr.sa_family != dev->type)
4506 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4507 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4508 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4512 if (ops->ndo_set_config) {
4513 if (!netif_device_present(dev))
4515 return ops->ndo_set_config(dev, &ifr->ifr_map);
4520 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4521 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4523 if (!netif_device_present(dev))
4525 return dev_mc_add_global(dev, ifr->ifr_hwaddr.sa_data);
4528 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4529 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4531 if (!netif_device_present(dev))
4533 return dev_mc_del_global(dev, ifr->ifr_hwaddr.sa_data);
4536 if (ifr->ifr_qlen < 0)
4538 dev->tx_queue_len = ifr->ifr_qlen;
4542 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4543 return dev_change_name(dev, ifr->ifr_newname);
4546 * Unknown or private ioctl
4549 if ((cmd >= SIOCDEVPRIVATE &&
4550 cmd <= SIOCDEVPRIVATE + 15) ||
4551 cmd == SIOCBONDENSLAVE ||
4552 cmd == SIOCBONDRELEASE ||
4553 cmd == SIOCBONDSETHWADDR ||
4554 cmd == SIOCBONDSLAVEINFOQUERY ||
4555 cmd == SIOCBONDINFOQUERY ||
4556 cmd == SIOCBONDCHANGEACTIVE ||
4557 cmd == SIOCGMIIPHY ||
4558 cmd == SIOCGMIIREG ||
4559 cmd == SIOCSMIIREG ||
4560 cmd == SIOCBRADDIF ||
4561 cmd == SIOCBRDELIF ||
4562 cmd == SIOCSHWTSTAMP ||
4563 cmd == SIOCWANDEV) {
4565 if (ops->ndo_do_ioctl) {
4566 if (netif_device_present(dev))
4567 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4579 * This function handles all "interface"-type I/O control requests. The actual
4580 * 'doing' part of this is dev_ifsioc above.
4584 * dev_ioctl - network device ioctl
4585 * @net: the applicable net namespace
4586 * @cmd: command to issue
4587 * @arg: pointer to a struct ifreq in user space
4589 * Issue ioctl functions to devices. This is normally called by the
4590 * user space syscall interfaces but can sometimes be useful for
4591 * other purposes. The return value is the return from the syscall if
4592 * positive or a negative errno code on error.
4595 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4601 /* One special case: SIOCGIFCONF takes ifconf argument
4602 and requires shared lock, because it sleeps writing
4606 if (cmd == SIOCGIFCONF) {
4608 ret = dev_ifconf(net, (char __user *) arg);
4612 if (cmd == SIOCGIFNAME)
4613 return dev_ifname(net, (struct ifreq __user *)arg);
4615 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4618 ifr.ifr_name[IFNAMSIZ-1] = 0;
4620 colon = strchr(ifr.ifr_name, ':');
4625 * See which interface the caller is talking about.
4630 * These ioctl calls:
4631 * - can be done by all.
4632 * - atomic and do not require locking.
4643 dev_load(net, ifr.ifr_name);
4645 ret = dev_ifsioc_locked(net, &ifr, cmd);
4650 if (copy_to_user(arg, &ifr,
4651 sizeof(struct ifreq)))
4657 dev_load(net, ifr.ifr_name);
4659 ret = dev_ethtool(net, &ifr);
4664 if (copy_to_user(arg, &ifr,
4665 sizeof(struct ifreq)))
4671 * These ioctl calls:
4672 * - require superuser power.
4673 * - require strict serialization.
4679 if (!capable(CAP_NET_ADMIN))
4681 dev_load(net, ifr.ifr_name);
4683 ret = dev_ifsioc(net, &ifr, cmd);
4688 if (copy_to_user(arg, &ifr,
4689 sizeof(struct ifreq)))
4695 * These ioctl calls:
4696 * - require superuser power.
4697 * - require strict serialization.
4698 * - do not return a value
4708 case SIOCSIFHWBROADCAST:
4711 case SIOCBONDENSLAVE:
4712 case SIOCBONDRELEASE:
4713 case SIOCBONDSETHWADDR:
4714 case SIOCBONDCHANGEACTIVE:
4718 if (!capable(CAP_NET_ADMIN))
4721 case SIOCBONDSLAVEINFOQUERY:
4722 case SIOCBONDINFOQUERY:
4723 dev_load(net, ifr.ifr_name);
4725 ret = dev_ifsioc(net, &ifr, cmd);
4730 /* Get the per device memory space. We can add this but
4731 * currently do not support it */
4733 /* Set the per device memory buffer space.
