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 <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/kmod.h>
111 #include <linux/module.h>
112 #include <linux/netpoll.h>
113 #include <linux/rcupdate.h>
114 #include <linux/delay.h>
115 #include <net/wext.h>
116 #include <net/iw_handler.h>
117 #include <asm/current.h>
118 #include <linux/audit.h>
119 #include <linux/dmaengine.h>
120 #include <linux/err.h>
121 #include <linux/ctype.h>
122 #include <linux/if_arp.h>
123 #include <linux/if_vlan.h>
124 #include <linux/ip.h>
126 #include <linux/ipv6.h>
127 #include <linux/in.h>
128 #include <linux/jhash.h>
129 #include <linux/random.h>
130 #include <trace/events/napi.h>
132 #include "net-sysfs.h"
134 /* Instead of increasing this, you should create a hash table. */
135 #define MAX_GRO_SKBS 8
137 /* This should be increased if a protocol with a bigger head is added. */
138 #define GRO_MAX_HEAD (MAX_HEADER + 128)
141 * The list of packet types we will receive (as opposed to discard)
142 * and the routines to invoke.
144 * Why 16. Because with 16 the only overlap we get on a hash of the
145 * low nibble of the protocol value is RARP/SNAP/X.25.
147 * NOTE: That is no longer true with the addition of VLAN tags. Not
148 * sure which should go first, but I bet it won't make much
149 * difference if we are running VLANs. The good news is that
150 * this protocol won't be in the list unless compiled in, so
151 * the average user (w/out VLANs) will not be adversely affected.
168 #define PTYPE_HASH_SIZE (16)
169 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
171 static DEFINE_SPINLOCK(ptype_lock);
172 static struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
173 static struct list_head ptype_all __read_mostly; /* Taps */
176 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
179 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
181 * Writers must hold the rtnl semaphore while they loop through the
182 * dev_base_head list, and hold dev_base_lock for writing when they do the
183 * actual updates. This allows pure readers to access the list even
184 * while a writer is preparing to update it.
186 * To put it another way, dev_base_lock is held for writing only to
187 * protect against pure readers; the rtnl semaphore provides the
188 * protection against other writers.
190 * See, for example usages, register_netdevice() and
191 * unregister_netdevice(), which must be called with the rtnl
194 DEFINE_RWLOCK(dev_base_lock);
195 EXPORT_SYMBOL(dev_base_lock);
197 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
199 unsigned hash = full_name_hash(name, strnlen(name, IFNAMSIZ));
200 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
203 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
205 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
208 /* Device list insertion */
209 static int list_netdevice(struct net_device *dev)
211 struct net *net = dev_net(dev);
215 write_lock_bh(&dev_base_lock);
216 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
217 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
218 hlist_add_head_rcu(&dev->index_hlist,
219 dev_index_hash(net, dev->ifindex));
220 write_unlock_bh(&dev_base_lock);
224 /* Device list removal
225 * caller must respect a RCU grace period before freeing/reusing dev
227 static void unlist_netdevice(struct net_device *dev)
231 /* Unlink dev from the device chain */
232 write_lock_bh(&dev_base_lock);
233 list_del_rcu(&dev->dev_list);
234 hlist_del_rcu(&dev->name_hlist);
235 hlist_del_rcu(&dev->index_hlist);
236 write_unlock_bh(&dev_base_lock);
243 static RAW_NOTIFIER_HEAD(netdev_chain);
246 * Device drivers call our routines to queue packets here. We empty the
247 * queue in the local softnet handler.
250 DEFINE_PER_CPU(struct softnet_data, softnet_data);
251 EXPORT_PER_CPU_SYMBOL(softnet_data);
253 #ifdef CONFIG_LOCKDEP
255 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
256 * according to dev->type
258 static const unsigned short netdev_lock_type[] =
259 {ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
260 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
261 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
262 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
263 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
264 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
265 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
266 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
267 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
268 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
269 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
270 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
271 ARPHRD_FCFABRIC, ARPHRD_IEEE802_TR, ARPHRD_IEEE80211,
272 ARPHRD_IEEE80211_PRISM, ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET,
273 ARPHRD_PHONET_PIPE, ARPHRD_IEEE802154,
274 ARPHRD_VOID, ARPHRD_NONE};
276 static const char *const netdev_lock_name[] =
277 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
278 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
279 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
280 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
281 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
282 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
283 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
284 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
285 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
286 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
287 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
288 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
289 "_xmit_FCFABRIC", "_xmit_IEEE802_TR", "_xmit_IEEE80211",
290 "_xmit_IEEE80211_PRISM", "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET",
291 "_xmit_PHONET_PIPE", "_xmit_IEEE802154",
292 "_xmit_VOID", "_xmit_NONE"};
294 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
295 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
297 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
301 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
302 if (netdev_lock_type[i] == dev_type)
304 /* the last key is used by default */
305 return ARRAY_SIZE(netdev_lock_type) - 1;
308 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
309 unsigned short dev_type)
313 i = netdev_lock_pos(dev_type);
314 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
315 netdev_lock_name[i]);
318 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
322 i = netdev_lock_pos(dev->type);
323 lockdep_set_class_and_name(&dev->addr_list_lock,
324 &netdev_addr_lock_key[i],
325 netdev_lock_name[i]);
328 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
329 unsigned short dev_type)
332 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
337 /*******************************************************************************
339 Protocol management and registration routines
341 *******************************************************************************/
344 * Add a protocol ID to the list. Now that the input handler is
345 * smarter we can dispense with all the messy stuff that used to be
348 * BEWARE!!! Protocol handlers, mangling input packets,
349 * MUST BE last in hash buckets and checking protocol handlers
350 * MUST start from promiscuous ptype_all chain in net_bh.
351 * It is true now, do not change it.
352 * Explanation follows: if protocol handler, mangling packet, will
353 * be the first on list, it is not able to sense, that packet
354 * is cloned and should be copied-on-write, so that it will
355 * change it and subsequent readers will get broken packet.
360 * dev_add_pack - add packet handler
361 * @pt: packet type declaration
363 * Add a protocol handler to the networking stack. The passed &packet_type
364 * is linked into kernel lists and may not be freed until it has been
365 * removed from the kernel lists.
367 * This call does not sleep therefore it can not
368 * guarantee all CPU's that are in middle of receiving packets
369 * will see the new packet type (until the next received packet).
372 void dev_add_pack(struct packet_type *pt)
376 spin_lock_bh(&ptype_lock);
377 if (pt->type == htons(ETH_P_ALL))
378 list_add_rcu(&pt->list, &ptype_all);
380 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
381 list_add_rcu(&pt->list, &ptype_base[hash]);
383 spin_unlock_bh(&ptype_lock);
385 EXPORT_SYMBOL(dev_add_pack);
388 * __dev_remove_pack - remove packet handler
389 * @pt: packet type declaration
391 * Remove a protocol handler that was previously added to the kernel
392 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
393 * from the kernel lists and can be freed or reused once this function
396 * The packet type might still be in use by receivers
397 * and must not be freed until after all the CPU's have gone
398 * through a quiescent state.
400 void __dev_remove_pack(struct packet_type *pt)
402 struct list_head *head;
403 struct packet_type *pt1;
405 spin_lock_bh(&ptype_lock);
407 if (pt->type == htons(ETH_P_ALL))
410 head = &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
412 list_for_each_entry(pt1, head, list) {
414 list_del_rcu(&pt->list);
419 printk(KERN_WARNING "dev_remove_pack: %p not found.\n", pt);
421 spin_unlock_bh(&ptype_lock);
423 EXPORT_SYMBOL(__dev_remove_pack);
426 * dev_remove_pack - remove packet handler
427 * @pt: packet type declaration
429 * Remove a protocol handler that was previously added to the kernel
430 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
431 * from the kernel lists and can be freed or reused once this function
434 * This call sleeps to guarantee that no CPU is looking at the packet
437 void dev_remove_pack(struct packet_type *pt)
439 __dev_remove_pack(pt);
443 EXPORT_SYMBOL(dev_remove_pack);
445 /******************************************************************************
447 Device Boot-time Settings Routines
449 *******************************************************************************/
451 /* Boot time configuration table */
452 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
455 * netdev_boot_setup_add - add new setup entry
456 * @name: name of the device
457 * @map: configured settings for the device
459 * Adds new setup entry to the dev_boot_setup list. The function
460 * returns 0 on error and 1 on success. This is a generic routine to
463 static int netdev_boot_setup_add(char *name, struct ifmap *map)
465 struct netdev_boot_setup *s;
469 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
470 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
471 memset(s[i].name, 0, sizeof(s[i].name));
472 strlcpy(s[i].name, name, IFNAMSIZ);
473 memcpy(&s[i].map, map, sizeof(s[i].map));
478 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
482 * netdev_boot_setup_check - check boot time settings
483 * @dev: the netdevice
485 * Check boot time settings for the device.
486 * The found settings are set for the device to be used
487 * later in the device probing.
488 * Returns 0 if no settings found, 1 if they are.
490 int netdev_boot_setup_check(struct net_device *dev)
492 struct netdev_boot_setup *s = dev_boot_setup;
495 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
496 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
497 !strcmp(dev->name, s[i].name)) {
498 dev->irq = s[i].map.irq;
499 dev->base_addr = s[i].map.base_addr;
500 dev->mem_start = s[i].map.mem_start;
501 dev->mem_end = s[i].map.mem_end;
507 EXPORT_SYMBOL(netdev_boot_setup_check);
511 * netdev_boot_base - get address from boot time settings
512 * @prefix: prefix for network device
513 * @unit: id for network device
515 * Check boot time settings for the base address of device.
516 * The found settings are set for the device to be used
517 * later in the device probing.
518 * Returns 0 if no settings found.
520 unsigned long netdev_boot_base(const char *prefix, int unit)
522 const struct netdev_boot_setup *s = dev_boot_setup;
526 sprintf(name, "%s%d", prefix, unit);
529 * If device already registered then return base of 1
530 * to indicate not to probe for this interface
532 if (__dev_get_by_name(&init_net, name))
535 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
536 if (!strcmp(name, s[i].name))
537 return s[i].map.base_addr;
542 * Saves at boot time configured settings for any netdevice.
544 int __init netdev_boot_setup(char *str)
549 str = get_options(str, ARRAY_SIZE(ints), ints);
554 memset(&map, 0, sizeof(map));
558 map.base_addr = ints[2];
560 map.mem_start = ints[3];
562 map.mem_end = ints[4];
564 /* Add new entry to the list */
565 return netdev_boot_setup_add(str, &map);
568 __setup("netdev=", netdev_boot_setup);
570 /*******************************************************************************
572 Device Interface Subroutines
574 *******************************************************************************/
577 * __dev_get_by_name - find a device by its name
578 * @net: the applicable net namespace
579 * @name: name to find
581 * Find an interface by name. Must be called under RTNL semaphore
582 * or @dev_base_lock. If the name is found a pointer to the device
583 * is returned. If the name is not found then %NULL is returned. The
584 * reference counters are not incremented so the caller must be
585 * careful with locks.
588 struct net_device *__dev_get_by_name(struct net *net, const char *name)
590 struct hlist_node *p;
591 struct net_device *dev;
592 struct hlist_head *head = dev_name_hash(net, name);
594 hlist_for_each_entry(dev, p, head, name_hlist)
595 if (!strncmp(dev->name, name, IFNAMSIZ))
600 EXPORT_SYMBOL(__dev_get_by_name);
603 * dev_get_by_name_rcu - find a device by its name
604 * @net: the applicable net namespace
605 * @name: name to find
607 * Find an interface by name.
608 * If the name is found a pointer to the device is returned.
609 * If the name is not found then %NULL is returned.
610 * The reference counters are not incremented so the caller must be
611 * careful with locks. The caller must hold RCU lock.
614 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
616 struct hlist_node *p;
617 struct net_device *dev;
618 struct hlist_head *head = dev_name_hash(net, name);
620 hlist_for_each_entry_rcu(dev, p, head, name_hlist)
621 if (!strncmp(dev->name, name, IFNAMSIZ))
626 EXPORT_SYMBOL(dev_get_by_name_rcu);
629 * dev_get_by_name - find a device by its name
630 * @net: the applicable net namespace
631 * @name: name to find
633 * Find an interface by name. This can be called from any
634 * context and does its own locking. The returned handle has
635 * the usage count incremented and the caller must use dev_put() to
636 * release it when it is no longer needed. %NULL is returned if no
637 * matching device is found.
640 struct net_device *dev_get_by_name(struct net *net, const char *name)
642 struct net_device *dev;
645 dev = dev_get_by_name_rcu(net, name);
651 EXPORT_SYMBOL(dev_get_by_name);
654 * __dev_get_by_index - find a device by its ifindex
655 * @net: the applicable net namespace
656 * @ifindex: index of device
658 * Search for an interface by index. Returns %NULL if the device
659 * is not found or a pointer to the device. The device has not
660 * had its reference counter increased so the caller must be careful
661 * about locking. The caller must hold either the RTNL semaphore
665 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
667 struct hlist_node *p;
668 struct net_device *dev;
669 struct hlist_head *head = dev_index_hash(net, ifindex);
671 hlist_for_each_entry(dev, p, head, index_hlist)
672 if (dev->ifindex == ifindex)
677 EXPORT_SYMBOL(__dev_get_by_index);
680 * dev_get_by_index_rcu - find a device by its ifindex
681 * @net: the applicable net namespace
682 * @ifindex: index of device
684 * Search for an interface by index. Returns %NULL if the device
685 * is not found or a pointer to the device. The device has not
686 * had its reference counter increased so the caller must be careful
687 * about locking. The caller must hold RCU lock.
690 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
692 struct hlist_node *p;
693 struct net_device *dev;
694 struct hlist_head *head = dev_index_hash(net, ifindex);
696 hlist_for_each_entry_rcu(dev, p, head, index_hlist)
697 if (dev->ifindex == ifindex)
702 EXPORT_SYMBOL(dev_get_by_index_rcu);
706 * dev_get_by_index - find a device by its ifindex
707 * @net: the applicable net namespace
708 * @ifindex: index of device
710 * Search for an interface by index. Returns NULL if the device
711 * is not found or a pointer to the device. The device returned has
712 * had a reference added and the pointer is safe until the user calls
713 * dev_put to indicate they have finished with it.
716 struct net_device *dev_get_by_index(struct net *net, int ifindex)
718 struct net_device *dev;
721 dev = dev_get_by_index_rcu(net, ifindex);
727 EXPORT_SYMBOL(dev_get_by_index);
730 * dev_getbyhwaddr - find a device by its hardware address
731 * @net: the applicable net namespace
732 * @type: media type of device
733 * @ha: hardware address
735 * Search for an interface by MAC address. Returns NULL if the device
736 * is not found or a pointer to the device. The caller must hold the
737 * rtnl semaphore. The returned device has not had its ref count increased
738 * and the caller must therefore be careful about locking
741 * If the API was consistent this would be __dev_get_by_hwaddr
744 struct net_device *dev_getbyhwaddr(struct net *net, unsigned short type, char *ha)
746 struct net_device *dev;
750 for_each_netdev(net, dev)
751 if (dev->type == type &&
752 !memcmp(dev->dev_addr, ha, dev->addr_len))
757 EXPORT_SYMBOL(dev_getbyhwaddr);
759 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
761 struct net_device *dev;
764 for_each_netdev(net, dev)
765 if (dev->type == type)
770 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
772 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
774 struct net_device *dev;
777 dev = __dev_getfirstbyhwtype(net, type);
783 EXPORT_SYMBOL(dev_getfirstbyhwtype);
786 * dev_get_by_flags - find any device with given flags
787 * @net: the applicable net namespace
788 * @if_flags: IFF_* values
789 * @mask: bitmask of bits in if_flags to check
791 * Search for any interface with the given flags. Returns NULL if a device
792 * is not found or a pointer to the device. The device returned has
793 * had a reference added and the pointer is safe until the user calls
794 * dev_put to indicate they have finished with it.
797 struct net_device *dev_get_by_flags(struct net *net, unsigned short if_flags,
800 struct net_device *dev, *ret;
804 for_each_netdev_rcu(net, dev) {
805 if (((dev->flags ^ if_flags) & mask) == 0) {
814 EXPORT_SYMBOL(dev_get_by_flags);
817 * dev_valid_name - check if name is okay for network device
820 * Network device names need to be valid file names to
821 * to allow sysfs to work. We also disallow any kind of
824 int dev_valid_name(const char *name)
828 if (strlen(name) >= IFNAMSIZ)
830 if (!strcmp(name, ".") || !strcmp(name, ".."))
834 if (*name == '/' || isspace(*name))
840 EXPORT_SYMBOL(dev_valid_name);
843 * __dev_alloc_name - allocate a name for a device
844 * @net: network namespace to allocate the device name in
845 * @name: name format string
846 * @buf: scratch buffer and result name string
848 * Passed a format string - eg "lt%d" it will try and find a suitable
849 * id. It scans list of devices to build up a free map, then chooses
850 * the first empty slot. The caller must hold the dev_base or rtnl lock
851 * while allocating the name and adding the device in order to avoid
853 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
854 * Returns the number of the unit assigned or a negative errno code.
