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/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <linux/stat.h>
103 #include <linux/if_bridge.h>
104 #include <linux/if_macvlan.h>
106 #include <net/pkt_sched.h>
107 #include <net/checksum.h>
108 #include <net/xfrm.h>
109 #include <linux/highmem.h>
110 #include <linux/init.h>
111 #include <linux/kmod.h>
112 #include <linux/module.h>
113 #include <linux/netpoll.h>
114 #include <linux/rcupdate.h>
115 #include <linux/delay.h>
116 #include <net/wext.h>
117 #include <net/iw_handler.h>
118 #include <asm/current.h>
119 #include <linux/audit.h>
120 #include <linux/dmaengine.h>
121 #include <linux/err.h>
122 #include <linux/ctype.h>
123 #include <linux/if_arp.h>
124 #include <linux/if_vlan.h>
125 #include <linux/ip.h>
127 #include <linux/ipv6.h>
128 #include <linux/in.h>
129 #include <linux/jhash.h>
130 #include <linux/random.h>
131 #include <trace/events/napi.h>
133 #include "net-sysfs.h"
135 /* Instead of increasing this, you should create a hash table. */
136 #define MAX_GRO_SKBS 8
138 /* This should be increased if a protocol with a bigger head is added. */
139 #define GRO_MAX_HEAD (MAX_HEADER + 128)
142 * The list of packet types we will receive (as opposed to discard)
143 * and the routines to invoke.
145 * Why 16. Because with 16 the only overlap we get on a hash of the
146 * low nibble of the protocol value is RARP/SNAP/X.25.
148 * NOTE: That is no longer true with the addition of VLAN tags. Not
149 * sure which should go first, but I bet it won't make much
150 * difference if we are running VLANs. The good news is that
151 * this protocol won't be in the list unless compiled in, so
152 * the average user (w/out VLANs) will not be adversely affected.
169 #define PTYPE_HASH_SIZE (16)
170 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
172 static DEFINE_SPINLOCK(ptype_lock);
173 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
174 static struct list_head ptype_all __read_mostly; /* Taps */
177 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
180 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
182 * Writers must hold the rtnl semaphore while they loop through the
183 * dev_base_head list, and hold dev_base_lock for writing when they do the
184 * actual updates. This allows pure readers to access the list even
185 * while a writer is preparing to update it.
187 * To put it another way, dev_base_lock is held for writing only to
188 * protect against pure readers; the rtnl semaphore provides the
189 * protection against other writers.
191 * See, for example usages, register_netdevice() and
192 * unregister_netdevice(), which must be called with the rtnl
195 DEFINE_RWLOCK(dev_base_lock);
196 EXPORT_SYMBOL(dev_base_lock);
198 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
200 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
201 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
204 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
206 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
209 /* Device list insertion */
210 static int list_netdevice(struct net_device *dev)
212 struct net *net = dev_net(dev);
216 write_lock_bh(&dev_base_lock);
217 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
218 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
219 hlist_add_head_rcu(&dev->index_hlist,
220 dev_index_hash(net, dev->ifindex));
221 write_unlock_bh(&dev_base_lock);
225 /* Device list removal
226 * caller must respect a RCU grace period before freeing/reusing dev
228 static void unlist_netdevice(struct net_device *dev)
232 /* Unlink dev from the device chain */
233 write_lock_bh(&dev_base_lock);
234 list_del_rcu(&dev->dev_list);
235 hlist_del_rcu(&dev->name_hlist);
236 hlist_del_rcu(&dev->index_hlist);
237 write_unlock_bh(&dev_base_lock);
244 static RAW_NOTIFIER_HEAD(netdev_chain);
247 * Device drivers call our routines to queue packets here. We empty the
248 * queue in the local softnet handler.
251 DEFINE_PER_CPU(struct softnet_data, softnet_data);
252 EXPORT_PER_CPU_SYMBOL(softnet_data);
254 #ifdef CONFIG_LOCKDEP
256 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
257 * according to dev->type
259 static const unsigned short netdev_lock_type[] =
260 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
261 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
262 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
263 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
264 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
265 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
266 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
267 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
268 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
269 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
270 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
271 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
272 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
273 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
274 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
275 ARPHRD_VOID, ARPHRD_NONE};
277 static const char *const netdev_lock_name[] =
278 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
279 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
280 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
281 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
282 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
283 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
284 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
285 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
286 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
287 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
288 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
289 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
290 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
291 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
292 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
293 "_xmit_VOID", "_xmit_NONE"};
295 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
296 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
298 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
302 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
303 if (netdev_lock_type[i] == dev_type)
305 /* the last key is used by default */
306 return ARRAY_SIZE(netdev_lock_type) - 1;
309 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
310 unsigned short dev_type)
314 i = netdev_lock_pos(dev_type);
315 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
316 netdev_lock_name[i]);
319 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
323 i = netdev_lock_pos(dev->type);
324 lockdep_set_class_and_name(&dev->addr_list_lock,
325 &netdev_addr_lock_key[i],
326 netdev_lock_name[i]);
329 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
330 unsigned short dev_type)
333 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
338 /*******************************************************************************
340 Protocol management and registration routines
342 *******************************************************************************/
345 * Add a protocol ID to the list. Now that the input handler is
346 * smarter we can dispense with all the messy stuff that used to be
349 * BEWARE!!! Protocol handlers, mangling input packets,
350 * MUST BE last in hash buckets and checking protocol handlers
351 * MUST start from promiscuous ptype_all chain in net_bh.
352 * It is true now, do not change it.
353 * Explanation follows: if protocol handler, mangling packet, will
354 * be the first on list, it is not able to sense, that packet
355 * is cloned and should be copied-on-write, so that it will
356 * change it and subsequent readers will get broken packet.
361 * dev_add_pack - add packet handler
362 * @pt: packet type declaration
364 * Add a protocol handler to the networking stack. The passed &packet_type
365 * is linked into kernel lists and may not be freed until it has been
366 * removed from the kernel lists.
368 * This call does not sleep therefore it can not
369 * guarantee all CPU's that are in middle of receiving packets
370 * will see the new packet type (until the next received packet).
373 void dev_add_pack(struct packet_type *pt)
377 spin_lock_bh(&ptype_lock);
378 if (pt->type == htons(ETH_P_ALL))
379 list_add_rcu(&pt->list, &ptype_all);
381 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
382 list_add_rcu(&pt->list, &ptype_base[hash]);
384 spin_unlock_bh(&ptype_lock);
386 EXPORT_SYMBOL(dev_add_pack);
389 * __dev_remove_pack - remove packet handler
390 * @pt: packet type declaration
392 * Remove a protocol handler that was previously added to the kernel
393 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
394 * from the kernel lists and can be freed or reused once this function
397 * The packet type might still be in use by receivers
398 * and must not be freed until after all the CPU's have gone
399 * through a quiescent state.
401 void __dev_remove_pack(struct packet_type *pt)
403 struct list_head *head;
404 struct packet_type *pt1;
406 spin_lock_bh(&ptype_lock);
408 if (pt->type == htons(ETH_P_ALL))
411 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
413 list_for_each_entry(pt1, head, list) {
415 list_del_rcu(&pt->list);
420 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
422 spin_unlock_bh(&ptype_lock);
424 EXPORT_SYMBOL(__dev_remove_pack);
427 * dev_remove_pack - remove packet handler
428 * @pt: packet type declaration
430 * Remove a protocol handler that was previously added to the kernel
431 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
432 * from the kernel lists and can be freed or reused once this function
435 * This call sleeps to guarantee that no CPU is looking at the packet
438 void dev_remove_pack(struct packet_type *pt)
440 __dev_remove_pack(pt);
444 EXPORT_SYMBOL(dev_remove_pack);
446 /******************************************************************************
448 Device Boot-time Settings Routines
450 *******************************************************************************/
452 /* Boot time configuration table */
453 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
456 * netdev_boot_setup_add - add new setup entry
457 * @name: name of the device
458 * @map: configured settings for the device
460 * Adds new setup entry to the dev_boot_setup list. The function
461 * returns 0 on error and 1 on success. This is a generic routine to
464 static int netdev_boot_setup_add(char *name, struct ifmap *map)
466 struct netdev_boot_setup *s;
470 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
471 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
472 memset(s[i].name, 0, sizeof(s[i].name));
473 strlcpy(s[i].name, name, IFNAMSIZ);
474 memcpy(&s[i].map, map, sizeof(s[i].map));
479 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
483 * netdev_boot_setup_check - check boot time settings
484 * @dev: the netdevice
486 * Check boot time settings for the device.
487 * The found settings are set for the device to be used
488 * later in the device probing.
489 * Returns 0 if no settings found, 1 if they are.
491 int netdev_boot_setup_check(struct net_device *dev)
493 struct netdev_boot_setup *s = dev_boot_setup;
496 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
497 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
498 !strcmp(dev->name, s[i].name)) {
499 dev->irq = s[i].map.irq;
500 dev->base_addr = s[i].map.base_addr;
501 dev->mem_start = s[i].map.mem_start;
502 dev->mem_end = s[i].map.mem_end;
508 EXPORT_SYMBOL(netdev_boot_setup_check);
512 * netdev_boot_base - get address from boot time settings
513 * @prefix: prefix for network device
514 * @unit: id for network device
516 * Check boot time settings for the base address of device.
517 * The found settings are set for the device to be used
518 * later in the device probing.
519 * Returns 0 if no settings found.
521 unsigned long netdev_boot_base(const char *prefix, int unit)
523 const struct netdev_boot_setup *s = dev_boot_setup;
527 sprintf(name, "%s%d", prefix, unit);
530 * If device already registered then return base of 1
531 * to indicate not to probe for this interface
533 if (__dev_get_by_name(&init_net, name))
536 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
537 if (!strcmp(name, s[i].name))
538 return s[i].map.base_addr;
543 * Saves at boot time configured settings for any netdevice.
545 int __init netdev_boot_setup(char *str)
550 str = get_options(str, ARRAY_SIZE(ints), ints);
555 memset(&map, 0, sizeof(map));
559 map.base_addr = ints[2];
561 map.mem_start = ints[3];
563 map.mem_end = ints[4];
565 /* Add new entry to the list */
566 return netdev_boot_setup_add(str, &map);
569 __setup("netdev=", netdev_boot_setup);
571 /*******************************************************************************
573 Device Interface Subroutines
575 *******************************************************************************/
578 * __dev_get_by_name - find a device by its name
579 * @net: the applicable net namespace
580 * @name: name to find
582 * Find an interface by name. Must be called under RTNL semaphore
583 * or @dev_base_lock. If the name is found a pointer to the device
584 * is returned. If the name is not found then %NULL is returned. The
585 * reference counters are not incremented so the caller must be
586 * careful with locks.
589 struct net_device *__dev_get_by_name(struct net *net, const char *name)
591 struct hlist_node *p;
592 struct net_device *dev;
593 struct hlist_head *head = dev_name_hash(net, name);
595 hlist_for_each_entry(dev, p, head, name_hlist)
596 if (!strncmp(dev->name, name, IFNAMSIZ))
601 EXPORT_SYMBOL(__dev_get_by_name);
604 * dev_get_by_name_rcu - find a device by its name
605 * @net: the applicable net namespace
606 * @name: name to find
608 * Find an interface by name.
609 * If the name is found a pointer to the device is returned.
610 * If the name is not found then %NULL is returned.
611 * The reference counters are not incremented so the caller must be
612 * careful with locks. The caller must hold RCU lock.
615 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
617 struct hlist_node *p;
618 struct net_device *dev;
619 struct hlist_head *head = dev_name_hash(net, name);
621 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
622 if (!strncmp(dev->name, name, IFNAMSIZ))
627 EXPORT_SYMBOL(dev_get_by_name_rcu);
630 * dev_get_by_name - find a device by its name
631 * @net: the applicable net namespace
632 * @name: name to find
634 * Find an interface by name. This can be called from any
635 * context and does its own locking. The returned handle has
636 * the usage count incremented and the caller must use dev_put() to
637 * release it when it is no longer needed. %NULL is returned if no
638 * matching device is found.
641 struct net_device *dev_get_by_name(struct net *net, const char *name)
643 struct net_device *dev;
646 dev = dev_get_by_name_rcu(net, name);
652 EXPORT_SYMBOL(dev_get_by_name);
655 * __dev_get_by_index - find a device by its ifindex
656 * @net: the applicable net namespace
657 * @ifindex: index of device
659 * Search for an interface by index. Returns %NULL if the device
660 * is not found or a pointer to the device. The device has not
661 * had its reference counter increased so the caller must be careful
662 * about locking. The caller must hold either the RTNL semaphore
666 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
668 struct hlist_node *p;
669 struct net_device *dev;
670 struct hlist_head *head = dev_index_hash(net, ifindex);
672 hlist_for_each_entry(dev, p, head, index_hlist)
673 if (dev->ifindex == ifindex)
678 EXPORT_SYMBOL(__dev_get_by_index);
681 * dev_get_by_index_rcu - find a device by its ifindex
682 * @net: the applicable net namespace
683 * @ifindex: index of device
685 * Search for an interface by index. Returns %NULL if the device
686 * is not found or a pointer to the device. The device has not
687 * had its reference counter increased so the caller must be careful
688 * about locking. The caller must hold RCU lock.
691 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
693 struct hlist_node *p;
694 struct net_device *dev;
695 struct hlist_head *head = dev_index_hash(net, ifindex);
697 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
698 if (dev->ifindex == ifindex)
703 EXPORT_SYMBOL(dev_get_by_index_rcu);
707 * dev_get_by_index - find a device by its ifindex
708 * @net: the applicable net namespace
709 * @ifindex: index of device
711 * Search for an interface by index. Returns NULL if the device
712 * is not found or a pointer to the device. The device returned has
713 * had a reference added and the pointer is safe until the user calls
714 * dev_put to indicate they have finished with it.
717 struct net_device *dev_get_by_index(struct net *net, int ifindex)
719 struct net_device *dev;
722 dev = dev_get_by_index_rcu(net, ifindex);
728 EXPORT_SYMBOL(dev_get_by_index);
731 * dev_getbyhwaddr - find a device by its hardware address
732 * @net: the applicable net namespace
733 * @type: media type of device
734 * @ha: hardware address
736 * Search for an interface by MAC address. Returns NULL if the device
737 * is not found or a pointer to the device. The caller must hold the
738 * rtnl semaphore. The returned device has not had its ref count increased
739 * and the caller must therefore be careful about locking
742 * If the API was consistent this would be __dev_get_by_hwaddr
745 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
747 struct net_device *dev;
751 for_each_netdev(net, dev)
752 if (dev->type == type &&
753 !memcmp(dev->dev_addr, ha, dev->addr_len))
758 EXPORT_SYMBOL(dev_getbyhwaddr);
760 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
762 struct net_device *dev;
765 for_each_netdev(net, dev)
766 if (dev->type == type)
771 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
773 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
775 struct net_device *dev;
778 dev = __dev_getfirstbyhwtype(net, type);
784 EXPORT_SYMBOL(dev_getfirstbyhwtype);
787 * dev_get_by_flags - find any device with given flags
788 * @net: the applicable net namespace
789 * @if_flags: IFF_* values
790 * @mask: bitmask of bits in if_flags to check
792 * Search for any interface with the given flags. Returns NULL if a device
793 * is not found or a pointer to the device. The device returned has
794 * had a reference added and the pointer is safe until the user calls
795 * dev_put to indicate they have finished with it.