4734 * Not applicable in our case */
4739 * Unknown or private ioctl.
4742 if (cmd == SIOCWANDEV ||
4743 (cmd >= SIOCDEVPRIVATE &&
4744 cmd <= SIOCDEVPRIVATE + 15)) {
4745 dev_load(net, ifr.ifr_name);
4747 ret = dev_ifsioc(net, &ifr, cmd);
4749 if (!ret && copy_to_user(arg, &ifr,
4750 sizeof(struct ifreq)))
4754 /* Take care of Wireless Extensions */
4755 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4756 return wext_handle_ioctl(net, &ifr, cmd, arg);
4763 * dev_new_index - allocate an ifindex
4764 * @net: the applicable net namespace
4766 * Returns a suitable unique value for a new device interface
4767 * number. The caller must hold the rtnl semaphore or the
4768 * dev_base_lock to be sure it remains unique.
4770 static int dev_new_index(struct net *net)
4776 if (!__dev_get_by_index(net, ifindex))
4781 /* Delayed registration/unregisteration */
4782 static LIST_HEAD(net_todo_list);
4784 static void net_set_todo(struct net_device *dev)
4786 list_add_tail(&dev->todo_list, &net_todo_list);
4789 static void rollback_registered_many(struct list_head *head)
4791 struct net_device *dev, *tmp;
4793 BUG_ON(dev_boot_phase);
4796 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4797 /* Some devices call without registering
4798 * for initialization unwind. Remove those
4799 * devices and proceed with the remaining.
4801 if (dev->reg_state == NETREG_UNINITIALIZED) {
4802 pr_debug("unregister_netdevice: device %s/%p never "
4803 "was registered\n", dev->name, dev);
4806 list_del(&dev->unreg_list);
4810 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4812 /* If device is running, close it first. */
4815 /* And unlink it from device chain. */
4816 unlist_netdevice(dev);
4818 dev->reg_state = NETREG_UNREGISTERING;
4823 list_for_each_entry(dev, head, unreg_list) {
4824 /* Shutdown queueing discipline. */
4828 /* Notify protocols, that we are about to destroy
4829 this device. They should clean all the things.
4831 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4833 if (!dev->rtnl_link_ops ||
4834 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
4835 rtmsg_ifinfo(RTM_DELLINK, dev, ~0U);
4838 * Flush the unicast and multicast chains
4843 if (dev->netdev_ops->ndo_uninit)
4844 dev->netdev_ops->ndo_uninit(dev);
4846 /* Notifier chain MUST detach us from master device. */
4847 WARN_ON(dev->master);
4849 /* Remove entries from kobject tree */
4850 netdev_unregister_kobject(dev);
4853 /* Process any work delayed until the end of the batch */
4854 dev = list_first_entry(head, struct net_device, unreg_list);
4855 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4859 list_for_each_entry(dev, head, unreg_list)
4863 static void rollback_registered(struct net_device *dev)
4867 list_add(&dev->unreg_list, &single);
4868 rollback_registered_many(&single);
4871 static void __netdev_init_queue_locks_one(struct net_device *dev,
4872 struct netdev_queue *dev_queue,
4875 spin_lock_init(&dev_queue->_xmit_lock);
4876 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4877 dev_queue->xmit_lock_owner = -1;
4880 static void netdev_init_queue_locks(struct net_device *dev)
4882 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4883 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4886 unsigned long netdev_fix_features(unsigned long features, const char *name)
4888 /* Fix illegal SG+CSUM combinations. */
4889 if ((features & NETIF_F_SG) &&
4890 !(features & NETIF_F_ALL_CSUM)) {
4892 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4893 "checksum feature.\n", name);
4894 features &= ~NETIF_F_SG;
4897 /* TSO requires that SG is present as well. */
4898 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4900 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4901 "SG feature.\n", name);
4902 features &= ~NETIF_F_TSO;
4905 if (features & NETIF_F_UFO) {
4906 if (!(features & NETIF_F_GEN_CSUM)) {
4908 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4909 "since no NETIF_F_HW_CSUM feature.\n",
4911 features &= ~NETIF_F_UFO;
4914 if (!(features & NETIF_F_SG)) {
4916 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4917 "since no NETIF_F_SG feature.\n", name);
4918 features &= ~NETIF_F_UFO;
4924 EXPORT_SYMBOL(netdev_fix_features);
4927 * netif_stacked_transfer_operstate - transfer operstate
4928 * @rootdev: the root or lower level device to transfer state from
4929 * @dev: the device to transfer operstate to
4931 * Transfer operational state from root to device. This is normally
4932 * called when a stacking relationship exists between the root
4933 * device and the device(a leaf device).