857 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
861 const int max_netdevices = 8*PAGE_SIZE;
862 unsigned long *inuse;
863 struct net_device *d;
865 p = strnchr(name, IFNAMSIZ-1, '%');
868 * Verify the string as this thing may have come from
869 * the user. There must be either one "%d" and no other "%"
872 if (p[1] != 'd' || strchr(p + 2, '%'))
875 /* Use one page as a bit array of possible slots */
876 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
880 for_each_netdev(net, d) {
881 if (!sscanf(d->name, name, &i))
883 if (i < 0 || i >= max_netdevices)
886 /* avoid cases where sscanf is not exact inverse of printf */
887 snprintf(buf, IFNAMSIZ, name, i);
888 if (!strncmp(buf, d->name, IFNAMSIZ))
892 i = find_first_zero_bit(inuse, max_netdevices);
893 free_page((unsigned long) inuse);
896 snprintf(buf, IFNAMSIZ, name, i);
897 if (!__dev_get_by_name(net, buf))
900 /* It is possible to run out of possible slots
901 * when the name is long and there isn't enough space left
902 * for the digits, or if all bits are used.
908 * dev_alloc_name - allocate a name for a device
910 * @name: name format string
912 * Passed a format string - eg "lt%d" it will try and find a suitable
913 * id. It scans list of devices to build up a free map, then chooses
914 * the first empty slot. The caller must hold the dev_base or rtnl lock
915 * while allocating the name and adding the device in order to avoid
917 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
918 * Returns the number of the unit assigned or a negative errno code.
921 int dev_alloc_name(struct net_device *dev, const char *name)
927 BUG_ON(!dev_net(dev));
929 ret = __dev_alloc_name(net, name, buf);
931 strlcpy(dev->name, buf, IFNAMSIZ);
934 EXPORT_SYMBOL(dev_alloc_name);
938 * dev_change_name - change name of a device
940 * @newname: name (or format string) must be at least IFNAMSIZ
942 * Change name of a device, can pass format strings "eth%d".
945 int dev_change_name(struct net_device *dev, const char *newname)
947 char oldname[IFNAMSIZ];
953 BUG_ON(!dev_net(dev));
956 if (dev->flags & IFF_UP)
959 if (!dev_valid_name(newname))
962 if (strncmp(newname, dev->name, IFNAMSIZ) == 0)
965 memcpy(oldname, dev->name, IFNAMSIZ);
967 if (strchr(newname, '%')) {
968 err = dev_alloc_name(dev, newname);
971 } else if (__dev_get_by_name(net, newname))
974 strlcpy(dev->name, newname, IFNAMSIZ);
977 /* For now only devices in the initial network namespace
980 if (net == &init_net) {
981 ret = device_rename(&dev->dev, dev->name);
983 memcpy(dev->name, oldname, IFNAMSIZ);
988 write_lock_bh(&dev_base_lock);
989 hlist_del(&dev->name_hlist);
990 write_unlock_bh(&dev_base_lock);
994 write_lock_bh(&dev_base_lock);
995 hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
996 write_unlock_bh(&dev_base_lock);
998 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
999 ret = notifier_to_errno(ret);
1004 "%s: name change rollback failed: %d.\n",
1008 memcpy(dev->name, oldname, IFNAMSIZ);
1017 * dev_set_alias - change ifalias of a device
1019 * @alias: name up to IFALIASZ
1020 * @len: limit of bytes to copy from info
1022 * Set ifalias for a device,
1024 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1028 if (len >= IFALIASZ)
1033 kfree(dev->ifalias);
1034 dev->ifalias = NULL;
1039 dev->ifalias = krealloc(dev->ifalias, len + 1, GFP_KERNEL);
1043 strlcpy(dev->ifalias, alias, len+1);
1049 * netdev_features_change - device changes features
1050 * @dev: device to cause notification
1052 * Called to indicate a device has changed features.
1054 void netdev_features_change(struct net_device *dev)
1056 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1058 EXPORT_SYMBOL(netdev_features_change);
1061 * netdev_state_change - device changes state
1062 * @dev: device to cause notification
1064 * Called to indicate a device has changed state. This function calls
1065 * the notifier chains for netdev_chain and sends a NEWLINK message
1066 * to the routing socket.
1068 void netdev_state_change(struct net_device *dev)
1070 if (dev->flags & IFF_UP) {
1071 call_netdevice_notifiers(NETDEV_CHANGE, dev);
1072 rtmsg_ifinfo(RTM_NEWLINK, dev, 0);
1075 EXPORT_SYMBOL(netdev_state_change);
1077 void netdev_bonding_change(struct net_device *dev, unsigned long event)
1079 call_netdevice_notifiers(event, dev);
1081 EXPORT_SYMBOL(netdev_bonding_change);
1084 * dev_load - load a network module
1085 * @net: the applicable net namespace
1086 * @name: name of interface
1088 * If a network interface is not present and the process has suitable
1089 * privileges this function loads the module. If module loading is not
1090 * available in this kernel then it becomes a nop.
1093 void dev_load(struct net *net, const char *name)
1095 struct net_device *dev;
1098 dev = dev_get_by_name_rcu(net, name);
1101 if (!dev && capable(CAP_NET_ADMIN))
1102 request_module("%s", name);
1104 EXPORT_SYMBOL(dev_load);
1107 * dev_open - prepare an interface for use.
1108 * @dev: device to open
1110 * Takes a device from down to up state. The device's private open
1111 * function is invoked and then the multicast lists are loaded. Finally
1112 * the device is moved into the up state and a %NETDEV_UP message is
1113 * sent to the netdev notifier chain.
1115 * Calling this function on an active interface is a nop. On a failure
1116 * a negative errno code is returned.
1118 int dev_open(struct net_device *dev)
1120 const struct net_device_ops *ops = dev->netdev_ops;
1129 if (dev->flags & IFF_UP)
1133 * Is it even present?
1135 if (!netif_device_present(dev))
1138 ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1139 ret = notifier_to_errno(ret);
1144 * Call device private open method
1146 set_bit(__LINK_STATE_START, &dev->state);
1148 if (ops->ndo_validate_addr)
1149 ret = ops->ndo_validate_addr(dev);
1151 if (!ret && ops->ndo_open)
1152 ret = ops->ndo_open(dev);
1155 * If it went open OK then:
1159 clear_bit(__LINK_STATE_START, &dev->state);
1164 dev->flags |= IFF_UP;
1169 net_dmaengine_get();
1172 * Initialize multicasting status
1174 dev_set_rx_mode(dev);
1177 * Wakeup transmit queue engine
1182 * ... and announce new interface.
1184 call_netdevice_notifiers(NETDEV_UP, dev);
1189 EXPORT_SYMBOL(dev_open);
1192 * dev_close - shutdown an interface.
1193 * @dev: device to shutdown
1195 * This function moves an active device into down state. A
1196 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1197 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1200 int dev_close(struct net_device *dev)
1202 const struct net_device_ops *ops = dev->netdev_ops;
1207 if (!(dev->flags & IFF_UP))
1211 * Tell people we are going down, so that they can
1212 * prepare to death, when device is still operating.
1214 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1216 clear_bit(__LINK_STATE_START, &dev->state);
1218 /* Synchronize to scheduled poll. We cannot touch poll list,
1219 * it can be even on different cpu. So just clear netif_running().
1221 * dev->stop() will invoke napi_disable() on all of it's
1222 * napi_struct instances on this device.
1224 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1226 dev_deactivate(dev);
1229 * Call the device specific close. This cannot fail.
1230 * Only if device is UP
1232 * We allow it to be called even after a DETACH hot-plug
1239 * Device is now down.
1242 dev->flags &= ~IFF_UP;
1245 * Tell people we are down
1247 call_netdevice_notifiers(NETDEV_DOWN, dev);
1252 net_dmaengine_put();
1256 EXPORT_SYMBOL(dev_close);
1260 * dev_disable_lro - disable Large Receive Offload on a device
1263 * Disable Large Receive Offload (LRO) on a net device. Must be
1264 * called under RTNL. This is needed if received packets may be
1265 * forwarded to another interface.
1267 void dev_disable_lro(struct net_device *dev)
1269 if (dev->ethtool_ops && dev->ethtool_ops->get_flags &&
1270 dev->ethtool_ops->set_flags) {
1271 u32 flags = dev->ethtool_ops->get_flags(dev);
1272 if (flags & ETH_FLAG_LRO) {
1273 flags &= ~ETH_FLAG_LRO;
1274 dev->ethtool_ops->set_flags(dev, flags);
1277 WARN_ON(dev->features & NETIF_F_LRO);
1279 EXPORT_SYMBOL(dev_disable_lro);
1282 static int dev_boot_phase = 1;
1285 * Device change register/unregister. These are not inline or static
1286 * as we export them to the world.
1290 * register_netdevice_notifier - register a network notifier block
1293 * Register a notifier to be called when network device events occur.
1294 * The notifier passed is linked into the kernel structures and must
1295 * not be reused until it has been unregistered. A negative errno code
1296 * is returned on a failure.
1298 * When registered all registration and up events are replayed
1299 * to the new notifier to allow device to have a race free
1300 * view of the network device list.
1303 int register_netdevice_notifier(struct notifier_block *nb)
1305 struct net_device *dev;
1306 struct net_device *last;
1311 err = raw_notifier_chain_register(&netdev_chain, nb);
1317 for_each_netdev(net, dev) {
1318 err = nb->notifier_call(nb, NETDEV_REGISTER, dev);
1319 err = notifier_to_errno(err);
1323 if (!(dev->flags & IFF_UP))
1326 nb->notifier_call(nb, NETDEV_UP, dev);
1337 for_each_netdev(net, dev) {
1341 if (dev->flags & IFF_UP) {
1342 nb->notifier_call(nb, NETDEV_GOING_DOWN, dev);
1343 nb->notifier_call(nb, NETDEV_DOWN, dev);
1345 nb->notifier_call(nb, NETDEV_UNREGISTER, dev);
1349 raw_notifier_chain_unregister(&netdev_chain, nb);
1352 EXPORT_SYMBOL(register_netdevice_notifier);
1355 * unregister_netdevice_notifier - unregister a network notifier block
1358 * Unregister a notifier previously registered by
1359 * register_netdevice_notifier(). The notifier is unlinked into the
1360 * kernel structures and may then be reused. A negative errno code
1361 * is returned on a failure.
1364 int unregister_netdevice_notifier(struct notifier_block *nb)
1369 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1373 EXPORT_SYMBOL(unregister_netdevice_notifier);
1376 * call_netdevice_notifiers - call all network notifier blocks
1377 * @val: value passed unmodified to notifier function
1378 * @dev: net_device pointer passed unmodified to notifier function
1380 * Call all network notifier blocks. Parameters and return value
1381 * are as for raw_notifier_call_chain().
1384 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1386 return raw_notifier_call_chain(&netdev_chain, val, dev);
1389 /* When > 0 there are consumers of rx skb time stamps */
1390 static atomic_t netstamp_needed = ATOMIC_INIT(0);
1392 void net_enable_timestamp(void)
1394 atomic_inc(&netstamp_needed);
1396 EXPORT_SYMBOL(net_enable_timestamp);
1398 void net_disable_timestamp(void)
1400 atomic_dec(&netstamp_needed);
1402 EXPORT_SYMBOL(net_disable_timestamp);
1404 static inline void net_timestamp(struct sk_buff *skb)
1406 if (atomic_read(&netstamp_needed))
1407 __net_timestamp(skb);
1409 skb->tstamp.tv64 = 0;
1413 * Support routine. Sends outgoing frames to any network
1414 * taps currently in use.
1417 static void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1419 struct packet_type *ptype;
1421 #ifdef CONFIG_NET_CLS_ACT
1422 if (!(skb->tstamp.tv64 && (G_TC_FROM(skb->tc_verd) & AT_INGRESS)))
1429 list_for_each_entry_rcu(ptype, &ptype_all, list) {
1430 /* Never send packets back to the socket
1431 * they originated from - MvS (miquels@drinkel.ow.org)
1433 if ((ptype->dev == dev || !ptype->dev) &&
1434 (ptype->af_packet_priv == NULL ||
1435 (struct sock *)ptype->af_packet_priv != skb->sk)) {
1436 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
1440 /* skb->nh should be correctly
1441 set by sender, so that the second statement is
1442 just protection against buggy protocols.
1444 skb_reset_mac_header(skb2);
1446 if (skb_network_header(skb2) < skb2->data ||
1447 skb2->network_header > skb2->tail) {
1448 if (net_ratelimit())
1449 printk(KERN_CRIT "protocol %04x is "
1451 skb2->protocol, dev->name);
1452 skb_reset_network_header(skb2);
1455 skb2->transport_header = skb2->network_header;
1456 skb2->pkt_type = PACKET_OUTGOING;
1457 ptype->func(skb2, skb->dev, ptype, skb->dev);
1464 static inline void __netif_reschedule(struct Qdisc *q)
1466 struct softnet_data *sd;
1467 unsigned long flags;
1469 local_irq_save(flags);
1470 sd = &__get_cpu_var(softnet_data);
1471 q->next_sched = sd->output_queue;
1472 sd->output_queue = q;
1473 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1474 local_irq_restore(flags);
1477 void __netif_schedule(struct Qdisc *q)
1479 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
1480 __netif_reschedule(q);
1482 EXPORT_SYMBOL(__netif_schedule);
1484 void dev_kfree_skb_irq(struct sk_buff *skb)
1486 if (atomic_dec_and_test(&skb->users)) {
1487 struct softnet_data *sd;
1488 unsigned long flags;
1490 local_irq_save(flags);
1491 sd = &__get_cpu_var(softnet_data);
1492 skb->next = sd->completion_queue;
1493 sd->completion_queue = skb;
1494 raise_softirq_irqoff(NET_TX_SOFTIRQ);
1495 local_irq_restore(flags);
1498 EXPORT_SYMBOL(dev_kfree_skb_irq);
1500 void dev_kfree_skb_any(struct sk_buff *skb)
1502 if (in_irq() || irqs_disabled())
1503 dev_kfree_skb_irq(skb);
1507 EXPORT_SYMBOL(dev_kfree_skb_any);
1511 * netif_device_detach - mark device as removed
1512 * @dev: network device
1514 * Mark device as removed from system and therefore no longer available.
1516 void netif_device_detach(struct net_device *dev)
1518 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
1519 netif_running(dev)) {
1520 netif_tx_stop_all_queues(dev);
1523 EXPORT_SYMBOL(netif_device_detach);
1526 * netif_device_attach - mark device as attached
1527 * @dev: network device
1529 * Mark device as attached from system and restart if needed.
1531 void netif_device_attach(struct net_device *dev)
1533 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
1534 netif_running(dev)) {
1535 netif_tx_wake_all_queues(dev);
1536 __netdev_watchdog_up(dev);
1539 EXPORT_SYMBOL(netif_device_attach);
1541 static bool can_checksum_protocol(unsigned long features, __be16 protocol)
1543 return ((features & NETIF_F_GEN_CSUM) ||
1544 ((features & NETIF_F_IP_CSUM) &&
1545 protocol == htons(ETH_P_IP)) ||
1546 ((features & NETIF_F_IPV6_CSUM) &&
1547 protocol == htons(ETH_P_IPV6)) ||
1548 ((features & NETIF_F_FCOE_CRC) &&
1549 protocol == htons(ETH_P_FCOE)));
1552 static bool dev_can_checksum(struct net_device *dev, struct sk_buff *skb)
1554 if (can_checksum_protocol(dev->features, skb->protocol))
1557 if (skb->protocol == htons(ETH_P_8021Q)) {
1558 struct vlan_ethhdr *veh = (struct vlan_ethhdr *)skb->data;
1559 if (can_checksum_protocol(dev->features & dev->vlan_features,
1560 veh->h_vlan_encapsulated_proto))
1568 * Invalidate hardware checksum when packet is to be mangled, and
1569 * complete checksum manually on outgoing path.
1571 int skb_checksum_help(struct sk_buff *skb)
1574 int ret = 0, offset;
1576 if (skb->ip_summed == CHECKSUM_COMPLETE)
1577 goto out_set_summed;
1579 if (unlikely(skb_shinfo(skb)->gso_size)) {
1580 /* Let GSO fix up the checksum. */
1581 goto out_set_summed;
1584 offset = skb->csum_start - skb_headroom(skb);
1585 BUG_ON(offset >= skb_headlen(skb));
1586 csum = skb_checksum(skb, offset, skb->len - offset, 0);
1588 offset += skb->csum_offset;
1589 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
1591 if (skb_cloned(skb) &&
1592 !skb_clone_writable(skb, offset + sizeof(__sum16))) {
1593 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1598 *(__sum16 *)(skb->data + offset) = csum_fold(csum);
1600 skb->ip_summed = CHECKSUM_NONE;
1604 EXPORT_SYMBOL(skb_checksum_help);
1607 * skb_gso_segment - Perform segmentation on skb.
1608 * @skb: buffer to segment
1609 * @features: features for the output path (see dev->features)
1611 * This function segments the given skb and returns a list of segments.
1613 * It may return NULL if the skb requires no segmentation. This is
1614 * only possible when GSO is used for verifying header integrity.