798 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
801 struct net_device *dev, *ret;
805 for_each_netdev_rcu(net, dev) {
806 if (((dev->flags ^ if_flags) & mask) == 0) {
815 EXPORT_SYMBOL(dev_get_by_flags);
818 * dev_valid_name - check if name is okay for network device
821 * Network device names need to be valid file names to
822 * to allow sysfs to work. We also disallow any kind of
825 int dev_valid_name(const char *name)
829 if (strlen(name) >= IFNAMSIZ)
831 if (!strcmp(name, ".") || !strcmp(name, ".."))
835 if (*name == '/' || isspace(*name))
841 EXPORT_SYMBOL(dev_valid_name);
844 * __dev_alloc_name - allocate a name for a device
845 * @net: network namespace to allocate the device name in
846 * @name: name format string
847 * @buf: scratch buffer and result name string
849 * Passed a format string - eg "lt%d" it will try and find a suitable
850 * id. It scans list of devices to build up a free map, then chooses
851 * the first empty slot. The caller must hold the dev_base or rtnl lock
852 * while allocating the name and adding the device in order to avoid
854 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
855 * Returns the number of the unit assigned or a negative errno code.
858 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
862 const int max_netdevices = 8*PAGE_SIZE;
863 unsigned long *inuse;
864 struct net_device *d;
866 p = strnchr(name, IFNAMSIZ-1, '%');
869 * Verify the string as this thing may have come from
870 * the user. There must be either one "%d" and no other "%"
873 if (p[1] != 'd' || strchr(p + 2, '%'))
876 /* Use one page as a bit array of possible slots */
877 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
881 for_each_netdev(net, d) {
882 if (!sscanf(d->name, name, &i))
884 if (i < 0 || i >= max_netdevices)
887 /* avoid cases where sscanf is not exact inverse of printf */
888 snprintf(buf, IFNAMSIZ, name, i);
889 if (!strncmp(buf, d->name, IFNAMSIZ))
893 i = find_first_zero_bit(inuse, max_netdevices);
894 free_page((unsigned long) inuse);
898 snprintf(buf, IFNAMSIZ, name, i);
899 if (!__dev_get_by_name(net, buf))
902 /* It is possible to run out of possible slots
903 * when the name is long and there isn't enough space left
904 * for the digits, or if all bits are used.
910 * dev_alloc_name - allocate a name for a device
912 * @name: name format string
914 * Passed a format string - eg "lt%d" it will try and find a suitable
915 * id. It scans list of devices to build up a free map, then chooses
916 * the first empty slot. The caller must hold the dev_base or rtnl lock
917 * while allocating the name and adding the device in order to avoid
919 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
920 * Returns the number of the unit assigned or a negative errno code.
923 int dev_alloc_name(struct net_device *dev, const char *name)
929 BUG_ON(!dev_net(dev));
931 ret = __dev_alloc_name(net, name, buf);
933 strlcpy(dev->name, buf, IFNAMSIZ);
936 EXPORT_SYMBOL(dev_alloc_name);
938 static int dev_get_valid_name(struct net *net, const char *name, char *buf,
941 if (!dev_valid_name(name))
944 if (fmt && strchr(name, '%'))
945 return __dev_alloc_name(net, name, buf);
946 else if (__dev_get_by_name(net, name))
948 else if (buf != name)
949 strlcpy(buf, name, IFNAMSIZ);
955 * dev_change_name - change name of a device
957 * @newname: name (or format string) must be at least IFNAMSIZ
959 * Change name of a device, can pass format strings "eth%d".
962 int dev_change_name(struct net_device *dev, const char *newname)
964 char oldname[IFNAMSIZ];
970 BUG_ON(!dev_net(dev));
973 if (dev->flags & IFF_UP)
976 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
979 memcpy(oldname, dev->name, IFNAMSIZ);
981 err = dev_get_valid_name(net, newname, dev->name, 1);
986 /* For now only devices in the initial network namespace
989 if (net_eq(net, &init_net)) {
990 ret = device_rename(&dev->dev, dev->name);
992 memcpy(dev->name, oldname, IFNAMSIZ);
997 write_lock_bh(&dev_base_lock);
998 hlist_del(&dev->name_hlist);
999 write_unlock_bh(&dev_base_lock);
1003 write_lock_bh(&dev_base_lock);
1004 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1005 write_unlock_bh(&dev_base_lock);
1007 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1008 ret = notifier_to_errno(ret);
1011 /* err >= 0 after dev_alloc_name() or stores the first errno */
1014 memcpy(dev->name, oldname, IFNAMSIZ);
1018 "%s: name change rollback failed: %d.\n",
1027 * dev_set_alias - change ifalias of a device
1029 * @alias: name up to IFALIASZ
1030 * @len: limit of bytes to copy from info
1032 * Set ifalias for a device,
1034 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1038 if (len >= IFALIASZ)
1043 kfree(dev->ifalias);
1044 dev->ifalias = NULL;
1049 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1053 strlcpy(dev->ifalias, alias, len+1);
1059 * netdev_features_change - device changes features
1060 * @dev: device to cause notification
1062 * Called to indicate a device has changed features.
1064 void netdev_features_change(struct net_device *dev)
1066 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1068 EXPORT_SYMBOL(netdev_features_change);
1071 * netdev_state_change - device changes state
1072 * @dev: device to cause notification
1074 * Called to indicate a device has changed state. This function calls
1075 * the notifier chains for netdev_chain and sends a NEWLINK message
1076 * to the routing socket.
1078 void netdev_state_change(struct net_device *dev)
1080 if (dev->flags & IFF_UP) {
1081 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1082 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1085 EXPORT_SYMBOL(netdev_state_change);
1087 void netdev_bonding_change(struct net_device *dev, unsigned long event)
1089 call_netdevice_notifiers(event, dev);
1091 EXPORT_SYMBOL(netdev_bonding_change);
1094 * dev_load - load a network module
1095 * @net: the applicable net namespace
1096 * @name: name of interface
1098 * If a network interface is not present and the process has suitable
1099 * privileges this function loads the module. If module loading is not
1100 * available in this kernel then it becomes a nop.
1103 void dev_load(struct net *net, const char *name)
1105 struct net_device *dev;
1108 dev = dev_get_by_name_rcu(net, name);
1111 if (!dev && capable(CAP_NET_ADMIN))
1112 request_module("%s", name);
1114 EXPORT_SYMBOL(dev_load);
1117 * dev_open - prepare an interface for use.
1118 * @dev: device to open
1120 * Takes a device from down to up state. The device's private open
1121 * function is invoked and then the multicast lists are loaded. Finally
1122 * the device is moved into the up state and a %NETDEV_UP message is
1123 * sent to the netdev notifier chain.
1125 * Calling this function on an active interface is a nop. On a failure
1126 * a negative errno code is returned.
1128 int dev_open(struct net_device *dev)
1130 const struct net_device_ops *ops = dev->netdev_ops;
1139 if (dev->flags & IFF_UP)
1143 * Is it even present?
1145 if (!netif_device_present(dev))
1148 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1149 ret = notifier_to_errno(ret);
1154 * Call device private open method
1156 set_bit(__LINK_STATE_START, &dev->state);
1158 if (ops->ndo_validate_addr)
1159 ret = ops->ndo_validate_addr(dev);
1161 if (!ret && ops->ndo_open)
1162 ret = ops->ndo_open(dev);
1165 * If it went open OK then:
1169 clear_bit(__LINK_STATE_START, &dev->state);
1174 dev->flags |= IFF_UP;
1179 net_dmaengine_get();
1182 * Initialize multicasting status
1184 dev_set_rx_mode(dev);
1187 * Wakeup transmit queue engine
1192 * ... and announce new interface.
1194 call_netdevice_notifiers(NETDEV_UP, dev);
1199 EXPORT_SYMBOL(dev_open);
1202 * dev_close - shutdown an interface.
1203 * @dev: device to shutdown
1205 * This function moves an active device into down state. A
1206 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1207 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1210 int dev_close(struct net_device *dev)
1212 const struct net_device_ops *ops = dev->netdev_ops;
1217 if (!(dev->flags & IFF_UP))
1221 * Tell people we are going down, so that they can
1222 * prepare to death, when device is still operating.
1224 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1226 clear_bit(__LINK_STATE_START, &dev->state);
1228 /* Synchronize to scheduled poll. We cannot touch poll list,
1229 * it can be even on different cpu. So just clear netif_running().
1231 * dev->stop() will invoke napi_disable() on all of it's
1232 * napi_struct instances on this device.
1234 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1236 dev_deactivate(dev);
1239 * Call the device specific close. This cannot fail.
1240 * Only if device is UP
1242 * We allow it to be called even after a DETACH hot-plug
1249 * Device is now down.
1252 dev->flags &= ~IFF_UP;
1255 * Tell people we are down
1257 call_netdevice_notifiers(NETDEV_DOWN, dev);
1262 net_dmaengine_put();
1266 EXPORT_SYMBOL(dev_close);
1270 * dev_disable_lro - disable Large Receive Offload on a device
1273 * Disable Large Receive Offload (LRO) on a net device. Must be
1274 * called under RTNL. This is needed if received packets may be
1275 * forwarded to another interface.
1277 void dev_disable_lro(struct net_device *dev)
1279 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1280 dev->ethtool_ops->set_flags) {
1281 u32 flags = dev->ethtool_ops->get_flags(dev);
1282 if (flags & ETH_FLAG_LRO) {
1283 flags &= ~ETH_FLAG_LRO;
1284 dev->ethtool_ops->set_flags(dev, flags);
1287 WARN_ON(dev->features & NETIF_F_LRO);
1289 EXPORT_SYMBOL(dev_disable_lro);
1292 static int dev_boot_phase = 1;
1295 * Device change register/unregister. These are not inline or static
1296 * as we export them to the world.
1300 * register_netdevice_notifier - register a network notifier block
1303 * Register a notifier to be called when network device events occur.
1304 * The notifier passed is linked into the kernel structures and must
1305 * not be reused until it has been unregistered. A negative errno code
1306 * is returned on a failure.
1308 * When registered all registration and up events are replayed
1309 * to the new notifier to allow device to have a race free
1310 * view of the network device list.
1313 int register_netdevice_notifier(struct notifier_block *nb)
1315 struct net_device *dev;
1316 struct net_device *last;
1321 err = raw_notifier_chain_register(&netdev_chain, nb);
1327 for_each_netdev(net, dev) {
1328 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1329 err = notifier_to_errno(err);
1333 if (!(dev->flags & IFF_UP))
1336 nb->notifier_call(nb, NETDEV_UP, dev);
1347 for_each_netdev(net, dev) {
1351 if (dev->flags & IFF_UP) {
1352 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1353 nb->notifier_call(nb, NETDEV_DOWN, dev);
1355 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1356 nb->notifier_call(nb, NETDEV_UNREGISTER_BATCH, dev);
1360 raw_notifier_chain_unregister(&netdev_chain, nb);
1363 EXPORT_SYMBOL(register_netdevice_notifier);
1366 * unregister_netdevice_notifier - unregister a network notifier block
1369 * Unregister a notifier previously registered by
1370 * register_netdevice_notifier(). The notifier is unlinked into the
1371 * kernel structures and may then be reused. A negative errno code
1372 * is returned on a failure.
1375 int unregister_netdevice_notifier(struct notifier_block *nb)
1380 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1384 EXPORT_SYMBOL(unregister_netdevice_notifier);
1387 * call_netdevice_notifiers - call all network notifier blocks
1388 * @val: value passed unmodified to notifier function
1389 * @dev: net_device pointer passed unmodified to notifier function
1391 * Call all network notifier blocks. Parameters and return value
1392 * are as for raw_notifier_call_chain().
1395 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1397 return raw_notifier_call_chain(&netdev_chain, val, dev);
1400 /* When > 0 there are consumers of rx skb time stamps */
1401 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1403 void net_enable_timestamp(void)
1405 atomic_inc(&netstamp_needed);
1407 EXPORT_SYMBOL(net_enable_timestamp);
1409 void net_disable_timestamp(void)
1411 atomic_dec(&netstamp_needed);
1413 EXPORT_SYMBOL(net_disable_timestamp);
1415 static inline void net_timestamp(struct sk_buff *skb)
1417 if (atomic_read(&netstamp_needed))
1418 __net_timestamp(skb);
1420 skb->tstamp.tv64 = 0;
1424 * dev_forward_skb - loopback an skb to another netif
1426 * @dev: destination network device
1427 * @skb: buffer to forward
1430 * NET_RX_SUCCESS (no congestion)
1431 * NET_RX_DROP (packet was dropped)
1433 * dev_forward_skb can be used for injecting an skb from the
1434 * start_xmit function of one device into the receive queue
1435 * of another device.
1437 * The receiving device may be in another namespace, so
1438 * we have to clear all information in the skb that could
1439 * impact namespace isolation.
1441 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1445 if (!(dev->flags & IFF_UP))
1448 if (skb->len > (dev->mtu + dev->hard_header_len))
1452 skb->tstamp.tv64 = 0;
1453 skb->pkt_type = PACKET_HOST;
1454 skb->protocol = eth_type_trans(skb, dev);
1458 return netif_rx(skb);
1460 EXPORT_SYMBOL_GPL(dev_forward_skb);
1463 * Support routine. Sends outgoing frames to any network
1464 * taps currently in use.
1467 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1469 struct packet_type *ptype;
1471 #ifdef CONFIG_NET_CLS_ACT
1472 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1479 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1480 /* Never send packets back to the socket
1481 * they originated from - MvS (miquels@drinkel.ow.org)
1483 if ((ptype->dev == dev || !ptype->dev) &&
1484 (ptype->af_packet_priv == NULL ||
1485 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1486 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1490 /* skb->nh should be correctly
1491 set by sender, so that the second statement is
1492 just protection against buggy protocols.
1494 skb_reset_mac_header(skb2);
1496 if (skb_network_header(skb2) < skb2->data ||
1497 skb2->network_header > skb2->tail) {
1498 if (net_ratelimit())
1499 printk(KERN_CRIT "protocol %04x is "
1501 skb2->protocol, dev->name);
1502 skb_reset_network_header(skb2);
1505 skb2->transport_header = skb2->network_header;
1506 skb2->pkt_type = PACKET_OUTGOING;
1507 ptype->func(skb2, skb->dev, ptype, skb->dev);
1514 static inline void __netif_reschedule(struct Qdisc *q)
1516 struct softnet_data *sd;
1517 unsigned long flags;
1519 local_irq_save(flags);
1520 sd = &__get_cpu_var(softnet_data);
1521 q->next_sched = sd->output_queue;
1522 sd->output_queue = q;
1523 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1524 local_irq_restore(flags);
1527 void __netif_schedule(struct Qdisc *q)
1529 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1530 __netif_reschedule(q);
1532 EXPORT_SYMBOL(__netif_schedule);
1534 void dev_kfree_skb_irq(struct sk_buff *skb)
1536 if (atomic_dec_and_test(&skb->users)) {
1537 struct softnet_data *sd;
1538 unsigned long flags;
1540 local_irq_save(flags);
1541 sd = &__get_cpu_var(softnet_data);
1542 skb->next = sd->completion_queue;
1543 sd->completion_queue = skb;
1544 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1545 local_irq_restore(flags);
1548 EXPORT_SYMBOL(dev_kfree_skb_irq);
1550 void dev_kfree_skb_any(struct sk_buff *skb)
1552 if (in_irq() || irqs_disabled())
1553 dev_kfree_skb_irq(skb);
1557 EXPORT_SYMBOL(dev_kfree_skb_any);
1561 * netif_device_detach - mark device as removed
1562 * @dev: network device
1564 * Mark device as removed from system and therefore no longer available.