4935 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4936 struct net_device *dev)
4938 if (rootdev->operstate == IF_OPER_DORMANT)
4939 netif_dormant_on(dev);
4941 netif_dormant_off(dev);
4943 if (netif_carrier_ok(rootdev)) {
4944 if (!netif_carrier_ok(dev))
4945 netif_carrier_on(dev);
4947 if (netif_carrier_ok(dev))
4948 netif_carrier_off(dev);
4951 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4954 * register_netdevice - register a network device
4955 * @dev: device to register
4957 * Take a completed network device structure and add it to the kernel
4958 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4959 * chain. 0 is returned on success. A negative errno code is returned
4960 * on a failure to set up the device, or if the name is a duplicate.
4962 * Callers must hold the rtnl semaphore. You may want
4963 * register_netdev() instead of this.
4966 * The locking appears insufficient to guarantee two parallel registers
4967 * will not get the same name.
4970 int register_netdevice(struct net_device *dev)
4973 struct net *net = dev_net(dev);
4975 BUG_ON(dev_boot_phase);
4980 /* When net_device's are persistent, this will be fatal. */
4981 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4984 spin_lock_init(&dev->addr_list_lock);
4985 netdev_set_addr_lockdep_class(dev);
4986 netdev_init_queue_locks(dev);
4991 if (!dev->num_rx_queues) {
4993 * Allocate a single RX queue if driver never called
4997 dev->_rx = kzalloc(sizeof(struct netdev_rx_queue), GFP_KERNEL);
5003 dev->_rx->first = dev->_rx;
5004 atomic_set(&dev->_rx->count, 1);
5005 dev->num_rx_queues = 1;
5008 /* Init, if this function is available */
5009 if (dev->netdev_ops->ndo_init) {
5010 ret = dev->netdev_ops->ndo_init(dev);
5018 ret = dev_get_valid_name(dev, dev->name, 0);
5022 dev->ifindex = dev_new_index(net);
5023 if (dev->iflink == -1)
5024 dev->iflink = dev->ifindex;
5026 /* Fix illegal checksum combinations */
5027 if ((dev->features & NETIF_F_HW_CSUM) &&
5028 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5029 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5031 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5034 if ((dev->features & NETIF_F_NO_CSUM) &&
5035 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5036 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5038 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5041 dev->features = netdev_fix_features(dev->features, dev->name);
5043 /* Enable software GSO if SG is supported. */
5044 if (dev->features & NETIF_F_SG)
5045 dev->features |= NETIF_F_GSO;
5047 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5048 ret = notifier_to_errno(ret);
5052 ret = netdev_register_kobject(dev);
5055 dev->reg_state = NETREG_REGISTERED;
5058 * Default initial state at registry is that the
5059 * device is present.
5062 set_bit(__LINK_STATE_PRESENT, &dev->state);
5064 dev_init_scheduler(dev);
5066 list_netdevice(dev);
5068 /* Notify protocols, that a new device appeared. */
5069 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5070 ret = notifier_to_errno(ret);
5072 rollback_registered(dev);
5073 dev->reg_state = NETREG_UNREGISTERED;
5076 * Prevent userspace races by waiting until the network
5077 * device is fully setup before sending notifications.