1616 struct sk_buff *skb_gso_segment(struct sk_buff *skb, int features)
1618 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
1619 struct packet_type *ptype;
1620 __be16 type = skb->protocol;
1623 skb_reset_mac_header(skb);
1624 skb->mac_len = skb->network_header - skb->mac_header;
1625 __skb_pull(skb, skb->mac_len);
1627 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1628 struct net_device *dev = skb->dev;
1629 struct ethtool_drvinfo info = {};
1631 if (dev && dev->ethtool_ops && dev->ethtool_ops->get_drvinfo)
1632 dev->ethtool_ops->get_drvinfo(dev, &info);
1634 WARN(1, "%s: caps=(0x%lx, 0x%lx) len=%d data_len=%d "
1636 info.driver, dev ? dev->features : 0L,
1637 skb->sk ? skb->sk->sk_route_caps : 0L,
1638 skb->len, skb->data_len, skb->ip_summed);
1640 if (skb_header_cloned(skb) &&
1641 (err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))
1642 return ERR_PTR(err);
1646 list_for_each_entry_rcu(ptype,
1647 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
1648 if (ptype->type == type && !ptype->dev && ptype->gso_segment) {
1649 if (unlikely(skb->ip_summed != CHECKSUM_PARTIAL)) {
1650 err = ptype->gso_send_check(skb);
1651 segs = ERR_PTR(err);
1652 if (err || skb_gso_ok(skb, features))
1654 __skb_push(skb, (skb->data -
1655 skb_network_header(skb)));
1657 segs = ptype->gso_segment(skb, features);
1663 __skb_push(skb, skb->data - skb_mac_header(skb));
1667 EXPORT_SYMBOL(skb_gso_segment);
1669 /* Take action when hardware reception checksum errors are detected. */
1671 void netdev_rx_csum_fault(struct net_device *dev)
1673 if (net_ratelimit()) {
1674 printk(KERN_ERR "%s: hw csum failure.\n",
1675 dev ? dev->name : "<unknown>");
1679 EXPORT_SYMBOL(netdev_rx_csum_fault);
1682 /* Actually, we should eliminate this check as soon as we know, that:
1683 * 1. IOMMU is present and allows to map all the memory.
1684 * 2. No high memory really exists on this machine.
1687 static inline int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
1689 #ifdef CONFIG_HIGHMEM
1692 if (dev->features & NETIF_F_HIGHDMA)
1695 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1696 if (PageHighMem(skb_shinfo(skb)->frags[i].page))
1704 void (*destructor)(struct sk_buff *skb);
1707 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
1709 static void dev_gso_skb_destructor(struct sk_buff *skb)
1711 struct dev_gso_cb *cb;
1714 struct sk_buff *nskb = skb->next;
1716 skb->next = nskb->next;
1719 } while (skb->next);
1721 cb = DEV_GSO_CB(skb);
1723 cb->destructor(skb);
1727 * dev_gso_segment - Perform emulated hardware segmentation on skb.
1728 * @skb: buffer to segment
1730 * This function segments the given skb and stores the list of segments
1733 static int dev_gso_segment(struct sk_buff *skb)
1735 struct net_device *dev = skb->dev;
1736 struct sk_buff *segs;
1737 int features = dev->features & ~(illegal_highdma(dev, skb) ?
1740 segs = skb_gso_segment(skb, features);
1742 /* Verifying header integrity only. */
1747 return PTR_ERR(segs);
1750 DEV_GSO_CB(skb)->destructor = skb->destructor;
1751 skb->destructor = dev_gso_skb_destructor;
1756 int dev_hard_start_xmit(struct sk_buff *skb, struct net_device *dev,
1757 struct netdev_queue *txq)
1759 const struct net_device_ops *ops = dev->netdev_ops;
1760 int rc = NETDEV_TX_OK;
1762 if (likely(!skb->next)) {
1763 if (!list_empty(&ptype_all))
1764 dev_queue_xmit_nit(skb, dev);
1766 if (netif_needs_gso(dev, skb)) {
1767 if (unlikely(dev_gso_segment(skb)))
1774 * If device doesnt need skb->dst, release it right now while
1775 * its hot in this cpu cache
1777 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
1780 rc = ops->ndo_start_xmit(skb, dev);
1781 if (rc == NETDEV_TX_OK)
1782 txq_trans_update(txq);
1784 * TODO: if skb_orphan() was called by
1785 * dev->hard_start_xmit() (for example, the unmodified
1786 * igb driver does that; bnx2 doesn't), then
1787 * skb_tx_software_timestamp() will be unable to send
1788 * back the time stamp.
1790 * How can this be prevented? Always create another
1791 * reference to the socket before calling
1792 * dev->hard_start_xmit()? Prevent that skb_orphan()
1793 * does anything in dev->hard_start_xmit() by clearing
1794 * the skb destructor before the call and restoring it
1795 * afterwards, then doing the skb_orphan() ourselves?
1802 struct sk_buff *nskb = skb->next;
1804 skb->next = nskb->next;
1806 rc = ops->ndo_start_xmit(nskb, dev);
1807 if (unlikely(rc != NETDEV_TX_OK)) {
1808 if (rc & ~NETDEV_TX_MASK)
1809 goto out_kfree_gso_skb;
1810 nskb->next = skb->next;
1814 txq_trans_update(txq);
1815 if (unlikely(netif_tx_queue_stopped(txq) && skb->next))
1816 return NETDEV_TX_BUSY;
1817 } while (skb->next);
1820 if (likely(skb->next == NULL))
1821 skb->destructor = DEV_GSO_CB(skb)->destructor;
1827 static u32 skb_tx_hashrnd;
1829 u16 skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb)
1833 if (skb_rx_queue_recorded(skb)) {
1834 hash = skb_get_rx_queue(skb);
1835 while (unlikely(hash >= dev->real_num_tx_queues))
1836 hash -= dev->real_num_tx_queues;
1840 if (skb->sk && skb->sk->sk_hash)
1841 hash = skb->sk->sk_hash;
1843 hash = skb->protocol;
1845 hash = jhash_1word(hash, skb_tx_hashrnd);
1847 return (u16) (((u64) hash * dev->real_num_tx_queues) >> 32);
1849 EXPORT_SYMBOL(skb_tx_hash);
1851 static inline u16 dev_cap_txqueue(struct net_device *dev, u16 queue_index)
1853 if (unlikely(queue_index >= dev->real_num_tx_queues)) {
1854 if (net_ratelimit()) {
1855 WARN(1, "%s selects TX queue %d, but "
1856 "real number of TX queues is %d\n",
1857 dev->name, queue_index,
1858 dev->real_num_tx_queues);
1865 static struct netdev_queue *dev_pick_tx(struct net_device *dev,
1866 struct sk_buff *skb)
1869 struct sock *sk = skb->sk;
1871 if (sk_tx_queue_recorded(sk)) {
1872 queue_index = sk_tx_queue_get(sk);
1874 const struct net_device_ops *ops = dev->netdev_ops;
1876 if (ops->ndo_select_queue) {
1877 queue_index = ops->ndo_select_queue(dev, skb);
1878 queue_index = dev_cap_txqueue(dev, queue_index);
1881 if (dev->real_num_tx_queues > 1)
1882 queue_index = skb_tx_hash(dev, skb);
1884 if (sk && sk->sk_dst_cache)
1885 sk_tx_queue_set(sk, queue_index);
1889 skb_set_queue_mapping(skb, queue_index);
1890 return netdev_get_tx_queue(dev, queue_index);
1893 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
1894 struct net_device *dev,
1895 struct netdev_queue *txq)
1897 spinlock_t *root_lock = qdisc_lock(q);
1900 spin_lock(root_lock);
1901 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
1904 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
1905 !test_and_set_bit(__QDISC_STATE_RUNNING, &q->state)) {
1907 * This is a work-conserving queue; there are no old skbs
1908 * waiting to be sent out; and the qdisc is not running -
1909 * xmit the skb directly.
1911 __qdisc_update_bstats(q, skb->len);
1912 if (sch_direct_xmit(skb, q, dev, txq, root_lock))
1915 clear_bit(__QDISC_STATE_RUNNING, &q->state);
1917 rc = NET_XMIT_SUCCESS;
1919 rc = qdisc_enqueue_root(skb, q);
1922 spin_unlock(root_lock);
1928 * dev_queue_xmit - transmit a buffer
1929 * @skb: buffer to transmit
1931 * Queue a buffer for transmission to a network device. The caller must
1932 * have set the device and priority and built the buffer before calling
1933 * this function. The function can be called from an interrupt.
1935 * A negative errno code is returned on a failure. A success does not
1936 * guarantee the frame will be transmitted as it may be dropped due
1937 * to congestion or traffic shaping.
1939 * -----------------------------------------------------------------------------------
1940 * I notice this method can also return errors from the queue disciplines,
1941 * including NET_XMIT_DROP, which is a positive value. So, errors can also
1944 * Regardless of the return value, the skb is consumed, so it is currently
1945 * difficult to retry a send to this method. (You can bump the ref count
1946 * before sending to hold a reference for retry if you are careful.)
1948 * When calling this method, interrupts MUST be enabled. This is because
1949 * the BH enable code must have IRQs enabled so that it will not deadlock.
1952 int dev_queue_xmit(struct sk_buff *skb)
1954 struct net_device *dev = skb->dev;
1955 struct netdev_queue *txq;
1959 /* GSO will handle the following emulations directly. */
1960 if (netif_needs_gso(dev, skb))
1963 if (skb_has_frags(skb) &&
1964 !(dev->features & NETIF_F_FRAGLIST) &&
1965 __skb_linearize(skb))
1968 /* Fragmented skb is linearized if device does not support SG,
1969 * or if at least one of fragments is in highmem and device
1970 * does not support DMA from it.
1972 if (skb_shinfo(skb)->nr_frags &&
1973 (!(dev->features & NETIF_F_SG) || illegal_highdma(dev, skb)) &&
1974 __skb_linearize(skb))
1977 /* If packet is not checksummed and device does not support
1978 * checksumming for this protocol, complete checksumming here.
1980 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1981 skb_set_transport_header(skb, skb->csum_start -
1983 if (!dev_can_checksum(dev, skb) && skb_checksum_help(skb))
1988 /* Disable soft irqs for various locks below. Also
1989 * stops preemption for RCU.
1993 txq = dev_pick_tx(dev, skb);
1994 q = rcu_dereference(txq->qdisc);
1996 #ifdef CONFIG_NET_CLS_ACT
1997 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_EGRESS);
2000 rc = __dev_xmit_skb(skb, q, dev, txq);
2004 /* The device has no queue. Common case for software devices:
2005 loopback, all the sorts of tunnels...
2007 Really, it is unlikely that netif_tx_lock protection is necessary
2008 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2010 However, it is possible, that they rely on protection
2013 Check this and shot the lock. It is not prone from deadlocks.
2014 Either shot noqueue qdisc, it is even simpler 8)
2016 if (dev->flags & IFF_UP) {
2017 int cpu = smp_processor_id(); /* ok because BHs are off */
2019 if (txq->xmit_lock_owner != cpu) {
2021 HARD_TX_LOCK(dev, txq, cpu);
2023 if (!netif_tx_queue_stopped(txq)) {
2024 rc = dev_hard_start_xmit(skb, dev, txq);
2025 if (dev_xmit_complete(rc)) {
2026 HARD_TX_UNLOCK(dev, txq);
2030 HARD_TX_UNLOCK(dev, txq);
2031 if (net_ratelimit())
2032 printk(KERN_CRIT "Virtual device %s asks to "
2033 "queue packet!\n", dev->name);
2035 /* Recursion is detected! It is possible,
2037 if (net_ratelimit())
2038 printk(KERN_CRIT "Dead loop on virtual device "
2039 "%s, fix it urgently!\n", dev->name);
2044 rcu_read_unlock_bh();
2050 rcu_read_unlock_bh();
2053 EXPORT_SYMBOL(dev_queue_xmit);
2056 /*=======================================================================
2058 =======================================================================*/
2060 int netdev_max_backlog __read_mostly = 1000;
2061 int netdev_budget __read_mostly = 300;
2062 int weight_p __read_mostly = 64; /* old backlog weight */
2064 DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
2068 * netif_rx - post buffer to the network code
2069 * @skb: buffer to post
2071 * This function receives a packet from a device driver and queues it for
2072 * the upper (protocol) levels to process. It always succeeds. The buffer
2073 * may be dropped during processing for congestion control or by the
2077 * NET_RX_SUCCESS (no congestion)
2078 * NET_RX_DROP (packet was dropped)
2082 int netif_rx(struct sk_buff *skb)
2084 struct softnet_data *queue;
2085 unsigned long flags;
2087 /* if netpoll wants it, pretend we never saw it */
2088 if (netpoll_rx(skb))
2091 if (!skb->tstamp.tv64)
2095 * The code is rearranged so that the path is the most
2096 * short when CPU is congested, but is still operating.
2098 local_irq_save(flags);
2099 queue = &__get_cpu_var(softnet_data);
2101 __get_cpu_var(netdev_rx_stat).total++;
2102 if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
2103 if (queue->input_pkt_queue.qlen) {
2105 __skb_queue_tail(&queue->input_pkt_queue, skb);
2106 local_irq_restore(flags);
2107 return NET_RX_SUCCESS;
2110 napi_schedule(&queue->backlog);
2114 __get_cpu_var(netdev_rx_stat).dropped++;
2115 local_irq_restore(flags);
2120 EXPORT_SYMBOL(netif_rx);
2122 int netif_rx_ni(struct sk_buff *skb)
2127 err = netif_rx(skb);
2128 if (local_softirq_pending())
2134 EXPORT_SYMBOL(netif_rx_ni);
2136 static void net_tx_action(struct softirq_action *h)
2138 struct softnet_data *sd = &__get_cpu_var(softnet_data);
2140 if (sd->completion_queue) {
2141 struct sk_buff *clist;
2143 local_irq_disable();
2144 clist = sd->completion_queue;
2145 sd->completion_queue = NULL;
2149 struct sk_buff *skb = clist;
2150 clist = clist->next;
2152 WARN_ON(atomic_read(&skb->users));
2157 if (sd->output_queue) {
2160 local_irq_disable();
2161 head = sd->output_queue;
2162 sd->output_queue = NULL;
2166 struct Qdisc *q = head;
2167 spinlock_t *root_lock;
2169 head = head->next_sched;
2171 root_lock = qdisc_lock(q);
2172 if (spin_trylock(root_lock)) {
2173 smp_mb__before_clear_bit();
2174 clear_bit(__QDISC_STATE_SCHED,
2177 spin_unlock(root_lock);
2179 if (!test_bit(__QDISC_STATE_DEACTIVATED,
2181 __netif_reschedule(q);
2183 smp_mb__before_clear_bit();
2184 clear_bit(__QDISC_STATE_SCHED,
2192 static inline int deliver_skb(struct sk_buff *skb,
2193 struct packet_type *pt_prev,
2194 struct net_device *orig_dev)
2196 atomic_inc(&skb->users);
2197 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2200 #if defined(CONFIG_BRIDGE) || defined (CONFIG_BRIDGE_MODULE)
2202 #if defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE)
2203 /* This hook is defined here for ATM LANE */
2204 int (*br_fdb_test_addr_hook)(struct net_device *dev,
2205 unsigned char *addr) __read_mostly;
2206 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
2210 * If bridge module is loaded call bridging hook.
2211 * returns NULL if packet was consumed.
2213 struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,
2214 struct sk_buff *skb) __read_mostly;
2215 EXPORT_SYMBOL_GPL(br_handle_frame_hook);
2217 static inline struct sk_buff *handle_bridge(struct sk_buff *skb,
2218 struct packet_type **pt_prev, int *ret,
2219 struct net_device *orig_dev)
2221 struct net_bridge_port *port;
2223 if (skb->pkt_type == PACKET_LOOPBACK ||
2224 (port = rcu_dereference(skb->dev->br_port)) == NULL)
2228 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2232 return br_handle_frame_hook(port, skb);
2235 #define handle_bridge(skb, pt_prev, ret, orig_dev) (skb)
2238 #if defined(CONFIG_MACVLAN) || defined(CONFIG_MACVLAN_MODULE)
2239 struct sk_buff *(*macvlan_handle_frame_hook)(struct sk_buff *skb) __read_mostly;
2240 EXPORT_SYMBOL_GPL(macvlan_handle_frame_hook);
2242 static inline struct sk_buff *handle_macvlan(struct sk_buff *skb,
2243 struct packet_type **pt_prev,
2245 struct net_device *orig_dev)
2247 if (skb->dev->macvlan_port == NULL)
2251 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2254 return macvlan_handle_frame_hook(skb);
2257 #define handle_macvlan(skb, pt_prev, ret, orig_dev) (skb)
2260 #ifdef CONFIG_NET_CLS_ACT
2261 /* TODO: Maybe we should just force sch_ingress to be compiled in
2262 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
2263 * a compare and 2 stores extra right now if we dont have it on
2264 * but have CONFIG_NET_CLS_ACT
2265 * NOTE: This doesnt stop any functionality; if you dont have
2266 * the ingress scheduler, you just cant add policies on ingress.