1566 void netif_device_detach(struct net_device *dev)
1568 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1569 netif_running(dev)) {
1570 netif_tx_stop_all_queues(dev);
1573 EXPORT_SYMBOL(netif_device_detach);
1576 * netif_device_attach - mark device as attached
1577 * @dev: network device
1579 * Mark device as attached from system and restart if needed.
1581 void netif_device_attach(struct net_device *dev)
1583 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1584 netif_running(dev)) {
1585 netif_tx_wake_all_queues(dev);
1586 __netdev_watchdog_up(dev);
1589 EXPORT_SYMBOL(netif_device_attach);
1591 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1593 return ((features & NETIF_F_GEN_CSUM) ||
1594 ((features & NETIF_F_IP_CSUM) &&
1595 protocol == htons(ETH_P_IP)) ||
1596 ((features & NETIF_F_IPV6_CSUM) &&
1597 protocol == htons(ETH_P_IPV6)) ||
1598 ((features & NETIF_F_FCOE_CRC) &&
1599 protocol == htons(ETH_P_FCOE)));
1602 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1604 if (can_checksum_protocol(dev->features, skb->protocol))
1607 if (skb->protocol == htons(ETH_P_8021Q)) {
1608 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1609 if (can_checksum_protocol(dev->features & dev->vlan_features,
1610 veh->h_vlan_encapsulated_proto))
1618 * Invalidate hardware checksum when packet is to be mangled, and
1619 * complete checksum manually on outgoing path.
1621 int skb_checksum_help(struct sk_buff *skb)
1624 int ret = 0, offset;
1626 if (skb->ip_summed == CHECKSUM_COMPLETE)
1627 goto out_set_summed;
1629 if (unlikely(skb_shinfo(skb)->gso_size)) {
1630 /* Let GSO fix up the checksum. */
1631 goto out_set_summed;
1634 offset = skb->csum_start - skb_headroom(skb);
1635 BUG_ON(offset >= skb_headlen(skb));
1636 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1638 offset += skb->csum_offset;
1639 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1641 if (skb_cloned(skb) &&
1642 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1643 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1648 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1650 skb->ip_summed = CHECKSUM_NONE;
1654 EXPORT_SYMBOL(skb_checksum_help);
1657 * skb_gso_segment - Perform segmentation on skb.
1658 * @skb: buffer to segment
1659 * @features: features for the output path (see dev->features)
1661 * This function segments the given skb and returns a list of segments.
1663 * It may return NULL if the skb requires no segmentation. This is
1664 * only possible when GSO is used for verifying header integrity.
1666 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1668 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1669 struct packet_type *ptype;
1670 __be16 type = skb->protocol;
1673 skb_reset_mac_header(skb);
1674 skb->mac_len = skb->network_header - skb->mac_header;
1675 __skb_pull(skb, skb->mac_len);
1677 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1678 struct net_device *dev = skb->dev;
1679 struct ethtool_drvinfo info = {};
1681 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1682 dev->ethtool_ops->get_drvinfo(dev, &info);
1684 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1686 info.driver, dev ? dev->features : 0L,
1687 skb->sk ? skb->sk->sk_route_caps : 0L,
1688 skb->len, skb->data_len, skb->ip_summed);
1690 if (skb_header_cloned(skb) &&
1691 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1692 return ERR_PTR(err);
1696 list_for_each_entry_rcu(ptype,
1697 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1698 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1699 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1700 err = ptype->gso_send_check(skb);
1701 segs = ERR_PTR(err);
1702 if (err || skb_gso_ok(skb, features))
1704 __skb_push(skb, (skb->data -
1705 skb_network_header(skb)));
1707 segs = ptype->gso_segment(skb, features);
1713 __skb_push(skb, skb->data - skb_mac_header(skb));
1717 EXPORT_SYMBOL(skb_gso_segment);
1719 /* Take action when hardware reception checksum errors are detected. */
1721 void netdev_rx_csum_fault(struct net_device *dev)
1723 if (net_ratelimit()) {
1724 printk(KERN_ERR "%s: hw csum failure.\n",
1725 dev ? dev->name : "<unknown>");
1729 EXPORT_SYMBOL(netdev_rx_csum_fault);
1732 /* Actually, we should eliminate this check as soon as we know, that:
1733 * 1. IOMMU is present and allows to map all the memory.
1734 * 2. No high memory really exists on this machine.
1737 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1739 #ifdef CONFIG_HIGHMEM
1742 if (dev->features & NETIF_F_HIGHDMA)
1745 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1746 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1754 void (*destructor)(struct sk_buff *skb);
1757 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1759 static void dev_gso_skb_destructor(struct sk_buff *skb)
1761 struct dev_gso_cb *cb;
1764 struct sk_buff *nskb = skb->next;
1766 skb->next = nskb->next;
1769 } while (skb->next);
1771 cb = DEV_GSO_CB(skb);
1773 cb->destructor(skb);
1777 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1778 * @skb: buffer to segment
1780 * This function segments the given skb and stores the list of segments
1783 static int dev_gso_segment(struct sk_buff *skb)
1785 struct net_device *dev = skb->dev;
1786 struct sk_buff *segs;
1787 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1790 segs = skb_gso_segment(skb, features);
1792 /* Verifying header integrity only. */
1797 return PTR_ERR(segs);
1800 DEV_GSO_CB(skb)->destructor = skb->destructor;
1801 skb->destructor = dev_gso_skb_destructor;
1806 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1807 struct netdev_queue *txq)
1809 const struct net_device_ops *ops = dev->netdev_ops;
1810 int rc = NETDEV_TX_OK;
1812 if (likely(!skb->next)) {
1813 if (!list_empty(&ptype_all))
1814 dev_queue_xmit_nit(skb, dev);
1816 if (netif_needs_gso(dev, skb)) {
1817 if (unlikely(dev_gso_segment(skb)))
1824 * If device doesnt need skb->dst, release it right now while
1825 * its hot in this cpu cache
1827 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1830 rc = ops->ndo_start_xmit(skb, dev);
1831 if (rc == NETDEV_TX_OK)
1832 txq_trans_update(txq);
1834 * TODO: if skb_orphan() was called by
1835 * dev->hard_start_xmit() (for example, the unmodified
1836 * igb driver does that; bnx2 doesn't), then
1837 * skb_tx_software_timestamp() will be unable to send
1838 * back the time stamp.
1840 * How can this be prevented? Always create another
1841 * reference to the socket before calling
1842 * dev->hard_start_xmit()? Prevent that skb_orphan()
1843 * does anything in dev->hard_start_xmit() by clearing
1844 * the skb destructor before the call and restoring it
1845 * afterwards, then doing the skb_orphan() ourselves?
1852 struct sk_buff *nskb = skb->next;
1854 skb->next = nskb->next;
1858 * If device doesnt need nskb->dst, release it right now while
1859 * its hot in this cpu cache
1861 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1864 rc = ops->ndo_start_xmit(nskb, dev);
1865 if (unlikely(rc != NETDEV_TX_OK)) {
1866 if (rc & ~NETDEV_TX_MASK)
1867 goto out_kfree_gso_skb;
1868 nskb->next = skb->next;
1872 txq_trans_update(txq);
1873 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1874 return NETDEV_TX_BUSY;
1875 } while (skb->next);
1878 if (likely(skb->next == NULL))
1879 skb->destructor = DEV_GSO_CB(skb)->destructor;
1885 static u32 skb_tx_hashrnd;
1887 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1891 if (skb_rx_queue_recorded(skb)) {
1892 hash = skb_get_rx_queue(skb);
1893 while (unlikely(hash >= dev->real_num_tx_queues))
1894 hash -= dev->real_num_tx_queues;
1898 if (skb->sk && skb->sk->sk_hash)
1899 hash = skb->sk->sk_hash;
1901 hash = skb->protocol;
1903 hash = jhash_1word(hash, skb_tx_hashrnd);
1905 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1907 EXPORT_SYMBOL(skb_tx_hash);
1909 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1911 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1912 if (net_ratelimit()) {
1913 WARN(1, "%s selects TX queue %d, but "
1914 "real number of TX queues is %d\n",
1915 dev->name, queue_index,
1916 dev->real_num_tx_queues);
1923 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1924 struct sk_buff *skb)
1927 struct sock *sk = skb->sk;
1929 if (sk_tx_queue_recorded(sk)) {
1930 queue_index = sk_tx_queue_get(sk);
1932 const struct net_device_ops *ops = dev->netdev_ops;
1934 if (ops->ndo_select_queue) {
1935 queue_index = ops->ndo_select_queue(dev, skb);
1936 queue_index = dev_cap_txqueue(dev, queue_index);
1939 if (dev->real_num_tx_queues > 1)
1940 queue_index = skb_tx_hash(dev, skb);
1942 if (sk && sk->sk_dst_cache)
1943 sk_tx_queue_set(sk, queue_index);
1947 skb_set_queue_mapping(skb, queue_index);
1948 return netdev_get_tx_queue(dev, queue_index);
1951 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
1952 struct net_device *dev,
1953 struct netdev_queue *txq)
1955 spinlock_t *root_lock = qdisc_lock(q);
1958 spin_lock(root_lock);
1959 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1962 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
1963 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
1965 * This is a work-conserving queue; there are no old skbs
1966 * waiting to be sent out; and the qdisc is not running -
1967 * xmit the skb directly.
1969 __qdisc_update_bstats(q, skb->len);
1970 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
1973 clear_bit(__QDISC_STATE_RUNNING, &q->state);
1975 rc = NET_XMIT_SUCCESS;
1977 rc = qdisc_enqueue_root(skb, q);
1980 spin_unlock(root_lock);
1986 * Returns true if either:
1987 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
1988 * 2. skb is fragmented and the device does not support SG, or if
1989 * at least one of fragments is in highmem and device does not
1990 * support DMA from it.
1992 static inline int skb_needs_linearize(struct sk_buff *skb,
1993 struct net_device *dev)
1995 return (skb_has_frags(skb) && !(dev->features & NETIF_F_FRAGLIST)) ||
1996 (skb_shinfo(skb)->nr_frags && (!(dev->features & NETIF_F_SG) ||
1997 illegal_highdma(dev, skb)));
2001 * dev_queue_xmit - transmit a buffer
2002 * @skb: buffer to transmit
2004 * Queue a buffer for transmission to a network device. The caller must
2005 * have set the device and priority and built the buffer before calling
2006 * this function. The function can be called from an interrupt.
2008 * A negative errno code is returned on a failure. A success does not
2009 * guarantee the frame will be transmitted as it may be dropped due
2010 * to congestion or traffic shaping.
2012 * -----------------------------------------------------------------------------------
2013 * I notice this method can also return errors from the queue disciplines,
2014 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2017 * Regardless of the return value, the skb is consumed, so it is currently
2018 * difficult to retry a send to this method. (You can bump the ref count
2019 * before sending to hold a reference for retry if you are careful.)
2021 * When calling this method, interrupts MUST be enabled. This is because
2022 * the BH enable code must have IRQs enabled so that it will not deadlock.
2025 int dev_queue_xmit(struct sk_buff *skb)
2027 struct net_device *dev = skb->dev;
2028 struct netdev_queue *txq;
2032 /* GSO will handle the following emulations directly. */
2033 if (netif_needs_gso(dev, skb))
2036 /* Convert a paged skb to linear, if required */
2037 if (skb_needs_linearize(skb, dev) && __skb_linearize(skb))
2040 /* If packet is not checksummed and device does not support
2041 * checksumming for this protocol, complete checksumming here.
2043 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2044 skb_set_transport_header(skb, skb->csum_start -
2046 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
2051 /* Disable soft irqs for various locks below. Also
2052 * stops preemption for RCU.
2056 txq = dev_pick_tx(dev, skb);
2057 q = rcu_dereference(txq->qdisc);
2059 #ifdef CONFIG_NET_CLS_ACT
2060 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2063 rc = __dev_xmit_skb(skb, q, dev, txq);
2067 /* The device has no queue. Common case for software devices:
2068 loopback, all the sorts of tunnels...
2070 Really, it is unlikely that netif_tx_lock protection is necessary
2071 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2073 However, it is possible, that they rely on protection
2076 Check this and shot the lock. It is not prone from deadlocks.
2077 Either shot noqueue qdisc, it is even simpler 8)
2079 if (dev->flags & IFF_UP) {
2080 int cpu = smp_processor_id(); /* ok because BHs are off */
2082 if (txq->xmit_lock_owner != cpu) {
2084 HARD_TX_LOCK(dev, txq, cpu);
2086 if (!netif_tx_queue_stopped(txq)) {
2087 rc = dev_hard_start_xmit(skb, dev, txq);
2088 if (dev_xmit_complete(rc)) {
2089 HARD_TX_UNLOCK(dev, txq);
2093 HARD_TX_UNLOCK(dev, txq);
2094 if (net_ratelimit())
2095 printk(KERN_CRIT "Virtual device %s asks to "
2096 "queue packet!\n", dev->name);
2098 /* Recursion is detected! It is possible,
2100 if (net_ratelimit())
2101 printk(KERN_CRIT "Dead loop on virtual device "
2102 "%s, fix it urgently!\n", dev->name);
2107 rcu_read_unlock_bh();
2113 rcu_read_unlock_bh();
2116 EXPORT_SYMBOL(dev_queue_xmit);
2119 /*=======================================================================
2121 =======================================================================*/
2123 int netdev_max_backlog __read_mostly = 1000;
2124 int netdev_budget __read_mostly = 300;
2125 int weight_p __read_mostly = 64; /* old backlog weight */
2127 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2131 * netif_rx - post buffer to the network code
2132 * @skb: buffer to post
2134 * This function receives a packet from a device driver and queues it for
2135 * the upper (protocol) levels to process. It always succeeds. The buffer
2136 * may be dropped during processing for congestion control or by the
2140 * NET_RX_SUCCESS (no congestion)
2141 * NET_RX_DROP (packet was dropped)
2145 int netif_rx(struct sk_buff *skb)
2147 struct softnet_data *queue;
2148 unsigned long flags;
2150 /* if netpoll wants it, pretend we never saw it */
2151 if (netpoll_rx(skb))
2154 if (!skb->tstamp.tv64)
2158 * The code is rearranged so that the path is the most
2159 * short when CPU is congested, but is still operating.
2161 local_irq_save(flags);
2162 queue = &__get_cpu_var(softnet_data);
2164 __get_cpu_var(netdev_rx_stat).total++;
2165 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2166 if (queue->input_pkt_queue.qlen) {
2168 __skb_queue_tail(&queue->input_pkt_queue, skb);
2169 local_irq_restore(flags);
2170 return NET_RX_SUCCESS;
2173 napi_schedule(&queue->backlog);
2177 __get_cpu_var(netdev_rx_stat).dropped++;
2178 local_irq_restore(flags);
2183 EXPORT_SYMBOL(netif_rx);
2185 int netif_rx_ni(struct sk_buff *skb)
2190 err = netif_rx(skb);
2191 if (local_softirq_pending())
2197 EXPORT_SYMBOL(netif_rx_ni);
2199 static void net_tx_action(struct softirq_action *h)
2201 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2203 if (sd->completion_queue) {
2204 struct sk_buff *clist;
2206 local_irq_disable();
2207 clist = sd->completion_queue;
2208 sd->completion_queue = NULL;
2212 struct sk_buff *skb = clist;
2213 clist = clist->next;
2215 WARN_ON(atomic_read(&skb->users));
2220 if (sd->output_queue) {
2223 local_irq_disable();
2224 head = sd->output_queue;
2225 sd->output_queue = NULL;
2229 struct Qdisc *q = head;
2230 spinlock_t *root_lock;
2232 head = head->next_sched;
2234 root_lock = qdisc_lock(q);
2235 if (spin_trylock(root_lock)) {
2236 smp_mb__before_clear_bit();
2237 clear_bit(__QDISC_STATE_SCHED,
2240 spin_unlock(root_lock);
2242 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2244 __netif_reschedule(q);
2246 smp_mb__before_clear_bit();
2247 clear_bit(__QDISC_STATE_SCHED,
2255 static inline int deliver_skb(struct sk_buff *skb,
2256 struct packet_type *pt_prev,
2257 struct net_device *orig_dev)
2259 atomic_inc(&skb->users);
2260 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2263 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2265 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2266 /* This hook is defined here for ATM LANE */
2267 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2268 unsigned char *addr) __read_mostly;
2269 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2273 * If bridge module is loaded call bridging hook.