5079 if (!dev->rtnl_link_ops ||
5080 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
5081 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5087 if (dev->netdev_ops->ndo_uninit)
5088 dev->netdev_ops->ndo_uninit(dev);
5091 EXPORT_SYMBOL(register_netdevice);
5094 * init_dummy_netdev - init a dummy network device for NAPI
5095 * @dev: device to init
5097 * This takes a network device structure and initialize the minimum
5098 * amount of fields so it can be used to schedule NAPI polls without
5099 * registering a full blown interface. This is to be used by drivers
5100 * that need to tie several hardware interfaces to a single NAPI
5101 * poll scheduler due to HW limitations.
5103 int init_dummy_netdev(struct net_device *dev)
5105 /* Clear everything. Note we don't initialize spinlocks
5106 * are they aren't supposed to be taken by any of the
5107 * NAPI code and this dummy netdev is supposed to be
5108 * only ever used for NAPI polls
5110 memset(dev, 0, sizeof(struct net_device));
5112 /* make sure we BUG if trying to hit standard
5113 * register/unregister code path
5115 dev->reg_state = NETREG_DUMMY;
5117 /* initialize the ref count */
5118 atomic_set(&dev->refcnt, 1);
5120 /* NAPI wants this */
5121 INIT_LIST_HEAD(&dev->napi_list);
5123 /* a dummy interface is started by default */
5124 set_bit(__LINK_STATE_PRESENT, &dev->state);
5125 set_bit(__LINK_STATE_START, &dev->state);
5129 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5133 * register_netdev - register a network device
5134 * @dev: device to register
5136 * Take a completed network device structure and add it to the kernel
5137 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5138 * chain. 0 is returned on success. A negative errno code is returned
5139 * on a failure to set up the device, or if the name is a duplicate.
5141 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5142 * and expands the device name if you passed a format string to
5145 int register_netdev(struct net_device *dev)
5152 * If the name is a format string the caller wants us to do a
5155 if (strchr(dev->name, '%')) {
5156 err = dev_alloc_name(dev, dev->name);
5161 err = register_netdevice(dev);
5166 EXPORT_SYMBOL(register_netdev);
5169 * netdev_wait_allrefs - wait until all references are gone.
5171 * This is called when unregistering network devices.
5173 * Any protocol or device that holds a reference should register
5174 * for netdevice notification, and cleanup and put back the
5175 * reference if they receive an UNREGISTER event.
5176 * We can get stuck here if buggy protocols don't correctly
5179 static void netdev_wait_allrefs(struct net_device *dev)
5181 unsigned long rebroadcast_time, warning_time;
5183 linkwatch_forget_dev(dev);
5185 rebroadcast_time = warning_time = jiffies;
5186 while (atomic_read(&dev->refcnt) != 0) {
5187 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5190 /* Rebroadcast unregister notification */
5191 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5192 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5193 * should have already handle it the first time */
5195 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5197 /* We must not have linkwatch events
5198 * pending on unregister. If this
5199 * happens, we simply run the queue
5200 * unscheduled, resulting in a noop
5203 linkwatch_run_queue();
5208 rebroadcast_time = jiffies;
5213 if (time_after(jiffies, warning_time + 10 * HZ)) {
5214 printk(KERN_EMERG "unregister_netdevice: "
5215 "waiting for %s to become free. Usage "
5217 dev->name, atomic_read(&dev->refcnt));
5218 warning_time = jiffies;
5227 * register_netdevice(x1);
5228 * register_netdevice(x2);
5230 * unregister_netdevice(y1);
5231 * unregister_netdevice(y2);
5237 * We are invoked by rtnl_unlock().
5238 * This allows us to deal with problems:
5239 * 1) We can delete sysfs objects which invoke hotplug
5240 * without deadlocking with linkwatch via keventd.
5241 * 2) Since we run with the RTNL semaphore not held, we can sleep
5242 * safely in order to wait for the netdev refcnt to drop to zero.
5244 * We must not return until all unregister events added during
5245 * the interval the lock was held have been completed.