2269 static int ing_filter(struct sk_buff *skb)
2271 struct net_device *dev = skb->dev;
2272 u32 ttl = G_TC_RTTL(skb->tc_verd);
2273 struct netdev_queue *rxq;
2274 int result = TC_ACT_OK;
2277 if (MAX_RED_LOOP < ttl++) {
2279 "Redir loop detected Dropping packet (%d->%d)\n",
2280 skb->iif, dev->ifindex);
2284 skb->tc_verd = SET_TC_RTTL(skb->tc_verd, ttl);
2285 skb->tc_verd = SET_TC_AT(skb->tc_verd, AT_INGRESS);
2287 rxq = &dev->rx_queue;
2290 if (q != &noop_qdisc) {
2291 spin_lock(qdisc_lock(q));
2292 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED, &q->state)))
2293 result = qdisc_enqueue_root(skb, q);
2294 spin_unlock(qdisc_lock(q));
2300 static inline struct sk_buff *handle_ing(struct sk_buff *skb,
2301 struct packet_type **pt_prev,
2302 int *ret, struct net_device *orig_dev)
2304 if (skb->dev->rx_queue.qdisc == &noop_qdisc)
2308 *ret = deliver_skb(skb, *pt_prev, orig_dev);
2311 /* Huh? Why does turning on AF_PACKET affect this? */
2312 skb->tc_verd = SET_TC_OK2MUNGE(skb->tc_verd);
2315 switch (ing_filter(skb)) {
2329 * netif_nit_deliver - deliver received packets to network taps
2332 * This function is used to deliver incoming packets to network
2333 * taps. It should be used when the normal netif_receive_skb path
2334 * is bypassed, for example because of VLAN acceleration.
2336 void netif_nit_deliver(struct sk_buff *skb)
2338 struct packet_type *ptype;
2340 if (list_empty(&ptype_all))
2343 skb_reset_network_header(skb);
2344 skb_reset_transport_header(skb);
2345 skb->mac_len = skb->network_header - skb->mac_header;
2348 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2349 if (!ptype->dev || ptype->dev == skb->dev)
2350 deliver_skb(skb, ptype, skb->dev);
2356 * netif_receive_skb - process receive buffer from network
2357 * @skb: buffer to process
2359 * netif_receive_skb() is the main receive data processing function.
2360 * It always succeeds. The buffer may be dropped during processing
2361 * for congestion control or by the protocol layers.
2363 * This function may only be called from softirq context and interrupts
2364 * should be enabled.
2366 * Return values (usually ignored):
2367 * NET_RX_SUCCESS: no congestion
2368 * NET_RX_DROP: packet was dropped
2370 int netif_receive_skb(struct sk_buff *skb)
2372 struct packet_type *ptype, *pt_prev;
2373 struct net_device *orig_dev;
2374 struct net_device *null_or_orig;
2375 int ret = NET_RX_DROP;
2378 if (!skb->tstamp.tv64)
2381 if (vlan_tx_tag_present(skb) && vlan_hwaccel_do_receive(skb))
2382 return NET_RX_SUCCESS;
2384 /* if we've gotten here through NAPI, check netpoll */
2385 if (netpoll_receive_skb(skb))
2389 skb->iif = skb->dev->ifindex;
2391 null_or_orig = NULL;
2392 orig_dev = skb->dev;
2393 if (orig_dev->master) {
2394 if (skb_bond_should_drop(skb))
2395 null_or_orig = orig_dev; /* deliver only exact match */
2397 skb->dev = orig_dev->master;
2400 __get_cpu_var(netdev_rx_stat).total++;
2402 skb_reset_network_header(skb);
2403 skb_reset_transport_header(skb);
2404 skb->mac_len = skb->network_header - skb->mac_header;
2410 #ifdef CONFIG_NET_CLS_ACT
2411 if (skb->tc_verd & TC_NCLS) {
2412 skb->tc_verd = CLR_TC_NCLS(skb->tc_verd);
2417 list_for_each_entry_rcu(ptype, &ptype_all, list) {
2418 if (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2419 ptype->dev == orig_dev) {
2421 ret = deliver_skb(skb, pt_prev, orig_dev);
2426 #ifdef CONFIG_NET_CLS_ACT
2427 skb = handle_ing(skb, &pt_prev, &ret, orig_dev);
2433 skb = handle_bridge(skb, &pt_prev, &ret, orig_dev);
2436 skb = handle_macvlan(skb, &pt_prev, &ret, orig_dev);
2440 type = skb->protocol;
2441 list_for_each_entry_rcu(ptype,
2442 &ptype_base[ntohs(type) & PTYPE_HASH_MASK], list) {
2443 if (ptype->type == type &&
2444 (ptype->dev == null_or_orig || ptype->dev == skb->dev ||
2445 ptype->dev == orig_dev)) {
2447 ret = deliver_skb(skb, pt_prev, orig_dev);
2453 ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2456 /* Jamal, now you will not able to escape explaining
2457 * me how you were going to use this. :-)
2466 EXPORT_SYMBOL(netif_receive_skb);
2468 /* Network device is going away, flush any packets still pending */
2469 static void flush_backlog(void *arg)
2471 struct net_device *dev = arg;
2472 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2473 struct sk_buff *skb, *tmp;
2475 skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
2476 if (skb->dev == dev) {
2477 __skb_unlink(skb, &queue->input_pkt_queue);
2482 static int napi_gro_complete(struct sk_buff *skb)
2484 struct packet_type *ptype;
2485 __be16 type = skb->protocol;
2486 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2489 if (NAPI_GRO_CB(skb)->count == 1) {
2490 skb_shinfo(skb)->gso_size = 0;
2495 list_for_each_entry_rcu(ptype, head, list) {
2496 if (ptype->type != type || ptype->dev || !ptype->gro_complete)
2499 err = ptype->gro_complete(skb);
2505 WARN_ON(&ptype->list == head);
2507 return NET_RX_SUCCESS;
2511 return netif_receive_skb(skb);
2514 void napi_gro_flush(struct napi_struct *napi)
2516 struct sk_buff *skb, *next;
2518 for (skb = napi->gro_list; skb; skb = next) {
2521 napi_gro_complete(skb);
2524 napi->gro_count = 0;
2525 napi->gro_list = NULL;
2527 EXPORT_SYMBOL(napi_gro_flush);
2529 enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2531 struct sk_buff **pp = NULL;
2532 struct packet_type *ptype;
2533 __be16 type = skb->protocol;
2534 struct list_head *head = &ptype_base[ntohs(type) & PTYPE_HASH_MASK];
2537 enum gro_result ret;
2539 if (!(skb->dev->features & NETIF_F_GRO))
2542 if (skb_is_gso(skb) || skb_has_frags(skb))
2546 list_for_each_entry_rcu(ptype, head, list) {
2547 if (ptype->type != type || ptype->dev || !ptype->gro_receive)
2550 skb_set_network_header(skb, skb_gro_offset(skb));
2551 mac_len = skb->network_header - skb->mac_header;
2552 skb->mac_len = mac_len;
2553 NAPI_GRO_CB(skb)->same_flow = 0;
2554 NAPI_GRO_CB(skb)->flush = 0;
2555 NAPI_GRO_CB(skb)->free = 0;
2557 pp = ptype->gro_receive(&napi->gro_list, skb);
2562 if (&ptype->list == head)
2565 same_flow = NAPI_GRO_CB(skb)->same_flow;
2566 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
2569 struct sk_buff *nskb = *pp;
2573 napi_gro_complete(nskb);
2580 if (NAPI_GRO_CB(skb)->flush || napi->gro_count >= MAX_GRO_SKBS)
2584 NAPI_GRO_CB(skb)->count = 1;
2585 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
2586 skb->next = napi->gro_list;
2587 napi->gro_list = skb;
2591 if (skb_headlen(skb) < skb_gro_offset(skb)) {
2592 int grow = skb_gro_offset(skb) - skb_headlen(skb);
2594 BUG_ON(skb->end - skb->tail < grow);
2596 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
2599 skb->data_len -= grow;
2601 skb_shinfo(skb)->frags[0].page_offset += grow;
2602 skb_shinfo(skb)->frags[0].size -= grow;
2604 if (unlikely(!skb_shinfo(skb)->frags[0].size)) {
2605 put_page(skb_shinfo(skb)->frags[0].page);
2606 memmove(skb_shinfo(skb)->frags,
2607 skb_shinfo(skb)->frags + 1,
2608 --skb_shinfo(skb)->nr_frags);
2619 EXPORT_SYMBOL(dev_gro_receive);
2622 __napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2626 if (netpoll_rx_on(skb))
2629 for (p = napi->gro_list; p; p = p->next) {
2630 NAPI_GRO_CB(p)->same_flow = (p->dev == skb->dev)
2631 && !compare_ether_header(skb_mac_header(p),
2632 skb_gro_mac_header(skb));
2633 NAPI_GRO_CB(p)->flush = 0;
2636 return dev_gro_receive(napi, skb);
2639 gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
2643 if (netif_receive_skb(skb))
2648 case GRO_MERGED_FREE:
2659 EXPORT_SYMBOL(napi_skb_finish);
2661 void skb_gro_reset_offset(struct sk_buff *skb)
2663 NAPI_GRO_CB(skb)->data_offset = 0;
2664 NAPI_GRO_CB(skb)->frag0 = NULL;
2665 NAPI_GRO_CB(skb)->frag0_len = 0;
2667 if (skb->mac_header == skb->tail &&
2668 !PageHighMem(skb_shinfo(skb)->frags[0].page)) {
2669 NAPI_GRO_CB(skb)->frag0 =
2670 page_address(skb_shinfo(skb)->frags[0].page) +
2671 skb_shinfo(skb)->frags[0].page_offset;
2672 NAPI_GRO_CB(skb)->frag0_len = skb_shinfo(skb)->frags[0].size;
2675 EXPORT_SYMBOL(skb_gro_reset_offset);
2677 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
2679 skb_gro_reset_offset(skb);
2681 return napi_skb_finish(__napi_gro_receive(napi, skb), skb);
2683 EXPORT_SYMBOL(napi_gro_receive);
2685 void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
2687 __skb_pull(skb, skb_headlen(skb));
2688 skb_reserve(skb, NET_IP_ALIGN - skb_headroom(skb));
2692 EXPORT_SYMBOL(napi_reuse_skb);
2694 struct sk_buff *napi_get_frags(struct napi_struct *napi)
2696 struct sk_buff *skb = napi->skb;
2699 skb = netdev_alloc_skb_ip_align(napi->dev, GRO_MAX_HEAD);
2705 EXPORT_SYMBOL(napi_get_frags);
2707 gro_result_t napi_frags_finish(struct napi_struct *napi, struct sk_buff *skb,
2713 skb->protocol = eth_type_trans(skb, napi->dev);
2715 if (ret == GRO_HELD)
2716 skb_gro_pull(skb, -ETH_HLEN);
2717 else if (netif_receive_skb(skb))
2722 case GRO_MERGED_FREE:
2723 napi_reuse_skb(napi, skb);
2732 EXPORT_SYMBOL(napi_frags_finish);
2734 struct sk_buff *napi_frags_skb(struct napi_struct *napi)
2736 struct sk_buff *skb = napi->skb;
2743 skb_reset_mac_header(skb);
2744 skb_gro_reset_offset(skb);
2746 off = skb_gro_offset(skb);
2747 hlen = off + sizeof(*eth);
2748 eth = skb_gro_header_fast(skb, off);
2749 if (skb_gro_header_hard(skb, hlen)) {
2750 eth = skb_gro_header_slow(skb, hlen, off);
2751 if (unlikely(!eth)) {
2752 napi_reuse_skb(napi, skb);
2758 skb_gro_pull(skb, sizeof(*eth));
2761 * This works because the only protocols we care about don't require
2762 * special handling. We'll fix it up properly at the end.
2764 skb->protocol = eth->h_proto;
2769 EXPORT_SYMBOL(napi_frags_skb);
2771 gro_result_t napi_gro_frags(struct napi_struct *napi)
2773 struct sk_buff *skb = napi_frags_skb(napi);
2778 return napi_frags_finish(napi, skb, __napi_gro_receive(napi, skb));
2780 EXPORT_SYMBOL(napi_gro_frags);
2782 static int process_backlog(struct napi_struct *napi, int quota)
2785 struct softnet_data *queue = &__get_cpu_var(softnet_data);
2786 unsigned long start_time = jiffies;
2788 napi->weight = weight_p;
2790 struct sk_buff *skb;
2792 local_irq_disable();
2793 skb = __skb_dequeue(&queue->input_pkt_queue);
2795 __napi_complete(napi);
2801 netif_receive_skb(skb);
2802 } while (++work < quota && jiffies == start_time);
2808 * __napi_schedule - schedule for receive
2809 * @n: entry to schedule
2811 * The entry's receive function will be scheduled to run
2813 void __napi_schedule(struct napi_struct *n)
2815 unsigned long flags;
2817 local_irq_save(flags);
2818 list_add_tail(&n->poll_list, &__get_cpu_var(softnet_data).poll_list);
2819 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2820 local_irq_restore(flags);
2822 EXPORT_SYMBOL(__napi_schedule);
2824 void __napi_complete(struct napi_struct *n)
2826 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
2827 BUG_ON(n->gro_list);
2829 list_del(&n->poll_list);
2830 smp_mb__before_clear_bit();
2831 clear_bit(NAPI_STATE_SCHED, &n->state);
2833 EXPORT_SYMBOL(__napi_complete);
2835 void napi_complete(struct napi_struct *n)
2837 unsigned long flags;
2840 * don't let napi dequeue from the cpu poll list
2841 * just in case its running on a different cpu
2843 if (unlikely(test_bit(NAPI_STATE_NPSVC, &n->state)))
2847 local_irq_save(flags);
2849 local_irq_restore(flags);
2851 EXPORT_SYMBOL(napi_complete);
2853 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
2854 int (*poll)(struct napi_struct *, int), int weight)
2856 INIT_LIST_HEAD(&napi->poll_list);
2857 napi->gro_count = 0;
2858 napi->gro_list = NULL;
2861 napi->weight = weight;
2862 list_add(&napi->dev_list, &dev->napi_list);
2864 #ifdef CONFIG_NETPOLL
2865 spin_lock_init(&napi->poll_lock);
2866 napi->poll_owner = -1;
2868 set_bit(NAPI_STATE_SCHED, &napi->state);
2870 EXPORT_SYMBOL(netif_napi_add);
2872 void netif_napi_del(struct napi_struct *napi)
2874 struct sk_buff *skb, *next;
2876 list_del_init(&napi->dev_list);
2877 napi_free_frags(napi);
2879 for (skb = napi->gro_list; skb; skb = next) {
2885 napi->gro_list = NULL;
2886 napi->gro_count = 0;
2888 EXPORT_SYMBOL(netif_napi_del);
2891 static void net_rx_action(struct softirq_action *h)
2893 struct list_head *list = &__get_cpu_var(softnet_data).poll_list;
2894 unsigned long time_limit = jiffies + 2;
2895 int budget = netdev_budget;
2898 local_irq_disable();
2900 while (!list_empty(list)) {
2901 struct napi_struct *n;
2904 /* If softirq window is exhuasted then punt.
2905 * Allow this to run for 2 jiffies since which will allow
2906 * an average latency of 1.5/HZ.
2908 if (unlikely(budget <= 0 || time_after(jiffies, time_limit)))
2913 /* Even though interrupts have been re-enabled, this
2914 * access is safe because interrupts can only add new
2915 * entries to the tail of this list, and only ->poll()
2916 * calls can remove this head entry from the list.
2918 n = list_entry(list->next, struct napi_struct, poll_list);
2920 have = netpoll_poll_lock(n);
2924 /* This NAPI_STATE_SCHED test is for avoiding a race
2925 * with netpoll's poll_napi(). Only the entity which
2926 * obtains the lock and sees NAPI_STATE_SCHED set will
2927 * actually make the ->poll() call. Therefore we avoid
2928 * accidently calling ->poll() when NAPI is not scheduled.
2931 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
2932 work = n->poll(n, weight);
2936 WARN_ON_ONCE(work > weight);
2940 local_irq_disable();
2942 /* Drivers must not modify the NAPI state if they
2943 * consume the entire weight. In such cases this code
2944 * still "owns" the NAPI instance and therefore can
2945 * move the instance around on the list at-will.
2947 if (unlikely(work == weight)) {
2948 if (unlikely(napi_disable_pending(n))) {
2951 local_irq_disable();
2953 list_move_tail(&n->poll_list, list);
2956 netpoll_poll_unlock(have);
2961 #ifdef CONFIG_NET_DMA
2963 * There may not be any more sk_buffs coming right now, so push
2964 * any pending DMA copies to hardware
2966 dma_issue_pending_all();
2972 __get_cpu_var(netdev_rx_stat).time_squeeze++;
2973 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
2977 static gifconf_func_t *gifconf_list[NPROTO];
2980 * register_gifconf - register a SIOCGIF handler
2981 * @family: Address family
2982 * @gifconf: Function handler
2984 * Register protocol dependent address dumping routines. The handler
2985 * that is passed must not be freed or reused until it has been replaced
2986 * by another handler.