2274 * returns NULL if packet was consumed.
2276 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2277 struct sk_buff *skb) __read_mostly;
2278 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2280 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2281 struct packet_type **pt_prev, int *ret,
2282 struct net_device *orig_dev)
2284 struct net_bridge_port *port;
2286 if (skb->pkt_type == PACKET_LOOPBACK ||
2287 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2291 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2295 return br_handle_frame_hook(port, skb);
2298 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2301 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2302 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2303 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2305 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2306 struct packet_type **pt_prev,
2308 struct net_device *orig_dev)
2310 if (skb->dev->macvlan_port == NULL)
2314 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2317 return macvlan_handle_frame_hook(skb);
2320 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2323 #ifdef CONFIG_NET_CLS_ACT
2324 /* TODO: Maybe we should just force sch_ingress to be compiled in
2325 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2326 * a compare and 2 stores extra right now if we dont have it on
2327 * but have CONFIG_NET_CLS_ACT
2328 * NOTE: This doesnt stop any functionality; if you dont have
2329 * the ingress scheduler, you just cant add policies on ingress.
2332 static int ing_filter(struct sk_buff *skb)
2334 struct net_device *dev = skb->dev;
2335 u32 ttl = G_TC_RTTL(skb->tc_verd);
2336 struct netdev_queue *rxq;
2337 int result = TC_ACT_OK;
2340 if (MAX_RED_LOOP < ttl++) {
2342 "Redir loop detected Dropping packet (%d->%d)\n",
2343 skb->skb_iif, dev->ifindex);
2347 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2348 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2350 rxq = &dev->rx_queue;
2353 if (q != &noop_qdisc) {
2354 spin_lock(qdisc_lock(q));
2355 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2356 result = qdisc_enqueue_root(skb, q);
2357 spin_unlock(qdisc_lock(q));
2363 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2364 struct packet_type **pt_prev,
2365 int *ret, struct net_device *orig_dev)
2367 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2371 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2374 /* Huh? Why does turning on AF_PACKET affect this? */
2375 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2378 switch (ing_filter(skb)) {
2392 * netif_nit_deliver - deliver received packets to network taps
2395 * This function is used to deliver incoming packets to network
2396 * taps. It should be used when the normal netif_receive_skb path
2397 * is bypassed, for example because of VLAN acceleration.
2399 void netif_nit_deliver(struct sk_buff *skb)
2401 struct packet_type *ptype;
2403 if (list_empty(&ptype_all))
2406 skb_reset_network_header(skb);
2407 skb_reset_transport_header(skb);
2408 skb->mac_len = skb->network_header - skb->mac_header;
2411 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2412 if (!ptype->dev || ptype->dev == skb->dev)
2413 deliver_skb(skb, ptype, skb->dev);
2419 * netif_receive_skb - process receive buffer from network
2420 * @skb: buffer to process
2422 * netif_receive_skb() is the main receive data processing function.
2423 * It always succeeds. The buffer may be dropped during processing
2424 * for congestion control or by the protocol layers.
2426 * This function may only be called from softirq context and interrupts
2427 * should be enabled.
2429 * Return values (usually ignored):
2430 * NET_RX_SUCCESS: no congestion
2431 * NET_RX_DROP: packet was dropped
2433 int netif_receive_skb(struct sk_buff *skb)
2435 struct packet_type *ptype, *pt_prev;
2436 struct net_device *orig_dev;
2437 struct net_device *null_or_orig;
2438 struct net_device *null_or_bond;
2439 int ret = NET_RX_DROP;
2442 if (!skb->tstamp.tv64)
2445 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2446 return NET_RX_SUCCESS;
2448 /* if we've gotten here through NAPI, check netpoll */
2449 if (netpoll_receive_skb(skb))
2453 skb->skb_iif = skb->dev->ifindex;
2455 null_or_orig = NULL;
2456 orig_dev = skb->dev;
2457 if (orig_dev->master) {
2458 if (skb_bond_should_drop(skb))
2459 null_or_orig = orig_dev; /* deliver only exact match */
2461 skb->dev = orig_dev->master;
2464 __get_cpu_var(netdev_rx_stat).total++;
2466 skb_reset_network_header(skb);
2467 skb_reset_transport_header(skb);
2468 skb->mac_len = skb->network_header - skb->mac_header;
2474 #ifdef CONFIG_NET_CLS_ACT
2475 if (skb->tc_verd & TC_NCLS) {
2476 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2481 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2482 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2483 ptype->dev == orig_dev) {
2485 ret = deliver_skb(skb, pt_prev, orig_dev);
2490 #ifdef CONFIG_NET_CLS_ACT
2491 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2497 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2500 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2505 * Make sure frames received on VLAN interfaces stacked on
2506 * bonding interfaces still make their way to any base bonding
2507 * device that may have registered for a specific ptype. The
2508 * handler may have to adjust skb->dev and orig_dev.
2510 null_or_bond = NULL;
2511 if ((skb->dev->priv_flags & IFF_802_1Q_VLAN) &&
2512 (vlan_dev_real_dev(skb->dev)->priv_flags & IFF_BONDING)) {
2513 null_or_bond = vlan_dev_real_dev(skb->dev);
2516 type = skb->protocol;
2517 list_for_each_entry_rcu(ptype,
2518 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2519 if (ptype->type == type && (ptype->dev == null_or_orig ||
2520 ptype->dev == skb->dev || ptype->dev == orig_dev ||
2521 ptype->dev == null_or_bond)) {
2523 ret = deliver_skb(skb, pt_prev, orig_dev);
2529 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2532 /* Jamal, now you will not able to escape explaining
2533 * me how you were going to use this. :-)
2542 EXPORT_SYMBOL(netif_receive_skb);
2544 /* Network device is going away, flush any packets still pending */
2545 static void flush_backlog(void *arg)
2547 struct net_device *dev = arg;
2548 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2549 struct sk_buff *skb, *tmp;
2551 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2552 if (skb->dev == dev) {
2553 __skb_unlink(skb, &queue->input_pkt_queue);
2558 static int napi_gro_complete(struct sk_buff *skb)
2560 struct packet_type *ptype;
2561 __be16 type = skb->protocol;
2562 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2565 if (NAPI_GRO_CB(skb)->count == 1) {
2566 skb_shinfo(skb)->gso_size = 0;
2571 list_for_each_entry_rcu(ptype, head, list) {
2572 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2575 err = ptype->gro_complete(skb);
2581 WARN_ON(&ptype->list == head);
2583 return NET_RX_SUCCESS;
2587 return netif_receive_skb(skb);
2590 static void napi_gro_flush(struct napi_struct *napi)
2592 struct sk_buff *skb, *next;
2594 for (skb = napi->gro_list; skb; skb = next) {
2597 napi_gro_complete(skb);
2600 napi->gro_count = 0;
2601 napi->gro_list = NULL;
2604 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2606 struct sk_buff **pp = NULL;
2607 struct packet_type *ptype;
2608 __be16 type = skb->protocol;
2609 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2612 enum gro_result ret;
2614 if (!(skb->dev->features & NETIF_F_GRO))
2617 if (skb_is_gso(skb) || skb_has_frags(skb))
2621 list_for_each_entry_rcu(ptype, head, list) {
2622 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2625 skb_set_network_header(skb, skb_gro_offset(skb));
2626 mac_len = skb->network_header - skb->mac_header;
2627 skb->mac_len = mac_len;
2628 NAPI_GRO_CB(skb)->same_flow = 0;
2629 NAPI_GRO_CB(skb)->flush = 0;
2630 NAPI_GRO_CB(skb)->free = 0;
2632 pp = ptype->gro_receive(&napi->gro_list, skb);
2637 if (&ptype->list == head)
2640 same_flow = NAPI_GRO_CB(skb)->same_flow;
2641 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2644 struct sk_buff *nskb = *pp;
2648 napi_gro_complete(nskb);
2655 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2659 NAPI_GRO_CB(skb)->count = 1;
2660 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2661 skb->next = napi->gro_list;
2662 napi->gro_list = skb;
2666 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2667 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2669 BUG_ON(skb->end - skb->tail < grow);
2671 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2674 skb->data_len -= grow;
2676 skb_shinfo(skb)->frags[0].page_offset += grow;
2677 skb_shinfo(skb)->frags[0].size -= grow;
2679 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2680 put_page(skb_shinfo(skb)->frags[0].page);
2681 memmove(skb_shinfo(skb)->frags,
2682 skb_shinfo(skb)->frags + 1,
2683 --skb_shinfo(skb)->nr_frags);
2694 EXPORT_SYMBOL(dev_gro_receive);
2697 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2701 if (netpoll_rx_on(skb))
2704 for (p = napi->gro_list; p; p = p->next) {
2705 NAPI_GRO_CB(p)->same_flow =
2706 (p->dev == skb->dev) &&
2707 !compare_ether_header(skb_mac_header(p),
2708 skb_gro_mac_header(skb));
2709 NAPI_GRO_CB(p)->flush = 0;
2712 return dev_gro_receive(napi, skb);
2715 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
2719 if (netif_receive_skb(skb))
2724 case GRO_MERGED_FREE:
2735 EXPORT_SYMBOL(napi_skb_finish);
2737 void skb_gro_reset_offset(struct sk_buff *skb)
2739 NAPI_GRO_CB(skb)->data_offset = 0;
2740 NAPI_GRO_CB(skb)->frag0 = NULL;
2741 NAPI_GRO_CB(skb)->frag0_len = 0;
2743 if (skb->mac_header == skb->tail &&
2744 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2745 NAPI_GRO_CB(skb)->frag0 =
2746 page_address(skb_shinfo(skb)->frags[0].page) +
2747 skb_shinfo(skb)->frags[0].page_offset;
2748 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2751 EXPORT_SYMBOL(skb_gro_reset_offset);
2753 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2755 skb_gro_reset_offset(skb);
2757 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2759 EXPORT_SYMBOL(napi_gro_receive);
2761 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2763 __skb_pull(skb, skb_headlen(skb));
2764 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2768 EXPORT_SYMBOL(napi_reuse_skb);
2770 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2772 struct sk_buff *skb = napi->skb;
2775 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
2781 EXPORT_SYMBOL(napi_get_frags);
2783 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
2789 skb->protocol = eth_type_trans(skb, napi->dev);
2791 if (ret == GRO_HELD)
2792 skb_gro_pull(skb, -ETH_HLEN);
2793 else if (netif_receive_skb(skb))
2798 case GRO_MERGED_FREE:
2799 napi_reuse_skb(napi, skb);
2808 EXPORT_SYMBOL(napi_frags_finish);
2810 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2812 struct sk_buff *skb = napi->skb;
2819 skb_reset_mac_header(skb);
2820 skb_gro_reset_offset(skb);
2822 off = skb_gro_offset(skb);
2823 hlen = off + sizeof(*eth);
2824 eth = skb_gro_header_fast(skb, off);
2825 if (skb_gro_header_hard(skb, hlen)) {
2826 eth = skb_gro_header_slow(skb, hlen, off);
2827 if (unlikely(!eth)) {
2828 napi_reuse_skb(napi, skb);
2834 skb_gro_pull(skb, sizeof(*eth));
2837 * This works because the only protocols we care about don't require
2838 * special handling. We'll fix it up properly at the end.
2840 skb->protocol = eth->h_proto;
2845 EXPORT_SYMBOL(napi_frags_skb);
2847 gro_result_t napi_gro_frags(struct napi_struct *napi)
2849 struct sk_buff *skb = napi_frags_skb(napi);
2854 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2856 EXPORT_SYMBOL(napi_gro_frags);
2858 static int process_backlog(struct napi_struct *napi, int quota)
2861 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2862 unsigned long start_time = jiffies;
2864 napi->weight = weight_p;
2866 struct sk_buff *skb;
2868 local_irq_disable();
2869 skb = __skb_dequeue(&queue->input_pkt_queue);
2871 __napi_complete(napi);
2877 netif_receive_skb(skb);
2878 } while (++work < quota && jiffies == start_time);
2884 * __napi_schedule - schedule for receive
2885 * @n: entry to schedule
2887 * The entry's receive function will be scheduled to run
2889 void __napi_schedule(struct napi_struct *n)
2891 unsigned long flags;
2893 local_irq_save(flags);
2894 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2895 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2896 local_irq_restore(flags);
2898 EXPORT_SYMBOL(__napi_schedule);
2900 void __napi_complete(struct napi_struct *n)
2902 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2903 BUG_ON(n->gro_list);
2905 list_del(&n->poll_list);
2906 smp_mb__before_clear_bit();
2907 clear_bit(NAPI_STATE_SCHED, &n->state);
2909 EXPORT_SYMBOL(__napi_complete);
2911 void napi_complete(struct napi_struct *n)
2913 unsigned long flags;
2916 * don't let napi dequeue from the cpu poll list
2917 * just in case its running on a different cpu
2919 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2923 local_irq_save(flags);
2925 local_irq_restore(flags);
2927 EXPORT_SYMBOL(napi_complete);
2929 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2930 int (*poll)(struct napi_struct *, int), int weight)
2932 INIT_LIST_HEAD(&napi->poll_list);
2933 napi->gro_count = 0;
2934 napi->gro_list = NULL;
2937 napi->weight = weight;
2938 list_add(&napi->dev_list, &dev->napi_list);
2940 #ifdef CONFIG_NETPOLL
2941 spin_lock_init(&napi->poll_lock);
2942 napi->poll_owner = -1;
2944 set_bit(NAPI_STATE_SCHED, &napi->state);
2946 EXPORT_SYMBOL(netif_napi_add);
2948 void netif_napi_del(struct napi_struct *napi)
2950 struct sk_buff *skb, *next;
2952 list_del_init(&napi->dev_list);
2953 napi_free_frags(napi);
2955 for (skb = napi->gro_list; skb; skb = next) {
2961 napi->gro_list = NULL;
2962 napi->gro_count = 0;
2964 EXPORT_SYMBOL(netif_napi_del);
2967 static void net_rx_action(struct softirq_action *h)
2969 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2970 unsigned long time_limit = jiffies + 2;
2971 int budget = netdev_budget;
2974 local_irq_disable();
2976 while (!list_empty(list)) {
2977 struct napi_struct *n;
2980 /* If softirq window is exhuasted then punt.
2981 * Allow this to run for 2 jiffies since which will allow
2982 * an average latency of 1.5/HZ.
2984 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2989 /* Even though interrupts have been re-enabled, this
2990 * access is safe because interrupts can only add new
2991 * entries to the tail of this list, and only ->poll()
2992 * calls can remove this head entry from the list.