5247 void netdev_run_todo(void)
5249 struct list_head list;
5251 /* Snapshot list, allow later requests */
5252 list_replace_init(&net_todo_list, &list);
5256 while (!list_empty(&list)) {
5257 struct net_device *dev
5258 = list_first_entry(&list, struct net_device, todo_list);
5259 list_del(&dev->todo_list);
5261 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5262 printk(KERN_ERR "network todo '%s' but state %d\n",
5263 dev->name, dev->reg_state);
5268 dev->reg_state = NETREG_UNREGISTERED;
5270 on_each_cpu(flush_backlog, dev, 1);
5272 netdev_wait_allrefs(dev);
5275 BUG_ON(atomic_read(&dev->refcnt));
5276 WARN_ON(dev->ip_ptr);
5277 WARN_ON(dev->ip6_ptr);
5278 WARN_ON(dev->dn_ptr);
5280 if (dev->destructor)
5281 dev->destructor(dev);
5283 /* Free network device */
5284 kobject_put(&dev->dev.kobj);
5289 * dev_txq_stats_fold - fold tx_queues stats
5290 * @dev: device to get statistics from
5291 * @stats: struct net_device_stats to hold results
5293 void dev_txq_stats_fold(const struct net_device *dev,
5294 struct net_device_stats *stats)
5296 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5298 struct netdev_queue *txq;
5300 for (i = 0; i < dev->num_tx_queues; i++) {
5301 txq = netdev_get_tx_queue(dev, i);
5302 tx_bytes += txq->tx_bytes;
5303 tx_packets += txq->tx_packets;
5304 tx_dropped += txq->tx_dropped;
5306 if (tx_bytes || tx_packets || tx_dropped) {
5307 stats->tx_bytes = tx_bytes;
5308 stats->tx_packets = tx_packets;
5309 stats->tx_dropped = tx_dropped;
5312 EXPORT_SYMBOL(dev_txq_stats_fold);
5315 * dev_get_stats - get network device statistics
5316 * @dev: device to get statistics from
5318 * Get network statistics from device. The device driver may provide
5319 * its own method by setting dev->netdev_ops->get_stats; otherwise
5320 * the internal statistics structure is used.
5322 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5324 const struct net_device_ops *ops = dev->netdev_ops;
5326 if (ops->ndo_get_stats)
5327 return ops->ndo_get_stats(dev);
5329 dev_txq_stats_fold(dev, &dev->stats);
5332 EXPORT_SYMBOL(dev_get_stats);
5334 static void netdev_init_one_queue(struct net_device *dev,
5335 struct netdev_queue *queue,
5341 static void netdev_init_queues(struct net_device *dev)
5343 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5344 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5345 spin_lock_init(&dev->tx_global_lock);
5349 * alloc_netdev_mq - allocate network device
5350 * @sizeof_priv: size of private data to allocate space for
5351 * @name: device name format string
5352 * @setup: callback to initialize device
5353 * @queue_count: the number of subqueues to allocate
5355 * Allocates a struct net_device with private data area for driver use
5356 * and performs basic initialization. Also allocates subquue structs
5357 * for each queue on the device at the end of the netdevice.
5359 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5360 void (*setup)(struct net_device *), unsigned int queue_count)
5362 struct netdev_queue *tx;
5363 struct net_device *dev;
5365 struct net_device *p;
5367 struct netdev_rx_queue *rx;
5371 BUG_ON(strlen(name) >= sizeof(dev->name));
5373 alloc_size = sizeof(struct net_device);
5375 /* ensure 32-byte alignment of private area */
5376 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5377 alloc_size += sizeof_priv;
5379 /* ensure 32-byte alignment of whole construct */
5380 alloc_size += NETDEV_ALIGN - 1;
5382 p = kzalloc(alloc_size, GFP_KERNEL);
5384 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5388 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5390 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5396 rx = kcalloc(queue_count, sizeof(struct netdev_rx_queue), GFP_KERNEL);
5398 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5403 atomic_set(&rx->count, queue_count);
5406 * Set a pointer to first element in the array which holds the
5409 for (i = 0; i < queue_count; i++)
5413 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5414 dev->padded = (char *)dev - (char *)p;
5416 if (dev_addr_init(dev))
5422 dev_net_set(dev, &init_net);
5425 dev->num_tx_queues = queue_count;
5426 dev->real_num_tx_queues = queue_count;
5430 dev->num_rx_queues = queue_count;
5433 dev->gso_max_size = GSO_MAX_SIZE;
5435 netdev_init_queues(dev);
5437 INIT_LIST_HEAD(&dev->ethtool_ntuple_list.list);
5438 dev->ethtool_ntuple_list.count = 0;
5439 INIT_LIST_HEAD(&dev->napi_list);
5440 INIT_LIST_HEAD(&dev->unreg_list);
5441 INIT_LIST_HEAD(&dev->link_watch_list);
5442 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5444 strcpy(dev->name, name);
5457 EXPORT_SYMBOL(alloc_netdev_mq);
5460 * free_netdev - free network device
5463 * This function does the last stage of destroying an allocated device
5464 * interface. The reference to the device object is released.