2988 int register_gifconf(unsigned int family, gifconf_func_t *gifconf)
2990 if (family >= NPROTO)
2992 gifconf_list[family] = gifconf;
2995 EXPORT_SYMBOL(register_gifconf);
2999 * Map an interface index to its name (SIOCGIFNAME)
3003 * We need this ioctl for efficient implementation of the
3004 * if_indextoname() function required by the IPv6 API. Without
3005 * it, we would have to search all the interfaces to find a
3009 static int dev_ifname(struct net *net, struct ifreq __user *arg)
3011 struct net_device *dev;
3015 * Fetch the caller's info block.
3018 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
3022 dev = dev_get_by_index_rcu(net, ifr.ifr_ifindex);
3028 strcpy(ifr.ifr_name, dev->name);
3031 if (copy_to_user(arg, &ifr, sizeof(struct ifreq)))
3037 * Perform a SIOCGIFCONF call. This structure will change
3038 * size eventually, and there is nothing I can do about it.
3039 * Thus we will need a 'compatibility mode'.
3042 static int dev_ifconf(struct net *net, char __user *arg)
3045 struct net_device *dev;
3052 * Fetch the caller's info block.
3055 if (copy_from_user(&ifc, arg, sizeof(struct ifconf)))
3062 * Loop over the interfaces, and write an info block for each.
3066 for_each_netdev(net, dev) {
3067 for (i = 0; i < NPROTO; i++) {
3068 if (gifconf_list[i]) {
3071 done = gifconf_list[i](dev, NULL, 0);
3073 done = gifconf_list[i](dev, pos + total,
3083 * All done. Write the updated control block back to the caller.
3085 ifc.ifc_len = total;
3088 * Both BSD and Solaris return 0 here, so we do too.
3090 return copy_to_user(arg, &ifc, sizeof(struct ifconf)) ? -EFAULT : 0;
3093 #ifdef CONFIG_PROC_FS
3095 * This is invoked by the /proc filesystem handler to display a device
3098 void *dev_seq_start(struct seq_file *seq, loff_t *pos)
3101 struct net *net = seq_file_net(seq);
3103 struct net_device *dev;
3107 return SEQ_START_TOKEN;
3110 for_each_netdev_rcu(net, dev)
3117 void *dev_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3119 struct net_device *dev = (v == SEQ_START_TOKEN) ?
3120 first_net_device(seq_file_net(seq)) :
3121 next_net_device((struct net_device *)v);
3124 return rcu_dereference(dev);
3127 void dev_seq_stop(struct seq_file *seq, void *v)
3133 static void dev_seq_printf_stats(struct seq_file *seq, struct net_device *dev)
3135 const struct net_device_stats *stats = dev_get_stats(dev);
3137 seq_printf(seq, "%6s:%8lu %7lu %4lu %4lu %4lu %5lu %10lu %9lu "
3138 "%8lu %7lu %4lu %4lu %4lu %5lu %7lu %10lu\n",
3139 dev->name, stats->rx_bytes, stats->rx_packets,
3141 stats->rx_dropped + stats->rx_missed_errors,
3142 stats->rx_fifo_errors,
3143 stats->rx_length_errors + stats->rx_over_errors +
3144 stats->rx_crc_errors + stats->rx_frame_errors,
3145 stats->rx_compressed, stats->multicast,
3146 stats->tx_bytes, stats->tx_packets,
3147 stats->tx_errors, stats->tx_dropped,
3148 stats->tx_fifo_errors, stats->collisions,
3149 stats->tx_carrier_errors +
3150 stats->tx_aborted_errors +
3151 stats->tx_window_errors +
3152 stats->tx_heartbeat_errors,
3153 stats->tx_compressed);
3157 * Called from the PROCfs module. This now uses the new arbitrary sized
3158 * /proc/net interface to create /proc/net/dev
3160 static int dev_seq_show(struct seq_file *seq, void *v)
3162 if (v == SEQ_START_TOKEN)
3163 seq_puts(seq, "Inter-| Receive "
3165 " face |bytes packets errs drop fifo frame "
3166 "compressed multicast|bytes packets errs "
3167 "drop fifo colls carrier compressed\n");
3169 dev_seq_printf_stats(seq, v);
3173 static struct netif_rx_stats *softnet_get_online(loff_t *pos)
3175 struct netif_rx_stats *rc = NULL;
3177 while (*pos < nr_cpu_ids)
3178 if (cpu_online(*pos)) {
3179 rc = &per_cpu(netdev_rx_stat, *pos);
3186 static void *softnet_seq_start(struct seq_file *seq, loff_t *pos)
3188 return softnet_get_online(pos);
3191 static void *softnet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3194 return softnet_get_online(pos);
3197 static void softnet_seq_stop(struct seq_file *seq, void *v)
3201 static int softnet_seq_show(struct seq_file *seq, void *v)
3203 struct netif_rx_stats *s = v;
3205 seq_printf(seq, "%08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
3206 s->total, s->dropped, s->time_squeeze, 0,
3207 0, 0, 0, 0, /* was fastroute */
3212 static const struct seq_operations dev_seq_ops = {
3213 .start = dev_seq_start,
3214 .next = dev_seq_next,
3215 .stop = dev_seq_stop,
3216 .show = dev_seq_show,
3219 static int dev_seq_open(struct inode *inode, struct file *file)
3221 return seq_open_net(inode, file, &dev_seq_ops,
3222 sizeof(struct seq_net_private));
3225 static const struct file_operations dev_seq_fops = {
3226 .owner = THIS_MODULE,
3227 .open = dev_seq_open,
3229 .llseek = seq_lseek,
3230 .release = seq_release_net,
3233 static const struct seq_operations softnet_seq_ops = {
3234 .start = softnet_seq_start,
3235 .next = softnet_seq_next,
3236 .stop = softnet_seq_stop,
3237 .show = softnet_seq_show,
3240 static int softnet_seq_open(struct inode *inode, struct file *file)
3242 return seq_open(file, &softnet_seq_ops);
3245 static const struct file_operations softnet_seq_fops = {
3246 .owner = THIS_MODULE,
3247 .open = softnet_seq_open,
3249 .llseek = seq_lseek,
3250 .release = seq_release,
3253 static void *ptype_get_idx(loff_t pos)
3255 struct packet_type *pt = NULL;
3259 list_for_each_entry_rcu(pt, &ptype_all, list) {
3265 for (t = 0; t < PTYPE_HASH_SIZE; t++) {
3266 list_for_each_entry_rcu(pt, &ptype_base[t], list) {
3275 static void *ptype_seq_start(struct seq_file *seq, loff_t *pos)
3279 return *pos ? ptype_get_idx(*pos - 1) : SEQ_START_TOKEN;
3282 static void *ptype_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3284 struct packet_type *pt;
3285 struct list_head *nxt;
3289 if (v == SEQ_START_TOKEN)
3290 return ptype_get_idx(0);
3293 nxt = pt->list.next;
3294 if (pt->type == htons(ETH_P_ALL)) {
3295 if (nxt != &ptype_all)
3298 nxt = ptype_base[0].next;
3300 hash = ntohs(pt->type) & PTYPE_HASH_MASK;
3302 while (nxt == &ptype_base[hash]) {
3303 if (++hash >= PTYPE_HASH_SIZE)
3305 nxt = ptype_base[hash].next;
3308 return list_entry(nxt, struct packet_type, list);
3311 static void ptype_seq_stop(struct seq_file *seq, void *v)
3317 static int ptype_seq_show(struct seq_file *seq, void *v)
3319 struct packet_type *pt = v;
3321 if (v == SEQ_START_TOKEN)
3322 seq_puts(seq, "Type Device Function\n");
3323 else if (pt->dev == NULL || dev_net(pt->dev) == seq_file_net(seq)) {
3324 if (pt->type == htons(ETH_P_ALL))
3325 seq_puts(seq, "ALL ");
3327 seq_printf(seq, "%04x", ntohs(pt->type));
3329 seq_printf(seq, " %-8s %pF\n",
3330 pt->dev ? pt->dev->name : "", pt->func);
3336 static const struct seq_operations ptype_seq_ops = {
3337 .start = ptype_seq_start,
3338 .next = ptype_seq_next,
3339 .stop = ptype_seq_stop,
3340 .show = ptype_seq_show,
3343 static int ptype_seq_open(struct inode *inode, struct file *file)
3345 return seq_open_net(inode, file, &ptype_seq_ops,
3346 sizeof(struct seq_net_private));
3349 static const struct file_operations ptype_seq_fops = {
3350 .owner = THIS_MODULE,
3351 .open = ptype_seq_open,
3353 .llseek = seq_lseek,
3354 .release = seq_release_net,
3358 static int __net_init dev_proc_net_init(struct net *net)
3362 if (!proc_net_fops_create(net, "dev", S_IRUGO, &dev_seq_fops))
3364 if (!proc_net_fops_create(net, "softnet_stat", S_IRUGO, &softnet_seq_fops))
3366 if (!proc_net_fops_create(net, "ptype", S_IRUGO, &ptype_seq_fops))
3369 if (wext_proc_init(net))
3375 proc_net_remove(net, "ptype");
3377 proc_net_remove(net, "softnet_stat");
3379 proc_net_remove(net, "dev");
3383 static void __net_exit dev_proc_net_exit(struct net *net)
3385 wext_proc_exit(net);
3387 proc_net_remove(net, "ptype");
3388 proc_net_remove(net, "softnet_stat");
3389 proc_net_remove(net, "dev");
3392 static struct pernet_operations __net_initdata dev_proc_ops = {
3393 .init = dev_proc_net_init,
3394 .exit = dev_proc_net_exit,
3397 static int __init dev_proc_init(void)
3399 return register_pernet_subsys(&dev_proc_ops);
3402 #define dev_proc_init() 0
3403 #endif /* CONFIG_PROC_FS */
3407 * netdev_set_master - set up master/slave pair
3408 * @slave: slave device
3409 * @master: new master device
3411 * Changes the master device of the slave. Pass %NULL to break the
3412 * bonding. The caller must hold the RTNL semaphore. On a failure
3413 * a negative errno code is returned. On success the reference counts
3414 * are adjusted, %RTM_NEWLINK is sent to the routing socket and the
3415 * function returns zero.
3417 int netdev_set_master(struct net_device *slave, struct net_device *master)
3419 struct net_device *old = slave->master;
3429 slave->master = master;
3437 slave->flags |= IFF_SLAVE;
3439 slave->flags &= ~IFF_SLAVE;
3441 rtmsg_ifinfo(RTM_NEWLINK, slave, IFF_SLAVE);
3444 EXPORT_SYMBOL(netdev_set_master);
3446 static void dev_change_rx_flags(struct net_device *dev, int flags)
3448 const struct net_device_ops *ops = dev->netdev_ops;
3450 if ((dev->flags & IFF_UP) && ops->ndo_change_rx_flags)
3451 ops->ndo_change_rx_flags(dev, flags);
3454 static int __dev_set_promiscuity(struct net_device *dev, int inc)
3456 unsigned short old_flags = dev->flags;
3462 dev->flags |= IFF_PROMISC;
3463 dev->promiscuity += inc;
3464 if (dev->promiscuity == 0) {
3467 * If inc causes overflow, untouch promisc and return error.
3470 dev->flags &= ~IFF_PROMISC;
3472 dev->promiscuity -= inc;
3473 printk(KERN_WARNING "%s: promiscuity touches roof, "
3474 "set promiscuity failed, promiscuity feature "
3475 "of device might be broken.\n", dev->name);
3479 if (dev->flags != old_flags) {
3480 printk(KERN_INFO "device %s %s promiscuous mode\n",
3481 dev->name, (dev->flags & IFF_PROMISC) ? "entered" :
3483 if (audit_enabled) {
3484 current_uid_gid(&uid, &gid);
3485 audit_log(current->audit_context, GFP_ATOMIC,
3486 AUDIT_ANOM_PROMISCUOUS,
3487 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
3488 dev->name, (dev->flags & IFF_PROMISC),
3489 (old_flags & IFF_PROMISC),
3490 audit_get_loginuid(current),
3492 audit_get_sessionid(current));
3495 dev_change_rx_flags(dev, IFF_PROMISC);
3501 * dev_set_promiscuity - update promiscuity count on a device
3505 * Add or remove promiscuity from a device. While the count in the device
3506 * remains above zero the interface remains promiscuous. Once it hits zero
3507 * the device reverts back to normal filtering operation. A negative inc
3508 * value is used to drop promiscuity on the device.
3509 * Return 0 if successful or a negative errno code on error.
3511 int dev_set_promiscuity(struct net_device *dev, int inc)
3513 unsigned short old_flags = dev->flags;
3516 err = __dev_set_promiscuity(dev, inc);
3519 if (dev->flags != old_flags)
3520 dev_set_rx_mode(dev);
3523 EXPORT_SYMBOL(dev_set_promiscuity);
3526 * dev_set_allmulti - update allmulti count on a device
3530 * Add or remove reception of all multicast frames to a device. While the
3531 * count in the device remains above zero the interface remains listening
3532 * to all interfaces. Once it hits zero the device reverts back to normal
3533 * filtering operation. A negative @inc value is used to drop the counter
3534 * when releasing a resource needing all multicasts.
3535 * Return 0 if successful or a negative errno code on error.
3538 int dev_set_allmulti(struct net_device *dev, int inc)
3540 unsigned short old_flags = dev->flags;
3544 dev->flags |= IFF_ALLMULTI;
3545 dev->allmulti += inc;
3546 if (dev->allmulti == 0) {
3549 * If inc causes overflow, untouch allmulti and return error.
3552 dev->flags &= ~IFF_ALLMULTI;
3554 dev->allmulti -= inc;
3555 printk(KERN_WARNING "%s: allmulti touches roof, "
3556 "set allmulti failed, allmulti feature of "
3557 "device might be broken.\n", dev->name);
3561 if (dev->flags ^ old_flags) {
3562 dev_change_rx_flags(dev, IFF_ALLMULTI);
3563 dev_set_rx_mode(dev);
3567 EXPORT_SYMBOL(dev_set_allmulti);
3570 * Upload unicast and multicast address lists to device and
3571 * configure RX filtering. When the device doesn't support unicast
3572 * filtering it is put in promiscuous mode while unicast addresses
3575 void __dev_set_rx_mode(struct net_device *dev)
3577 const struct net_device_ops *ops = dev->netdev_ops;
3579 /* dev_open will call this function so the list will stay sane. */
3580 if (!(dev->flags&IFF_UP))
3583 if (!netif_device_present(dev))
3586 if (ops->ndo_set_rx_mode)
3587 ops->ndo_set_rx_mode(dev);
3589 /* Unicast addresses changes may only happen under the rtnl,
3590 * therefore calling __dev_set_promiscuity here is safe.