2994 n = list_entry(list->next, struct napi_struct, poll_list);
2996 have = netpoll_poll_lock(n);
3000 /* This NAPI_STATE_SCHED test is for avoiding a race
3001 * with netpoll's poll_napi(). Only the entity which
3002 * obtains the lock and sees NAPI_STATE_SCHED set will
3003 * actually make the ->poll() call. Therefore we avoid
3004 * accidently calling ->poll() when NAPI is not scheduled.
3007 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
3008 work = n->poll(n, weight);
3012 WARN_ON_ONCE(work > weight);
3016 local_irq_disable();
3018 /* Drivers must not modify the NAPI state if they
3019 * consume the entire weight. In such cases this code
3020 * still "owns" the NAPI instance and therefore can
3021 * move the instance around on the list at-will.
3023 if (unlikely(work == weight)) {
3024 if (unlikely(napi_disable_pending(n))) {
3027 local_irq_disable();
3029 list_move_tail(&n->poll_list, list);
3032 netpoll_poll_unlock(have);
3037 #ifdef CONFIG_NET_DMA
3039 * There may not be any more sk_buffs coming right now, so push
3040 * any pending DMA copies to hardware
3042 dma_issue_pending_all();
3048 __get_cpu_var(netdev_rx_stat).time_squeeze++;
3049 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3053 static gifconf_func_t *gifconf_list[NPROTO];
3056 * register_gifconf - register a SIOCGIF handler
3057 * @family: Address family
3058 * @gifconf: Function handler
3060 * Register protocol dependent address dumping routines. The handler
3061 * that is passed must not be freed or reused until it has been replaced
3062 * by another handler.
3064 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
3066 if (family >= NPROTO)
3068 gifconf_list[family] = gifconf;
3071 EXPORT_SYMBOL(register_gifconf);
3075 * Map an interface index to its name (SIOCGIFNAME)
3079 * We need this ioctl for efficient implementation of the
3080 * if_indextoname() function required by the IPv6 API. Without
3081 * it, we would have to search all the interfaces to find a
3085 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3087 struct net_device *dev;
3091 * Fetch the caller's info block.
3094 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3098 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3104 strcpy(ifr.ifr_name, dev->name);
3107 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3113 * Perform a SIOCGIFCONF call. This structure will change
3114 * size eventually, and there is nothing I can do about it.
3115 * Thus we will need a 'compatibility mode'.
3118 static int dev_ifconf(struct net *net, char __user *arg)
3121 struct net_device *dev;
3128 * Fetch the caller's info block.
3131 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3138 * Loop over the interfaces, and write an info block for each.
3142 for_each_netdev(net, dev) {
3143 for (i = 0; i < NPROTO; i++) {
3144 if (gifconf_list[i]) {
3147 done = gifconf_list[i](dev, NULL, 0);
3149 done = gifconf_list[i](dev, pos + total,
3159 * All done. Write the updated control block back to the caller.
3161 ifc.ifc_len = total;
3164 * Both BSD and Solaris return 0 here, so we do too.
3166 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3169 #ifdef CONFIG_PROC_FS
3171 * This is invoked by the /proc filesystem handler to display a device
3174 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3177 struct net *net = seq_file_net(seq);
3179 struct net_device *dev;
3183 return SEQ_START_TOKEN;
3186 for_each_netdev_rcu(net, dev)
3193 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3195 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3196 first_net_device(seq_file_net(seq)) :
3197 next_net_device((struct net_device *)v);
3200 return rcu_dereference(dev);
3203 void dev_seq_stop(struct seq_file *seq, void *v)
3209 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3211 const struct net_device_stats *stats = dev_get_stats(dev);
3213 seq_printf(seq, "%6s: %7lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3214 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3215 dev->name, stats->rx_bytes, stats->rx_packets,
3217 stats->rx_dropped + stats->rx_missed_errors,
3218 stats->rx_fifo_errors,
3219 stats->rx_length_errors + stats->rx_over_errors +
3220 stats->rx_crc_errors + stats->rx_frame_errors,
3221 stats->rx_compressed, stats->multicast,
3222 stats->tx_bytes, stats->tx_packets,
3223 stats->tx_errors, stats->tx_dropped,
3224 stats->tx_fifo_errors, stats->collisions,
3225 stats->tx_carrier_errors +
3226 stats->tx_aborted_errors +
3227 stats->tx_window_errors +
3228 stats->tx_heartbeat_errors,
3229 stats->tx_compressed);
3233 * Called from the PROCfs module. This now uses the new arbitrary sized
3234 * /proc/net interface to create /proc/net/dev
3236 static int dev_seq_show(struct seq_file *seq, void *v)
3238 if (v == SEQ_START_TOKEN)
3239 seq_puts(seq, "Inter-| Receive "
3241 " face |bytes packets errs drop fifo frame "
3242 "compressed multicast|bytes packets errs "
3243 "drop fifo colls carrier compressed\n");
3245 dev_seq_printf_stats(seq, v);
3249 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3251 struct netif_rx_stats *rc = NULL;
3253 while (*pos < nr_cpu_ids)
3254 if (cpu_online(*pos)) {
3255 rc = &per_cpu(netdev_rx_stat, *pos);
3262 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3264 return softnet_get_online(pos);
3267 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3270 return softnet_get_online(pos);
3273 static void softnet_seq_stop(struct seq_file *seq, void *v)
3277 static int softnet_seq_show(struct seq_file *seq, void *v)
3279 struct netif_rx_stats *s = v;
3281 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3282 s->total, s->dropped, s->time_squeeze, 0,
3283 0, 0, 0, 0, /* was fastroute */
3288 static const struct seq_operations dev_seq_ops = {
3289 .start = dev_seq_start,
3290 .next = dev_seq_next,
3291 .stop = dev_seq_stop,
3292 .show = dev_seq_show,
3295 static int dev_seq_open(struct inode *inode, struct file *file)
3297 return seq_open_net(inode, file, &dev_seq_ops,
3298 sizeof(struct seq_net_private));
3301 static const struct file_operations dev_seq_fops = {
3302 .owner = THIS_MODULE,
3303 .open = dev_seq_open,
3305 .llseek = seq_lseek,
3306 .release = seq_release_net,
3309 static const struct seq_operations softnet_seq_ops = {
3310 .start = softnet_seq_start,
3311 .next = softnet_seq_next,
3312 .stop = softnet_seq_stop,
3313 .show = softnet_seq_show,
3316 static int softnet_seq_open(struct inode *inode, struct file *file)
3318 return seq_open(file, &softnet_seq_ops);
3321 static const struct file_operations softnet_seq_fops = {
3322 .owner = THIS_MODULE,
3323 .open = softnet_seq_open,
3325 .llseek = seq_lseek,
3326 .release = seq_release,
3329 static void *ptype_get_idx(loff_t pos)
3331 struct packet_type *pt = NULL;
3335 list_for_each_entry_rcu(pt, &ptype_all, list) {
3341 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3342 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3351 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3355 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3358 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3360 struct packet_type *pt;
3361 struct list_head *nxt;
3365 if (v == SEQ_START_TOKEN)
3366 return ptype_get_idx(0);
3369 nxt = pt->list.next;
3370 if (pt->type == htons(ETH_P_ALL)) {
3371 if (nxt != &ptype_all)
3374 nxt = ptype_base[0].next;
3376 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3378 while (nxt == &ptype_base[hash]) {
3379 if (++hash >= PTYPE_HASH_SIZE)
3381 nxt = ptype_base[hash].next;
3384 return list_entry(nxt, struct packet_type, list);
3387 static void ptype_seq_stop(struct seq_file *seq, void *v)
3393 static int ptype_seq_show(struct seq_file *seq, void *v)
3395 struct packet_type *pt = v;
3397 if (v == SEQ_START_TOKEN)
3398 seq_puts(seq, "Type Device Function\n");
3399 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3400 if (pt->type == htons(ETH_P_ALL))
3401 seq_puts(seq, "ALL ");
3403 seq_printf(seq, "%04x", ntohs(pt->type));
3405 seq_printf(seq, " %-8s %pF\n",
3406 pt->dev ? pt->dev->name : "", pt->func);
3412 static const struct seq_operations ptype_seq_ops = {
3413 .start = ptype_seq_start,
3414 .next = ptype_seq_next,
3415 .stop = ptype_seq_stop,
3416 .show = ptype_seq_show,
3419 static int ptype_seq_open(struct inode *inode, struct file *file)
3421 return seq_open_net(inode, file, &ptype_seq_ops,
3422 sizeof(struct seq_net_private));
3425 static const struct file_operations ptype_seq_fops = {
3426 .owner = THIS_MODULE,
3427 .open = ptype_seq_open,
3429 .llseek = seq_lseek,
3430 .release = seq_release_net,
3434 static int __net_init dev_proc_net_init(struct net *net)
3438 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3440 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3442 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3445 if (wext_proc_init(net))
3451 proc_net_remove(net, "ptype");
3453 proc_net_remove(net, "softnet_stat");
3455 proc_net_remove(net, "dev");
3459 static void __net_exit dev_proc_net_exit(struct net *net)
3461 wext_proc_exit(net);
3463 proc_net_remove(net, "ptype");
3464 proc_net_remove(net, "softnet_stat");
3465 proc_net_remove(net, "dev");
3468 static struct pernet_operations __net_initdata dev_proc_ops = {
3469 .init = dev_proc_net_init,
3470 .exit = dev_proc_net_exit,
3473 static int __init dev_proc_init(void)
3475 return register_pernet_subsys(&dev_proc_ops);
3478 #define dev_proc_init() 0
3479 #endif /* CONFIG_PROC_FS */
3483 * netdev_set_master - set up master/slave pair
3484 * @slave: slave device
3485 * @master: new master device
3487 * Changes the master device of the slave. Pass %NULL to break the
3488 * bonding. The caller must hold the RTNL semaphore. On a failure
3489 * a negative errno code is returned. On success the reference counts
3490 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3491 * function returns zero.
3493 int netdev_set_master(struct net_device *slave, struct net_device *master)
3495 struct net_device *old = slave->master;
3505 slave->master = master;
3513 slave->flags |= IFF_SLAVE;
3515 slave->flags &= ~IFF_SLAVE;
3517 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3520 EXPORT_SYMBOL(netdev_set_master);
3522 static void dev_change_rx_flags(struct net_device *dev, int flags)
3524 const struct net_device_ops *ops = dev->netdev_ops;
3526 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3527 ops->ndo_change_rx_flags(dev, flags);
3530 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3532 unsigned short old_flags = dev->flags;
3538 dev->flags |= IFF_PROMISC;
3539 dev->promiscuity += inc;
3540 if (dev->promiscuity == 0) {
3543 * If inc causes overflow, untouch promisc and return error.
3546 dev->flags &= ~IFF_PROMISC;
3548 dev->promiscuity -= inc;
3549 printk(KERN_WARNING "%s: promiscuity touches roof, "
3550 "set promiscuity failed, promiscuity feature "
3551 "of device might be broken.\n", dev->name);
3555 if (dev->flags != old_flags) {
3556 printk(KERN_INFO "device %s %s promiscuous mode\n",
3557 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3559 if (audit_enabled) {
3560 current_uid_gid(&uid, &gid);
3561 audit_log(current->audit_context, GFP_ATOMIC,
3562 AUDIT_ANOM_PROMISCUOUS,
3563 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3564 dev->name, (dev->flags & IFF_PROMISC),
3565 (old_flags & IFF_PROMISC),
3566 audit_get_loginuid(current),
3568 audit_get_sessionid(current));
3571 dev_change_rx_flags(dev, IFF_PROMISC);
3577 * dev_set_promiscuity - update promiscuity count on a device
3581 * Add or remove promiscuity from a device. While the count in the device
3582 * remains above zero the interface remains promiscuous. Once it hits zero
3583 * the device reverts back to normal filtering operation. A negative inc
3584 * value is used to drop promiscuity on the device.
3585 * Return 0 if successful or a negative errno code on error.
3587 int dev_set_promiscuity(struct net_device *dev, int inc)
3589 unsigned short old_flags = dev->flags;
3592 err = __dev_set_promiscuity(dev, inc);
3595 if (dev->flags != old_flags)
3596 dev_set_rx_mode(dev);
3599 EXPORT_SYMBOL(dev_set_promiscuity);
3602 * dev_set_allmulti - update allmulti count on a device
3606 * Add or remove reception of all multicast frames to a device. While the
3607 * count in the device remains above zero the interface remains listening
3608 * to all interfaces. Once it hits zero the device reverts back to normal
3609 * filtering operation. A negative @inc value is used to drop the counter
3610 * when releasing a resource needing all multicasts.
3611 * Return 0 if successful or a negative errno code on error.
3614 int dev_set_allmulti(struct net_device *dev, int inc)
3616 unsigned short old_flags = dev->flags;
3620 dev->flags |= IFF_ALLMULTI;
3621 dev->allmulti += inc;
3622 if (dev->allmulti == 0) {
3625 * If inc causes overflow, untouch allmulti and return error.
3628 dev->flags &= ~IFF_ALLMULTI;
3630 dev->allmulti -= inc;
3631 printk(KERN_WARNING "%s: allmulti touches roof, "
3632 "set allmulti failed, allmulti feature of "
3633 "device might be broken.\n", dev->name);
3637 if (dev->flags ^ old_flags) {
3638 dev_change_rx_flags(dev, IFF_ALLMULTI);
3639 dev_set_rx_mode(dev);
3643 EXPORT_SYMBOL(dev_set_allmulti);
3646 * Upload unicast and multicast address lists to device and
3647 * configure RX filtering. When the device doesn't support unicast
3648 * filtering it is put in promiscuous mode while unicast addresses
3651 void __dev_set_rx_mode(struct net_device *dev)
3653 const struct net_device_ops *ops = dev->netdev_ops;
3655 /* dev_open will call this function so the list will stay sane. */
3656 if (!(dev->flags&IFF_UP))
3659 if (!netif_device_present(dev))
3662 if (ops->ndo_set_rx_mode)
3663 ops->ndo_set_rx_mode(dev);
3665 /* Unicast addresses changes may only happen under the rtnl,
3666 * therefore calling __dev_set_promiscuity here is safe.