5465 * If this is the last reference then it will be freed.
5467 void free_netdev(struct net_device *dev)
5469 struct napi_struct *p, *n;
5471 release_net(dev_net(dev));
5475 /* Flush device addresses */
5476 dev_addr_flush(dev);
5478 /* Clear ethtool n-tuple list */
5479 ethtool_ntuple_flush(dev);
5481 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5484 /* Compatibility with error handling in drivers */
5485 if (dev->reg_state == NETREG_UNINITIALIZED) {
5486 kfree((char *)dev - dev->padded);
5490 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5491 dev->reg_state = NETREG_RELEASED;
5493 /* will free via device release */
5494 put_device(&dev->dev);
5496 EXPORT_SYMBOL(free_netdev);
5499 * synchronize_net - Synchronize with packet receive processing
5501 * Wait for packets currently being received to be done.
5502 * Does not block later packets from starting.
5504 void synchronize_net(void)
5509 EXPORT_SYMBOL(synchronize_net);
5512 * unregister_netdevice_queue - remove device from the kernel
5516 * This function shuts down a device interface and removes it
5517 * from the kernel tables.
5518 * If head not NULL, device is queued to be unregistered later.
5520 * Callers must hold the rtnl semaphore. You may want
5521 * unregister_netdev() instead of this.
5524 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5529 list_move_tail(&dev->unreg_list, head);
5531 rollback_registered(dev);
5532 /* Finish processing unregister after unlock */
5536 EXPORT_SYMBOL(unregister_netdevice_queue);
5539 * unregister_netdevice_many - unregister many devices
5540 * @head: list of devices
5542 void unregister_netdevice_many(struct list_head *head)
5544 struct net_device *dev;
5546 if (!list_empty(head)) {
5547 rollback_registered_many(head);
5548 list_for_each_entry(dev, head, unreg_list)
5552 EXPORT_SYMBOL(unregister_netdevice_many);
5555 * unregister_netdev - remove device from the kernel
5558 * This function shuts down a device interface and removes it
5559 * from the kernel tables.
5561 * This is just a wrapper for unregister_netdevice that takes
5562 * the rtnl semaphore. In general you want to use this and not
5563 * unregister_netdevice.
5565 void unregister_netdev(struct net_device *dev)
5568 unregister_netdevice(dev);
5571 EXPORT_SYMBOL(unregister_netdev);
5574 * dev_change_net_namespace - move device to different nethost namespace
5576 * @net: network namespace
5577 * @pat: If not NULL name pattern to try if the current device name
5578 * is already taken in the destination network namespace.
5580 * This function shuts down a device interface and moves it
5581 * to a new network namespace. On success 0 is returned, on
5582 * a failure a netagive errno code is returned.
5584 * Callers must hold the rtnl semaphore.
5587 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5593 /* Don't allow namespace local devices to be moved. */
5595 if (dev->features & NETIF_F_NETNS_LOCAL)
5598 /* Ensure the device has been registrered */
5600 if (dev->reg_state != NETREG_REGISTERED)
5603 /* Get out if there is nothing todo */
5605 if (net_eq(dev_net(dev), net))
5608 /* Pick the destination device name, and ensure
5609 * we can use it in the destination network namespace.
5612 if (__dev_get_by_name(net, dev->name)) {
5613 /* We get here if we can't use the current device name */
5616 if (dev_get_valid_name(dev, pat, 1))
5621 * And now a mini version of register_netdevice unregister_netdevice.
5624 /* If device is running close it first. */
5627 /* And unlink it from device chain */
5629 unlist_netdevice(dev);
5633 /* Shutdown queueing discipline. */
5636 /* Notify protocols, that we are about to destroy
5637 this device. They should clean all the things.