3592 if (dev->uc.count > 0 && !dev->uc_promisc) {
3593 __dev_set_promiscuity(dev, 1);
3594 dev->uc_promisc = 1;
3595 } else if (dev->uc.count == 0 && dev->uc_promisc) {
3596 __dev_set_promiscuity(dev, -1);
3597 dev->uc_promisc = 0;
3600 if (ops->ndo_set_multicast_list)
3601 ops->ndo_set_multicast_list(dev);
3605 void dev_set_rx_mode(struct net_device *dev)
3607 netif_addr_lock_bh(dev);
3608 __dev_set_rx_mode(dev);
3609 netif_addr_unlock_bh(dev);
3612 /* hw addresses list handling functions */
3614 static int __hw_addr_add(struct netdev_hw_addr_list *list, unsigned char *addr,
3615 int addr_len, unsigned char addr_type)
3617 struct netdev_hw_addr *ha;
3620 if (addr_len > MAX_ADDR_LEN)
3623 list_for_each_entry(ha, &list->list, list) {
3624 if (!memcmp(ha->addr, addr, addr_len) &&
3625 ha->type == addr_type) {
3632 alloc_size = sizeof(*ha);
3633 if (alloc_size < L1_CACHE_BYTES)
3634 alloc_size = L1_CACHE_BYTES;
3635 ha = kmalloc(alloc_size, GFP_ATOMIC);
3638 memcpy(ha->addr, addr, addr_len);
3639 ha->type = addr_type;
3642 list_add_tail_rcu(&ha->list, &list->list);
3647 static void ha_rcu_free(struct rcu_head *head)
3649 struct netdev_hw_addr *ha;
3651 ha = container_of(head, struct netdev_hw_addr, rcu_head);
3655 static int __hw_addr_del(struct netdev_hw_addr_list *list, unsigned char *addr,
3656 int addr_len, unsigned char addr_type)
3658 struct netdev_hw_addr *ha;
3660 list_for_each_entry(ha, &list->list, list) {
3661 if (!memcmp(ha->addr, addr, addr_len) &&
3662 (ha->type == addr_type || !addr_type)) {
3665 list_del_rcu(&ha->list);
3666 call_rcu(&ha->rcu_head, ha_rcu_free);
3674 static int __hw_addr_add_multiple(struct netdev_hw_addr_list *to_list,
3675 struct netdev_hw_addr_list *from_list,
3677 unsigned char addr_type)
3680 struct netdev_hw_addr *ha, *ha2;
3683 list_for_each_entry(ha, &from_list->list, list) {
3684 type = addr_type ? addr_type : ha->type;
3685 err = __hw_addr_add(to_list, ha->addr, addr_len, type);
3692 list_for_each_entry(ha2, &from_list->list, list) {
3695 type = addr_type ? addr_type : ha2->type;
3696 __hw_addr_del(to_list, ha2->addr, addr_len, type);
3701 static void __hw_addr_del_multiple(struct netdev_hw_addr_list *to_list,
3702 struct netdev_hw_addr_list *from_list,
3704 unsigned char addr_type)
3706 struct netdev_hw_addr *ha;
3709 list_for_each_entry(ha, &from_list->list, list) {
3710 type = addr_type ? addr_type : ha->type;
3711 __hw_addr_del(to_list, ha->addr, addr_len, addr_type);
3715 static int __hw_addr_sync(struct netdev_hw_addr_list *to_list,
3716 struct netdev_hw_addr_list *from_list,
3720 struct netdev_hw_addr *ha, *tmp;
3722 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3724 err = __hw_addr_add(to_list, ha->addr,
3725 addr_len, ha->type);
3730 } else if (ha->refcount == 1) {
3731 __hw_addr_del(to_list, ha->addr, addr_len, ha->type);
3732 __hw_addr_del(from_list, ha->addr, addr_len, ha->type);
3738 static void __hw_addr_unsync(struct netdev_hw_addr_list *to_list,
3739 struct netdev_hw_addr_list *from_list,
3742 struct netdev_hw_addr *ha, *tmp;
3744 list_for_each_entry_safe(ha, tmp, &from_list->list, list) {
3746 __hw_addr_del(to_list, ha->addr,
3747 addr_len, ha->type);
3749 __hw_addr_del(from_list, ha->addr,
3750 addr_len, ha->type);
3755 static void __hw_addr_flush(struct netdev_hw_addr_list *list)
3757 struct netdev_hw_addr *ha, *tmp;
3759 list_for_each_entry_safe(ha, tmp, &list->list, list) {
3760 list_del_rcu(&ha->list);
3761 call_rcu(&ha->rcu_head, ha_rcu_free);
3766 static void __hw_addr_init(struct netdev_hw_addr_list *list)
3768 INIT_LIST_HEAD(&list->list);
3772 /* Device addresses handling functions */
3774 static void dev_addr_flush(struct net_device *dev)
3776 /* rtnl_mutex must be held here */
3778 __hw_addr_flush(&dev->dev_addrs);
3779 dev->dev_addr = NULL;
3782 static int dev_addr_init(struct net_device *dev)
3784 unsigned char addr[MAX_ADDR_LEN];
3785 struct netdev_hw_addr *ha;
3788 /* rtnl_mutex must be held here */
3790 __hw_addr_init(&dev->dev_addrs);
3791 memset(addr, 0, sizeof(addr));
3792 err = __hw_addr_add(&dev->dev_addrs, addr, sizeof(addr),
3793 NETDEV_HW_ADDR_T_LAN);
3796 * Get the first (previously created) address from the list
3797 * and set dev_addr pointer to this location.
3799 ha = list_first_entry(&dev->dev_addrs.list,
3800 struct netdev_hw_addr, list);
3801 dev->dev_addr = ha->addr;
3807 * dev_addr_add - Add a device address
3809 * @addr: address to add
3810 * @addr_type: address type
3812 * Add a device address to the device or increase the reference count if
3813 * it already exists.
3815 * The caller must hold the rtnl_mutex.
3817 int dev_addr_add(struct net_device *dev, unsigned char *addr,
3818 unsigned char addr_type)
3824 err = __hw_addr_add(&dev->dev_addrs, addr, dev->addr_len, addr_type);
3826 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3829 EXPORT_SYMBOL(dev_addr_add);
3832 * dev_addr_del - Release a device address.
3834 * @addr: address to delete
3835 * @addr_type: address type
3837 * Release reference to a device address and remove it from the device
3838 * if the reference count drops to zero.
3840 * The caller must hold the rtnl_mutex.
3842 int dev_addr_del(struct net_device *dev, unsigned char *addr,
3843 unsigned char addr_type)
3846 struct netdev_hw_addr *ha;
3851 * We can not remove the first address from the list because
3852 * dev->dev_addr points to that.
3854 ha = list_first_entry(&dev->dev_addrs.list,
3855 struct netdev_hw_addr, list);
3856 if (ha->addr == dev->dev_addr && ha->refcount == 1)
3859 err = __hw_addr_del(&dev->dev_addrs, addr, dev->addr_len,
3862 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
3865 EXPORT_SYMBOL(dev_addr_del);
3868 * dev_addr_add_multiple - Add device addresses from another device
3869 * @to_dev: device to which addresses will be added
3870 * @from_dev: device from which addresses will be added
3871 * @addr_type: address type - 0 means type will be used from from_dev
3873 * Add device addresses of the one device to another.
3875 * The caller must hold the rtnl_mutex.
3877 int dev_addr_add_multiple(struct net_device *to_dev,
3878 struct net_device *from_dev,
3879 unsigned char addr_type)
3885 if (from_dev->addr_len != to_dev->addr_len)
3887 err = __hw_addr_add_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3888 to_dev->addr_len, addr_type);
3890 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3893 EXPORT_SYMBOL(dev_addr_add_multiple);
3896 * dev_addr_del_multiple - Delete device addresses by another device
3897 * @to_dev: device where the addresses will be deleted
3898 * @from_dev: device by which addresses the addresses will be deleted
3899 * @addr_type: address type - 0 means type will used from from_dev
3901 * Deletes addresses in to device by the list of addresses in from device.
3903 * The caller must hold the rtnl_mutex.
3905 int dev_addr_del_multiple(struct net_device *to_dev,
3906 struct net_device *from_dev,
3907 unsigned char addr_type)
3911 if (from_dev->addr_len != to_dev->addr_len)
3913 __hw_addr_del_multiple(&to_dev->dev_addrs, &from_dev->dev_addrs,
3914 to_dev->addr_len, addr_type);
3915 call_netdevice_notifiers(NETDEV_CHANGEADDR, to_dev);
3918 EXPORT_SYMBOL(dev_addr_del_multiple);
3920 /* multicast addresses handling functions */
3922 int __dev_addr_delete(struct dev_addr_list **list, int *count,
3923 void *addr, int alen, int glbl)
3925 struct dev_addr_list *da;
3927 for (; (da = *list) != NULL; list = &da->next) {
3928 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3929 alen == da->da_addrlen) {
3931 int old_glbl = da->da_gusers;
3948 int __dev_addr_add(struct dev_addr_list **list, int *count,
3949 void *addr, int alen, int glbl)
3951 struct dev_addr_list *da;
3953 for (da = *list; da != NULL; da = da->next) {
3954 if (memcmp(da->da_addr, addr, da->da_addrlen) == 0 &&
3955 da->da_addrlen == alen) {
3957 int old_glbl = da->da_gusers;
3967 da = kzalloc(sizeof(*da), GFP_ATOMIC);
3970 memcpy(da->da_addr, addr, alen);
3971 da->da_addrlen = alen;
3973 da->da_gusers = glbl ? 1 : 0;
3981 * dev_unicast_delete - Release secondary unicast address.
3983 * @addr: address to delete
3985 * Release reference to a secondary unicast address and remove it
3986 * from the device if the reference count drops to zero.
3988 * The caller must hold the rtnl_mutex.
3990 int dev_unicast_delete(struct net_device *dev, void *addr)
3996 netif_addr_lock_bh(dev);
3997 err = __hw_addr_del(&dev->uc, addr, dev->addr_len,
3998 NETDEV_HW_ADDR_T_UNICAST);
4000 __dev_set_rx_mode(dev);
4001 netif_addr_unlock_bh(dev);
4004 EXPORT_SYMBOL(dev_unicast_delete);
4007 * dev_unicast_add - add a secondary unicast address
4009 * @addr: address to add
4011 * Add a secondary unicast address to the device or increase
4012 * the reference count if it already exists.
4014 * The caller must hold the rtnl_mutex.
4016 int dev_unicast_add(struct net_device *dev, void *addr)
4022 netif_addr_lock_bh(dev);
4023 err = __hw_addr_add(&dev->uc, addr, dev->addr_len,
4024 NETDEV_HW_ADDR_T_UNICAST);
4026 __dev_set_rx_mode(dev);
4027 netif_addr_unlock_bh(dev);
4030 EXPORT_SYMBOL(dev_unicast_add);
4032 int __dev_addr_sync(struct dev_addr_list **to, int *to_count,
4033 struct dev_addr_list **from, int *from_count)
4035 struct dev_addr_list *da, *next;
4039 while (da != NULL) {
4041 if (!da->da_synced) {
4042 err = __dev_addr_add(to, to_count,
4043 da->da_addr, da->da_addrlen, 0);
4048 } else if (da->da_users == 1) {
4049 __dev_addr_delete(to, to_count,
4050 da->da_addr, da->da_addrlen, 0);
4051 __dev_addr_delete(from, from_count,
4052 da->da_addr, da->da_addrlen, 0);
4058 EXPORT_SYMBOL_GPL(__dev_addr_sync);
4060 void __dev_addr_unsync(struct dev_addr_list **to, int *to_count,
4061 struct dev_addr_list **from, int *from_count)
4063 struct dev_addr_list *da, *next;
4066 while (da != NULL) {
4068 if (da->da_synced) {
4069 __dev_addr_delete(to, to_count,
4070 da->da_addr, da->da_addrlen, 0);
4072 __dev_addr_delete(from, from_count,
4073 da->da_addr, da->da_addrlen, 0);
4078 EXPORT_SYMBOL_GPL(__dev_addr_unsync);
4081 * dev_unicast_sync - Synchronize device's unicast list to another device
4082 * @to: destination device
4083 * @from: source device
4085 * Add newly added addresses to the destination device and release
4086 * addresses that have no users left. The source device must be
4087 * locked by netif_tx_lock_bh.
4089 * This function is intended to be called from the dev->set_rx_mode
4090 * function of layered software devices.
4092 int dev_unicast_sync(struct net_device *to, struct net_device *from)
4096 if (to->addr_len != from->addr_len)
4099 netif_addr_lock_bh(to);
4100 err = __hw_addr_sync(&to->uc, &from->uc, to->addr_len);
4102 __dev_set_rx_mode(to);
4103 netif_addr_unlock_bh(to);
4106 EXPORT_SYMBOL(dev_unicast_sync);
4109 * dev_unicast_unsync - Remove synchronized addresses from the destination device
4110 * @to: destination device
4111 * @from: source device
4113 * Remove all addresses that were added to the destination device by
4114 * dev_unicast_sync(). This function is intended to be called from the
4115 * dev->stop function of layered software devices.
4117 void dev_unicast_unsync(struct net_device *to, struct net_device *from)
4119 if (to->addr_len != from->addr_len)
4122 netif_addr_lock_bh(from);
4123 netif_addr_lock(to);
4124 __hw_addr_unsync(&to->uc, &from->uc, to->addr_len);
4125 __dev_set_rx_mode(to);
4126 netif_addr_unlock(to);
4127 netif_addr_unlock_bh(from);
4129 EXPORT_SYMBOL(dev_unicast_unsync);
4131 static void dev_unicast_flush(struct net_device *dev)
4133 netif_addr_lock_bh(dev);
4134 __hw_addr_flush(&dev->uc);
4135 netif_addr_unlock_bh(dev);
4138 static void dev_unicast_init(struct net_device *dev)
4140 __hw_addr_init(&dev->uc);
4144 static void __dev_addr_discard(struct dev_addr_list **list)
4146 struct dev_addr_list *tmp;
4148 while (*list != NULL) {
4151 if (tmp->da_users > tmp->da_gusers)
4152 printk("__dev_addr_discard: address leakage! "
4153 "da_users=%d\n", tmp->da_users);
4158 static void dev_addr_discard(struct net_device *dev)
4160 netif_addr_lock_bh(dev);
4162 __dev_addr_discard(&dev->mc_list);
4165 netif_addr_unlock_bh(dev);
4169 * dev_get_flags - get flags reported to userspace
4172 * Get the combination of flag bits exported through APIs to userspace.
4174 unsigned dev_get_flags(const struct net_device *dev)
4178 flags = (dev->flags & ~(IFF_PROMISC |
4183 (dev->gflags & (IFF_PROMISC |
4186 if (netif_running(dev)) {
4187 if (netif_oper_up(dev))
4188 flags |= IFF_RUNNING;
4189 if (netif_carrier_ok(dev))
4190 flags |= IFF_LOWER_UP;
4191 if (netif_dormant(dev))
4192 flags |= IFF_DORMANT;
4197 EXPORT_SYMBOL(dev_get_flags);
4200 * dev_change_flags - change device settings
4202 * @flags: device state flags
4204 * Change settings on device based state flags. The flags are
4205 * in the userspace exported format.
4207 int dev_change_flags(struct net_device *dev, unsigned flags)
4210 int old_flags = dev->flags;
4215 * Set the flags on our device.
4218 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
4219 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
4221 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
4225 * Load in the correct multicast list now the flags have changed.
4228 if ((old_flags ^ flags) & IFF_MULTICAST)
4229 dev_change_rx_flags(dev, IFF_MULTICAST);
4231 dev_set_rx_mode(dev);
4234 * Have we downed the interface. We handle IFF_UP ourselves
4235 * according to user attempts to set it, rather than blindly
4240 if ((old_flags ^ flags) & IFF_UP) { /* Bit is different ? */
4241 ret = ((old_flags & IFF_UP) ? dev_close : dev_open)(dev);
4244 dev_set_rx_mode(dev);
4247 if (dev->flags & IFF_UP &&
4248 ((old_flags ^ dev->flags) & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI |
4250 call_netdevice_notifiers(NETDEV_CHANGE, dev);
4252 if ((flags ^ dev->gflags) & IFF_PROMISC) {
4253 int inc = (flags & IFF_PROMISC) ? 1 : -1;
4255 dev->gflags ^= IFF_PROMISC;
4256 dev_set_promiscuity(dev, inc);
4259 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4260 is important. Some (broken) drivers set IFF_PROMISC, when
4261 IFF_ALLMULTI is requested not asking us and not reporting.
4263 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
4264 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
4266 dev->gflags ^= IFF_ALLMULTI;
4267 dev_set_allmulti(dev, inc);
4270 /* Exclude state transition flags, already notified */
4271 changes = (old_flags ^ dev->flags) & ~(IFF_UP | IFF_RUNNING);
4273 rtmsg_ifinfo(RTM_NEWLINK, dev, changes);
4277 EXPORT_SYMBOL(dev_change_flags);
4280 * dev_set_mtu - Change maximum transfer unit
4282 * @new_mtu: new transfer unit
4284 * Change the maximum transfer size of the network device.