3668 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
3669 __dev_set_promiscuity(dev, 1);
3670 dev->uc_promisc = 1;
3671 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
3672 __dev_set_promiscuity(dev, -1);
3673 dev->uc_promisc = 0;
3676 if (ops->ndo_set_multicast_list)
3677 ops->ndo_set_multicast_list(dev);
3681 void dev_set_rx_mode(struct net_device *dev)
3683 netif_addr_lock_bh(dev);
3684 __dev_set_rx_mode(dev);
3685 netif_addr_unlock_bh(dev);
3688 /* hw addresses list handling functions */
3690 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3691 int addr_len, unsigned char addr_type)
3693 struct netdev_hw_addr *ha;
3696 if (addr_len > MAX_ADDR_LEN)
3699 list_for_each_entry(ha, &list->list, list) {
3700 if (!memcmp(ha->addr, addr, addr_len) &&
3701 ha->type == addr_type) {
3708 alloc_size = sizeof(*ha);
3709 if (alloc_size < L1_CACHE_BYTES)
3710 alloc_size = L1_CACHE_BYTES;
3711 ha = kmalloc(alloc_size, GFP_ATOMIC);
3714 memcpy(ha->addr, addr, addr_len);
3715 ha->type = addr_type;
3718 list_add_tail_rcu(&ha->list, &list->list);
3723 static void ha_rcu_free(struct rcu_head *head)
3725 struct netdev_hw_addr *ha;
3727 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3731 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3732 int addr_len, unsigned char addr_type)
3734 struct netdev_hw_addr *ha;
3736 list_for_each_entry(ha, &list->list, list) {
3737 if (!memcmp(ha->addr, addr, addr_len) &&
3738 (ha->type == addr_type || !addr_type)) {
3741 list_del_rcu(&ha->list);
3742 call_rcu(&ha->rcu_head, ha_rcu_free);
3750 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
3751 struct netdev_hw_addr_list *from_list,
3753 unsigned char addr_type)
3756 struct netdev_hw_addr *ha, *ha2;
3759 list_for_each_entry(ha, &from_list->list, list) {
3760 type = addr_type ? addr_type : ha->type;
3761 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
3768 list_for_each_entry(ha2, &from_list->list, list) {
3771 type = addr_type ? addr_type : ha2->type;
3772 __hw_addr_del(to_list, ha2->addr, addr_len, type);
3777 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
3778 struct netdev_hw_addr_list *from_list,
3780 unsigned char addr_type)
3782 struct netdev_hw_addr *ha;
3785 list_for_each_entry(ha, &from_list->list, list) {
3786 type = addr_type ? addr_type : ha->type;
3787 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
3791 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
3792 struct netdev_hw_addr_list *from_list,
3796 struct netdev_hw_addr *ha, *tmp;
3798 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3800 err = __hw_addr_add(to_list, ha->addr,
3801 addr_len, ha->type);
3806 } else if (ha->refcount == 1) {
3807 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
3808 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
3814 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
3815 struct netdev_hw_addr_list *from_list,
3818 struct netdev_hw_addr *ha, *tmp;
3820 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3822 __hw_addr_del(to_list, ha->addr,
3823 addr_len, ha->type);
3825 __hw_addr_del(from_list, ha->addr,
3826 addr_len, ha->type);
3831 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
3833 struct netdev_hw_addr *ha, *tmp;
3835 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3836 list_del_rcu(&ha->list);
3837 call_rcu(&ha->rcu_head, ha_rcu_free);
3842 static void __hw_addr_init(struct netdev_hw_addr_list *list)
3844 INIT_LIST_HEAD(&list->list);
3848 /* Device addresses handling functions */
3850 static void dev_addr_flush(struct net_device *dev)
3852 /* rtnl_mutex must be held here */
3854 __hw_addr_flush(&dev->dev_addrs);
3855 dev->dev_addr = NULL;
3858 static int dev_addr_init(struct net_device *dev)
3860 unsigned char addr[MAX_ADDR_LEN];
3861 struct netdev_hw_addr *ha;
3864 /* rtnl_mutex must be held here */
3866 __hw_addr_init(&dev->dev_addrs);
3867 memset(addr, 0, sizeof(addr));
3868 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
3869 NETDEV_HW_ADDR_T_LAN);
3872 * Get the first (previously created) address from the list
3873 * and set dev_addr pointer to this location.
3875 ha = list_first_entry(&dev->dev_addrs.list,
3876 struct netdev_hw_addr, list);
3877 dev->dev_addr = ha->addr;
3883 * dev_addr_add - Add a device address
3885 * @addr: address to add
3886 * @addr_type: address type
3888 * Add a device address to the device or increase the reference count if
3889 * it already exists.
3891 * The caller must hold the rtnl_mutex.
3893 int dev_addr_add(struct net_device *dev, unsigned char *addr,
3894 unsigned char addr_type)
3900 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
3902 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3905 EXPORT_SYMBOL(dev_addr_add);
3908 * dev_addr_del - Release a device address.
3910 * @addr: address to delete
3911 * @addr_type: address type
3913 * Release reference to a device address and remove it from the device
3914 * if the reference count drops to zero.
3916 * The caller must hold the rtnl_mutex.
3918 int dev_addr_del(struct net_device *dev, unsigned char *addr,
3919 unsigned char addr_type)
3922 struct netdev_hw_addr *ha;
3927 * We can not remove the first address from the list because
3928 * dev->dev_addr points to that.
3930 ha = list_first_entry(&dev->dev_addrs.list,
3931 struct netdev_hw_addr, list);
3932 if (ha->addr == dev->dev_addr && ha->refcount == 1)
3935 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
3938 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3941 EXPORT_SYMBOL(dev_addr_del);
3944 * dev_addr_add_multiple - Add device addresses from another device
3945 * @to_dev: device to which addresses will be added
3946 * @from_dev: device from which addresses will be added
3947 * @addr_type: address type - 0 means type will be used from from_dev
3949 * Add device addresses of the one device to another.
3951 * The caller must hold the rtnl_mutex.
3953 int dev_addr_add_multiple(struct net_device *to_dev,
3954 struct net_device *from_dev,
3955 unsigned char addr_type)
3961 if (from_dev->addr_len != to_dev->addr_len)
3963 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3964 to_dev->addr_len, addr_type);
3966 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3969 EXPORT_SYMBOL(dev_addr_add_multiple);
3972 * dev_addr_del_multiple - Delete device addresses by another device
3973 * @to_dev: device where the addresses will be deleted
3974 * @from_dev: device by which addresses the addresses will be deleted
3975 * @addr_type: address type - 0 means type will used from from_dev
3977 * Deletes addresses in to device by the list of addresses in from device.
3979 * The caller must hold the rtnl_mutex.
3981 int dev_addr_del_multiple(struct net_device *to_dev,
3982 struct net_device *from_dev,
3983 unsigned char addr_type)
3987 if (from_dev->addr_len != to_dev->addr_len)
3989 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3990 to_dev->addr_len, addr_type);
3991 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3994 EXPORT_SYMBOL(dev_addr_del_multiple);
3996 /* multicast addresses handling functions */
3998 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3999 void *addr, int alen, int glbl)
4001 struct dev_addr_list *da;
4003 for (; (da = *list) != NULL; list = &da->next) {
4004 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
4005 alen == da->da_addrlen) {
4007 int old_glbl = da->da_gusers;
4024 int __dev_addr_add(struct dev_addr_list **list, int *count,
4025 void *addr, int alen, int glbl)
4027 struct dev_addr_list *da;
4029 for (da = *list; da != NULL; da = da->next) {
4030 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
4031 da->da_addrlen == alen) {
4033 int old_glbl = da->da_gusers;
4043 da = kzalloc(sizeof(*da), GFP_ATOMIC);
4046 memcpy(da->da_addr, addr, alen);
4047 da->da_addrlen = alen;
4049 da->da_gusers = glbl ? 1 : 0;
4057 * dev_unicast_delete - Release secondary unicast address.
4059 * @addr: address to delete
4061 * Release reference to a secondary unicast address and remove it
4062 * from the device if the reference count drops to zero.
4064 * The caller must hold the rtnl_mutex.
4066 int dev_unicast_delete(struct net_device *dev, void *addr)
4072 netif_addr_lock_bh(dev);
4073 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
4074 NETDEV_HW_ADDR_T_UNICAST);
4076 __dev_set_rx_mode(dev);
4077 netif_addr_unlock_bh(dev);
4080 EXPORT_SYMBOL(dev_unicast_delete);
4083 * dev_unicast_add - add a secondary unicast address
4085 * @addr: address to add
4087 * Add a secondary unicast address to the device or increase
4088 * the reference count if it already exists.
4090 * The caller must hold the rtnl_mutex.
4092 int dev_unicast_add(struct net_device *dev, void *addr)
4098 netif_addr_lock_bh(dev);
4099 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
4100 NETDEV_HW_ADDR_T_UNICAST);
4102 __dev_set_rx_mode(dev);
4103 netif_addr_unlock_bh(dev);
4106 EXPORT_SYMBOL(dev_unicast_add);
4108 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
4109 struct dev_addr_list **from, int *from_count)
4111 struct dev_addr_list *da, *next;
4115 while (da != NULL) {
4117 if (!da->da_synced) {
4118 err = __dev_addr_add(to, to_count,
4119 da->da_addr, da->da_addrlen, 0);
4124 } else if (da->da_users == 1) {
4125 __dev_addr_delete(to, to_count,
4126 da->da_addr, da->da_addrlen, 0);
4127 __dev_addr_delete(from, from_count,
4128 da->da_addr, da->da_addrlen, 0);
4134 EXPORT_SYMBOL_GPL(__dev_addr_sync);
4136 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
4137 struct dev_addr_list **from, int *from_count)
4139 struct dev_addr_list *da, *next;
4142 while (da != NULL) {
4144 if (da->da_synced) {
4145 __dev_addr_delete(to, to_count,
4146 da->da_addr, da->da_addrlen, 0);
4148 __dev_addr_delete(from, from_count,
4149 da->da_addr, da->da_addrlen, 0);
4154 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
4157 * dev_unicast_sync - Synchronize device's unicast list to another device
4158 * @to: destination device
4159 * @from: source device
4161 * Add newly added addresses to the destination device and release
4162 * addresses that have no users left. The source device must be
4163 * locked by netif_tx_lock_bh.
4165 * This function is intended to be called from the dev->set_rx_mode
4166 * function of layered software devices.
4168 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4172 if (to->addr_len != from->addr_len)
4175 netif_addr_lock_bh(to);
4176 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4178 __dev_set_rx_mode(to);
4179 netif_addr_unlock_bh(to);
4182 EXPORT_SYMBOL(dev_unicast_sync);
4185 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4186 * @to: destination device
4187 * @from: source device
4189 * Remove all addresses that were added to the destination device by
4190 * dev_unicast_sync(). This function is intended to be called from the
4191 * dev->stop function of layered software devices.
4193 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4195 if (to->addr_len != from->addr_len)
4198 netif_addr_lock_bh(from);
4199 netif_addr_lock(to);
4200 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4201 __dev_set_rx_mode(to);
4202 netif_addr_unlock(to);
4203 netif_addr_unlock_bh(from);
4205 EXPORT_SYMBOL(dev_unicast_unsync);
4207 static void dev_unicast_flush(struct net_device *dev)
4209 netif_addr_lock_bh(dev);
4210 __hw_addr_flush(&dev->uc);
4211 netif_addr_unlock_bh(dev);
4214 static void dev_unicast_init(struct net_device *dev)
4216 __hw_addr_init(&dev->uc);
4220 static void __dev_addr_discard(struct dev_addr_list **list)
4222 struct dev_addr_list *tmp;
4224 while (*list != NULL) {
4227 if (tmp->da_users > tmp->da_gusers)
4228 printk("__dev_addr_discard: address leakage! "
4229 "da_users=%d\n", tmp->da_users);
4234 static void dev_addr_discard(struct net_device *dev)
4236 netif_addr_lock_bh(dev);
4238 __dev_addr_discard(&dev->mc_list);
4241 netif_addr_unlock_bh(dev);
4245 * dev_get_flags - get flags reported to userspace
4248 * Get the combination of flag bits exported through APIs to userspace.
4250 unsigned dev_get_flags(const struct net_device *dev)
4254 flags = (dev->flags & ~(IFF_PROMISC |
4259 (dev->gflags & (IFF_PROMISC |
4262 if (netif_running(dev)) {
4263 if (netif_oper_up(dev))
4264 flags |= IFF_RUNNING;
4265 if (netif_carrier_ok(dev))
4266 flags |= IFF_LOWER_UP;
4267 if (netif_dormant(dev))
4268 flags |= IFF_DORMANT;
4273 EXPORT_SYMBOL(dev_get_flags);
4276 * dev_change_flags - change device settings
4278 * @flags: device state flags
4280 * Change settings on device based state flags. The flags are
4281 * in the userspace exported format.
4283 int dev_change_flags(struct net_device *dev, unsigned flags)
4286 int old_flags = dev->flags;
4291 * Set the flags on our device.
4294 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4295 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4297 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4301 * Load in the correct multicast list now the flags have changed.
4304 if ((old_flags ^ flags) & IFF_MULTICAST)
4305 dev_change_rx_flags(dev, IFF_MULTICAST);
4307 dev_set_rx_mode(dev);
4310 * Have we downed the interface. We handle IFF_UP ourselves
4311 * according to user attempts to set it, rather than blindly
4316 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4317 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
4320 dev_set_rx_mode(dev);
4323 if (dev->flags & IFF_UP &&
4324 ((old_flags ^ dev->flags) & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
4326 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4328 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4329 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4331 dev->gflags ^= IFF_PROMISC;
4332 dev_set_promiscuity(dev, inc);
4335 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4336 is important. Some (broken) drivers set IFF_PROMISC, when
4337 IFF_ALLMULTI is requested not asking us and not reporting.
4339 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4340 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4342 dev->gflags ^= IFF_ALLMULTI;
4343 dev_set_allmulti(dev, inc);
4346 /* Exclude state transition flags, already notified */
4347 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
4349 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4353 EXPORT_SYMBOL(dev_change_flags);
4356 * dev_set_mtu - Change maximum transfer unit
4358 * @new_mtu: new transfer unit
4360 * Change the maximum transfer size of the network device.
4362 int dev_set_mtu(struct net_device *dev, int new_mtu)
4364 const struct net_device_ops *ops = dev->netdev_ops;
4367 if (new_mtu == dev->mtu)
4370 /* MTU must be positive. */
4374 if (!netif_device_present(dev))
4378 if (ops->ndo_change_mtu)
4379 err = ops->ndo_change_mtu(dev, new_mtu);
4383 if (!err && dev->flags & IFF_UP)
4384 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4387 EXPORT_SYMBOL(dev_set_mtu);
4390 * dev_set_mac_address - Change Media Access Control Address
4394 * Change the hardware (MAC) address of the device
4396 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4398 const struct net_device_ops *ops = dev->netdev_ops;
4401 if (!ops->ndo_set_mac_address)
4403 if (sa->sa_family != dev->type)
4405 if (!netif_device_present(dev))
4407 err = ops->ndo_set_mac_address(dev, sa);
4409 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4412 EXPORT_SYMBOL(dev_set_mac_address);
4415 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4417 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4420 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4426 case SIOCGIFFLAGS: /* Get interface flags */
4427 ifr->ifr_flags = (short) dev_get_flags(dev);
4430 case SIOCGIFMETRIC: /* Get the metric on the interface
4431 (currently unused) */
4432 ifr->ifr_metric = 0;
4435 case SIOCGIFMTU: /* Get the MTU of a device */
4436 ifr->ifr_mtu = dev->mtu;
4441 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4443 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4444 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4445 ifr->ifr_hwaddr.sa_family = dev->type;
4453 ifr->ifr_map.mem_start = dev->mem_start;
4454 ifr->ifr_map.mem_end = dev->mem_end;
4455 ifr->ifr_map.base_addr = dev->base_addr;
4456 ifr->ifr_map.irq = dev->irq;
4457 ifr->ifr_map.dma = dev->dma;
4458 ifr->ifr_map.port = dev->if_port;
4462 ifr->ifr_ifindex = dev->ifindex;
4466 ifr->ifr_qlen = dev->tx_queue_len;
4470 /* dev_ioctl() should ensure this case
4482 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4484 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4487 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4488 const struct net_device_ops *ops;
4493 ops = dev->netdev_ops;
4496 case SIOCSIFFLAGS: /* Set interface flags */
4497 return dev_change_flags(dev, ifr->ifr_flags);
4499 case SIOCSIFMETRIC: /* Set the metric on the interface
4500 (currently unused) */
4503 case SIOCSIFMTU: /* Set the MTU of a device */
4504 return dev_set_mtu(dev, ifr->ifr_mtu);
4507 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4509 case SIOCSIFHWBROADCAST:
4510 if (ifr->ifr_hwaddr.sa_family != dev->type)
4512 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4513 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4514 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4518 if (ops->ndo_set_config) {
4519 if (!netif_device_present(dev))
4521 return ops->ndo_set_config(dev, &ifr->ifr_map);
4526 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4527 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4529 if (!netif_device_present(dev))
4531 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4535 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4536 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4538 if (!netif_device_present(dev))
4540 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4544 if (ifr->ifr_qlen < 0)
4546 dev->tx_queue_len = ifr->ifr_qlen;
4550 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4551 return dev_change_name(dev, ifr->ifr_newname);
4554 * Unknown or private ioctl
4557 if ((cmd >= SIOCDEVPRIVATE &&
4558 cmd <= SIOCDEVPRIVATE + 15) ||
4559 cmd == SIOCBONDENSLAVE ||
4560 cmd == SIOCBONDRELEASE ||
4561 cmd == SIOCBONDSETHWADDR ||
4562 cmd == SIOCBONDSLAVEINFOQUERY ||
4563 cmd == SIOCBONDINFOQUERY ||
4564 cmd == SIOCBONDCHANGEACTIVE ||
4565 cmd == SIOCGMIIPHY ||
4566 cmd == SIOCGMIIREG ||
4567 cmd == SIOCSMIIREG ||
4568 cmd == SIOCBRADDIF ||
4569 cmd == SIOCBRDELIF ||
4570 cmd == SIOCSHWTSTAMP ||
4571 cmd == SIOCWANDEV) {
4573 if (ops->ndo_do_ioctl) {
4574 if (netif_device_present(dev))
4575 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4587 * This function handles all "interface"-type I/O control requests. The actual
4588 * 'doing' part of this is dev_ifsioc above.