5639 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5640 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5643 * Flush the unicast and multicast chains
5648 /* Actually switch the network namespace */
5649 dev_net_set(dev, net);
5651 /* If there is an ifindex conflict assign a new one */
5652 if (__dev_get_by_index(net, dev->ifindex)) {
5653 int iflink = (dev->iflink == dev->ifindex);
5654 dev->ifindex = dev_new_index(net);
5656 dev->iflink = dev->ifindex;
5659 /* Fixup kobjects */
5660 err = device_rename(&dev->dev, dev->name);
5663 /* Add the device back in the hashes */
5664 list_netdevice(dev);
5666 /* Notify protocols, that a new device appeared. */
5667 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5670 * Prevent userspace races by waiting until the network
5671 * device is fully setup before sending notifications.
5673 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5680 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5682 static int dev_cpu_callback(struct notifier_block *nfb,
5683 unsigned long action,
5686 struct sk_buff **list_skb;
5687 struct sk_buff *skb;
5688 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5689 struct softnet_data *sd, *oldsd;
5691 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5694 local_irq_disable();
5695 cpu = smp_processor_id();
5696 sd = &per_cpu(softnet_data, cpu);
5697 oldsd = &per_cpu(softnet_data, oldcpu);
5699 /* Find end of our completion_queue. */
5700 list_skb = &sd->completion_queue;
5702 list_skb = &(*list_skb)->next;
5703 /* Append completion queue from offline CPU. */
5704 *list_skb = oldsd->completion_queue;
5705 oldsd->completion_queue = NULL;
5707 /* Append output queue from offline CPU. */
5708 if (oldsd->output_queue) {
5709 *sd->output_queue_tailp = oldsd->output_queue;
5710 sd->output_queue_tailp = oldsd->output_queue_tailp;
5711 oldsd->output_queue = NULL;
5712 oldsd->output_queue_tailp = &oldsd->output_queue;
5715 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5718 /* Process offline CPU's input_pkt_queue */
5719 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
5721 input_queue_head_incr(oldsd);
5723 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
5725 input_queue_head_incr(oldsd);
5733 * netdev_increment_features - increment feature set by one
5734 * @all: current feature set
5735 * @one: new feature set
5736 * @mask: mask feature set
5738 * Computes a new feature set after adding a device with feature set
5739 * @one to the master device with current feature set @all. Will not
5740 * enable anything that is off in @mask. Returns the new feature set.
5742 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5745 /* If device needs checksumming, downgrade to it. */
5746 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5747 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5748 else if (mask & NETIF_F_ALL_CSUM) {
5749 /* If one device supports v4/v6 checksumming, set for all. */
5750 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5751 !(all & NETIF_F_GEN_CSUM)) {
5752 all &= ~NETIF_F_ALL_CSUM;
5753 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5756 /* If one device supports hw checksumming, set for all. */
5757 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5758 all &= ~NETIF_F_ALL_CSUM;
5759 all |= NETIF_F_HW_CSUM;
5763 one |= NETIF_F_ALL_CSUM;
5765 one |= all & NETIF_F_ONE_FOR_ALL;
5766 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5767 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5771 EXPORT_SYMBOL(netdev_increment_features);
5773 static struct hlist_head *netdev_create_hash(void)
5776 struct hlist_head *hash;
5778 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5780 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5781 INIT_HLIST_HEAD(&hash[i]);
5786 /* Initialize per network namespace state */
5787 static int __net_init netdev_init(struct net *net)
5789 INIT_LIST_HEAD(&net->dev_base_head);
5791 net->dev_name_head = netdev_create_hash();
5792 if (net->dev_name_head == NULL)
5795 net->dev_index_head = netdev_create_hash();
5796 if (net->dev_index_head == NULL)
5802 kfree(net->dev_name_head);
5808 * netdev_drivername - network driver for the device
5809 * @dev: network device
5810 * @buffer: buffer for resulting name
5811 * @len: size of buffer
5813 * Determine network driver for device.