4286 int dev_set_mtu(struct net_device *dev, int new_mtu)
4288 const struct net_device_ops *ops = dev->netdev_ops;
4291 if (new_mtu == dev->mtu)
4294 /* MTU must be positive. */
4298 if (!netif_device_present(dev))
4302 if (ops->ndo_change_mtu)
4303 err = ops->ndo_change_mtu(dev, new_mtu);
4307 if (!err && dev->flags & IFF_UP)
4308 call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
4311 EXPORT_SYMBOL(dev_set_mtu);
4314 * dev_set_mac_address - Change Media Access Control Address
4318 * Change the hardware (MAC) address of the device
4320 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
4322 const struct net_device_ops *ops = dev->netdev_ops;
4325 if (!ops->ndo_set_mac_address)
4327 if (sa->sa_family != dev->type)
4329 if (!netif_device_present(dev))
4331 err = ops->ndo_set_mac_address(dev, sa);
4333 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4336 EXPORT_SYMBOL(dev_set_mac_address);
4339 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4341 static int dev_ifsioc_locked(struct net *net, struct ifreq *ifr, unsigned int cmd)
4344 struct net_device *dev = dev_get_by_name_rcu(net, ifr->ifr_name);
4350 case SIOCGIFFLAGS: /* Get interface flags */
4351 ifr->ifr_flags = (short) dev_get_flags(dev);
4354 case SIOCGIFMETRIC: /* Get the metric on the interface
4355 (currently unused) */
4356 ifr->ifr_metric = 0;
4359 case SIOCGIFMTU: /* Get the MTU of a device */
4360 ifr->ifr_mtu = dev->mtu;
4365 memset(ifr->ifr_hwaddr.sa_data, 0, sizeof ifr->ifr_hwaddr.sa_data);
4367 memcpy(ifr->ifr_hwaddr.sa_data, dev->dev_addr,
4368 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4369 ifr->ifr_hwaddr.sa_family = dev->type;
4377 ifr->ifr_map.mem_start = dev->mem_start;
4378 ifr->ifr_map.mem_end = dev->mem_end;
4379 ifr->ifr_map.base_addr = dev->base_addr;
4380 ifr->ifr_map.irq = dev->irq;
4381 ifr->ifr_map.dma = dev->dma;
4382 ifr->ifr_map.port = dev->if_port;
4386 ifr->ifr_ifindex = dev->ifindex;
4390 ifr->ifr_qlen = dev->tx_queue_len;
4394 /* dev_ioctl() should ensure this case
4406 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4408 static int dev_ifsioc(struct net *net, struct ifreq *ifr, unsigned int cmd)
4411 struct net_device *dev = __dev_get_by_name(net, ifr->ifr_name);
4412 const struct net_device_ops *ops;
4417 ops = dev->netdev_ops;
4420 case SIOCSIFFLAGS: /* Set interface flags */
4421 return dev_change_flags(dev, ifr->ifr_flags);
4423 case SIOCSIFMETRIC: /* Set the metric on the interface
4424 (currently unused) */
4427 case SIOCSIFMTU: /* Set the MTU of a device */
4428 return dev_set_mtu(dev, ifr->ifr_mtu);
4431 return dev_set_mac_address(dev, &ifr->ifr_hwaddr);
4433 case SIOCSIFHWBROADCAST:
4434 if (ifr->ifr_hwaddr.sa_family != dev->type)
4436 memcpy(dev->broadcast, ifr->ifr_hwaddr.sa_data,
4437 min(sizeof ifr->ifr_hwaddr.sa_data, (size_t) dev->addr_len));
4438 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
4442 if (ops->ndo_set_config) {
4443 if (!netif_device_present(dev))
4445 return ops->ndo_set_config(dev, &ifr->ifr_map);
4450 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4451 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4453 if (!netif_device_present(dev))
4455 return dev_mc_add(dev, ifr->ifr_hwaddr.sa_data,
4459 if ((!ops->ndo_set_multicast_list && !ops->ndo_set_rx_mode) ||
4460 ifr->ifr_hwaddr.sa_family != AF_UNSPEC)
4462 if (!netif_device_present(dev))
4464 return dev_mc_delete(dev, ifr->ifr_hwaddr.sa_data,
4468 if (ifr->ifr_qlen < 0)
4470 dev->tx_queue_len = ifr->ifr_qlen;
4474 ifr->ifr_newname[IFNAMSIZ-1] = '\0';
4475 return dev_change_name(dev, ifr->ifr_newname);
4478 * Unknown or private ioctl
4481 if ((cmd >= SIOCDEVPRIVATE &&
4482 cmd <= SIOCDEVPRIVATE + 15) ||
4483 cmd == SIOCBONDENSLAVE ||
4484 cmd == SIOCBONDRELEASE ||
4485 cmd == SIOCBONDSETHWADDR ||
4486 cmd == SIOCBONDSLAVEINFOQUERY ||
4487 cmd == SIOCBONDINFOQUERY ||
4488 cmd == SIOCBONDCHANGEACTIVE ||
4489 cmd == SIOCGMIIPHY ||
4490 cmd == SIOCGMIIREG ||
4491 cmd == SIOCSMIIREG ||
4492 cmd == SIOCBRADDIF ||
4493 cmd == SIOCBRDELIF ||
4494 cmd == SIOCSHWTSTAMP ||
4495 cmd == SIOCWANDEV) {
4497 if (ops->ndo_do_ioctl) {
4498 if (netif_device_present(dev))
4499 err = ops->ndo_do_ioctl(dev, ifr, cmd);
4511 * This function handles all "interface"-type I/O control requests. The actual
4512 * 'doing' part of this is dev_ifsioc above.
4516 * dev_ioctl - network device ioctl
4517 * @net: the applicable net namespace
4518 * @cmd: command to issue
4519 * @arg: pointer to a struct ifreq in user space
4521 * Issue ioctl functions to devices. This is normally called by the
4522 * user space syscall interfaces but can sometimes be useful for
4523 * other purposes. The return value is the return from the syscall if
4524 * positive or a negative errno code on error.
4527 int dev_ioctl(struct net *net, unsigned int cmd, void __user *arg)
4533 /* One special case: SIOCGIFCONF takes ifconf argument
4534 and requires shared lock, because it sleeps writing
4538 if (cmd == SIOCGIFCONF) {
4540 ret = dev_ifconf(net, (char __user *) arg);
4544 if (cmd == SIOCGIFNAME)
4545 return dev_ifname(net, (struct ifreq __user *)arg);
4547 if (copy_from_user(&ifr, arg, sizeof(struct ifreq)))
4550 ifr.ifr_name[IFNAMSIZ-1] = 0;
4552 colon = strchr(ifr.ifr_name, ':');
4557 * See which interface the caller is talking about.
4562 * These ioctl calls:
4563 * - can be done by all.
4564 * - atomic and do not require locking.
4575 dev_load(net, ifr.ifr_name);
4577 ret = dev_ifsioc_locked(net, &ifr, cmd);
4582 if (copy_to_user(arg, &ifr,
4583 sizeof(struct ifreq)))
4589 dev_load(net, ifr.ifr_name);
4591 ret = dev_ethtool(net, &ifr);
4596 if (copy_to_user(arg, &ifr,
4597 sizeof(struct ifreq)))
4603 * These ioctl calls:
4604 * - require superuser power.
4605 * - require strict serialization.
4611 if (!capable(CAP_NET_ADMIN))
4613 dev_load(net, ifr.ifr_name);
4615 ret = dev_ifsioc(net, &ifr, cmd);
4620 if (copy_to_user(arg, &ifr,
4621 sizeof(struct ifreq)))
4627 * These ioctl calls:
4628 * - require superuser power.
4629 * - require strict serialization.
4630 * - do not return a value
4640 case SIOCSIFHWBROADCAST:
4643 case SIOCBONDENSLAVE:
4644 case SIOCBONDRELEASE:
4645 case SIOCBONDSETHWADDR:
4646 case SIOCBONDCHANGEACTIVE:
4650 if (!capable(CAP_NET_ADMIN))
4653 case SIOCBONDSLAVEINFOQUERY:
4654 case SIOCBONDINFOQUERY:
4655 dev_load(net, ifr.ifr_name);
4657 ret = dev_ifsioc(net, &ifr, cmd);
4662 /* Get the per device memory space. We can add this but
4663 * currently do not support it */
4665 /* Set the per device memory buffer space.
4666 * Not applicable in our case */
4671 * Unknown or private ioctl.
4674 if (cmd == SIOCWANDEV ||
4675 (cmd >= SIOCDEVPRIVATE &&
4676 cmd <= SIOCDEVPRIVATE + 15)) {
4677 dev_load(net, ifr.ifr_name);
4679 ret = dev_ifsioc(net, &ifr, cmd);
4681 if (!ret && copy_to_user(arg, &ifr,
4682 sizeof(struct ifreq)))
4686 /* Take care of Wireless Extensions */
4687 if (cmd >= SIOCIWFIRST && cmd <= SIOCIWLAST)
4688 return wext_handle_ioctl(net, &ifr, cmd, arg);
4695 * dev_new_index - allocate an ifindex
4696 * @net: the applicable net namespace
4698 * Returns a suitable unique value for a new device interface
4699 * number. The caller must hold the rtnl semaphore or the
4700 * dev_base_lock to be sure it remains unique.
4702 static int dev_new_index(struct net *net)
4708 if (!__dev_get_by_index(net, ifindex))
4713 /* Delayed registration/unregisteration */
4714 static LIST_HEAD(net_todo_list);
4716 static void net_set_todo(struct net_device *dev)
4718 list_add_tail(&dev->todo_list, &net_todo_list);
4721 static void rollback_registered_many(struct list_head *head)
4723 struct net_device *dev;
4725 BUG_ON(dev_boot_phase);
4728 list_for_each_entry(dev, head, unreg_list) {
4729 /* Some devices call without registering
4730 * for initialization unwind.
4732 if (dev->reg_state == NETREG_UNINITIALIZED) {
4733 pr_debug("unregister_netdevice: device %s/%p never "
4734 "was registered\n", dev->name, dev);
4740 BUG_ON(dev->reg_state != NETREG_REGISTERED);
4742 /* If device is running, close it first. */
4745 /* And unlink it from device chain. */
4746 unlist_netdevice(dev);
4748 dev->reg_state = NETREG_UNREGISTERING;
4753 list_for_each_entry(dev, head, unreg_list) {
4754 /* Shutdown queueing discipline. */
4758 /* Notify protocols, that we are about to destroy
4759 this device. They should clean all the things.
4761 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
4764 * Flush the unicast and multicast chains
4766 dev_unicast_flush(dev);
4767 dev_addr_discard(dev);
4769 if (dev->netdev_ops->ndo_uninit)
4770 dev->netdev_ops->ndo_uninit(dev);
4772 /* Notifier chain MUST detach us from master device. */
4773 WARN_ON(dev->master);
4775 /* Remove entries from kobject tree */
4776 netdev_unregister_kobject(dev);
4781 list_for_each_entry(dev, head, unreg_list)
4785 static void rollback_registered(struct net_device *dev)
4789 list_add(&dev->unreg_list, &single);
4790 rollback_registered_many(&single);
4793 static void __netdev_init_queue_locks_one(struct net_device *dev,
4794 struct netdev_queue *dev_queue,
4797 spin_lock_init(&dev_queue->_xmit_lock);
4798 netdev_set_xmit_lockdep_class(&dev_queue->_xmit_lock, dev->type);
4799 dev_queue->xmit_lock_owner = -1;
4802 static void netdev_init_queue_locks(struct net_device *dev)
4804 netdev_for_each_tx_queue(dev, __netdev_init_queue_locks_one, NULL);
4805 __netdev_init_queue_locks_one(dev, &dev->rx_queue, NULL);
4808 unsigned long netdev_fix_features(unsigned long features, const char *name)
4810 /* Fix illegal SG+CSUM combinations. */
4811 if ((features & NETIF_F_SG) &&
4812 !(features & NETIF_F_ALL_CSUM)) {
4814 printk(KERN_NOTICE "%s: Dropping NETIF_F_SG since no "
4815 "checksum feature.\n", name);
4816 features &= ~NETIF_F_SG;
4819 /* TSO requires that SG is present as well. */
4820 if ((features & NETIF_F_TSO) && !(features & NETIF_F_SG)) {
4822 printk(KERN_NOTICE "%s: Dropping NETIF_F_TSO since no "
4823 "SG feature.\n", name);
4824 features &= ~NETIF_F_TSO;
4827 if (features & NETIF_F_UFO) {
4828 if (!(features & NETIF_F_GEN_CSUM)) {
4830 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4831 "since no NETIF_F_HW_CSUM feature.\n",
4833 features &= ~NETIF_F_UFO;
4836 if (!(features & NETIF_F_SG)) {
4838 printk(KERN_ERR "%s: Dropping NETIF_F_UFO "
4839 "since no NETIF_F_SG feature.\n", name);
4840 features &= ~NETIF_F_UFO;
4846 EXPORT_SYMBOL(netdev_fix_features);
4849 * register_netdevice - register a network device
4850 * @dev: device to register
4852 * Take a completed network device structure and add it to the kernel
4853 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
4854 * chain. 0 is returned on success. A negative errno code is returned
4855 * on a failure to set up the device, or if the name is a duplicate.
4857 * Callers must hold the rtnl semaphore. You may want
4858 * register_netdev() instead of this.
4861 * The locking appears insufficient to guarantee two parallel registers
4862 * will not get the same name.
4865 int register_netdevice(struct net_device *dev)
4867 struct hlist_head *head;
4868 struct hlist_node *p;
4870 struct net *net = dev_net(dev);
4872 BUG_ON(dev_boot_phase);
4877 /* When net_device's are persistent, this will be fatal. */
4878 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
4881 spin_lock_init(&dev->addr_list_lock);
4882 netdev_set_addr_lockdep_class(dev);
4883 netdev_init_queue_locks(dev);
4887 /* Init, if this function is available */
4888 if (dev->netdev_ops->ndo_init) {
4889 ret = dev->netdev_ops->ndo_init(dev);
4897 if (!dev_valid_name(dev->name)) {
4902 dev->ifindex = dev_new_index(net);
4903 if (dev->iflink == -1)
4904 dev->iflink = dev->ifindex;
4906 /* Check for existence of name */
4907 head = dev_name_hash(net, dev->name);
4908 hlist_for_each(p, head) {
4909 struct net_device *d
4910 = hlist_entry(p, struct net_device, name_hlist);
4911 if (!strncmp(d->name, dev->name, IFNAMSIZ)) {
4917 /* Fix illegal checksum combinations */
4918 if ((dev->features & NETIF_F_HW_CSUM) &&
4919 (dev->features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4920 printk(KERN_NOTICE "%s: mixed HW and IP checksum settings.\n",
4922 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
4925 if ((dev->features & NETIF_F_NO_CSUM) &&
4926 (dev->features & (NETIF_F_HW_CSUM|NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
4927 printk(KERN_NOTICE "%s: mixed no checksumming and other settings.\n",
4929 dev->features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM|NETIF_F_HW_CSUM);
4932 dev->features = netdev_fix_features(dev->features, dev->name);
4934 /* Enable software GSO if SG is supported. */
4935 if (dev->features & NETIF_F_SG)
4936 dev->features |= NETIF_F_GSO;
4938 netdev_initialize_kobject(dev);
4940 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
4941 ret = notifier_to_errno(ret);
4945 ret = netdev_register_kobject(dev);
4948 dev->reg_state = NETREG_REGISTERED;
4951 * Default initial state at registry is that the
4952 * device is present.
4955 set_bit(__LINK_STATE_PRESENT, &dev->state);
4957 dev_init_scheduler(dev);
4959 list_netdevice(dev);
4961 /* Notify protocols, that a new device appeared. */
4962 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
4963 ret = notifier_to_errno(ret);
4965 rollback_registered(dev);
4966 dev->reg_state = NETREG_UNREGISTERED;
4973 if (dev->netdev_ops->ndo_uninit)
4974 dev->netdev_ops->ndo_uninit(dev);
4977 EXPORT_SYMBOL(register_netdevice);
4980 * init_dummy_netdev - init a dummy network device for NAPI
4981 * @dev: device to init
4983 * This takes a network device structure and initialize the minimum
4984 * amount of fields so it can be used to schedule NAPI polls without
4985 * registering a full blown interface. This is to be used by drivers
4986 * that need to tie several hardware interfaces to a single NAPI
4987 * poll scheduler due to HW limitations.
4989 int init_dummy_netdev(struct net_device *dev)
4991 /* Clear everything. Note we don't initialize spinlocks
4992 * are they aren't supposed to be taken by any of the
4993 * NAPI code and this dummy netdev is supposed to be
4994 * only ever used for NAPI polls
4996 memset(dev, 0, sizeof(struct net_device));
4998 /* make sure we BUG if trying to hit standard
4999 * register/unregister code path
5001 dev->reg_state = NETREG_DUMMY;
5003 /* initialize the ref count */
5004 atomic_set(&dev->refcnt, 1);
5006 /* NAPI wants this */
5007 INIT_LIST_HEAD(&dev->napi_list);
5009 /* a dummy interface is started by default */
5010 set_bit(__LINK_STATE_PRESENT, &dev->state);
5011 set_bit(__LINK_STATE_START, &dev->state);
5015 EXPORT_SYMBOL_GPL(init_dummy_netdev);
5019 * register_netdev - register a network device
5020 * @dev: device to register
5022 * Take a completed network device structure and add it to the kernel
5023 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5024 * chain. 0 is returned on success. A negative errno code is returned
5025 * on a failure to set up the device, or if the name is a duplicate.
5027 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5028 * and expands the device name if you passed a format string to
5031 int register_netdev(struct net_device *dev)
5038 * If the name is a format string the caller wants us to do a
5041 if (strchr(dev->name, '%')) {
5042 err = dev_alloc_name(dev, dev->name);
5047 err = register_netdevice(dev);
5052 EXPORT_SYMBOL(register_netdev);
5055 * netdev_wait_allrefs - wait until all references are gone.
5057 * This is called when unregistering network devices.
5059 * Any protocol or device that holds a reference should register
5060 * for netdevice notification, and cleanup and put back the
5061 * reference if they receive an UNREGISTER event.
5062 * We can get stuck here if buggy protocols don't correctly
5065 static void netdev_wait_allrefs(struct net_device *dev)
5067 unsigned long rebroadcast_time, warning_time;
5069 rebroadcast_time = warning_time = jiffies;
5070 while (atomic_read(&dev->refcnt) != 0) {
5071 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
5074 /* Rebroadcast unregister notification */
5075 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5077 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
5079 /* We must not have linkwatch events
5080 * pending on unregister. If this
5081 * happens, we simply run the queue
5082 * unscheduled, resulting in a noop
5085 linkwatch_run_queue();
5090 rebroadcast_time = jiffies;
5095 if (time_after(jiffies, warning_time + 10 * HZ)) {
5096 printk(KERN_EMERG "unregister_netdevice: "
5097 "waiting for %s to become free. Usage "
5099 dev->name, atomic_read(&dev->refcnt));
5100 warning_time = jiffies;
5109 * register_netdevice(x1);
5110 * register_netdevice(x2);
5112 * unregister_netdevice(y1);
5113 * unregister_netdevice(y2);
5119 * We are invoked by rtnl_unlock().
5120 * This allows us to deal with problems:
5121 * 1) We can delete sysfs objects which invoke hotplug
5122 * without deadlocking with linkwatch via keventd.