4592 * dev_ioctl - network device ioctl
4593 * @net: the applicable net namespace
4594 * @cmd: command to issue
4595 * @arg: pointer to a struct ifreq in user space
4597 * Issue ioctl functions to devices. This is normally called by the
4598 * user space syscall interfaces but can sometimes be useful for
4599 * other purposes. The return value is the return from the syscall if
4600 * positive or a negative errno code on error.
4603 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4609 /* One special case: SIOCGIFCONF takes ifconf argument
4610 and requires shared lock, because it sleeps writing
4614 if (cmd == SIOCGIFCONF) {
4616 ret = dev_ifconf(net, (char __user *) arg);
4620 if (cmd == SIOCGIFNAME)
4621 return dev_ifname(net, (struct ifreq __user *)arg);
4623 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4626 ifr.ifr_name[IFNAMSIZ-1] = 0;
4628 colon = strchr(ifr.ifr_name, ':');
4633 * See which interface the caller is talking about.
4638 * These ioctl calls:
4639 * - can be done by all.
4640 * - atomic and do not require locking.
4651 dev_load(net, ifr.ifr_name);
4653 ret = dev_ifsioc_locked(net, &ifr, cmd);
4658 if (copy_to_user(arg, &ifr,
4659 sizeof(struct ifreq)))
4665 dev_load(net, ifr.ifr_name);
4667 ret = dev_ethtool(net, &ifr);
4672 if (copy_to_user(arg, &ifr,
4673 sizeof(struct ifreq)))
4679 * These ioctl calls:
4680 * - require superuser power.
4681 * - require strict serialization.
4687 if (!capable(CAP_NET_ADMIN))
4689 dev_load(net, ifr.ifr_name);
4691 ret = dev_ifsioc(net, &ifr, cmd);
4696 if (copy_to_user(arg, &ifr,
4697 sizeof(struct ifreq)))
4703 * These ioctl calls:
4704 * - require superuser power.
4705 * - require strict serialization.
4706 * - do not return a value
4716 case SIOCSIFHWBROADCAST:
4719 case SIOCBONDENSLAVE:
4720 case SIOCBONDRELEASE:
4721 case SIOCBONDSETHWADDR:
4722 case SIOCBONDCHANGEACTIVE:
4726 if (!capable(CAP_NET_ADMIN))
4729 case SIOCBONDSLAVEINFOQUERY:
4730 case SIOCBONDINFOQUERY:
4731 dev_load(net, ifr.ifr_name);
4733 ret = dev_ifsioc(net, &ifr, cmd);
4738 /* Get the per device memory space. We can add this but
4739 * currently do not support it */
4741 /* Set the per device memory buffer space.
4742 * Not applicable in our case */
4747 * Unknown or private ioctl.
4750 if (cmd == SIOCWANDEV ||
4751 (cmd >= SIOCDEVPRIVATE &&
4752 cmd <= SIOCDEVPRIVATE + 15)) {
4753 dev_load(net, ifr.ifr_name);
4755 ret = dev_ifsioc(net, &ifr, cmd);
4757 if (!ret && copy_to_user(arg, &ifr,
4758 sizeof(struct ifreq)))
4762 /* Take care of Wireless Extensions */
4763 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4764 return wext_handle_ioctl(net, &ifr, cmd, arg);
4771 * dev_new_index - allocate an ifindex
4772 * @net: the applicable net namespace
4774 * Returns a suitable unique value for a new device interface
4775 * number. The caller must hold the rtnl semaphore or the
4776 * dev_base_lock to be sure it remains unique.
4778 static int dev_new_index(struct net *net)
4784 if (!__dev_get_by_index(net, ifindex))
4789 /* Delayed registration/unregisteration */
4790 static LIST_HEAD(net_todo_list);
4792 static void net_set_todo(struct net_device *dev)
4794 list_add_tail(&dev->todo_list, &net_todo_list);
4797 static void rollback_registered_many(struct list_head *head)
4799 struct net_device *dev, *tmp;
4801 BUG_ON(dev_boot_phase);
4804 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
4805 /* Some devices call without registering
4806 * for initialization unwind. Remove those
4807 * devices and proceed with the remaining.
4809 if (dev->reg_state == NETREG_UNINITIALIZED) {
4810 pr_debug("unregister_netdevice: device %s/%p never "
4811 "was registered\n", dev->name, dev);
4814 list_del(&dev->unreg_list);
4818 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4820 /* If device is running, close it first. */
4823 /* And unlink it from device chain. */
4824 unlist_netdevice(dev);
4826 dev->reg_state = NETREG_UNREGISTERING;
4831 list_for_each_entry(dev, head, unreg_list) {
4832 /* Shutdown queueing discipline. */
4836 /* Notify protocols, that we are about to destroy
4837 this device. They should clean all the things.
4839 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4842 * Flush the unicast and multicast chains
4844 dev_unicast_flush(dev);
4845 dev_addr_discard(dev);
4847 if (dev->netdev_ops->ndo_uninit)
4848 dev->netdev_ops->ndo_uninit(dev);
4850 /* Notifier chain MUST detach us from master device. */
4851 WARN_ON(dev->master);
4853 /* Remove entries from kobject tree */
4854 netdev_unregister_kobject(dev);
4857 /* Process any work delayed until the end of the batch */
4858 dev = list_entry(head->next, struct net_device, unreg_list);
4859 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
4863 list_for_each_entry(dev, head, unreg_list)
4867 static void rollback_registered(struct net_device *dev)
4871 list_add(&dev->unreg_list, &single);
4872 rollback_registered_many(&single);
4875 static void __netdev_init_queue_locks_one(struct net_device *dev,
4876 struct netdev_queue *dev_queue,
4879 spin_lock_init(&dev_queue->_xmit_lock);
4880 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4881 dev_queue->xmit_lock_owner = -1;
4884 static void netdev_init_queue_locks(struct net_device *dev)
4886 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4887 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4890 unsigned long netdev_fix_features(unsigned long features, const char *name)
4892 /* Fix illegal SG+CSUM combinations. */
4893 if ((features & NETIF_F_SG) &&
4894 !(features & NETIF_F_ALL_CSUM)) {
4896 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4897 "checksum feature.\n", name);
4898 features &= ~NETIF_F_SG;
4901 /* TSO requires that SG is present as well. */
4902 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4904 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4905 "SG feature.\n", name);
4906 features &= ~NETIF_F_TSO;
4909 if (features & NETIF_F_UFO) {
4910 if (!(features & NETIF_F_GEN_CSUM)) {
4912 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4913 "since no NETIF_F_HW_CSUM feature.\n",
4915 features &= ~NETIF_F_UFO;
4918 if (!(features & NETIF_F_SG)) {
4920 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4921 "since no NETIF_F_SG feature.\n", name);
4922 features &= ~NETIF_F_UFO;
4928 EXPORT_SYMBOL(netdev_fix_features);
4931 * netif_stacked_transfer_operstate - transfer operstate
4932 * @rootdev: the root or lower level device to transfer state from
4933 * @dev: the device to transfer operstate to
4935 * Transfer operational state from root to device. This is normally
4936 * called when a stacking relationship exists between the root
4937 * device and the device(a leaf device).
4939 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
4940 struct net_device *dev)
4942 if (rootdev->operstate == IF_OPER_DORMANT)
4943 netif_dormant_on(dev);
4945 netif_dormant_off(dev);
4947 if (netif_carrier_ok(rootdev)) {
4948 if (!netif_carrier_ok(dev))
4949 netif_carrier_on(dev);
4951 if (netif_carrier_ok(dev))
4952 netif_carrier_off(dev);
4955 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
4958 * register_netdevice - register a network device
4959 * @dev: device to register
4961 * Take a completed network device structure and add it to the kernel
4962 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4963 * chain. 0 is returned on success. A negative errno code is returned
4964 * on a failure to set up the device, or if the name is a duplicate.
4966 * Callers must hold the rtnl semaphore. You may want
4967 * register_netdev() instead of this.
4970 * The locking appears insufficient to guarantee two parallel registers
4971 * will not get the same name.
4974 int register_netdevice(struct net_device *dev)
4977 struct net *net = dev_net(dev);
4979 BUG_ON(dev_boot_phase);
4984 /* When net_device's are persistent, this will be fatal. */
4985 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4988 spin_lock_init(&dev->addr_list_lock);
4989 netdev_set_addr_lockdep_class(dev);
4990 netdev_init_queue_locks(dev);
4994 /* Init, if this function is available */
4995 if (dev->netdev_ops->ndo_init) {
4996 ret = dev->netdev_ops->ndo_init(dev);
5004 ret = dev_get_valid_name(net, dev->name, dev->name, 0);
5008 dev->ifindex = dev_new_index(net);
5009 if (dev->iflink == -1)
5010 dev->iflink = dev->ifindex;
5012 /* Fix illegal checksum combinations */
5013 if ((dev->features & NETIF_F_HW_CSUM) &&
5014 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5015 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
5017 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
5020 if ((dev->features & NETIF_F_NO_CSUM) &&
5021 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
5022 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
5024 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
5027 dev->features = netdev_fix_features(dev->features, dev->name);
5029 /* Enable software GSO if SG is supported. */
5030 if (dev->features & NETIF_F_SG)
5031 dev->features |= NETIF_F_GSO;
5033 netdev_initialize_kobject(dev);
5035 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
5036 ret = notifier_to_errno(ret);
5040 ret = netdev_register_kobject(dev);
5043 dev->reg_state = NETREG_REGISTERED;
5046 * Default initial state at registry is that the
5047 * device is present.
5050 set_bit(__LINK_STATE_PRESENT, &dev->state);
5052 dev_init_scheduler(dev);
5054 list_netdevice(dev);
5056 /* Notify protocols, that a new device appeared. */
5057 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
5058 ret = notifier_to_errno(ret);
5060 rollback_registered(dev);
5061 dev->reg_state = NETREG_UNREGISTERED;
5064 * Prevent userspace races by waiting until the network
5065 * device is fully setup before sending notifications.
5067 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5073 if (dev->netdev_ops->ndo_uninit)
5074 dev->netdev_ops->ndo_uninit(dev);
5077 EXPORT_SYMBOL(register_netdevice);
5080 * init_dummy_netdev - init a dummy network device for NAPI
5081 * @dev: device to init
5083 * This takes a network device structure and initialize the minimum
5084 * amount of fields so it can be used to schedule NAPI polls without
5085 * registering a full blown interface. This is to be used by drivers
5086 * that need to tie several hardware interfaces to a single NAPI
5087 * poll scheduler due to HW limitations.
5089 int init_dummy_netdev(struct net_device *dev)
5091 /* Clear everything. Note we don't initialize spinlocks
5092 * are they aren't supposed to be taken by any of the
5093 * NAPI code and this dummy netdev is supposed to be
5094 * only ever used for NAPI polls
5096 memset(dev, 0, sizeof(struct net_device));
5098 /* make sure we BUG if trying to hit standard
5099 * register/unregister code path
5101 dev->reg_state = NETREG_DUMMY;
5103 /* initialize the ref count */
5104 atomic_set(&dev->refcnt, 1);
5106 /* NAPI wants this */
5107 INIT_LIST_HEAD(&dev->napi_list);
5109 /* a dummy interface is started by default */
5110 set_bit(__LINK_STATE_PRESENT, &dev->state);
5111 set_bit(__LINK_STATE_START, &dev->state);
5115 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5119 * register_netdev - register a network device
5120 * @dev: device to register
5122 * Take a completed network device structure and add it to the kernel
5123 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5124 * chain. 0 is returned on success. A negative errno code is returned
5125 * on a failure to set up the device, or if the name is a duplicate.
5127 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5128 * and expands the device name if you passed a format string to
5131 int register_netdev(struct net_device *dev)
5138 * If the name is a format string the caller wants us to do a
5141 if (strchr(dev->name, '%')) {
5142 err = dev_alloc_name(dev, dev->name);
5147 err = register_netdevice(dev);
5152 EXPORT_SYMBOL(register_netdev);
5155 * netdev_wait_allrefs - wait until all references are gone.
5157 * This is called when unregistering network devices.
5159 * Any protocol or device that holds a reference should register
5160 * for netdevice notification, and cleanup and put back the
5161 * reference if they receive an UNREGISTER event.
5162 * We can get stuck here if buggy protocols don't correctly
5165 static void netdev_wait_allrefs(struct net_device *dev)
5167 unsigned long rebroadcast_time, warning_time;
5169 linkwatch_forget_dev(dev);
5171 rebroadcast_time = warning_time = jiffies;
5172 while (atomic_read(&dev->refcnt) != 0) {
5173 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5176 /* Rebroadcast unregister notification */
5177 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5178 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5179 * should have already handle it the first time */
5181 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5183 /* We must not have linkwatch events
5184 * pending on unregister. If this
5185 * happens, we simply run the queue
5186 * unscheduled, resulting in a noop
5189 linkwatch_run_queue();
5194 rebroadcast_time = jiffies;
5199 if (time_after(jiffies, warning_time + 10 * HZ)) {
5200 printk(KERN_EMERG "unregister_netdevice: "
5201 "waiting for %s to become free. Usage "
5203 dev->name, atomic_read(&dev->refcnt));
5204 warning_time = jiffies;
5213 * register_netdevice(x1);
5214 * register_netdevice(x2);
5216 * unregister_netdevice(y1);
5217 * unregister_netdevice(y2);
5223 * We are invoked by rtnl_unlock().
5224 * This allows us to deal with problems:
5225 * 1) We can delete sysfs objects which invoke hotplug
5226 * without deadlocking with linkwatch via keventd.
5227 * 2) Since we run with the RTNL semaphore not held, we can sleep
5228 * safely in order to wait for the netdev refcnt to drop to zero.
5230 * We must not return until all unregister events added during
5231 * the interval the lock was held have been completed.