5815 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5817 const struct device_driver *driver;
5818 const struct device *parent;
5820 if (len <= 0 || !buffer)
5824 parent = dev->dev.parent;
5829 driver = parent->driver;
5830 if (driver && driver->name)
5831 strlcpy(buffer, driver->name, len);
5835 static void __net_exit netdev_exit(struct net *net)
5837 kfree(net->dev_name_head);
5838 kfree(net->dev_index_head);
5841 static struct pernet_operations __net_initdata netdev_net_ops = {
5842 .init = netdev_init,
5843 .exit = netdev_exit,
5846 static void __net_exit default_device_exit(struct net *net)
5848 struct net_device *dev, *aux;
5850 * Push all migratable network devices back to the
5851 * initial network namespace
5854 for_each_netdev_safe(net, dev, aux) {
5856 char fb_name[IFNAMSIZ];
5858 /* Ignore unmoveable devices (i.e. loopback) */
5859 if (dev->features & NETIF_F_NETNS_LOCAL)
5862 /* Leave virtual devices for the generic cleanup */
5863 if (dev->rtnl_link_ops)
5866 /* Push remaing network devices to init_net */
5867 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5868 err = dev_change_net_namespace(dev, &init_net, fb_name);
5870 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5871 __func__, dev->name, err);
5878 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5880 /* At exit all network devices most be removed from a network
5881 * namespace. Do this in the reverse order of registeration.
5882 * Do this across as many network namespaces as possible to
5883 * improve batching efficiency.
5885 struct net_device *dev;
5887 LIST_HEAD(dev_kill_list);
5890 list_for_each_entry(net, net_list, exit_list) {
5891 for_each_netdev_reverse(net, dev) {
5892 if (dev->rtnl_link_ops)
5893 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5895 unregister_netdevice_queue(dev, &dev_kill_list);
5898 unregister_netdevice_many(&dev_kill_list);
5902 static struct pernet_operations __net_initdata default_device_ops = {
5903 .exit = default_device_exit,
5904 .exit_batch = default_device_exit_batch,
5908 * Initialize the DEV module. At boot time this walks the device list and
5909 * unhooks any devices that fail to initialise (normally hardware not
5910 * present) and leaves us with a valid list of present and active devices.
5915 * This is called single threaded during boot, so no need
5916 * to take the rtnl semaphore.
5918 static int __init net_dev_init(void)
5920 int i, rc = -ENOMEM;
5922 BUG_ON(!dev_boot_phase);
5924 if (dev_proc_init())
5927 if (netdev_kobject_init())
5930 INIT_LIST_HEAD(&ptype_all);
5931 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5932 INIT_LIST_HEAD(&ptype_base[i]);
5934 if (register_pernet_subsys(&netdev_net_ops))
5938 * Initialise the packet receive queues.
5941 for_each_possible_cpu(i) {
5942 struct softnet_data *sd = &per_cpu(softnet_data, i);
5944 memset(sd, 0, sizeof(*sd));
5945 skb_queue_head_init(&sd->input_pkt_queue);
5946 skb_queue_head_init(&sd->process_queue);
5947 sd->completion_queue = NULL;
5948 INIT_LIST_HEAD(&sd->poll_list);
5949 sd->output_queue = NULL;
5950 sd->output_queue_tailp = &sd->output_queue;
5952 sd->csd.func = rps_trigger_softirq;
5958 sd->backlog.poll = process_backlog;
5959 sd->backlog.weight = weight_p;
5960 sd->backlog.gro_list = NULL;
5961 sd->backlog.gro_count = 0;
5966 /* The loopback device is special if any other network devices
5967 * is present in a network namespace the loopback device must
5968 * be present. Since we now dynamically allocate and free the
5969 * loopback device ensure this invariant is maintained by
5970 * keeping the loopback device as the first device on the
5971 * list of network devices. Ensuring the loopback devices
5972 * is the first device that appears and the last network device
5975 if (register_pernet_device(&loopback_net_ops))
5978 if (register_pernet_device(&default_device_ops))
5981 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5982 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5984 hotcpu_notifier(dev_cpu_callback, 0);
5992 subsys_initcall(net_dev_init);
5994 static int __init initialize_hashrnd(void)
5996 get_random_bytes(&hashrnd, sizeof(hashrnd));
6000 late_initcall_sync(initialize_hashrnd);