5123 * 2) Since we run with the RTNL semaphore not held, we can sleep
5124 * safely in order to wait for the netdev refcnt to drop to zero.
5126 * We must not return until all unregister events added during
5127 * the interval the lock was held have been completed.
5129 void netdev_run_todo(void)
5131 struct list_head list;
5133 /* Snapshot list, allow later requests */
5134 list_replace_init(&net_todo_list, &list);
5138 while (!list_empty(&list)) {
5139 struct net_device *dev
5140 = list_entry(list.next, struct net_device, todo_list);
5141 list_del(&dev->todo_list);
5143 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
5144 printk(KERN_ERR "network todo '%s' but state %d\n",
5145 dev->name, dev->reg_state);
5150 dev->reg_state = NETREG_UNREGISTERED;
5152 on_each_cpu(flush_backlog, dev, 1);
5154 netdev_wait_allrefs(dev);
5157 BUG_ON(atomic_read(&dev->refcnt));
5158 WARN_ON(dev->ip_ptr);
5159 WARN_ON(dev->ip6_ptr);
5160 WARN_ON(dev->dn_ptr);
5162 if (dev->destructor)
5163 dev->destructor(dev);
5165 /* Free network device */
5166 kobject_put(&dev->dev.kobj);
5171 * dev_get_stats - get network device statistics
5172 * @dev: device to get statistics from
5174 * Get network statistics from device. The device driver may provide
5175 * its own method by setting dev->netdev_ops->get_stats; otherwise
5176 * the internal statistics structure is used.
5178 const struct net_device_stats *dev_get_stats(struct net_device *dev)
5180 const struct net_device_ops *ops = dev->netdev_ops;
5182 if (ops->ndo_get_stats)
5183 return ops->ndo_get_stats(dev);
5185 unsigned long tx_bytes = 0, tx_packets = 0, tx_dropped = 0;
5186 struct net_device_stats *stats = &dev->stats;
5188 struct netdev_queue *txq;
5190 for (i = 0; i < dev->num_tx_queues; i++) {
5191 txq = netdev_get_tx_queue(dev, i);
5192 tx_bytes += txq->tx_bytes;
5193 tx_packets += txq->tx_packets;
5194 tx_dropped += txq->tx_dropped;
5196 if (tx_bytes || tx_packets || tx_dropped) {
5197 stats->tx_bytes = tx_bytes;
5198 stats->tx_packets = tx_packets;
5199 stats->tx_dropped = tx_dropped;
5204 EXPORT_SYMBOL(dev_get_stats);
5206 static void netdev_init_one_queue(struct net_device *dev,
5207 struct netdev_queue *queue,
5213 static void netdev_init_queues(struct net_device *dev)
5215 netdev_init_one_queue(dev, &dev->rx_queue, NULL);
5216 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
5217 spin_lock_init(&dev->tx_global_lock);
5221 * alloc_netdev_mq - allocate network device
5222 * @sizeof_priv: size of private data to allocate space for
5223 * @name: device name format string
5224 * @setup: callback to initialize device
5225 * @queue_count: the number of subqueues to allocate
5227 * Allocates a struct net_device with private data area for driver use
5228 * and performs basic initialization. Also allocates subquue structs
5229 * for each queue on the device at the end of the netdevice.
5231 struct net_device *alloc_netdev_mq(int sizeof_priv, const char *name,
5232 void (*setup)(struct net_device *), unsigned int queue_count)
5234 struct netdev_queue *tx;
5235 struct net_device *dev;
5237 struct net_device *p;
5239 BUG_ON(strlen(name) >= sizeof(dev->name));
5241 alloc_size = sizeof(struct net_device);
5243 /* ensure 32-byte alignment of private area */
5244 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
5245 alloc_size += sizeof_priv;
5247 /* ensure 32-byte alignment of whole construct */
5248 alloc_size += NETDEV_ALIGN - 1;
5250 p = kzalloc(alloc_size, GFP_KERNEL);
5252 printk(KERN_ERR "alloc_netdev: Unable to allocate device.\n");
5256 tx = kcalloc(queue_count, sizeof(struct netdev_queue), GFP_KERNEL);
5258 printk(KERN_ERR "alloc_netdev: Unable to allocate "
5263 dev = PTR_ALIGN(p, NETDEV_ALIGN);
5264 dev->padded = (char *)dev - (char *)p;
5266 if (dev_addr_init(dev))
5269 dev_unicast_init(dev);
5271 dev_net_set(dev, &init_net);
5274 dev->num_tx_queues = queue_count;
5275 dev->real_num_tx_queues = queue_count;
5277 dev->gso_max_size = GSO_MAX_SIZE;
5279 netdev_init_queues(dev);
5281 INIT_LIST_HEAD(&dev->napi_list);
5282 INIT_LIST_HEAD(&dev->unreg_list);
5283 dev->priv_flags = IFF_XMIT_DST_RELEASE;
5285 strcpy(dev->name, name);
5295 EXPORT_SYMBOL(alloc_netdev_mq);
5298 * free_netdev - free network device
5301 * This function does the last stage of destroying an allocated device
5302 * interface. The reference to the device object is released.
5303 * If this is the last reference then it will be freed.
5305 void free_netdev(struct net_device *dev)
5307 struct napi_struct *p, *n;
5309 release_net(dev_net(dev));
5313 /* Flush device addresses */
5314 dev_addr_flush(dev);
5316 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
5319 /* Compatibility with error handling in drivers */
5320 if (dev->reg_state == NETREG_UNINITIALIZED) {
5321 kfree((char *)dev - dev->padded);
5325 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
5326 dev->reg_state = NETREG_RELEASED;
5328 /* will free via device release */
5329 put_device(&dev->dev);
5331 EXPORT_SYMBOL(free_netdev);
5334 * synchronize_net - Synchronize with packet receive processing
5336 * Wait for packets currently being received to be done.
5337 * Does not block later packets from starting.
5339 void synchronize_net(void)
5344 EXPORT_SYMBOL(synchronize_net);
5347 * unregister_netdevice_queue - remove device from the kernel
5351 * This function shuts down a device interface and removes it
5352 * from the kernel tables.
5353 * If head not NULL, device is queued to be unregistered later.
5355 * Callers must hold the rtnl semaphore. You may want
5356 * unregister_netdev() instead of this.
5359 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
5364 list_move_tail(&dev->unreg_list, head);
5366 rollback_registered(dev);
5367 /* Finish processing unregister after unlock */
5371 EXPORT_SYMBOL(unregister_netdevice_queue);
5374 * unregister_netdevice_many - unregister many devices
5375 * @head: list of devices
5378 void unregister_netdevice_many(struct list_head *head)
5380 struct net_device *dev;
5382 if (!list_empty(head)) {
5383 rollback_registered_many(head);
5384 list_for_each_entry(dev, head, unreg_list)
5388 EXPORT_SYMBOL(unregister_netdevice_many);
5391 * unregister_netdev - remove device from the kernel
5394 * This function shuts down a device interface and removes it
5395 * from the kernel tables.
5397 * This is just a wrapper for unregister_netdevice that takes
5398 * the rtnl semaphore. In general you want to use this and not
5399 * unregister_netdevice.
5401 void unregister_netdev(struct net_device *dev)
5404 unregister_netdevice(dev);
5407 EXPORT_SYMBOL(unregister_netdev);
5410 * dev_change_net_namespace - move device to different nethost namespace
5412 * @net: network namespace
5413 * @pat: If not NULL name pattern to try if the current device name
5414 * is already taken in the destination network namespace.
5416 * This function shuts down a device interface and moves it
5417 * to a new network namespace. On success 0 is returned, on
5418 * a failure a netagive errno code is returned.
5420 * Callers must hold the rtnl semaphore.
5423 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
5426 const char *destname;
5431 /* Don't allow namespace local devices to be moved. */
5433 if (dev->features & NETIF_F_NETNS_LOCAL)
5437 /* Don't allow real devices to be moved when sysfs
5441 if (dev->dev.parent)
5445 /* Ensure the device has been registrered */
5447 if (dev->reg_state != NETREG_REGISTERED)
5450 /* Get out if there is nothing todo */
5452 if (net_eq(dev_net(dev), net))
5455 /* Pick the destination device name, and ensure
5456 * we can use it in the destination network namespace.
5459 destname = dev->name;
5460 if (__dev_get_by_name(net, destname)) {
5461 /* We get here if we can't use the current device name */
5464 if (!dev_valid_name(pat))
5466 if (strchr(pat, '%')) {
5467 if (__dev_alloc_name(net, pat, buf) < 0)
5472 if (__dev_get_by_name(net, destname))
5477 * And now a mini version of register_netdevice unregister_netdevice.
5480 /* If device is running close it first. */
5483 /* And unlink it from device chain */
5485 unlist_netdevice(dev);
5489 /* Shutdown queueing discipline. */
5492 /* Notify protocols, that we are about to destroy
5493 this device. They should clean all the things.
5495 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
5498 * Flush the unicast and multicast chains
5500 dev_unicast_flush(dev);
5501 dev_addr_discard(dev);
5503 netdev_unregister_kobject(dev);
5505 /* Actually switch the network namespace */
5506 dev_net_set(dev, net);
5508 /* Assign the new device name */
5509 if (destname != dev->name)
5510 strcpy(dev->name, destname);
5512 /* If there is an ifindex conflict assign a new one */
5513 if (__dev_get_by_index(net, dev->ifindex)) {
5514 int iflink = (dev->iflink == dev->ifindex);
5515 dev->ifindex = dev_new_index(net);
5517 dev->iflink = dev->ifindex;
5520 /* Fixup kobjects */
5521 err = netdev_register_kobject(dev);
5524 /* Add the device back in the hashes */
5525 list_netdevice(dev);
5527 /* Notify protocols, that a new device appeared. */
5528 call_netdevice_notifiers(NETDEV_REGISTER, dev);
5535 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
5537 static int dev_cpu_callback(struct notifier_block *nfb,
5538 unsigned long action,
5541 struct sk_buff **list_skb;
5542 struct Qdisc **list_net;
5543 struct sk_buff *skb;
5544 unsigned int cpu, oldcpu = (unsigned long)ocpu;
5545 struct softnet_data *sd, *oldsd;
5547 if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
5550 local_irq_disable();
5551 cpu = smp_processor_id();
5552 sd = &per_cpu(softnet_data, cpu);
5553 oldsd = &per_cpu(softnet_data, oldcpu);
5555 /* Find end of our completion_queue. */
5556 list_skb = &sd->completion_queue;
5558 list_skb = &(*list_skb)->next;
5559 /* Append completion queue from offline CPU. */
5560 *list_skb = oldsd->completion_queue;
5561 oldsd->completion_queue = NULL;
5563 /* Find end of our output_queue. */
5564 list_net = &sd->output_queue;
5566 list_net = &(*list_net)->next_sched;
5567 /* Append output queue from offline CPU. */
5568 *list_net = oldsd->output_queue;
5569 oldsd->output_queue = NULL;
5571 raise_softirq_irqoff(NET_TX_SOFTIRQ);
5574 /* Process offline CPU's input_pkt_queue */
5575 while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
5583 * netdev_increment_features - increment feature set by one
5584 * @all: current feature set
5585 * @one: new feature set
5586 * @mask: mask feature set
5588 * Computes a new feature set after adding a device with feature set
5589 * @one to the master device with current feature set @all. Will not
5590 * enable anything that is off in @mask. Returns the new feature set.
5592 unsigned long netdev_increment_features(unsigned long all, unsigned long one,
5595 /* If device needs checksumming, downgrade to it. */
5596 if (all & NETIF_F_NO_CSUM && !(one & NETIF_F_NO_CSUM))
5597 all ^= NETIF_F_NO_CSUM | (one & NETIF_F_ALL_CSUM);
5598 else if (mask & NETIF_F_ALL_CSUM) {
5599 /* If one device supports v4/v6 checksumming, set for all. */
5600 if (one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM) &&
5601 !(all & NETIF_F_GEN_CSUM)) {
5602 all &= ~NETIF_F_ALL_CSUM;
5603 all |= one & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
5606 /* If one device supports hw checksumming, set for all. */
5607 if (one & NETIF_F_GEN_CSUM && !(all & NETIF_F_GEN_CSUM)) {
5608 all &= ~NETIF_F_ALL_CSUM;
5609 all |= NETIF_F_HW_CSUM;
5613 one |= NETIF_F_ALL_CSUM;
5615 one |= all & NETIF_F_ONE_FOR_ALL;
5616 all &= one | NETIF_F_LLTX | NETIF_F_GSO | NETIF_F_UFO;
5617 all |= one & mask & NETIF_F_ONE_FOR_ALL;
5621 EXPORT_SYMBOL(netdev_increment_features);
5623 static struct hlist_head *netdev_create_hash(void)
5626 struct hlist_head *hash;
5628 hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL);
5630 for (i = 0; i < NETDEV_HASHENTRIES; i++)
5631 INIT_HLIST_HEAD(&hash[i]);
5636 /* Initialize per network namespace state */
5637 static int __net_init netdev_init(struct net *net)
5639 INIT_LIST_HEAD(&net->dev_base_head);
5641 net->dev_name_head = netdev_create_hash();
5642 if (net->dev_name_head == NULL)
5645 net->dev_index_head = netdev_create_hash();
5646 if (net->dev_index_head == NULL)
5652 kfree(net->dev_name_head);
5658 * netdev_drivername - network driver for the device
5659 * @dev: network device
5660 * @buffer: buffer for resulting name
5661 * @len: size of buffer
5663 * Determine network driver for device.
5665 char *netdev_drivername(const struct net_device *dev, char *buffer, int len)
5667 const struct device_driver *driver;
5668 const struct device *parent;
5670 if (len <= 0 || !buffer)
5674 parent = dev->dev.parent;
5679 driver = parent->driver;
5680 if (driver && driver->name)
5681 strlcpy(buffer, driver->name, len);
5685 static void __net_exit netdev_exit(struct net *net)
5687 kfree(net->dev_name_head);
5688 kfree(net->dev_index_head);
5691 static struct pernet_operations __net_initdata netdev_net_ops = {
5692 .init = netdev_init,
5693 .exit = netdev_exit,
5696 static void __net_exit default_device_exit(struct net *net)
5698 struct net_device *dev;
5700 * Push all migratable of the network devices back to the
5701 * initial network namespace
5705 for_each_netdev(net, dev) {
5707 char fb_name[IFNAMSIZ];
5709 /* Ignore unmoveable devices (i.e. loopback) */
5710 if (dev->features & NETIF_F_NETNS_LOCAL)
5713 /* Delete virtual devices */
5714 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) {
5715 dev->rtnl_link_ops->dellink(dev, NULL);
5719 /* Push remaing network devices to init_net */
5720 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
5721 err = dev_change_net_namespace(dev, &init_net, fb_name);
5723 printk(KERN_EMERG "%s: failed to move %s to init_net: %d\n",
5724 __func__, dev->name, err);
5732 static struct pernet_operations __net_initdata default_device_ops = {
5733 .exit = default_device_exit,
5737 * Initialize the DEV module. At boot time this walks the device list and
5738 * unhooks any devices that fail to initialise (normally hardware not
5739 * present) and leaves us with a valid list of present and active devices.
5744 * This is called single threaded during boot, so no need
5745 * to take the rtnl semaphore.
5747 static int __init net_dev_init(void)
5749 int i, rc = -ENOMEM;
5751 BUG_ON(!dev_boot_phase);
5753 if (dev_proc_init())
5756 if (netdev_kobject_init())
5759 INIT_LIST_HEAD(&ptype_all);
5760 for (i = 0; i < PTYPE_HASH_SIZE; i++)
5761 INIT_LIST_HEAD(&ptype_base[i]);
5763 if (register_pernet_subsys(&netdev_net_ops))
5767 * Initialise the packet receive queues.
5770 for_each_possible_cpu(i) {
5771 struct softnet_data *queue;
5773 queue = &per_cpu(softnet_data, i);
5774 skb_queue_head_init(&queue->input_pkt_queue);
5775 queue->completion_queue = NULL;
5776 INIT_LIST_HEAD(&queue->poll_list);
5778 queue->backlog.poll = process_backlog;
5779 queue->backlog.weight = weight_p;
5780 queue->backlog.gro_list = NULL;
5781 queue->backlog.gro_count = 0;
5786 /* The loopback device is special if any other network devices
5787 * is present in a network namespace the loopback device must
5788 * be present. Since we now dynamically allocate and free the
5789 * loopback device ensure this invariant is maintained by
5790 * keeping the loopback device as the first device on the
5791 * list of network devices. Ensuring the loopback devices
5792 * is the first device that appears and the last network device
5795 if (register_pernet_device(&loopback_net_ops))
5798 if (register_pernet_device(&default_device_ops))
5801 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
5802 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
5804 hotcpu_notifier(dev_cpu_callback, 0);
5812 subsys_initcall(net_dev_init);
5814 static int __init initialize_hashrnd(void)
5816 get_random_bytes(&skb_tx_hashrnd, sizeof(skb_tx_hashrnd));
5820 late_initcall_sync(initialize_hashrnd);