5233 void netdev_run_todo(void)
5235 struct list_head list;
5237 /* Snapshot list, allow later requests */
5238 list_replace_init(&net_todo_list, &list);
5242 while (!list_empty(&list)) {
5243 struct net_device *dev
5244 = list_entry(list.next, struct net_device, todo_list);
5245 list_del(&dev->todo_list);
5247 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5248 printk(KERN_ERR "network todo '%s' but state %d\n",
5249 dev->name, dev->reg_state);
5254 dev->reg_state = NETREG_UNREGISTERED;
5256 on_each_cpu(flush_backlog, dev, 1);
5258 netdev_wait_allrefs(dev);
5261 BUG_ON(atomic_read(&dev->refcnt));
5262 WARN_ON(dev->ip_ptr);
5263 WARN_ON(dev->ip6_ptr);
5264 WARN_ON(dev->dn_ptr);
5266 if (dev->destructor)
5267 dev->destructor(dev);
5269 /* Free network device */
5270 kobject_put(&dev->dev.kobj);
5275 * dev_txq_stats_fold - fold tx_queues stats
5276 * @dev: device to get statistics from
5277 * @stats: struct net_device_stats to hold results
5279 void dev_txq_stats_fold(const struct net_device *dev,
5280 struct net_device_stats *stats)
5282 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5284 struct netdev_queue *txq;
5286 for (i = 0; i < dev->num_tx_queues; i++) {
5287 txq = netdev_get_tx_queue(dev, i);
5288 tx_bytes += txq->tx_bytes;
5289 tx_packets += txq->tx_packets;
5290 tx_dropped += txq->tx_dropped;
5292 if (tx_bytes || tx_packets || tx_dropped) {
5293 stats->tx_bytes = tx_bytes;
5294 stats->tx_packets = tx_packets;
5295 stats->tx_dropped = tx_dropped;
5298 EXPORT_SYMBOL(dev_txq_stats_fold);
5301 * dev_get_stats - get network device statistics
5302 * @dev: device to get statistics from
5304 * Get network statistics from device. The device driver may provide
5305 * its own method by setting dev->netdev_ops->get_stats; otherwise
5306 * the internal statistics structure is used.
5308 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5310 const struct net_device_ops *ops = dev->netdev_ops;
5312 if (ops->ndo_get_stats)
5313 return ops->ndo_get_stats(dev);
5315 dev_txq_stats_fold(dev, &dev->stats);
5318 EXPORT_SYMBOL(dev_get_stats);
5320 static void netdev_init_one_queue(struct net_device *dev,
5321 struct netdev_queue *queue,
5327 static void netdev_init_queues(struct net_device *dev)
5329 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5330 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5331 spin_lock_init(&dev->tx_global_lock);
5335 * alloc_netdev_mq - allocate network device
5336 * @sizeof_priv: size of private data to allocate space for
5337 * @name: device name format string
5338 * @setup: callback to initialize device
5339 * @queue_count: the number of subqueues to allocate
5341 * Allocates a struct net_device with private data area for driver use
5342 * and performs basic initialization. Also allocates subquue structs
5343 * for each queue on the device at the end of the netdevice.
5345 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5346 void (*setup)(struct net_device *), unsigned int queue_count)
5348 struct netdev_queue *tx;
5349 struct net_device *dev;
5351 struct net_device *p;
5353 BUG_ON(strlen(name) >= sizeof(dev->name));
5355 alloc_size = sizeof(struct net_device);
5357 /* ensure 32-byte alignment of private area */
5358 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5359 alloc_size += sizeof_priv;
5361 /* ensure 32-byte alignment of whole construct */
5362 alloc_size += NETDEV_ALIGN - 1;
5364 p = kzalloc(alloc_size, GFP_KERNEL);
5366 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5370 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5372 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5377 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5378 dev->padded = (char *)dev - (char *)p;
5380 if (dev_addr_init(dev))
5383 dev_unicast_init(dev);
5385 dev_net_set(dev, &init_net);
5388 dev->num_tx_queues = queue_count;
5389 dev->real_num_tx_queues = queue_count;
5391 dev->gso_max_size = GSO_MAX_SIZE;
5393 netdev_init_queues(dev);
5395 INIT_LIST_HEAD(&dev->napi_list);
5396 INIT_LIST_HEAD(&dev->unreg_list);
5397 INIT_LIST_HEAD(&dev->link_watch_list);
5398 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5400 strcpy(dev->name, name);
5410 EXPORT_SYMBOL(alloc_netdev_mq);
5413 * free_netdev - free network device
5416 * This function does the last stage of destroying an allocated device
5417 * interface. The reference to the device object is released.
5418 * If this is the last reference then it will be freed.
5420 void free_netdev(struct net_device *dev)
5422 struct napi_struct *p, *n;
5424 release_net(dev_net(dev));
5428 /* Flush device addresses */
5429 dev_addr_flush(dev);
5431 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5434 /* Compatibility with error handling in drivers */
5435 if (dev->reg_state == NETREG_UNINITIALIZED) {
5436 kfree((char *)dev - dev->padded);
5440 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5441 dev->reg_state = NETREG_RELEASED;
5443 /* will free via device release */
5444 put_device(&dev->dev);
5446 EXPORT_SYMBOL(free_netdev);
5449 * synchronize_net - Synchronize with packet receive processing
5451 * Wait for packets currently being received to be done.
5452 * Does not block later packets from starting.
5454 void synchronize_net(void)
5459 EXPORT_SYMBOL(synchronize_net);
5462 * unregister_netdevice_queue - remove device from the kernel
5466 * This function shuts down a device interface and removes it
5467 * from the kernel tables.
5468 * If head not NULL, device is queued to be unregistered later.
5470 * Callers must hold the rtnl semaphore. You may want
5471 * unregister_netdev() instead of this.
5474 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5479 list_move_tail(&dev->unreg_list, head);
5481 rollback_registered(dev);
5482 /* Finish processing unregister after unlock */
5486 EXPORT_SYMBOL(unregister_netdevice_queue);
5489 * unregister_netdevice_many - unregister many devices
5490 * @head: list of devices
5492 void unregister_netdevice_many(struct list_head *head)
5494 struct net_device *dev;
5496 if (!list_empty(head)) {
5497 rollback_registered_many(head);
5498 list_for_each_entry(dev, head, unreg_list)
5502 EXPORT_SYMBOL(unregister_netdevice_many);
5505 * unregister_netdev - remove device from the kernel
5508 * This function shuts down a device interface and removes it
5509 * from the kernel tables.
5511 * This is just a wrapper for unregister_netdevice that takes
5512 * the rtnl semaphore. In general you want to use this and not
5513 * unregister_netdevice.
5515 void unregister_netdev(struct net_device *dev)
5518 unregister_netdevice(dev);
5521 EXPORT_SYMBOL(unregister_netdev);
5524 * dev_change_net_namespace - move device to different nethost namespace
5526 * @net: network namespace
5527 * @pat: If not NULL name pattern to try if the current device name
5528 * is already taken in the destination network namespace.
5530 * This function shuts down a device interface and moves it
5531 * to a new network namespace. On success 0 is returned, on
5532 * a failure a netagive errno code is returned.
5534 * Callers must hold the rtnl semaphore.
5537 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5543 /* Don't allow namespace local devices to be moved. */
5545 if (dev->features & NETIF_F_NETNS_LOCAL)
5549 /* Don't allow real devices to be moved when sysfs
5553 if (dev->dev.parent)
5557 /* Ensure the device has been registrered */
5559 if (dev->reg_state != NETREG_REGISTERED)
5562 /* Get out if there is nothing todo */
5564 if (net_eq(dev_net(dev), net))
5567 /* Pick the destination device name, and ensure
5568 * we can use it in the destination network namespace.
5571 if (__dev_get_by_name(net, dev->name)) {
5572 /* We get here if we can't use the current device name */
5575 if (dev_get_valid_name(net, pat, dev->name, 1))
5580 * And now a mini version of register_netdevice unregister_netdevice.
5583 /* If device is running close it first. */
5586 /* And unlink it from device chain */
5588 unlist_netdevice(dev);
5592 /* Shutdown queueing discipline. */
5595 /* Notify protocols, that we are about to destroy
5596 this device. They should clean all the things.
5598 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5599 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH, dev);
5602 * Flush the unicast and multicast chains
5604 dev_unicast_flush(dev);
5605 dev_addr_discard(dev);
5607 netdev_unregister_kobject(dev);
5609 /* Actually switch the network namespace */
5610 dev_net_set(dev, net);
5612 /* If there is an ifindex conflict assign a new one */
5613 if (__dev_get_by_index(net, dev->ifindex)) {
5614 int iflink = (dev->iflink == dev->ifindex);
5615 dev->ifindex = dev_new_index(net);
5617 dev->iflink = dev->ifindex;
5620 /* Fixup kobjects */
5621 err = netdev_register_kobject(dev);
5624 /* Add the device back in the hashes */
5625 list_netdevice(dev);
5627 /* Notify protocols, that a new device appeared. */
5628 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5631 * Prevent userspace races by waiting until the network
5632 * device is fully setup before sending notifications.
5634 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U);
5641 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5643 static int dev_cpu_callback(struct notifier_block *nfb,
5644 unsigned long action,
5647 struct sk_buff **list_skb;
5648 struct Qdisc **list_net;
5649 struct sk_buff *skb;
5650 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5651 struct softnet_data *sd, *oldsd;
5653 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5656 local_irq_disable();
5657 cpu = smp_processor_id();
5658 sd = &per_cpu(softnet_data, cpu);
5659 oldsd = &per_cpu(softnet_data, oldcpu);
5661 /* Find end of our completion_queue. */
5662 list_skb = &sd->completion_queue;
5664 list_skb = &(*list_skb)->next;
5665 /* Append completion queue from offline CPU. */
5666 *list_skb = oldsd->completion_queue;
5667 oldsd->completion_queue = NULL;
5669 /* Find end of our output_queue. */
5670 list_net = &sd->output_queue;
5672 list_net = &(*list_net)->next_sched;
5673 /* Append output queue from offline CPU. */
5674 *list_net = oldsd->output_queue;
5675 oldsd->output_queue = NULL;
5677 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5680 /* Process offline CPU's input_pkt_queue */
5681 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5689 * netdev_increment_features - increment feature set by one
5690 * @all: current feature set
5691 * @one: new feature set
5692 * @mask: mask feature set
5694 * Computes a new feature set after adding a device with feature set
5695 * @one to the master device with current feature set @all. Will not
5696 * enable anything that is off in @mask. Returns the new feature set.
5698 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5701 /* If device needs checksumming, downgrade to it. */
5702 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5703 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5704 else if (mask & NETIF_F_ALL_CSUM) {
5705 /* If one device supports v4/v6 checksumming, set for all. */
5706 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5707 !(all & NETIF_F_GEN_CSUM)) {
5708 all &= ~NETIF_F_ALL_CSUM;
5709 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5712 /* If one device supports hw checksumming, set for all. */
5713 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5714 all &= ~NETIF_F_ALL_CSUM;
5715 all |= NETIF_F_HW_CSUM;
5719 one |= NETIF_F_ALL_CSUM;
5721 one |= all & NETIF_F_ONE_FOR_ALL;
5722 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5723 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5727 EXPORT_SYMBOL(netdev_increment_features);
5729 static struct hlist_head *netdev_create_hash(void)
5732 struct hlist_head *hash;
5734 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5736 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5737 INIT_HLIST_HEAD(&hash[i]);
5742 /* Initialize per network namespace state */
5743 static int __net_init netdev_init(struct net *net)
5745 INIT_LIST_HEAD(&net->dev_base_head);
5747 net->dev_name_head = netdev_create_hash();
5748 if (net->dev_name_head == NULL)
5751 net->dev_index_head = netdev_create_hash();
5752 if (net->dev_index_head == NULL)
5758 kfree(net->dev_name_head);
5764 * netdev_drivername - network driver for the device
5765 * @dev: network device
5766 * @buffer: buffer for resulting name
5767 * @len: size of buffer
5769 * Determine network driver for device.
5771 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5773 const struct device_driver *driver;
5774 const struct device *parent;
5776 if (len <= 0 || !buffer)
5780 parent = dev->dev.parent;
5785 driver = parent->driver;
5786 if (driver && driver->name)
5787 strlcpy(buffer, driver->name, len);
5791 static void __net_exit netdev_exit(struct net *net)
5793 kfree(net->dev_name_head);
5794 kfree(net->dev_index_head);
5797 static struct pernet_operations __net_initdata netdev_net_ops = {
5798 .init = netdev_init,
5799 .exit = netdev_exit,
5802 static void __net_exit default_device_exit(struct net *net)
5804 struct net_device *dev, *aux;
5806 * Push all migratable network devices back to the
5807 * initial network namespace
5810 for_each_netdev_safe(net, dev, aux) {
5812 char fb_name[IFNAMSIZ];
5814 /* Ignore unmoveable devices (i.e. loopback) */
5815 if (dev->features & NETIF_F_NETNS_LOCAL)
5818 /* Leave virtual devices for the generic cleanup */
5819 if (dev->rtnl_link_ops)
5822 /* Push remaing network devices to init_net */
5823 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5824 err = dev_change_net_namespace(dev, &init_net, fb_name);
5826 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5827 __func__, dev->name, err);
5834 static void __net_exit default_device_exit_batch(struct list_head *net_list)
5836 /* At exit all network devices most be removed from a network
5837 * namespace. Do this in the reverse order of registeration.
5838 * Do this across as many network namespaces as possible to
5839 * improve batching efficiency.
5841 struct net_device *dev;
5843 LIST_HEAD(dev_kill_list);
5846 list_for_each_entry(net, net_list, exit_list) {
5847 for_each_netdev_reverse(net, dev) {
5848 if (dev->rtnl_link_ops)
5849 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
5851 unregister_netdevice_queue(dev, &dev_kill_list);
5854 unregister_netdevice_many(&dev_kill_list);
5858 static struct pernet_operations __net_initdata default_device_ops = {
5859 .exit = default_device_exit,
5860 .exit_batch = default_device_exit_batch,
5864 * Initialize the DEV module. At boot time this walks the device list and
5865 * unhooks any devices that fail to initialise (normally hardware not
5866 * present) and leaves us with a valid list of present and active devices.
5871 * This is called single threaded during boot, so no need
5872 * to take the rtnl semaphore.
5874 static int __init net_dev_init(void)
5876 int i, rc = -ENOMEM;
5878 BUG_ON(!dev_boot_phase);
5880 if (dev_proc_init())
5883 if (netdev_kobject_init())
5886 INIT_LIST_HEAD(&ptype_all);
5887 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5888 INIT_LIST_HEAD(&ptype_base[i]);
5890 if (register_pernet_subsys(&netdev_net_ops))
5894 * Initialise the packet receive queues.
5897 for_each_possible_cpu(i) {
5898 struct softnet_data *queue;
5900 queue = &per_cpu(softnet_data, i);
5901 skb_queue_head_init(&queue->input_pkt_queue);
5902 queue->completion_queue = NULL;
5903 INIT_LIST_HEAD(&queue->poll_list);
5905 queue->backlog.poll = process_backlog;
5906 queue->backlog.weight = weight_p;
5907 queue->backlog.gro_list = NULL;
5908 queue->backlog.gro_count = 0;
5913 /* The loopback device is special if any other network devices
5914 * is present in a network namespace the loopback device must
5915 * be present. Since we now dynamically allocate and free the
5916 * loopback device ensure this invariant is maintained by
5917 * keeping the loopback device as the first device on the
5918 * list of network devices. Ensuring the loopback devices
5919 * is the first device that appears and the last network device
5922 if (register_pernet_device(&loopback_net_ops))
5925 if (register_pernet_device(&default_device_ops))
5928 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5929 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5931 hotcpu_notifier(dev_cpu_callback, 0);
5939 subsys_initcall(net_dev_init);
5941 static int __init initialize_hashrnd(void)
5943 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5947 late_initcall_sync(initialize_hashrnd);